Optimized Solar Photovoltaic Generation in a Real

Local Distribution Network

Hamidreza Sadeghian, Mir Hadi Athari, Zhifang Wang
Department of Electrical and Computer Engineering
Virginia Commonwealth University
Richmond, VA, USA
Email: {sadeghianh}, {atharih}, {zfwang}@vcu.edu

Abstract— Remarkable penetration of renewable energy in optimum size of the DG. These methodologies are either based
electric networks, despite its valuable opportunities, such as on analytical tools or on optimization programming methods.
power loss reduction and loadability improvements, has raised In [5], the authors presented an analytical approach to
concerns for system operators. Such huge penetration can lead to determine the optimal allocation for the DG with an objective
a violation of the grid requirements, such as voltage and current
of loss minimization for distribution and transmission
limits and reverse power flow. Optimal placement and sizing of
Distributed Generation (DG) are one of the best ways to networks. In [6], a new meta-heuristic Fireworks Algorithm
strengthen the efficiency of the power systems. This paper builds (FWA) is implemented on network reconfiguration problem to
a simulation model for the local distribution network based on minimize the power loss and enhance the voltage profile of the
obtained load profiles, GIS information, solar insolation, feeder system. A simple search algorithm is proposed in [7] for
and voltage settings, and define the optimization problem of solar optimal sizing and placement of DG for a network system,
PVDG installation to determine the optimal siting and sizing for based on losses and cost function as an objective function. The
different penetration levels with different objective functions. The method is simple but time-consuming for searching both the
objective functions include voltage profile improvement and best location and optimum size. In [8], the author considered
energy loss minimization and the considered constraints include
the minimum loss and generation cost as a parameter in
the physical distribution network constraints (AC power flow),
the PV capacity constraint, and the voltage and reverse power addition to DG power limits to determine the optimal size and
flow constraints. location of the DG. The method is accurate but very tedious
and mathematical computation needs much computation time.
Index Terms – PV distributed generation, optimal allocation, loss Authors in [9], presented an effective technique for optimal
reduction, voltage improvement. placement of Photovoltaic (PV) array and network
reconfiguration in radial distribution network simultaneously
I. INTRODUCTION to diminish the total real power loss and enhance the voltage
Over the past decade, employment of solar power for level of the network. The population-based Differential
electricity generation has grown dramatically due to its Evaluation (DE) Algorithm has been implemented to identify
economic benefits. However, due to variable nature of PV an optimal switching combination, optimal location, size and a
generation, the integration of a large amount of PV in a close number of PV module in the distribution system. In [10], a
geographic region will have various negative effects on the stochastic approach based on kernel density estimation is
operation of distribution feeders. The most common potential proposed to identify the optimal location for the PV plant
concerns caused by solar power are steady-state overvoltage, installation in distribution systems so that the voltage deviation
impacts on system losses, and issues with voltage regulating and network losses are minimized. In order to demonstrate the
devices, protection, and voltage fluctuation. Generally, power effectiveness of the proposed method, the model of a real
loss minimization and voltage stability improvement are distribution feeder has been used. The test case system is
important areas of power systems due to existing transmission
located in Walterboro, SC, USA, which is composed of 38-bus
financial loss of utility, network reliability, and power system
and includes a photovoltaic plant. Authors in [11], have
blackouts. Therefore, optimal allocation of PV generation is
necessary to support grid voltage regulation and improve the proposed a new approach to studying the impact of high PV
performance of distribution networks [1]–[4]. penetration on a distribution network and its hosting capacity.
The proposed method combines high-resolution resource
For the PV distributed generation (PVDG), there have been assessment using sky imagery with power system simulation
numerous studies to achieve the optimum allocation of the on real distribution models to study the impacts of up to 200%
system. As it mentioned, the optimal site and size of DG PV penetration level on voltage excursions, line losses, and tap
reflects the maximum loss reduction and improvement in changing operations. A new dual-index-based analytical
voltage profile of distribution system. Different methodologies approach to determine the optimal location, size and power
have been developed to determine the optimum location and factor of DG unit for reducing power losses and enhancing
loadability is presented in [12]. This index is defined as a
combination of active and reactive power loss indices by OpenEI dataset includes hourly load profiles for all types of
optimally assigning a weight to each index. DG placement buildings (e.g., residential base load, residential low load, large
with optimal power factor and appropriate weights of active office, small office, quick service restaurant, small hotel, etc.)
and reactive power losses can significantly reduce losses and during a year. The peak load for each load type is determined
better enhance loadability and voltage profiles. based on statistical analysis of available realistic data. Fig. 2
To conclude, there has been a lack of research on PVDG shows the empirical probability density function (PDF) of peak
allocation based on large-scale real-world feeders that load for base residential load and the approximated fit
incorporates real-time solar insolation data along with time- distribution. The empirical PDF is approximated with Log-
series analysis in the U.S. and this paper aims to address this Logistic distribution with parameters shown in Fig. 2.
gap. The main contributions of this study are a) to build a
simulation model for the local distribution network based on
obtained load profiles, GIS information, solar insolation,
feeder and voltage settings, b) to define the optimization
problem of solar PVDG installation to determine the optimal
siting and sizing and c) to analyze the optimal siting/sizing
results obtained from optimization procedure. Note that the
objective functions include voltage profile improvement and
energy loss minimization and the considered constraints
include the physical distribution network constraints (AC
power flow), the PV capacity constraint, and the voltage and
reverse power flow constraints.
The rest of the paper is organized as follows. In section II
the description of data analysis for electrical network modeling Figure 1. Base map of study area and sub-regions
and simulation is presented. Section III describes the problem
formulation. Simulation scenarios and results are provided in
section IV and section V discusses the results and concludes the
paper.
II. DATA ANALYSIS AND ELECTRICAL NETWORK
MODELING
The hypothetical electrical distribution model is based on a
local utility distribution network in Virginia with a summer
peak of 23,260 kW feeding 1902 customers by classes of 1429
residential, 397 small commercial, and 76 large
commercial/industrial. The network voltage level is 12.5 kV.
For security reasons, the local utility could not provide specific
information on electrical network structure and design.
Therefore, we tried to model the system based on the logical
alignment of the electrical system and statistical analysis of Figure 2. Empirical PDF of base residential load and approximated fit
available data from the local utility and Open Energy distribution
Information (OpenEI).
By considering provided actual peak load by the local utility
By obtaining GIS layers that identified structures and and scaling bus load profiles, aggregated load profiles for all
zoning information for all properties in the area surrounding the categories are shown in Fig. 3.
substation and an iterative re-drawing process, a collection of
buildings around the substation that perfectly matched the local 25000
Larg Commercial
utility customer data for the substation captured. For electrical
Residential
network modeling, first, study area divided to six sub-regions 20000
based on study area map and electrical lines and then main Smal Commertial

distribution feeders and branch lines which recognized from Total Demand
Demand (kW)

15000
Google map, logically assumed to feed these areas. An exact
number of building types in each area derived from GIS layers. 10000
(Fig. 1).
A. Load buses and load profiles 5000

In order to consider load buses, buildings in the study area

are grouped based on geographical locations. Next, load profile 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
of each bus is calculated by aggregating of hourly load profiles
Time of a day
of all buildings connected to that bus [13]. Note that, different
bus types are determined based on connected building types. Figure 3. Aggregated load profiles for all categories
B. Solar radiation potential
Solar insolation was determined using LiDAR elevation
source data from the US Geological Survey (USGS) “The
National Map” (TNM) Download Manager service (U.S.
Geological Survey, 2016). This data comes in the form of CVS
point files, which can then be converted into a raster file in
ArcGIS. Once the LiDAR data is in raster form, the “Area
Solar Radiation” tool in ArcGIS can convert that data into a
solar insolation raster, as shown in below in Fig. 4. Given that
these raster values stem from the LiDAR data, they take into
account shading from trees and other obstructions. For
buildings with pitched roofs, the process also reflects that
south-facing surfaces receive higher average annual insolation.
The solar insolation raster data can then be converted back into
point data, showing the average annual solar insolation for a
given area, in terms of watt-hours of solar radiation per year. Figure 5. Distribution of low, medium, and high Insolation on buildings

The final steps of this part were to identify the potential

amount of distributed PV that could be installed in the study
area and the ensuing amount of potential electricity production.
We first assumed that PV would only be installed buildings that
fall into the “high-insolation” category. This meant that 510 of
the residential buildings (36%), 119 of the small commercial
buildings (30%), and 26 of the large commercial and industrial
buildings (34%) would be eligible for PV installations.
Aggregating the values per building to determine the total
potential installed solar power capacity and electricity
generation in the study area, by building type category shows
that that the total potential PV coverage of 24,830 kW is equal
to 107% of the total sub-station area peak load. The potential
annual energy production of 32,310 MWh is equal to 28% of
the estimated annual energy demand of 114,758 MWh for the
Figure 4. Solar insolation raster values in study Area entire substation service area. These figures indicate that a
substantial portion of the study area’s electricity demands could
At this point, the average insolation could be calculated for be met through local distributed solar power, assuming that
each building. However, simply calculating the average distribution grid impacts could be mitigated. By considering
insolation for an entire building surface would under-estimate high-insolation buildings we can assume buses with the
the potential of a PV array placed on that building, as in the real potential to install PV systems.
world PV arrays are only placed in optimal locations that will
receive adequate insulation to maximize their cost- TABLE I. PERCENT OF BUILDING COVERAGE FOR ALL CATEGORIES
effectiveness. Therefore, a query applied so that the insolation Percent of Building Coverage with Insolation
points layer would only display points with an average solar Building Type above 4.6 kWh/m2/day
insolation of greater than 4.6 kWh/m2/ day. This threshold value Low Medium High
produced clearly discernable patterns of high-insolation points. Residential 0 – 15% 16 – 33% 34% or above
Next, the percentage of each building that is covered with those Small Commercial 0 – 15% 16 – 33% 34% or above
Large Commercial /
high-insolation points is calculated, and divided the buildings Industrial
0 – 25% 26 – 40% 41% or above
into three categories based on their density of high-insolation
coverage, as shown in Table 1. Figure 5 shows the distribution
of low, medium, and high-insolation buildings (defined by the C. Electrical parameters
density of high-insolation points on each roof) within a selected For calculating reactive power profile for each bus,
neighborhood of single-family attached housing. The map statistical evaluation of power factor for loads from given sub-
demonstrates that the buildings with minimal shading and east- station hourly data is performed. It is found that the power
west orientation (i.e., those with south-facing rooftops) are factor for aggregated loads follows a normal distribution with
more likely to be designated as high-insolation in this µ=0.8233 and σ= 0.0193. Next, given the active demanded
methodology. power for each load, the reactive power profile is calculated
according to random power factors generated from a normal
distribution. For the electrical network, line parameters are
calculated based on the estimated distance between buses, and
conductor electrical characteristics. The schematic diagram of
the electric networks is shown in Fig.6.
 0.95 ≤ |𝑣𝑖 | ≤ 1.05 (11)
where ℒ𝑃𝑉 = [ℓ1 , ℓ2 , … , ℓ𝑛 ]𝑇 ℓ𝑖 ∈ {0,1} is PVDG
location vector, ω is weighting factor,
𝑚𝑎𝑥
𝑃𝑃𝑉 = [𝑃1𝑃𝑉 , 𝑃2𝑃𝑉 , … , 𝑃𝑛𝑃𝑉 ] is PVDG maximum
capacity vector, 𝑌𝑏𝑢𝑠 is network admittance matrix, 𝑖 is
vector of bus injected current, 𝑣 is bus voltage vector, 𝐼(𝑡)
is solar insolation at time t, 𝑃̃ 𝑚𝑎𝑥
𝑃𝑉,𝑖 is PV installation limit
𝑡
for bus i derived from solar data analysis, and 𝑖𝑖𝑗 is
current flowing from bus i to j at time t, L is the number of
𝑚𝑎𝑥
lines, n is the total number of buses, and 𝑃𝑠𝑢𝑏 is defined as
substation peak load which is 23,260 kW in the experimented
case. Note that Eqs. (5) and (6) are network constraints
enforced by AC power flow and network operation constraints,
respectively. Eqs. (7-9) are PVDG installation constraints.
Figure 6. Schematic diagram of distribution network
With consideration for the effect of temperature, the output
generated power (kW) by PV arrays is determined using the
III. PROBLEM FORMULATION model from [15]. Note that in this study, both direct and diffuse
The main objective of the proposed method is to determine beam irradiance data are considered in the calculation of PV
the optimal placement and sizing of PV systems that minimizes array output.
the multi-objective function of power loss and voltage deviation
subject to distributed PV constraints and operational constraints IV. SIMULATION SCENARIOS AND RESULTS
of a distribution network such as avoiding reverse power flows Four scenarios based on different penetration ratios of PV
in the network. PV penetration ratio is defined based on are considered to study the impacts of PV on voltage and
substation peak load and is as follows current profiles, network power loss, and reverse power flow.
𝑚𝑎𝑥 The Improved Particle Swarm Optimization (IPSO) is used to
∑𝑁
𝑖=1 𝑃𝑃𝑉 𝑖 (1)
solve the objective functions of optimum PVDG placement and
𝛼(%) = ∗ 100
max (∑𝑀 sizing [15].
𝑗=1 𝑃𝐿𝑜𝑎𝑑 𝑗 )
All the simulations and load profiles carried out for June 24
where 𝑃𝑃𝑉 and 𝑃𝐿𝑜𝑎𝑑 are PV panel output power (kW) and
base on the actual occurrence time of maximum substation load.
electrical load demand (kW), respectively. In this paper
It should be mentioned that for this day, total system energy
penetration ratios are considered as 5, 10, 20, and 50%.
consumption is 421.22 (MWh) and total energy loss is
The total real energy loss of radial distribution system can calculated 26.99 (MWh) which is 6.40% of total system energy
be calculated as [8] consumption. Desired system operation voltage is set to 1 p.u.,
24 𝐿 so for modeled system, the voltage deviation (VD) is calculated
𝐸𝑙𝑜𝑠𝑠 = ∑ ∑ |𝑖𝐿𝑡 |2 𝑅𝐿 (2) as 1.8315. All solar potential buses are a candidate to have PV
𝑡=1 𝑙=1
installation and the installation capacity limit is proportional to
where 𝑖𝐿𝑡 is current flowing through line L at time t and 𝑅𝐿 is calculated solar power potential in data analysis section. Also,
resistance of line L. The formulation for voltage profile in this way the physical limitation on rooftop area for buildings
improvement (𝑣𝐷 ) with 𝑣𝑛𝑡 as the voltage of bus n at time t is is taken into account. In order to apply different objectives of
as follow [14]: PVDG systems implementation, all possible objective
24 𝑛 functions are investigated in the simulation section. Therefore,
𝑣𝐷 = ∑ ∑ (𝑣𝑖𝑡 − 1)2 (3) in the comparative performance study, four individual
𝑡=1 𝑙=1
objectives are considered: I) Voltage improvement and energy
The multi-objective function can be formulated as follows, loss reduction with reverse power flow constraint. II) Voltage
the minimum of the objective function implies the best PV-DG improvement and energy loss reduction. III) Energy loss
allocation for minimizing power losses and enhancing reduction VI) Voltage improvement. The total energy loss and
loadability and voltage profiles. voltage improvements for different penetration levels for all
objective functions are presented in Table II. As shown in Table
min𝑚𝑎𝑥
𝑓 = 𝐸𝑙𝑜𝑠𝑠 + 𝜔 ∗ 𝑣𝐷 (4) II, total energy loss is decreased by the deployment of PVDG
ℒ𝑃𝑉 ,𝑃𝑃𝑉
Subject to: systems. However, the analysis shows that voltage deviation for
(5) the network is not a strictly decreasing function and its
𝑓(𝑃𝐿 , 𝑃𝑃𝑉 , 𝑣|𝑌𝑏𝑢𝑠 ) = 0
minimum occurs at 20% penetration ratio. This means,
𝑖 = ℎ(𝑣|𝑌𝑏𝑢𝑠 ) (6)
increasing penetration ratio of PVDGs always would not
𝑃𝑃𝑉 (𝑡) = 𝑔(𝑃𝑚𝑎𝑥
𝑃𝑉 , 𝐼(𝑡)) (7) necessarily result in improved voltage profiles for distribution
0 ≤ 𝑃𝑚𝑎𝑥
𝑖 ≤ ̃
𝑃𝑚𝑎𝑥
𝑃𝑉,𝑖 (8) network and for penetration ratios above a certain threshold,
𝑇 𝑚𝑎𝑥 there would be voltage quality and reverse power flow issues
1 𝑃𝑃𝑉 ≤ 𝛼. 𝑃𝑚𝑎𝑥𝑠𝑢𝑏 (9)
𝑡
𝑖𝑖𝑗 ≥0 ∀𝑖 <𝑗 (10) that need to be addressed. It is worthwhile to note that all the
buses for PV installation are selected from PV potential buses challenges to the insertion of distributed generation, especially
derived in data analysis section. to the use of photovoltaic technology, is the utilization of its
There are all types of residential, small commercial and benefits without losses in reliability and with satisfactory
large commercial buses in the optimum solution. However, operation of electrical power systems. In this context, voltage
large commercial buses because of large PV installation profile along the feeders and magnitude and direction of current
capacity and alignment of their profile with solar irradiance through lines are of great importance. Therefore, The main
profile, play an effective role in the optimum solutions. Voltage contributions of this study is building a simulation model for
violation is considered as a constraint for all PV installation the local distribution network based on obtained load profiles,
scenarios, so there is not any violation in voltage. Fig. 7 shows GIS information, solar insolation, feeder and voltage settings,
voltage profiles for all PV penetration ratio. and define the optimization problem of solar PVDG installation
TABLE II. TOTAL ENERGY LOSS AND VOLTAGE DEVIATION FOR ALL to determine the optimal siting and sizing . Optimum solutions
SCENARIOS is found for different penetration levels with different objective
functions. As the future extension of this study we can consider
Voltage
Objective
Energy loss
deviation Random
annual analysis along with economic analysis to suggest best
α strategy for PVDG allocation.
function # Imp. installation
kWh
(%)
𝑣𝐷 Imp. (%)
REFERENCES
I 26,081 3.39 1.7372 5.15 Energy loss [1] M. M. Haque and P. Wolfs, “A review of high PV penetrations in LV
II 26,067 3.44 1.7375 5.13 26,358 distribution networks: Present status, impacts and mitigation
5%
III 25,996 3.70 1.7325 5.41 𝑣𝐷 measures,” Renew. Sustain. Energy Rev., vol. 62, pp. 1195–1208,
VI 26,083 3.38 1.7250 5.81 1.7688 2016.
[2] M. J. E. Alam, K. M. Muttaqi, and D. Sutanto, “A comprehensive
I 25,837 4.29 1.6759 8.50 Energy loss assessment tool for solar PV impacts on low voltage three phase
II 25,216 6.59 1.6701 8.81 25,914 distribution networks,” 2012 2nd Int. Conf. Dev. Renew. Energy
10%
III 25,233 6.53 1.6711 8.76 𝑣𝐷 Technol., pp. 1–5, 2012.
VI 25,289 6.32 1.6628 9.21 1.6927 [3] G. Guerra, J. a Martinez, and S. Member, “A Monte Carlo Method for
I 24,092 10.76 1.6102 12.08 Energy loss Optimum Placement of Photovoltaic Generation Using a Multicore
Computing Environment,” PES Gen. Meet. Conf. Expo. 2014
II 24,167 10.48 1.6192 11.59 25,187
20% IEEE. IEEE, pp. 1–5, 2014.
III 23,798 11.85 1.5858 13.42 𝑣𝐷 [4] T. Tayjasanant and V. Hengsritawat, “Comparative evaluation of DG and
VI 24,001 11.09 1.5848 13.47 1.6494 PV-DG capacity allocation in a distribution system,” Proc. Int.
I 22,998 14.81 1.6295 11.03 Energy loss Conf. Harmon. Qual. Power, ICHQP, pp. 293–298, 2012.
II 22.268 17.51 1.6218 11.45 23,338 [5] C. Wang and M. H. Nehrir, “Analytical Approaches for Optimal
50% Placement of Distributed Generation Sources in Power Systems,”
III 22,131 18.02 1.6228 11.40 𝑣𝐷 IEEE Trans. Power Syst., vol. 19, no. 4, pp. 2068–2076, Nov.2004.
VI 22,428 16.92 1.6206 11.52 1.6306 [6] A. Mohamed Imran and M. Kowsalya, “A new power system
reconfiguration scheme for power loss minimization and voltage
profile enhancement using Fireworks Algorithm,” Int. J. Electr.
Power Energy Syst., vol. 62, pp. 312–322, 2014.
[7] S. Ghosh, S. P. Ghoshal, and S. Ghosh, “Optimal sizing and placement of
distributed generation in a network system,” Int. J. Electr. Power
Energy Syst., vol. 32, no. 8, pp. 849–856, 2010.
[8] A. Medina, J. C. Hernández, and F. Jurado, “Optimal placement and
sizing procedure for PV systems on radial distribution systems,”
2006 Int. Conf. Power Syst. Technol. POWERCON2006, 2007.
[9] M. Dixit, P. Kundu, and H. R. Jariwala, “Optimal Placement of PV Array
in Distribution System for Power Loss Minimization Considering
Feeder Reconfiguration,” 2016.
[10] Y. Chen, M. Strothers, and A. Benigni, “A Stochastic Approach to
Optimum Placement of Photovoltaic Generation in Distribution
Feeder,” 2016.
[11] A. Nguyen et al., “High PV penetration impacts on five local
distribution networks using high resolution solar resource
assessment with sky imager and quasi-steady state distribution
system simulations,” Sol. Energy, vol. 132, pp. 221–235, 2016.
[12] D. Q. Hung and N. Mithulananthan, “Loss reduction and loadability
enhancement with DG: A dual-index analytical approach,” Appl.
Energy, vol. 115, pp. 233–241, 2014.
Figure 7. Voltage profile of feeder F for all scenario [13] “Index of /datasets/files/961/pub.” [Online]. Available:
http://en.openei.org/datasets/files/961/pub/.
V. CONCLUSION AND RECOMMENDATIONS [14] S. Kucuksari et al., “An Integrated GIS, optimization and simulation
framework for optimal PV size and location in campus area
Optimal allocation of PVDG with different objective environments,” Appl. Energy, vol. 113, pp. 1601–1613, 2014.
functions including voltage profile improvement and energy [15] M. H. Athari and M. M. Ardehali, “Operational performance of energy
loss reduction is presented in this paper. The optimization storage as function of electricity prices for on-grid hybrid
renewable energy system by optimized fuzzy logic controller,”
problem is solved considering both individual objective
Renew. Energy, vol. 85, pp. 890–902, 2016.
function and multi-objective functions. One of the greatest