You are on page 1of 6

APPLICATION OF NEURAL NETWORKS TO POWER SYSTEMS

PROF. D.P. Kothari


Centre for Energy Studies
LIT. Delhi. New Delhi-110016, INDIA.

luwtract - In recent years, ANNs (Artificial power systems to seek NN solutions to some of
Neural Networks) have a#racted considerable their more complicated or unsolved problems.
attenton as candidates for computational system NN applications to power systems can be
due to the variety of advantages they offer over the categorised under three main areas: regression,
colIventional m - o n a l systems. Among classifhation and combinatorial ophmjzation.
these advantages, the ability to memorise, rapidity Applications involving regression include transient
and robustness are the most profound and stability analysis, load forecasting, s y n c h r o m
interesting pmpert~es which have attracted machine m o d e - contingency screening and
attention in many fields. The paper critically harmonic evaluation Applications involving
raiews the ANN related publications involving classifications include harmonic load
typical power system problems during last ckade. identification, stataic and dynarmc s b t y
A brief overview of the ANN theory, difiemt analysis. The area of combinational optmizition
models and their apphcations is given. includes unit commitment and capacitor control.

INTRODUCTION OVERVIEW OF ARTIF'ICAL NEURAL


The paper describes a overview on ANNs NETWORKS:
in power systems. Artificial neural networks Artificial neural networki are made up of
(ANNs) are the distributed processing systems that simple highly inte~onnected processing units
have been inspired by the biological nerve system. called neurons each of which perform two
They consist of a p u p of units called "neurow" functions: aggregation of its inputs from other
that are analogous to nerve neurous. Each neuron n m n s or the external environment and
i s c o - t o each other with the weights. For generation of an output from the aggregated inputs.
example, the inductive learning process finds out A connection between a pair of neurons has an
the weights so that the relationship between input associated numerical strength called synaptic
and output variables is d e t e r m i d , The methds weight. The development of ANN involves two
throw up new possibilities of parallel or distributed phases: training or learning phase and testing
complting. phase. Training of ANN is done by presenting the
network with examples called trauzing patterns.
Recent years have witnessed a rapdly During training, the synaptic weights get m d f i e d
growing intemt in two important to model the given problem As soon as the
network has learnt the problem it may be tested
Artificial intelligence technologes viz., expert with new unknown patterns and its efliciency can
systems (ES) and ANNs. In recent be checked. (testing phase). Dependmg upon the
years, another term 'intelligent control' has come to training Imparted, ANN can be classified as
embrace diverse methodologies combining supeMsed ANN or unsupervised ANN.
conventional control and emergent techniques
based on physiological metaphors, such as ANNs, Supervised ANN
fuzzy logic, genetic algorithms and a wide variety The supervised ANN requires the sets of
of search and oghrmzation techniques [20]. In th~~ inputs and the outputs for its training. During the
paper we will mainly concentrate on ANNs and training, the output from the ANN is compared
their application to power systems. with the desired output (target) and the difference
ANNs have been studied for many years (error) i s reduced b y employing some 1 -
with the hope of unkrstanding and achieving algorithm. This training is repeated till the actual
human-like computational performance. output acquires an acceptable level. Supervised
Appeahg benefits include massive parallelism, ANN may be a fd forward or nonrecurrent
architectural modularity, fast speed, high fault network such as Multi Layer Perceptron (MLP),
tolerance and adaptrve capability. These have Functional Link Net (FLN) and Radial Basis
lured researchers from controls, robotics and Function (RBS), or a feedback or recurrent ANN

621
such as Hopfield network, often used in power Unsupervised ANN
system applications. The artiticlal neural network which does
not require a supenisor or teacher for training is
Multi-layer Perceptron Model: known as unsupervised ANN. In competitive or
tt comprises an input layer, one or more unsupervised learning units of the outplt layer
hidden layers) and an output-layer. It is used compete for the chance to respond to a given input
practically in all power system applications and is pattem. Kohonen's Self-organizing Feature Map
trained by a back propagation (BP) algorithm [7]. (SOFM) and Adaphve Resonance Theory (ART)
HP is an iterative, gradient search, supervised are examples of unsupermed learning.
algorithm and consists of three phases: forward
execution, back propagation of the error and ANN APPLICATIONS TO POWER
weight update. It works well with a sigmoidal SYSTEMS
activation rule. Several key features that distinguish NNs
from other AI techniques are: learning by example
Fancttoaal l i n k Network: in real time, distributed memory and associated
The FLN [8] not only increases learning mall, fault tolerance and graceful degradation,
rate but simplifies the learning algorithm. The real time pattern recognition, intelligent aSSOciatiOn
inputs are expanded and are used for training with and synksls.
die actual input data. Despite its nummus advantages,
distributed memory causes a major flaw in NNs,
RBF Network Model: because knowledge in a NN is stured as a ptttern
Radial basis function model [9] consists of weights and connections. Other stumbhg
of three layers: the input, hidden and output layers. blocks and how moE work is going on to tackle
The training of RBF network requires less them is discussed in the next section.
computation time since only the second layer This section &ids with typical ANN
weights have to be calculated using an error signal. aplication areas in power systems. The areas in
order of d e " g amount of work already
Parallel Setf-organizing Hierarchical Neural plMished are:
Networic Ihning Security assesrment, fault
Parallel, self-organizing, hierarchical detectioddiagonosis, control, analysis, protection
neural network (PSHNN) are multistage networks and design. Most popllar prohlems are: (i) load
[10], in which each stage neural network (SNN) is forecastin& (ii) security assessment and (iii) Edult
usually a 3-layered feed forward ANN having deteuioddiagonosis.
linear input and output units and nonlinear hidden
units. The revised back propagation algorithm is Load Forecasting:
used to train each SNN. The training of the Load forecasting is a suitable problem for
PSHNN is carried out for a number of sweeps till ANN apphcation due to the availability of
convergence is achieved. historical load data on the utility databases. ANN
schemes using peroepbon network and SoFM
aopwda Mode& have been successful in short-term [46,49] as well
Hopseld model [I11 is a recu~~ent neural as long-term load foI.ecasting with impressive
network (RNN) having feedback paths from their accuracy. A combined use of Unsupervised and
W p t s to their inplts and amsists of a single layex supervised learning was done for short-term load
of l x u m % actiog both as output and input using forecashng [12]. Dash et al. 1131 used " a n -
s d f a g a n b h g associatve memory. Neurons with filter based A" algorithm for faster convergence
gm&d response (or sigmoidal input-outpt and impcoved prediction acamcy. The RBF
dation) ~IE used in Hopfield Neural Network. neqwork was found superior to MLP or BP model
The outplt of each newon is linked with in terms of training time and itccuilcy.
the previous value of its om activation and ANN does not need additional memory
therefore individual neurons have time dependent for storing the history of load pa#ems. 1%
behaviour. It on recognize p;rtterns by matching improvement in accuracy of STLF can save upto
new uputs with ~ ~ y stored patterns. The Rs. 700 million for a typical power utdity.
Hopfield maid is particulsrly used for apphcation
to combinatorial optimization problems such as Security Assessment:
unit commitment. static and aynarmc security assessrment
often require on-line computation. In order to

622
evaluate solution efficiently, the nonlinear Liang and Hsu [19] pmp0sed NN bts.d
mapping of MLP is utilized to reduce approach for the schedulq of h y d m - g m d m .
computational burden and &al with the system hourly loads and the natural inflow of each
Charactensb*'cs of power systems. This allows us to mervoir were c o n s i w as inputs to the ANN.
cany out on-line monitoring/asse,ssment in
transient, mall signal stability, and voltage Power System Stabilizer Design:
instability. Power system stabilizer (PSS) has been
Though contmgency ranking and widely used in modern power systems to provide
sensitivity factor methods have reduced the datlllping for lower kquency oscillatm in the
number of critical contmgencies to be computed, powa system MLP based P.S.S. was praposed by
ANNs have played a challenging role in security several workers (211. In Ref. [22], a frray
area. kchniqw and " besed metlaod for Pss a " l
In ref. [14], a 4-layered feed forward by superconducting magnetic energy storage have
ANN trained with BP algorithm was discussed for been developed.
predicting bus voltages foB0wing an outage. The
Ps and Qls that af€ect bus voltages most were Load Flow:
selected as the inputs to ANN using an entropy LF is a must for s o w a large " b e r
function. Ghosh et al. [15] designed a feedback of power system problems. Kalra et al. [23]
ANN for line-now contingency " g . A New developed a East load flow method besed on MLP
type of performance index, the severity index was model with real and reactive load &mands at load
considered as output of the neural network buses as mpds. The autpd nodes pmvided
Jayasurya [16] proposed an ANN to | v I and 6 at all PQ buses. Ref. [24] presents
provide an energy measure which is an indication
of the power system's proximity to voltage an MLP based adaptive loss emhution a @ "
collapse. for p w e r transntss''on system.

Fault Detection/Diiosis: Voltage and Reactive Power C O W


Fault &ection/diagnosis is one of In Ref [25], Kojima et al. pmposed an
challenging problems in power systems. MLP RNN based algorithm for learning the imcrse
identifies the type and location of facults with a dynamics and apphed the algorithm to VQ control
given set of power system conditions, call4 "neun, V Q r , It was found to be more
measurements, alanms, etc. KN (Kohonen net) is StaMe and accUme.
applied to handle the classification of fhlt
patterns. The dqposis of the power apgaratus is There are many other power system
done to judge what kinds of faults the apparatus problems for whicb ANN is increasingly being
suffered from. KN is inferior to MLP in tenns of used such as load modelling [42,50], HVDC
the solution accuacy due to umpewised learning. [36,48], power system rehabdity studies [MI,
RBF and BP models [171 were developed for Wt topology [39], nonconventional energy source
dugnosis problem. The BP network had given [35], load fiqwncy control [33], maintenance
superior performance while training of RBF schedtuiing [31], unit co"itmenl [30].
network was much faster as compared to BP
network STUMBLING BLOCKS
ANN faces several problems to be solved
inspite of attractive features discussed & e r . l l e
Economic Load Dispatch: main difticulities with ANN implementation are
Park et al. [18] presented a method to discussed below.
solve ELD problem with piecewise quadtatrc cost
function using Hopfield NN. The ANN based Optilmlal Structure of ANN
appIoach turned out to be much simpler and It is essential to &dout the network size
accurate. Ref. [29] deals with combined ED and (the number of input and outprt neurons in MLP,
emission Qspatch using impnrved BPNN. and the number of output neurons in KN). Further,
Adaptwe Hopfield NN is recently used for ELD pruning is necessary to obtain more msonable
[41,471 models.

Hydroelectric Generation Scheduling:

623
o d i a e Efficient Learning Algoritbm: It is known that ES, FS (ibzzy system),
The BP algorithm requires many iteration genek algorithm (GA) and ANN have' strong
counts. It is not suitable for on-line learning c r d e n w s to deal with uncertain, lllample and
scheme. It is necessary to have a learning noise polluted data ANN is a good approximator
algorithm with better convergence. Efficient of non-linear M o n s and pexforming well for
global optmuahon technique is needed to non-linear regression. FS, GA, chaotic dynamics
evaluate the weights. are " l y used t o r e c b r c e the trainmg t i m e and
pruning of the ANN, Fine training or
Alleviation of "Curse of Dimensionality": defuzzi6cation can be done by ANN.
The direct application of ANN to large
scale real-size power system requires large-scale CONCLUSION
ANN. It is not easy to findout the optunal weights Neural networks are robust. Even if
in terms of accuacy and the computational effort. inputdata are not complete or bave some noise,
the ANN can still give good results. ANNs have
Consideration of Network Topdogies: adaptivity and can adJust t o the new e l l -
We have to cope with n m o r k topologies easily. Modern control techniques like adaptwe,
if the problem is related to transmisson lines. variable strucaue, H (infinite) a " l s can be
Actually, ANN applications to one or several learned by ANN from a series of training sets and
snapshot are not convincing. once the learning phase is over ANN can be used
as a robust controller. Special purpose hardware
Necessary Amount of Learning Data: designed to implement and evaluate ANN
The efficient g u t & h e for finding out the technologies and vanety of NNS (eg. recurrent) are
necessary amount of the learning data is not bemg seriously tried for solving PS problem.
avadable in constructing ANN. The inefficient An overview on ANNs in power system
data creates inappmpriate models while too much has been presented in this paper. This paper has
data needs excesive computation time. focussed on MLP, HN and KN as typical ANNs.
It is normal that in spite of good It may be seen that, MLP is the most poprlar
performance on training data, worse performance owmg to the surpeMsed learning that is superior
is obtained on test data. This may be due to the in terms of accuracy. As the apphcation areas,
fact that the training data is not utuformly load forecasting, security assessment and fault
distributed. The accuracy of ANN model depends detedion/dia~s were of main importance
on the number of t " g patterns in a gven range. though there exits a varie~ of application areas.
The present day p r i m a l Qfficulties and their
Other Factors: proposed solutions are reviewed Finally, the
In literature no systematic procsdure is integration of ANNs with other emerging
available on choice of initial weights assigned to technologies such as FS, GA etc. was chscussed as
the interconnections between two nodes in neural a future research direction.
network. Ref. [26] proposes use of some Wctions
for thts. REF'ERENCES:
The present AI implementation can reach
the perfection only if it acquires the level of human 1. I.J. Nagrath and D.P. Kothari,'Tbwer
competence. Unllke ES, ANN lmplementation Sytem Engineering", Tata
M e r s from a lack of end-user interaction. Kalra McGraw-Hill, New Delhi, 1994.
et al. [27] presented various models concerning the 2
synergsm of ES and ANN. NPSC Tutorial Course, I.I.T. Delhi,
It is important to handle normalization of 1994.
input data so that feature extraction is obtamed and 3.
D.P. Sengupta et al.,"Recent Advances in
solution accuracy is improved. Control and Management of Energy
Systems", Ch. 15 by P.K. Dash, Interhe y
FUTURE WORK: 1993.
Research would continue to enhance 4. M.El-Sharkawi and D. Niebur
ANN performance. It is a natural research (Eds.),'TEm Tutorial Come on ANNs
W o n to make use of other emerging with Applications to Power Systems", %
technologes [30,32,34,43,44,45] to overcome TP 112-0.
drawbacks o f m s .

624
5. L. Srivastava, S.N. Slngh and J. using HopEeld NN", IEEE Traos. pbwer
Sharma,"ANN Aapphation in Power System, Vo1.8, No.3, Aug. 1993, pp.
systems: An overview and Key Issues", 1030-1038.
Proc. Int. & d . CERA 97, pp. 3 9 7 4 3 . 19. hey-Hsun Liang and Yuan-
6. T. Dillon and D. Niebur (Eds.),"Neural Y.h.Hsu,"Scheduhg of Hydroelectfk
Net Appliations in Power Systems", generaton using A " , IEE Roc.-
CRL Publishing Limited, Lecies, U.K., GTD, Vol.141, N0.5, Sept. 1994, pp.
1995. 452-458.
7. D.E. Remelhart,*Rmllel Distributed 20. D.A Linkws and HO.
Processing", cambridge, M.A, MlT Nyongesa,"Leanung Systems in
Press, 1988. Intelligent control: An Apjxaid of
8. Y.H. Pa0," Adaphve Pattem Recognition Fuzzy, Neural and Genetic Algorithm
and Neural Networks", Addison - Wesley control Applications", IEE proc.control
Pub. Co. Inc., 1989. Theory Appl., Vol.143, No.4, July 19%,
9. D.K. Panaweera et al.,"l\pplication of m. 367-385.
Radml Basis Function NN Model for 21. L. Guan et al.,"ANN Power system
Short-term Load Forecasting", IEE Proc. Stabilizer Trained with au Ixqmved BP
GTD, V01.142, N0.1, Jan. 1995, p~. 45- Algorithm", IEE Proc. GTD, Vol.143,
50. N0.2, March 1996, p~. 135-141.
10. O.K. Ersoy and S.Wee Deng "Paralle1 22. Y. Kawakita et al., T ower-System
Self-organizing Hierclucal Neural S t a b k n g cantto1 by SMES using
Networks with Continuous Inputs and Fuzzy Techniques and NNs", Elect Engg.
Outputs", IEEE T m . Neural Networks, i n J - V01. 114, N0.2,1994, p~. 9-17.
Vo1.6, N0.5, sept'95. 23. P.K. Kalra, S.C. SrivastaVa, S.K. J o e
11. P.D. Wassennan,"Neural Computing- N. Kumar 'and R. Wagnes,"ANN Based
theory and Applications", Van Nostrand Load Flow M o d & , Proc. 4th Expert
R e d o l d , New York, 1989. System Apphcation to Power System,
12. M. Qunanovic, et Melbourne, Australta, 1993.
al.,"Unsupervised/supervised Learning 24. T.S. Sidhu and 2. Ao,"On-ljne
Concept for 24 Hour Load Forecasting", Evaluation of capacity and Energy
IEE Proc. Pt. C, Vol.14, No.4, July 1993, Losses in Power TransnisSion Systems
p~. 311-318. by using ANNs", IEEE Trans. Power
13 P.K. Dash et al.,"Powerdemand Delivery, Vol.10, No.4, Oct. 1995, pp,
Forecasting Using a Neural Network with 1913-1919.
an Adaptwe Learning L EE 25. Y. Kojima et al.,"Voltage and Reactive
Proc. GTD, Vo1.142, No.6, Nov. 1?95, Power control Using Recumznt Neural
pp. 560-568. Networks", Electrical Engg. in lapan,
14. Y.Y. Hsu and C.C. Yang,"Fast Voltage V01.114, No.4,1994, p~. 119-128.
Estimation u + g an ANN", Electric 26. G. S i n & S.C. S r i " , P.K Kalra and
Power System Research, 27; 1993, pp.1- D.M. Vinod Kumar,',"A Fast Approack to
9. ANN Trainhg and its Apphation to
15. S. Ghosh and B.H. Choudhary,"Design of Economic Load Despatch", Electric
an ANN for Fast Line Flow Contingency Machines and Power Systems, 1995, pp-
Ranlang", Electric Power and Energy 13-24.
Systems, Vo1.18, No.5, 1996, pp. 271- 27. P.K. Kalra, S.C. Srivastava and S.K.
277. Joshi,"Synergism of Expea System and
16. B. Jayasurya,"ANNs for Power System ANN for Power System Aplications",
Steady-state Voltage Instability Proc. XV NSC, Roorkee, March, 1992.
Evaluation'*, Electric Power System 28. P.S. Kulkami, S.G. Tarnelau and D.P.
Research, Vo1.29,1994, pp. 85-90. Kothari,"Radial Basis Function Neural
17. D.K. Ranaweera,"Companson of NN Network Application to Economic
Models for Fault Diagnosis of Power Generation scheduling with
Systems", Electric Power System Transmission Losses", M. Cod. on
Resear~h, 29,1994, pp. 99-104. Power, Dec. 1997, D e & p~. 92-97.
18. J.H. Park etal.,"Economic Load Dispatch
for Piecewise Quadrahc Cost Function

625
29. RS. Kulkami, A.G. Kothari and D.P. S&duhng'',IEEE Trans. on P.S.. Vol.14,
Kothan,"Co"ed Economic and No.2, May 1999, p~. 765-771.
Emission I)lgtatch using Impwved BP 42. AP. Alvas da Silva et al.,"A New
NW, Int. J. uf E m , Vd.28, No.1, Constructive ANN and its Application to
January, 2000. E W Load Representation", IEEE
30. Z. Ouyang and S.M. S h a h i W u r , " A Tmns. on P.S., Vo1.12, No 4, Nov'97, pp.
Hybrid ANN-DP Aapproach to Unit 1569-1575.
Commitment", IEEE PES 1991 Summer 43. R Kfiosla and T. Dlbn,"Learrung
Power Meeting, 91 SM 438-2, PWRS, Knowledge and Strategy of a Neuro-
San Diego, CA,Jul. 1991. Expert System Architecture in Alarm
31. H. Sasala, et al.,"A Solution Method for Processing", ibid, 1610-1618.
Mamtenance Schedulmg of Thermal 44. Y.Y. Hsu and F.C. Lw"A Combined
units by ANNs", m. of NNCEPI, pp. ANN-Fuzzy DP approach to Reactive
185-190, SMot'd, 1992. PowerNoItage Control in a Distribution
32. AS. C m h e k a r a et al.,"A N e w - Substation", IEEE Trans. on PS, Vo1.13,
expert System for Planning and Load No.4, NOV. 1998, p~. 1265-1271.
Forecasting of Distribuhon System", 45. M.A Abido and Y.L. Abdd-Mogid"A
Electric Power and Energy System, Hybnd Neuro-Fuzzy Power System
VoI.21, NOS, J~ne 1999, p~. 309-314. Stabdizer for M u l k k Fbwer
33. D.K. C h a m & , P.S. Satsang and P.K. Systems", ibid, pp. 1323-1330.
Kafra,"Load Frequency Control: A 46. J. Vermaak and E.C. Botha,'aecxlrrent
Generalized NN Approach", Electric Neural Netwodcs for Short-term Load
Power and Energ~ System, Vol. 21, Forecastmg", lEEE Trans. on P.S.,
N0.6, August 1999, pp. 405415. V01.13, No I, Feb. 1998, p. 126-132.
34. G. Ramaknshna' and N.D. Rao,"Adaptive 47. K.Y.Lee et al.,"Adaptme Hopfied N a d
Nemo-fuzzy Inference System for Networks for Economic Load Dispatch",
VoWar Control in Distribution IEEE Trans. m PS, Vol. 13, N0.2, May
Systems", Elect. Power Syst. Res., 1998, p~. 519-526.
V01.49, N0. 2, Match 1999, p~. 87-98. 48. K.G. Nareadra et a f . , ' ? n v q n into
35. F. Girand and Z.M. Salameh,"" an ANN Based on-line current
Madehng of Gust Effects on a Gnd- controller for an HVDC TtansnissiOn
&exactive Wind Energy Conversions Link", IEEE Trans. on P.S., V01.12,
system with Battery Storage", EPSR, No.4, NOV. 1997, pp. 1425-1431.
VolSO, N0.3, June 1999, pp. 155-162. 49. AS. AlFuhaid et &. ,"cascaded ANNs for
36. P.K Dash et al.,"A NN Based Feedback Short-term Laxi Forecasbng", ibid, pp.
Liaesrising Contdler for HVDC Links", 1524-1529.
QPSR, Vd.50, N0.2, May 1999, p~. 125- 50. T. Hiyama et al.,"ANN Based Dynatnic
132. Load MOdelling", &id, 1576-1583.
37. D.M Vinod Kumar and S.C.
Srivastava,"Power System State
Forecasbiag using ANNs", EMPS,
Vd.27, No.6, June, 1999, pp. 653464.
38. B. Changaroon and S.C. Sriwstava,"A
Hybrid PSS using Neuro-identifier and
predictive Controller'', ibid, pp. 637-651.
39. M. Abd-H-Aal and Abd-El-r
C~"Power Network Toplopy
Ramgmtion using NNs", EMPS, VoL27,
N0.2,1999, p~. 195-208.
40. N. Amjady and M. J3san, "Evaluation of
Power systems Rdiabdity by an ANN",
IEEE Trans. on P.S., Vol.14, NaL,
Feb'99, pp. 287-292.
41. M.P. Walsb and
M3CYMalley,"Anpented HspseM
Network for Cm Gegeratw

626