Professional Documents
Culture Documents
Working Group
C4.605
February 2014
MODELLING AND
AGGREGATION OF
LOADS IN FLEXIBLE
POWER NETWORKS
WG C4.605
Members
Jovica V. Milanović, Convenor (GB), Julija Matevosiyan, Secretary (US), Anish Gaikwad, Web
Officer (US); Members: Alberto Borghetti (IT), Saša Ž. Djokić (GB), Zhao Yang Dong (AU), Andrew
Halley (AU), Lidija M. Korunović (RS), Sergio Martinez Villanueva (ES), Jin Ma (CN), Pouyan
Pourbeik (US), Fernanda Resende (PT), Stefan Sterpu (FR), Fortunato Villella (BE), Koji
Yamashita (JP); Corresponding Members: Odin Auer (AR), Karim Karoui (BE), Dimitry Kosterev
(US), Shu Kwan Leung (AU), Dumisani Mtolo (ZA), Samila Mat Zali (MY); Contributors: Adam
Collin (GB), Yizheng Xu (GB); Reviewers: Hans Abildgaard (DK), Jose Conto (US), Marian
Piekutowski (AU), Walter Sattinger (CH), Toshio Inoue (JP), William Hung (GB)
Copyright © 2013
“Ownership of a CIGRE publication, whether in paper form or on electronic support only infers right
of use for personal purposes. Total or partial reproduction of the publication for use other than
personal and transfer to a third party are prohibited, except if explicitly agreed by CIGRE; hence
circulation on any intranet or other company network is forbidden”.
Disclaimer notice
“CIGRE gives no warranty or assurance about the contents of this publication, nor does it accept
any responsibility, as to the accuracy or exhaustiveness of the information. All implied warranties
and conditions are excluded to the maximum extent permitted by law”.
ISBN : 978-2-85873-261-6
Modelling and Aggregation of Loads in Flexible Power Networks
Appendix 2-A Examples of the Effect of Load Modeling on Power System Dynamics ...........98
Appendix 2-B Overview of two exemplary structures used in the measurement based
approach .............................................................................................................................................. 101
Appendix 2-C Combined Measurement and Component based Load Modelling Approach
................................................................................................................................................................ 106
Appendix 2-D Trend in Change in Dynamic Voltage Behaviour with Change in Induction
Motor Ratio or Fault Duration [14] .................................................................................................. 107
Appendix 2-E Additional Procedure of Extracting the Measurements for Deriving More
Reliable Load Model Parameters [30] ........................................................................................... 110
Appendix 2-F Detailed Results of the International Survey on Load Modelling .................... 112
Appendix 3-A Reported Parameters of Existing Load Models ................................................. 123
Appendix 3-B Models of Power Electronics Interfaced Loads ................................................... 132
Appendix 3-C Model of Directly Connected Single-phase Induction Motors ......................... 134
Appendix 3-D WECC Residential Air-Conditioning Stalling Motor Model ............................. 136
Appendix 3-E Model of Distributed Energy Storage Systems [63] .......................................... 137
Appendix 4-A Practical example of measurement based load modelling............................. 139
Appendix 4-B Load models with load self-disconnection ........................................................... 143
Appendix 5-A Example of Load Aggregation Methodology.................................................... 145
Appendix 5-B Generic UK LV Network Configurations and Component Values ................... 153
Appendix 5-C Additional residential load curves and models ................................................. 155
Appendix 5-D Commercial load curves ......................................................................................... 157
Appendix 5-E Reported load curves for different load classes ................................................ 158
Appendix 6-A Aggregated Models of Distributed Generation and Active Distribution
Network Cells for Power System Studies – Literature Overview .............................................. 162
Appendix 6-B Black and Grey-Box Based Dynamic Equivalent Models ................................. 169
Appendix 6-C Dynamic Equivalents for Micro Grids................................................................... 175
Appendix 7 Bibliography on Load Modelling .............................................................................. 179
Appendix 8 List of publications based on the work presented in the WG report ................ 190
Page 2
Modelling and Aggregation of Loads in Flexible Power Networks
EXECUTIVE SUMMARY
The power system research community and industry acknowledge the importance of accurate load modelling
for power system studies, however, many still use typical representation of static loads by the constant
impedance/current/power load models, while dynamic loads, if represented, are usually modelled with an
induction motor (IM) model. The last systematic update of load models was performed in the mid 1990s, since
when significant changes have occurred in the structure, type and composition of loads at all network buses.
General inadequacy of currently used load models was highlighted in several unsuccessful attempts to
reproduce the behaviour observed in recent blackouts during the corresponding “post-mortem” simulations and
analysis.
Over the last several years, there has been a renewed interest in both industry and academia for load
modelling due to appearance of new types of loads, offering increased efficiency and controllability. Different
types of modern non-linear power electronic loads are now responsible for a significant part of the total
demand in almost all load sectors. Furthermore, there are currently no appropriate load models available for
the correct representation of various directly connected and inverter-interfaced micro and small-scale
distributed generation technologies, which, in some of the future network scenarios, may strongly impact real
and reactive power demands and behaviour in future network scenarios, as they would be installed in large
numbers.
In a response to this renewed interest in load modeling, CIGRÉ Study Committee C4 established, in late 2009,
the Working Group (WG) C4.605: “Modelling and Aggregation of Loads in Flexible Power Networks”. The
WG started work in February 2010 with the aim to: i) provide a critical and updated overview of existing
load models and their parameters for power system studies at all voltage levels, and identify types of loads
and load classes for which adequate load models are presently missing; ii) provide a comprehensive overview
of existing methodologies for load modeling, with a critical overview of component based and measurement
based approaches, clearly identifying their advantages and disadvantages; iii) develop a set of
recommendations and step-by-step procedures for load model development and validation, using either
component based or measurement based approaches, or their combination; iv) develop load models for all
typical devices and classes of customers for which there are no existing models and recommend their typical
parameter values and ranges; v) provide recommendations on developing equivalent static and dynamic
models of networks with significant amount of distributed generation, including equivalent models of micro-grids
and active distribution network cells.
This report summarises major results of the WG achieved between February 2010 and February 2013. It starts
with a critical overview of two existing most widely used methodologies for load modelling, the measurement
based and the component based approaches. They are described in detail and their major advantages and
disadvantages discussed.
Following this, the report summarizes existing static and dynamic load models, the load classes these load
models are valid for and load parameters that can be found in literature. The most frequently used static load
models, the exponential, second order polynomial and linear load model, are discussed in detail. A
comprehensive static load model that can be used for large voltage variations as well as a static load model
of induction motors typically used to represent load consisting predominantly of induction motors are also
discussed. Dynamic load models, including exponential dynamic load model, dynamic model of induction motor
and different variants of composite dynamic load model, are discussed as well.
A comprehensive questionnaire on load modeling practices was developed by the WG and distributed during
the summer/autumn of 2010 to more than 160 utilities and system operators from over 50 countries on all
continents. The report summarizes some of the key findings of that questionnaire, based on 97 responses
(60.6% response rate) received by September 2011, and provides the full report on survey in the appendix.
The survey revealed that the constant real and reactive power load model (constant P and Q) is the most
Page 3
Modelling and Aggregation of Loads in Flexible Power Networks
widely used load model for steady state power system studies. It also showed that the static load models are
most commonly used even for dynamic system studies and that only about 30% of utilities and transmission
system operators represent dynamic load by some form of induction motor model. The analysis of existing load
models presented in the introductory part of the report is supported by findings of this survey.
The report further develops a set of recommendations for load model development, for both measurement
based and component based approaches. For the measurement based approach, it provides: i) types of signals
to be measured (instantaneous or RMS values of voltages, currents or powers, frequency, etc.) or data to be
collected, ii) data collection procedure, measurement requirements in terms of sampling rate, duration and
location of monitoring equipment, iii) methodology of handling measured signals and conversion to the required
values (e.g., formulae for conversion of measured voltage and current responses into power responses), iv) data
filtering and processing, including required accuracy of filtering procedures, v) handling data where measured
signals are clearly distorted due to presence of higher harmonics or noise, vi) type of field tests needed, vii)
typical load model structures to be used and suitable parameter fitting procedures, and for the first time, viii)
handling the issues resulting from load self-disconnection following system disturbances. For the component
based approach, the chapter provides: i) types of individual loads that should be considered for data
collection, ii) type of data to be collected, iii) mathematical models of individual devices/loads that should be
considered, iv) methods for aggregating different devices, v) dealing with missing data and vi) representation
of the distribution and subtransmission network.
Due to their number, volume and complexity, system loads connected at medium voltage (MV) and low voltage
(LV) distribution networks, i.e., at bulk load supply points (BLSPs), are generally represented as aggregate load
models in power system analysis. These aggregate load models will include all network components (overhead
lines, cables, transformers, etc.), the actual connected load and, possibly, some micro, small and medium-scale
distributed generation systems, e.g. “micro-generation” connected at LV. The aggregate load model is,
therefore, incorporating a significant “non-load” portion of the network, which must be correctly represented
during the aggregation. One special category of aggregate load models are models of different Load
Sectors, also known as Classes of Customers, or Load Classes, which are generally defined as an aggregation
or collection of different types of loads, or load categories, representing the typical structure and composition
of electrical devices and equipment found in a specific end-use application, where similar activities and tasks
are performed, e.g. in residential or commercial load sector applications. This similarity in performed tasks and
activities usually results in inherent similarities in characteristics and patterns of active and reactive power
demands of end-users within the same load sector, which in turn allows use of similar aggregate load models
for the representation of their aggregate demands.
Aggregate load connected at network BLSPs will typically consist of several load sectors and possibly further
sub-sectors, which must be identified during the aggregation. System loads can be generally grouped into the
three main load sectors (classes of customers): residential, commercial and industrial, and a variety of other,
typically smaller load sectors/classes: public, agricultural, service industry, etc. Variations in type, location and
size of the network and electrical installations where residential, commercial and industrial loads are present,
as well as in the ways these loads are used in the buildings containing them, will introduce further load sub-
sectors, e.g. commercial load sector can be subdivided into office, retail, education and other sub-sectors. Due
to the importance of these load classes for the operation of future power networks, where greater emphasis
will be placed on active participation of small(er) customers, the report describes and discusses in detail
aggregate models for residential and commercial load sectors and corresponding sub-sectors.
Large scale integration of inverter interfaced small generation units with power ratings less than a few tens of
kilowatts in LV networks and Distributed Generation (DG) in MV networks calls for the development of
equivalent mathematical models of Active Distribution Network Cells (ADNC) and Micro Grids (MG) suitable to
represent ADNC and MG in steady state and dynamic studies of large power networks. Since ADNC and MG
properties cannot be adequately modelled using conventional dynamic aggregation methods, the report
Page 4
Modelling and Aggregation of Loads in Flexible Power Networks
suggests exploiting systems identification techniques for this purpose. Considering that there is no current
industrial practice on development or use of this type of aggregate models, the recommendations made in this
report are based on relatively limited academic research experience, presenting a pioneering work in this
field. Two approaches are recommended to derive dynamic equivalents for ADNC and MG. The first one is a
“black-box” modelling approach, based on Artificial Neural Networks (ANN) that tries to exploit the full
response of the ADNC or MG after a disturbance. The second one is a “grey-box” modelling approach,
exploiting the physical behaviour of the different components of the ADNC or MG. Recommended equivalent
models could reduce the complexity of the ADNC and MG models and make them computationally feasible for
application in large power system steady state and dynamic studies.
Finally, the report identifies potential areas for further research and development in the field of power system
load modelling. These include i) real-time load model identification, ii) introduction of new load classes to
reflect significant participation of relatively new load devices, such as electronic-based load, electric vehicles,
storage devices, etc., iii) extension of current ADNC dynamic equivalents to include additional models of DG
units and characterized by increasing flexibility to allow mimicking the interaction with the automation systems
of future smart distribution networks, iv) development of aggregated models for ADNC and MG operated as
integrated energy systems, when a number of energy carriers such as gas, heat, electricity and potentially
hydrogen will be optimally generated and dispatched to supply manifold of responsive and controllable loads,
such as reversible heat pumps, electric vehicles, etc. v) exploiting computer intelligence methods for efficient
power system load model development while taking advantage of the smart metering infrastructure that will
supply large amount of data, vi) setting up pilot installations and experimental facilities for the purpose of
load model validation.
All specific tasks identified in the Terms of Reference of the WG have been completed. Additionally, an
international survey on industry practice on load modelling, involving 97 utilities from all five continents, has
been carried out and its findings summarized in the report. The survey’s results not only complement the work
of the WG and serve a consolidated summary of the present modeling practices in the industry. Additionally,
during the course of the work the members of the WG have widely disseminated the results of the work and
published 1 international journal paper (IEEE Transactions on Power Systems) and 12 international conference
papers. Finally, it should be noted that many regions continue to improve their load modeling practices, most
notably is the Western Electricity Coordinating Council in the USA.
Page 5
Modelling and Aggregation of Loads in Flexible Power Networks
Demand Factor:
The ratio of the maximum demand of a group of loads during a specified time period to the corresponding
total installed power of those loads. (At the transmission level: The ratio between system peak load and sum of
substation peak load.)
Page 6
Modelling and Aggregation of Loads in Flexible Power Networks
Generation:
The total active and reactive power output of a generator, or power plant.
Inverter-based Load:
Inverter based load is the load which uses inverters for the connection or interfacing to the main supply
system such as single-phase and three-phase adjustable speed drives.
Load:
An electrical device or item of equipment, or any combination and number of these, connected in parallel to a
power supply system and specifically designed to consume active power supplied at a dedicated point of
delivery for end-use consumption of electricity. Also denotes actual active and reactive powers consumed by a
load (i.e. electricity demand by a device). In that context, “load” and “demand” are interchangeable terms in
this report.
Load demand at a system bus is the sum of the actual demands of loads connected at the bus, as well as all
loads and losses in the network downstream of that connection point.
Note: As defined above, the, term “load” represents the power system component which during normal operation, is not
generating active power, nor is participating in the transmission or distribution of power. Depending on the point of interest and
type of power system study for which an appropriate load model is required (e.g. analysis of transmission system performance),
the aggregate representation of the load may include not only the connected active power-consuming devices, but also other
distribution and transmission network components present in the aggregated part of the system (i.e. downstream of the point of
aggregation). This is particularly true in the context of “system load”, when the corresponding aggregate load model, may
include transmission/distribution transformers and lines, capacitor banks, various control and voltage regulation equipment, as
well as distributed generation.
Page 7
Modelling and Aggregation of Loads in Flexible Power Networks
Load Bus:
A bus in the modelled system/network, which has a load connected to it and is a net consumer/importer of
active power, and where total active and reactive power demands of the connected load are either known or
predetermined.
Note: When some generation is connected at the load bus (e.g. distributed generation), the local active power demand may be
lower than the local active power generation, e.g. during the minimum loading conditions. In that case, or during this time, the
bus cannot be treated as a load bus, as it is a net exporter of active power (see definition of Active Distribution Network Cell).
Load Characteristics:
A set of parameters and/or functional relationships describing and characterising the electrical behaviour and
response of the load to changes in system operating and loading conditions, including following small and large
disturbances.
Note: The two most important load characteristics for the purpose of this report are variations of active and reactive power
demands over the time and as a function of system voltage and system frequency. However, load models may also include
other characteristics of interest (e.g. fundamental or harmonic currents, total harmonic distortion, efficiency, etc.).
Load Component:
An individual electrical device or item of equipment, or a group of such items, typically used in the same end-
use applications, which consume active and reactive power and respond to variations in voltage and frequency
in a similar way.
Load Factor:
The ratio of the actual consumption of a load within a specified time period (e.g. in kWh) to the consumption
that would result from the continuous maximum demand of that load in the same period. Also denotes the ratio
Page 8
Modelling and Aggregation of Loads in Flexible Power Networks
of the average load (e.g. in kW) supplied during a specified time period to the peak load occurring in that
period. (Load factor is also sometimes referred to as Equivalent full-load hours)
Note: This term should only be used when the actual/average/maximum demands and period of time to which they relate are
specified.
Load Model:
An analytical, mathematical, equivalent-circuit based, physical-component based, or otherwise established or
formulated representation of a load, which can be used for the analysis, prediction or estimation of relevant
load characteristics in power system studies and subsequent analysis of system-load interactions.
Load Structure:
The specification of main, or all important, load model categories, or load components, or end-use types of
loads, which are present in the modelled group of loads (e.g. in an aggregate load, or in a model of a specific
load class/sector) connected at one point of delivery for end-use consumption of electricity.
Micro-generation:
A small distributed generating system, representing one or several generating units, with a total rated power
output of typically up to 50kW-100kW, connected to a low voltage distribution system through a power
electronic-based or other control interface, and typically utilising some of the low-carbon technology (e.g.
renewable energy sources or high efficiency fossil fuel-based combined heat and power applications).
Page 9
Modelling and Aggregation of Loads in Flexible Power Networks
System Load:
The total active and reactive power demanded from a power system, or a particular part of the system,
including corresponding losses due to the transmission and distribution of electricity.
Page 10
Modelling and Aggregation of Loads in Flexible Power Networks
Chapter 1 Introduction
Load modeling is one of the most important aspects of power system modeling. Most of the load models used
today were developed several decades ago, and have not been adequately updated after the subsequent
changes in load structure and load characteristics. The last systematic update of load models, on an industry-
wide level, was performed in the mid 1990s. Additionally, load models and their parameters currently used by
utilities and system operators for power system analysis are generally not in public domain, and there is a level
of uncertainty regarding industry acceptance of research efforts in this area. There have also been some
recent, and concerted efforts in the development of load models in some regions (e.g. Western Electricity
Coordinating Council [1]), however, these efforts have focused primarily on the specific region in which they
have been developed.
Although the majority of current power system research and industry acknowledges the importance of accurate
load modelling for power system studies, they still use typical static load models (constant
impedance/current/power) while dynamic load models, if used, are usually in the form of standard induction
motor (IM) models.
Load characteristics have significant influence on both steady state and dynamic performance of power systems
[2–4]. Accordingly, correct analysis of power systems require accurate load models, together with appropriate
representations of generation, transmission and distribution parts of the system. Load modelling, however, is not
a simple task, as there is a significant number of factors that should be taken into account such as the diversity
in types and characteristics of the loads, the lack of information on the load structure and the difficulties during
the assessment and validation of the load model. Furthermore, spatial and temporal load variability needs to
also be taken into consideration to assess system behaviour at different times of the year and different times of
the day, as well as in different demographic and geographic regions. Moreover, aggregate load models at
MV and HV bulk supply buses adopted for power system studies include implicitly the distribution transformers,
shunt compensation and the distribution network feeders, often without accounting appropriately for dynamics
of operation of tap changing transformers and other voltage regulators that may be deployed at lower
voltage levels.
General inadequacy of currently used load models was highlighted in several unsuccessful attempts to
reconstruct recent blackouts in the corresponding “post-mortem” simulations and analysis. During the power
system stability analysis, the emphasis is mainly placed on modelling power generating units, while load models
are regarded as of secondary importance, although the influence of load representation on the stability was
recognized a long time ago. Power system engineers began to pay more attention to the load modelling since
the Swedish blackout of 1983, as inappropriate representation of system loads has usually led to the
discrepancies between the recorded and simulated system responses.
The computer simulation of the power system using appropriate models of its components is one of the most
important tools to understand the system dynamic behaviour and guide its operation and planning. Load
models play a vital role in these simulations. Different load models may lead to different simulation results. Too
conservative load model assumptions may lead to over-expenditure or situations that require exorbitant levels
of investment to solve a problem that may be highly unlikely, whereas too simple and optimistic a load model
may result in serious problems not being identified and thus leaving significant vulnerabilities in the system.
These effects of the load model on the power system dynamics have been well documented all over the world
and raised serious concerns related to the load modelling work. Some of the examples are discussed below.
On August, 10th, 1996, a serious power failure occurred in the Western Systems Coordinating Council (WSCC)
system resulting in its break-up into four islands, with loss of 30,390MW affecting 7.49 million customers in
Western North America. The system lost its stability with increased oscillations. The post-fault simulation based
analysis of the event carried out by Bonneville Power Administration (BPA) and using the WSCC database
showed a very stable system. To match the simulation results with the measurements, the experts from BPA
Page 11
Modelling and Aggregation of Loads in Flexible Power Networks
modified the Pacific HVDC Intertie model, modelled Automatic generation Control (AGC), blocked certain
turbine-governor models and also made changes to the voltage controls on the Lower Columbia generators.
After all the above changes had been made, however, the simulation results still showed more damping in the
system than the actual situation. In these simulations, the loads in Northwest and Canada were initially
represented primarily by a constant current real power and constant impedance reactive power load model.
When the load models were changed to a combination of the induction motor models and various static loads,
the simulated and measured responses showed very good agreement [5]. Similar analysis carried out by
Powertech Labs Inc. concluded “Our analysis has shown that by far the two most critical modelling elements in
reproducing this oscillatory disturbance are load characteristics and generator excitation controls. Both are
relatively uncertain (particularly load) and changes to either can profoundly change the system response”.
More information about this outage is given in Appendix 2-A.
As mentioned above, the inappropriate load model may cause quantitative analysis errors, even if it does not
give a completely wrong judgment on the system stability. In the study of dynamic stability, in the Taiwan
Power System, the effects of load models on the critical modes during an unstable low frequency oscillation
were investigated. When the complete dynamic load model was used in the simulation, the results were much
worse than the field measurements. When the composite load models were used, the results were in close
agreement with the field-measurements. When the exponential load model was used, the results could not
predict at all the undamped oscillations in the system [6].
In the study of transient stability of the power system in the Northeast of China following a three-phase short
circuit test, three different load models were applied and different simulation results were observed compared
to measurements [7]. Further details are provided in Appendix 2-A.
Finally, in the study of voltage stability in the Argentinean power transmission system, the static load model
could not predict the voltage collapse and gave a more optimistic analysis result, while the composite load
models with the induction motor models included captured very well the observed voltage dynamics.
In summary, different load models can make quite a difference to simulation results. Since the digital simulation
is the fundamental tool of power system analysis and control, developing the appropriate load models is
essential for ensuring secure and economic operation of power system. Examples of load modelling
inadequacies on power system analysis are given in Appendix 2-A.
In addition to the real life examples discussed above, the recent renewed interest in load modelling is also
fuelled by the appearance of new types of loads, offering increased efficiency and controllability. Different
types of modern non-linear power electronic loads are now responsible for a significant part of the total
demand of the residential load sector in many power systems. Similarly, there are currently no appropriate
load models available for the correct representation of various directly connected and inverter-interfaced
micro and small-scale distributed generation technologies, which, in some of the future network scenarios, will
strongly impact real and reactive power demands, as they will be installed in large numbers.
In a response to this renewed interest in load modeling, CIGRÉ Study Committee C4 established in late 2009
the Working Group (WG) C4.605 on Modelling and Aggregation of Loads in Flexible Power Networks. The
group started its work in February 2010, with an objective to: i) provide a critical and updated overview of
existing load models and their parameters for power system studies at all voltage levels, and identify the
types of loads and load classes for which adequate load models are presently missing; ii) provide a
comprehensive overview of existing methodologies for load modeling, with a critical overview of component
based and measurement based approaches, clearly identifying their advantages and disadvantages;
iii) develop a set of recommendations and step-by-step procedures for load model development and
validation using either component based or measurement based approaches, or their combination; iv) develop
load models for all typical devices and classes of customers for which there are no existing models and
recommend their typical parameter values and ranges; v) provide recommendations on developing equivalent
Page 12
Modelling and Aggregation of Loads in Flexible Power Networks
static and dynamic models of networks with a significant amount of distributed generation, including equivalent
models of micro grids and active distribution network cells.
The main objective of WG C4.605 is therefore to identify existing load models and to provide guidelines for
development of new and more realistic generic static and dynamic models of power system loads for the
analysis of transmission and distribution networks (i.e. networks with voltage levels ranging from 6kV, or 11kV,
to 400kV, or above). Generally, the network buses where considered loads are connected are assumed to be
predominantly passive (i.e. without significant generation connected to them). A part of the work, however, is
dedicated to the representation of “load” models at network buses which during a specified period may
contain a high active component, i.e., act as net exporters of active power to the network (due to, e.g., a high
penetration of distributed generation).
The intended applications of the load models presented in this report include both, steady state analysis and
small/large disturbance studies of power networks. The emphasis is on the aggregate load models for
representing power system loads at the bulk supply points. Therefore, single phase loads and load imbalance
are not considered. Only the fundamental frequency component is taken into account in load model derivation.
In order to inform the WG activities and to establish a critical overview of the current approaches to load
modelling by the international industry, a comprehensive questionnaire on load modeling practices was
developed by the WG and distributed during the summer/autumn of 2010 to more than 160 utilities and
system operators from over 50 countries on all continents. The results of the survey, based on 97 responses
(60.6% response rate) received by September 2011 are summarized in this report. The survey revealed that
the constant real and reactive power load model (constant P and Q) is the most widely used load model for
steady state power system studies. It also showed that the static load models are most commonly used even for
dynamic system studies and that only about 30% of utilities and transmission system operators represent
dynamic load by some form of induction motor model. The full analysis of the results of the survey is given in
Appendix 2-F.
In addition to this introductory chapter, the report contains further seven chapters and a number of appendices,
where in-depth analysis, case studies and a range of tables with different load model parameters are given.
Chapter 2 summarizes and discusses the existing methodologies for load modelling, with a detailed discussion
of the component based and measurement based approaches. Chapter 2 also describes briefly the possibility
to combine the measurement based and component based load modelling approaches in order to obtain more
accurate load models.
Chapter 3 provides an overview of existing load models divided into two groups - static and dynamic load
models. The chapter also lists the most frequently used load models and load classes for which these models are
valid. The areas of application of load models and identified load model parameters from the existing open
source literature are also specified. At the end of the chapter, some of the results of the survey on load
modelling practices are presented and discussed.
Chapter 4 develops a set of recommendations for load model development, for both measurement based and
component based approaches.
Chapter 5 provides models of different Load Sectors, also known as Classes of Customers, or Load Classes,
which are generally defined as an aggregation or collection of different types of loads, or load categories,
representing the typical structure and composition of electrical devices and equipment found in a specific end-
use application, where similar activities and tasks are performed.
Chapter 6 presents equivalent mathematical models of Active Distribution Network Cells (ADNC) and Micro
Grids (MG), suitable for their representation in steady state and dynamic studies of large power networks.
Page 13
Modelling and Aggregation of Loads in Flexible Power Networks
Chapter 7 brings together major conclusions of the report and provides suggestions for further research and
development in the area of power system load modelling.
Finally, the Appendices to this report provide supporting information and various examples and details of the
material discussed in the main body of the report.
Page 14
Modelling and Aggregation of Loads in Flexible Power Networks
2.1 Introduction
In order to identify the current industry practices for development of power system load models, and to support
findings coming from a literature review, a questionnaire was developed and distributed to more than 160
utilities and system operators from over 50 countries on all continents.
One of the questions contained in the survey was intended to clarify the adequacy of presently available load
models. The responses indicated that most engineers are satisfied with the load models available in existing
software packages (85% on average) and have not developed their own models for system studies (78% on
average).
The survey also revealed that utilities and system operators updated load model parameters used in
simulations relatively frequently. In 41% of cases, the load model parameters were updated within the last five
years.
Taking into account these findings, it can be concluded that many utilities and system operators haven’t used
load models different from those available in commercial software, but have changed parameters applied to
those models in order to improve the accuracy of the simulation results.
During the past few decades, data capture and storage capabilities of various measurement devices have
significantly improved and the number of measurement devices deployed in power systems has substantially
increased. This has unlocked greater possibilities for the measurement and capture of data suitable for load
model development and in particular, an opportunity to improve the quality of model parameters applied. The
responses to the questionnaire revealed that the power industry is taking full advantage of available
monitoring systems for such purposes. It has been reported that in over 50% of cases load model parameters
are being identified based on field measurements.
This chapter provides a critical overview of existing methodologies for load model development, with a
detailed discussion on the component based and measurement based approaches, clearly identifying their
advantages and disadvantages. Furthermore, considering the findings of the survey, the chapter discusses
existing approaches to the load modelling that are based on the use of measurement devices, and the methods
used for identifying load model parameters from field data.
Note: The responses to questions Q4, Q5, Q7 and Q8 of the questionnaire completed by utilities and system operators are
summarised above. Chapter 3 and Chapter 6 will provide further information on responses to questions Q1, Q2, Q3 and Q9,
respectively.
Spatial and temporal diversity in the types of loads connected to the power system. Load variations are
stochastic [7] and significantly different measured data are typically captured for the same load at
different times, including time of day, day of week and across seasons.
Highly non-linear and discontinuous behaviour of some loads. Some examples are the stalling of residential
air conditioners, high sensitivity to dynamic stability limits causing tripping of motor loads, as well as self-
disconnection of power electronic loads above or below certain voltage thresholds.
Page 15
Modelling and Aggregation of Loads in Flexible Power Networks
Lack of accurate information about the structure and composition of aggregate loads on a continuous basis.
As previously mentioned, measured data can be used for identification of the fraction/percentage of each
load component in the load mix, as well as for derivation of load model parameters. Measurements required
for the former are mainly intermittently metered data (e.g. captured every 5 minutes [14]), while those
required for derivation of load model parameters are usually continuously measured data (or data captured
at a high sampling rate [14]). Considering that end-use survey and literature can also be taken advantage of
identifying load components and their composition, as well as to derive load model parameters, general load
modelling procedures are classified into six different approaches as shown in Figure 2-2.
All possible approaches that can be derived from Figure 2-2, taking into account the different paths that can
be taken from the top to the bottom of the diagram, are briefly described below:
C E:
This approach is applied to a load model which consists of multiple (more than two) components, e.g. the ZIP
load model and the composite load model, or a combination of an induction motor model and the ZIP load
model. After the load model structure and composition are estimated using the intermittently metered data, the
load model parameters are derived from the continuously measured data.
Page 16
Modelling and Aggregation of Loads in Flexible Power Networks
C D:
This approach also applies to a load model which consists of multiple components. After the load model
structure and composition are estimated using the intermittently metered data, the load model parameters are
derived from the information/knowledge available in literature, or from results coming from end-use surveys.
A B C
Assumption of the multiple Assumption of single load Assumption of the multiple
load model structure using model structure load model structure using
end use survey or literature E.g. Exponential static model measurements
D E
Derivation of the load model Derivation of the load model
parameters using survey or literature parameters using measurements
A E:
As for the two preceding examples, this approach is applied to a load model which consists of multiple
components. After the load model structure and composition are estimated from the result coming from end-use
surveys or literature, the load model parameters are derived from the continuously measured data.
A D:
This is another approach which applies to a load model consisting of multiple components. After the load model
structure and composition are determined based on the result of end-use surveys or literature, the load model
parameters are then also derived from the same information and knowledge base.
B E:
This approach is applied to a load model that ultimately consists of only a single model component, e.g.
exponential load model, or constant power/current/impedance load model. The load model
structure/composition is not required (it is assumed), and the load model parameters are then derived from (i.e.
matched with) continuously measured data.
B D:
This alternative approach applies to a load model having a single model component. Again, the load model
structure/composition is known or assumed, but the load model parameters are derived from the result of end
use surveys or literature.
The results of the international survey conducted by the WG [8] relating to applied methodologies for load
model developments, reveal some interesting information about the current practices of utilities and system
operators. Approximately 50% use measurement data, while about 40% of them use end-use surveys or
literature (see Figure 2-F-4). In addition, more than half use load models which consists of only single model
component. Therefore, B E and B D appear to be the two most popular (most widely used) load
modelling approaches presently being applied.
Collecting adequate information and data from measurements alone is difficult. This is because in large
interconnected bulk power systems, even the loss of a major power plant or circuit may lead to only slight
changes in voltage and frequency. Moreover, such large disturbances do not occur frequently. Intentionally
applied disturbances, where this is even possible, generally cannot be large or repeated on regular basis.
Page 17
Modelling and Aggregation of Loads in Flexible Power Networks
Given these difficulties, many utilities have used simplistic load models comprising voltage-dependent static
polynomial or exponential load representations. The former is essentially a combination of constant impedance
(Z), constant current (I) and constant power (P) loads (commonly referred to a ZIP load model), while the latter
expresses power dependence with voltage as an exponential function with corresponding exponent.
Such models can be used for the analysis of system steady state performance, but they are certainly not an
appropriate for correct load representation for dynamic studies, particularly if there is a high penetration of
induction motor loads (e.g. residential air conditioners). More recently, some utilities (mainly in the USA) have
used more detailed load models that capture both the static and dynamic responses. Capturing the dynamic
response of loads is critical for stability analysis, especially for simulating slow voltage recovery phenomenon
due to the stalling of residential air conditioners (single-phase IM loads) following a system disturbance [9],
[10]. The process of identifying a correct bulk system load model (i.e. a correct aggregate load model) can be
broken down into the following problems:
Identifying a load model that can represent the structure of the given load as a combination of static
and dynamic load components.
Determining the percentages of static and dynamic load components and their parameters.
It should be pointed out that the above two challenges are related to not only the combination of static and
dynamic load models, but also to any other load model which consists of multiple load components, such as the
ZIP load model.
To overcome the above two issues and to obtain data on load characteristics, two approaches have been
predominantly used by the industry in the past [11]:
Component based approach (bottom-up or knowledge-based approach)
Measurement based approach (top-down or behaviour-based approach)
In general, the component based approach is applied to a load model which consists of multiple model
components, e.g. the ZIP load model and the composite load model in parallel. The measurement based
approach can be applied to load models consisting of single and multiple components
Page 18
Modelling and Aggregation of Loads in Flexible Power Networks
Metered demand at the load bus, which is typically available at least every hour (in some cases more often),
can be used to determine load class split. This can be combined with laboratory test results for individual loads
using AI (Artificial Intelligence) techniques such as fuzzy regression method [14] or similar. Further discussion and
more details of this are provided in Chapter 4.
Transmission bus
Distribution bus
95% 5% 0% 0% 0% 0% 0% 0% 100% 0%
LM SM Z P I LM SM Z P I
Load characteristics
Figure 2-3: An Example of a Component Based Load Modelling Approach (adopted from [12]).
After identifying load classes, the next step is to derive load components and their percentage contribution
within each load class. Each load class contains typical load components that account for the majority of the
power consumed by end users. For example, lighting, air conditioning, space heating, water heating and
refrigeration. The challenge is to identify percentage contributions of each load component within the
considered load class. Obtaining this information, which is different for each network and geographic location,
and which also changes with time, is typically a time-consuming and complex process, and essentially relies on
customer surveys. To overcome this issues, a simpler approach was devised in [15]. It requires the following
information: i) Typical load composition of each load class (fractions of load consumed by each load
component); ii) Combination of load classes at each bus.
Much of the data requirements were determined and documented in [14]. As one of the latest data sets, the
load component/category percentages for different load classes for various geographic regions within the
Western Electricity Coordinating Council (WECC) in the United States are currently provided by the WECC's
Load Modeling Task Force (LMTF) [15-16]. The data was compiled for both summer and winter seasons. Though
approximate, reference [14] is the most complete source of information available for the utilities in WECC and
it is also publicly available. An Asian country also provides a fraction of typical load components for residential
commercial and industrial areas based on end use surveys as shown in Figure 2-4 [17]. Examples of
composition for different load classes are given in the Appendix 2-B.
Although the above approach might be applied to any region, typical compositions can differ. According to the
most recent worldwide survey [8], there seems to be no discrimination in modelling different load classes,
except in US and Oceania power systems. In the case of Oceania, industrial load is simply distinguished from
other loads by inclusion of induction motors.
A promising approach is the use of models representing actual electrical circuits of all load components, which
are divided into only five general load categories (see Figure 2 3) based on their electrical characteristics (and
not on their end-use). As circuit-based models allow for more accurate representation of load current/power
responses to voltage variations, this approach provides a more detailed aggregate load model [18], [19]. For
the approach presented in [12], each load component has its specific static and dynamic characteristics. Load
characteristics emulate electrical behaviour of the load component relative to changes in voltage and
frequency. Load characteristics comprise static (constant impedance, constant current and constant power) and
dynamic (one or more types of induction motor) mathematical models. For example, resistive load components,
Page 19
Modelling and Aggregation of Loads in Flexible Power Networks
such as electrical heaters and incandescent lamps, are modelled as constant impedance loads, while inverter-
based load components are modelled as constant power loads unless the voltage drop at the load bus is
significant.
Industries
Motors , 69% 8% 11% 12%
443TWh/year
Residential
273TWh/year 37% 21% Heater, 25% 18%
Commercial
194TWh/year 41% Lamps, 34% Others , 25%
Total
910TWh/year 53% 17% 13% 17%
Figure 2-4: Typical Load Components for Typical Load Classes, (adopted from [17]).
In order to correctly aggregate load in medium-voltage networks, feeders transformers and reactive power
compensation equipment should be included into the aggregate load model. Therefore, as shown in Figure 2-3,
the composite load model not only consists of the equivalent load components, but also incorporates the
substation, on-load tap changing transformer (OLTC) and other downstream equipment. It follows that the
changes in both load composition and network configuration will cause changes in the characteristics of an
aggregate load fed from a bulk supply point.
Page 20
Modelling and Aggregation of Loads in Flexible Power Networks
If the information on load model structure is obtained for one substation, the same load model structure
and composition cannot be directly applied to another substation.
Combinations of load classes and load composition data are not normally used by, nor available to
power system operators and planners. Large-scale load surveys are usually required to gather data
of sufficient quantity and quality.
If a new or undefined load type, which does not belong to any previously defined load component, is
connected to the system, this will result in an error when identification of load model parameters is
performed.
Even if the fraction of each load component is the same, their parameters can be different and vary
widely depending on such variables as age, the manufacturer, end-use application etc. An example is
the inertia constant of small induction motors.
In the case of transmission system operators, whose assets are typically associated with voltages at 110
kV or higher, ownership and location of load devices in customer facilities may not be directly
accessible. This can make it difficult for transmission system operators to apply the component based
load modelling approach.
Bus Voltage
× Medium Voltage Level
Feeders
Disturbance Current
Recorder
Figure 2-5: Outline of an Example Digital Disturbance Recorder (adopted from [26]).
Page 21
Modelling and Aggregation of Loads in Flexible Power Networks
The parameters of the load model are estimated by fitting measured data to the assumed model structure
using parameter identification and curve fitting techniques. Complex estimation techniques may be needed to
obtain parameters for more complicated models. This may also be true depending on the types and number of
disturbances in recording data sets. The process of identification involves identifying a suitable mathematical
model, i.e. a performance (objective) function, and appropriate load model parameters, so that it can replicate
the dynamic response of the loads during and following a disturbance. This is done by observing the
relationship between the change in voltage and/or frequency with the corresponding changes in load active
and reactive power demand. Figure 2-6 shows a representative flowchart of the measurement based load
modelling approach. Note that Figure 2-6 is an example (generic) flowchart, and that various derivative
flowcharts have also been used for the measurement based approach (See Chapter 4).
Step 1 Data Collection
No
Are parameters adequate?
Yes
Step 6 Selection of derived Load Model Parameters
Step 1
Collect time stamped system disturbance data (time-domain voltages and currents for each phase and, if
possible, frequency, active and reactive power). The necessary characteristics of an event that make it suitable
for load modelling are described in Appendix 2-B.
A typical location for data collection is the secondary side of a medium-voltage transformer (Figure 2-5)
supplying radially connected loads.
Step 2
Signal processing techniques, such as a Discrete Fourier Transform (DFT)-based signal processing algorithm [25],
are applied in order to calculate fundamental components of voltage, current, active power and reactive
power. Typical sampling rates for different devices are provided in Table 4-3 in Chapter 4. It should be noted
though that these algorithms inherently result in some filtering of the input data.
Step 3
Select a suitable load model and the corresponding load model structure. The initially adopted load model
structure may be changed in Step (5) if appropriate load model parameters cannot be obtained.
Step 4
Run an optimisation routine. Nonlinear optimisation techniques that minimise/maximise the performance
(objective) function are normally used to determine the parameters of the load model.
Given the performance (objective) function including model parameters to be estimated, the values of the
parameters are selected through an optimisation process such that the error between the measured response
and model response is minimised. Different optimisation techniques have been used in the past to estimate the
parameters, e.g., a genetic algorithm [27], support vector theories [28], simulated annealing [29], etc.
Page 22
Modelling and Aggregation of Loads in Flexible Power Networks
As with any optimisation process, the user has to specify initial values of the parameters that need to be
estimated. It is very important, therefore, to select reasonable initial values and their bounds, as this may affect
convergence and accuracy of the optimisation routine and the final values of the estimated parameters.
Step 5
Validate the derived load model using commercially available or otherwise developed time-domain simulation
tools. The simulated power responses coming from the proposed model are compared with the actual recorded
active and reactive power responses. If the responses do not satisfy pre-defined accuracy criteria, the
procedure should be repeated by selecting different initial estimates, including changing the assumed load
model if deemed appropriate.
Step 6
If appropriate model parameters are obtained, they are ultimately accepted as the final load model
parameters for implementation in system studies. If not, another load model structure is selected and
corresponding performance function derived and the procedure is repeated from Step 3. If however, an
appropriate load model cannot be found, it may be that the measured (or conditioned/filtered) data are not
suitable for use in the measurement based approach. Alternate data should be obtained and the procedure
reapplied on the new data set.
The generic flow chart describing the measurement based approach and shown in Figure 2-6 can be generally
applied to derive any load model. This approach has been implemented for fitting parameters to an
exponential load model using different tools developed by different companies, e.g. EPRI [30], CRIEPI [31] and
TRACTEBEL Engineering [32]. The same approach has been recently used to fit parameters to a composite load
model using the Load Modelling Parameter Derivation (LMPD) routine developed by EPRI.
The measurement based approach discussed above may be modified depending on the load models and the
specifications of available measurement data. Two examples of slightly modified approaches are given below:
The performance function can be modified to include a large number of events/sets of data [33]. The method
divides available data into two groups: 1) Data for deriving load model parameters, 2) Data for validating
load model parameters. In this case, the flow chart in Figure 2-6 would change, as there would be no need to
repeat the estimation process in Steps 4 and 5.
If the measurement recording time is too long for the type of the load model that needs to be developed (e.g.
a load model particularly intended for voltage stability studies or for angular stability studies) and excess
data exists within the data set, another activity can be included between Step 3 and Step 4 of Figure 2 6 [34]
(See Appendix 2-E) The additional step is to extract/shorten suitable measurement data from the overall data
set in a way that makes it suitable for deriving load model parameters for the intended model.
Page 23
Modelling and Aggregation of Loads in Flexible Power Networks
2.5 Summary
This chapter provided a critical overview and clear identification of advantages and disadvantages of the two
most widely used methodologies for load modelling, being the component based and measurement based
approaches. The main advantage of the component based approach is that the load model can be developed
from customer survey data on load composition (or even based on the data form literature) at each load bus
and known characteristics of participating load components. The main disadvantage is the difficulty of
establishing accurate load composition and its variation with time at any given bus. The main advantage of the
measurement based approach is the use of recorded data from the actual system without the need to know
actual load composition. The main disadvantage is usually the lack of appropriate (large and sufficiently time
varied) disturbance measurement data required for use in load model validation and parameter estimation
processes.
Page 24
Modelling and Aggregation of Loads in Flexible Power Networks
3.1 Introduction
As outlined in Chapter 2, there are two main approaches for the development of load models being the
component based and measurement based approaches. The first one implies a knowledge of the relevant
modelling parameters applicable to a large number of individual load components, as well as the relative
participation of those components in the total load at a given bus. The measurement based approach, on the
other hand, does not require any knowledge about load composition and instead, relies on field measurements
to develop load model(s) to represent aggregate loads at power system buses. For either approach, it is
essential that the end result is a load model that accurately represents the actual load behaviour at the system
bus.
This chapter presents the most frequently used load models, irrespective of the approach used for their
development. Existing load models can be conveniently divided into two basic groups - static and dynamic.
Static load models include exponential, polynomial, linear, comprehensive, static induction motor and power
electronic-interfaced models. Dynamic load models include exponential dynamic load model, dynamic induction
motor (IM) models, transfer function IM model, composite, distribution, bulk power bus load and distributed
energy storage system (DESS) models.
Static load models describe the relationship between active and reactive power drawn by given load as an
algeblic function of voltage and frequency. To model the dynamic properties of the load its dynamic model can
be formulated independently from its static load model, or, if the dynamic load model is known, the static load
model can be easily derived from the available dynamic load model. However, the opposite is not true. The
dynamic load model cannot usually be formulated from the known static load model. Therefore, in case of
static load, “model of the static load” is the same as the “static load model”, while both “static load model” and
“dynamic load model” can be formulated in case of the “dynamic load”. The adjectives “static” and “dynamic”
refer both to type of studies and to the types of load models used in those studies. Static studies, such as power
flow, (or load flow) refer to steady state analysis. Dynamic studies refer to the full range of stability analyses
[11] involving the transient responses of the system including the connected load, typically following
disturbances of various types. Static loads are assumed to respond instantaneously to a change in supply
voltage and/or frequency. Dynamic loads exhibit time-dependent responses, which are determined by the
previous states/conditions of both the system and the load itself. The dynamic response is also influenced by the
interactions and exchange of energy between the system and the load during and after the transition from the
previous state/condition to the next. Dynamic load models are usually described in differential equation form
relating real and reactive power with voltage and frequency. It should be pointed out, though, that both static
and dynamic load models can be used in dynamic studies.
In reality, there are no “static loads”, as all loads will respond to a large disturbance or change in supply
voltage/frequency in a finite way, taking some time for the transition from pre-disturbance to post-disturbance
states. However, “dynamic loads” may be expressed in the form of a “static load model” for: i) steady state
power system analysis, when the changes of system conditions having an influence on the load characteristics
are either very slow or small (can be assumed as constant in the considered time interval); ii) When a response
of the modelled load to a voltage and/or frequency changes is very fast (e.g. cannot be captured by the
measuring equipment); iii) If interest is in the load responses after the initial transient. An example of this
duality is induction motor load, for which both static and dynamic load models can be formulated and used in
the corresponding power system studies.
The chapter also lists most frequently used load models and load classes for which these models are valid. The
areas of application of load models and typical load model parameters identified from the existing open
Page 25
Modelling and Aggregation of Loads in Flexible Power Networks
source literature are also presented. At the end of the chapter, some of the results of the survey on load
modelling practices are presented and discussed.
Q fQ (U , f ) (3.2)
Static load models are mostly used for representing resistive load devices, lighting, general residential load
and other similar aggregate loads that lack the participation of large induction motors and electrical drives in
the overall load mix. They are most often implemented in power flow calculations and voltage stability studies.
Existing static load models can be divided into groups as shown in Figure 3-1, being exponential, polynomial,
linear, comprehensive, static induction motor and power electronic-interfaced load models.
The generic mathematical formulations of a dynamic load model are the same as (3.1) and (3.2) with the
exception of the inclusion of time dependence in the right hand side of the equality sign in (3.1) and (3.2).
Dynamic load models are typically used for representing the loads having a significant participation of
induction motors and electrical drives. Dynamic load models can be classified into the following groups:
exponential, induction motor (IM), transfer function based IM model, composite, distribution, bulk power bus
load and distributed energy storage systems (DESS) models.
Page 26
Modelling and Aggregation of Loads in Flexible Power Networks
kqu kqf
U f
Q Qn (3.4)
Un fn
where P and Q are the real and reactive power drawn by the load at voltage U and frequency f, Pn and Qn
are the real and reactive power drawn under rated voltage (Un) and frequency conditions (fn ), and the
exponents kpu, kqu, kpf and kqf describe the change in load demand in response to variations in supply voltage
and frequency away from nominal, respectively. Parameters of the exponential load model (and many of the
other models described in this chapter), as reported in available open access literature, are summarised in
Appendix 3-A.
A load dependence on frequency is, however, often neglected since voltage changes are much more frequent
and more pronounced than the changes in system frequency. With this simplification, (3.3) and (3.4) become:
k pu
U
P Pn (3.5)
Un
kqu
U
Q Qn (3.6)
Un
Simulation of small power systems, with a small equivalent inertia, or power systems dominated by slow
acting governor controls (typical of hydro generators relative to thermal plant). In both cases, system
frequency may vary by several or more percentage points following credible contingency events.
Parameters kpu and kqu represent the partial derivatives of real and reactive power with respect to voltage in
the vicinity of rated voltage, respectively [35], or real and reactive power sensitivities to voltage [36]. These
parameters indicate the magnitude of real and reactive power changes in percent for a one percent in voltage
in the vicinity of rated voltage [37]. If the voltage exponents in (3.5) and (3.6) are set to 0, 1 or 2, the load
exhibits constant power, constant current or constant impedance characteristics, respectively. Although these
load types predominantly have theoretical significance, they are and have been the most widely used load
models in power system studies. This has been confirmed through the findings of the international survey carried
out by this WG. Some resistive load devices, such as hot plates, heaters etc. are typically modelled as constant
impedance load [11], [38], [39].
Frequency dependency terms in (3.3) and (3.4) can be modified at constant voltage Un by Taylor series
expansion since the frequency change is much smaller than voltage. The alternate form of the model is obtained
[35].
k pu
U
P Pn 1 k pf f (3.7)
Un
kqu
U
Q Qn 1 k qf f (3.8)
Un
where f represents the relative frequency change in frequency (f- fn) / fn.
Page 27
Modelling and Aggregation of Loads in Flexible Power Networks
dQ
Qc Pn Qn (3.10)
dP
Note that positive Qc in this equation denotes capacitive load, while positive Q denotes inductive load. The
method requires that the compensation can be assumed to be constant.
U 2 U
P Pn p1 p2 p3 (3.11)
U n Un
U 2 U
Q Qn q1 q2 q3 (3.12)
U n Un
This model is also called the “ZIP model”, since it consists of constant impedance (Z), constant current (I) and
constant power (P) load components. Parameters p1 and q1 represent the relative participation of constant
impedance load, p2 and q2 the relative participation of constant current load, and p3 and q3 relative
participation of constant power load. Load participation of every load component (Z, I, P) in the total load is in
the range from 0 and 1 p.u. such that their overall sum is 1 p.u. This variant is called the “constrained ZIP
model” [38]. In another variant, the individual pi and qi parameters can be larger than 1 p.u. and/or less than
0, but their sum must still equal 1 p.u. Although the parameters of this variant may not look intuitive, the model
may be more accurate than the constrained ZIP model. As such this variation is often referred to as the
“accurate ZIP model” [37].
One frequently used form of the polynomial load model with frequency dependence taken into account [11] is:
U 2 U
P Pn p1 p2 p3 1 k pf f ,
3
U n
p i 1 (3.13)
Un i 1
U 2 U (3.14)
Q Qn q1 q2 q3 1 kqf f ,
3
U n
q i 1
Un i 1
Page 28
Modelling and Aggregation of Loads in Flexible Power Networks
The variant is analogous to that shown in equations (3.7) and (3.8), but with additional terms for voltage
dependence allowing participation of more than just one response characteristic.
U 3
P Pn a0 a1 , p i 1 (3.15)
Un i 1
U 3
Q Qn b0 b1 , q i 1 (3.16)
Un i 1
The research presented in [41] showed however that reactive power typically varies according to the ZIP
model. The same finding was confirmed by [23], where it was recommended to use the linear model for real
power but a polynomial model for reactive power.
It should be noted that in some studies [24], [41], [42], real (P0) and reactive (Q0) power at the initial (pre-
disturbance) voltage (U0) are used, instead of real (Pn) and reactive power (Qn) at rated voltage, in formulae
for exponential, polynomial and linear load models.
Page 29
Modelling and Aggregation of Loads in Flexible Power Networks
Where:
2
U U
PZIP p1 p2 p3 (3.18)
Un Un
a1
U
PEX1 p4 1 k pf1 f (3.19)
Un
a2
U
PEX 2 p5 1 k pf 2 f (3.20)
Un
Reactive power is modelled using similar expressions. To capture dynamic behaviour of loads, the exponents a1
and a2 become voltage dependent below a set threshold value of bus voltage. This is mainly intended to
capture the dynamic behaviour of the load. Typically, in most commercial software tools, the constant power
and constant current load components in PZIP are switched to an elliptical current-voltage characteristic or
constant impedance representation when the voltage drops below a certain voltage threshold (typically 0.7
p.u. or lower), that can be selected manually by the software tool user, to ensure numerical stability of the
network solution in the simulation. If the load bus voltage is very low, as in the case of a fault, the power
consumed by the load cannot remain constant, as current will dramatically increase and it will cause a failure of
calculation. Figure 3-2 illustrates how constant power load characteristic (both for active and reactive power) is
automatically switched to an elliptical characteristic when the load bus voltage goes below 0.7 p.u. after a
voltage dip is provoked by a fault.
2.0
P [MW]
1.5
1.0
0.5
0.0
Q [Mvar]
0.6
0.4
0.2
0.0 0.7 p.u. Voltage Threshold
1.6
V [p.u.]
1.2
0.8
0.4
0.0
0 2 4 6 8 10
Time [s]
Figure 3-2: Example of automatic change of load characteristic in the case of a significance voltage drop
Page 30
Modelling and Aggregation of Loads in Flexible Power Networks
Equations for the real and reactive power that an induction motor consumes can be used to form a
corresponding static load model:
R U2
P Rs r 2
s Rr (3.21)
X s X r
2
s
R
s
U2 U2
Q X s X s 2
Rr Xs (3.22)
Rs X s X r
2
s
Page 31
Modelling and Aggregation of Loads in Flexible Power Networks
power electronic devices (some examples are presented in Appendix 3-B) and are therefore complex, need
long simulation times and require specialised simulation software (e.g. PSPice, EMTDC).
In comparison, the equivalent circuit models, on the other hand, are much simpler as they generally consists of
an uncontrolled front-end diode bridge rectifier (“power electronic interface”), input impedance (R and L), dc
link capacitor (Cdc) and an equivalent resistance (req), The basic topology is shown in Figure 3-4. In the
corresponding equivalent circuit model, the input impedance and value of Cdc are based on typical (or
estimated) component values present in the circuit. The equivalent resistance is given by an analytical
relationship which describes the behaviour of all components beyond the Cdc, and is usually determined by
investigating the current, voltage or power characteristics of the modelled power electronic device at the dc
link. Further details about modelling of such loads are provided in Appendix 3-B.
D1 D3
Rsys Lsys R L
iin
AC Power Supply
System
D4 D2
Power electronic-interfaced loads can be generally represented with exponential and polynomial models as
detailed in Appendix 3-A. Besides the aforementioned power electronic load categories, there are also plug-in
chargers for electric vehicles that will need to be considered going forward. These represent an important
power electronic load category, which is expected to significantly increase in the (near) future, particularly in
the residential load sector.
Page 32
Modelling and Aggregation of Loads in Flexible Power Networks
The response of reactive power (Q) can be represented using the same form as equations (3.23) and (3.24)
with corresponding change in parameters.
The exponential dynamic load model can give inadequate results when used to reproduce the short-term
response of loads having a high percentage of IMs because it does not take into account the inherent coupling
between real and reactive power absorbed by IMs. The model however does give adequate results for long
term voltage stability studies and for situations where there are limited voltage variations at the load buses (as
typically experienced in distribution networks supplied through on load tap changing transformers [52]).
In studies where only small disturbances are considered [50], the linearised form of the exponential dynamic
load model can be used (only the real power model is shown below):
t
Tp s 1
(3.25)
Pl 0 s s U
P
U0 Tp s 1
d ds
uds Rs ids s qs (3.26)
d
d qs
uqs Rs iqs s ds (3.27)
d
d dr
udr Rr idr s qr (3.28)
d
d qr
uqr Rr iqr s dr (3.29)
d
Page 33
Modelling and Aggregation of Loads in Flexible Power Networks
d M e M
(3.30)
d bTm
In practical applications, the stator transients are neglected in the above model and the model reduces to a
third order with (3.26) and (3.27) effectively converting to algebraic equations. Fifth order IM models should
only be used for larger induction motors [54], or when the influence of IMs is very important with respect to the
other parts of the load,. Otherwise a third order dynamic model is usually sufficient.
Directly connected single-phase induction motors are widely used in low-voltage applications, in devices such as
refrigerators, freezers, fans, pumps, dishwashers and washing machines. They are often referred to as
asymmetrical two-phase induction motors, as an auxiliary winding is included and used during start-up. They
can be represented using the d-q model given in [55]. An application model for this type of IM load is
presented in Appendix 3-C.
Page 34
Modelling and Aggregation of Loads in Flexible Power Networks
k pf Tpf s k pu Tpu s
P s f s U s (3.35)
1 T1s 1 T1s
P s K pu 1 T3 p s
(3.37)
U s 1 T s 1 T s
1p 2p
Q s K qu 1 T3q s
(3.38)
U s 1 T s 1 T s
1q 2q
P s K 1 T4 p s 1 T5 p s
(3.39)
U s 1 T s 1 T s 1 T s
1p 2p 3p
Q s K 1 T4 q s 1 T5q s
(3.40)
U s 1 T s 1 T s 1 T s
1q 2q 3q
Through comparisons of responses against field measurements coming from network buses, it was found that the
second-order and third-order load models are better at capturing load behaviour during transients, compared
to the first-order load model.
The above “combined IM - static load” model is a transition from relatively simple “single facet” dynamic load
models to something more complex. The "composite load model" is probably the most widely used and most
versatile dynamic load model available to practitioners in recent years.
A similar composite model is used in [54], with static load represented with conductance (GSL) and susceptance
(BSL) in parallel. Sometimes, the composite load model may incorporate equations for static load, equations of
Page 35
Modelling and Aggregation of Loads in Flexible Power Networks
an equivalent IM model (typically third order model) and equations for an equivalent synchronous motor [11].
The required complexity of the model depends on load composition, type of phenomenon which should be
analysed, required accuracy of results, etc.
There are several variants of composite load model. In [59], and [60] the static load component is represented
by a ZIP model. This model is depicted in Figure 3-6 with separated constant power, constant current and
constant impedance load components. The static part of Figure 3-6 can be described as [60].
2
U * U
P P
s
*
PI
Z
*
PP
*
(3.43)
0
U 0
U
2
U * U
Q Q
*
s QI
*
Z QQ
*
(3.44)
U0 U0
with
Qmotor
QZ* QI* QP* 1 (3.46)
Q0
where Kpm is the ratio of initial motor load and initial real load of the bus, Qmot is the initial reactive power
consumed by the motor and Q0 is the initial reactive load of the bus.
The parameter M lf that reflects the effect of the dynamic components on the total load can also be defined as
Pmotor U0
M lf , where S motorBase is the equivalent motor nominal apparent power, U 0 and U Base are
S motorBase U Base
initial bus voltage and base voltage, respectively. Thus, the dynamic part of the load is described as:
Page 36
Modelling and Aggregation of Loads in Flexible Power Networks
dEd' 1
' Ed' ( X X ' ) I q ( 1) Eq' (3.47)
dt T
dEq' 1
Eq' ( X X ' ) I d ( 1) Ed' (3.48)
dt T'
d 1
( A 2 B C )T0 ( Ed' I d Eq' I q ) (3.49
dt 2H
1
Id Rs (U d Ed' ) X ' (U q Eq' ) (3.50)
R X '2
2
s
1
Iq Rs (U q Eq' ) X ' (U d Ed' ) (3.51)
R X '2
2
s
where:
T ' ( X r X m ) Rr
Xr = Xm + Xr
X Xs Xm
X ' X s X m X r ( X m X r )
A B C 1
Additionally, composite load models can include saturation characteristics of transformers and motors [61] to
account for reactive power losses due to losses in the iron core of transformers and motors.
Page 37
Modelling and Aggregation of Loads in Flexible Power Networks
device connected to transformer secondary with susceptance BC, feeder with resistance Rfed and reactance Xfed,
conductance GL, corresponding to the resistive part of the load, and induction motor whose stator resistance
and reactance are R1 and X1, magnetizing reactance is Xm, first rotor winding resistance and reactance are R2
and X2, and assuming a double cage rotor, second rotor winding resistance and reactance are R3 and X3.
The tap changer is modelled as a controller that adapts the transformer ratio (modelled as a
continuous variable) between user-defined limits to control the low voltage level to a reference value specified
as a parameter. The controller acts following a time constant that can be specified by the user.
It should be noted that there are variants in modelling of equivalent distribution loads. One of these variants is
the WECC composite load model described in [1] and [10]. It has four motors (A, B, C and D), static and
electronic load, and shunt capacitance at the secondary of the transformer, as shown in Figure 3-8. Motor A
represents three-phase motors driving constant torque loads, such as commercial and industrial air-conditioning
and refrigeration compressors. Motor B represents so-called “speed squared” motors with large inertia such as
motors driving fans. Motor C represents so-called “speed squared” motors with low inertia such as water
pumps. Motor D represents single-phase motors driving constant torque loads such as residential air-
conditioners, refrigerators and heat pump compressors. Appendix 3-D provides details on modelling of such
small motors, which are prone to stall.
Variable frequency drive controlled motors are classified as electronic loads. Static and electronic loads are
represented by polynomial models, but some three-phase electronic loads may self-disconnect (turn-off) during
voltage sags depending on the design of the inverter circuit (see Chapter 4 for more details).
The use of Variable Speed Drives (VSD) to connect large synchronous and asynchronous motors (in the range of
hundreds of kW to tens of MW) in the oil and gas industry has considerably grown in the last few years. In
parallel, an increasing number of smaller motors (e.g. refrigerators and air conditioners) that utilise inverter
technologies are being installed in the networks.
The correct representation of such industrial/residential loads is of fundamental importance when studying the
effect of network disturbances on the dynamic behaviour of not only the loads themselves, but also the
Page 38
Modelling and Aggregation of Loads in Flexible Power Networks
connecting network. This is particularly true for isolated industrial systems or in regions of the network where
large industrial plants are installed. The typical application of such models would be for studies related to
industrial plant performance during system disturbances (from the connection point looking inward), and studies
which aim to investigate the impact of such devices/plant on overall system stability and performance.
where kp and kq are characteristic constants, Pdrop and Qdrop are the amounts of load drop related to minimum
bus voltage, and Pdyn and Qdyn are the magnitudes of the dynamic load components. The conductance of the
motor, G(t), follows the equation with the initial condition G(t=0)=1.
dG / dt 1/ T GV 2 1 (3.54)
In this equation, T is the time constant of the dynamic load component. It is assumed that susceptance (B) of the
motor also satisfies equation (3.54).
In [62] one variant of the model (3.52)–(3.54) is presented. It includes one more equation for the susceptance B
dynamics, and two damping time constants (Tp and Tq) for motor conductance and susceptance following a
fault.
Figure 3-9: Load Model Structure of Bulk Power Bus (adopted from [39]).
Page 39
Modelling and Aggregation of Loads in Flexible Power Networks
Page 40
Modelling and Aggregation of Loads in Flexible Power Networks
There is a relatively even distribution of different load models used in dynamic system studies for modelling
real power demand, but static load models are again dominant. Constant power and constant current load
models account for about 42% of all used models. A similar dominance of static load models is observed in the
modelling of reactive power demand. In this case, constant power and constant impedance load models account
for 45% of all used load models used. For modelling both real and reactive power demand, about 30% of the
reported models represent dynamic load by some form of induction motor model. Appendix 2-F presents the
types of load models used in static and dynamic studies as reported across in different continents.
Table 3-1: Types of load models used in static and dynamic power system studies
Constant Constant Constant ZIP Exponential ZIP model with Exponential model with Detailed composite
Load model
P(Q) I Z model model IM IM model
Although the same type of load model can be used for representing loads at different buses, its parameters
could vary depending on the modelled load class (e.g. residential, industrial, commercial, etc.). However, the
current practice in 75% of utilities and system operators is that they do not discriminate between different load
classes. This could be explained by the difficulties that transmission system operators (in particular) encounter in
obtaining accurate information on load classes at different network buses. A possible explanation of this could
be because the data are typically owned by distribution system companies.
Another interesting observation from the survey is how frequently load model parameters are updated.
According to the responses to question three (Q3), utilities and system operators are very well aware of the
significance of load modelling issue as they update load model parameters relatively frequently. In 41% of
cases, load model parameters were updated within the last five years. Further information from the
questionnaire regarding the use of different load models for different load classes, and the updating of load
model parameters, is given in Appendix 2-F.
3.6 Summary
The chapter presented the typical existing load models that describe static or dynamic behaviour of different
load devices and load classes. Provided as appendix materials are the parameters of the different load
models found in publicly available literature. Based on the completed review of existing load models and the
results of the international survey, it was found that the most frequently used static load models are
exponential, second order polynomial, and linear model with frequency dependence neglected. The constant
PQ load model is the most widely used for steady state analysis of power system. A comprehensive static load
model is typically used for large voltage variations, while a static load model of induction motors (and loads
dominated by induction motors) is valid depending on the type of analysis being undertaken. One of the most
frequently used dynamic load models is exponential dynamic load model. It is predominantly used to model
residential load. Induction motor dynamic load model is used to model load having a large participation of
induction motors. The composite load model is adequate for representing loads consisting of both static load
components as well as induction motors. There are different variants of composite dynamic models but ZIP-
induction motor model is the most commonly used. Parameters of selected load model can be different even for
the same load class and the same season, since load structure depends on many factors such as economic,
social, climate etc. It is worth noting that the international survey on industry practices revealed that static load
models (typically of constant PQ or ZIP) are the most commonly used for dynamic studies with only about 30%
of utilities and transmission system operators using some form of IM model to represent dynamic loads.
Page 41
Modelling and Aggregation of Loads in Flexible Power Networks
4.1 Introduction
Following the review of existing methodologies for load model development provided in Chapter 2, and the
overview of load models used by industry as given in Chapter 3, this chapter provides recommendations for the
development of aggregate load models at bulk power supply buses for power system steady state and
dynamic studies. The first part of the chapter focuses on methodologies for applying the measurement based
load modelling approach, while the second half discusses methodologies for component based load modelling.
Page 42
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 4-1 and Figure 4-2 summarise the general approach to measurement based load modelling developed
in [12], [15], [16].
Accepted Load
Model
Figure 4-1: Simplified Schematic for Developing Load Models Based on Identification Aggregation (adopted from
[64], [65]).
Data Collection
Event
Selection
Data
Processing
Model
Validation
Figure 4-2: Measurement Based Load Modelling Approach, (adopted from [12]).
1. Data Collection
Ideally, time stamped, time domain voltages and currents for each phase following system disturbance
should be recorded. Different commercially available data acquisition devices, such as Power-Quality (PQ)
monitors, digital fault recorders (DFR) and digital relays, can be used for collecting system disturbance
data. The ideal location for collecting data is at the LV side of distribution substation transformers. Either a
single feeder or multiple feeders can be monitored depending on the load model development process
being considered.
Page 43
Modelling and Aggregation of Loads in Flexible Power Networks
3. Data Processing
A Discrete Fourier Transform (DFT) based sliding window algorithm is used to convert three-phase voltages
and currents into positive sequence, per unit voltage, current, real power and reactive power. Sampling
rate of 1 kHz, i.e., every 1ms, or 2 kHz, i.e., every 0.5ms, are typically sufficient for the optimization
algorithm. Note that the sliding window algorithm inherently results in filtering of the input data.
Page 44
Modelling and Aggregation of Loads in Flexible Power Networks
5. Model Validation
Once a reasonable set of model parameters are obtained, they can be benchmarked using commercial
power system analysis software. A validation process should be implemented to assess the level of
agreement between simulated and measured load responses, ideally using data that has been guaranteed
from the load model development process. If the performance of the model is determined to be adequate,
it can be accepted for further general use in system studies. If not, further fine tuning of the model should
be attempted via Step 4 of the process, whereby alternate model parameters are estimated, or if needed,
a different model topology is selected.
An example of a top-down approach to measurement based load modelling is provided in Appendix 4-A.
It should be noted that these identification methodologies are common for various types of stability studies:
transient stability, small disturbance stability and frequency stability, if a suitable load model is selected.
Page 45
Modelling and Aggregation of Loads in Flexible Power Networks
the response of the aggregate load, including variations coming from the operation of downstream tap
changers and voltage regulation devices (which may or may not be directly controllable during the staged
filed tests.) To mitigate such issues, it is recommended that several consecutive step change in voltage be
carried out during each testing period in order to obtain a reliable data set.
T 2´ 1000/5/5/5A F1
110kV F2
network CT F12
110 10´ 1.5%/10.5kV
31.5 MVA, Ynd5
VT 10 0.1 0.1
uk=14.29% / / kV
3 3 3
Recording
equipment
Figure 4-3: Simplified Diagram of Equipment Connections for Field Tests (adopted from [69]).
For staged tests performed in a laboratory environment, the required step change in voltage can be produced
in several different ways, including the switching of parallel transformers as already described above. In a
laboratory, it may also be possible to directly after the turns ratio of the supply transformer. An advantage of
this method is that a step change in voltage in the order of 30% can be produced. Other methods include
inserting resistors between the source and the supply transformer, and connecting resistor in parallel at the
secondary side of transformer. The conservative aspect of laboratory testing is that there is no impact on real
time network operators or end-use customers which need to be managed.
Examples of measurement based load modelling are given in Appendix 4-A and Appendix 4-B.
In most cases, the average of the three phase voltages can be used as the terminal voltage at the load for
load model identification purposes (noting that some voltage imbalance may exist in practice).
EPRI’s measurement based approach, discussed in the introductory part of this chapter, requires disturbance
(event) data from data acquisition devices installed at a selection of representative distribution feeders or
Page 46
Modelling and Aggregation of Loads in Flexible Power Networks
substations. The events include the aggregate response of loads connected to a distribution feeder due to
either an upstream transmission fault or a fault on an adjacent feeder.
The ideal location for collecting events for the measurement based approach is at the LV side of a distribution
substation transformer. For each feeder for which a load model is to be identified, the following quantities
should ideally be measured:
Waveform (time domain) bus voltages for all the three phases
Waveform (time domain) current data for all the three phases
Many of the commercially available monitoring devices can be used for data collection, in particular power
quality monitors (PQMs) and digital fault recorders (DFRs). A minimum set of requirements that monitoring
device used for data collection for the measurement based approach should fulfil is given in Table 4-1.
A number of software tools can be used for load modelling purposes. Some of them are listed in Table 4-12,
however, this is not an exhaustive list of software tools.
Ideally, it would be good to collect many cycles or even a few seconds worth of post-fault data for the EPRI’s
LMPD algorithm. PQMs typically do not store more than 16 cycles of post-fault data, which is a notable
limitation. Digital fault recorders (DFR) can store seconds worth of post-fault data and are more suitable;
however, they are also typically more expensive than PQMs and not as widely installed as PQMs on
distribution feeders.
The sampling rate required for load modelling depends on the actual load models to be developed. It could
range from 1 ms (1 kHz) to 1 s (1 Hz). Depending on the load model, the dynamics to be being represented by
the model, the availability of data, and the capability of the measurement equipment, the sampling rate can
be determined accordingly. The accuracy of the load model developed ultimately depends on the sampling
rate used during data capture. With lower sampling rates, such as 1 Hz, static load models or exponential (1st
order) dynamic load models (typically for voltage stability studies) can be developed. The quality of dynamic
load models (as may be used for transient stability studies) developed with this sampling rate may not be
appropriate. Load models used for transient stability studies should be developed using data having a sample
rate of between 50 Hz and 1kHz. When high sampling rates are to be used, care should be taken when
considering the length of the recording window as data storage limitations may be an issue. A possible solution
to this is to use monitors with variable sampling rates, where slower sampling is used to record events over
longer time period, and faster sampling is applied for short time periods immediately after triggered events.
Tables 4-3 and 4-4 summarise different measurement devices, their associated sampling rates, and their
implications on load modelling. Recommendations are also given in view of different power system stability
issues.
Page 47
Modelling and Aggregation of Loads in Flexible Power Networks
be used for these purposes. As discussed in the previous section, depending on the type of load models
involved, the sampling frequency should be at least 50 Hz.
Table 4-1: Minimum Requirement of Hardware Configuration for Data Recording Devices (naturally occurring large system disturbances)
INPUT SPECIFICATIONS
Number of digital channels Not necessary for load modelling however the position of certain circuit breakers can be useful
additional information depending on substation topology and the types of field tests being
undertaken (if any), e.g. transformer switching.
Sampling rate At least 960 samples per second per channel (sampling rate of 1kHz or higher). Many specific
data acquisition systems have adjustable sampling rates
Pre-fault Recording Time (cycles) At least 2 to 5 cycles of pre-fault data. Several seconds of data is preferred.
Post-fault Recording time (cycles) Post-fault is a function of maximum storage capability and reset threshold. In the case of large
disturbances (e.g., faults), several seconds of data may be sufficient for the development of
dynamic models. Although , a several minutes data set is needed in the case where load models
are to be used for long term voltage stability studies. .
Trigger Threshold The user should be able to set the voltage or frequency level that initiates an event record.
HARDWARE SPECIFICATIONS
Hard drive storage This is largely dependent on whether the device can be downloaded remotely to recover
stored data.
Ethernet connection, network Chosen in order to facilitate the configuration updating of the unit and the download of data.
protocols
FILE FORMATS
The data recorded in system disturbances need to be processed and filtered before load model identification
analysis is performed. For load model development based on fault data, the key steps involved include: i)
Identification of data recorded at LV or HV side. ii) Data pre-processing to select data which contain sufficient
voltage and power variations during recorded disturbances. iii) Calculation of real and reactive power values
using recorded data (which may be three phase voltages and currents). iv) Load model identification based on
the processed measurement data.
Filtering may be needed if the original measurement data contain significant noise. However, filtering has to be
carried out with care to ensure that the useful transients contained in the data series are adequately preserved.
In most cases, the measurement noise is Gaussian noise, so simple averaging filters can be used to make the
signals useful. For example, in a data series measured at 100 Hz, values of 5 to10 adjacent data points can
be averaged. The average value will replace the original measurement value and used for load model
identification purposes. An averaging filter is very effective even for higher sampling rates since the random
Page 48
Modelling and Aggregation of Loads in Flexible Power Networks
noise is the major target of filtering process. The application of Fast Fourier Transform (FFT) can then be used to
extract the positive sequence component of recorded currents and voltages for further processing.
Table 4-2: Example software tools that can be used for load modelling (non exhaustive list)
SOFTWARE DESCRIPTION
DiGSILENT PowerFactory Transmission, distribution modelling and analysis, optimisation toolkit can
be used for identification of measurement based load modelling
Eurostag Comprehensive power system analysis tool
CPAT (CRIEPI's Power Analysis Tools) Transmission network modelling and analysis
Phasor estimation is one of the most popular topics in the field of power system estimation. There are many
algorithms that have been developed to estimate the frequency and fundamental phasors of voltage and
current signals. These include level-crossing technique [75], Kalman filtering [76], adaptive notch filtering [77],
Shank’s method [78], Short-Time Fourier Transform (STFT) based modified algorithms, etc. Among those
techniques, level-crossing technique, infinite impulse response (IIR), short time Fourier transform, and Finite
Impulse Response (FIR) based dynamic phasor estimator are commonly used. Detailed descriptions of these
techniques are given in Appendix 4-A.
For the measurement based load modelling approach, fundamental frequency (50Hz or 60 Hz) data are
needed. The focus of load modelling described in this report is on positive sequence load models that can be
used in performing steady state and stability studies using commercial power system analysis software (see
Table 4-2). Therefore, the three phase voltage and current data recorded as field measurement need to be
Page 49
Modelling and Aggregation of Loads in Flexible Power Networks
processed to extract the fundamental frequency quantities. A Discrete Fourier Transform (DFT) based sliding
window algorithm is a useful tool for such purposes.
It should be noted, however that there are situations when it becomes necessary to consider harmonics and the
modelling of non-linear load behaviours. These may include studies related to quantifying harmonic levels,
capacitor placement investigations and filter design. Such studies are performed at both transmission and
distribution levels and is important from a system compatibility perspective. These types of studies are outside
of the scope of this report.
Page 50
Modelling and Aggregation of Loads in Flexible Power Networks
implementation of fast capacitive MVAr reserves such as Static VAr Compensator (SVC) or special protection
systems based on undervoltage activated load shedding. Such countermeasures may constitute a large capital
investment. It should be noted that residential air-conditioner compressor motors tend to stall very rapidly (in a
Table 4-4: Measurement Options and Their Implications on Load Modelling
OK
Short term
voltage Insufficient data resolution
(Insufficient data resolution
stability
if sampling rate is 5-6Hz),
Long term
May be OK depending May be OK depending on
voltage OK
on record length record length
stability
(including pre-event
buffer) and LSU
Good for all types of load Good for all types of load May be OK
Transient capacity
Insufficient data resolution modelling activities, modelling activities
stability (Insufficient data resolution
if sampling rate is 5-6Hz),
Recommended application
OK if record length
Frequency
Insufficient data resolution (including pre-event OK
stability
buffer is ≥ 10 sec)
Page 51
Modelling and Aggregation of Loads in Flexible Power Networks
100
90
Non-inverter-based
80 high-pressure Air-conditioner Electric hot water
discharge lamps for industrial use supply system
70 Control sequencer Inverter-based
Voltage [%]
for industrial use air-conditioner
60 Inverter-based
Electric hot water fluorescent lamp
Miniature relay Timer
50 supply system Digital
Electromagnetic television
40 switch
Induction
DVD recorder heating cooker
Control sequencer
30 Induction
for industrial use
heating cooker
Personal Inverter-based
20 Computer high-pressure
discharge lamps
Inverter-based
10 fluorescent lamp
Digital television
0
0.001 0.01 0.1 1
Voltage Sag Duration [s]
Figure 4-4: Individual Load Self-disconnection Characteristics Following a Voltage Sag [79].
200
VU-N[V]
100
0
-100
-200
20
10
I [A]
0
-10
-20 Restart 800 Compressor During Restart
Stoppage
P [W]
200
400
P [W]
100
0
0
-100 200
Q [var]
100
Preparation of Restart 100
Q [var]
50
0 0
-50 -100
0 5 10 15 20 25 -200 0 200 400 600 800
Time [s] Time [s]
Figure 4-5: Example of Load Self-disconnection Characteristics Following a Voltage Sag [79] (Left: Digital Television Under 80% Voltage Sag with 500
ms duration, Right: Inverter-based Air-conditioner Under 40% Voltage Sag with 500 ms Duration).
few cycles of disturbance initiation[9]) and thus the solution to the problem may be quite difficult. Where a
large portion of the load are residential air-conditioners, the only way to prevent stalling is to have enough
short-circuit capability at or near the load centre to maintain an adequate voltage profile during and following
network events. The investigation of this problem is quite complex as some air-conditioner system may trip due
to their own overcurrent protection, leading to a non-uniform response from load devices which are essentially
equivalent until exposed to such extreme operating conditions.
The advent of inverter-based appliances and their fast disconnection, or idling for several minutes, following
system disturbances, might influence power system behaviour. For example, according to the test results shown
in Figure 4-5, as the conventional induction motors are replaced with the inverter driven motors, many load
types became more easily disconnected from the power system [79]. Their impact on the dynamic security
assessment must be carefully considered by the grid operator.
In low load density areas, or when load centres are remotely located from each other, the fault-induced
voltage sags (dips) have a stronger influence around the location of the disturbance. Generally, load self-
Page 52
Modelling and Aggregation of Loads in Flexible Power Networks
disconnection facilitates the voltage recovery in the network and offers more flexibility to grid operators to
secure the voltage stability of the transmission system.
Net load self-disconnection amount: P+ P’
The effects of large portions of load being subject to disconnection need to be carefully considered. Two
possible consequences are: (i) It may be beneficial in that the unintentional loss of load may partially relieve
the system and arrest the voltage reduction, thus limiting the subsequent stalling of other motors, and (ii) The
unintentional reduction in load may excessively relieve the system following fault clearance, resulting in surplus
reactive power that would need to be absorbed by local dynamic reactive control devices (SVCs etc.).
Depending on the amount and location of reserve capability, overvoltages may occur in parts of the system
that have low short circuit levels or are electrically remote from reactive control devices. Activation of
synchronous generator underexcitation limiters may also warrant consideration depending on local stability
issues. While the determination of such load behaviours is not an easy task, it is considered important especially
when large portion of load are at risk and/or the load is highly concentrated in a particular region of the
power system.
Page 53
Modelling and Aggregation of Loads in Flexible Power Networks
data is normally collected at a high (or extra high) voltage bus in the system. For developing load
models for use in both voltage and angular stability studies, the recording devices should be set to
collect data for tens of seconds, so that slow voltage recovery events can also be captured.
Events data coming from acquisition devices installed at a few representative distribution feeders or
substations. These events include an aggregate response of loads connected to a distribution feeder
due to either an upstream transmission fault or a fault on an adjacent feeder. This data can be
collected using a PQM or a portable DFR.
Load composition and load component information is essential for building a system-wide load model. System
event data is not as critical, although it is eventually needed to test the performance of the load model being
developed. Note that the collection of system events is a time-consuming process due to the random nature of
network faults. Even if event recordings are available, they may not be suitable for model validation or
deriving load model parameters. Therefore, a long term commitment is essential for the collection of system
event data that are to be used for load modelling purposes.
The aggregate response of loads connected to a distribution feeder that are induced due to either an
upstream transmission fault or a fault on an adjacent feeder can be used for load modelling activities. Such
data would include motor dynamics and the response of static loads and would allow the portion of each type
to be estimated.
It is important to distinguish between load model validation carried out at an individual feeder or a specific
distribution voltage bus and validation that may be undertaken at a broader system level, i.e. at one or more
Extra High Voltage (EHV) buses in the system.
Page 54
Modelling and Aggregation of Loads in Flexible Power Networks
The load composition at the bulk supply bus reflects the percentage of different load types/categories
participating in the overall demand at a given point in time. The participation of different customers in the
overall demand at bulk supply buses varies with time and it depends on the type of the load class involved,
devices constituting demand and the nature of processes in end user facilities. While global participation of
different classes of demand can be established to a certain extent based on metering data, it is very difficult
Page 55
Modelling and Aggregation of Loads in Flexible Power Networks
to establish participation of different load types within each class at any given time. The essential requirement
for developing appropriate load models at bulk supply points, and consequently the load response to system
disturbances at given point in time (in addition to individual load component models) is information about
demand composition at that time.
Figure 4-7 shows a decomposed daily loading curve (DDLC) in terms of different load types for commercial
load for a working day. It is derived from daily demand data for different types of loads [86]. The peak
demand is normalized to 100 for ease of comparison.
The key information needed for estimating load model at the bulk supply bus is the type and percentage of
different load categories rather than appliance or end-users involved, i.e.: devices for which individual load
models are available. Therefore, a daily load-by-category loading curve is needed instead of that shown in
Figure 4-7. Considering demand decomposition shown in Figure 4-7, it is obvious that the same load category
(e.g., IM) may be included in different load types as identified in the figure.
The load types listed in Figure 4-7 therefore can therefore be categorized as follows:
Air conditioning load, motor load, process load, refrigerator load and ventilation load are grouped
into the category of motor load.
Figure 4-7: Decomposed Daily Loading Curve for Commercial Load Sector.
Cooking, water and space heating load can be categorized as resistive load.
Due to different definitions of miscellaneous load, miscellaneous can be classified as a single category in its
own right. Following this categorisation, new DDLC can be produced as shown in Figure 4-8. From this figure, it
can be easily determined which load category is participating in the total demand at a given bus and what
share of the total demand it constitutes. Since individual load models of each load category are known, and
there are fewer of them, it is much easier to estimate aggregate load model at any chosen time by combining
participating load categories after applying appropriate weighting factors (based on participation in total
demand).
A similar approach can be used to convert conventional (based on load classes) DDLC (as shown in Figure 4-9
a)) into DDLC based on different load categories for a general network bus (as shown in Figure 4-9 b)), or to
develop DDLC for any season of the year. Further detail about decomposition of daily load curves and
development of aggregate load models for different load classes is provided in Chapter 5 and 0.
Page 56
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 4-8: DDLC Based on Load Categories for Commercial Load Class.
Figure 4-9: DDLC at Bulk Supply Bus Based on a) Load Class Mix, b) Load Category Mix.
Table 4-6 presents the percentage of basic load models adopted in the Western regions of the United States
to represent a residential load.
Two specific issues that one should consider are
The aggregation of individual loads in a single model of corresponding type.
The inclusion into the aggregate model of other network components that are present in distribution and
sub-transmission networks, e.g., feeders, transformers equipped with on load tap changers, capacitors,
etc.
Several methods to group the same type of load into a single aggregate model are presented in literature, in
particular those related to aggregation of induction motors. The classical reference for aggregation of loads
represented by exponential models is [87], while references [84], [88–92] deal with aggregation of induction
motors.
An important part of developing aggregate load models at bulk supply buses is the representation of
distribution and subtransmission network elements. The main aspects that should be considered are:
The feeder representation to capture voltage drop and real and reactive power losses in primary
distribution circuits, distribution transformers, and secondary distribution circuits;
The representation of shunt reactive compensation both at the substation and feeder ends with the
relevant regulation controls (where these exist).
Page 57
Modelling and Aggregation of Loads in Flexible Power Networks
100%
Constant I 0 0 0 0 0 0
(fluorescent)
Constant P 0 0 0 100% 0 0 0
Page 58
Modelling and Aggregation of Loads in Flexible Power Networks
RACs are in general equipped with magnetic contactors that trip the unit when the supply voltage drops
below a certain threshold, typically between 40% and 52% of nominal voltage. This voltage is
independent of outdoor temperature.
The stalling threshold voltage of RAC motors is between 50% and 73% of nominal voltage, depending
on outdoor temperature.
RAC compressors can stall for voltage sag as short as 3 cycles if the voltage reduces to below the
stalling threshold voltage for this duration. This means they will almost certainly stall for any
transmission fault causing a large voltage dip in local supply voltage.
Thermal Overload Protection (TOL) disconnects the motor if it remains stalled for a considerable amount
of time (seconds to tens of seconds). TOL operation time is inversely proportional to the applied
voltage at the RAC terminals.
Once a unit stalls, TOL invariably operates for units with reciprocating compressors. In the case of scroll
units, the unit may recover from the stalled mode without TOL operation if the voltage recovers quickly
enough (roughly above 70%).
RAC motors have a very low moment of inertia, which makes them quite vulnerable to stalling.
Voltage sags may affect the efficiency of the RAC if it goes into “no load” mode.
A model for Residential Air Conditioners that incorporates the above characteristic is given in [10].
When the model of the individual load components is chosen, a static model consistent with the developed
dynamic model should be defined to calculate the initial steady state condition. This issue is particularly
important for induction motor models. In [83], the following procedure is proposed. Time-domain simulations are
carried out in which the motor models are subjected to step-wise voltage variations, making their terminal
voltages vary in the range of 0.8 to 1.0 p.u., and the quasi steady-state responses are recorded. The obtained
P-V and Q-V characteristics are then fitted with an exponential model of the following form:
P Pn aU bU aU
0 bU 0 (4.3)
Q rU0 Pn cU dU cU
0 dU 0 (4.4)
Table 4-7 shows the results presented in [83] for the case of the data set proposed in [37] for small motor and
large motor dynamic models.
Table 4-7: Parameters of static models of induction motor models (adopted from [83])
a b â c ã d ä r
0.6
Small motor 0.86 0 0.14 0.09 0.8 1.6 0.2 -3.3 0.627 U n
Large motors 0.86 0 0.14 0.09 0.747 1.54 0.253 -3.07 0.519 U n0.35
Once the reactive power absorbed in steady state by all the load components at the bus of interest is
determined, the reactive power output of fixed capacitors should be calculated such that the net reactive load
at the bus matches the power flow data. This last step is typically done using commercially available power
system analysis software.
Page 59
Modelling and Aggregation of Loads in Flexible Power Networks
Collect system event data at one or more EHV buses in the system using either DFRs or PMUs. These
devices should be set to collect continuous data for tens of seconds post-disturbance with adequate
pre-disturbance record length to establish the initial steady state conditions.
If an event recording for a major fault is available, at one or more EHV buses in the system, then by
comparing the actual recording to the simulated response one can obtain a better understanding of
how good the load model is. The model can be refined if the measured and simulated responses
don’t match with sufficient accuracy. This is an iterative process and engineering judgment needs to
be carefully applied. This would require the additional activities of obtaining system information,
such as status of generators and other equipment in the surrounding region at the time of the
disturbance, to be able to faithfully simulate the event for comparison with the measured system
response.
It should be noted that in the real power system, there will always be variability (i.e. temporal and spatial
variations in load due to climate and human behaviour) and uncertainty in the load model data. One can never
have exact information about all the loads that are connected to the system at any given time. It is not possible
to achieve an exact match between simulated and measured data at each load bus in the system and some
discrepancy is to be expected. What is more important, however, is to compare the measured and simulated
responses at a few EHV buses in the system where the measurements were taken. If there is a reasonable match
between two sets of data/responses, then it indicates that the developed load models are appropriate for the
system. If the match is not satisfactory, then the load model parameters (most likely percentages of static and
dynamic load components) should be adjusted as necessary.
During the model validation process, it is particularly important to assess sensitivity to variations in different
parameters. Despite the best modelling efforts, the characteristics of actual loads make it difficult to remove all
the uncertainties involved. To deal with these uncertainties, sensitivity analysis is always recommended to assess
the effect of varying model parameters and the ability to reproduce actual measured responses.
Page 60
Modelling and Aggregation of Loads in Flexible Power Networks
The sensitivity analysis should identify the set of key parameters for the study of interest so that the load model
tuning process can specifically focus on refining those parameters. In general, the sensitivity analysis should
include an assessment of the effects of varying the amount of dynamic load component, induction motor load in
particular, in the aggregate load model and the assessment of the effect of changes in motor parameters, e.g.,
stalling voltage as a function of temperature, or motor inertia. Additional analysis could include the sensitivity
of the load model to self-disconnection and reconnection of part of the system load, as this could have
significant effect on longer term dynamic studies, in particular, voltage and frequency stability studies.
4.5 Summary
In this chapter, the measurement based and component based load modelling approaches have been
presented. Recommendations are made with respect to all stages of the load model development process for
both approaches. The issues of selection of monitoring location for both, continuous monitoring and staged field
tests, selection and processing of monitoring data, sampling rates, signal filtering, selection of model structure
and self-disconnection of part of the load are discussed in detail. The step-by-step recommendations for system
load model development given in this chapter could be used by engineers in distribution and transmission
companies for development of appropriate load models.
Page 61
Modelling and Aggregation of Loads in Flexible Power Networks
5.1 Introduction
A “Load Sector”, also known as a “Class of Customer” or a Load Class 1 , is generally defined as an
aggregation or collection of different types of loads (or different load categories, see Chapter 3 and
Appendix 5-A for a general overview), representing a typical structure and composition of electrical devices
and equipment found in a specific end-use application, where similar activities and tasks are performed. This
usually results in inherent similarities in characteristics and patterns of active and reactive power demands of
end-users, allowing the use of the same or similar aggregate load models for the representation of their
aggregate demands. In order to provide a more accurate load model, one load sector may be further divided
into several sub-sectors.
In power system studies, load models representing a group of loads from the same or different load
categories/sectors connected to a single network bus (being the so called “bulk load supply point, BLSP) are
usually termed as “Aggregate Load Models” or “Models of Aggregate Load”. They are often in a simplified
form, e.g. use exponential or polynomial analytical expressions, in order to reduce the required computation
times and overall complexity of the analysis of transmission and distribution networks. Accordingly, a model of
a load sector is always an aggregate load model. This chapter mainly considers one special category, being
load models of typical load sectors.
Both in related publications (e.g. [3], [14], [30]), and in available statistical data (e.g. [96] and [97]), the
corresponding representation of aggregate system loads are generally divided into three load sectors:
residential (or domestic), commercial (or general service) and industrial. The residential load sector is generally
defined as houses or buildings for dwellings, whose sole purpose is to provide residency to the occupants. The
commercial load sector consists of public, private and voluntary establishments and businesses, which are
generally aimed at providing a specific service to the public. The commercial load sector does not involve
general product manufacturing and material processing activities, which are part of the industrial sector.
Accordingly, industrial sector loads are defined as equipment and devices employed in any business that
performs raw or any other level of material processing, fabrication or manufacturing, or similar activities.
It is generally not possible to define general (i.e. generic) aggregate load models for the industrial sector, as
the specific industrial processes and activities performed at different sites are usually not comparable across
the whole industrial load sector. On the other hand, generic aggregate load models may be defined for
residential and commercial load sectors, as there is less diversity within the loads and the general load structure
is similar. However, this assumption is only applicable if load sub-sectors are introduced and defined, as the
electrical characteristics of end-use equipment and their actual active and reactive power demands will vary
between different users within the same load sector. Furthermore, the hourly and daily (24-hour load profile),
weekly and monthly (difference between weekdays, weekends and holidays) and seasonal (summer-winter)
variations will also differ within a given load sector. This suggests that it is often necessary to subdivide one
specific load sector into corresponding “load sub-sectors” in order to correctly represent involved temporal,
spatial and functional variations in their active and reactive power demands and other relevant characteristics.
Variations in general electrical characteristics and active/reactive power demands within each load sector
have been reported in a number of previous studies (e.g. [3], [98–104]). However, the production of detailed,
time-varying load models for use in power system studies from this data is not readily available in existing
literature. Therefore, 0 describes a general methodology for producing aggregate component based load
models for different load sectors and sub-sectors, where load structure and composition are determined from
available statistics and/or measured data (e.g. data collected by “smart meters”). Particular attention is given
1 In this report, terms “Load Sector”, “Class of Customer” and “Load Class” are interchangeable.
Page 62
Modelling and Aggregation of Loads in Flexible Power Networks
to the residential and commercial load sectors, as the industrial load sector will require specific load models
which are generally determined by the actual manufacturing processes and involved end-use devices which
form the loads.
Page 63
Modelling and Aggregation of Loads in Flexible Power Networks
medium strength (although not significantly reduced compared to urban areas) and often radial in nature. It is
further characterised by a medium concentration/density of powers per m2/km2 and a medium to high
penetration of modern loads. The transformer ratings will be lower than in the urban sub-sector and the length
of cable or overhead lines will be further increased due to the lower density of housing. Small/micro solar/PV,
wind and CHP generation technologies can be expected to exist as space restrictions become less onerous.
Cooking Resistive
Heating Resistive
Directly connected three-phase motors
HVAC
Drive controlled three-phase motors
ICT equipment SMPS
GIL
Interior lighting
Energy efficient lighting – CFL and LFL
Lifts/elevators Three phase motors
Power electronics SMPS
Refrigeration Single-phase motors
Abbreviations: HVAC – heating, ventilation and air-conditioning, ICT – information and communication technologies, GIL – general
incandescent lamps, SMPS – switch-mode power supplies, CFL – compact fluorescent lamps, LFL – linear fluorescent lamps.
Load trends
The residential load profile has two distinctive peaks – one in the morning, when the majority of the population
are getting up and commencing their day, and one in the late afternoon and early evening hours, when
dwellings are most heavily occupied. It is also possible for peaks to occur during the day, for example at lunch
time or during the night-time, due to electric heating that is remotely/locally controlled by off-peak tariff
meters. The morning and evening peaks are usually dominant as shown in Table 5-1.
2 It is often hard to make a clear distinction between rural residential and small agricultural load sectors.
Page 64
Modelling and Aggregation of Loads in Flexible Power Networks
Daily (i.e. weekday/weekend) and seasonal variations will exist within each sub-sector. However, as occupant’s
habits, as well as the actual load devices present, are likely to be similar, these variations are expected to be
consistent and similar between all sub-sectors. In the UK, seasonal variations are caused by longer times for use
of heating, lighting and cooking equipment during the winter seasons and air-conditioning and
refrigerator/cooling loads during summer. This is a direct result of the temperature variations and of the extent
of available daylight. Generally, the seasonal variations will be significantly influenced by the climate zone in
which modelled aggregate load is located.
Variations between individual weekdays, although likely to happen (e.g. due to a public holiday, or some
important social event influencing work-hours schedule) are in this report considered negligible. For residential
load profiles, however, significant variations between the weekday and weekends can be expected. The main
difference is that dwelling occupancy will increase during daytime hours at the weekend, ultimately resulting in
increased demand. Demand will also be influenced by prevailing weather conditions and "cultural norms". For
example in some countries, Sunday daytime load demand is expected to be higher than Saturday daytime
load due to the “Sunday lunch effect”. Additionally, there is a less pronounced morning peak in weekend load
profiles as occupants will become active at different times, resulting in a more gradual increase in the load
demand. However, the evening peak for weekdays and weekends is expected to be similar. (Note: this
behaviour may be different in different regions of the world)
Retail
Commercial offices 36.03%
13.41%
Other
5.08%
Figure 5-1: Contribution of Sub-sectors to Commercial Load Sector Electricity Consumption (UK data, based on [105]).
Commercial sub-sectors are not defined with respect to location as there is much greater variability in building
location than in the residential sector. However, some sub-sectors are only expected to be found in certain
locations, which are outlined where applicable in the discussion below. The following sections describe each of
the commercial sub-sectors and their specific characteristics in terms of load composition and load trends over
time.
Page 65
Modelling and Aggregation of Loads in Flexible Power Networks
Cooking Resistive
Exterior lighting HID
Heating Resistive
Directly connected single-phase motors
HVAC Directly connected three-phase motors
Drive controlled three-phase motors
ICT equipment SMPS
Interior lighting Energy efficient lighting: CFL and LFL
Lifts/elevators Three-phase motors
Refrigeration Single-phase motors
Abbreviations: HID – high intensity discharge.
: included; : not included
Commercial Office Sub-sector – Load Trends
The opening hours of a given commercial premise will dictate its daily load profile (e.g. standard office hours
in the UK, USA, Australia and many other countries are Monday to Friday, from 09:00 – 17:00). In the hours
prior to opening, certain systems will automatically ‘switch-on’ (e.g. HVAC or space heating systems) causing a
gradual increase in demand prior to the actual opening time. A ‘Switch-off’ effect is also present, but is
generally less pronounced.
During the night-time hours, a ‘base load’ continues to be present, e.g. ICT-support systems and night-lighting.
This will exhibit small, if any, seasonal variations. Therefore, the total seasonal variations of the commercial
office sub-sector are not expected to vary in wide ranges, as winter heating loads are replaced with cooling
(air-conditioning) loads in summer. However, seasonal variations for ‘small’ commercial offices may be more
pronounced, due to the lack of highly automated air-conditioning systems and the likelihood of such devices
being manually switched based on personal comfort levels.
Page 66
Modelling and Aggregation of Loads in Flexible Power Networks
The weekend load profile is assumed to be constant and equal to the evening ‘base loading’ conditions.
Transport classification
Load type Load category/sub-category
Small Large
Cooking Resistive
Exterior lighting HID
Heating Resistive
Directly connected three-phase motors
HVAC
Drive controlled three-phase motors
ICT equipment SMPS
Interior lighting Energy efficient lighting: CFL and LFL
Lifts/elevators Three-phase motors
Refrigeration Single-phase motors
Page 67
Modelling and Aggregation of Loads in Flexible Power Networks
some universities to have their own CHP schemes. Universities are usually expected to be present in cities and
large towns, i.e. highly urban and urban areas, although some campuses may be located at the outskirts of
large cities.
Table 5-4: Loads present in education load sub-sector
Cooking Resistive
Exterior lighting HID
Heating Resistive
Directly connected three-phase motors
HVAC
Drive controlled three-phase motors
ICT equipment SMPS
Interior lighting Energy efficient lighting
Lifts/elevators Three-phase motors
Refrigeration Single-phase motor
Three-phase lab equipment Three-phase motors
Page 68
Modelling and Aggregation of Loads in Flexible Power Networks
‘Large’ health facilities will have a larger ‘base load’ because of the presence of various health-critical systems.
This will contribute to a significant night-time load, which will also include the night-time staffing requirements of
such facilities.
Table 5-5: Loads present in health load sub-sector
Cooking Resistive
Exterior lighting HID
Heating Resistive
Directly connected three-phase motors
HVAC
Drive controlled three-phase motors
ICT equipment SMPS
Interior lighting Energy efficient lighting – CFL and LFL
Lifts/elevators Three-phase motors
Refrigeration Single-phase motors
Three-phase equipment Three-phase motors
Hotel/catering classification
Load type Load category/sub-category
Small Large
Cooking Resistive
Exterior lighting HID
Heating Resistive
Directly connected single-phase motors
HVAC Directly connected three-phase motors
Drive controlled three-phase motors
ICT equipment SMPS
Interior lighting Energy efficient lighting: CFL and LFL
Lifts/elevators Three phase motors
Power electronics SMPS
Refrigeration Single-phase motors
Page 69
Modelling and Aggregation of Loads in Flexible Power Networks
Page 70
Modelling and Aggregation of Loads in Flexible Power Networks
It is possible to divide this sub-sector into two main types being community/public recreation facilities, and sport
grounds/stadiums. Community/public recreation facilities would typically include gyms and associated sports
resources. Although the loads found in both types will be similar, the percentage contribution will be different.
However for both applications, lighting is the dominant load. The contribution from HVAC systems will be
determined by the size of the facility. HVAC is not expected to be present in high percentages in stadiums, but
will be present in ‘large’ public recreation facilities, together with demand for hot water (showers and so forth).
Cooking Resistive
Exterior lighting HID
Heating Resistive
Directly connected single-phase motors
HVAC Directly connected three-phase motors
Drive controlled three-phase motors
IT equipment SMPS
Interior lighting Energy efficient lighting – CFL and LFL
Refrigeration Single-phase motors
Page 71
Modelling and Aggregation of Loads in Flexible Power Networks
Warehouse classification
Load type Load category/sub-category
Controlled env. Uncontrolled env.
Cooking
Exterior lighting HID
Heating Resistive
HVAC Directly connected single-phase motors
ICT equipment SMPS
Interior lighting Energy efficient lighting - CFL and LFL
Directly connected single-phase motors
Refrigeration
Directly connected three-phase motors
5.3 Summary
This chapter has presented a detailed description of the two most diverse load sectors (classes of customers),
commercial and residential, which are most likely to undergo the biggest changes in load types participating in
the load class mix in the future. Commercial and residential load sectors (classes of loads) are described in
detail, including all relevant sub-sectors. Additional material is provided in Appendix 5.
Page 72
Modelling and Aggregation of Loads in Flexible Power Networks
6.1 Introduction
Active Distributed Network Cells (ADNC) are distribution networks with a significant amount of distributed
generation (DG) which at specific periods of time (e.g. at minimum loading conditions) is a net exporter of
active power, but at other times (e.g. at maximum loading conditions) may be a net importer of active power.
Micro Grid (MG) is a particular example of ADNC. The MG comprises a Low Voltage (LV) network with loads
and several micro generation systems (usually with power electronic interfaces) connected to it, and a
hierarchical control with management system comprising of two main control levels, being the local and central
control mode, allowing the MG operation as a flexible active cell either when interconnected with the Medium
Voltage (MV) distribution network or when isolated from it.
Large-scale integration of various DG technologies into distribution networks leads to a gradual transition from
passive to active distribution networks, with a large number of small to medium size generating units of various
technologies spread over both MV and LV levels. Approximating the dynamics of these networks using passive
lumped loads, as has been done previously, lacks the accuracy when simulating the dynamic behaviour of
complex ADNC. Although most of the work done in this field has been focused on distribution networks
connecting wind farms, aggregated models able to appropriately account for the ADNC dynamics (which may
comprise a large number of active components with various technologies and sizes) have been developed. The
aim has been to replace the whole detailed model (or a part of it)with a reduced order model exhibiting
similar dynamic characteristics as the whole original ADNC. The ADNC should be represented through a
combination of aggregated passive components and aggregated active components, so that power system
operators can assess their impact, with a considerable computational time saving, but keeping the required
accuracy regarding the phenomenon under study.
Two methods to derive aggregated models of power systems (commonly known as dynamic equivalents) are
system reduction and system identification techniques [106], [107].
Reduction techniques are based on aggregation and elimination of some components of the detailed model of
the system. The two mostly found in literature are known as modal analysis [108], [109], and coherency based
aggregation methods [110–112]. Modal analysis aims to reduce the system by aggregating similar modes and
eliminating those modes that are not of interest. There are three main drawbacks: this approach is time
consuming, the reduced order models do not have structural identity (they are purely mathematical in nature)
and linear methods cannot properly capture complex power system dynamics following major disturbances. As
an alternative to modal analysis based reduction techniques, coherency based aggregation became very
popular by being able to identify coherent generators based on rotor angle swings, to aggregate those
coherent generators into a single equivalent generator through the Zhukov’s method [113], [114] and to reduce
the network by means of the Dimos method [114]. Several methods based on linear models have been used for
coherency identification.[106], with techniques including inspection of time responses, pattern recognition, closest
unstable equilibrium point, Lyapunov function, weakly coupled subsystems and modes of low frequency
oscillations. Coherency identification without linearisation has also been attempted, as discussed in [115–119].
In system identification techniques, the dynamic equivalents are determined using a data set comprised of
either by simulated data or real system measurements at a boundary bus between the study system and the
system to be reduced. After a model structure is chosen, its parameters are adjusted so that the model response
matches the measurements. Similar procedures have been adopted for aggregation of loads based on
measurement based models (see Chapter 4). Among system identification techniques, Artificial Neural Networks
(ANN) are the most preferred measurement based method due to its high inherent ability for modelling
Page 73
Modelling and Aggregation of Loads in Flexible Power Networks
nonlinear dynamic systems such as power system dynamic equivalents, with the system dynamic properties
being obtained only from data [120–127].
On model performance, it is expected that the dynamic equivalent obtained from coherency based methods
would be considerably more reliable and accurate than those set up by system identification method, because
it is determined from a physical model rather than an approximation based on measurements. However, the
reduction based equivalent method requires the physical knowledge about the network structure, its parameters,
and the system operating status. In some cases, the available physical knowledge may not be sufficient to
develop accurate and reliable dynamic equivalents [124–126], [128]. Moreover, the application of coherency
based methods is restricted to power systems with a relatively small number of large synchronous generators
concentrated in a few areas. In addition, when there is a need to update the dynamic equivalent, the tuning
task may prove difficult since there is a little indication about which parameters should be adjusted [129],
[130]. Also, dynamic equivalents derived from one class of event are typically not very successful for other
classes [131], [132].
DG technologies based on induction machines or connected to a network through power electronic interfaces
have quite different control systems which can be difficult to aggregate [133]. Some DG technologies do not
have rotating parts (such as fuel cells and PV systems). Therefore, the term coherency widely adopted by
classic reduction methods becomes less meaningful since induction generators do not have synchronizing torques
and the power electronic based interfaces can almost completely separate the dynamic behaviour of the
generator from the network. This results in different dynamic behaviour of DG strongly influenced by the
controllers of their power electronic converters. Thus, system identification techniques have widen to derive
aggregated models of complex power systems that include distribution networks integrating large amounts of
DG that are not limited to synchronous generators..
It should be pointed out that they are currently no aggregated models of ADNCs and MGs that are in practical
use by utilities. The aggregated modelling of ADNCs and MGs is in the earliest stages and recommendations
for model development presented in this chapter are based on findings coming largely from academic
research. This chapter presents several recommended modelling approaches in detail. A comprehensive
literature review on the state of the art in aggregated DG modelling and ADNC equivalencing is presented in
0, along with conclusions and recommendations. (Note: Aggregation of different generation technologies should
be done with care in case of scaling the response to other situations, e.g., wind and solar generation should be
handled separately as their production pattern is determined by the meteorological conditions.)
6.2 Methodology to Develop Aggregate Models of ADNC and MG Using Nonlinear Techniques
System identification based methods cover a number of issues ranging from data processing to validation of
the measurement based model. With a given observed data set the main tasks of system identification are two-
fold: i) Selecting the model structure; ii) Estimating the model parameters in a way that is consistent with the
identification criterion. Another important step, however, is to evaluate the model performance regarding its
intended purpose, i.e., model validation [134], [135]. Thus a general conceptual approach involves the main
steps described in the sections that follow.
Page 74
Modelling and Aggregation of Loads in Flexible Power Networks
Therefore, a suitable data acquisition process is required that deals with both input and output observations
(either measured or simulated) of the system to be reduced. Once decided where and what to measure, the
next steps comprise the sampling time definition, ( T ), and also the required number of measurements, ( N ).
Thus, the data set of system inputs and outputs, u kT and y kT , respectively, is given by::
Z N u 1 , y 1 u 2 , y 2
u N , y N (6.1)
It should be assumed that there are always signals that cannot be controlled, such as measurement noise and
possibly uncontrolled inputs, commonly denoted by v kT , that affect the system output.
In formal terms, after data set collection the user seeks a relationship between past observations ( u k 1 and
y k 1 ) and future outputs, y k , as:
y k g u k 1 , y k 1 v k (6.2)
It has been assumed that g belongs to a family of functions that is parameterized in terms of a finite number
of parameters commonly denoted by , i.e. the model structure, M . Also, it should be useful to write g as a
concatenation of two mappings. One that takes the number of past observations and maps them into a vector
k , of fixed dimensions and another that takes this vector to the space of the model outputs,
g u k 1 , y k 1; g k , ; . Thus, the model structure selection process should be decomposed into two
partial problems [136]: i) How to choose the regression vector k , from past inputs and outputs; ii) How to
choose the nonlinear mapping g , from the regressor space to the output space. Depending on the system
identification based approach used, relying on “black-box” or “grey-box” methodologies, both partial
problems should be considered or only the last one, respectively.
Page 75
Modelling and Aggregation of Loads in Flexible Power Networks
vk
System to be +
+ y k
reduced
k + k
-
Aggregated
model Criterion
g k , y k |
Page 76
Modelling and Aggregation of Loads in Flexible Power Networks
In addition to the ANN itself, the model requires two supplementary functions, being a mapping function to
prepare the ANN inputs, and a de-mapping function to process the outputs from the ANN to calculate complex
power. These two functions represent the interface between the ANN and the retained network. The equivalent
model interacts with the retained system through the boundary buses. It is perturbed by the voltages at these
buses and reacts by supplying the corresponding complex power at each time interval. The ANN itself acts as a
Norton model, where the normalised deviations of voltage are used as the main inputs and the normalised
deviations of currents represent the outputs. In addition to the input voltages, past values of currents and
voltages are also introduced at the input layer to achieve the recurrent structure. Thus, the ANN is able to
capture the dynamic nature of the original system and to maintain the continuous-time operation of the entire
network.
Figure 6-2: Basic Principles of the Recurrent ANN Based Dynamic Equivalent Approach.
The current rather than power is used as output from the ANN, as it represents an independent variable,
whereas power depends on voltage, which is an input to the ANN. Therefore, the use of current gives better
convergence in the training process compared to complex power due to the complete decoupling between
outputs and inputs of the ANN. The use of normalised deviations as inputs and outputs in the ANN allows the
use of the equivalent model under new initial power-flow condition. The ANN in this case represents a
normalised model scaled on initial conditions at the boundary buses. Augmenting this feature with the
independent representation of active and passive elements results in a universal model, which is capable of
simulating the original system under different operating conditions.
Developing a recurrent ANN based dynamic equivalent model involves performing common system
identification procedures, namely i) Data preparation; ii) Definition of the ANN structure; iii) Training
procedure; iv) Model validation. Each of these tasks is briefly discussed below.
Data Preparation
Several three-phase short circuits were simulated at different locations in the retained network using the
detailed model of the system implemented in the Power System Dynamics (PSD) simulation package [137],. The
simulation of each fault is carried out for 10s, which is enough to restore the steady state conditions after the
fault clearance. A 10ms integration time step is used in the simulation, which results in 1000 patterns with each
fault. Complex voltages and injected currents at the boundary-buses are stored during the fault simulation and
subsequently used to prepare suitable patterns for training the ANN.
Definition of the ANN Structure
Since the ANN is required to capture the dynamics of the replaced network, characterised by currents,
recursive loops are used to feedback deviations of normalised current to the input layer with time delays. In
Page 77
Modelling and Aggregation of Loads in Flexible Power Networks
addition, the ANN interacts with the retained network and recognises its dynamic behaviour through the
voltages. Therefore, past values of normalised voltage deviations are also used at the input layer. With two
boundary buses, the ANN has 4 main inputs representing the real and imaginary components of normalised
voltage-deviations. Furthermore, normalised deviations of these voltages at four previous time intervals are
received at the input neurons. On the other hand, four outputs representing the real and imaginary components
of normalised deviations of currents are obtained from the ANN. Four recursive loops with delay actions from
each output are fed back to the input layer of the ANN. This results in a total number of 20 inputs in addition to
16 recurrent loops. The ANN contains two hidden layers with 10 and 5 neurons respectively. All hidden layers
comprise neurons with nonlinear-sigmoid activation functions, while neurons in the output layer have linear
functions.
Training Procedure
Patterns corresponding to eight different three-phase short circuits are used in the training process. Time-
histories of voltage variables are involved in the input features to the network, while real recurrent loops from
output current variables are considered. A special program is developed to train the ANN with actual recurrent
loops. With 1000 pattern belonging to each waveform, 8000 patterns are used in the training process, which is
accomplished offline without actual interaction with the retained network. However, the test and validation of
the developed dynamic model are based on real-online interactions. The results from the training process (i.e.
biases and weights) are saved to be used in the implementation procedures.
Model Validation
Since the equivalent model is intended for online applications, it has to be implemented in the PSD package as
an unconventional power source [137] in such a way that it interacts with the retained subsystem at each time
step in a similar way to that of the original network. Thus, the recurrent ANN based dynamic equivalent is
simulated in a so-called regulator file and integrated into the network through one or more buses. The
processing within the regulator files is accomplished using pre-constructed blocks to describe the dynamic
behaviour of the element, as described in 0. After replacing the external system by the ANN-based dynamic
model, the performance is investigated online and compared to that of the full system. For this purpose,
disturbances not used in the training procedure should be used in order to evaluate the generalisation
capability of the ANN. Detailed results can be found in [138–140].
Page 78
Modelling and Aggregation of Loads in Flexible Power Networks
constant impedance, constant power and a constant current load model (ZIP load model), accounting for the
static load part, connected in parallel with an induction motor model accounting for the dynamic load part. The
composite load model is referred to as the ZIP-IM load model [33], [59], [144], [145]; iii) The mechanical
torques of both generator and induction motor are assumed to be constant.
PL jQ L
IM
ZIP
P jQ
CONVERTER SG
PG jQ G
The proposed converter-connected generator model can be used to represent micro-turbines and wind turbines,
especially the direct drive synchronous generator type [146–149]. The third-order of synchronous generator
model was chosen as it is adequate to represent the dynamic behaviour of a synchronous generator [148–
150], while preserving the low order of the overall equivalent model, which was the main goal of development
of equivalent model of ADNC. In addition, the full converter model used a simple model that is sufficient to
represent the converter without neglecting its main principle, i.e. that the real power flowing through the
converter is balanced. The chosen converter model preserves the dynamic characteristics represented by the
DC-link equation (the capacitor linking between inverter and rectifier) [146], [147]. A more detailed
description is presented in 0.
Estimation of Parameters
The parameter estimation procedure is shown in Figure 6-4 [141], [142]. Initially, the pre-processing procedure
is performed on the input and output signals (voltage, v(t), frequency, f(t), real power P(t), and reactive power
Q(t)). The pre-processing procedure, also known as data conditioning, filters out the noise contained in the
signals. After the data conditioning process, the filtered input signals, voltage and frequency (V, f) are
imported into the nonlinear grey-box model together with the initial values of model parameters to be
estimated. The System Identification Toolbox in MATLAB is used to develop the grey-box model and to produce
the output responses of real and reactive power ( Pˆ , Qˆ ). The simulated output responses ( Pˆ , Qˆ ) are obtained
from the model using estimated model parameters based on nonlinear least square optimization, according to
the specified identification criterion. The parameter estimation procedure is performed in MATLAB environment.
Nonlinear least square is an optimization-based technique used to search for the best model parameters by
fitting a curve through data. It fits data to any equation that defines Y as a function of X and one or more
parameters. It finds the values of those parameters generating the curve that come closest to the data
(minimizing the sum of the squares of the vertical distances between data points and curve) [151]. This
technique requires a model of the analysed signal. For dynamic equivalent of ADNC, the signal model is
defined by the grey-box model structure. Nonlinear least square optimization is generally used where the goal
is to minimize the difference between the physical observation and the prediction from the mathematical model.
More precisely, the goal is to determine the best values of the unknown nonlinear parameters of the model in
order to minimize the squared errors between the measured values of the signal and the computed ones. The
squared error functions for measured and simulated output are defined as follows:
Page 79
Modelling and Aggregation of Loads in Flexible Power Networks
1 n 2 1 n
min P min Pkm Pks
2
kP min
n k 1 n k 1
n
(6.3)
1 1 n
min Q min kQ min Qkm Qks
2 2
n k 1 n k 1
where Pm and Qm are measured active and reactive power, Ps and Qs are simulated active and reactive power
from the grey-box model.
V, f +
Initial parameters P̂, Q̂
Grey-box model
_
The parameter estimation process is performed through iterations. In successive iterations, the sum of squared
errors between the simulated and measured output signals is calculated. Based on this, the parameters are
updated through the optimization algorithm. The iterative process and parameter tuning continues until the
squared errors (ε) are within a predefined threshold. Three algorithms are available in MATLAB software for
solving nonlinear least square problems: trust-region, Levenberg-Marquardt and Gauss-Newton [152]. The
Levenberg-Marquardt algorithm was used for estimating grey-box model parameters in [141–143].
Model Validation
Similar to the black box approach, the model validation can be performed using the comparison between the
simulated output of the model corresponding to the original input with the original output of the system. The
developed equivalent model has to be implemented in any simulation software for large system application
under various dynamic system studies. The effectiveness and accuracy of the developed model can be
confirmed by comparing the responses from original external system with responses from developed equivalent
model. The results of model validation are given in 0
Page 80
Modelling and Aggregation of Loads in Flexible Power Networks
for detailed analysis and the external subsystem that will be replaced by the equivalent model [153–155], as
depicted in Figure 6-5.
Boundary
bus
Internal subsystem – MV network External subsystem - MicroGrid
LV
MV
IDE, IQE
MG slow dynamics
equivalent model
I D, I Q
IDS, IQS
MC
VD, VQ
AC
DC
Storage
device
Figure 6-5: The Network Equivalent Model Including the MG Equivalent Aggregated Model.
As it can be observed from Figure 6-5, the dynamic system to be identified consists of a set of differential and
algebraic equations, corresponding to the dynamic models of the several micro generation systems describing
the state evolution over time of the physical system – the Micro Grid [154] according to the MG concept
presented in 0. Therefore, the dynamic equivalent to be developed will replace the MG detailed model, which
is assumed to be the external subsystem, by a reduced order model according to the following guidelines: i)
The MG dynamic equivalent must be an accurate representation of the detailed model concerning the transient
analysis to be performed; ii) The cost of building the dynamic equivalent must be much smaller than the cost of
performing the transient analysis using the MG detailed model; iii) The obtained dynamic equivalent has to be
integrated into dynamic simulation tools.
The MG detailed model is excited through generated disturbance scenarios into the internal subsystem and the
corresponding MG dynamics are captured by means of the electrical variables measured at the system
boundary bus. Thus, the MG dynamic equivalents will react to the boundary bus voltage and system frequency
changes by varying the injected current into the retained subsystem.
Based on the physical insights regarding the MG operation, two different time scales regarding the MG
dynamic response can be distinguished [153–155]. The main storage device interfaced through a Voltage
Source Inverter (VSI) has fast dynamic responses and the controllable micro-generation systems interfaced
through PQ inverter controls have slow dynamic responses. Therefore, suitable dynamic equivalents for MG
comprise the detailed model of the main storage device VSI control and the MG slow dynamics equivalent
model corresponding to the aggregated model of the remaining MG components, as it can be observed from
Figure 6-5. Then, the system identification process will also identify the equivalent model able to represent the
MG slow dynamic behaviour.
Page 81
Modelling and Aggregation of Loads in Flexible Power Networks
a significant advantage, especially when there is a limited understanding of the relations between system
variables.
Data Processing
As the data set is the basis of any successful identification procedure, a numerical experiment should be
designed to produce a set of samples that describe how the system behaves over its entire range of operation.
For this purpose the MMG detailed model is used taking into account the following issues: i) The design of input
signals which lead to an informative data set; ii) The techniques for preparing data for neural network
modelling.
Based on the system definition presented before, as well as in the engineering expertise, adequate input
signals have been designed. Thus, after MG islanding, several perturbations occurring in the MV network are
simulated and both the input and output signals are measured according to a suitable sample time. As already
mentioned previously, boundary bus voltages expressed in the synchronous reference frame, D Q , and
system angular frequency are considered as inputs, while the boundary bus injected current from the tie line,
expressed in the D Q reference frame, are considered as outputs. Thus, the TDNN is disturbed by both
boundary bus voltage and network frequency variations. It reacts to these variations by varying the injected
currents into the boundary bus, operating according to the principles of a Norton model [154].
Boundary bus voltages, system frequency and injected currents are stored in a database during the
perturbation simulations in order to build suitable training patterns. Since the data set is almost the only source
of information to build the TDNN based dynamic equivalent model, the number of samples should be large
enough in order to form appropriate training and validation data sets.
Since signals are likely to be measured in different physical units, it is recommended to remove the mean and
scale all signals to the same variance to avoid the tendency that the signal of largest magnitude will be too
dominating. Moreover, scaling makes the training algorithm numerically robust and leads to a faster
convergence and tends to give better models [156].
In order to generate a more robust TDNN, which is able to simulate the MG dynamic behaviour under different
operating conditions, normalized deviations of voltage, system frequency and currents from the corresponding
steady state analysis could be used [121], [126], [138], [139], [156].
Page 82
Modelling and Aggregation of Loads in Flexible Power Networks
When an unlimited amount of training data is available, the MLP neural network architecture determination
relies on fully connected neural networks. In this case, the architecture selection is reduced to a matter of
choosing a number of hidden units and the activation functions types of an MLP neural network. A regularization
technique is further applied in order to deal with the bias/variance dilemma.
Model Validation
After the training procedure, the performance of the TDNN based equivalent models with less generalization
error is evaluated in the dynamic simulation platform. A MG slow dynamics equivalent model is then embedded
in the validation module forming the MMG equivalent model. The model performance is evaluated by
comparing its response following perturbations that occur in the retained subsystem not used during the training
phase with the response obtained using the MG detailed model. Results of the MG equivalent model based on
TDNN can be found in [154].
R K1 K2 T2 TD (6.4)
Since the MG slow dynamics equivalent model should be a current source, the instantaneous power theory
[158] was used in order to determine the network injected current as depicted in Figure 6-6. A more detailed
description is presented in Appendix 6-C.
Page 83
Modelling and Aggregation of Loads in Flexible Power Networks
VD
VQ I DR
1 Tm ,max Instantaneous
t power
R
theory
grid +
+ m
a K2 mb 1 Tm Pm I QR
- T2 s 1 TD s 1
+ Qref
K
1
ref 1 s Tm ,min
Page 84
Modelling and Aggregation of Loads in Flexible Power Networks
Mutation
Evaluation function
Changing multimachine Power system
Selection power system matrices matrices in D-Q
i
Dynamic simulation of
MMG equivalent model
Hi
- +
Target
6.5 Summary
This chapter presented two approaches to develop dynamic equivalents of ADNC and MG, namely black box
(ANN based) and grey box (physical) approach. As a new area of power system modelling work, with very
few reports and research papers available and limited practical use by utilities of ADNCs and MGs
aggregated models, recommendations for development of aggregate models of ADNCs and MGs presented in
this chapter are based on findings of academic research. The main conclusions of the chapter can be
summarised as follows: i) Both black box (the recurrent ANN based) and grey box (based to a certain extent
on physical understanding of the structure and composition of the network) methods can be used for
development of aggregate dynamic models. ii) Both derived dynamic equivalent models are in a simple linear
or nonlinear form e.g. state space form which is very flexible and compatible with various commercial
simulation tools like PSS/E, DIgSILENT PowerFactory, MATLAB etc.; iii) The input data for both approaches can
be either simulated or measured network responses at the boundary buses regardless of the size and
complexity of the network. iv) The models facilitate significant simplification of dynamic analysis of large and
complex power systems with interconnected ADNC containing DG of diverse technologies; v) Considerable
computational effort and frequent user interaction is required to derive the TDNN based MG dynamic
equivalent which is a very time consuming task, where the initial values of both inputs and outputs of TDNN, as
well as their maximum deviations from initial values, have to be updated whenever the initial steady state
conditions are changed and the model domain of validity is restricted to the test system used to generate the
data set; vi) The grey box approach requires significantly less computational effort to develop the model and
the models domain of validity is largely extended.
Page 85
Modelling and Aggregation of Loads in Flexible Power Networks
7.1 Conclusions
This report presented the major results and conclusions of the group’s three year work. It presents a critical
detailed overview of the two existing and most widely used methodologies for load modelling, the
measurement based and component based approaches and identifies their major advantages and
disadvantages. This overview clearly indicates the need for a hybrid approach in the future that will combine
the individual strengths of the existing approaches, taking into account the data acquisition capabilities offered
by modern measurement systems. The report also summarises the existing static and dynamic load models, the
load classes that these load models are valid for, and the large amount of data for load model parameters
that can be found in public literature.
The analysis of existing load modelling methodologies and load models presented in Chapters 2 and 3, is
supported by findings of an international survey on load modelling performed by the WG. A comprehensive
questionnaire on load modelling practices was developed by the WG and distributed during the
summer/autumn of 2010 to more than 160 utilities and system operators from over 50 countries on all
continents. The report summarises some of the key findings of that questionnaire, based on 97 responses
(60.6% response rate) received by September 2011. The survey revealed that the constant real and reactive
power load model (constant P and Q) is the most widely used load model for steady state power system
studies. It also showed that static load models are still the most commonly used even for dynamic system studies
and that only about 30% of utilities and transmission system operators represent dynamic load by some form
of induction motor model. Therefore, it can be considered that the number of models used, in actual studies, is
limited although the variety of load models is wide.
The report then presented, in Chapter 4, a set of recommendations for load model development, using both the
measurement based and component based approaches. The recommendations were based on a combination of
best practises coming from past analysis, and the practical experience of the WG members. The step-by-step
recommendations for load model development given in Chapter 4 can be used by engineers in distribution and
transmission companies for developing load models within their own systems.
Accurate representation of aggregate loads for power system analysis requires detailed analysis of the loads
connected downstream of the point of aggregation. Although system loads are typically classify in one of the
three general load sectors (residential, commercial and industrial), this is typically not sufficient to accurately
describe the diversity in load structure and load composition, as well as typical load profiles within each load
sector. The general load sectors should be further divided into sub-sectors, for which similar generic aggregate
load models can be developed.
Large scale integration of inverter-interfaced small generation units, with power ratings less than a few tens of
kilowatts in LV networks and Distributed Generation (DG) in MV networks, calls for the development of
equivalent mathematical models of Active Distribution Network Cells (ADNC) and Micro Grids (MG). Such
models should be suitable for representing ADNC and MG in steady state and dynamic studies of large power
networks. As specific ADNC and MG properties cannot be adequately modelled using conventional dynamic
equivalencing methods, the report suggests exploiting system identification techniques for this purpose.
Considering that there is no current industrial practice on development or use of these types of aggregate
models, the recommendations made in this report are based on relatively limited academic research
experience. Two approaches are recommended to derive dynamic equivalents for DNC and MG. The first is a
black-box modelling approach based on Artificial Neural Networks (ANN) that tries to exploit the full response
of the MG when excited after a disturbance. The second is a grey box modelling approach exploiting the
physical behaviour of the different components of the ADNC or MG. The chapter also provides conclusions
related to modelling of ADNC and MG for large power system studies.
Page 86
Modelling and Aggregation of Loads in Flexible Power Networks
Finally, in spite of every effort taken to provide as comprehensive report on load modelling as possible some
issues might have been missed. It is also important to re-emphasise that not all phenomena can be studied with
one simple/single load model. The user should ultimately decide on the type of model to use based on the
phenomena which is to be analysed, e.g., steady state, small disturbance or large disturbance power system
analysis would require different types of load models to be used.
Page 87
Modelling and Aggregation of Loads in Flexible Power Networks
References
[1] WECC Load Modelling Task Force, “Composite Load Model for Dynamic Simulations: Report 1,” 2012.
[2] K. Walve, “Modelisation des composantes de reseaux soumis de fortes perturbations,” in Rapport CIGRE 38-18, 1986.
[3] IEEE Task Force on Load Representation for Dynamic Performance, “Load representation for dynamic performance
analysis (of power systems),” IEEE Transactions on Power Systems, vol. 8, no. 2, pp. 472–482, May 1993.
[4] B. Khodabakhchian and G.-T. Vuong, “Modeling a mixed residential-commercial load for simulations involving large
disturbances,” IEEE Transactions on Power Systems, vol. 12, no. 2, pp. 791–796, May 1997.
[5] D. N. Kosterev, C. W. Taylor, and W. A. Mittelstadt, “Model validation for the August 10, 1996 WSCC system
outage,” IEEE Transactions on Power Systems, vol. 14, no. 3, pp. 967–979, 1999.
[6] W.-S. Kao, “The effect of load models on unstable low-frequency oscillation damping in Taipower system experience
w/wo power system stabilizers,” IEEE Transactions on Power Systems, vol. 16, no. 3, pp. 463–472, 2001.
[7] D. Han, J. Ma, R. He, and Z.-Y. Dong, “A Real Application of Measurement-Based Load Modeling in Large-Scale Power
Grids and its Validation,” IEEE Transactions on Power Systems, vol. 24, no. 4, pp. 1756–1764, Nov. 2009.
[8] K. Yamashita, S. M. Villanueva, and J. Milanović, “Initial Results of International Survey on Industrial Practice on Power
System Load Modelling Conducted by CIGRE WG C4.605,” in CIGRE symposium, Bologna, Italy, 2011.
[9] A. M. Gaikwad, R. J. Bravo, D. Kosterev, S. Yang, A. Maitra, P. Pourbeik, B. Agrawal, R. Yinger, and D. Brooks, “Results
of residential air conditioner testing in WECC,” in 2008 IEEE Power and Energy Society General Meeting - Conversion
and Delivery of Electrical Energy in the 21st Century, 2008, pp. 1–9.
[10] D. Kosterev, A. Meklin, J. Undrill, B. Lesieutre, W. Price, D. Chassin, R. Bravo, and S. Yang, “Load modeling in power
system studies: WECC progress update,” in 2008 IEEE Power and Energy Society General Meeting - Conversion and
Delivery of Electrical Energy in the 21st Century, 2008, pp. 1–8.
[11] P. Kundur, Power System Stability and Control. New York: McGraw-Hill, Inc., 1994.
[12] P. Pourbeik and A. Gaikwad, “A Summary of EPRI Load Modelling Efforts,” 2011.
[13] J. L. Aguero, M. Beroqui, and S. Achilles, “Aluminum plant. Load modeling for stability studies,” in 199 IEEE Power
Engineering Society Summer Meeting. Conference Proceedings (Cat. No.99CH36364), vol. 2, pp. 1330–1335.
[14] EPRI, “Advanced Load Modeling – Entergy Pilot Study,” 2004.
[15] EPRI, “Comprehensive Load Modeling for System Planning Studies”, EPRI Product ID 1015999,Palo Alto, USA, 2009.
[16] EPRI, “Load Modeling Using a Measurement Based Approach: Phase-2”, EPRI Product ID 1016255, Palo Alto, USA,
2007.
[17] K. Yamashita, Y. Kitauchi, and K. Katsuragi, “Influence of voltage sags on dynamic voltage behaviour with massive
inverter-based loads,” in IEEE PES Innovative Smart Grid Technologies, 2012, pp. 1–8.
[18] A. J. Collin, I. Hernando-Gil, J. L. Acosta, and S. Z. Djokic, “An 11 kV steady state residential aggregate load model.
Part 1: Aggregation methodology,” in 2011 IEEE Trondheim PowerTech, 2011, pp. 1–8.
[19] A. J. Collin, J. L. Acosta, I. Hernando-Gil, and S. Z. Djokic, “An 11 kV steady state residential aggregate load model.
Part 2: Microgeneration and demand-side management,” in 2011 IEEE Trondheim PowerTech, 2011, pp. 1–8.
[20] C. Cresswell, S. Djokic, and S. Munshi, “Analytical Modeling of Adjustable Speed Drive Load for Power System
Studies,” in 2007 IEEE Lausanne Power Tech, 2007, pp. 1899–1904.
[21] D. Karlsson and D. J. Hill, “Modelling and identification of nonlinear dynamic loads in power systems,” IEEE Transactions
on Power Systems, vol. 9, no. 1, pp. 157–166, 1994.
[22] P. C. Krause, Analysis of Electric Machinery. New York: McGraw-Hill, Inc., 1986.
[23] T. Frantz, T. Gentile, S. Ihara, N. Simons, and M. Waldron, “Load Behaviour Observed in LILCO and RG&E Systems,”
IEEE Transactions on Power Apparatus and Systems, vol. PAS-103, no. 4, pp. 819–831, Apr. 1984.
[24] K. Yamashita, M. Asada, and K. Yoshimura, “A development of dynamic load model parameter derivation method,” in
2009 IEEE Power & Energy Society General Meeting, 2009, pp. 1–8.
[25] EPRI, “Power Quality Monitoring Data Mining for Power System Load Modelling,” 2006.
[26] K. Yamashita, S. Djokic, J. Matevosyan, F. O. Resende, L. M. Korunovic, Z. Y. Dong, and J. V. Milanovic, “Modelling and
Aggregation of Loads in Flexible Power Networks – Scope and Status of the Work of CIGRE WG C4.605,” in IFAC
Symposium on Power Plants and Power Systems Control, 2012.
[27] S. Z. Zhu, Z. Y. Dong, K. P. Wong, and Z. H. Wang, “Power system dynamic load identification and stability,” in
PowerCon 2000. 2000 International Conference on Power System Technology. Proceedings (Cat. No.00EX409), vol. 1,
pp. 13–18.
Page 88
Modelling and Aggregation of Loads in Flexible Power Networks
[28] J. Ma, Z. Y. Dong, and P. Zhang, “Using a support vector machine (SVM) to improve generalization ability of load
model parameters,” in 2009 IEEE/PES Power Systems Conference and Exposition, 2009, pp. 1–8.
[29] V. Knyazkin, C. A. Canizares, and L. H. Soder, “On the Parameter Estimation and Modeling of Aggregate Power
System Loads,” IEEE Transactions on Power Systems, vol. 19, no. 2, pp. 1023–1031, May 2004.
[30] EPRI, “Measurement-Based Load Modeling,” 2006.
[31] Power System Study Group, “Integrated Analysis Software for Bulk Power System Stability.”
[32] TRACTEBEL Engineering GDFSUEZ, “No Title,” www.eurostag.be/en/products/eurostag/the-reference-power-system-
dynamic-simulation/. .
[33] H. Renmu, M. Jin, and D. J. Hill, “Composite Load Modeling via Measurement Approach,” IEEE Transactions on Power
Systems, vol. 21, no. 2, pp. 663–672, May 2006.
[34] K. Yamashita, K. Tokumitsu, A. Koyama, and M. Tatematsu, “Automatic Extraction Method Suitable for Deriving Load
Model Parameters,” in IEEE IPEC, 2012.
[35] J. Ribeiro and F. Lange, “A New Aggregation Method for Determining Composite Load Characteristics,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-101, no. 8, pp. 2869–2875, Aug. 1982.
[36] K. Lidija and S. Dobrivoje, “Load modeling in distribution networks,” Facta universitatis - series: Electronics and
Energetics, vol. 15, no. 3, pp. 419–427, 2002.
[37] C. Taylor, Power System Voltage Stability. New York: McGraw-Hill, Inc., 1994.
[38] L. Hajagos and B. Danai, “Laboratory Measurement of Modern Loads Subjected to Large Voltage Changes for Use in
Voltage Stability Studies,” 1996.
[39] S. Ihara, M. Tani, and K. Tomiyama, “Residential load characteristics observed at KEPCO power system,” IEEE
Transactions on Power Systems, vol. 9, no. 2, pp. 1092–1101, May 1994.
[40] K. Yamashita, M. Asada, and K. Yoshimura, “A development of dynamic load model parameter derivation method,” in
2009 IEEE Power & Energy Society General Meeting, 2009, pp. 1–8.
[41] Y. Baghzouz and C. Quist, “Determination of static load models from LTC and capacitor switching tests,” in 2000 Power
Engineering Society Summer Meeting (Cat. No.00CH37134), vol. 1, pp. 389–394.
[42] IEEE Task Force on Load Representation for Dynamic Performance, “Bibliography on load models for power flow and
dynamic performance simulation,” IEEE Transactions on Power Systems, vol. 10, no. 1, pp. 523–538, 1995.
[43] M. Kostić, “The Influence of network voltage on loads and rational consumption of electrical energy,” Beograd, Serbia,
1997.
[44] D. Tasić, “The Influence of Static and Dynamic Load Characteristic on Voltage Collapse in Electrical Power System,”
University of Niš, 1996.
[45] S. Z. Djokic, A. J. Collin, and C. E. Cresswell, “The Future of Residential Lighting: Shift from Incandescent to CFL to LED
light sources,” in Illuminating Engineering Society of North America Annual Conference, 2009.
[46] C. Cresswell and S. Z. Djokic, “Representation of Directly Connected and Drive-controlled Induction Motors. Part 1:
Single-phase Load Models; Part 2 Three-phase Load Models,” in 18th International Conference on Electrical Machines,
2008.
[47] C. Cresswell and S. Z. Djokic, “Steady State Models of Low-Energy Consumption Light Sources,” in 16th Power System
Computation Conference, 2008.
[48] C. Cresswell, S. Z. Djokic, K. Ochije, and E. Maspherson, “Modelling of Non-Linear Electronic Loads for Power System
Studies: A Qualitative Approach,” in 19th International Conference on Electrical Distribution (CIRED), 2007.
[49] D. J. Hill, “Nonlinear dynamic load models with recovery for voltage stability studies,” IEEE Transactions on Power
Systems, vol. 8, no. 1, pp. 166–176, 1993.
[50] J. V. Milanović, “The Influence of Loads on Power System Electromechanical Oscillations,” The University of Newcastle,
Australia, 1996.
[51] D. Karlsson and D. J. Hill, “Modelling and identification of nonlinear dynamic loads in power systems,” IEEE Transactions
on Power Systems, vol. 9, no. 1, pp. 157–166, 1994.
[52] A. Borghetti, R. Caldon, and C. A. Nucci, “Generic dynamic load models in long-term voltage stability studies,”
International Journal of Electrical Power & Energy Systems, vol. 22, no. 4, pp. 291–301, May 2000.
[53] I. R. Navarro, O. Samuelsson, and S. Lindahl, “Automatic determination of parameters in dynamic load models from
normal operation data,” in 2003 IEEE Power Engineering Society General Meeting (IEEE Cat. No.03CH37491), pp.
1375–1378.
[54] T. Y. J. Lem and R. T. H. Alden, “Comparison of experimental and aggregate induction motor responses,” IEEE
Transactions on Power Systems, vol. 9, no. 4, pp. 1895–1900, 1994.
Page 89
Modelling and Aggregation of Loads in Flexible Power Networks
[55] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of electric machinery and drive systems, Second. New York: Wiley-
IEEE Press, 2002.
[56] C.-J. Lin, A. Y.-T. Chen, C.-Y. Chiou, C.-H. Huang, H.-D. Chiang, J.-C. Wang, and L. Fekih-Ahmed, “Dynamic load models
in power systems using the measurement approach,” IEEE Transactions on Power Systems, vol. 8, no. 1, pp. 309–315,
1993.
[57] Y. Li, H.-D. Chiang, B.-K. Choi, Y.-T. Chen, D.-H. Huang, and M. G. Lauby, “Representative static load models for
transient stability analysis: development and examination,” IET Generation, Transmission & Distribution, vol. 1, no. 3, p.
422, 2007.
[58] S. A. Y. Sabir and D. c. Lee, “Dynamic Load Models Derived from Date Acquired During System Transients,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-101, no. 9, pp. 3365–3372, Sep. 1982.
[59] H.-D. Chiang, J.-C. Wang, C.-T. Huang, Y.-T. Chen, and C.-H. Huang, “Development of a dynamic ZIP-motor load
model from on-line field measurements,” International Journal of Electrical Power & Energy Systems, vol. 19, no. 7, pp.
459–468, Oct. 1997.
[60] H. Renmu, M. Jin, and D. J. Hill, “Composite Load Modeling via Measurement Approach,” IEEE Transactions on Power
Systems, vol. 21, no. 2, pp. 663–672, May 2006.
[61] G. Shackshaft, O. C. Symons, and J. G. Hadwick, “General-purpose model of power-system loads,” Proceedings of the
Institution of Electrical Engineers, vol. 124, no. 8, p. 715, 1977.
[62] K. Tomiyama, S. Ueoka, T. Takano, I. Iyoda, K. Matsuno, K. Temma, and J. J. Paserba, “Modeling of load during and
after system faults based on actual field data,” in 2003 IEEE Power Engineering Society General Meeting (IEEE Cat.
No.03CH37491), pp. 1385–1391.
[63] G. Delille, “Participation of Storage in the Advanced Management of Power Systems. Organizational, Technical and
Economic Approaches in Distribution Grids,” PhD thesis, University of Lille, France, 2010.
[64] D.-Q. Ma and P. Ju, “A novel approach to dynamic load modelling,” IEEE Transactions on Power Systems, vol. 4, no. 2,
pp. 396–402, May 1989.
[65] “Load modeling and dynamics,” Electra, no. 130, pp. 122–141, 1990.
[66] L. Angquist, B. Lundin, and J. Samuelsson, “Power oscillation damping using controlled reactive power compensation-a
comparison between series and shunt approaches,” IEEE Transactions on Power Systems, vol. 8, no. 2, pp. 687–700,
May 1993.
[67] J. V. Milanović, “On unreliability of exponential load models,” Electric Power Systems Research, vol. 49, no. 1, pp. 1–9,
Feb. 1999.
[68] Q. H. Wu and D. W. Shimmin, “Measurement based power system load modeling using a population diversity genetic
algorithm,” in POWERCON ’98. 1998 International Conference on Power System Technology. Proceedings (Cat.
No.98EX151), vol. 1, pp. 771–775.
[69] D. P. Stojanović, L. M. Korunović, and J. V. Milanović, “Dynamic load modelling based on measurements in medium
voltage distribution network,” Electric Power Systems Research, vol. 78, no. 2, pp. 228–238, Feb. 2008.
[70] S. M. Kay and S. L. Marple, “Spectrum analysis—A modern perspective,” Proceedings of the IEEE, vol. 69, no.
11, pp. 1380–1419, 1981.
[71] M. Zhou, V. A. Centeno, J. S. Thorp, and A. G. Phadke, “An Alternative for Including Phasor Measurements in State
Estimators,” IEEE Transactions on Power Systems, vol. 21, no. 4, pp. 1930–1937, Nov. 2006.
[72] B. Milosevic and M. Begovic, “Voltage-stability protection and control using a wide-area network of phasor
measurements,” IEEE Transactions on Power Systems, vol. 18, no. 1, pp. 121–127, Feb. 2003.
[73] J.-A. Jiang, “A novel adaptive PMU-based transmission-line relay-design and EMTP simulation results,” IEEE Transactions
on Power Delivery, vol. 17, no. 4, pp. 930–937, Oct. 2002.
[74] “C37.118-2005 - IEEE Standard for Synchrophasors for Power Systems,” 2006.
[75] C. t. Nguyen and K. Srinivasan, “A New Technique for Rapid Tracking of Frequency Deviations Based on Level
Crossings,” IEEE Transactions on Power Apparatus and Systems, vol. PAS-103, no. 8, pp. 2230–2236, Aug. 1984.
[76] P. K. Dash, A. K. Pradhan, and G. Panda, “Frequency estimation of distorted power system signals using extended
complex Kalman filter,” IEEE Transactions on Power Delivery, vol. 14, no. 3, pp. 761–766, Jul. 1999.
[77] M. Mojiri and A. R. Bakhshai, “An Adaptive Notch Filter for Frequency Estimation of a Periodic Signal,” IEEE
Transactions on Automatic Control, vol. 49, no. 2, pp. 314–318, Feb. 2004.
[78] A. T. Muoz and J. A. de la O Serna, “Shanks’ Method for Dynamic Phasor Estimation,” IEEE Transactions on
Instrumentation and Measurement, vol. 57, no. 4, pp. 813–819, Apr. 2008.
[79] K. Yamashita, H. Kobayashi, and Y. Kitauchi, “Identification of Individual Load Self-disconnection Following a Voltage
Sag,” in Asia Oceania Regional Council of CIGRE Technical Meeting, 2011.
Page 90
Modelling and Aggregation of Loads in Flexible Power Networks
[80] P. Pourbeik and B. Agrawal, “A hybrid model for representing air-conditioner compressor motor behaviour in power
system studies,” in 2008 IEEE Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in
the 21st Century, 2008, pp. 1–8.
[81] K. Yamashita and O. Sakamoto, “A study on dynamic behaviour of load supply system including synchronous
generators with and without load drop,” in IEEE PES General Meeting, 2010, pp. 1–7.
[82] W. W. Price, K. A. Wirgau, A. Murdoch, J. V. Mitsche, E. Vaahedi, and M. El-Kady, “Load modeling for power flow
and transient stability computer studies,” IEEE Transactions on Power Systems, vol. 3, no. 1, pp. 180–187, 1988.
[83] K. Morison and H. Hamadani, “Practical issues in load modeling for voltage stability studies,” in 2003 IEEE Power
Engineering Society General Meeting (IEEE Cat. No.03CH37491), pp. 1392–1397.
[84] F. Nozari, M. D. Kankam, and W. W. Price, “Aggregation of Induction Motors for Transient Stability Load Modeling,”
IEEE Transactions on Power Systems, vol. 2, no. 4, pp. 1096–1103, 1987.
[85] A. Capasso, W. Grattieri, R. Lamedica, and A. Prudenzi, “A bottom-up approach to residential load modeling,” IEEE
Transactions on Power Systems, vol. 9, no. 2, pp. 957–964, May 1994.
[86] WECC Load Modelling Task Force, “Load Composition Data,” [online] Available at:
http://en.sourceforge.jp/projects/sfnet_lcm-data/downloads/loadcomposition-excel-1_7_8.zip/.
[87] G. J. Berg, “Power-system load representation,” Proceedings of the Institution of Electrical Engineers, vol. 120, no. 3, p.
344, 1973.
[88] F. Iliceto and A. Capasso, “Dynamic Equibalents of Asynchronous Motor Loads in System Stability Studies,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-93, no. 5, pp. 1650–1659, Sep. 1974.
[89] M. M. Abdel Hakim and G. J. Berg, “Dynamic single-unit representation of induction motor groups,” IEEE Transactions on
Power Apparatus and Systems, vol. 95, no. 1, pp. 155–165, Jan. 1976.
[90] G. Rogers, J. Di Manno, and R. T. H. Alden, “An Aggregate Induction Motor Model for Industrial Plants,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-103, no. 4, pp. 683–690, Apr. 1984.
[91] M. Taleb, M. Akbaba, and E. A. Abdullah, “Aggregation of induction machines for power system dynamic studies,” IEEE
Transactions on Power Systems, vol. 9, no. 4, pp. 2042–2048, 1994.
[92] K. W. Louie, J. R. Marti, and H. W. Dommel, “Aggregation of Induction Motors in a Power System Based on Some
Special Operating Conditions,” in 2007 Canadian Conference on Electrical and Computer Engineering, 2007, pp. 1429–
1432.
[93] R. Aresi, B. Delfino, G. B. Denegri, M. Invernizzi, and S. Massucco, “Dynamic equivalents of tap-changing-under-load
transformers in a transmission system,” IEE Proceedings - Generation, Transmission and Distribution, vol. 146, no. 1, p. 45,
1999.
[94] P. Ju and X. Y. Zhou, “Dynamic equivalents of distribution systems for voltage stability studies,” IEE Proceedings -
Generation, Transmission and Distribution, vol. 148, no. 1, p. 49, 2001.
[95] T. Cutsem, Q. Renoy, and D. Lefebvre, “Modelling the short-term and long-term aggregate response of multiple loads
fed through a sub-transmission network,” in 2006 IEEE PES Power Systems Conference and Exposition, 2006, pp. 569–
575.
[96] Dept. for Energy and Climate Change, “Energy consumption in the UK 2010,” London, UK, 2010.
[97] P. Waide and S. Tanishima, “Light’s Labours Lost – Policies for energy efficient lighting,” Paris, France, 2006.
[98] C. Norén and J. Pyrko, “Typical load shapes for Swedish schools and hotels,” Energy and Buildings, vol. 28, no. 2, pp.
145–157, Oct. 1998.
[99] C. Norén, “Typical load shapes for six categories of Swedish commercial buildings,” 2007.
[100] J. P. Zimmermann, “End-use metering campaign in 400 households In Sweden Assessment of the Potential Electricity
Savings,” in Enertech, 2009.
[101] H. Akbari, J. Eto, S. Konopacki, A. Afzal, K. Heinemeier, and L. Rainer, “Integrated Estimation of Commercial Sector
End.Use Load Shapes and Energy Use Intensifies in the PG&E Service Area,” 1993.
[102] C. N. Jardine, “Synthesis of High Resolution Domestic Electricity Load Profiles,” in First International Conference and
Workshop on Micro-Cogeneration Technologies and Applications, 2008.
[103] R. Yao and K. Steemers, “A method of formulating energy load profile for domestic buildings in the UK,” Energy and
Buildings, vol. 37, no. 6, pp. 663–671, Jun. 2005.
[104] Building Research Establishment, “The impact of changing energy use patterns in buildings on peak electricity demand
in the UK,” 2008.
[105] National Statistics and DECC, “Energy Consumption in the UK, Service sector data Tables, 2009 update, Table 5.6,”
2009.
Page 91
Modelling and Aggregation of Loads in Flexible Power Networks
[106] F. Milano and K. Srivastava, “Dynamic REI equivalents for short circuit and transient stability analyses,” Electric Power
Systems Research, vol. 79, no. 6, pp. 878–887, Jun. 2009.
[107] J. M. Ramirez, B. V. Hernández, and R. E. Correa, “Dynamic equivalence by an optimal strategy,” Electric Power
Systems Research, vol. 84, no. 1, pp. 58–64, Mar. 2012.
[108] J. Undrill and A. Turner, “Construction of Power System lectromechanical Equivalents by Modal Analysis,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-90, no. 5, pp. 2049–2059, Sep. 1971.
[109] B. Marinescu, B. Mallem, and L. Rouco, “Large-Scale Power System Dynamic Equivalents Based on Standard and
Border Synchrony,” IEEE Transactions on Power Systems, vol. 25, no. 4, pp. 1873–1882, Nov. 2010.
[110] A. M. Miah, “Study of a coherency-based simple dynamic equivalent for transient stability assessment,” IET Generation,
Transmission & Distribution, vol. 5, no. 4, p. 405, 2011.
[111] R. Podmore and A. Germond, “Development of dynamic equivalents for transient stability studies,” 1977.
[112] E. J. S. Pires de Souza, “Identification of coherent generators considering the electrical proximity for drastic dynamic
equivalents,” Electric Power Systems Research, vol. 78, no. 7, pp. 1169–1174, Jul. 2008.
[113] J. Machowski, A. Cichy, F. Gubina, and P. Omahen, “External subsystem equivalent model for steady-state and
dynamic security assessment,” IEEE Transactions on Power Systems, vol. 3, no. 4, pp. 1456–1463, 1988.
[114] J. Machowski, J. W. Bialek, and J. R. Bumby, External Subsystem Equivalent Model for Steady-State and Dynamic Security
Assessment. Chichester, UK: John Wiley & Sons, Ltd, 1997. (same title as previous?)
[115] J. MACHOWSKI, “Dynamic equivalents for transient stability studies of electrical power systems,” International Journal
of Electrical Power & Energy Systems, vol. 7, no. 4, pp. 215–224, Oct. 1985.
[116] J. H. Chow, R. Galarza, P. Accari, and W. W. Price, “Inertial and slow coherency aggregation algorithms for power
system dynamic model reduction,” IEEE Transactions on Power Systems, vol. 10, no. 2, pp. 680–685, May 1995.
[117] G. N. Ramaswamy, L. Rouco, O. Fillatre, G. C. Verghese, P. Panciatici, B. C. Lesieutre, and D. Peltier, “Synchronic modal
equivalencing (SME) for structure-preserving dynamic equivalents,” IEEE Transactions on Power Systems, vol. 11, no. 1,
pp. 19–29, 1996.
[118] G. N. Ramaswamy, C. Evrard, G. C. Verghese, O. Fillatre, and B. C. Lesieutre, “Extensions, simplifications, and tests of
synchronic modal equivalencing (SME),” IEEE Transactions on Power Systems, vol. 12, no. 2, pp. 896–905, May 1997.
[119] R. J. Galarza, J. H. Chow, W. W. Price, A. W. Hargrave, and P. M. Hirsch, “Aggregation of exciter models for
constructing power system dynamic equivalents,” IEEE Transactions on Power Systems, vol. 13, no. 3, pp. 782–788,
1998.
[120] H. Shakouri G. and H. R. Radmanesh, “Identification of a continuous time nonlinear state space model for the external
power system dynamic equivalent by neural networks,” International Journal of Electrical Power & Energy Systems, vol.
31, no. 7–8, pp. 334–344, Sep. 2009.
[121] E. De Tuglie, L. Guida, F. Torelli, D. Lucarella, M. Pozzi, and G. Vimercati, “Identification of dynamic voltage-current
power system equivalents through artificial neural networks,” in Bulk Power System Dynamics and Control-VI, 2004.
[122] P. Sowa, A. M. Azmy, and I. Erlich, “Dynamic equivalents for calculation of power system restoration,” in APE ’04,
2004, pp. 7–9.
[123] A. H. M. A. Rahim and A. J. Al-Ramadhan, “Dynamic equivalent of external power system and its parameter estimation
through artificial neural networks,” International Journal of Electrical Power & Energy Systems, vol. 24, no. 2, pp. 113–
120, Feb. 2002.
[124] A. M. Stankovicc, A. T. Saric, and M. Milosevic, “Identification of nonparametric dynamic power system equivalents with
artificial neural networks,” IEEE Transactions on Power Systems, vol. 18, no. 4, pp. 1478–1486, Nov. 2003.
[125] A. M. Stankovic and A. T. Saric, “Transient Power System Analysis With Measurement-Based Gray Box and Hybrid
Dynamic Equivalents,” IEEE Transactions on Power Systems, vol. 19, no. 1, pp. 455–462, Feb. 2004.
[126] O.-C. Y. Lino, “Robust recurrent neural network-based dynamic equivalencing in power system,” in IEEE PES Power
Systems Conference and Exposition, 2004., pp. 677–686.
[127] J. M. Ramirez and V. H. Benitez, “Dynamic Equivalents by RHONN,” Electric Power Components and Systems, vol. 35, no.
4, pp. 377–391, May 2007.
[128] A. M. Miah, “Simple dynamic equivalent for fast online transient stability assessment,” IEE Proceedings - Generation,
Transmission and Distribution, vol. 145, no. 1, p. 49, 1998.
[129] M. Burth, G. C. Verghese, and M. Velez-Reyes, “Subset selection for improved parameter estimation in on-line
identification of a synchronous generator,” IEEE Transactions on Power Systems, vol. 14, no. 1, pp. 218–225, 1999.
[130] I. A. Hiskens, “Nonlinear dynamic model evaluation from disturbance measurements,” IEEE Transactions on Power
Systems, vol. 16, no. 4, pp. 702–710, 2001.
Page 92
Modelling and Aggregation of Loads in Flexible Power Networks
[131] W. W. Price, A. W. Hargrave, B. J. Hurysz, J. H. Chow, and P. M. Hirsch, “Large-scale system testing of a power system
dynamic equivalencing program,” IEEE Transactions on Power Systems, vol. 13, no. 3, pp. 768–774, 1998.
[132] W. T. Brown and W. J. Cloues, “Combination Load-Flow and Stability Equivalent for Power System Representation on
A-C Network Analyzers,” Transactions of the American Institute of Electrical Engineers. Part III: Power Apparatus and
Systems, vol. 74, no. 3, pp. 782–787, Jan. 1955.
[133] A. Ishchenko, A. Jokic, J. M. A. Myrzik, and W. L. Kling, “Dynamic reduction of distribution networks with dispersed
generation,” in 2005 International Conference on Future Power Systems, 2005, p. 7 pp.–7.
[134] L. Ljung, System Identification: Theory for the user, Second. Prentice Hall, Inc., 1999.
[135] L. Ljung and J. Sjoberg, “A system identification perspective on neural nets,” in Neural Networks for Signal Processing II
Proceedings of the 1992 IEEE Workshop, pp. 423–435.
[136] J. Sjöberg, Q. Zhang, L. Ljung, A. Benveniste, B. Delyon, P.-Y. Glorennec, H. Hjalmarsson, and A. Juditsky, “Nonlinear
black-box modeling in system identification: a unified overview,” Automatica, vol. 31, no. 12, pp. 1691–1724, Dec.
1995.
[137] I. Erlich, “Analysis and simulation of dynamic behaviour of power systems.” 1995.
[138] A. M. Azmy and I. Erlich, “Identification of dynamic equivalents for distribu-tion power networks using recurrent ANNs,”
in IEEE PES Power Systems Conference and Exposition, 2004., pp. 721–726.
[139] A. M. Azmy, I. Erlich, and P. Sowa, “Artificial neural network-based dynamic equivalents for distribution systems
containing active sources,” IEE Proceedings - Generation, Transmission and Distribution, vol. 151, no. 6, p. 681, 2004.
[140] A. M. Amzy, “Simulation and Management of Distributed Generating Units using Intelligent Techniques,” University of
Duinsburg-Essen, 2005.
[141] S. M. Zali and J. V. Milanovic, “Modelling of distribution network cell based on grey-box approach,” in 7th
Mediterranean Conference and Exhibition on Power Generation, Transmission, Distribution and Energy Conversion
(MedPower 2010), 2010, pp. 106–106.
[142] S. M. Zali and J. V. Milanovic, “Validation of developed grey-box model of distribution network cell,” in 7th
Mediterranean Conference and Exhibition on Power Generation, Transmission, Distribution and Energy Conversion
(MedPower 2010), 2010, pp. 107–107.
[143] S. M. Zali, N. C. Woolley, and J. V. Milanovic, “Development of equivalent dynamic model of distribution network using
clustering procedure,” in 17th Power Systems Computation Conference, 2011.
[144] M. G. Lauby, “Development of composite load models of power systems using on-line measurement data,” in 2006 IEEE
Power Engineering Society General Meeting, 2006, p. 8 pp.
[145] P. Ju, F. Wu, Z.-Y. Shao, X.-P. Zhang, H.-J. Fu, P.-F. Zhang, N.-Q. He, and J.-D. Han, “Composite load models based on
field measurements and their applications in dynamic analysis,” IET Generation, Transmission & Distribution, vol. 1, no. 5,
p. 724, 2007.
[146] F. Wu, X.-P. Zhang, and P. Ju, “Impact of wind turbines on power system stability,” in 2007 iREP Symposium - Bulk
Power System Dynamics and Control - VII. Revitalizing Operational Reliability, 2007, pp. 1–7.
[147] F. Wu, X.-P. Zhang, and P. Ju, “Modeling and control of the wind turbine with the Direct Drive Permanent Magnet
Generator integrated to power grid,” in 2008 Third International Conference on Electric Utility Deregulation and
Restructuring and Power Technologies, 2008, pp. 57–60.
[148] X. Yu, Z. Jiang, and A. Abbasi, “Dynamic modeling and control design of microturbine distributed generation systems,”
in 2009 IEEE International Electric Machines and Drives Conference, 2009, pp. 1239–1243.
[149] H. R. Baghaee, M. Mirsalim, M. J. Sanjari, and G. B. Gharehpetian, “Effect of type and interconnection of DG units in
the fault current level of distribution networks,” in 2008 13th International Power Electronics and Motion Control
Conference, 2008, pp. 313–319.
[150] M. Dehghani and S. K. Y. Nikravesh, “Nonlinear state space model identification of synchronous generators,” Electric
Power Systems Research, vol. 78, no. 5, pp. 926–940, May 2008.
[151] W. Sun and Y. -x. Yuan, Optimization Theory and Methods: Nonlinear Programming. New York: Springer Inc., 2006.
[152] L. Ljung, System Identification ToolboxTM 7 User’s Guide. The Mathworks, Inc, 2010.
[153] F. O. Resende, C. L. Moreira, and J. A. Peças Lopes, “Identification of Dynamic Equivalents for MicroGrids with High
Penetration of Solar Energy using ANNs,” in 3rd. European Conference of PV-Hybrid and Mini-Grid, 2006.
[154] F. de O. Resende, “Contributions for Microgrids Dynamic Modelling and Operation,” PhD thesis, Porto, 2007.
[155] F. O. Resende and J. A. P. Lopes, “Development of Dynamic Equivalents for MicroGrids using System Identification
Theory,” in 2007 IEEE Lausanne Power Tech, 2007, pp. 1033–1038.
Page 93
Modelling and Aggregation of Loads in Flexible Power Networks
[156] O. Nelles, Nonlinear System Identification. From Classical Approaches to Neural Networks and Fuzzy Models. New York:
Springer Inc., 2001.
[157] H. Demuth and M. Beale, Neural Network Toolbox: User’s Guide, Version 3.0. The Mathworks, Inc, 1998.
[158] H. Akagi, Y. Kanazawa, and A. Nabae, “Instantaneous Reactive Power Compensators Comprising Switching Devices
without Energy Storage Components,” IEEE Transactions on Industry Applications, vol. IA-20, no. 3, pp. 625–630, May
1984.
[159] G. Amico, V. Sinagra, and G. Lo Giudice, “Tools for Analysing Voltage Collapses,” in XIII ERIAC-Cigre Conference,
2009.
[160] S. J. Ranade, A. Ellis, and J. Mechenbier, “The development of power system load models from measurements,” in 2001
IEEE/PES Transmission and Distribution Conference and Exposition. Developing New Perspectives (Cat. No.01CH37294),
vol. 1, pp. 201–206.
[161] IEEE Task Force on Load Representation for Dynamic Performance, “Standard load models for power flow and
dynamic performance simulation,” IEEE Transactions on Power Systems, vol. 10, no. 3, pp. 1302–1313, 1995.
[162] PSS/E®, “Program Application Guide,” 2010.
[163] DIgSilent, “Technical Documentation: General Load Model,” 2008.
[164] G. Energy, “www.ge-energy.com/products_and_services/products /concorda_software_suite/.” .
[165] W. Zhongxi and Z. Xiaoxin, “Power System Analysis Software Package (PSASP)-an integrated power system analysis
tool,” in POWERCON ’98. 1998 International Conference on Power System Technology. Proceedings (Cat. No.98EX151),
vol. 1, pp. 7–11.
[166] Manitoba HVDC Research Centre, “https://pscad.com/products/pscad.” .
[167] L. Korunović and D. Stojanović, “Load Model Parameters on Low and Middle Voltage in Distribution Networks,”
Elektroprivreda, vol. 55, no. 2, pp. 46–56, 2002.
[168] L. M. Korunovic, D. P. Stojanovic, and J. V. Milanovic, “Identification of static load characteristics based on
measurements in medium-voltage distribution network,” IET Generation, Transmission & Distribution, vol. 2, no. 2, p. 227,
2008.
[169] J. Ribeiro and F. Lange, “A New Aggregation Method for Determining Composite Load Characteristics,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-101, no. 8, pp. 2869–2875, Aug. 1982.
[170] L. M. Korunović and D. P. Stojanović, “Dynamic Load Modelling of Some Low Voltage Devices”, Facta Universitatis
(Niš), Series: Electronics and Energetics, Vol. 22, No. 1, April 2009, pp. 61-70.
[171] S. A. Y. Sabir and D. c. Lee, “Dynamic Load Models Derived from Date Acquired During System Transients,” IEEE
Transactions on Power Apparatus and Systems, vol. PAS-101, no. 9, pp. 3365–3372, Sep. 1982.
[172] J. Ping and M. Daqiang, The Load Modeling of the Power System, Second. Beijing, China: China Water & Power Press,
2008.
[173] W.-S. Kao, C.-T. Huang, and C.-Y. Chiou, “Dynamic load modeling in Taipower system stability studies,” IEEE
Transactions on Power Systems, vol. 10, no. 2, pp. 907–914, May 1995.
[174] W. W. Price, “PSLF Load Model Update,” in GE presentation at WECC Modeling & Validation WG Meeting, 2008.
[175] Y. A. Jebril, I. Ibrahim, S. A. Shaban, S. A. Al Dessi, K. Karoui, F. Depieereux, A. Szekut, and F. Villella, “Development of
a detailed dynamic load model and implementation staged testing of DEWA network,” in GCC Cigre.
[176] UK Energy Research Council, “Energy Data Centre.”
[177] Scottish Power, “Private communication measurements.” 2011.
[178] R. Stamminger, “Synergy Potential of Smart Appliances.” 2008.
[179] R. E. Brown and J. G. Koomey, “Electricity Use in California: Past Trends and Present Usage Patterns,” Energy Policy,
vol. 31, no. 9, pp. 849–864, 2003.
[180] J. Gershuny, D. Lader, and S. Short, “The Time Use Survey, 2005: How we spend our time.”
[181] Dept. for Environment Food and Rural Affairs, “BNDL01: Domestic Lighting Government Standards Evidence Base
2009: Key Inputs,” London, UK.
[182] IEC-61000-3-12, “[13] Electromagnetic compatibility (EMC), Part 3-12: Limits for Harmonic Currents Produced by
Equipment Connected to Public Low-voltage Systems with Input Current Greater than 16 A and Less than 75 A Per-
phase,” 2004.
[183] EUROPA European Union, “Member States approve the phasing out of incandescent bulbs by 2012,” 2009.
Page 94
Modelling and Aggregation of Loads in Flexible Power Networks
[184] Public Law 110-140, “H.R. 6 Energy Independence and Security Act of 2007.” [online] Available at:
http://www.gpo.gov/fdsys/pkg/PLAW-110publ140/pdf/PLAW-110publ140.pdf.
[185] L. N. F. De Villiers, “DEFICIENCIES OF PRACTICES ESKOM CURRENTLY USES FOR SETTING OUT-OF-STEP RELAYS.” 26-
Oct-2006.
[186] J. P. Zimmermann, “End-use metering campaign in 400 households in Sweden assessment of the potential electricity
savings,” Ingenierie Energetique et Fluides, 2009.
[187] ADEME, CCE, and CRES, “End-use metering campaign in 400 households of the European Community-assessment of the
potential electricity savings,” 2002.
[188] L. Dittmar, “Intergration of energy saving technologies and learning curves into the basic model of regional electricity
liberalisation,” University of Flensburg, Germany, 2006.
[189] R. Stamminger and R. Friedrich-Wilhelms, “Synergy Potential of Smart Appliances,” Bonn, Germany, 2008.
[190] C. Chao-Shun, W. Tsung-Hsien, L. Chung-Chieh, and T. Yenn-Minn, “The application of load models of electric
appliances to distribution system analysis,” IEEE Transactions on Power Systems, vol. 10, no. 3, pp. 1376–1382, 1995.
[191] J. C. Boemer, M. Gubescu, and W. L. Kling, “Dynamic Models for Transient Stability Anlysis of Transmission and
Distribution Systems with Distributed Generation: an Overview,” in 2011 IEEE Trondheim PowerTech, 2009.
[192] CIGRE WG C4.601, “Modelling and Dynamic Behaviour of Wind Generation as it Relates to Power System Control
and Dynamic Performance,” 2007.
[193] M. Pöller and S. Achilles, “Aggregated wind park models for analyzing power system dynamics,” in International
Workshop on Large Scale Integration of Wind Power and Transmission Networks for Offshore Windfarms, 2003.
[194] J. Usaola, P. Ledesma, J. M. Rodriguez, J. L. Fernandez, D. Beato, R. Iturbe, and J. R. Wilhelmi, “Transient stability
studies in grids with great wind power penetration. Modelling issues and operation requirements,” in IEEE PES General
Meeting, 2003, pp. 1534–1541.
[195] L. M. Fernández, C. A. García, J. R. Saenz, and F. Jurado, “Equivalent models of wind farms by using aggregated wind
turbines and equivalent winds,” Energy Conversion and Management, vol. 50, no. 3, pp. 691–704, Mar. 2009.
[196] V. Akhmatov and H. Knudsen, “An aggregate model of a grid-connected, large-scale, offshore wind farm for power
stability investigations—importance of windmill mechanical system,” International Journal of Electrical Power & Energy
Systems, vol. 24, no. 9, pp. 709–717, Nov. 2002.
[197] V. Akhmatov, “An aggregated model of a large wind farm with variable-speed wind turbines equipped with doubly-
fed induction generators,” Wind Engineering, vol. 28, no. 4, pp. 479–486, 2004.
[198] M. V. A. Nunes, J. A. PecasLopes, H. H. Zurn, U. H. Bezerra, and R. G. Almeida, “Influence of the Variable-Speed Wind
Generators in Transient Stability Margin of the Conventional Generators Integrated in Electrical Grids,” IEEE
Transactions on Energy Conversion, vol. 19, no. 4, pp. 692–701, Dec. 2004.
[199] J. O. G. Tande, “Grid Integration of Wind Farms,” Wind Energy, vol. 6, no. 3, pp. 281–295, Jul. 2003.
[200] M. Ali, I.-S. Ilie, J. V Milanovic, and G. Chicco, “Probabilistic clustering of wind generators,” in IEEE PES General Meeting,
2010, pp. 1–6.
[201] L. M. Fernández, F. Jurado, and J. R. Saenz, “Aggregated dynamic model for wind farms with doubly fed induction
generator wind turbines,” Renewable Energy, vol. 33, no. 1, pp. 129–140, Jan. 2008.
[202] L. M. Fernández, J. R. Saenz, and F. Jurado, “Dynamic models of wind farms with fixed speed wind turbines,”
Renewable Energy, vol. 31, no. 8, pp. 1203–1230, Jul. 2006.
[203] J. Conroy and R. Watson, “Aggregate modelling of wind farms containing full-converter wind turbine generators with
permanent magnet synchronous machines: transient stability studies,” IET Renewable Power Generation, vol. 3, no. 1, p.
39, 2009.
[204] F. Rodríguez-Bobada, P. Ledesma, S. Martínez, L. Coronado, and E. Prieto, “Simplified wind generator model for
Transmission System Operator planning studies,” in International Workshop on Large Scale Integration of Wind Power
and on Transmission Networks for Offshore Wind Farms, 2005.
[205] J. L. Acosta, K. Combe, S. Z. Djokic, and I. Hernando-Gil, “Performance Assessment of Micro and Small-Scale Wind
Turbines in Urban Areas,” IEEE Systems Journal, vol. 6, no. 1, pp. 152–163, Mar. 2012.
[206] J. L. Acosta, K. Combe, S. Z. Djokic, and I. Hernando-Gil, “Performance Assessment of Micro and Small-Scale Wind
Turbines in Urban Areas,” IEEE Systems Journal, vol. 6, no. 1, pp. 152–163, Mar. 2012.
[207] J. L. Acosta, A. J. Collin, B. Hayes, and S. Z. Djokic, “Micro and small-scale wind generation in urban distribution
networks,” in 7th Mediterranean Conference and Exhibition on Power Generation, Transmission, Distribution and Energy
Conversion (MedPower 2010), 2010, pp. 160–160.
Page 95
Modelling and Aggregation of Loads in Flexible Power Networks
[208] D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo, “A compensation-based power flow method for weakly
meshed distribution and transmission networks,” IEEE Transactions on Power Systems, vol. 3, no. 2, pp. 753–762, May
1988.
[209] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution systems for loss reduction and load balancing,” IEEE
Transactions on Power Delivery, vol. 4, no. 2, pp. 1401–1407, Apr. 1989.
[210] R. P. Broadwater, A. Chandrasekaran, C. T. Huddleston, and A. H. Khan, “Power flow analysis of unbalanced
multiphase radial distribution systems,” Electric Power Systems Research, vol. 14, no. 1, pp. 23–33, Feb. 1988.
[211] M. H. Haque, “A general load flow method for distribution systems,” Electric Power Systems Research, vol. 54, no. 1, pp.
47–54, Apr. 2000.
[212] D. Rajičić and R. Taleski, “Two novel methods for radial and weakly meshed network analysis,” Electric Power Systems
Research, vol. 48, no. 2, pp. 79–87, Dec. 1998.
[213] A. Abur, “REI-Equivalent Based Decomposition Method for Multi-Area TTC Computation,” in 2005/2006 PES TD, pp.
506–510.
[214] A. C. Neto, A. B. Rodrigues, R. B. Prada, and M. da Guia da Silva, “External Equivalent for Electric Power Distribution
Networks With Radial Topology,” IEEE Transactions on Power Systems, vol. 23, no. 3, pp. 889–895, Aug. 2008.
[215] A. Ishchenko, A. Jokic, J. M. A. Myrzik, and W. L. Kling, “Dynamic reduction of distribution networks with dispersed
generation,” in 2005 International Conference on Future Power Systems, 2005, p. 7 pp.–7.
[216] A. Ishchenko, J. M. A. Myrzik, and W. L. Kling, “Dynamic equivalencing of distribution networks with dispersed
generation using Hankel norm approximation,” IET Generation, Transmission & Distribution, vol. 1, no. 5, p. 818, 2007.
[217] A. Ishchenko, J. M. A. Myrzik, and W. L. Kling, “Dynamic Equivalencing of Distribution Networks with Dispersed
Generation Using Krylov Methods,” in 2007 IEEE Lausanne Power Tech, 2007, pp. 2023–2028.
[218] D. Chaniotis and M. A. Pai, “Model Reduction in Power Systems Using Krylov Subspace Methods,” IEEE Transactions on
Power Systems, vol. 20, no. 2, pp. 888–894, May 2005.
[219] J.-Y. Kim, D.-J. Won, and S.-I. Moon, “Development of the dynamic equivalent model for large power system,” in 2001
Power Engineering Society Summer Meeting. Conference Proceedings (Cat. No.01CH37262), vol. 2, pp. 973–977.
[220] X. Feng, Z. Lubosny, and J. Bialek, “Dynamic Equivalencing of Distribution Network with High Penetration of Distributed
Generation,” in Proceedings of the 41st International Universities Power Engineering Conference, 2006, pp. 467–471.
[221] X. Feng, Z. Lubosny, and J. W. Bialek, “Identification based Dynamic Equivalencing,” in 2007 IEEE Lausanne Power Tech,
2007, pp. 267–272.
[222] S. M. Zali and J. V. Milanovic, “Dynamic equivalent model of Distribution Network Cell using Prony analysis and
Nonlinear least square optimization,” in 2009 IEEE Bucharest PowerTech, 2009, pp. 1–6.
[223] G. Diaz, C. Gonzalez-Moran, J. Gomez-Aleixandre, and A. Diez, “Composite Loads in Stand-Alone Inverter-Based
Microgrids—Modeling Procedure and Effects on Load Margin,” IEEE Transactions on Power Systems, vol. 25, no. 2, pp.
894–905, May 2010.
[224] European SmartGrids Technology Platform, “Vision and Strategy for Europe’s Electricity Networks of the Future.”
[225] P1547.4/D12, “IEEE Draft Guide for Design, Operation and Integration of Distributed Resource Island Systems with
Electric Power Systems,” 2011.
[226] “European research project MICROGRIDS.”
[227] “European research project MORE MICROGRIDS.”
[228] M. Barnes, J. Kondoh, H. Asano, J. Oyarzabal, G. Ventakaramanan, R. Lasseter, N. Hatziargyriou, and T. Green, “Real-
World MicroGrids-An Overview,” in 2007 IEEE International Conference on System of Systems Engineering, 2007, pp.
1–8.
[229] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, “Microgrids,” IEEE Power and Energy Magazine, vol. 5, no. 4,
pp. 78–94, Jul. 2007.
[230] “Consortium for Electric Reliability Technology Solutions.”
[231] J. A. Pecas Lopes, C. L. Moreira, and A. G. Madureira, “Defining Control Strategies for MicroGrids Islanded
Operation,” IEEE Transactions on Power Systems, vol. 21, no. 2, pp. 916–924, May 2006.
[232] J. A. Peças Lopes, C. L. Moreira, and F. O. Resende, “Control strategies for microgrids black start and islanded
operation,” International Journal of Distributed Energy Resources, vol. 1, pp. 242–261, 2005.
[233] C. L. Moreira, F. O. Resende, and J. A. P. Lopes, “Using Low Voltage MicroGrids for Service Restoration,” IEEE
Transactions on Power Systems, vol. 22, no. 1, pp. 395–403, Feb. 2007.
Page 96
Modelling and Aggregation of Loads in Flexible Power Networks
[234] S. Barsali, M. Ceraolo, P. Pelacchi, and D. Poli, “Control techniques of Dispersed Generators to improve the continuity
of electricity supply,” in 2002 IEEE Power Engineering Society Winter Meeting. Conference Proceedings (Cat.
No.02CH37309), vol. 2, pp. 789–794.
[235] R. Caldon, F. Rossetto, and R. Turri, “Analysis of dynamic performance of dispersed generation connected through
inverter to distribution networks,” 17th International Conference on Electricity Distribution - CIRED, Barcelona, 2003.
[236] M. Y. El-Sharkh, A. Rahman, M. S. Alam, A. A. Sakla, P. C. Byrne, and T. Thomas, “Analysis of Active and Reactive
Power Control of a Stand-Alone PEM Fuel Cell Power Plant,” IEEE Transactions on Power Systems, vol. 19, no. 4, pp.
2022–2028, Nov. 2004.
[237] F. Katiraei, M. R. Iravani, and P. W. Lehn, “Micro-Grid Autonomous Operation During and Subsequent to Islanding
Process,” IEEE Transactions on Power Delivery, vol. 20, no. 1, pp. 248–257, Jan. 2005.
[238] R. Lasseter and P. Piagi, “Providing premium power through distributed resources,” in Proceedings of the 33rd Annual
Hawaii International Conference on System Sciences, vol. vol.1, p. 9.
[239] J. A. Peças Lopes, S. A. Polenz, C. L. Moreira, and R. Cherkaoui, “Identification of control and management strategies
for LV unbalanced microgrids with plugged-in electric vehicles,” Electric Power Systems Research, vol. 80, no. 8, pp.
898–906, Aug. 2010.
[240] P. G. Barbosa, I. Misaka, and E. H. Watanabe, “Advanced var compensators using PWM - voltage source inverters,” in
2o Congresso Brasileiro de Electrónica de Potência - COBEP/93, 1993.
[241] P. G. Barbosa, L. G. B. Rolim, E. H. Watanabe, and R. Hanitsch, “Control strategy for grid-connected DC-AC converters
with load power factor correction,” IEE Proceedings - Generation, Transmission and Distribution, vol. 145, no. 5, p. 487,
1998.
[242] T. Shimada, S. Agematsu, T. Shoji, T. Funabashi, H. Otoguro, and A. Ametani, “Combining power system load models at
a busbar,” in 2000 Power Engineering Society Summer Meeting (Cat. No.00CH37134), vol. 1, pp. 383–388.
[243] P. Pillay, S. M. A. Sabur, and M. M. Haq, “A model for induction motor aggregation for power system studies,” Electric
Power Systems Research, vol. 42, no. 3, pp. 225–228, Sep. 1997.
Page 97
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 2-A-1: Recording of Actual Malin 500-kV voltage. Figure 2-A-2: Simulated Malin 500-kV voltage using WSCC
data base.
Page 98
Modelling and Aggregation of Loads in Flexible Power Networks
Table 2-A-1: The Effect of Load Models on the Critical Mode Eigenvalues During the Unstable Low Frequency Oscillation Event
Page 99
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 2-A-5: Bus voltage angle and amplitude of the simulated results and the measured data at LY 500kV bus
Page 100
Modelling and Aggregation of Loads in Flexible Power Networks
The “Optimized” parameters in Figure 2-B-1 are those which are optimized during the non-linear optimization
process. For these parameters, the user has to provide an initial estimate. The “Derived” parameters are those
which are calculated either based on user input or during the optimization process, but are not optimized.
“Input Data” is the voltage and current at the secondary of the substation transformer where monitors are
located. The “Derived Data” node is the bus where the actual load response is modelled. The complete set of
input and output parameters required for the single machine structure is shown in Figure 2-B-2.
The salient features of this model are:
INITIAL ESTIMATES AND BOUNDS
The optimization process is sensitive to the initial estimates and the bounds on the parameters that are to be
optimized. In any optimization process, multiple solutions exist and the answer depends on the initial estimate of
the parameters. In the last phase of the project [16], a literature review was performed to come up with
different sets of machine and static parameters that have been used in previous load modelling efforts. These
parameters can be used as initial estimates. Different initial estimates available in Load Model Parameter
Derivation (LMPD) for both the structures are given in [16]. In addition, the user can change lower and upper
bounds on the parameters as well.
The detailed equations for calculating Qmotor , Qstatic , and Q feeder are given in [16]. Sign of Qcapacitor follows
load convention. That is, a. negative value indicates vars being supplied to the system and a positive value
Page 101
Modelling and Aggregation of Loads in Flexible Power Networks
indicates vars being absorbed (in essence an inductor). In practice, Qcapacitor should always come out to be
negative.
Optimized Parameters
Calculated Parameters
User Input
Voltage drop
Feeder X/R Rfeeder Distribution Feeder Resistance
Machine power factor
Static load power factor Xfeeder Distribution Feeder Reactance
Initial guesses
Bounds of parameters Kq % of Dynamic Machine Reactive Power
Figure 2-B-2: 1-Input and Output for the LMPD Algorithm – 1-Machine Structure
Note that Qcapacitor includes the net effect of substation capacitor banks as well as capacitor banks along the
feeder. Depending on the time of the day, capacitor banks would be in or out of service. In some cases, utility
engineers might have accurate information of the status of the banks at the time of the event. In that case,
Qcapacitor may be a constant value supplied by the user. If Qcapacitor is entered by the user, the match for reactive
power will be sensitive to the accuracy of Qcapacitor . Therefore it is recommended that user should not try to
force this value unless he or she is confident. The present version of the LMPD algorithm does not allow the user
to enter a fixed value of Qcapacitor .
Page 102
Modelling and Aggregation of Loads in Flexible Power Networks
Percentage of dynamic reactive power is not an optimized parameter. It is calculated based on the real power
consumed by the motor and static loads, and load power factors as shown in the following equations.
Qmotor
Kq (2-B.2)
Qstatic Qmotor
Pmotor 2 (1 pf _ motor 2 )
Qmotor (2-B.3)
pf _ motor 2
Pstatic 2 (1 pf _ static 2 )
Qstatic (2-B.4)
pf _ static 2
Pmotor is the induction motor real power at the load bus while Pstatic is the real power consumed by static load.
T T0 ( A 2 B C ) (2-B.5)
Where, T0 is the steady state torque. From the results obtained so far, there is no advantage in using
polynomial torque representation for the mechanical load.
STATIC LOAD
The static load equations are given by:
P P0 (a pV 2 b pV c p ) (2-B.6)
Where ap(aq),bp(bq) and cp(cq) are constant impedance, constant current and constant power coefficients of real
(reactive) power, respectively. The coefficients are not percentages of each static load type. They are
algebraic coefficients and can take negative values as well. This is because in reality there are many loads
(compact fluorescent light, discharge lighting, electronic loads etc.) that cannot necessarily be categorized as an
ideal constant impedance or constant current or constant power load. Thus, representing loads as a combination
of constant impedance (Z), constant current (I) and constant power (P) (ZIP model), loses some physical
significance and the exponential model may be just as applicable in these cases. The default bounds on the
static coefficients are 0 and 1. However, in some cases a lower bound of -1 on the static parameters can give
a better fit. Some literature suggests using upper bounds greater than 1 to allow more flexibility in the static
reactive model. Having a wide range of coefficients for the reactive model allows for non-linearity in the load
response [160]. Bounds greater than 1 were not used for the test cases. The default value of static load power
Page 103
Modelling and Aggregation of Loads in Flexible Power Networks
factor is 0.9. For some static loads the power factor can be as high as or higher than 0.95. However, because
we are trying to model the static load at the substation a value of 0.9 is more realistic to take into account
reactive losses as we go down in voltage (i.e. transformation that is not explicitly modelled).
The complete set of input and output parameters required for the two machine structure is shown in Figure 2-B-
4. The purpose of adding the second machine is to be able to separate out typically large (and heavier)
industrial motors loads from typically smaller (and much lighter inertia) residential motor loads such as
residential air conditioners. This structure can be used for summer peak events to explicitly represent residential
air conditioners.
Most of the parameters are already described in the previous section. Some of the features unique to the 2-
machine structure are described in this section.
Page 104
Modelling and Aggregation of Loads in Flexible Power Networks
Care should be taken in setting the bounds for Kp for the two machines. The default bounds on Kpl are set to
0.2 and 0.8 (same as the 1-machine structure). However, these bounds should be changed as needed.
Optimized Parameters
Small Motor
kp % of Dynamic Machine
Real Power
rs Stator Resistance
rr Rotor Resistance
xm Magnetizing Reactance
Calculated Parameters
Figure 2-B-4: Input and Output for LMPD Algorithm – 2-motor Structure
Kps has bounds of 0.1 and 0.3. The upper bound of 0.3 is probably the upper limit of the percentage of
residential air conditioners on a hot summer day. However, these bounds should be changed as needed based
on the season and time of the day, and any survey data that is available to give a better understanding of the
actual potential range of values for a given utility.
Page 105
Modelling and Aggregation of Loads in Flexible Power Networks
Page 106
Modelling and Aggregation of Loads in Flexible Power Networks
0.8
C D1 C D1
0.6 0.6
D2 E D2 E
0.4 0.4
F G F G
0.2 0.2
H I1 H I1
0 I2 J 0 I2 J
0 10 20 30 40 0 10 20 30 40
Induction motor ratio (%) D3 Induction motor ratio (%) D3
(a) Initial voltage during voltage sags (b) Minimum voltage during voltage sags
1 1.1
A B 1.08 A B
Voltage (d) (p.u.)
Voltage (c) (p.u.)
0.9
C D1 1.06 C D1
0.8 1.04
0.7 D2 E 1.02 D2 E
0.6 F G 1 F G
0.98
0.5 H I1 0.96 H I1
0.4 0.94 I2 J
I2 J
0 10 20 30 40 0 10 20 30 40
Induction motor ratio (%) D3 Induction motor ratio (%) D3
(c) Initial recovery voltage (d) Maximum voltage after voltage sags
Page 107
Modelling and Aggregation of Loads in Flexible Power Networks
1.1
1.08 A B
1.0
Voltage (p.u.)
0.8
Induction motor ratio: 0%
0.6 Induction motor ratio: 10%
0.4 Induction motor ratio: 20%
Induction motor ratio: 30%
0.2
Induction motor ratio: 40%
0.0
0 1 2 3 4 5
Time in second
Figure 2-D-3: Voltage response after voltage sags at point H
0.9
0.8 C D1
0.7 D2 E
0.6 F G
0.5 H I1
0.4 I2 J
0.05 0.1 0.15
Fault duration (s) D3
Figure 2-D-4: Evaluation items of dynamic voltage behaviour following voltage sags
Table 2-D-1 shows the mean values of the voltage change at all bus load voltages of points A to I for a 10%
change in IM ratio (from 40% to 30%, from 30% to 20%, from 20% to 10%, and from 10% to 0%). The table
also shows the mean values of the voltage change at all bus load voltages for a 50 ms change in fault duration
(from 150 ms to 100 ms, from 100 ms to 50 ms). As shown in Table 2-D-1, the IM ratio does not affect b), d),
and e), but does affect a) and c).
The trend in the change in dynamic voltage behaviour resulting from the change in load self-disconnection
amount and load voltage characteristics index is illustrated based on the sensitivity analysis, respectively in
Figure 2-D-6.
Page 108
Modelling and Aggregation of Loads in Flexible Power Networks
1.2
Voltage (p.u.)
0.8
Fault duration: 50ms
0.4 Fault duration: 100ms
Fault duration: 150ms
0.0
0 1 2 3 4 5
Time in second
Figure 2-D-5: Voltage response after voltage sags at point H
Table 2-D-1: Dynamic voltage behaviour resulting from change in induction motor ratio and fault duration
Page 109
Modelling and Aggregation of Loads in Flexible Power Networks
0.98
0.97
0.96
0.95
-0.060
-0.070
Q [pu]
-0.080
-0.090
-0.100
1.000
V [pu]
0.996
0.992
0.988
0.1
0
F [Hz]
-0.1
-0.2
-0.3
-0.4
-0.5
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time in second
(a) without the extraction method (b) with the extraction method
Figure 2-E-1: The example measured data and the model response
Figure 2-E-1 shows the measured data and the response of active and reactive power load model without the
developed extraction method for large frequency change. Figure 2-E-1(a) shows the measured data and the
response of active and reactive power load model without the developed extraction method using FIS. At 11
seconds, a significant load drop occurs caused by probably a feeder trip. When the measured data to be
identified the load model parameters includes this load change at 11 seconds, the mismatch between measured
data (blue solid line) and the response of the load models (red solid line) becomes large as shown in Figure 2-
E-1(a)). That results in the improper load model parameter derivation for this case and shows the necessity of
removing the measured data which include load change caused by the change in load structures.
Figure 2-E-1(b) shows the measured data and the response of active and reactive power load model with the
developed extraction method using FIS. It is recognized that the model response after applying the extraction
method coincide with the measurement data, while the model response without applying the method is not
Page 110
Modelling and Aggregation of Loads in Flexible Power Networks
identical with the measured data. Therefore, it can be concluded that the developed extraction method can
extract the measured data which are suitable for load model parameter derivation. Figure 2-E-2 shows the
measured data and the response of active and reactive power load model without the developed extraction
method for small frequency change. It also reveals that the developed extraction method can extract those
measured data which are suitable for load model parameter derivation.
The developed extraction method using FIS is expected to derive more appropriate load model parameters
more speedy. The development of the automated process enables to achieve unified and neutral judgment.
1.010
P [pu]
1.005 Measured
Simulated
1.000
-0.256
Q [pu]
-0.258
-0.260
-0.262
1.004
V [pu]
1.003
1.002
1.001
1.000
F [Hz]
0.04
0.02
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Time in second
(a) without the extraction method (b) with the extraction method
Figure 2-E-2: The example measured data and the model response with and without the developed extraction method for small frequency change
Page 111
Modelling and Aggregation of Loads in Flexible Power Networks
Page 112
Modelling and Aggregation of Loads in Flexible Power Networks
0 1 2
V V V
P P1 P2
V
P3
V
(2-F.2)
V0 0 0
where: np=0 → Constant P (Power); np=1 → Constant I (Current); np=2 → Constant Z (Impedance); P and P0:
Actual and initial real power of modelled load; V and V0: Actual and initial load bus voltage magnitudes; Pi,
i=1-3: corresponding shares of constant P, I or Z load, and P1 + P2 + P3 = P0.
The widely accepted practice for power flow analysis of electric networks is to assume that the distribution
system tap changing transformers and voltage regulators have brought bus voltages close to nominal values
(i.e. close to 1 p.u. value). In this case, loads may be treated as a constant real and reactive power demands,
and constant PQ load model can be used. Therefore, as expected, the responses to this question (Q1, Figure 2-
F-1) identified constant power PQ load model as by far the most dominant type of load model used in steady
state power system studies (84% of all responses).
Table 2-F-2: The participants and response rates of the survey
Africa 21 7 33.3 7
Americas 34 17 50.0 18
Asia 35 24 68.6 25
Europe 62 41 66.1 42
Oceania 8 8 100.0 8
Page 113
Modelling and Aggregation of Loads in Flexible Power Networks
Overall 84% 6% 8%
Figure 2-F-1: Load models used for steady state power system studies
Page 114
Modelling and Aggregation of Loads in Flexible Power Networks
Oceania 14% 43% 14% 29% Oceania 15% 15% 38% 31%
Asia 15% 32% 6% 21% 3% 18% 6% Asia 12% 41% 18% 6% 18% 6%
Americas 13% 13% 26% 4% 30% 13% Americas 13% 13% 26% 4% 30% 13%
Europe 29% 8% 6% 21% 10% 16% 8% Europe 31% 16% 21% 6% 16% 8%
Overall 23% 19% 4% 19% 7% 16% 10% Overall 23% 22% 19% 9% 17% 10%
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
Figure 2-F-3: Q3: Use of different load models for different Figure 2-F-4: Q4: Approaches for load model parameter
load classes in dynamic power system studies identification
It is interesting to note that in 19% of the cases (on average), the utilities rely solely on model parameters
available in the literature. This highlights the importance of publishing the results of load model parameter
identification studies based on measurements by utilities around the world. These results are likely to be used
by other utilities, e.g. those that might not have resources to conduct their own load modelling studies.
Page 115
Modelling and Aggregation of Loads in Flexible Power Networks
Page 116
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 2-F-7: Q7: Adequacy of available load models for Figure 2-F-8: Q8: Extent of use of user-defined load models
system stability studies
Page 117
Modelling and Aggregation of Loads in Flexible Power Networks
V
0
V
1
V
2
V
np
P P1 P2 P3 P0
V V V V
0 0 0 0
0 1 2 nq (2-F.3)
V V V V
Q Q1 Q2 Q3 Q0
V0 V0 V0 V0
P1 ´ 0 P2 ´ 1 P3 ´ 2 Q ´ 0 Q2 ´ 1 Q3 ´ 2
np , nq 1 (2-F.4)
P1 P2 P3 Q1 Q2 Q3
where: np and nq are active and reactive power coefficients of the equivalent exponential load model, and all
other quantities are defined in (2-F.1) and (2-F.2).
Page 118
Modelling and Aggregation of Loads in Flexible Power Networks
The conversion by (2-F.4) is based on Taylor expansion, which is more accurate if the voltage changes are
smaller. Table 2-F-3 shows an example of the conversion and equivalent np and nq coefficients for five network
operators and utilities from North and South America. (Note: Some respondents from Americas did not provide
any specific information for the selected ZIP model as the answer to Q2.) On the other hand, Table 2-F-4 shows
four examples of the coefficients of exponential load models used by one respondent from Europe. (Note: In
this case, the conversion is not performed, as this respondent indicated four exponential load models with the
corresponding coefficients as the actually used load models).
Table 2-F-3: Parameters of zip model and equivalent exponential model: for all respondents from North and South America
Table 2-F-4: Examples of Exponential Load Model Parameters Provided by One Respondent from Europe
Following the above conversion procedure, currently used static load models for dynamic stability studies,
shown in Figure 2-F-2, are converted into equivalent exponential load models shown in Figure 2-F-10.
The conversion of static load models used by the survey respondents in dynamic system stability studies into
equivalent exponential load models is further statistically analysed in Table 2-F-5 and Table 2-F-7, as well as
Page 119
Modelling and Aggregation of Loads in Flexible Power Networks
in Figure 2-F-11 and Figure 2-F-12. These results show minimum, maximum, median and mean values of
coefficients np and nq, their standard deviations and ranges for different continents and for the World as a
whole.
Constant P Constant I Constant P Constant Z
Constant Z ZIP Model ZIP Model Exponential Model
Exponential Model ZIP Model with IM ZIP Model with IM Exponential Model with IM
Exponential Model with IM Detailed Composite Model Detailed Composite Model
Before Before
23% 19% 4% 19% 7% 16% 10% 23% 22% 19% 9% 17% 10%
Conversion Conversion
After After
Static Load Model 72% 16% 10% Static Load Model 73% 17% 10%
Conversion Conversion
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
(a) real power load model (b) reactive power load model
Figure 2-F-10: Identified prevalent static load models used for dynamic power system studies, represented as equivalent exponential load model
It can be seen that in case of Asia, Americas, Africa and Europe, np varies from 0 to 2, while in Oceania it
ranges from 0 to 1.7. Mean values of np are less than 1 for all continents except for Oceania, where it is
slightly greater than 1. Consequently, mean value of np for the World is approximately 0.7, while its median
value is 1.0 (Figure 2-F-10(a)).
Standard deviations are of the same order as the corresponding mean values, indicating a large dispersion of
np coefficient for all continents and worldwide. Using these standard deviations and corresponding mean values,
somewhat narrower ranges of np are obtained according to the formulae:
np np , i.e. np np ; np (2-F.5)
where: np is mean value and is standard deviation. These ranges are shown in the last column in Table 2-F-5,
and the same approach is repeated for calculated equivalent nq coefficient.
Table 2-F-6: The analysis of values of NP coefficient without conversion
For comparison, Table 2-F-6 presents the results of statistical analysis of processing of np excluding those values
that are obtained by conversion. Most of the differences between Table 2-F-5 and Table 2-F-6 are small, but
generally greater standard deviations and therefore wider ranges are obtained in Table 2-F-6.
Page 120
Modelling and Aggregation of Loads in Flexible Power Networks
mean median
50
40
Frequency
30
20
10
0
-0.5 0 0.5 1 1.5 2 2.5
np
Figure 2-F-11: Histogram of np coefficients of equivalent exponential models for static load models used worldwide in dynamic power system studies
Table 2-F-7 shows that the values of nq are spread over wider range than the values of np. Mean values of nq
vary from 0.79 for Africa, to 2.25 for Oceania; while the mean worldwide value of nq is 1.3 (worldwide
median value is 1.8, Figure 2-F-12). Standard deviations of nq values are again of the same order as the
corresponding mean values, confirming a rather widely spread of nq values. Table 2-F-8 presents the results of
the analysis of nq values that are not obtained by conversion. As expected, the greater standard deviations
and therefore wider ranges are obtained in Table 2-F-8 in comparison with corresponding standard deviations
in Table 2-F-7. Exception is in the case of Oceania where none of nq values is obtained by conversion.
Page 121
Modelling and Aggregation of Loads in Flexible Power Networks
mean median
50
40
Frequency
30
20
10
0
-1 0 1 2 3 4 5
nq
Figure 2-F-12: Histogram of nq coefficients of equivalent exponential models for static load models used worldwide in dynamic power system studies
Since the dispersion of parameters np and nq used in dynamic studies is significant, both mean values and
median values can be treated as prevalent, i.e. typical. For example, prevalent value of np for Europe is
approximately 0.6, while for Americas and the World it is approximately 0.7/1.0 (Note: mean value/median
value). Prevalent or typical value of nq is almost 1/0.4 for Europe, but it is approximately 2.3/2.8 for Oceania
and 1.3/1.8 globally.
2-F.5 CONCLUSIONS
The survey indicated that there are large differences between the load models used for static and dynamic
power system studies by different system operators and utilities. Therefore, additional analysis of the received
responses is performed, including the calculation of equivalent exponential load model, in order to identify
prevalent (i.e. dominant or overall average) load models used by system operators and utilities in different
continents and worldwide.
The following main conclusions can be drawn from the results of the survey:
About 70% of utilities and system operators around the World use only static load model for power
system stability studies.
About 30% of utilities and system operators use some form of induction motor model to represent
dynamic loads in power system stability studies.
Dominant practice in the USA is to use a combination of static (typically ZIP) and dynamic load model
(typically induction motor), while use of static load models is prevalent in the rest of the World.
In about 40% of the cases, currently used load model parameters have been updated within the last
five years.
Typical static load model used in steady state studies is constant power (PQ) load for both real (active)
and reactive power, i.e., np = nq=0.
Typical values of np and nq coefficients of the equivalent exponential static load models used in
dynamic studies worldwide are 0.7 and 1.3, respectively.
Most of the utilities and system operators represent distributed generation as a negative load in system
studies, without modelling them explicitly. Some of them, however, recognize the importance of
appropriate DG models, particularly for future power system studies.
Page 122
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-2: Parameters of the model (3.5)–(3.6) for the load device/load class
Page 123
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-3: Parameters of the model (3.7)–(3.8) for the load device/load class
Table 3-A-4: Parameters of the model (3.11)-(3.12) for the load device/load class
Page 124
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-5: Load model and identified parameters for the load device/load class
U
2
U
Q Qn q1 q 2
q 3 1 k qf f
U n Un
Residential load [41] (2000)
U
P Pn a 0 a1 with capacitors OFF
U n a0=0.55, a1=0.45, b0=9.2, b1=-20.4, b2=12.2
with capacitors ON
U
2
Q Qn b0 b1
U a0=0.51, a1=0.49, b0=9.5, b1=-21.4, b2=13.2
b2
Un Un
Primarily residential load
Q 0,5Pn 1 q1U q2 U 2 summer: p1=1.1, q1=4.7, q2=23
Page 125
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-6: IEEE recommended typical induction motor data [161] (1995)
Page 126
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-8: Parameters for models of many load components [37] (1994)
Page 127
Modelling and Aggregation of Loads in Flexible Power Networks
Type Rs Xs Xm Rr Xr Tj á P S0
I 0 0.295 3.497 0.02 0.12 2.0 0.15 2.0 0.0116
II 0.013 0.11 3.0 0.012 0.12 2.0 0.15 2.0 0.011
T M =K L [α+(1-α)(1- S0)P], K L is a coefficient, α is the damp torque coefficient which is independent of the
speed, P is the exponential coefficient which relates to the damp torque.
Table 3-A-10: Induction motor parameters recommended by WSCC [172] (2008)
Page 128
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-13: Exponential and polynomial model parameters for power electronic (SMPS) load category
Exponential
PF1 Polynomial model
Load Type model
(cap)
kpu kqu p1 p2 p3 q1 q2 q3
SMPS with no PFC 0.994 0 2.36 0 0 1 -3.63 9.88 -7.25
SMPS with p-PFC 0.97 0 -0.5 0 0 1 0.45 -1.44 1.99
SMPS with a-PFC 1.00 0 N/A 0 0 1 0 0 0
Note: PFC, p-PFC and a-PFC are abbreviations for power factor correction, active power factor correction and passive
power factor correction, respectively. PF1 is displacement/fundamental power factor (capacitive in case of SMPS load).
Table 3-A-13 to Table 3-A-17 present the results of exponential (3.5)–(3.6) and polynomial (3.11)–(3.12)
model interpretations of power electronics interfaced loads: SMPS load category, energy efficient light sources
(CFL and LED) load category, single-phase drive-controlled motor load category and three-phase drive-
controlled motor load category [20], [45–48] (2007-2009). The parameters of exponential and polynomial
load model parameters are obtained on the basis of data analysis from steady-states for different values of
supply voltage.
Table 3-A-14: Exponential and polynomial model parameters for energy efficient lighting (CFL/LED) load category
Note: Load models of high-intensity discharge (HID) light sources, which are often used in commercial load sector, are
currently being developed.
Table 3-A-15: Exponential and polynomial model parameters for single-phase drive-controlled motor (SASD) load category
Note: Abbreviations CT, LT, QT and CP denote constant torque, linear torque, quadratic torque and constant power
mechanical load, respectively.
Page 129
Modelling and Aggregation of Loads in Flexible Power Networks
Table 3-A-16: Exponential and polynomial model parameters for three-phase drive-controlled motor (ASD) operated in continuous mode
Table 3-A-17: Exponential and polynomial model parameters for three-phase drive-controlled motor (ASD) operated in discontinuous mode
The results of processing parameters of exponential static load model (3.5)-(3.6) for different low voltage
devices from existing literature showed that they attain normal probability distribution [240]. The centres of
probability of kpu and kqu are 0.158 and 0.292, respectively. However, the parameters vary in very wide
ranges and corresponding standard deviations are more than 2.5 times larger than the corresponding centers
of probability. Much larger standard deviations are obtained for the parameters of polynomial/ZIP model
(3.11)-(3.12). Furthermore, the analysis showed that the load devices of the same type can have quite different
parameter values because the parameters can vary depending on the manufacturer, used electrical circuits,
production process, auxiliary components, operating conditions, etc.
Table 3-A-18 and Table 3-A-19 present load models and the parameters for residential and commercial load
class, respectively, found in existing literature.
Page 130
Modelling and Aggregation of Loads in Flexible Power Networks
Variant or
Ref Load model Values
season
With motors PF = 0.82, ZP = 0.21, IP = 0.15, PP = 0.64, ZQ = 4.74,
ZIP
included IQ = -6.19, PQ = 2.45
[14]
With motor
PF = 0.88, ZP = 0.29, IP = 0.15, PP = 0.56, ZQ = 4.5,
ZIP + motor modelled
IQ = -6.19, PQ = 2.45
separately
Page 131
Modelling and Aggregation of Loads in Flexible Power Networks
Bridge Inverter
Lemi
Toroid esr Lres sqrt((i_tube^2)/time))
i_tube i_tube^2
Rin N N IT
1 3
System
Cdc N Cres
Impedance
2 VT/IT
v_tube R_tube
Supply system
(a) (b)
Figure 3-B-3: DQ equivalent circuit for three-phase induction motor: a) d-axis b) q-axis
Switch-mode power supply (SMPS) loads are modern power electronic devices (“consumer electronics”, PCs, TVs,
CD/DVD players, etc.) that are sensitive to voltage variations and require a regulated dc voltage supply. The
SMPS include a front-end rectifier, dc link and dc-dc converter operating with feedback control to provide
regulated output, as well as other power electronic components which should be accurately represented in the
full-circuit load model. For the SMPS load category, the equivalent-circuit model is introduced based on the
principle that, for a steady-state dc load, this equipment operates as a constant power load, as the SMPS is
able to regulate the dc load voltage over a range of supply system voltages. The implementation of the
equivalent resistance for this load category is represented by the equation
2
vdc
req (3-B.1)
Prated
where: vdc is the instantaneous value of dc link voltage and Prated is the rated power of the supplied dc load.
Energy efficient lighting (CFL/LED) load category is currently dominated by compact fluorescent lamps (CFLs), but
it is expected that (organic) light-emitting diode, (O)LED, light sources will increase in popularity and in numbers
Page 132
Modelling and Aggregation of Loads in Flexible Power Networks
in the future. Currently implemented LED light source circuit topologies consist of a diode bridge rectifier
directly supplying the LED chain (the current characteristics of which can be easily modelled using the diode
equation), and is therefore no need to simplify the circuit, as it already has the form of the equivalent-circuit
model. Standard circuit topology of modern CFLs consists of an electronic ballast circuit with self-oscillating
inverter, used for controlling the voltage across the fluorescent tube as presented in the full circuit CFL model in
Figure 3-B-1. It is possible to represent steady state behaviour of all components behind the Cdc with an
equivalent resistance. The mathematical formulation of this resistance changes depending on instantaneous dc
link voltage vdc, which is determined by the state of charge of the Cdc. One formulation of the resistance should
be used during charging and the other during the discharging stage:
Single-phase and three-phase adjustable speed drive (ASD) controlled motors may also be represented by the
equivalent circuit models. For these two load categories, the equivalent resistance is determined by the motor
loading conditions, which can be represented with the four general types of mechanical loads: constant torque,
linear torque, quadratic torque, and constant power. The final results are again obtained after investigating
the relationship between the active power/current drawn at the dc link and the dc link voltage for considered
mechanical loading conditions. These results are then converted to the equivalent resistance using the standard
relationship (3-B.1) with Pdc (the active power drawn at the dc link) instead of Prated.
The vast majority of single-phase ASDs consist of a single-phase front end rectifier and a three phase inverter
controlling a small three-phase motor. Therefore, the full circuit model consists of a 4th order dq motor model
connected to front-end rectifier/inverter. Three-phase ASDs can be divided into two general categories: those
that operate in continuous mode and those that operate in discontinuous modes. In continuous current conduction
mode, the dc link filter inductor current never falls to zero, while in discontinuous current conduction mode the dc
link filter inductor current does fall to zero. Mode of operation is determined by the dc link filter components
(Ldc and Cdc).
Page 133
Modelling and Aggregation of Loads in Flexible Power Networks
rs X ss 0
b
X ms 0
s
vqs b i s
s
X mS s
qs
0 rS X SS 0
vds b b ids
's (3-C.1)
1 r ' 1 r ' i ' s
vqr X ms X ms rr' X rr X RR qr
v ' s b n b b n b 's
dr idr
n r X ms X mS '
n r X rr rR' '
X RR
b b b b
The values used for typical single-phase induction motor (SPIM) parameters in Table 3-C-1. The parameters of
exponential and ZIP model which are developed from this model are presented in Table 3-C-2. In this table
abbreviations IR and CR denote motors without run capacitors (inductor run SPIM) and motors with them
(capacitor run SPIM), respectively. The abbreviations CT, LT, QT and CP are the types of motor mechanical
loads: constant torque, quadratic torque, linear torque and constant mechanical power, respectively.
Table 3-C-1: Values of SPIM [55]
Parameter (Ù)
Main winding Magnetising Auxiliary winding
Page 134
Modelling and Aggregation of Loads in Flexible Power Networks
Page 135
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 3-D-1: The stalling motor model used in EPRI’s composite load model [80]
The functionality of motor D from WECC composite load model (that represents single-phase motors mainly
residential air conditioners) is implemented using what is called a “performance model”. In this methodology the
actual measured response of a single-phase a/c compressor motor to a gradual ramp down and up in voltage
is taken and two pairs of polynomial equations are fitted to the response. One pair is fitted to the real power
response and one pair to the reactive power response. The polynomial equations are quadratic equations. One
represents the behaviour of the unit while running normally and the other when the unit has stalled (effectively
a constant impedance curve). This can be seen in Figure 3-D-3, where we see the two regions of operation of
the motor: normal and stalling. The transition from the normal running curve to the stalled curve is based on the
stalling voltage, Vstall. That is, the voltage at which the motor stalls. So once again a state-transition algorithm
Page 136
Modelling and Aggregation of Loads in Flexible Power Networks
determines when the voltage has fallen below the stall voltage and thus switches the model response from one
curve to the other.
Detailed description of the final development of the recently released and approved WECC composite load
model “cmpldw” can be found in [1].
Figure 3-D-2: State transition diagram for stalling motor model [80]
Figure 3-D-3: Performance model approach (reproduced with permission from [174]).
Page 137
Modelling and Aggregation of Loads in Flexible Power Networks
component is used to control the voltage of the DC-bus. The reactive power of the DESS is controlled through
the quadratic current component irq of the inverter. The use of a controlled DC voltage and the general layout
of the control scheme are chosen in a way to be able to facilitate any future upgrade of this DESS model to a
multi-source power station.
Page 138
Modelling and Aggregation of Loads in Flexible Power Networks
HV
3-phase V& I
in HV
Meshed
MV
X
3-phase V& I
Radial
for one feeder
An 8 kHz sampling frequency was adopted to derive continuously the 10 cycle RMS average values. Following each
detected disturbance, the signals are recorded during 130 s with a 4kHz sampling frequency (see Figure 4-A-
2).
trigger
time
In this particular case, a sampling rate used for collecting data was 16 samples/cycle/channel (1kHz). Almost
all modern dedicated monitoring devices can sample at this, or much higher, rates.
Page 139
Modelling and Aggregation of Loads in Flexible Power Networks
10
8
s
9.50 9.75 10.00 10.25 10.50 10.75 11.00 11.25 11.50 11.75 12.00
[DFR_signal_20080913_093743_02 (imported)] Direct Voltage Amplitude(kV)
3
1
s
9.50 9.75 10.00 10.25 10.50 10.75 11.00 11.25 11.50 11.75 12.00
[DFR_signal_20080913_093743_02 (imported)] Direct Current Amplitude(kA)
51
50
s
9.50 9.75 10.00 10.25 10.50 10.75 11.00 11.25 11.50 11.75 12.00
[DFR_signal_20080913_093743_02 (imported)] Actual frequency (Hz)
30
20
s
9.50 9.75 10.00 10.25 10.50 10.75 11.00 11.25 11.50 11.75 12.00
[DFR_signal_20080913_093743_02 (imported)] Direct Active Power(MW)
10
-0 s
9.50 9.75 10.00 10.25 10.50 10.75 11.00 11.25 11.50 11.75 12.00
[DFR_signal_20080913_093743_02 (imported)] Direct Reactive Power (Mvar)
Page 140
Modelling and Aggregation of Loads in Flexible Power Networks
A-1). The unit is disconnected within few cycles of STI detection. This activation time is below usual transmission
systems fault clearing times.
Table 4-A-1:Sensitivity to voltage sag amplitude and duration
Appliance 1 Appliance 2
Duration -30% -40% -50% -80% Duration -30% -40% -50% -80%
1c-1c-1c OK OK OK OK 1c-1c-1c OK OK OK OK
Sensitivity to voltage swell: Two of four appliances tripped for voltages above 125% of the nominal phase-neutral voltage.
If the STI protection is not activated, the stalling of the large induction motor will activate the current overload
protection (COP) that is set to trip after 1.5 s with a current higher than 140% of nominal one.
Page 141
Modelling and Aggregation of Loads in Flexible Power Networks
Vav
34.4
34.3 VfitFFT
Vav, KV
34.2
34.1
34.0
33.9
33.8
33.7
33.6
42
Power1
MW
40 P1
PfitFFT
38
0 1000 2000 3000 4000 5000 6000
22
20
18
Qpower1
MVAR
16 Q1
14 QfitFFT
12
0 1000 2000 3000 4000 5000 6000
Time, s
Power1
P1fitFFT
42.0 P1fitSG200
P1fitSG500
41.5 P1fitSG1000
P1fit200
P1fit500
41.0
P1fit1000
40.5
Power1
40.0
39.5
39.0
38.5
All three methods were implemented; the SG and AA were implemented with different windows of 200s, 500s
and 1000s, and the resulting curves for active power are shown in Figure 4-A-6. It is important to note that all
the methods yielded curves with similar trends but with varying accuracies. This would indicate that the major
trends have been captured by the filtering. For the AA method, the most obvious effect would be the averaging
has reduced the peaks quite drastically. In addition, as the window for the averaging increases, the
discrepancy between the FFT smoothed signal and that processed by AA increases as expected.
The next step would be to decide between the filtering methods. FFT filtering is deemed not to be suitable due
to the large errors that are introduced at the start and the end of the filtered data; typical of the FFT due to
offsets. For the AA method, the most obvious effect is the averaging has reduced the peaks quite drastically. In
addition, as the window for the averaging increases, the discrepancy between the FFT smoothed signal and
that processed by AA increases as expected.
Page 142
Modelling and Aggregation of Loads in Flexible Power Networks
⍺% of the total supplied load can be disconnected in case of large voltage sags;
(100-α)%
in 3ph α(100-β)/3%
(α.β)% for each phase
Single, two and three phase faults of various depth can be simulated in the HV system to assess the voltage sag
seen by in MV and LV taking into account the grounding and connection arrangements of HV/MV transformers
and MV/LV transformers.
The resulting voltage sags are compared to load sensitivity voltage ranges [Vtrip_min;Vtrip_max]. Above Vtrip_max, 100%
of the load will remain connected, while below Vtrip_min, 100% of the load subjected to EUVLS will be disconnected.
Vph 0.4 kV Vmes 11 kV
100.00
1 p.u.
Load shed in % of the total load
80.00
0.00
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Positive voltage (p.u.)
Phase 1
Phase 2
Three phase load Vpos 11kV Load shed envelope Recording campaign (11kV)
Phase 3
disconnection window
Single phase load ESG implemented load shed envelope
disconnection windows
Figure 4-B-2: EUVLS disconnection windows Figure 4-B-3: EUVLS envelope characteristic
This model leads to voltage disconnection windows (see Figure 4-B-2). The amount of load disconnected is
related to the MV positive magnitude of voltage sag seen from the MV side. This characteristic shows a
piecewise linear function shape and depends on the type of electrical appliances supplied by the MV
substation.
Page 143
Modelling and Aggregation of Loads in Flexible Power Networks
The availability of recordings allows identifying a EUVLS curve by tuning Vtrip_min and Vtrip_max levels in LV and
MV and the proportions α and .
The EUVLS envelope can be adjusted to cover various severity scenarios and discretized in few steps to simplify
the parameters coding in the simulation tool (see Figure 4-B-3). This model justifies the EUVLS phenomenon
observed for positive voltage falling just below 0.9 p.u..
100 100%
Residual Voltage [%]
Load Self-disconnection
Starting Voltage 80%
(Residual) Voltage [%]
60%
50 Load Self-disconnection
Saturation Voltage
40%
20%
0 50 100 [%]
0%
Load Self-disconnection Ratio
0% 20% 40% 60% 80% 100%
Maximum Load Self-disconnection Amount Load self-disconnection amount [%]
Figure 4-B-4: An image of a piecewise linear approximation of self-disconnection characteristics and self-disconnection of system loads obtained from
measurements on 29 August 2008 [17].
Page 144
Modelling and Aggregation of Loads in Flexible Power Networks
Page 145
Modelling and Aggregation of Loads in Flexible Power Networks
Input data
Load structure
Models of main
and Network data
load categories
Load curves
Load aggregation
(Low-voltage)
Network representation
(with low-voltage aggregate load connected)
Figure 5-A-1: The “flow chart” of the proposed load aggregation methodology.
The type of the target power system analysis/study will determine the actual form of the aggregate MV load
model. For steady-state power system analysis, for example, the MV load model will typically take
exponential or polynomial/ZIP forms. However, by considering the distribution network impedance, it is
possible to represent component-based models (e.g. motor models) at higher voltage-levels.
5-A.2 Data requirements for identifying load structure and load compositio n
The development of accurate aggregate load models for every load sector and load sub-sector requires either
extensive long-term and wide-scale measurement campaigns (as in various measurement-based modelling
approaches), or access to the representative and detailed statistical data (as in various component-based
approaches). The information that should be collected is usually related to active and reactive power demands
of the aggregate load (and individual load components in the aggregate demands) with certain resolution –
typically 30-min or 60-min intervals, but sometimes shorter periods may be required (e.g. 1-min intervals).
If measurements are used, and if sufficient processing, storage and metering resources are available,
instantaneous current and voltage waveforms may be used to accurately identify the actual load
composition/mix, using various non-invasive approaches (e.g. “load signatures” and “demand pattern
identification”). Both in component and measurement-based modelling approaches it is important to allow the
inclusion of hourly, daily, weekly and seasonal variations in the final aggregate load model.
Page 146
Modelling and Aggregation of Loads in Flexible Power Networks
- resistive loads
- single-phase and three-phase directly-connected motor loads
- dc power supplies, or switch-mode power supply (SMPS) loads
- energy efficient lighting, consisting of compact fluorescent lamps (CFL) loads and light-emitting diode
light sources (LED) loads
- single-phase and three-phase drive-controlled motors, or adjustable speed drive (ASD) loads
In addition to the loads, and as a part of the aggregate LV network/load representation, some micro and
small-scale distributed generation (DG), e.g. micro-PV, micro-Wind, micro-CHP etc., may be connected in
parallel to the LV loads within the end-users’ premises, while medium and large-scale DG units may be
connected at MV and higher voltage levels. This is illustrated in Figure 5-A-2.
Load models of resistive and directly-connected motor load categories are commonly available in existing
literature and simple to implement, as they draw continuous sinusoidal currents from the supply. Accordingly,
they can be accurately represented using the standard steady state and dynamic load model formulations (e.g.
exponential or polynomial/ZIP load models and associated power factors – very close to unity for resistive
loads). The three other load categories (SMPS loads, CFL/LED loads and ASD loads), however, represent non-
linear power electronic equipment, which require different load models and more detailed simulations for
accurate representation. The loads from these categories have increased significantly in numbers in recent
years, and their accurate (aggregate) models for the use in power system studies and analysis are missing in
existing literature (see Chapter 3). Additionally, the influence of harmonic legislation, differences in
technologies and circuit topology variations effectively introduced several sub-categories within each category
of these loads. An overview of the load categories and sub-categories is given in Table 5-A-1 (see also Figure
5-A-2 and Chapter 3 for more detail), while the subsequent text provides further details.
Bulk
Supply
System
11kV Busbar
Transformer
Medium/ Line/Feeder
Impedance
Large DG
Low-voltage Busbar
CT QT LT CP
Figure 5-A-2: Representation and composition of aggregate load based on the five general load categories (DG included).
Page 147
Modelling and Aggregation of Loads in Flexible Power Networks
no-PFC Devices with rated power less than or equal to 75 W are not
expected to have power factor correction (PFC) circuit.
Power electronics – SMPS loads a-PFC
Above 75 W, p-PFC is currently more common, but the contribution of
p-PFC a-PFC is expected to increase in future
CFL Low- and high-power variations possible. For CFLs, higher power
LED devices are expected to contain vf-PFC.
Energy efficient lighting – CFL/LED loads
LFL The details of LED loads are not commonly established, due to the
HID infancy of this technology
Page 148
Modelling and Aggregation of Loads in Flexible Power Networks
100
Consumer
90 electronics / ICT
Cooking
80
Wet
Time (hr)
Figure 5-A-3: Decomposition of the load curve into end-use load types for the aggregate UK residential load sector during average loading conditions,
where 1 p.u. (100%) corresponds to the peak winter demand, [104].
Step two – conversion of “load type” structure into main “load modelling categories”
The determined load structure must then be converted from “end-use load types” into the general or main “load
modelling categories” (and sub-categories), discussed in Chapter 3 and Section 5-A.3. This again may be
achieved without direct measurements, e.g. by using national statistics ([96], [181]) and associated legislation
(e.g. [182]).
“Wet load”
Wet load consists of dishwashers, tumble-dryers, washer-dryers and washing machines. Due to the high running
torque requirements, tumble-dryers, washer-dryers and washing machines will require run capacitors. The
running torque of dishwashers does not require a run capacitor, so they will utilise inductor run SPIMs. All wet
loads are assumed to have constant torque (CT) mechanical loading.
“Cold load”
The cold load type covers all variants of refrigerators and freezers. As the compressors used in such devices do
not require high running torque, it is expected that 100% of this load uses inductor run SPIM. The motor
mechanical loading is represented by quadratic torque conditions.
Information and communication technologies (ICT) load
This load category covers all types of home computers and related information/communication technology
equipment (e.g. modems, printers, etc.). This load is dominated by desktop computers and monitors, all with
relatively large rated powers. Based on the implemented power factor correction (PFC) circuit (active, passive
or none), electrical equipment/load here can be divided into three categories: load with a-PFC, load with p-
PFC and load with no-PFC.
Many laptop chargers will not have to satisfy harmonic legislation, due to the low rated powers. Laptop
chargers that do have to adhere to legislation will have a-PFC, as the size and weight of the p-PFC inductor is
not suited to portable applications. As mentioned, desktop computers have larger rated powers and the vast
majority will have to have PFC, as there is no benefit from using a-PFC. Monitors and printers of high enough
rated power will incorporate p-PFC. All other loads in this load type will have low-rated powers (≤ 75 W) and
are expected to have no-PFC.
Consumer electronics
This load category is dominated by televisions/TV sets, with the remaining load consisting of various devices
with rated power below 75 W (e.g. CD/DVD players). As with the ICT load type, these loads also can be
divided into SMPS loads with a-PFC, with p-PFC and with no-PFC.
Page 149
Modelling and Aggregation of Loads in Flexible Power Networks
Larger (primary) TV’s will incorporate p-PFC, with more recent technologies likely to have a-PFC. Smaller
(secondary) TV’s below 75 W will have no-PFC.
Cooking load
Hobs and ovens are modelled as resistive load category, while microwaves are modelled as p-PFC SMPS.
Lighting load
Many lighting solutions exist, and national statistics must be used to determine the expected contribution for
each location. In the residential sector, the majority of lighting is either from general incandescent lamps (GILs)
or compact fluorescent lamps (CFLs) light sources. The contribution from CFLs is expected to increase in the near
future, due to a recent ban and phase-out of GILs (EU [183], US [184]), with light-emitting diode (LED) light
sources expected to increase in the future.
Applying the above assumptions for the UK case, it is possible to convert the load type demand profile (Figure
5-A-3) into a load category demand profile. This is shown in Figure 5-A-4.
100
SMPS:
90 a_PFC
p_PFC
80
no_PFC
70
Contribution to load (%)
SPIM:
60 IR QT
IR CT
50
CR CT
40
Resistive:
30
20
Lighting:
10 GIL
CFL
0
2 4 6 8 10 12 14 16 18 20 22 24
Time (hr)
Figure 5-A-4: Decomposition of the aggregate UK residential load curve for average loading conditions into general load categories and sub-categories.
1.8 np
nq
Exponential model coefficient value
1.6
1.4
1.2
1.0
0.8
0.6
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr)
Page 150
Modelling and Aggregation of Loads in Flexible Power Networks
1.0
2.0 Constant impedance coefficient, Zq
Constant impedance coefficient, ZP
0.9 Constant current coefficient, Iq
Constant current coefficient, IP
0.8 Constant power coefficient, PP 1.5 Constant power coefficient, Pq
0.7
1.0
0.6
0.5 0.5
0.4
0.0
0.3
0.2 -0.5
0.1
-1.0
0.0 0 2 4 6 8 10 12 14 16 18 20 22 24
0 2 4 6 8 10 12 14 16 18 20 22 24 Time (hr)
Time (hr)
33
8 9 10 11 12 13 14 15
Page 151
Modelling and Aggregation of Loads in Flexible Power Networks
1.6
LV Aggregate Model (0.4kV) 1.8
MV Aggregate Model (11kV)
1.4
1.2
1.2
1.0
1.0
0.8 0.8
0.6
0.6
0.4
0 2 4 6 8 10 12 14 16 18 20 22 24
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr)
Time (hr)
1.0 2.0
LV Aggregate Model (0.4kV): ZP IP PP LV Aggregate Model (0.4kV): ZQ IQ PQ
0.9
MV Aggregate Model (11kV): ZP IP PP MV Aggregate Model (11kV): ZQ IQ PQ
1.5
0.8
0.7 1.0
0.6
0.5
0.5
0.4 0.0
0.3
-0.5
0.2
0.1 -1.0
0.0
0 2 4 6 8 10 12 14 16 18 20 22 24
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr) Time (hr)
Page 152
Modelling and Aggregation of Loads in Flexible Power Networks
L L
HU18 (16 cust.) U18 (6 cust.)
C 76m E 90m
A A B L C D E L
HU19 U19
20m 16m 43m 27m 26m 53m (4 cust.)
(20 cust.) D 82m
B 66m
C C E E
30m 11m 40m 17m
L L L L (15 cust.) (15 cust.)
(33 cust.) (33 cust.)
HU14 HU15 U14 U15
HU16 HU17 U16 U17
11kV 0.4kV (21 cust.)
11kV 0.4kV (8 cust.) (8 cust.)
(21 cust.) L L L L
A A B D D E
9m 64m 71m 15m 80m 85m
L U9 L U10
HU9 HU10 (9 cust.)
(16 cust.) (8 cust.) (6 cust.)
D 22m E 32m
L L E E
C C
41m 6m 53m 12m
1000kVA D 26m 500kVA E 41m
(20 cust.) (10 cust.) (12 cust.) (6 cust.) U11
Transformer HU11 Transformer U4 E 70m
HU4 HU7 B 55m L (8 cust.)
U7 L (6 cust.)
ZT=1.1 + j4.62 L L ZT=2.04 + j9.28 L L
A A A A A A A A B C D E
(p.u. on 100MVA) (p.u. on 100MVA) 32m 90m 41m 35m 33m 17m
22m 72m 30m 25m 23m 10m
B 50m B 55m B 30m E 73m E 67m E 93m E 68m E 40m
B 60m B 75m
C C E E
L L L L
14m 40m L 24m 52m L
L L HU8 L L U8
HU12 L (12 cust.) U12 U13
L (22 cust.) HU13 (12 cust.)
HU5 HU6 (22 cust.) (14 cust.) U5 U6 (9 cust.)
HU3 (24 cust.) (14 cust.) U3 (14 cust.) (9 cust.) L
L (12 cust.) U1
(20 cust.) HU1
(31 cust.) E 36m (15 cust.)
B 26m D
A
63.6m 78.6m E 85m
B 70m
L L U2
HU2
* L type line length = 30m (12 cust.)
* L type line length = 30m (27 cust.)
M R7
L SU1 M M M
L L
(5 customers)
R1 R3 R5 * M type line length = 30m
SU3 SU5
(10 customers) (8 customers)
* L type line length = 35m
Page 153
Modelling and Aggregation of Loads in Flexible Power Networks
Cross Maximum
Positive sequence Neutral Zero-phase sequence
LV Line type Sectional sustained
Z Z Z
Area current
(CSA) Rph Xph Rneutral R0 X0 Izph
Id. Configuration
(mm2) (Ω/km) (Amps)
A 300 0.1 0.073 0.168 0.593 0.042 465
B Underground Line (Cable) 185 0.164 0.074 0.168 0.656 0.05 355
C 120 0.253 0.071 0.253 1.012 0.046 280
D EPR or XLPE 95 0.320 0.075 0.320 1.280 0.051 245
E 0.6/1 kV 4x(CSA) Al / Cu (earth) CNE 70 0.443 0.076 0.443 1.772 0.052 205
F 35 0.87 0.085 0.87 3.481 0.058 156
G 120 0.284 0.083 - 1.136 0.417 261
Overhead Line
H 95 0.32 0.085 - - - 228
I 70 0.497 0.086 0.63 2.387 0.447 195
Aerial Bundled Conductor (ABC)
J XLPE 4x(CSA) Al 50 0.397 0.279 - - - 168
K 35 0.574 0.294 - - - 148
Service Connection
L 35 0.851 0.041 0.9 3.404 0.03 120
PVC or XLPE
M 0.6/1 kV 1x(CSA) Al / Cu (neutral / earth) CNE 25 1.191 0.043 1.26 4.766 0.03 100
Public Lighting
N 25 1.18 0.043 0.9 4.72 0.03 100
PVC or XLPE
0.6/1 kV 1x(CSA) Cu / Cu (neutral / earth) CNE
Page 154
Modelling and Aggregation of Loads in Flexible Power Networks
70 70
(a) decomposition of minimum loading conditions into load (b) decomposition of minimum loading conditions into load
types, [104]. categories and sub-categories
100
Consumer
100
90 electronics / ICT SMPS:
Cooking 90 a_PFC
80 p_PFC
Wet 80
no_PFC
Active power demand (p.u.)
70
70
Contribution to load (%)
Cold SPIM:
60 60 RSIR_QT
Storage DHW
RSIR_CT
50 50
RSCR_CT
Direct DHW
40 40
Resistive:
Top up heating 30
30
Direct heating 20
20 Lighting:
Storage heating 10 GIL
10 CFL
0
Lighting 2 4 6 8 10 12 14 16 18 20 22 24
0
2 4 6 8 10 12 14 16 18 20 22 24 Time (hr)
Time (hr)
(c) decomposition of maximum load conditions into load (d) decomposition of maximum load conditions into load
types, [104]. categories and sub-categories
Figure 5-C-1: Characteristic loading conditions for residential load sector
1.6 1.6
1.4 1.4
1.2 1.2
1.0 1.0
0.8 0.8
0.6 0.6
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr) Time (hr)
Page 155
Modelling and Aggregation of Loads in Flexible Power Networks
1.0
2.0 Constant impedance coefficient, Zq
Constant impedance coefficient, ZP
0.9 Constant current coefficient, Iq
Constant current coefficient, IP
0.8 Constant power coefficient, PP 1.5 Constant power coefficient, Pq
0.7
1.0
0.6
0.5 0.5
0.4
0.0
0.3
0.2
-0.5
0.1
-1.0
0.0
0 2 4 6 8 10 12 14 16 18 20 22 24
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr)
Time (hr)
(a) minimum loading conditions active power coefficients (b) minimum loading conditions reactive power coefficients
1.0
Constant impedance coefficient, ZP Constant impedance coefficient, Zq
0.9 2.0 Constant current coefficient, Iq
Constant current coefficient, IP
0.8 Constant power coefficient, PP Constant power coefficient, Pq
1.5
Polynomial model coeffcient
Polynomial model coeffcient
0.7
1.0
0.6
0.5 0.5
0.4
0.0
0.3
0.2 -0.5
0.1
-1.0
0.0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
(c) maximum loading conditions active power coefficients (d) maximum loading conditions reactive power coefficients
Figure 5-C-3: Low-voltage aggregate polynomial load models for minimum and maximum residential loading conditions
Page 156
Modelling and Aggregation of Loads in Flexible Power Networks
ICT + CE
80 ICT + CE 80
Catering
Catering
70 70 Cold
Cold
Direct WH
Direct HW 60
60 Direct SH
Direct SH
Lighting
50 Lighting 50
40 40
30 30
20 20
10 10
0
0
0 2 4 6 8 10 12 14 16 18 20 22 24
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr) Time (hr)
(a) decomposition of minimum loading conditions into (b) decomposition of average loading conditions into load types,
load types, [104]. [104].
100 100
Lifts 3PIM Drive
90
ICT + CE SMPS:
Active power demand (% of peak load)
Catering npPFC
Contribution to load (% of peak load)
80 80
Cold pPFC
70 Direct WH aPFC
Direct SH 3PIM
60 60
Lighting SPIM
50 R
LFL
40 40 HID
CFL
30
GIL
20 20
10
0 0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hr) Time (hr)
(c) decomposition of maximum loading conditions into (d) decomposition of maximum loading conditions into load
load types, [104]. categories and sub-categories
Figure 5-D-1: Low-voltage aggregate polynomial load models for minimum and maximum residential loading conditions
Page 157
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 5-E-1: Load class mix in US [185] Figure 5-E-2: Residential load composition data in US [185]
Figure 5-E-3: Overall load composition data in California Figure 5-E-4: Residential load composition data in
[179] California [179]
Figure 5-E-5: Commercial load composition data Figure 5-E-6: Industrial load composition data (California)
(California)
Page 158
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 5-E-7: Decomposed load curves in Swedish house with direct electric heating (workday and holiday) [186]
Figure 5-E-8: Decomposed load curves in Swedish house without direct electric heating (workday and holiday) [186]
Figure 5-E-9: Decomposed load curves in Swedish Apartment (workday and holiday) [186]
Figure 5-E-10: Decomposed daily loading curves (DDLC) in Figure 5-E-11: DDLC in a “typical ” house in Greece [187]
“typical” house in Denmark [187]
Page 159
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 5-E-12: DDLC in a typical house in Italy [187] Figure 5-E-13: DDLC in a typical house in Portugal [187]
Figure 5-E-14: Load class mix under total load in winter Figure 5-E-15: Residential load composition data in winter
Germany [188] Germany [188]
Figure 5-E-16: Residential load composition in Italy (with Figure 5-E-17: Residential load composition in Spain
Air conditioners) [189] (summer)
Figure 5-E-18: Residential load composition in summer in Figure 5-E-19: Residential load composition in summer in
Scandinavia New Member States
Page 160
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 5-E-20: Residential load composition in Summer Figure 5-E-21: Residential load composition in Summer UK
Germany/Austria
Figure 5-E-22: Residential load decomposition data in Taiwan, China [190] (This curve is created in 1995)
Page 161
Modelling and Aggregation of Loads in Flexible Power Networks
3 It is assumed that individual WTG parameters are provided in per unit on individual machine base
Page 162
Modelling and Aggregation of Loads in Flexible Power Networks
supply, synthetic inertia4, etc. in accordance with grid code requirements [203], [204]. This type of models are
not component based models but performance-based models.
A further consideration in aggregated modelling is the representation of the WF cable network. In [193],
[195], [201] and [202] an approximate method is presented using short circuit fault calculations for deriving
the values of a series reactance that represents the cable network. This reactance is placed in series with the
WF transformer. The size of this reactance is much smaller than that of the leakage reactance of the WF
transformer, so it may also be neglected without a significant loss of accuracy.
4Synthetic inertia or simulated inertia - a facility provided to replicate the effect of inertia of a synchronous generating unit to
a prescribed level of performance.
Page 163
Modelling and Aggregation of Loads in Flexible Power Networks
400
400
Power Output (W/m ²) Generic 1
180 180
Power Output (W/m ²)
160
Figure 6-A-1: Comparison of four generic and a number of actual wind turbine power curves, all normalised using the corresponding swept areas: a) to
d) individual generic and the closest matches with actual wind turbines, adopted from [206].
Page 164
Modelling and Aggregation of Loads in Flexible Power Networks
For dynamic equivalent models for ADNC, strictly mathematical techniques that originate from control systems
theory have been proposed in [215], [216], where they perform quite well. Most of these model order
reduction methods, also known as Singular Value Decomposition (SVD) based methods, are based on modal
analysis to directly identify and preserve the modes of interest regarding the dynamic behaviour of distribution
networks while the others are eliminated. Thus, a balanced realization method with truncation performed
through singular perturbations theory was proposed in [215] and Hankel norm approximation, focusing on the
observability and controllability properties of the system, was used in [216]. The method of Hankel norm
approximation was considered the most efficient out of SVD based methods and, therefore, it was successfully
applied for dynamic equivalency purposes of a medium size distribution network with DG of various types and
size, including WTGs (squirrel cage induction generator and doubly-fed induction generator), split shaft
microturbines, combine heat and power (CHP) plants equipped with microturbine units and a diesel generator.
Another family of model reduction techniques is the Krylov methods, also referred as moment matching
methods, which are based on the leading coefficients of a power series expansion of the reduced system
transfer function around a user-defined point that have to match those of the original system transfer function
[119], [217], [218]. These methods have certain advantages and disadvantages when compared with strictly
mathematical techniques. Generally, SVD based methods hardly applicable to systems of very large
dimensions (thousands of states) since singular values of the detailed model of the system have to be computed
and such a computation might be cumbersome for large order systems. The Krylov methods are iterative in
nature and can be used for systems with large number of states. However, these methods do not provide the
bounds for the approximation error while SVD based methods do so. Therefore the selection of the reduction
method has to be performed in accordance with both the desired goals and the system structure. Krylov
methods are simple, they can be automated and they can provide a significant order reduction while retaining
small error with respect to the original model. They are especially profitable when applied to large systems.
Dynamic equivalents obtained by means of strictly mathematical techniques or Krylov methods lose direct
physical interpretation. The system is represented in state space form with states being some linear combination
of real physical variables. This raises the problem of how to integrate such equivalents into software for power
system simulations. It can be done by the introduction of controlled current sources in the lines connecting the
study network with the neighbouring external power system. The controllers of such current sources will
represent dynamic relationship between the current and the voltage [215–217]. Although the procedure for
obtaining them can be automated [216], [217], new reduced order models should be derived for different
operating conditions.
Since restrictions may arise when applying these methods to high dimensional systems [219] and linear models
provide a limited accuracy to represent nonlinear ADNC with effective dynamic impact in the study system
when major disturbances occur, system identification based methods have been proposed recently in the
technical literature [138–143], [220–223]. These methods consist in defining suitable model structures based on
the available prior knowledge and physical insights about the detailed system to be reduced, with the model
parameters being estimated from simulated data by means of a self-adaptive learning procedure. For this
purpose an identification criterion is defined, based on the error between output trajectories of the physical
system detailed model and the corresponding aggregated model, in order to measure how well the model fits
the system response. Then, with a proper optimization algorithm the model parameters are adjusted so that the
identification criterion is minimized. The resulting aggregated models present behaviours close to those
exhibited by physical systems, even if they do not have physical correspondence with them. Depending on the
physical insights effectively used to define the model structure black box modelling and grey box modelling
approaches have been reported on the literature.
Black-box modelling approaches have been exploited in [138–140], [220–222], avoiding the need of
detailed information about the network structure and parameters and also the need of complex mathematical
analysis. These facts represent a significant advantage especially when there is a limited understanding of the
relationships between system variables.
Page 165
Modelling and Aggregation of Loads in Flexible Power Networks
Thus, in [220], [221] the models are developed in the form of state space and auto-regressive model with
exogenous input (ARX), with simplicity being their main advantage. The whole distribution network model
containing steam, diesel and hydro generators was implemented in PSS/E dynamic tool to simulate the system
behaviour following different types of disturbances taking place on the study system. Simulated data
comprising time series voltage and frequency are used as inputs and time series of active and reactive power
are used as outputs. These time series are imported into the MATLAB System Identification Toolbox in order to
perform the estimation of the model parameters. These models were introduced later in PSS/E to replace the
distribution network for validation purposes. The performance of the proposed model is highly dependent on
the type and location of the disturbance.
A generic nonlinear dynamic equivalent model based on recurrent Artificial Neural Networks (ANN) was
proposed in [138], [139] and used to replace a MV distribution network containing several tens of DG units,
such as fuel cells and micro turbines with several ratings, connected near the end user terminals at the LV levels.
The model structure is defined through the recurrent ANN structure, which allows capturing the dynamic
behaviour of the replaced ADNC and to enable the interaction with the study system through the boundary
buses at a wide range of operating conditions, keeping the continuous time operation of the entire network. The
ANN acts as a Norton model, where the normalized deviations of boundary bus voltages are used as the main
inputs and the normalized deviations of currents represent the outputs. In addition to the input voltages, past
values of both current and voltages deviations are introduced as the input layer to achieve the recurrent
structure. Three-phase short-circuits are simulated in the study system using the system detailed model
implemented in the PSD simulation package [137]. During the faults, both complex voltages and injected
currents are measured at boundary buses and stored to be used subsequently to prepare suitable patterns for
ANN training purposes. The training process was accomplished offline. Passive loads are represented as
lumped equivalent elements connected at boundary nodes using constant impedance equivalent elements.
Loads voltage and frequency dependence can also be modelled using the general exponential relationships.
The separation between active and passive elements extends the validity of the equivalent model to simulate
changes regarding the generation and load conditions inside the replaced system itself, giving more flexibility
regarding power system analysis. However, the entire distribution system can also be replaced by the recurrent
ANN if there is a difficulty in representing the passive loads separately. The use of normalized deviations as
model inputs and outputs allows the equivalent model to be used under new initial power flow conditions and,
therefore, the ANN based aggregated model represents a normalized model scaled on initial operating
conditions at the boundary buses. Augmenting this feature with the independent representation of both active
and passive elements, a general model is derived with capability for simulating the original system under
different operating conditions.
The ADNC aggregated model proposed in [141–143] comprises a composite load model also referred as ZIP-
IM load model [60], connected in parallel with a converter interfaced generator. This model is intended to be
connected to the ADNC point of connection and is represented in the form of a sixth-order nonlinear state
space model resulting from the differential and algebraic equations that typically describe the dynamics of the
model individual components. Voltage and frequency are defined as the model inputs whereas both active and
reactive powers are used as the outputs. All the input and output signals are measured at the ADNC point of
connection following various disturbances simulated using the system detailed model implemented in DIgSILENT
PowerFactory software. These recorded signals are used for parameters estimation purposes through nonlinear
least square optimization techniques using the MATLAB System Identification Toolbox. The performance of the
ADNC aggregated model was evaluated by comparing its response with the one obtained from the system
detailed model implemented in DIgSILENT PowerFactory simulation tool, demonstrating the model effectiveness.
However, further developments are required in order to extend the dynamic equivalent considering the
presence of other inverter interfaced DG technologies, such as fuel cells and PV systems.
Page 166
Modelling and Aggregation of Loads in Flexible Power Networks
Page 167
Modelling and Aggregation of Loads in Flexible Power Networks
DG systems include modular and small scale generation systems based on renewable energy sources such as
Photovoltaics (PV) or micro wind generators and fuel-based generators in Combined Heat and Power (CHP)
applications, such as microturbines and fuel cells. Due to the type of energy conversion system used,
microsources are interfaced through power electronic converters to the LV network. Therefore, the inverter
dominated MG requires that power balance during transients have to be provided by energy storage devices,
either flywheels connected to the LV network through AC/DC/AC power electronic interfaces or batteries and
supercapacitors connected to the dc-link of microgeneration systems [231].
PV
Microturbine
MC
AC
DC
AC
AC
LC
MC PV
MC
DC
LV
LC AC
LC
MV
MC
Wind Generator
LC
Fuel Cell
MC
AC
DMS MGCC DC
MC MC
AC
AC
AC
DC LC
Storage
device
Microturbine
In order to achieve the desired flexibility, the MG is centrally controlled and managed by the Micro Grid
Central Controller (MGCC) installed in the LV side of the MV/LV distribution transformer, which communicates
with controllers located in a lower hierarchical level comprising local Microsource Controllers (MC) and Load
Controllers (LC). The MC takes advantage of the MS power electronic interface and can be enhanced with
various degrees of intelligence in order to control both voltage and frequency of the MG during transient
conditions based on only local information. The MGCC functions can range from monitoring the active and
reactive power of MS to assuming full responsibility of optimizing the MG operation by sending set points to
the MC and LC in order to control microgenerators and controllable loads, respectively [231].
Page 168
Modelling and Aggregation of Loads in Flexible Power Networks
interfaces, describing the MG dynamic behaviour only under balanced conditions. However, recently, models
able to simulate the MG operation under unbalanced conditions have been reported in the literature [239].
Aggregated models for MG have been successfully developed under the framework of the MORE MICRO
GRIDS Project, then focusing the MMG concept, exploiting system identification techniques [153–155].
Therefore, the system identification procedure consists on finding just another mathematical representation of
the MG built upon the corresponding MG detailed model, being the MG operated under balanced conditions
according to the SMO control approach, such that the MG is able to provide primary frequency control through
the VSI interfacing the main storage device and secondary load frequency control by means of the controllable
DG units. The MG dynamic equivalents are able to replace the MG in dynamic simulations following the
occurrence of several disturbances at the MV level, like MMG islanding and load following in islanded mode.
Short-circuits were not addressed since the contribution of power electronic based interfaces for short-circuit
currents is limited by their low over current capacity when compared to conventional generators. The MG
equivalent model is basically divided into two parts taking into account the dynamics going on the several
active devices, i.e. fast dynamic response and slow dynamic response. The MG fast dynamic response is
represented by the detailed model of the Voltage Source Inverter (VSI) interfacing a constant voltage source
that represents the main storage device. On the other hand, the MG slow dynamic response is represented by
reduced order models that can be derived from two different approaches: Black box and grey box modelling.
A black box modelling approach based on ANN was proposed in [153]. The MG dynamic equivalent is based
on a Time Delay Neural Network (TDNN) model structure combining a Nonlinear Finite Impulse Response (NFIR)
input structure with a Multilayer Perceptron (MLP) neural network as the nonlinear mapping. The TDNN based
model was trained through the Levenberg-Marquardt method and Mean Square Error (MSE) criterion using the
MATLAB Neural Network Toolbox, with the MLP neural network structure being optimized using early stopping.
The training process is performed offline using a data set based on time series obtained from time domain
simulations using the MG detailed model implemented in Matlab®/Simulink® environment.
The grey box modelling approach was exploited in [155] to derive a physically based MG dynamic
equivalent. The selected model structure is similar to the active power control system found in diesel engine run
power generators. The parameter identification task is performed using Evolutionary Particle Swarm
Optimization (EPSO) as the global optimization tool and the Sum Square Error (SSE) as the identification
criterion. The loss function of each particle is evaluated using time domain simulation. For this purpose the model
structure was embedded in the dynamic simulation platform built in Matlab®/Simulink® environment, replacing
the MG slow dynamics detailed model. The EPSO sends the parameter vector to the dynamic simulation
platform, being its performance evaluated through the SSE calculated using the error of the time domain
responses of both the MG equivalent model and the MG detailed model. The SSE is then sent to the EPSO
algorithm in order to perform the search towards the optimum solution in the SSE sense. This modelling
approach is presented in detail in this chapter.
Page 169
Modelling and Aggregation of Loads in Flexible Power Networks
Figure 6-B-1: Implementation and interaction of the recurrent ANN based dynamic equivalent with the retained network
As it can be observed from Figure 6-B-1, at each time step state variables and boundary bus voltages are
captured and processed to get a complete input set to the ANN through the mapping function, f1 , as follows:
U i U i ,0
f1 U i , i 1, 2, ..., j
n
(6-B.1)
U i ,0
Where:
A block simulating the behaviour of the ANN is used to process the inputs with the help of the information about
the ANN structure, such as biases, weights and types of activation functions, to get the corresponding output
currents. The information describing the ANN structure is saved in a supplementary file to be used through the
simulation process. The outputs of the ANN are then used to define the corresponding normalized deviations of
currents for the active components, I na , computed through the demapping function f 2 , as:
I a , i I a ,i
0
f 2 I a ,i , i 1, 2, ..., j
n
0 (6-B.2)
I a ,i
Where:
I a ,i , I a ,i is the current of active sources at boundary bus i and its initial value;
0
Page 170
Modelling and Aggregation of Loads in Flexible Power Networks
Real and imaginary parts of U in are used as separate inputs to the ANN while real and imaginary
components of I na ,i are obtained separately at the output layer. The complex power is then calculated and
supplied to the retained network.
x Ax Bu f ( x)
(6-B.3)
y Cx Du f ( x)
where A, B, C and D are the coefficient matrices, x is the state vector, u is the input vector, y is the output vector
and f(x) is a function that represents the nonlinear parts of the model.
The schematic of the composite equivalent circuit for the ZIP-IM load model is shown in Figure 6-B-2 [141–143].
The dynamic part of this model is represented by internal voltage relationships of the third-order induction
motor [144].
Constant Constant Constant
power current impedance
loads loads loads
V IM
ZIP
Figure 6-B-2: The equivalent circuit of the ZIP-IM load model
The composite load model used in the DNC model is described by the following equations:
dEm 1
Bm Em CmV cos m (6-B.4)
dt Tdm
d m CV
r s m sin m (6-B.5)
dt Em
Tdm
dm 1 Em V
sin m Tm (6-B.6)
dt H m X m
Xm X X m
where Bm , Cm m , m is the subscript of motor, E' is voltage behind the transient reactance, T'd is
X m X m
d-axis time constant, V is the bus voltage, ωr is the angular velocity of rotor, ωs is the angular velocity of stator,
ω is the angular frequency, δ is the angle between E' and V, H is the inertia, X is the reactance, X' is the
transient reactance and Tm is the mechanical torque.
The output equations of the composite load model are expressed in the following way:
Page 171
Modelling and Aggregation of Loads in Flexible Power Networks
V 2 V
1
PL PZIP 0 PZ PI PP Bm Em sin m
Vo Vo
(6-B.7)
V 2 V
1
V2
QL QZIP 0 QZ QI QQ Bm Em cos m
Vo Vo X m
where PL and QL are the real and reactive power of the ZIP-IM model, respectively; PZIP0 and QZIP0 are the
real and reactive power of the static ZIP model at steady state,; PZ and QZ are the constant impedance part of
the ZIP model; PI and QI are the constant current part; and PP and QQ are the constant power part.
The converter-connected generator is composed of a second-order synchronous generator model and a back-
to-back full converter model [146], [147], [150]. The synchronous generator interfaces with the grid via a
back-to-back full converter as shown in Figure 6-B-3 [146], [147]. The real power flow through the converter is
balanced via the DC-link (the capacitor linking inverter and rectifier).
Generator
Grid side Inverter Rectifier side
IG Ig
+ + IDC +
VG VDC Vg
_ _ _
Figure 6-B-3: The back-to-back full converter model. Adopted from [146], [147]
The dynamic parts of the converter-connected generator can be described by (6-B.8) to (6-B.10).
dEg
dt
1
Tdg
EFD Eg X g X g I d
(6-B.8)
1
Tdg
EFD Bg Eg CgV cos g
dg
dt
1
Hg
Tm Te Dg
(6-B.9)
1
Hg
Tm Bg Eg sin g Dg
dVDC
dt
1
CVDC
VD I D VQ IQ Vd I d Vq I q (6-B.10)
Xg X X g
where Bg , Cg g , g is the subscript of generator, E' is voltage behind the transient reactance, T'd
X g X g
is d-axis time constant, V is the bus voltage, Vo is the nominal bus voltage, ωr is the angular velocity of rotor, ωs
is the angular velocity of stator, ω is the angular frequency, δ is the angle between E' and V, H is the inertia, X
is the reactance, X' is the transient reactance, Tm and Te is the mechanical and electrical torque respectively, EFD
is the excitation voltage, D is the damping factor, C is the capacitance, VDC is the capacitor DC voltage, IDC is
the capacitor DC current, VD and ID is the d-axis voltage and current of the grid side of converter respectively,
VQ and IQ is the q-axis voltage and current of the grid side of converter respectively, Vd and Id is the d-axis
Page 172
Modelling and Aggregation of Loads in Flexible Power Networks
voltage and current of the generator side of converter respectively, Vq and Iq is the q-axis voltage and current
of the generator side of converter respectively.
The output real and reactive power of the converter-connected generator are computed as follows:
where PG and QG are the active and reactive power of the converter-connected generator part.
Based on (6-B.4) to (6-B.11), the system states, inputs and outputs are defined as follows:
x1 Em
x
2 m
x u V P
x 3 m , u 1 , y
x4 Eg u2 s Q
x5 g
x6 VDC
The output equations (P and Q) of the equivalent model can be obtained from the following equation:
P PG PL
(6-B.12)
Q QG QL
Therefore, the nonlinear state space model of DNC in its final form can be described as follows:
Bm
T 0 0 0 0 0
dm Cm
0 0 1 0 0 0 T cos m 0
x1
dm
0
Am 0 0 0 0
0 x m sin m
C
1 0
Hm 2 Tm E
x3 u
0
x Bg 0
1
0 x4
0 (6-B.13)
0 0 0 0
u2 EFD
Tdg
x g cos g
C
0 0
Ag D Tdg
5
0 0 0 0 x6 0
Hg Hg 0 0
0
K1
0 0
0 0 0 0
CVDC
2
Page 173
Modelling and Aggregation of Loads in Flexible Power Networks
x
1
V P
V sin V
sin g I DC
x
2
P P
I
0 0 0
u
X x
0
X g V V P
m ZIP 0 Z
P
2
y
m
3
o o 1 ZIP 0 P
(6-B.14)
V cos V x Q Q V Q V V u Q Q
cos g I DC 0
ZIP 0 Q
X x
4 I 2
0 0 0
m
X g V V X
ZIP 0 Z 2
X m
g 5 o o g
x
6
Xm X m X m Xg X X g V V
where Bm , Cm , Bg , Cg g , Am sin m , Ag sin g and
X m X m X g X g X m X g
K1 VD I D VQ IQ Vd I d Vq I q
The unknown parameters in this equivalent model are represented as P1, P2, P3, ….., P20 for model simplicity
purpose. Finally, the nonlinear state space model of DNC can be summarized by matrix equation (6-B.15).
P1 0 0 0
0 0 x1 P7 0 0
0 1 0 0
0
0 x2 P8 1 0
P2 0 0 0
0 0 x3 0 0 u1 0
x
0 0 P3 00 0 x4 P9 0 u2 P10
0 0 P 4 P5 0 x5 0 0
0 0
0 0 0 0 P6 x6 0 0
0 0
(6-B.15)
x1
x
2
P11 0 0 P12 0 P13 x3 P17 0 u1 P19
y
P14 0 0 P15 0 P16 x4 P18 0 u2 P 20
x5
x6
Obviously the dynamic equivalent model of DNC as shown in (6-B.15) is a nonlinear model and has twenty
unknown parameters. These twenty parameters have to be identified using a suitable parameter estimation
procedure, custom-made or commercially available. Parameters P10, P19 and P20 featuring in (6-B.15) are
the nonlinear parts of the equivalent DNC model.
Page 174
Modelling and Aggregation of Loads in Flexible Power Networks
TDNN
v D
f1
q-1
VD , VQ
q-1
I DR , I QR vQ
MLP i D
q-1
Boundary neural
bus network iQ
q-1
q-1
q-1
f2
At each time step, the MG slow dynamics equivalent model recognizes the operating status of the retained
network through the boundary bus voltage and system frequency. The normalized voltage and system
frequency deviations, vD , vQ and , respectively, are computed through the function f1 . The TDNN is then
used to determine the corresponding normalized current deviations, iD and iQ . Therefore the current to be
injected into the retained network is computed using the function f 2 . These normalized deviations are computed
based on their initial steady state values as follows
VD VD 0 VQ VQ 0 0
vD ; vQ ; (6-C.1)
VD ,max VQ ,max max
0
I DR iD ´ I D,max I DR ; IQR iQ ´ IQ,max IQR 0
(6-C.2)
where vD , vQ , iD , iQ are the normalized deviations of both voltage and current D Q components;
is the normalized deviation of frequency;
VD ,max , VQ ,max , I DR ,max , I QR ,max are the maximum variations considered to normalize both voltage
and current D Q components;
max is the maximum frequency deviation considered to normalize frequency;
VD 0 , VQ 0 , I DR
0 0
, I QR are the initial steady state values of both voltage and current D Q components;
0 is the nominal value of system frequency.
The initial steady state values of boundary bus voltage and injected current of the MG slow dynamics
equivalent model are determined through the initial load flow calculations. Their maximum deviations as well as
Page 175
Modelling and Aggregation of Loads in Flexible Power Networks
frequency maximum deviation are obtained from the dynamic simulation of the largest amount of load
connection and disconnection upon MMG islanding.
VQ I DR
Instantaneous
t power
theory
Pm I QR
Qref
Figure 6-C-2: Interface between the MG slow dynamics equivalent model and LV network
The current source assumes the role of the inverter, by determining the current from both the active power
delivered by the MG slow dynamics equivalent model, Pm , and a given reactive power, Qref , which
corresponds either to a pre-defined value linked to a given MS power factor or a reactive power set-point
sent by the MGCC.
The instantaneous power theory was proposed in [211] for control of active power filters and has been used to
control the PWM-VSI (Pulse Width Modulation – Voltage Source Inverter) or PWM-CSI (Pulse Width
Modulation – Current Source Inverter). Voltages or current reference signals employed to turn on and turn off
the switches of the inverter can be obtained from this theory [240], [241]. In this case, the instantaneous
voltages and currents in three-phase circuits are adequately expressed as instantaneous space vectors in abc
coordinates as depicted in Figure 6-C-3.
b axis axis
vb , ib v , i
2 / 3
v a , ia v , i
2 / 3
a axis axis
2 / 3
v c , ic
c axis
In a balanced three-phase system the abc space vectors are easily transformed into and coordinates
through the Clark transformation as follows:
va t
v t
C ´ vb t (6-C.3)
v t
vc t
Page 176
Modelling and Aggregation of Loads in Flexible Power Networks
ia t
i t
C ´ ib t (6-C.4)
i t ic t
where
va t , vb t and vc t are the instantaneous voltages in abc coordinates, respectively;
ia t , ib t and ic t are the instantaneous currents in abc coordinates, respectively;
v t and v t are the instantaneous voltages in coordinates, respectively;
i t and i t are the instantaneous currents in coordinates, respectively;
C is the Clark transformation given by
1 1
1 2
2
2
C ´ (6-C.5)
3
3 3
0 2
2
So, as described in [241], the instantaneous active and reactive powers are defined as:
p t v t v t i t
´ (6-C.6)
q t v t v t i t
where
p t is the instantaneous active power in W ;
q t is the instantaneous reactive power in VAr .
In systems with sinusoidal balanced voltages and currents, the average value of q t is equal to the
conventional reactive power and the instantaneous active power, p t , is always equal to the conventional
active power [158]. Thus, from equation (6-C.6) it is possible to obtain the currents reference signals to control
the PWM-CSI depicted in Figure 6-C-2 as follows
ia* t
* i t
ib t C ´ i t
1
(6-C.7)
ic* t
Where
1
i t v t v t Pm
´ (6-C.8)
i t v t v t Qref
Page 177
Modelling and Aggregation of Loads in Flexible Power Networks
And
1 0
3 1 3
C 1 ´ (6-C.9)
2 2 2
1 3
2
2
The references of currents, ia* , ib* and ic* are calculated instantaneously without any time delay by using the
boundary bus instantaneous voltages and both active and reactive power values. The procedure described
above was implemented in a Simulink S-function coded in MatLab m-file, following the scheme presented in
Figure 6-C-4.
va
VD V D Q
v
VD p.u.
abc
2 vb
´ Vbase v
VQ p.u. 3 vc
VQ V abc
t
ia*
Pm W i I D A I DR p.u.
Pm p.u. abc
1
v v ib* 1
S base v
Qref p.u. v ic* 2 ´ I base
D Q I A I QR p.u.
Qref VAr
i abc
Q
t
Page 178
Modelling and Aggregation of Loads in Flexible Power Networks
Page 179
Modelling and Aggregation of Loads in Flexible Power Networks
Bibliography References
[1] IEEE Task Force on Load Representation for Dynamic Performance, “Bibliography on load models for power flow and
dynamic performance simulation,” IEEE Transactions on Power Systems, vol. 10, no. 1, pp. 523–538, 1995.
[2] IEEE Task Force on Load Representation for Dynamic Performance, “Standard load models for power flow and
dynamic performance simulation,” IEEE Transactions on Power Systems, vol. 10, no. 3, pp. 1302–1313, 1995.
[3] IEEE Task Force on Load Representation for Dynamic Performance, “Load representation for dynamic performance
analysis (of power systems),” IEEE Transactions on Power Systems, vol. 8, no. 2, pp. 472–482, May 1993.
[4] T. Y. J. Lem and R. T. H. Alden, “Comparison of experimental and aggregate induction motor responses,” IEEE
Transactions on Power Systems, vol. 9, no. 4, pp. 1895–1900, 1994.
[5] T. Shimada, S. Agematsu, T. Shoji, T. Funabashi, H. Otoguro, and A. Ametani, “Combining power system load models at
a busbar,” in 2000 Power Engineering Society Summer Meeting (Cat. No.00CH37134), vol. 1, pp. 383–388.
[6] P. Pillay, S. M. A. Sabur, and M. M. Haq, “A model for induction motor aggregation for power system studies,” Electric
Power Systems Research, vol. 42, no. 3, pp. 225–228, Sep. 1997.
[7] K. W. Louie, “A new approach to compose load devices in electric power systems,” International Journal of Applied
Science and Engineering, vol. 2, no. 2, pp. 197–210, 2004.
[8] J. Y. Lim, P. S. Ji, A. Ozdemir, and C. Singh, “Component-based load modeling including capacitor banks,” in 2001
Power Engineering Society Summer Meeting. Conference Proceedings (Cat. No.01CH37262), vol. 2, pp. 1199–1204.
[9] O. M. Fahmy, A. S. Attia, and M. A. L. Badr, “A novel analytical model for electrical loads comprising static and
dynamic components,” Electric Power Systems Research, vol. 77, no. 10, pp. 1249–1256, Aug. 2007.
[10] W.-S. Kao, C.-J. J. Lin, C.-T. Huang, Y.-T. T. Chen, and C.-Y. Chiou, “Comparison of simulated power system dynamics
applying various load models with actual recorded data,” IEEE Transactions on Power Systems, vol. 9, no. 1, pp. 248–
254, 1994.
[11] K. Morison and H. Hamadani, “Practical issues in load modeling for voltage stability studies,” in 2003 IEEE Power
Engineering Society General Meeting (IEEE Cat. No.03CH37491), pp. 1392–1397.
[12] K. Louie, A novel aggregate load model for studying power system dynamic behaviour. 2003, pp. 563 – 568.
[13] I. Kasikci, M. Darwish, and P. Mehta, “Analytical load model for power system components,” in Proceedings of the IEEE
1999 International Conference on Power Electronics and Drive Systems. PEDS’99 (Cat. No.99TH8475), 1999, pp. 673–
676 vol.2.
[14] W.-S. Kao, “The effect of load models on unstable low-frequency oscillation damping in Taipower system experience
w/wo power system stabilizers,” IEEE Transactions on Power Systems, vol. 16, no. 3, pp. 463–472, 2001.
[15] S. Talebi Rafsanjani, M. Sedighizadeh, M. Kalantar, and M.R. Dadashzadeh, “Load Modelling for Power Flow and
Transient Stability Computer Studies with LOADMOD Software,” Jun. 2004.
[16] D. Brooks, A. Gaikwad, and P. Pourbeik, “Comprehensive Load Modeling for System Planning Studies,” Palo Alto, USA,
2009.
[17] D. Kosterev, A. Meklin, J. Undrill, B. Lesieutre, W. Price, D. Chassin, R. Bravo, and S. Yang, “Load modeling in power
system studies: WECC progress update,” in 2008 IEEE Power and Energy Society General Meeting - Conversion and
Delivery of Electrical Energy in the 21st Century, 2008, pp. 1–8.
[18] J. V. Milanović, “On unreliability of exponential load models,” Electric Power Systems Research, vol. 49, no. 1, pp. 1–9,
Feb. 1999.
[19] E. Bompard, E. Carpaneto, G. Chicco, and R. Napoli, “Contribution of loads with thermostatic control to the long-term
dynamics,” in PowerTech Budapest 99. Abstract Records. (Cat. No.99EX376), p. 114.
[20] S. Z. Djokic, A. J. Collin, and C. E. Cresswell, “The Future of Residential Lighting: Shift from Incandescent to CFL to LED
light sources,” in Illuminating Engineering Society of North America Annual Conference, 2009.
[21] S. Z. Djokic, C. E. Cresswell, and A. J. Collin, “Comparison of electrical characteristics of low-power and high-power CFL
and LED light sources,” in Proc. 2009 IES Annual Conference, 2009.
[22] C. Cresswell and S. Z. Djokic, “Representation of Directly Connected and Drive-controlled Induction Motors. Part 1:
Single-phase Load Models; Part 2 Three-phase Load Models,” in 18th International Conference on Electrical Machines,
2008.
[23] C. Cresswell and S. Z. Djokic, “Steady State Models of Low-Energy Consumption Light Sources,” in 16th Power System
Computation Conference, 2008.
[24] C. Cresswell, S. Djokic, and S. Munshi, “Analytical Modeling of Adjustable Speed Drive Load for Power System
Studies,” in 2007 IEEE Lausanne Power Tech, 2007, pp. 1899–1904.
Page 180
Modelling and Aggregation of Loads in Flexible Power Networks
[25] R. J. O’Keefe, R. P. Schulz, and F. B. Bhatt, “Improved representation of generator and load dynamics in the study of
voltage limited power system operations,” IEEE Transactions on Power Systems, vol. 12, no. 1, pp. 304–314, 1997.
[26] A. M. Stankovic and B. C. Lesieutre, “A probabilistic approach to aggregate induction machine modeling,” IEEE
Transactions on Power Systems, vol. 11, no. 4, pp. 1983–1989, 1996.
[27] A. M. Stankovic and B. C. Lesieutre, “Parametric variations in dynamic models of induction machine clusters,” IEEE
Transactions on Power Systems, vol. 12, no. 4, pp. 1549–1554, 1997.
[28] M. Reformat, D. Woodford, R. Wachal, and N. J. Tarko, “Non-linear load modeling for simulations in time domain,” in
8th International Conference on Harmonics and Quality of Power. Proceedings (Cat. No.98EX227), vol. 1, pp. 506–
510.
[29] A. P. Alves da Silva, C. Ferreira, and G. L. Torres, “Dynamic load modeling based on a nonparametric ANN,” in
Proceedings of International Conference on Intelligent System Application to Power Systems, pp. 55–59.
[30] A. P. Alves da Silva, C. Ferreira, A. C. Zambroni de Souza, and G. Lambert-Torres, “A new constructive ANN and its
application to electric load representation,” IEEE Transactions on Power Systems, vol. 12, no. 4, pp. 1569–1575, 1997.
[31] F. Nozari, M. D. Kankam, and W. W. Price, “Aggregation of Induction Motors for Transient Stability Load Modeling,”
IEEE Transactions on Power Systems, vol. 2, no. 4, pp. 1096–1103, Nov. 1987.
[32] A. J. Collin, S. Z. Djokic, H. F. Thomas, and J. Meyer, “Modelling of electric vehicle chargers for power system analysis,”
in 11th International Conference on Electrical Power Quality and Utilisation, 2011, pp. 1–6.
[33] S. Z. Djokic and I. Papic, “Smart Grid Implementation of Demand Side Management and Micro-Generation,”
International Journal of Energy Optimization and Engineering, vol. 1, no. 2, pp. 1–19, 2012.
[34] A. J. Collin, I. Hernando-Gil, J. L. Acosta, and S. Z. Djokic, “An 11 kV steady state residential aggregate load model.
Part 1: Aggregation methodology,” in 2011 IEEE Trondheim PowerTech, 2011, pp. 1–8.
[35] A. J. Collin, J. L. Acosta, I. Hernando-Gil, and S. Z. Djokic, “An 11 kV steady state residential aggregate load model.
Part 2: Microgeneration and demand-side management,” in 2011 IEEE Trondheim PowerTech, 2011, pp. 1–8.
[36] O. H. Abdalla, M. E. Bahgat, A. M. Serag, and M. A. El-Sharkawi, “Dynamic load modelling and aggregation in power
system simulation studies,” in 2008 12th International Middle-East Power System Conference, 2008, pp. 270–276.
[37] D. Salomonsson and A. Sannino, “Load modelling for steady-state and transient analysis of low-voltage DC systems,”
IET Electric Power Applications, vol. 1, no. 5, p. 690, 2007.
[38] M. Rylander, W. M. Grady, A. Arapostathis, and E. J. Powers, “Power Electronic Transient Load Model for Use in
Stability Studies of Electric Power Grids,” IEEE Transactions on Power Systems, vol. 25, no. 2, pp. 914–921, May 2010.
[39] F. Puyleart and S. Yang, “Load component database of household appliances and small office equipment,” in 2008
IEEE Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st Century,
2008, pp. 1–5.
[40] C.-S. S. Chen, T.-H. Wu, C.-C. C. Lee, and Y.-M. M. Tzeng, “The application of load models of electric appliances to
distribution system analysis,” IEEE Transactions on Power Systems, vol. 10, no. 3, pp. 1376–1382, 1995.
[41] A. M. Gole, A. Keri, C. Kwankpa, E. W. Gunther, H. W. Dommel, I. Hassan, J. R. Marti, J. A. Martinez, K. G. Fehrle, L.
Tang, M. F. McGranaghan, O. B. Nayak, P. F. Ribeiro, R. Iravani, and R. Lasseter, “Guidelines for modeling power
electronics in electric power engineering applications,” IEEE Transactions on Power Delivery, vol. 12, no. 1, pp. 505–
514, 1997.
[42] A. Mansoor, W. M. Grady, A. H. Chowdhury, and M. J. Samotyi, “An investigation of harmonics attenuation and
diversity among distributed single-phase power electronic loads,” IEEE Transactions on Power Delivery, vol. 10, no. 1,
pp. 467–473, 1995.
[43] A. J. Collin and S. Z. Djokic, “Voltage Disturbances and Inrush Current of DC Power Supplies,” Renewable Energy and
Power Quality Journal, vol. 8, 2010.
[44] A. J. Collin, J. L. Acosta, B. P. Hayes, and S. Z. Djokic, “Component-based aggregate load models for combined power
flow and harmonic analysis,” in 7th Mediterranean Conference and Exhibition on Power Generation, Transmission,
Distribution and Energy Conversion (MedPower 2010), 2010, pp. 171–171.
[45] A. J. Collin, C. E. Cresswell, and S. Z. Djokic, “Harmonic cancellation of modern switch-mode power supply load,” in
Proceedings of 14th International Conference on Harmonics and Quality of Power - ICHQP 2010, 2010, pp. 1–9.
[46] G. Delille, “Participation of Storage in the Advanced Management of Power Systems. Organizational, Technical and
Economic Approaches in Distribution Grids,” The University of Lille, France, 2010.
[47] Z. Y. Dong, A.Borghetti, K.Yamashita, A. Gaikwad, P. Pourbeik, and J.V.Milanovic, “CIGRE WG C4.065
Recommendations on Measurement Based and Component Based Load Modelling Practice,” in CIGRE SC C4
Colloquium in Japan, 2012.
Page 181
Modelling and Aggregation of Loads in Flexible Power Networks
[48] L. Hajagos and B. Danai, “Laboratory Measurement of Modern Loads Subjected to Large Voltage Changes for Use in
Voltage Stability Studies,” 1996.
[49] I. R. Navarro, O. Samuelsson, and S. Lindahl, “Automatic determination of parameters in dynamic load models from
normal operation data,” in 2003 IEEE Power Engineering Society General Meeting (IEEE Cat. No.03CH37491), pp.
1375–1378.
[50] S. Ihara, M. Tani, and K. Tomiyama, “Residential load characteristics observed at KEPCO power system,” IEEE
Transactions on Power Systems, vol. 9, no. 2, pp. 1092–1101, May 1994.
[51] H.-D. Chiang, J.-C. Wang, C.-H. C.-T. Huang, Y.-T. Chen, and C.-H. C.-T. Huang, “Development of a dynamic ZIP-motor
load model from on-line field measurements,” International Journal of Electrical Power & Energy Systems, vol. 19, no.
7, pp. 459–468, Oct. 1997.
[52] K. Tomiyama, S. Ueoka, T. Takano, I. Iyoda, K. Matsuno, K. Temma, and J. J. Paserba, “Modeling of load during and
after system faults based on actual field data,” in 2003 IEEE Power Engineering Society General Meeting (IEEE Cat.
No.03CH37491), pp. 1385–1391.
[53] Y. Li, H.-D. Chiang, B.-K. Choi, Y.-T. Chen, D.-H. Huang, and M. G. Lauby, “Representative static load models for
transient stability analysis: development and examination,” IET Generation, Transmission & Distribution, vol. 1, no. 3, p.
422, 2007.
[54] H. Renmu, M. Jin, and D. J. Hill, “Composite Load Modeling via Measurement Approach,” IEEE Transactions on Power
Systems, vol. 21, no. 2, pp. 663–672, May 2006.
[55] W.-S. Kao, C.-T. Huang, and C.-Y. Chiou, “Dynamic load modeling in Taipower system stability studies,” IEEE
Transactions on Power Systems, vol. 10, no. 2, pp. 907–914, May 1995.
[56] A. Keyhani and G. T. Heydt, “Composite neural network load models for power system stability analysis,” in IEEE PES
Power Systems Conference and Exposition, 2004., pp. 1574–1578.
[57] T. Hiyama, M. Tokieda, W. Hubbi, and H. Andou, “Artificial neural network based dynamic load modeling,” IEEE
Transactions on Power Systems, vol. 12, no. 4, pp. 1576–1583, 1997.
[58] J.-C. C. Wang, H.-D. D. Chiang, C.-L. W. Chang, A.-H. H. Liu, C.-H. C.-Y. Y. Huang, and C.-H. C.-Y. Y. Huang,
“Development of a frequency-dependent composite load model using the measurement approach,” IEEE Transactions
on Power Systems, vol. 9, no. 3, pp. 1546–1556, 1994.
[59] B.-K. Choi, H.-D. Chiang, Y. Li, H. Li, Y.-T. Chen, D.-H. Huang, and M. G. Lauby, “Measurement-Based Dynamic Load
Models: Derivation, Comparison, and Validation,” IEEE Transactions on Power Systems, vol. 21, no. 3, pp. 1276–1283,
Aug. 2006.
[60] Y. A. Jebril, A. I. Ibrahim, S. A. Shaban, S. A. Al Dessi, K. Karoui, F. Depierreux, A. Szekut, F.Villella, I. Ibrahim, F.
Depieereux, and F. Villella, “Development of a detailed dynamic load model and implementation staged testing of
DEWA network,” in Cigré GCC Power, 2010.
[61] D. Karlsson and D. J. Hill, “Modelling and identification of nonlinear dynamic loads in power systems,” IEEE Transactions
on Power Systems, vol. 9, no. 1, pp. 157–166, 1994.
[62] Y. Baghzouz and C. Quist, “Composite load model derivation from recorded field data,” in IEEE Power Engineering
Society. 1999 Winter Meeting (Cat. No.99CH36233), 1999, pp. 713–718 vol.1.
[63] K. Lidija and S. Dobrivoje, “Load modeling in distribution networks,” Facta universitatis - series: Electronics and
Energetics, vol. 15, no. 3, pp. 419–427, 2002.
[64] Y. Liang, C. O. Nwankpa, R. Fischl, A. DeVito, and S. C. Readinger, “Dynamic reactive load model,” IEEE Transactions
on Power Systems, vol. 13, no. 4, pp. 1365–1372, 1998.
[65] EPRI, “Measurement-Based Load Modeling,” 2006.
[66] Y. Baghzouz and C. Quist, “Determination of static load models from LTC and capacitor switching tests,” in 2000 Power
Engineering Society Summer Meeting (Cat. No.00CH37134), vol. 1, pp. 389–394.
[67] W. Xu, E. Vaahedi, Y. Mansour, and J. Tamby, “Voltage stability load parameter determination from field tests on BC
Hydro’s system,” IEEE Transactions on Power Systems, vol. 12, no. 3, pp. 1290–1297, 1997.
[68] S. J. Ranade, A. Ellis, and J. Mechenbier, “The development of power system load models from measurements,” in 2001
IEEE/PES Transmission and Distribution Conference and Exposition. Developing New Perspectives (Cat.
No.01CH37294), vol. 1, pp. 201–206.
[69] E. Agneholm and J. E. Daalder, “Load recovery in the pulp and paper industry following a disturbance,” IEEE
Transactions on Power Systems, vol. 15, no. 2, pp. 831–837, May 2000.
[70] Q. Ai, D. Gu, and C. Chen, “New Load Modeling Approaches Based on Field Tests for Fast Transient Stability
Calculations,” IEEE Transactions on Power Systems, vol. 21, no. 4, pp. 1864–1873, Nov. 2006.
Page 182
Modelling and Aggregation of Loads in Flexible Power Networks
[71] M. M. Begovic and R. Q. Mills, “Load identification and voltage stability monitoring,” IEEE Transactions on Power
Systems, vol. 10, no. 1, pp. 109–116, 1995.
[72] P. Ju, E. Handschin, and D. Karlsson, “Nonlinear dynamic load modelling: model and parameter estimation,” IEEE
Transactions on Power Systems, vol. 11, no. 4, pp. 1689–1697, 1996.
[73] V. Knyazkin, C. Canizares, and L. Soder, “On the parameter estimation of linear models of aggregate power system
loads,” in 2003 IEEE Power Engineering Society General Meeting (IEEE Cat. No.03CH37491), pp. 2392–2397.
[74] A. Qian and G. B. Shrestha, “An ANN-based load model for fast transient stability calculations,” Electric Power Systems
Research, vol. 76, no. 4, pp. 217–227, Jan. 2006.
[75] K. Srinivasan and C. Lafond, “Statistical analysis of load behaviour parameters at four major loads,” in Proceedings of
IEEE/PES Transmission and Distribution Conference, pp. 40–45.
[76] Z. Huang, J. Phillips, B. Lesieutre, D. Kosterev, S. Yang, M. Hoffman, A. Ellis, O. Ciniglio, A. Meklin, R. Hartwell, B. Wong,
P. Pourbeik, A. Gaikwad, A. Maitra, D. Brooks, N. Lu, and D. Hammerstrom, “Load Monitoring: CEC/LMTF Load
Research Program,” 2007.
[77] L. Korunović and D. Stojanović, “Load Model Parameters on Low and Middle Voltage in Distribution Networks,”
Elektroprivreda, vol. 55, no. 2, pp. 46–56, 2002.
[78] A. M. Gaikwad, R. J. Bravo, D. Kosterev, S. Yang, A. Maitra, P. Pourbeik, B. Agrawal, R. Yinger, and D. Brooks, “Results
of residential air conditioner testing in WECC,” in 2008 IEEE Power and Energy Society General Meeting - Conversion
and Delivery of Electrical Energy in the 21st Century, 2008, pp. 1–9.
[79] K. Karoui, Y. A. Jebril, A. I. Ibrahim, S. A. Shaban, and S. A. Al Dessi, “Load model development based on monitoring
and laboratory staged testing,” in 2011 IEEE GCC Conference and Exhibition (GCC), 2011, pp. 661–664.
[80] M. Kostić, “The Influence of network voltage on loads and rational consumption of electrical energy,” Beograd, Serbia,
1997.
[81] Z. Hongbin, H. Renmu, L. Li, and Z. Jian, “Application of different load models for the transient stability calculation,” in
Proceedings. International Conference on Power System Technology, vol. 4, pp. 2014–2018.
[82] C.-J. Lin, A. Y.-T. Chen, C.-Y. Chiou, C.-H. Huang, H.-D. Chiang, J.-C. Wang, and L. Fekih-Ahmed, “Dynamic load models
in power systems using the measurement approach,” IEEE Transactions on Power Systems, vol. 8, no. 1, pp. 309–315,
1993.
[83] J. W. O’Sullivan and M. J. O’Malley, “Identification and validation of dynamic global load model parameters for use
in power system frequency simulations,” IEEE Transactions on Power Systems, vol. 11, no. 2, pp. 851–857, May 1996.
[84] F. T. Dai, “Development of a dynamic power system load model,” in Seventh International Conference on AC and DC
Transmission, 2001, vol. 2001, pp. 344–349.
[85] B. Jianhua, G. Jingde, S. Shande, and Z. Shouzhen, “PQ-coupled difference equation model for power system load,” in
Proceedings of TENCON ’93. IEEE Region 10 International Conference on Computers, Communications and Automation,
pp. 170–174.
[86] Q. H. WU, J. Y. WEN, K. I. NUTTALL, D. W. SHIMMIN, and S. J. CHENG, “Power System Load Modeling by
Evolutionary Computation Based on System Measurements,” Electric Power Components and Systems, vol. 31, no. 5, pp.
423–439, May 2003.
[87] R. J. Thomas, “Power system dynamic load modeling using artificial neural networks,” IEEE Transactions on Power
Systems, vol. 9, no. 4, pp. 1868–1874, 1994.
[88] R. C. Leou, Z. L. Gaing, C. N. Lu, B. S. Chang, and C. L. Cheng, “Distribution system feeder cold load pickup model,”
Electric Power Systems Research, vol. 36, no. 3, pp. 163–168, Mar. 1996.
[89] L. T. M. Mota and A. A. Mota, “Load modeling at electric power distribution substations using dynamic load parameters
estimation,” International Journal of Electrical Power & Energy Systems, vol. 26, no. 10, pp. 805–811, Dec. 2004.
[90] EPRI, “Advanced Load Modeling – Entergy Pilot Study,” 2004.
[91] C. Lafond, K. Srinivasan, and M. Dusseault, “A dynamic model for paper mill load,” in Proceedings 1995 Canadian
Conference on Electrical and Computer Engineering, vol. 2, pp. 1077–1082.
[92] Q. H. Wu, “Measurement based power system load modelling by learning,” in UKACC International Conference on
Control (CONTROL ’98), 1998, vol. 1998, pp. 871–876.
[93] J. Y. Wen, L. Jiang, Q. H. Wu, and S. J. Cheng, “Power system load modeling by learning based on system
measurements,” IEEE Transactions on Power Delivery, vol. 18, no. 2, pp. 364–371, Apr. 2003.
[94] H. Yan, Z. Zabar, D. Czarkowski, L. Birenbaum, E. Levi, and J. Hajagos, “Experimental test of a load model in the
presence of harmonics,” IEE Proceedings - Generation, Transmission and Distribution, vol. 146, no. 2, p. 186, 1999.
Page 183
Modelling and Aggregation of Loads in Flexible Power Networks
[95] W. Jin, L. Xinran, S. Sheng, and Z. Xiangjun, “An improved algorithm for aggregate load modeling in voltage stability
studies,” in 2004 IEEE International Conference on Electric Utility Deregulation, Restructuring and Power Technologies.
Proceedings, 2004, pp. 421–426 Vol.1.
[96] P. Pourbeik and B. Agrawal, “A hybrid model for representing air-conditioner compressor motor behaviour in power
system studies,” in 2008 IEEE Power and Energy Society General Meeting - Conversion and Delivery of Electrical
Energy in the 21st Century, 2008, pp. 1–8.
[97] A. Maitra, A. Gaikwad, P. Pourbeik, and D. Brooks, “Load model parameter derivation using an automated algorithm
and measured data,” in 2008 IEEE Power and Energy Society General Meeting - Conversion and Delivery of Electrical
Energy in the 21st Century, 2008, pp. 1–7.
[98] EPRI, “Load Modeling Using a Measurement Based Approach: Phase-2”, EPRI Product ID 1016255, Palo Alto, USA,
2007.[PP1]
[99] V. Knyazkin, C. A. Canizares, and L. H. Soder, “On the Parameter Estimation and Modeling of Aggregate Power
System Loads,” IEEE Transactions on Power Systems, vol. 19, no. 2, pp. 1023–1031, May 2004.
[100] J. Deuse, J. Dubois, R. Fanna, and I. Hamza, “EWR under voltage load shedding scheme,” IEEE Transactions on Power
Systems, vol. 12, no. 4, pp. 1446–1454, 1997.
[101] I. A. Hamzah and J. A. Yasin, “Static var compensators (SVC) reouired to solve the problem of delayed voltage
recovery following faults in the power system of the Saudi Electricity Company, Western Region (SEC-WR),” in 2003
IEEE Bologna Power Tech Conference Proceedings,, vol. 4, pp. 487–494.
[102] C. S. Chen, J. C. Hwang, Y. M. Tzeng, C. W. Huang, and M. Y. Cho, “Determination of customer load characteristics by
load survey system at Taipower,” IEEE Transactions on Power Delivery, vol. 11, no. 3, pp. 1430–1436, Jul. 1996.
[103] I. R. Navarro, O. Samuelsson, and S. Lindahl, “Influence of normalization in dynamic reactive load models,” IEEE
Transactions on Power Systems, vol. 18, no. 2, pp. 972–973, May 2003.
[104] T. H. Ortmeyer, “Short term residential load data: measurement and analysis,” in 10th International Conference on
Harmonics and Quality of Power. Proceedings (Cat. No.02EX630), vol. 2, pp. 430–434.
[105] J. V. Milanovic and N. Jenkins, “The influence of measurement delays on estimated load model parameters,” in Ninth
International Conference on Harmonics and Quality of Power. Proceedings (Cat. No.00EX441), vol. 1, pp. 324–328.
[106] R.-F. F. Chang, R.-C. C. Leou, and C.-N. N. Lu, “Distribution transformer load modeling using load research data,” IEEE
Transactions on Power Delivery, vol. 17, no. 2, pp. 655–661, Apr. 2002.
[107] L. M. Korunovic, D. P. Stojanovic, and J. V. Milanovic, “Identification of static load characteristics based on
measurements in medium-voltage distribution network,” IET Generation, Transmission & Distribution, vol. 2, no. 2, p.
227, 2008.
[108] D. P. Stojanović, L. M. Korunović, and J. V. Milanović, “Dynamic load modelling based on measurements in medium
voltage distribution network,” Electric Power Systems Research, vol. 78, no. 2, pp. 228–238, Feb. 2008.
[109] B. Khodabakhchian and G.-T. Vuong, “Modeling a mixed residential-commercial load for simulations involving large
disturbances,” IEEE Transactions on Power Systems, vol. 12, no. 2, pp. 791–796, May 1997.
[110] E. Agneholm and J. Daalder, “Cold load pick-up of residential load,” IEE Proceedings - Generation, Transmission and
Distribution, vol. 147, no. 1, p. 44, 2000.
[111] Q. H. Wu and D. W. Shimmin, “Measurement based power system load modeling using a population diversity genetic
algorithm,” in POWERCON ’98. 1998 International Conference on Power System Technology. Proceedings (Cat.
No.98EX151), vol. 1, pp. 771–775.
[112] A. Dwyer, R. E. Nielsen, J. Stangl, and N. S. Markushevich, “Load to voltage dependency tests at B.C. Hydro,” IEEE
Transactions on Power Systems, vol. 10, no. 2, pp. 709–715, May 1995.
[113] L. M. Korunović, D. P. Stojanović, and M. P. Košarac, “Load Parameters in Low Voltage Distribution Networks,” in Proc.
XXXIX International Scientific Conference on Information, Communication and Energy Systems and Technologies (ICEST),
2004.
[114] M. Košarac, L. Korunović, and D. Stojanović, “Equivalent Load Parameters on 0.4kV Voltage Level,” in Proc. Regional
Conference and Exhibition on Electricity Distribution, 2004.
[115] L. Korunović and D. Stojanović, “Determination of Dynamic Load Model Parameters,” in Proc. Second Regional
Conference & Exhibition on Electricity Distribution, 2006.
[116] L. M. Korunovic and D. P. Stojanovic, “The effects of normalization of static load characteristics,” in 2009 IEEE Bucharest
PowerTech, 2009, pp. 1–6.
[117] L. Korunović, B. Nikolić, D. Nikolić, and M. Petronijević, “Load Modelling by using Normal Operation Data,” in Proc.
XLVI International Scientific Conference on Information, Communication and Energy Systems and Technologies (ICEST),
2011.
Page 184
Modelling and Aggregation of Loads in Flexible Power Networks
[118] J. Ma, X. Zheng, J. Xu, Y. Tang, and Z.-Y. Dong, “Instantaneous Measurements based Load Modeling,” in CIGRE SC C4
Colloquium in Japan, 2012.
[119] K. Yamashita, S. Z. Djokić, F. Villella, and J. V. Milanovic, “Self-disconnection and Self-recovery of Loads Due to
Voltage Sags and Short Interruptions,” in CIGRE Symposium in Portugal, 2013.
[120] B. C. Lesieutre, P. W. Sauer, and A. Pai, “Development and comparative study of induction machine based dynamic P,
Q load models,” IEEE Transactions on Power Systems, vol. 10, no. 1, pp. 182–191, 1995.
[121] R. Balanathan, N. C. Pahalawaththa, and U. D. Annakkage, “Modelling induction motor loads for voltage stability
analysis,” International Journal of Electrical Power & Energy Systems, vol. 24, no. 6, pp. 469–480, Aug. 2002.
[122] A. Borghetti, R. Caldon, A. Mari, and C. A. Nucci, “On dynamic load models for voltage stability studies,” IEEE
Transactions on Power Systems, vol. 12, no. 1, pp. 293–303, 1997.
[123] T. J. Overbye, “Effects of load modelling on analysis of power system voltage stability,” International Journal of
Electrical Power & Energy Systems, vol. 16, no. 5, pp. 329–338, Oct. 1994.
[124] K.-H. Tseng, W.-S. Kao, and J.-R. Lin, “Load model effects on distance relay settings,” IEEE Transactions on Power
Delivery, vol. 18, no. 4, pp. 1140–1146, Oct. 2003.
[125] A. Oonsivilai and M. E. El-Hawary, “Power system dynamic load modeling using adaptive-network-based fuzzy
inference system,” in Engineering Solutions for the Next Millennium. 1999 IEEE Canadian Conference on Electrical and
Computer Engineering (Cat. No.99TH8411), vol. 3, pp. 1217–1222.
[126] P. K. SATPATHY, “Impact of Various Load Models in Power System Bifurcation Analysis,” Electric Power Components
and Systems, vol. 31, no. 7, pp. 653–669, Jul. 2003.
[127] J. A. Martinez and J. Martin-Arnedo, “Advanced load models for voltage sag studies in distribution networks,” in IEEE
Power Engineering Society General Meeting, 2004., vol. 2, pp. 614–619.
[128] S. Arnborg, G. Andersson, D. J. Hill, and I. A. Hiskens, “On influence of load modelling for undervoltage load shedding
studies,” IEEE Transactions on Power Systems, vol. 13, no. 2, pp. 395–400, May 1998.
[129] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of electric machinery and drive systems, Second. New York: Wiley-
IEEE Press, 2002.
[130] C. A. Canizares, “On bifurcations, voltage collapse and load modeling,” IEEE Transactions on Power Systems, vol. 10,
no. 1, pp. 512–522, 1995.
[131] S. Z. Zhu, J. H. Zheng, S. D. Shen, and G. M. Luo, “Effect of load modeling on voltage stability,” in 2000 Power
Engineering Society Summer Meeting (Cat. No.00CH37134), vol. 1, pp. 395–400.
[132] Y. Kataoka, “State variable description of dynamic load models in electrical power system analysis,” IEEE Transactions
on Power Systems, vol. 10, no. 1, pp. 149–157, 1995.
[133] A. M. Azmy, I. Erlich, and P. Sowa, “Artificial neural network-based dynamic equivalents for distribution systems
containing active sources,” IEE Proceedings - Generation, Transmission and Distribution, vol. 151, no. 6, p. 681, 2004.
[134] J. . Wen, Q. . Wu, K. . Nuttall, D. . Shimmin, and S. . Cheng, “Construction of power system load models and network
equivalence using an evolutionary computation technique,” International Journal of Electrical Power & Energy Systems,
vol. 25, no. 4, pp. 293–299, May 2003.
[135] D. M. Vilathgamuwa and H. M. Wijekoon, “Representation of power system load dynamics with ANN for real-time
applications,” in 2003 IEEE Power Engineering Society General Meeting (IEEE Cat. No.03CH37491), pp. 1665–1672.
[136] M. Bostanci, J. Koplowitz, and C. W. Taylor, “Identification of power system load dynamics using artificial neural
networks,” IEEE Transactions on Power Systems, vol. 12, no. 4, pp. 1468–1473, 1997.
[137] T. Cutsem, Q. Renoy, and D. Lefebvre, “Modelling the short-term and long-term aggregate response of multiple loads
fed through a sub-transmission network,” in 2006 IEEE PES Power Systems Conference and Exposition, 2006, pp. 569–
575.
[138] M. Akbaba, M. Taleb, and A. Rumeli, “Improved estimation of induction machine parameters,” Electric Power Systems
Research, vol. 34, no. 1, pp. 65–73, Jul. 1995.
[139] Q. Liu, Y. Chen, and D. Duan, “The load modeling and parameters identification for voltage stability analysis,” in
Proceedings. International Conference on Power System Technology, vol. 4, pp. 2030–2033.
[140] P. Ju, E. Handschin, Z. N. Wei, and U. Schlucking, “Sequential parameter estimation of a simplified induction motor load
model,” IEEE Transactions on Power Systems, vol. 11, no. 1, pp. 319–324, 1996.
[141] S. A. Soliman, N. H. Abbasy, and M. E. El-Hawary, “Frequency domain modelling and identification of nonlinear loads
using a least error squares algorithm,” Electric Power Systems Research, vol. 40, no. 1, pp. 1–6, Jan. 1997.
[142] L. M. K. and D. P. Stojanović and D. P. Stojanović, “Dynamic Load Modelling of Some Low Voltage Devices.”
Page 185
Modelling and Aggregation of Loads in Flexible Power Networks
[143] S. Z. Zhu, Z. Y. Dong, K. P. Wong, and Z. H. Wang, “Power system dynamic load identification and stability,” in
PowerCon 2000. 2000 International Conference on Power System Technology. Proceedings (Cat. No.00EX409), vol. 1,
pp. 13–18.
[144] L. Loron and G. Laliberté, “Application of the extended Kalman filter to parameters estimation of induction motors.”
[145] S. Ahmed-Zaid, C. W. Taylor, and D. J. Sobajic, “Input-output model identification of small and large induction motors
using multilayer feedforward neural networks,” in 1997 IEEE International Electric Machines and Drives Conference
Record, pp. MD1/4.1–MD1/4.3.
[146] L. Ljung, System Identification ToolboxTM 7 User’s Guide. The Mathworks, Inc, 2010.
[147] S. J. Ranade, D. R. Sagi, and R. Adapa, “Load Understanding and Model Development,” in 2005/2006 PES TD, pp.
1315–1319.
[148] K. Yamashita, M. Asada, and K. Yoshimura, “A development of dynamic load model parameter derivation method,” in
2009 IEEE Power & Energy Society General Meeting, 2009, pp. 1–8.
[149] M. H. Haque, “Load flow solution of distribution systems with voltage dependent load models,” Electric Power Systems
Research, vol. 36, no. 3, pp. 151–156, Mar. 1996.
[150] J. D. Weber, T. J. Overbye, and C. L. DeMarco, “Inclusion of price dependent load models in the optimal power flow,”
in Proceedings of the Thirty-First Hawaii International Conference on System Sciences, vol. 3, pp. 62–70.
[151] M. M. EL-Metwally, A. A. EL-Emary, and M. EL-Azab, “Effect of load characteristics on maximum power transfer limit
for HV compensated transmission lines,” International Journal of Electrical Power & Energy Systems, vol. 26, no. 7, pp.
467–472, Sep. 2004.
[152] J. a. De Leon and B. Kehrli, “The Modeling Requirements for Short-Term Voltage Stability Studies,” in 2006 IEEE PES
Power Systems Conference and Exposition, 2006, pp. 582–588.
[153] K. W. Louie and J. R. Marti, “A Method to Improve the Performance of Conventional Static Load Models,” IEEE
Transactions on Power Systems, vol. 20, no. 1, pp. 507–508, Feb. 2005.
[154] R. H. Fletcher and A. Saeed, “Integrating engineering and economic analysis for conservation voltage reduction,” in
IEEE Power Engineering Society Summer Meeting,, vol. 2, pp. 725–730.
[155] P. Kundur, Power System Stability and Control. New York: McGraw-Hill, Inc., 1994.
[156] K. Koyanagi, R. Yokoyama, Y. Fukuhara, and T. Komukai, “Effects of characteristics of load and generator on power
system dynamic stability,” in Universities Power Engineering Conference, 2007, pp. 683–686.
[157] L. Pereira, D. Kosterev, P. Mackin, D. Davies, and J. Undrill, “An interim dynamic induction motor model for stability
studies in the WSCC,” IEEE Transactions on Power Systems, vol. 17, no. 4, pp. 1108–1115, Nov. 2002.
[158] J. W. Shaffer, “Air conditioner response to transmission faults,” IEEE Transactions on Power Systems, vol. 12, no. 2, pp.
614–621, May 1997.
[159] I. Hiskens, “Significance of Load Modelling in Power System Dynamics,” in , Proc. X Symposium of Specialists in Electric
Operational and Expansion Planning, 2006.
[160] Y. V. Makarov, V. A. Maslennikov, and D. J. Hill, “Revealing loads having the biggest influence on power system small
disturbance stability,” IEEE Transactions on Power Systems, vol. 11, no. 4, pp. 2018–2023, 1996.
[161] J. V. Milanovic and I. A. Hiskens, “Effects of load dynamics on power system damping,” IEEE Transactions on Power
Systems, vol. 10, no. 2, pp. 1022–1028, May 1995.
[162] B. R. Korte, A. Manglick, and J. W. Howarth, “Interconenction damping performance tests on the south-east Australian
power grid,” in Colloquium of CIGRE study committee 38, 1993.
[163] R. J. Davy and I. A. Hiskens, “Lyapunov functions for multimachine power systems with dynamic loads,” IEEE Transactions
on Circuits and Systems I: Fundamental Theory and Applications, vol. 44, no. 9, pp. 796–812, 1997.
[164] I. A. Hiskens and J. V. Milanovic, “Locating dynamic loads which significantly influence damping,” IEEE Transactions on
Power Systems, vol. 12, no. 1, pp. 255–261, 1997.
[165] E. Z. Zhou, “Power oscillation flow study of electric power systems,” International Journal of Electrical Power & Energy
Systems, vol. 17, no. 2, pp. 143–150, Apr. 1995.
[166] Y. V. Makarov and D. J. Hill, “A general method for small signal stability analysis,” IEEE Transactions on Power Systems,
vol. 13, no. 3, pp. 979–985, 1998.
[167] Y. V. Makarov, D. H. Popovic, and D. J. Hill, “Stabilization of transient processes in power systems by an eigenvalue
shift approach,” IEEE Transactions on Power Systems, vol. 13, no. 2, pp. 382–388, May 1998.
[168] J. V. Milanovic, “Oscillatory interaction between synchronous generator and local voltage-dependent load,” IEE
Proceedings - Generation, Transmission and Distribution, vol. 142, no. 5, p. 473, 1995.
Page 186
Modelling and Aggregation of Loads in Flexible Power Networks
[169] E. H. Allen and M. D. Ilic, “Interaction of transmission network and load phasor dynamics in electric power systems,” IEEE
Transactions on Circuits and Systems I: Fundamental Theory and Applications, vol. 47, no. 11, pp. 1613–1620, 2000.
[170] D. J. Trudnowski and J. E. Dagle, “Effects of generator and static-load nonlinearities on electromechanical oscillations,”
IEEE Transactions on Power Systems, vol. 12, no. 3, pp. 1283–1289, 1997.
[171] R. K. Varma, R. M. Mathur, G. J. Rogers, and P. Kundur, “Modeling effects of system frequency variation in long-term
stability studies,” IEEE Transactions on Power Systems, vol. 11, no. 2, pp. 827–832, May 1996.
[172] L. Angquist, B. Lundin, and J. Samuelsson, “Power oscillation damping using controlled reactive power compensation-a
comparison between series and shunt approaches,” IEEE Transactions on Power Systems, vol. 8, no. 2, pp. 687–700,
May 1993.
[173] P Kundur et al, “294 T 622 Investigation of low frequency-area oscillation problems in large interconnected power
systems,” Toronto, Canada, 1993.
[174] F. P. de Mello and J. W. Feltes, “Voltage oscillatory instability caused by induction motor loads,” IEEE Transactions on
Power Systems, vol. 11, no. 3, pp. 1279–1285, 1996.
[175] T. Omata and K. Uemura, “Aspects of voltage responses of induction motor loads,” IEEE Transactions on Power Systems,
vol. 13, no. 4, pp. 1337–1344, 1998.
[176] T. Omata and K. Uemura, “Effects of series impedance on power system load dynamics,” IEEE Transactions on Power
Systems, vol. 14, no. 3, pp. 1070–1077, 1999.
[177] S. M. Deckmann and V. F. da Costa, “A power sensitivity model for electromechanical oscillation studies,” IEEE
Transactions on Power Systems, vol. 9, no. 2, pp. 965–971, May 1994.
[178] D. J. Hill, I. A. Hiskens, and D. H. Popovic, “Stability analysis of power system loads with recovery dynamics,”
International Journal of Electrical Power & Energy Systems, vol. 16, no. 4, pp. 277–286, Aug. 1994.
[179] R. Lind and D. Karlsson, “Distribution system modelling for voltage stability studies,” IEEE Transactions on Power
Systems, vol. 11, no. 4, pp. 1677–1682, 1996.
[180] C. Taylor, Power System Voltage Stability. New York: McGraw-Hill, Inc., 1994.
[181] D. Tasić, “The Influence of Static and Dynamic Load Characteristic on Voltage Collapse in Electrical Power System,”
University of Niš, 1996.
[182] D. J. Hill, “Nonlinear dynamic load models with recovery for voltage stability studies,” IEEE Transactions on Power
Systems, vol. 8, no. 1, pp. 166–176, 1993.
[183] M. A. Pai, P. W. Sauer, B. C. Lesieutre, and R. Adapa, “Structural stability in power systems-effect of load models,”
IEEE Transactions on Power Systems, vol. 10, no. 2, pp. 609–615, May 1995.
[184] W. Xu and Y. Mansour, “Voltage stability analysis using generic dynamic load models,” IEEE Transactions on Power
Systems, vol. 9, no. 1, pp. 479–493, 1994.
[185] S.-C. Chen, “The effects of load models on power system stabilizers and static VAr compensators for the Taiwan power
system,” in Proceedings of EMPD ’98. 1998 International Conference on Energy Management and Power Delivery (Cat.
No.98EX137), vol. 1, pp. 298–303.
[186] C. P. Gupta, R. K. Varma, and S. C. Srivastava, “A method to determine closest Hopf bifurcation in power systems
considering exciter and load dynamics,” in Proceedings of EMPD ’98. 1998 International Conference on Energy
Management and Power Delivery (Cat. No.98EX137), vol. 1, pp. 293–297.
[187] V. Stewart and E. H. Camm, “Modeling of stalled motor loads for power system short-term voltage stability analysis,”
in IEEE Power Engineering Society General Meeting, 2005, pp. 1252–1257.
[188] P. Pourbeik and D. Wang, “Load modeling in voltage stability studies,” in IEEE Power Engineering Society General
Meeting, 2005, pp. 23–30.
[189] R. Aresi, B. Delfino, G. B. Denegri, M. Invernizzi, and S. Massucco, “Dynamic equivalents of tap-changing-under-load
transformers in a transmission system,” IEE Proceedings - Generation, Transmission and Distribution, vol. 146, no. 1, p.
45, 1999.
[190] A. Borghetti, R. Caldon, and C. A. Nucci, “Generic dynamic load models in long-term voltage stability studies,”
International Journal of Electrical Power & Energy Systems, vol. 22, no. 4, pp. 291–301, May 2000.
[191] G. B. Denegri, M. Invernizzi, and F. Milano, “Synthesis of an equivalent dynamic model for load areas with LTC
transformers,” in 2001 IEEE/PES Transmission and Distribution Conference and Exposition. Developing New Perspectives
(Cat. No.01CH37294), pp. 641–646.
[192] P. Ju and X. Y. Zhou, “Dynamic equivalents of distribution systems for voltage stability studies,” IEE Proceedings -
Generation, Transmission and Distribution, vol. 148, no. 1, p. 49, 2001.
[193] C. D. Vournas and G. A. Manos, “Modelling of stalling motors during voltage stability studies,” IEEE Transactions on
Power Systems, vol. 13, no. 3, pp. 775–781, 1998.
Page 187
Modelling and Aggregation of Loads in Flexible Power Networks
[194] D. Lee Hau Aik and G. Andersson, “Influence of load characteristics on the power/voltage stability of HVDC systems. II.
Stability margin sensitivity,” IEEE Transactions on Power Delivery, vol. 13, no. 4, pp. 1437–1444, 1998.
[195] D. Lee Hau Aik and G. Andersson, “Influence of load characteristics on the power/voltage stability of HVDC systems. I.
Basic equations and relationships,” IEEE Transactions on Power Delivery, vol. 13, no. 4, pp. 1437–1444, 1998.
[196] R. B. L. Guedes, A. C. P. Martins, L. F. C. Alberto, and N. G. Bretas, “An extended energy function for voltage collapse
analysis considering voltage dependent load models,” in 2003 IEEE Bologna Power Tech Conference Proceedings,, vol.
1, pp. 466–471.
[197] M. Z. El-Sadek, M. M. Dessouky, G. A. Mahmoud, and W. I. Rashed, “Load representation for steady-state voltage
stability studies,” Electric Power Systems Research, vol. 43, no. 3, pp. 187–195, Dec. 1997.
[198] J. A. Diaz de Leon and C. W. Taylor, “Understanding and solving short-term voltage stability problems,” in IEEE Power
Engineering Society Summer Meeting,, vol. 2, pp. 745–752.
[199] D. H. Popovic, D. J. Hill, and C. J. Parker, “Voltage security enhancement via coordinated control,” IEEE Transactions on
Power Systems, vol. 16, no. 1, pp. 127–135, 2001.
[200] S. Sivanagaraju, N. Sreenivasulu, and M. Vijayakumar, “Voltage stability for radial distribution networks,” The
Institution of Engineers (India), J. Electr. Eng., vol. 84, pp. 166–172, 2003.
[201] A. Gebreselassie and J. H. Chow, “Investigation of the effects of load models and generator voltage regulators on
voltage stability,” International Journal of Electrical Power & Energy Systems, vol. 16, no. 2, pp. 83–89, Apr. 1994.
[202] G. K. Morison, B. Gao, and P. Kundur, “Voltage stability analysis using static and dynamic approaches,” IEEE
Transactions on Power Systems, vol. 8, no. 3, pp. 1159–1171, 1993.
[203] Y. G. Zeng, “Voltage stability analysis considering dynamic load model,” in APSCOM-97. International Conference on
Advances in Power System Control, Operation and Management, 1997, vol. 1997, pp. 396–401.
[204] B. Delfino, G. B. Denegri, M. Invernizzi, and A. Morini, “Voltage stability of power systems: links between static and
dynamic approaches,” in Proceedings of 8th Mediterranean Electrotechnical Conference on Industrial Applications in
Power Systems, Computer Science and Telecommunications (MELECON 96), vol. 2, pp. 854–858.
[205] R. Balanathan, N. C. Pahalawaththa, U. D. Annakkage, and P. W. Sharp, “Undervoltage load shedding to avoid
voltage instability,” IEE Proceedings - Generation, Transmission and Distribution, vol. 145, no. 2, p. 175, 1998.
[206] I. A. Hiskens and J. V. Milanovic, “Load modelling in studies of power system damping,” IEEE Transactions on Power
Systems, vol. 10, no. 4, pp. 1781–1788, Nov. 1995.
[207] J. V. Milanović and I. A. Hiskens, “Effects of dynamic load model parameters on damping of oscillations in power
systems,” Electric Power Systems Research, vol. 33, no. 1, pp. 53–61, Apr. 1995.
[208] J. V. Milanovic and I. A. Hiskens, “Influence of load dynamics on power system oscillations,” in National Conference
Publication - Institution of Engineers, 1994, pp. 171–176.
[209] J. V. Milanović, “The Influence of Loads on Power System Electromechanical Oscillations,” The University of Newcastle,
Australia, 1996.
[210] A. Bendre, D. Divan, W. Kranz, and W. E. Brumsickle, “Are voltage sags destroying equipment?,” IEEE Industry
Applications Magazine, vol. 12, no. 4, pp. 12–21, Jul. 2006.
[211] K. Yamashita and O. Sakamoto, “A study on dynamic behaviour of load supply system including synchronous
generators with and without load drop,” in IEEE PES General Meeting, 2010, pp. 1–7.
[212] G. L. Chinn, “Modeling Stalled Induction Motors,” in 2005/2006 PES TD, pp. 1325–1328.
[213] R. Balanathan, “Influence of induction motor modelling for undervoltage load shedding studies,” in IEEE/PES
Transmission and Distribution Conference and Exhibition, vol. 2, pp. 1346–1351.
[214] C.-Y. Chiou, C.-H. Huang, A.-S. Liu, Y.-T. Chen, T.-H. Li, C.-J. Lin, H.-D. Chiang, and J.-L. Yuan, “Development of a micro-
processor-based transient data recording system for load behaviour analysis,” IEEE Transactions on Power Systems, vol.
8, no. 1, pp. 16–22, 1993.
[215] K. Tomiyama, J. P. Daniel, and S. Ihara, “Modeling air conditioner load for power system studies,” IEEE Transactions on
Power Systems, vol. 13, no. 2, pp. 414–421, May 1998.
[216] S. M. Shinners, Modern Control System Theory and Design, Second. New York: John Wiley & Sons, 1998.
[217] I. A. Hiskens and D. J. Hill, “Modelling of dynamic load behaviour,” in Proc. Of international Conference “Bulk Power
System Voltage Phenomena III”, 1994.
[218] J. Arrillaga and N. R. Watson, Computer Modelling of Electrical Power Systems, Second. New York: John Wiley &
Sons, 2001.
[219] G. J. Thaler and M. l. Wilcox, Electric Machines: Dynamics and Steady. New York: John Wiley & Sons, 1994.
[220] P. E. Sarachik, Principles of Linear Systems. London: Cambridge University Press, 1997.
Page 188
Modelling and Aggregation of Loads in Flexible Power Networks
[221] L. Ljung, System Identification: Theory for the user, Second. Prentice Hall, Inc., 1999.
[222] M. L. Coker and H. Kgasoane, “Load modeling,” in 1999 IEEE Africon. 5th Africon Conference in Africa (Cat.
No.99CH36342), vol. 2, pp. 663–668.
[223] TRACTEBEL Engineering GDFSUEZ, “No Title,” www.eurostag.be/en/products/eurostag/the-reference-power-system-
dynamic-simulation/. .
[224] W. W. Price, “PSLF Load Model Update,” in GE presentation at WECC Modeling & Validation WG Meeting, 2008.
[225] J. Matevosyan, S. Martinez Villanueva, S. Z. Djokic, J. L. Acosta, S. Mat Zali, F. O. Resende, and J. V. Milanovic,
“Aggregated models of wind-based generation and active distribution network cells for power system studies -
literature overview,” in 2011 IEEE Trondheim PowerTech, 2011, pp. 1–8.
[226] K. Yamashita, S. M. Villanueva, and J. Milanović, “Initial Results of International Survey on Industrial Practice on Power
System Load Modelling Conducted by CIGRE WG C4.605,” in CIGRE symposium, Bologna, Italy, 2011.
[227] K. Yamashita, S. Djokic, J. Matevosyan, F. O. Resende, L. M. Korunovic, Z. Y. Dong, and J. V. Milanovic, “Modelling and
Aggregation of Loads in Flexible Power Networks – Scope and Status of the Work of CIGRE WG C4.605,” in IFAC
Symposium on Power Plants and Power Systems Control, 2012.
[228] S. M. Villanueva, K. Yamashita, L. M. Korunović, S. Z. Djokić, J. Matevosyan, Z. Y. D. A. Borghetti, and J. V. Milanović,
“Modelling and Aggregation of Loads in Flexible Power Networks– Scope and Status of the Work of CIGRE WG
C4.605,” in CIGRE SC C4 Colloquium in Japan, 2012.
[229] L. M. Korunović, K. Yamashita, S. Z. Djokić, F. Villella, J. V. Milanović, A. Gaikwad, and S. M. Villanueva, “Overview of
Existing Load Models and Their Applications,” in CIGRE SC C4 Colloquium in Japan, 2012.
[230] K. Yamashita, S. M. Villanueva, S. Z. Djokić, J. Ma, A. Gaikwad, and J. V. Milanović, “Overview of Existing
Methodologies for Load Model Development,” in CIGRE SC C4 Colloquium in Japan, 2012.
[231] L. M. Korunovic, S. Sterpu, S. Djokic, K. Yamashita, S. M. Villanueva, and J. V. Milanovic, “Processing of load
parameters based on Existing Load Models,” in 2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT
Europe), 2012, pp. 1–6.
[232] J. V. Milanović, K. Yamashita, S. M. Villanueva, S. Z. Djokić, and L. M. Korunović, “International Industry Practice on
Power System Load Modeling,” Accepted for publication in IEEE Transactions for Power Systems.
Page 189
Modelling and Aggregation of Loads in Flexible Power Networks
[1] J.V. Milanović, K. Yamashita, S. Martínez Villanueva, S.Z. Djokić, L. M. Korunović “International Industry Practice on
Power System Load Modelling”, accepted for publication in IEEE Transactions on Power Systems
[3] K. Yamashita, S. M. Villanueva, and J. Milanović, “Initial Results of International Survey on Industrial Practice on Power
System Load Modelling Conducted by CIGRE WG C4.605,” in CIGRE symposium, Bologna, Italy, Sep., 2011.
[4] K. Yamashita, S. Djokic, J. Matevosyan, F. O. Resende, L. M. Korunovic, Z. Y. Dong, and J. V. Milanovic, “Modelling and
Aggregation of Loads in Flexible Power Networks – Scope and Status of the Work of CIGRE WG C4.605,” in IFAC
Symposium on Power Plants and Power Systems Control, Sep., 2012.
[5] S. Martínez Villanueva, K. Yamashita, L. M. Korunović, S. Z. Djokić, J. Matevosyan, A. Borghetti, Z.Y. Dong, J. V.
Milanović, “Modelling and Aggregation of Loads in Flexible Power Networks– Scope and Status of the Work of CIGRE
WG C4.605”, in Proc. CIGRE SC C4 Colloquium, Hakodate, Japan, Oct., 2012.
[6] L.M. Korunović, K.Yamashita, S. Z. Djokić, F. Villella, J.V. Milanović, A. Gaikwad, S. M. Villanueva, “Overview of
Existing Load Models and Their Applications”, in Proc. CIGRE SC C4 Colloquium, Hakodate, Japan, Oct., 2012.
[7] K. Yamashita, S.M. Villanueva, S.Z. Djokić, J. Ma, A. Gaikwad, J.V. Milanović, “Overview of Existing Methodologies for
Load Model Development”, in Proc. CIGRE SC C4 Colloquium, Hakodate, Japan, Oct., 2012.
[8] Z.Y. Dong, Alberto Borghetti, K. Yamashita, J.V. Milanvoic, “CIGRE WG C4.065 Recommendations on Measurement
Based and Component Based Load Modelling Practice”, in Proc. CIGRE SC C4 Colloquium, Hakodate, Japan, Oct.,
2012.
[9] L. M. Korunovic, S. Sterpu, S. Djokic, K. Yamashita, S. M. Villanueva, and J. V. Milanovic, “Processing of load parameters
based on Existing Load Models,” in 2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), Berlin,
Germany, 2012, pp. 1–6.
[10] K. Yamashita, S. Z. Djokić, F. Villella, J.V. Milanović “Self-disconnection and Self-recovery of Loads Due to Voltage
Sags and Short Interruptions”, in Proc. CIGRE Symposium, Lisbon, Portugal, Apr., 2013.
[11] S. Martínez Villanueva, K. Yamashita, J.V. Milanović, “Estado del arte en el modelado de cargas en grandes sistemas
eléctricos de potencia”, in Proc. CIGRE ERIAC, Foz de Iguazu, Brazil, May, 2013
[12] F. O. Resende, J. Matevosyan, J. V. Milanovic, “Application of Dynamic Equivalence Techniques to Derive Aggregated
Models of Active Distribution Network Cells and MicroGrids, A Critical Overview”, in Proc. IEEE PowerTech, Grenoble,
France, Jun., 2013
Page 190