IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO.

3, MAY 2018 2919

Optimal Planning and Design of Low-Voltage

Low-Power Solar DC Microgrids
Mashood Nasir , Saqib Iqbal , and Hassan Abbas Khan , Member, IEEE

Abstract—Low-voltage, low-power solar photovoltaic (PV) ηm Name plate efficiency of PV panel/module.

based dc microgrids are becoming very popular in nonelectrified ηP L Efficiency reduction factor due to panel losses.
regions of developing countries due to lower upfront costs com- I(t) Incident irradiance at time t (W/m2 ).
pared to utility grid alternatives and limited power needs of rural
occupants. The optimal planning of distribution architecture along Itc (t) Temperature compensated irradiance at time
with sizing of various system components such as solar panels, t (W/m2 ).
m
batteries, and distribution conductors is essential for minimizing Itc Mean temperature compensated irradiance
the system cost and enhance its utilization. In this paper, we de- (W/m2 ).
velop a framework for optimal planning and design of low-power i
Itc temperature compensated irradiance at ith dip
low-voltage dc microgrids for minimum upfront cost. The analy-
sis is based on region-specific irradiance and temperature profiles; (W/m2 ).
m
constraints in storage and distributions; distribution loss analy- ND Total number of irradiance recordings below Itc .
sis; and optimum component sizing (storage, conductor, and PV Tcell (t) Temperature of cell at time t (°C).
panel) requirements based upon an energy balance model for a 24- Tamb (t) Ambient Temperature at time t (°C).
h operation. We further analyze the merits of tailoring distribution G Conductance matrix of village distribution.
architecture for maximizing the system utility in the planning of
future microgrid deployments. Gxy Individual Elements of conductance matrix.
gxy Distributor conductance between two houses (Ω−1 ).
Index Terms—Distributed generation, Newton-Raphson Vx Voltage at node x (V).
method, rural-electrification, microgrid.
Vy Voltage at node y (V).
V m ax Maximum voltage after convergence (V).
NOMENCLATURE V m in Minimum voltage after convergence (V).
SOCm ax Maximum value of battery state of charge (%).
The notation used throughout the paper is stated below for
SOCm in Minimum allowable battery state of charge (%).
quick reference. Other symbols are defined as required.
P (t) Power demand of the house at time t (W).
A. Indexes Pm ax Maximum power demand of the house (W).
t Instant of time ranging from 1 to T . Pm in Minimum Power demand of the house (W).
i House number ranging from 1 to N . Ploss Distribution losses (W).
x Conductor bus number ranging from 1 to N . ηD Distribution efficiency.
y Conductor bus number ranging from 1 to N . ηP L Efficiency degradation due to PV panel losses.
j Irradiance recording ranging from 1 to ND . ηm PV module efficiency.
ηMP Efficiency of DC-DC converter in CPPU.
B. Parameters η Efficiency of DC-DC converter at each house.
N Number of houses in the village. ηB Charging/discharging cycle efficiency of battery.
A Area of installed PV panels (m2 ). VD Worst voltage dips for power distribution (W).
w1 Per kWh cost of battery ($).
Manuscript received March 20, 2017; revised July 12, 2017 and September 11, w2 Per kW cost of PV panel ($).
2017; accepted September 23, 2017. Date of publication September 27, 2017; w3 Per meter cost of distribution conductor ($).
date of current version April 17, 2018. This work was supported in part by the
International Growth Centre, U.K. under Grant 37315 and in part by the Lahore
University of Management Science, Pakistan. Paper no. TPWRS-00393-2017. C. Acronyms
(Corresponding author: Hassan Abbas Khan.)
M. Nasir and S. Iqbal are with the Department of Electrical Engineering, MPPT Maximum Power Point Tacking
Lahore University of Management Sciences, Lahore 54792, Pakistan (e-mail: CPPU Central Power Processing Unit
14060018@lums.edu.pk; 15060005@lums.edu.pk). AWG American Wire Gauge
H. A. Khan was with the School of Electrical Engineering, The University
of Manchester, Manchester M13 9PL, U.K. He is now with the Department LV Low-voltage
of Electrical Engineering, Lahore University of Management Sciences, Lahore LP Low-power
54792, Pakistan (e-mail: hassan.khan@lums.edu.pk). kWp Kilo Watt Peak
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. PSH Peak Sunlight hours
Digital Object Identifier 10.1109/TPWRS.2017.2757150 IVF Irradiance Volatility Factor

0885-8950 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
2920
D. Decision Variables
CB Battery energy capacity (Whr).
Sm ax Nameplate power capacity of PV panels (Wp ).
X Size of the distribution conductor (AWG).
EB (t) Time varying energy state of the battery (Whr).
ES B (t) Energy flow from solar to battery at time t (Whr).
ES L (t) Energy flow from solar PV to load at time t (Whr).
EB L (t) Energy flow from battery to load at time t (Whr).
I. INTRODUCTION
CCESS to electricity even at basic levels increases pro-
A ductivity and provides opportunities for economic devel-
opment. According to the International Energy Agency (IEA),
more than 500 million occupants in South Asia (India, Pakistan
and Bangladesh) and further 500 million in Africa have no ac-
cess to any form of electricity [1]. These occupants have to rely
on unhealthy resources, like kerosene oil, even for lighting pur-
poses and the use of these fuels have many documented adverse
effects [2]. Unfortunately, most of these people do not have a
choice in this regard. The major source of electricity i.e., the grid
is unviable for many of these isolated villages and large upfront
costs of electrification through national grids are prohibitively
high for most of these developing countries [3], [4]. Therefore, a
paradigm shift towards powering these villages through low cost
(and consequently low-power) distributed renewable resources
such as solar PV is seen in recent years [4]–[7]. Such microgrid
systems are primarily based on DC distribution due to an in-
herent advantage in terms of an increased operational efficiency
and economic viability for many remote locations where the grid
is unavailable or highly intermittent [8]–[10]. These integrated
systems also have a better levelized cost of electricity compared
to standalone systems as it takes advantage of usage diversity at
a village scale [5]. Therefore, the focus for rural electrification
has been shifted to off-grid electrification through low-cost solar
DC microgrids [5], [6], [11].
Numerous Mini- or Micro- grids of capacities ranging from
1 kWp to 200 kWp have been proposed in the literature [7],
[12]–[16]. Low power microgrids generally provide low power
provision of a few watts to a few tens of watts per home as
compared to high power microgrids which allow provisions of
several hundred watts or higher. Prominent practical implemen-
tations for low-cost rural electrification include micro solar PV
plants (referred as microgrids) in India including Chhattisgarh,
Sunderbans and Lakshadweep [15], [17]. However, the most
common commercial scale microgrid is the Mera Gao Power
(MGP) in India which provides 5 W of DC electricity to each
subscribing house in a village, for only 8-hrs per day, enough to
power an LED light and a mobile-phone charging point. MGP
has reportedly connected over 10,000 households spread across
400 villages [16], [18]. In 2012, Uttar Pradesh and Renewable
Energy Development Agency (UPNEDA), installed 1 kW DC
microgrids in 11 districts covering around 4,000 houses [19].
Another, recent successfully deployment in Africa includes up to
600 W microgrids providing lighting and mobile phone charg-
ing typically up to 10 W for 8-hr daily operation [20]. Other
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 3, MAY 2018
TABLE I
SPECIFIC REGIONS FOR ANALYSIS WITH IRRADIANCE PROFILES
Regions Annually Regions Annually
Averaged PSH Averaged PSH
1. Bihar 4.992 7. Jhansi 5.457
2. Delhi 5.035 8. Indore 5.493
3. Kolkata 5.063 9. Mumbai 5.604
4. Rajpur 5.280 10. Aurangabad 5.664
5. Ranchi 5.328 11. Jodhpur 5.808
6. Banglore 5.400
small container based solar solutions on 12 V and 24 V are also
being readily utilized in Africa [12], [13], [21].
While these (and other similar low voltage low power) sys-
tems are becoming popular for off-grid rural electrification, for-
mal analysis on optimal component sizing and loss evaluation
is not addressed in the literature. Generic microgrid systems
planned without taking the regional characteristics in to consid-
eration are significantly oversized and are not a good fit for all
environments and regions. Thus, for DC microgrids, there is a
need to plan efficient distribution schemes based upon the de-
tailed loss analysis along with quantification of optimal system
sizing incorporating local conditions for overall cost minimiza-
tion and enhanced system utilization.
This paper therefore focuses on optimal planning and design
of low-voltage off-grid DC microgrids with local PV genera-
tion and storage for low-power provisioning. Various critical
parameters that affect the optimal sizing of system components
are identified and analyzed for optimal system assessment. In
order to analyze the impact of distribution architecture on op-
timal component sizing, we related distribution losses to actual
structure of the village and spatial distribution of houses in it.
Two possible architectures of DC distribution (O-architecture
and C-architecture) have been analyzed for power flow and
associated distribution losses. So, our first contribution lies in
quantification of distribution losses with respect to distribution
architectures in line with common settlements in South Asia.
Our second contribution lies in the detailed analysis (and
assessment) of various locations in India, (given in Table I) from
optimal cost, component sizing and distribution architecture
planning perspective. It has been identified that optimal sizing of
the system components is highly dependent upon region-specific
time varying profile of irradiance and temperature throughout
the year. Unfortunately, it is a common practice to use mean
values of regional irradiance and temperature for system sizing
which may lead to in accurate system sizing [22], [23].
Our third and major contribution is the formulation of the
optimization framework for the selection of microgrid compo-
nents, including PV generation capacity (kWp ), battery storage
capacity (kWh) and distribution conductor sizing (AWG) for
minimum cost solution. To the best of our knowledge, optimal
planning of system components for DC Microgrids considering
the impact of distribution architecture has not been presented
in literature and we believe that it will be highly beneficial
for future DC microgrid deployments. Further, quantification
of losses for distribution architectures with spatial arrangement
NASIR et al.: OPTIMAL PLANNING AND DESIGN OF LOW-VOLTAGE LOW-POWER SOLAR DC MICROGRIDS 2921

will be critical in retrofitting current systems to enhance the

overall capacity and system efficiency.
Kindly note that the focus in this work is on lowest cost
topologies for basic rural electrification. High power micro-
grids (with household provisions for several hundred watts)
are generally unviable (due to high up front cost and lack of
micro-financing options) and not commonly implemented for
self-sustained rural electrification. Therefore, we restrict our
analysis to low-power and low voltage DC microgrids which Fig. 1. Topological diagram of linearly distributed C -architecture with PV
are considered safe for direct touch and potentially do not cre- generation and power processing and storage units (PPSU).
ate any electric shock or fire hazards when implemented at lower
voltage levels (24 V and 48 V) with 100 VA name plate capacity
[22]–[24]. Importantly, the methodology presented in this work,
for the optimal component sizing and optimal distribution ar-
chitecture planning, is generic and equally applicable for higher
power DC microgrids. However, lower voltages (24 V and 48 V)
are not suitable for very high power system, as distribution losses
would be very high for low distribution voltages.
Alternatively, for higher power systems, higher voltage levels
of LVDC (220 V, 325 V and 380V–400 V) must be employed Fig. 2. Topological diagram of linearly distributed O-architecture with
(PPSU).
to enhance the distribution efficiency, but they require sophisti-
cated protection schemes for their safe operation [23].
The organization of the rest of the paper is as follows. are wide apart. Therefore, an additional concern of PPSU place-
Section II presents an overview of existing structures of villages ment arises because it directly affects the distribution losses and
and associated arrangement of houses in developing regions. In associated optimal sizing calculations. Due to inclusion of an
Section III an energy balance model incorporating various sys- additional objective i.e., optimal placement of PV generation
tem losses for microgrid operation is presented. System model and storage unit, the overall objective function becomes a com-
along with objective function formulation and constraints identi- binatorial optimization problem (not part of the current scope
fication for optimal sizing of microgrid components is presented of work).
in Section IV. In Section VI, we present and discuss the optimal Typically, in order to electrify these villages, standard radial
component sizing for various locations in India based on the systems with single generation and storage hub is installed [15],
formulated framework. [16], [29], [30]. However, the distribution efficiency of these
systems can be significantly enhanced by considering a second
generation hub if the provision is available. Therefore, we con-
II. COMMON VILLAGE ORIENTATIONS sider linear distribution microgrid architecture with up to two
In order to design an efficient power distribution architecture generation and storage hubs. These proposed structures can then
that ensures the optimal power flow from source end to load end, be classified into C- distribution architecture and O- distribu-
it is important to analyze the spatial distribution (orientation) of tion architecture depending upon the number of generation and
houses commonly found in villages across the developing coun- storage hubs.
tries. Typically, two main arrangements of houses are found:
a) Linear arrangement in which houses are generally situated A. Linearly Distributed C-Architecture
alongside a central street/road.
The visual representation for linearly distributed C- archi-
b) Clustered arrangement in which houses are situated in
tecture is shown in Fig. 1. Distribution conductors are laid in
independent fields or in clusters of multiple huts/homes
a linear manner while generation and power processing and
[25].
storage units (PPSU) are placed at the start of a village, thus for-
Northern Africa and Namibia show highly clustered settle-
mulate a C-like structure and is termed as linearly distributed
ment and population distribution patterns [26]. However, in
C-architecture. This village architecture is a simplistic model
Asia, primarily in South East Asian countries such as India,
of villages found commonly in India [16].
Pakistan and Bangladesh, the most common forms of rural set-
tlements is linear arrangement of houses which are situated
across roads in order to facilitate access to infrastructure facili- B. Linearly Distributed O-Architecture
ties, markets and resources [27], [28]. In this work, we develop Based upon the availability of the land, PV generation unit
a system model for these linear architectures and formulate op- and PPSU may be located at both ends of the central street.
timization framework for planning and design of low-power Thus, such a structure in which conductors are laid in a linear
and low-voltage DC microgrids situated in these linear settle- manner, interconnecting generation and storage at both ends of
ments. However, in a clustered non-linear architecture, houses the house load formulates a linearly distributed O-architecture
may not be in close proximity and may exist in clusters which DC microgrid (Fig. 2). While keeping the overall generation and
2922 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 3, MAY 2018

Fig. 4. System diagram for energy flow in O-architecture with N houses.

Fig. 3. System diagram for energy flow in C -architecture with N houses.

the converter losses can be linearized and its efficiency can be

storage capacity the same, the introduction of this architecture considered as a constant for majority of its operational range
having two similar generation and similar storage units on sys- [35]. For instance Kolar et al. [35] shows that a typical DC-DC
tem enhances its overall efficiency as discussed in Section VI. converter exhibits slight efficiency variations (±2%), during its
operation in between 25 to 95 percent of its rated output power
III. ENERGY BALANCE MODEL FOR A DC MICROGRID loading. Therefore, the rating of CPPU converter is approxi-
mated to a fixed value. For variable loading scenarios, detailed
In order to design an optimal system, it is important to an-
analysis on converter losses may be included using (8) and (14)
alyze the system in terms of load requirements, converter re-
for accurate sizing estimations. Battery charging/discharging
quirements, supply availability, storage size and loss analysis.
cycle efficiency (ηB ) depends upon the battery technology and
In this section we evaluate various loss elements in the opera-
manufacturer specifications. For the system evaluation, lead acid
tion of microgrid. These losses mainly include i) panel losses
battery is considered, while its charging/discharging efficiency
in the PV output due to low irradiance, soiling and mismatch,
is considered constant for the simplicity of analysis. Distri-
ii) Degradation in PV output due to temperature, iii) DC/DC
bution efficiency is assessed by calculating distribution losses
converter losses, iv) battery charge/discharge cycle losses and
in the system through Newton-Raphson Method modified for
v) distribution losses.
DC power flow analysis [36]. Ploss (t) accounts for distribution
The output produced by PV panels is a function of incident
losses and is a function of a) distribution configuration in a vil-
irradiance and temperature. However, due to panel losses and
lage, b) distribution voltage level and c) permitted load levels
temperature degradations, output of PV panel is generally lower
to each household and must be critically analyze. In order to
than its name plate capacity. Low irradiance losses generally
quantify distribution losses, an N -house village is modeled as a
vary linearly with the peak sun hours (PSH) of the incident
combination of interconnection resistance of the laid conductors
irradiance. A mathematical model to quantify low irradiance
as shown in Figs. 3 and 4. For each configuration (C or O) of the
loss has been presented in our earlier work [31]. The degrada-
village, a unique conductance matrix G can be formulated de-
tion in output characteristics of PV Panels due to soiling and
pending upon the spatial distribution between house and length
mismatch of cells and temperature, discussed in [32], [33], are
of conductor laid and is given by (4) and (5).
also incorporated. Therefore, considering panel losses and the
temperature degradation effect, S(t) is given by (1) ⎡ ⎤
G11 G12 ......... G1N
S (t) = A · ηP L · ηm .Itc (t) (1) ⎢G ⎥
⎢ 21 G22 ......... G2N ⎥
Where, Itc is the temperature compensated irradiance and G=⎢
⎢ .. .. .. ..
⎥
⎥ ; G ∈ RN ×N
⎣ . . . . ⎦
depend upon I(t), and Tcell and is given by (3) [34].
GN 1 GN2 ......... GN N
Tcell (t) = Tamb (t) + (0.01875 ∗ I (t)) (2) (4)
Itc (t) = (1 − (Tcell − 25) ∗ 0.0045) I (t) (3) 

N
y =1 gxy ;∀x = y
The output from PV panels is processed through MPPT Gxy = (5)
−gxy ; ∀ x = y
DC/DC converter which incurs losses in PPSU along with fur-
ther DC/DC losses at the distribution panel of each house (see
Fig. 3). The efficiency of a typical DC-DC converter varies sig- This G-matrix can be used to calculate Ploss and worst voltage
nificantly at low power levels (percentage loading). However, dip (VD) as discussed in our earlier work [36], [37] and are given
NASIR et al.: OPTIMAL PLANNING AND DESIGN OF LOW-VOLTAGE LOW-POWER SOLAR DC MICROGRIDS 2923

by (6) and (7). Since the battery lifetime is dependent upon its depth of dis-
N N
charge (DOD), therefore, energy level of the battery EB (t) at
1  any time t is not allowed to go below its minimum energy level
Ploss = Gxy [Vx (Vx − Vy ) + Vy (Vy − Vx )] (6)
2 x=1 y =1 dictated by allowable minimum state of charge SOCm in .

V D = Vxm ax − Vxm in (7) EB ≥ SOCm in · CB ; ∀T (11)

For energy balance model, we consider a typical village

microgrid orientation consisting of N houses electrified via C. Equality Constraints
solar PV panels having maximum power generation capacity The constraints on generated solar energy are dictated by (12)
Sm ax (kWp ) and a battery storage system having energy capac- as the generated energy can either be used to supply load or to
ity CB (kWh). Solar PV generation S(t) will vary with time charge the battery including the losses encountered in the path
depending upon the input irradiance and ambient temperature, of power flow.
therefore, battery state of charge and associated state of energy
EB (t) will also vary with time depending upon S(t) and load ηMP S (t) = ES L (t) + EB L (t) ; ∀T (12)
requirements. Load demand of each house i at any time is given
by Pi (t).Therefore, for any time interval Δt, the balance of Constraints on battery energy are given by the net balance of
energy at microgrid is given by (8) influx and out-flux of energy and are given by (16)

N
 ΔEB = EB (t) − EB (t − 1) = ηB ES B (t) − EB L (t) ; ∀T
ηMP S (t) Δt = ηB ΔEB (t) + ηi Pi (t) Δt + Ploss (t) Δt (13)
i=1
(8) The constraints on load are defined by (14) such that load de-
mand must always be fulfilled either through battery or solar PV
Where, ηMP is the efficiency of DC/DC converter employed output. For constant household power loading scenario as dis-
at PPSU responsible for maximum power point tracking, bat- cussed in the current scope of the work, converter efficiency ηi is
tery charging and maintenance of distribution voltage and ηi is considered constant. Therefore, linearization of load constraints
the efficiency of DC/DC converter employed at the distribution does not result in any significant loss of accuracy. However, in
panel of each house. Since battery may take energy from PV case of variable household loading, non-linear accurate mod-
panels or may supply power to the load at any time t depending els for converter efficiency need to be included for accurate
upon the net energy flux in the battery, ΔEB (t) can be positive estimations [35].
or negative. ΔEB (t) will be negative in durations of no solar
power generation and the stored energy in the battery will be  N
T  T

Pi (t)
used to meet the load demand. Δt + Ploss Δt = ES L (t) + EB L (t) ; ∀T
t=1 i=1
ηi t=1

IV. SYSTEM MODEL FORMULATION FOR OPTIMAL (14)

COMPONENT SIZING
In order to optimally size various system components in-
cluding PV generation capacity Sm ax (kWp ), battery Storage
capacity CB (kWh) and conductor size X (AWG) for mini-
mum cost of installation, we consider the energy flow diagrams
for C- and O- distribution architectures (as dictated by (8)) as
shown in Figs. 3 and 4, respectively.
A. Objective Function
The overall optimization problem is therefore, the minimiza-
tion of objective function (9), subjected to the constraints defined
by (10)–(17).
The optimal PV size Sm ax is determined by the maximum
output power produced by PV array. Similarly, optimal battery
size CB is determined by the maximum energy state attained
by the battery, therefore (15) and (16) dictate the equality con-
straints associated with the objective function for the optimal
sizing of PV and battery respectively.
Sm ax = max(S) ; ∀T (15)
CB = max(EB ) ; ∀T (16)
A range of possible conductor sizes for DC distribution are
considered from American wire gauge (AWG) table [38]. There-
fore, constraints on Conductor size are given by (17).
 T


min ω1 CB + ω2 Sm ax + ω3 X X ∈ {4AW G, 6AW G, 8AW G, 10AW G, 12AW G,
(E S B ,E S L ,E B L ,E B ,S m a x ,C B ,X )
t=1
(9) 14AW G, 16AW G} (17)

For considered range of operation, the stated objective func-

B. Inequality Constraints tion along with equality and inequality constraints exhibit lin-
Non-negativity constraints on decision variables earity as shown by (9)–(17). Therefore, optimization problem
is written in standard linear form and is solved through linprog
ES B ≥ 0, ES L ≥ 0, EB L ≥ 0, CB ≥ 0, Sm ax ≥ 0; ∀T (10) function in MATLAB.
2924 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 3, MAY 2018

V. RESULTS AND DISCUSSION

For the current scope of work, we consider a typical linear
village structure in South East Asia having (typically) 40 houses
and distance between two consecutive houses is (typically) 10 m.
We consider both C- and O- architectures along with two distri-
bution voltages of 24 V and 48 V. Optimal selection of PV panel
size, battery storage capacity and conductor size is performed
for 5 W (case 1) and 10 W (case 2) provision and is based upon
problem formulated in Section V.
The idea is to account for the impact of instantaneous fluctua-
tions in PV generation (based upon timed fluctuations in instant
irradiance and temperature) and losses on system sizing, there-
fore, selection of time base is critical. Selection of smaller time
base results in better resolution to account for instantaneous
changes, however, requires more computational resources for Fig. 5. Distribution efficiency, η (right y-axis and worst voltage dip, VD (left
data processing. The proposed methodology is generic and can y-axis) for C -architecture and O-architecture with 5 W loading at different
gauge sizes, and different voltage levels.
be applied for any time base, based upon the availability of solar
irradiance data, load profile and computational resources for op-
timization problem solving. For current optimization problem,
discrete time interval of 1 hour with constant load demand for
each house is used as time base for all the calculations. Hourly
variations in the irradiance and temperature data are taken from
NREL [39]. According to the presented optimization problem
optimal PV, battery and conductor sizing is determined.
PV panel losses are considered 8% for both cases. Although
the converter losses are quadratic in nature but due to fix load
assumption and for simplicity in optimization problem formu-
lation, DC-DC converter losses for CPPU as well as individual
household distribution panel converter are considered 10% for
both cases. Battery cycle efficiency ηB is considered 95%, and is
assumed constant for simplicity of analysis. The cost of panels
per watt-peak ‘w1 ’ is taken as 600 $/kWp (including the cost
of mounting frame) and lead-acid battery cost per kWh ‘w2 ’ is
taken as 120 $/kWh and gauge sizes cost is taken at 1671, 1305,
Fig. 6. Optimal selection of conductor size with 5 W power provisions at
947, 588, 358, 250 and 204 $ for 4, 6, 8, 10, 12, 14 and 16AWG, different voltage levels and distribution architectures.
respectively [40]–[42]. While there are variations in these costs
but we took most commonly found prices at which sourcing
is readily available. For overall system cost calculation, fixed is also interesting to note that O-24 V and C-48 V have similar
cost of 300 $ has been included which accounts for CPPU and voltage dips and efficiencies for all wire sizes. This is because
converter cost. of the uniform loading and equal generation at both ends of the
O- architecture. In case of non-uniform loading and unequal
generation, the characteristics of C-24 V and O-48 V will not
A. Case 1: 24-7 5 W Supply to 40 Houses for a 365-Day necessarily resemble, therefore, detailed distribution loss anal-
Operation
ysis has to be performed as discussed in [36]. Along with the
Based upon the analysis presented in Section IV, the distri- selection of PV panel and battery size, one important parameter
bution efficiency, ηD and worst voltage dips, V D calculated for is the optimization of the conductor size. The cost of conduc-
following cases are plotted in Fig. 5. tor increases with its decreasing gauge thickness and vice versa
i. C-24 V (C-architecture with 24 V distribution) while an opposite trend for system distribution efficiency is ob-
ii. O-24 V (O-architecture with 48 V distribution) served. Therefore, in order to analyze this effect in terms of cost,
iii. C-48 V (C-architecture with 48 V distribution) Fig. 6 (based on 365-day study on the solar irradiance data of
iv. O-48 V (C-architecture with 48 V distribution) Bihar, India) is plotted. The cost of distribution losses is calcu-
VD is critical in terms of power electronic converter require- lated by taking the difference between overall system cost with
ment at each subscribing household. While, generally, a 20% distribution losses at a particular AWG and the overall system
input voltage variation capability is allowed in most power elec- cost with ideal conductor having zero distribution losses.
tronic converters, the performance is optimal close to the rated From Fig. 6 it can be seen that conductor cost decreases with
input voltages. For this particular analysis, we limit the optimum the increase in gauge size while the cost associated with the
component selection at 20% variations in the grid voltages. It distribution losses increases at higher gauge values. Thereby,
NASIR et al.: OPTIMAL PLANNING AND DESIGN OF LOW-VOLTAGE LOW-POWER SOLAR DC MICROGRIDS
Fig. 7. Optimal PV panel sizing of the system at 5 W loading (Case 1). Kindly,
note that each data point on the figure represents a region shown in Table I. For
instance, the first data point (from left) is for Bihar where PSH is 4.992 and so
on. (Followed in all subsequent figures).
our optimization problem calculates the optimal point such that
overall cost of the distribution losses and upfront cost of conduc-
tor is minimal. For instance, for C-24V distribution architecture,
10AWG is optimum compared to its operation at 12AWG where
its operation would be less efficient and sub-optimal from cost
perspective. Alternatively, at 8AWG the system will be more
efficient but at the resulting cost would be higher compared to
the optimal value at 10AWG. In the current case study, the two
cost functions i.e., cost of distribution losses and cost of con-
ductor intersect due to similar scale. However, in general (for
all cases), it may not be the case. The objective is to find the
minima rather than the intersection for calculating the optimal
conductor size and there may not be an intersection point for
certain cases where cost of conductor is considerably higher
than cost of distribution losses and vice versa.
For further analysis of the impact of region-specific data i.e.,
time varying incident irradiance and temperature profiles on the
cost and sizing of the system, results of a 365-day study for
multiple locations in India (Table I), depicting optimal panel
sizing, optimal battery capacity and overall optimal installation
cost are shown in Figs. 7–9 respectively. These areas are selected
based upon their spatial distribution on the map and variation in
annually averaged, daily PSH.
From Figs. 7–9, a few important observations can be made.
1) Optimal PV sizing varies approximately linearly with the
average daily PSH (Fig. 7).
2) Average daily PSH are not a direct measure for optimal
battery sizing (Fig. 8).
3) As the battery cost constitutes a predominant portion of
the overall cost of the system, the overall cost of the system
also does not vary linearly with the daily PSH (Fig. 9).
Fig. 8 and Table I verify that “optimal battery sizing is not
directly related with the average PSH and the information of
average PSH only is not sufficient for optimal sizing of battery
storage requirements.” Rather, a detailed analysis on the varia-
tions pattern of PSH is needed to calculate the optimal battery
storage requirements. For instance, in Fig. 8, for all four cases,
2925
Fig. 8. Optimal battery sizing of the system at 5 W loading on left Y- axis and
irradiance volatility factor on right Y- axis. (Case 1).
Fig. 9. Optimal installation cost of the system at 5 W loading (Case 1).
the battery requirements are lower for PSH = 5.3 in comparison
to PSH = 5.4 and PSH = 5.5. Thus, if a proper investigation
is not made for daily variations in PSH, it may leads to incor-
rect sizing estimations. Therefore, a detailed analysis must be
conducted to find the optimal battery sizing incorporating the
volatility of irradiance. In order to analyze this further an irra-
diance volatility factor (IVF) has been defined, which accounts
for the number of dips from the mean ‘temperature compen-
sated irradiance’ and their corresponding extent of deviation
normalized over mean irradiance.
ND
1   m i

IV F = m Itc − Itc (18)
N Itc i=1
Where, N is the total number of temperature compensated
irradiance recordings out of which ND numbers of recordings
m
are below mean temperature compensated irradiance (Itc ) and
i
(Itc ) is the temperature compensated irradiance of ith dip. Us-
ing (18), IVF has been calculated for various regions under
observation and it is found that battery sizing varies in direct
proportion with IVF as shown in Fig. 8. The value of IVF de-
pends on number of dips and their extent of deviation from the
mean intensity. IVF directly impacts the battery sizing and in
2926 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 3, MAY 2018

Fig. 11. Optimal selection of conductor size with 10 W power provisions

based upon the relative cost of distribution at different voltage levels and distri-
Fig. 10. Distribution efficiency η D and worst voltage dip VD for C- bution architectures.
architecture and O-architecture with 10 W loading at different gauge sizes,
and different voltage levels.

turn the overall cost of the system. For instance, in Fig. 8 it can
be seen that Kolkota with a relatively low PSH 5.06 (Table I)
compared to Banglore (PSH 5.4) requires lower battery size.
This is due to the reason that Kolkota has lower IVF (volatility
factor) compared to Banglore which directly translates into re-
duction in the battery size. Similarly, regions with higher IVF
require higher battery sizing despite of possibility of having high
mean irradiance. Therefore, the sizing of a PV system must in-
corporate the volatility of irradiance along with the mean values
for PSH.

B. Case 2: 24-7 10 W Supply to 40 Houses

for 365-Days Operation
The most critical aspect is the calculation of optimal sizing Fig. 12. Optimal installation cost of the system (case 2).
for higher power (10 W) is the worst voltage dip along with dis-
tribution efficiency of each distribution topology (see Fig. 10).
conductor sizes for 5 W operation (Fig. 6) are 10, 12, 12 and
With the increase in power provisioning, the distribution losses
16AWG for C-24, O-24, C-48 and O-48 configurations respec-
(I2 R) increase significantly compared to case 1 (5 W household
tively. While for 10 W provision, the gauge sizes are reduced
load). However, like case 1, C-48 V and O-24 V have very sim-
to 6, 10, 10 and 12AWG for C-24, O-24, C-48 and O-48 con-
ilar voltage dips and overall distribution efficiencies. Moreover,
figurations respectively. Therefore, the proposed optimization
distribution at lower voltage level i.e., 24 V with C-distribution
framework has adjusted the conductor size such that overall
architecture becomes practically infeasible due to higher losses
cost of the losses has been compensated by the selection of a
and higher voltage dips.
thicker conductor while keeping the overall cost of the system
The increased distribution losses can be compensated by se-
as minimum. Fig. 12 shows the overall system optimal cost for
lecting a thicker conductor which will increase the cost of the
various regions (with reference to their PSH). It is important to
system. Therefore, it becomes even more critical to optimally
note that due to the planning flexibility in the current framework,
size the conductor by taking the capital and relative cost of dis-
the overall cost of the system in high power provision follows
tribution into account. The proposed optimization framework
the same trend with relatively higher values in comparison to
therefore, enables the optimal selection of conductor size based
case 1. Further it has been observed that the overall panel and
upon the trade-off between cost of the conductor and its relative
battery requirements for case 2 also follow a similar trend as
cost of distribution (see Fig. 11).
compared with case 1 (results for case 2 are not shown here).
From the comparison of Fig. 6 and Fig. 11 it can be seen that
with the higher power provision, the optimal conductor selection
has been shifted to lower gauge size (thicker conductor). For in- VI. CONCLUSION
stance, the optimal conductor sizes for 5 W operation (Fig. 6) are A methodology for optimal planning and design of PV based
10, 12, 12 and 16AWG for C-24, O-24, From the comparison DC microgrids is formulated. The proposed framework al-
of Fig. 6 and Fig. 11 it can be seen that with the higher power lows optimal selection of 1) solar panel sizing, 2) storage size
provision, the optimal conductor selection has been shifted to and 3) conductor size for optimal cost solution. The depen-
lower gauge size (thicker conductor). For instance, the optimal dence of region specific temperature and variations in irradiance
NASIR et al.: OPTIMAL PLANNING AND DESIGN OF LOW-VOLTAGE LOW-POWER SOLAR DC MICROGRIDS
pattern on optimal system sizing is observed through analysis
of multiple locations in India. The quantification of system per-
[17]
formance with distribution of generation and storage units at
the two ends of the system (O-architecture) shows considerable
improvement on distribution efficiency (10% or higher in most [18]
cases). In particular, C-48 V (C-architecture with 48 V distribu-
tion) and O-24 V (O-architecture with 24 V distribution) show [19]
remarkable similarity in terms of optimal system costs.
[20]
The selection of distribution voltage (24 V or 48 V) and the
distributed architecture (C or O) is ultimately the choice for the
microgrid installers/planners and is influenced by power con- [21]
verter preferences and financial models. But, importantly, the
framework established in this work for optimal system topol- [22]
ogy (in multiple constraint scenarios) will be highly useful in
planning for new optimal low-voltage systems along with effi- [23]
ciency enhancements of many existing systems by retrofitting
distribution architectures.
[24]
REFERENCES [25]
[1] World Energy Outlook, Electricity Access, 2016. [Online]. Available: [26]
http://www.worldenergyoutlook.org/resources/energydevelopment/energ
yaccessdatabase/
[2] N. L. Lam, K. R. Smith, A. Gauthier, and M. N. Bates, “Kerosene: A review [27]
of household uses and their hazards in low-and middle-income countries,”
J. Toxicol. Environ. Health, Part B, vol. 15, pp. 396–432, 2012. [28]
[3] J. Thornburg, T. S. Ustun, and B. Krogh, “Smart microgrid operation
simulator for management and electrification planning,” in Proc. 2016 [29]
IEEE Power Energy Soc. PowerAfrica, 2016, pp. 1–5.
[4] K. Ubilla et al., “Smart microgrids as a solution for rural electrification:
Ensuring long-term sustainability through cadastre and business models,” [30]
IEEE Trans. Sustain. Energy, vol. 5, no. 4, pp. 1310–1318, Oct. 2014.
[5] A. Jhunjhunwala, A. Lolla, and P. Kaur, “Solar-dc microgrid for Indian
homes: A transforming power scenario,” IEEE Electrific. Mag., vol. 4, [31]
no. 2, pp. 10–19, Jun. 2016.
[6] K. Shenai, A. Jhunjhunwala, and P. Kaur, “Electrifying India: Using solar
dc microgrids,” IEEE Power Electron. Mag., vol. 3, no. 4, pp. 42–48, [32]
Dec. 2016.
[7] P. A. Madduri, J. Poon, J. Rosa, M. Podolsky, E. A. Brewer, and S.
R. Sanders, “Scalable DC microgrids for rural electrification in emerg- [33]
ing regions,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 4, no. 4,
pp. 1195–1205, Dec. 2016.
[8] H. Lotfi and A. Khodaei, “AC versus DC microgrid planning,” IEEE [34]
Trans. Smart Grid, vol. 8, no. 1, pp. 296–304, Jan. 2017.
[9] J. J. Justo, F. Mwasilu, J. Lee, and J.-W. Jung, “AC-microgrids versus [35]
DC-microgrids with distributed energy resources: A review,” Renewable
Sustain. Energy Rev., vol. 24, pp. 387–405, 2013.
[10] D. Soto and V. Modi, “Simulations of efficiency improvements using [36]
measured microgrid data,” in Proc. 2012 IEEE Global Humanitarian
Technol. Conf., 2012, pp. 369–374.
[11] V. M. Balijepalli, S. Khaparde, and C. Dobariya, “Deployment of micro- [37]
grids in India,” in Proc. IEEE Power Energy Soc. Gen. Meeting, 2010,
pp. 1–7.
[12] Z. Xu, M. Nthontho, and S. Chowdhury, “Rural electrification implemen-
tation strategies through microgrid approach in South African context,”
Int. J. Elect. Power Energy Syst., vol. 82, pp. 452–465, 2016. [38]
[13] C. L. Azimoh, P. Klintenberg, F. Wallin, B. Karlsson, and C. Mbohwa,
“Electricity for development: Mini-grid solution for rural electrification [39]
in South Africa,” Energy Convers. Manage., vol. 110, pp. 268–277, 2016.
[14] R. Singh and K. Shenai, “DC microgrids and the virtues of lo-
cal electricity,” IEEE Spectr., vol. 51, no. 2, Feb. 2014. [Online]. [40]
Available: https://spectrum.ieee.org/green-tech/buildings/dc-microgrids-
and-the-virtues-of-local-electricity [41]
[15] D. Palit, G. K. Sarangi, and P. Krithika, “Energising rural India using
distributed generation: The case of solar mini-grids in Chhattisgarh state,
India,” in Mini-Grids for Rural Electrification of Developing Countries. [42]
New York, NY, USA: Springer, 2014, pp. 313–342.
[16] D. Palit and G. K. Sarangi, “Renewable energy based mini-grids for
enhancing electricity access: Experiences and lessons from India,” in
2927
Proc. Int. Conf. Utility Exhib. Green Energy Sustain. Develop., Mar. 19–
21, 2014, pp. 1–8.
S. Mishra and O. Ray, “Advances in nanogrid technology and its integra-
tion into rural electrification in India,” in Proc. 2014 Int. Power Electron.
Conf. (IPEC-Hiroshima 2014—ECCE Asia), 2014, pp. 2707–2713.
J. Urpelainen, “Energy poverty and perceptions of solar power in marginal-
ized communities: Survey evidence from Uttar Pradesh, India,” Renewable
Energy, vol. 85, pp. 534–539, 2016.
A. K. Srivastava, “Solar minigrids in rural areas of Uttar Pradesh,” Akshay
Urja, vol. 4, pp. 16–17, 2013.
P. Loomba, S. Asgotraa, and R. Podmore, “DC solar microgrids: A suc-
cessful technology for rural sustainable development,” in Proc. 2016 IEEE
Power Energy Soc. PowerAfrica, 2016, pp. 204–208.
B. K. Blyden and W.-J. Lee, “Modified microgrid concept for rural elec-
trification in Africa,” in Proc. 2006 IEEE Power Eng. Soc. Gen. Meeting,
2006, pp. 1–5.
B. T. Patterson, “LVDC: From concept to practice: An EMerge per-
spective,” in Proc. IET Low Voltage, Direct Current Conf., Jun. 2015,
pp. 87–113. doi: 10.1049/ic.2015.0044.
D. Salomonsson, L. Soder, and A. Sannino, “Protection of low-voltage
DC microgrids,” IEEE Trans. Power Del., vol. 24, no. 3, pp. 1045–1053,
Jul. 2009.
National Fire Protection Association, Quincy, MA, USA, NFPA 70: Na-
tional Electrical Code, 2011.
A. R. Desai, Rural Sociology in India. Mumbai, India: Popular Prakashan,
1994.
C. Linard, M. Gilbert, R. W. Snow, A. M. Noor, and A. J. Tatem, “Pop-
ulation distribution, settlement patterns and accessibility across Africa in
2010,” PloS One, vol. 7, 2012, Art. no. e31743.
M. Silberfein, Rural Settlement Structure and African Development. Boul-
der, CO, USA: Westview, 1998.
K. R. Varshney et al., “Targeting villages for rural development using
satellite image analysis,” Big Data, vol. 3, pp. 41–53, 2015.
S. C. Bhattacharyya and D. Palit, “Mini-grid based off-grid electrification
to enhance electricity access in developing countries: What policies may
be required?” Energy Policy, vol. 94, pp. 166–178, 2016.
M. M. H. Sajeeb, A. Rahman, and S. Arif, “Feasibility analysis of solar
DC nano grid for off grid rural Bangladesh,” in Proc. 2015 3rd Int. Conf.
Green Energy Technol., 2015, pp. 1–5.
S. Pervaiz and H. A. Khan, “Low irradiance loss quantification in c-Si
panels for photovoltaic systems,” J. Renewable Sustain. Energy, vol. 7,
2015, Art. no. 013129.
M. R. Maghami, H. Hizam, C. Gomes, M. A. Radzi, M. I. Rezadad, and
S. Hajighorbani, “Power loss due to soiling on solar panel: A review,”
Renewable Sustain. Energy Rev., vol. 59, pp. 1307–1316, 2016.
E. Forniés, F. Naranjo, M. Mazo, and F. Ruiz, “The influence of mismatch
of solar cells on relative power loss of photovoltaic modules,” Solar En-
ergy, vol. 97, pp. 39–47, 2013.
C. S. Solanki, Solar Photovoltaics: Fundamentals, Technologies and Ap-
plications. New Delhi, India: PHI Learn. Pvt. Ltd., 2015.
J. Kolar, F. Krismer, Y. Lobsiger, J. Muhlethaler, T. Nussbaumer, and J.
Minibock, “Extreme efficiency power electronics,” in Proc. 2012 7th Int.
Conf. Integr. Power Electron. Syst., 2012, pp. 1–22.
M. Nasir, N. A. Zaffar, and H. A. Khan, “Analysis on central and dis-
tributed architectures of solar powered DC microgrids,” in Proc. 2016
Clemson Univ. Power Syst. Conf., 2016, pp. 1–6.
M. S. M. Hamza, S. Fazal, M. Nasir, and H. Khan, “Design and analy-
sis of solar PV based low-power low-voltage DC microgrid architectures
for rural electrification,” presented at the 2017 IEEE Power Energy Soci-
ety General Meeting Energizing More Secure, Resiliant Adaptable Grid,
Chicago, IL, USA, 2017.
American Wire Gauge Table. [Online]. Available: http://www.rapidtables.
com/calc/wire/wire-gauge-chart.htm. Accessed on: Mar. 1, 2017.
National Solar Radiation Database, Nat. Renewable Energy Lab., Golden,
CO, USA. [Online]. Available: https://nsrdb.nrel.gov/. Accessed on:
Mar. 6, 2017.
Southwire Company, LLC. [Online]. Available: http://www.southwire.
com/Southwire.htm. Accessed on: Mar. 6, 2017.
S. Matteson and E. Williams, “Residual learning rates in lead-acid batter-
ies: Effects on emerging technologies,” Energy Policy, vol. 85, pp. 71–79,
2015.
S. M. Schoenung and W. V. Hassenzahl, “Long-vs. short-term energy
storage technologies analysis: A life-cycle cost study: A study for the
DOE energy storage systems program,” Sandia Nat. Lab., Albuquerque,
NM, USA, Sandia Rep. SAND2003-2783, 2003.
2928
Mashood Nasir received the B.S. degree from the
University of Engineering and Technology, Lahore,
Pakistan, and the M.S. degree from the University of
Management and Technology (UMT), Lahore, Pak-
istan, in 2009 and 2011, respectively, both in electri-
cal engineering. He is currently working toward the
Ph.D. degree in the Department of Electrical Engi-
neering, Lahore University of Management Sciences,
Lahore, Pakistan. From 2011 to 2012, he was a Lec-
turer, and from 2013 to 2014, he was an Assistant
Professor in the Electrical Engineering Department,
UMT. His research interests include power electronics, electrical machines and
drives, grid integration of alternate energy resources, electrochemical energy
conversion and battery storage systems, ac/dc/hybrid microgrids, etc.
Saqib Iqbal received the B.E. degree from the Ghu-
lam Ishaq Khan Institute of Engineering Sciences
and Technology, Topi, Pakistan, in 2014, and the
M.S. degree from Lahore University of Management
and Sciences (LUMS), Lahore, Pakistan, in 2017,
both in electrical engineering. In 2017, he joined the
Syed Baber Ali School of Science and Engineering,
LUMS, as a Research Assistant. His research inter-
ests include smart grids, microgrids, and renewable
energy resources.
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 33, NO. 3, MAY 2018
Hassan Abbas Khan received the B.Eng. degree in
electronic engineering from the Ghulam Ishaq Khan
Institute of Engineering Sciences and Technology,
Topi, Pakistan, in 2005, the M.Sc. (with distinction)
degree, and Ph.D. degree in electrical and electronic
engineering from the School of Electrical Engineer-
ing, The University of Manchester, Manchester, U.K.
From 2005 to 2010, he was with the School of Elec-
trical Engineering, The University of Manchester. He
is currently an Assistant Professor in the Department
of Electrical Engineering, Lahore University of Man-
agement and Sciences, Lahore, Pakistan. His current research focuses on renew-
able energy and its uptake in developing countries. His core focus is on novel
grid architectures for low-cost rural electrification through solar energy. He is
also working on efficient and reliable solar PV deployments in urban settings to
maximize their performance ratios.