Computers and Electrical Engineering 85 (2020) 106676

Contents lists available at ScienceDirect

Computers and Electrical Engineering

journal homepage: www.elsevier.com/locate/compeleceng

Alternative Approaches for Analysing the Impact of

Distributed Generation on Shunt Compensated Radial
Medium Voltage Networks
I. Molver, S. Chowdhury
University of Cape Town, Cape Town, South Africa

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates the impact of distributed generation (DG) on the output of shunt
Received 26 June 2019 capacitors using power system analysis (PSA) tool DIgSILENT Powerfactory. The study then
Revised 3 December 2019
investigates two non-PSA techniques and one alternative PSA technique which makes use
Accepted 6 April 2020
of Load Flow Sensitivities (LFS) to perform voltage assessment and assess the effect of DG
on medium voltage (MV) shunt capacitors. The use of non-PSA techniques is crucial to
Keywords: utilities in developing countries like those found in sub-Saharan Africa. Globally the use
Distributed generation of PSA tools are part of standard practice and the use of LFS can be shown to reduce the
shunt connected capacitors steps involved in voltage analysis and provides additional benefit to using PSA tools. In this
DIgSILENT Powerfactory paper, authors use South African MV capacitor placement methodologies implemented by
non-PSA the national energy utility, Eskom.
voltage
sensitivities © 2020 Elsevier Ltd. All rights reserved.

1. Introduction

Shunt compensation on distribution networks is used in industry for the reduction of technical losses, power factor and
voltage improvement [1,2], where studies show that the contribution of technical losses can be as high as 13% of power pro-
duced [1]. There are numerous techniques outlined by research of the optimal placement of capacitors on a network. Some
of these techniques in [1] include Numerical, Analytical, Heuristic, Programming and Artificial Intelligence (AI), where AI
techniques include Genetic Algorithms, Simulation Algorithm, Particle Swarm Optimization (PSO), Artificial Neural Network
(ANN), Fuzzy and Ant Algorithm techniques [1, 3].
Literature shows that the voltage regulation of a capacitor is proportional to the voltage squared at the terminals of the
capacitor [4], hence the injection of reactive power is influenced by the voltage and not necessarily equal to the rated re-
active power. In [5] it is shown that power injected/absorbed by DG in a network influences the voltage, where capacitor
and DG placement are independent of each other. Continual optimisation of networks can result in the location of network
devices like capacitors to be moved. Due to the unpredictable movement and location of network devices, the entire voltage
profile of the network needs to be analysed when assessing the voltage impact of DG on a network and shunt compensa-
tion. For the assessment of MV network voltage, two non-PSA techniques and one PSA technique is investigated. To assess
the performance of each technique the results are compared to conventional PSA studies. This will test their accuracy and
usefulness to industry. Conventional PSA studies are completed using DIgSILENT Powerfactory.

E-mail addresses: [email protected] (I. Molver), [email protected] (S. Chowdhury).

https://doi.org/10.1016/j.compeleceng.2020.106676
0045-7906/© 2020 Elsevier Ltd. All rights reserved.
2 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

The first non-PSA technique uses formulae from [6]; these formulae are further tested under varying network conditions
and used to create voltage profiles for visualization. The second non-PSA technique is derived in this paper using rudimen-
tary voltage regulation formulae and tested under similar conditions.
The use of non-PSA techniques has its place in developing countries. In an article published on The World Bank website
(worldbank.org) titled “Making Power Affordable for Africa and Viable for Its Utilities”, a World Bank study was conducted on
39 sub-Saharan countries to recognize what factors would influence the financial viability of power utilities and aspects im-
pacting the affordability of energy. Statistically access to electricity in Africa is alarming where the availability of electricity
is low, leaving one in three Africans without electricity. Impoverished power providers are burdened with aging infrastruc-
ture and do not have the ability to service customers in a reliable manner. It is projected that should nothing be done more
Africans will be without power by 2030 [7]. There are many factors that form part of a solution to try and make power
utilities in developing countries, like those in sub-Saharan Africa, more financially viable where reducing overhead costs is
one of those factors. The development of non-PSA tools having the ability to perform rudimentary power system analysis
aids in reducing utility overhead costs in developing countries.
Currently in South Africa and possibly sub-Saharan Africa there is an aging skill base with exceedingly large young, entry
level engineers and technicians. For skilled engineers and technicians, the balance between meeting deadlines and training
new graduates can be a tough task where the latter often falls short. In an article published in The Guardian titled “Africa’s
shortage of engineering skills ‘will stunt its growth’”, it states that Africa faces a large deficit of engineering skills and it
requires a growth spurt with regards to this skill base [8]. This will further create pressure on the existing engineering skill
base to upskill/mentor the number of growing graduates potentially decreasing productivity. Training tools developed by
non-PSA techniques can be a potential solution to this problem and used to assist in skilling up young professionals before
transitioning to PSA software without having to invest in additional PSA software licenses.
To be inclusive there are African utilities and utilities globally where the use of PSA tools is part of standard practice.
This paper further explores the use of sensitivity analysis to aid/supplement power system studies and planning. The im-
pact on system performance caused by an electrical disturbance can be assessed using sensitivity analysis. The sensitivity
parameters can be defined as “dϕ /dP”, “dϕ /dQ”, “dV/dP” and “dV/dQ” which quantifies the power angle and voltage change
with respect to variations in active and reactive power. At a transmission level, the sensitivity margin is determined using
sensitivity analysis [9]. The voltage sensitivity factor is used to check the system voltages with respect to loading where a
high sensitivity means small changes in loading will cause large changes in voltage magnitude and ultimately indicates the
weakness of the bus [10]. In [9] voltage sensitivity coefficients with respect to active and reactive power changes is defined
and quantified using an analytical tool represented by closed-form mathematical expressions. A future development of the
study in [9] is to compare the technique to different models. The aim of the study in [9] was to assess the likelihood of
finding useful expressions to integrate into software tools to assess the possible contribution of DG to voltage regulation
by controlling their power exchange with the MV network. Today, voltage sensitivity results are available using PSA tools.
One aspect of this paper investigates the use of these readily available results for voltage analysis with specific focus on
DG voltage influence and its impact on capacitor output. These results are compared for accuracy to simulated results using
DIgSILENT Powerfactory. Additionally, a formula to use these sensitivity results is given which can be used in the develop-
ment of external tools.
For the investigation, Eskom capacitor placement methodologies in [2, 11] are used to align this study to industry prac-
tice. The impact of variable network conditions on the three techniques is shown. The study is arranged such that a base
case is formed using results from conventional a PSA technique to assess the impact of DG on MV capacitors. The results
produced by the three alternative techniques are then analysed and compared to the base case. The first non-PSA technique
(A) is the Point Voltage Analysis Technique (PVAT), the second non-PSA technique (B) is the Sequential Voltage Analysis
Technique (SVAT) and thirdly the alternative PSA technique (C) is the LFS technique.
The main body of the paper is organised into eight (8) sections. The first section introduces the research topic. Section
(2) discusses voltage rise in a DG-integrated network. Section (3) discusses capacitor placement practice in South African
networks. Section (4) discusses uniform and varying network feeder loading in the context of this work. Section (5) deals
with development of PVAT, SVAT and LFS techniques while section (6) presents the results. Section (7) presents the discus-
sion of results comparing the performance of PSA and non-PSA tools. Section (8) presents the conclusion focussing on the
benefits of the proposed techniques and makes future recommendations for improving the techniques. Finally, sections (9),
(10) and (11) present the Acknowledgement, References and Vitae of the authors.

2. Voltage Rise in a Network

In [5], it is shown that in the case of DG/power injection, the voltage rise is linear to point of DG/power injection and
constant beyond the point of injection. This is illustrated in Fig. 1, where the influence of voltage caused by DG injection is
governed by Equation (1). The factors influencing voltage are impedances and power flow, where the transmission network
has a high X/R ratio (strong network) and the distribution network usually has a low X/R ratio (weaker network) [12, 13].

λ RVnom
PDG
λ ≤ λDG
VDG =
2
(1)
λDG RVnom
PDG
2 λ > λDG
where,
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 3

Fig. 1. Typical voltage rise caused by DG injection

VDG is the per unit change in voltage produced by DG injection [p.u.]

R is the resistance of the line in []
PDG is the injected active power in [kW]
Vnom ( and Unom in Fig.1) is the nominal voltage of the line in [kV]
λ is location along the feeder (0 to 1) expressed as fraction of total feeder length
Network characteristics such as mechanical inertia and impedance influence the grid strength [13], measurable by the
short circuit ratio (SCR). SCR is an indication of the strength of a system at a particular point. Grid strength can influence
voltage variations where stronger grids tend to have lower changes in voltage when subjected to a power disturbance [12].
The probability of strong grids to be affected by flicker and harmonics is lower [13].

3. Capacitor Placement

Capacitor placement by South African utility Eskom is informed by [11] and largely represents the analytical approach.
Capacitors are placed on a network where the reactive power consumption exceeds 300kVAr at an approximate location
two-thirds from the feeder source. Further investigation in [2] has showcased improved capacitor placement methodologies
and influencing methodologies used in the utility by providing a more precise method to better justify the location of the
capacitor with regards to reactive power flow in a network. The reactive power distribution (RPD) is given the name β in
this paper, where values vary from 0 to 1. A uniform RPD is defined by β =0 and reactive power concentrated at the end of
the feeder is defined by β =1. The value of β ensures the selection of an optimal location for a capacitor, which can vary
from 0.67 (two-thirds down the line) to 1 (end of the line). A value of β ≥0.4 suggests a capacitor be placed at the tail end.
The capacitor size is also selectable using this technique. This method allows for assessment of capacitors to be placed on
network for voltage support neglecting losses.
In [12] it is shown that grid strength impacts the extent of voltage variation caused by power disturbances, where weak
grids exhibit higher voltage variations with power disturbances. Typically, higher voltage variations are seen towards the end
of the feeder. Comparing the study in [12], average stiff grids are applicable to Eskom MV networks. Capacitors on Eskom
network would typically be placed at any location from a point two-thirds down the line to the end of the feeder [2], hence
these two locations are used to place capacitors in the following study.

4. Network Loading and Load Distribution

Network loading is the amount of load consumed by the network and load distribution is the dispersal/clustering of the
load. Fig. 2 illustrates that network loading can be varying and may differ from feeder to feeder making it unpredictable. The
data shown in Fig. 2(a) is extracted from on field statistical metering and typical After Diversity Maximum Demand (ADMD)
information from Geo-Based Load Forecasting (GLF) tools. Fig. 2(b) shows yearly data from on field statistical metering.
Due to the varying nature of load consumed, two scenarios were used i.e. High Load – High Generation and Low Load –
High Generation. Fig. 3 illustrates the concept of load distribution, where Fig. 3(a) shows a uniformly distributed feeder
and Fig. 3(b) a non-uniformly distributed feeder with linearly increasing load [14]. The type of customer and location of
these customers impact the loading and its distribution. Eskom Urban type networks have a dense cluster of customers
supplied by short MV feeder and exhibit large low voltage (LV) drops. Eskom Rural feeders supply single customers or
small clusters of customers using long MV feeders where most of the voltage drop occurs on the MV [14, 15, 16, 18]. The
network attributes that impact voltage is mainly power flow, conductor size and length [17]. For the studies, varying network
impedance characterized by varying conductor sizes is used as this is a more practical representation of the feeder model.
Fig. 4 illustrates how load consumed varies on a network where 42 MV feeders were used to show the illustration. The
fall/drop in the curve illustrates load being consumed at that point on the network. Using Fig. 4, power flow for SVAT can be
4 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

Fig. 2. Varying loads: (a) Daily and (b) Yearly

Fig. 3. Notion of a radial feeder showing (a) uniform and (b) non-uniform load distribution [14]

Fig. 4. Actual load distribution of 42 MV feeders

visualised for a radial feeder with no DG and no capacitors on the network. Information is derived from on field statistical
metering.

5. Impact of DG on Shunt Capacitors

For the investigation studies were completed on test network shown in Fig. 5. Earlier it is stated that capacitors are to
be assessed at two locations i.e. two-thirds and tail end. These two locations represent buses 20 and 29 The base case is
established using DIgSILENT Powerfactory where the results are used to benchmark the results of the second part of the
study i.e. investigation of PVAT and SVAT non-PSA techniques. Detailed investigation of PVAT and SVAT can be found in the
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 5

Fig. 5. Test Network showing the location of capacitors and DGs

study conducted by Molver and Chowdhury in [16, 18]. The impact of DG on shunt connected capacitors is also assessed
using an alternative PSA technique, named the LFS technique.
To represent a practical network, varying network impedance was considered by selecting the conductor in the first
10km of the feeder to be Fox and the remaining conductor (20km) on the feeder to be Oak. To perform the assessment the
following scenarios are simulated using a uniformly distributed feeder under two loading conditions high load (typical 2.06
MW and 294 kVAr) and low load. Low load is known as network off peak loading and is taken as 20% of peak loading. This
is a typical observation from on field statistical metering. A fixed 300 kVAr shunt connected capacitor and 1.25 MW DG is
used (power factor = 1). The following scenarios were considered in the analysis:

1 No capacitors and no DG are placed on the network.

2 Then a capacitor and DG is placed on the network in isolation. Capacitor placed at buses 20 and 29. DG placed at buses
23 and 28
3 Capacitors assessed with DG on the network. Capacitor placed at buses 20 and 29. DG placed at buses 23 and 28.

For the High/Low Load-Low Generation scenario, it is unlikely that there would be significant voltage variation caused
by DG injection. This can be seen in Equation (2) where low P and Q will produce a small variance in voltage regulation.
(Vs − Vr ) (RP + XQ )
VR = ≈ (2)
Vr Vr 2

5.1. Simulation using DIgSILENT Powerfactory

For the simulation using DIgSILENT Powerfactory, conventional power system analysis is completed. For each scenario
mentioned in 4 capacitors and DGs are put into service at the relevant locations for the scenario and a simulation is run. At
the capacitor terminals the capacitor output and voltage are recorded. A voltage profile is also available for viewing.

5.2. Calculation using PVAT

The equations that form PVAT are found in literature [6], where Equations (3), (4), (5) and (6) are used to develop the
methodology to use PVAT [6, 5]. These equations have the ability to calculate the voltage at any location on a feeder without
requiring prerequisite voltage calculations. How these equations are used is narrated next to each equation. It is expected
that the resultant voltage will rise as active power injected into a network and flowing through a resistance will cause the
voltage to increase [5]. Should active/reactive power be absorbed by the DG, power flow changes and the generator behaves
like a load. In this case voltage depreciation is expected and P or Q is negative.

Uniform Z: Calculates the change in voltage at any location λ:

Result is subtracted from 1.03 p.u (Eskom regulation voltage)
v RP + XQ
 1

V ≈ = 2
. λ − λ2 (3)
Vnom Vnom 2

where

V is the change in voltage at λ [p.u.]

v is the change in voltage at λ [V]
6 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

Vnom is the nominal voltage of the feeder [V]

Uniform Z: Calculates the change in voltage caused by the injection of active power P (+). Q component is also available
as an additional term (λ XQ
2 ): Vnom
The result is added to the previous result from Eq. (3)

λ RVnom
PDG
λ ≤ λDG
VDG =
2
(4)
λDG RVnom
PDG
2 λ > λDG
Non-uniformh Z: Calculates the change in voltage at any location λ:
Result is subtracted from 1.03 p.u. ((Eskom regulation voltage)
RInλ P + XInλ Q 1
 
V ≈ 2
1− λ (5)
Vnom 2
Non-uniform Z: Calculates the change in voltage caused by the injection of active power P (+). Impedance up to λ is
used.
The result is added to the previous result from Eq. (5)
 RInλ P+XInλ Q
2 λ ≤ λDG
VDG ≈ RInλ
Vnom
P+XInλ Q (6)
DG
2
Vnom
DG
λ > λDG
Non-uniform Z: Calculates the change in voltage caused by the injection of reactive power Q (+). Impedance up to λ is
used.
The result is added to the previous result from Eq. (5)
 XInλ Qcap
λ ≤ λcap
Vcap ≈ Vnom
XInλcap Qcap (7)
Vnom
λ > λcap
where

Qcap is the Reactive Power output of the capacitor/DG [kVAr]

Xlnλ is the reactance of the feeder to the point λ [Ω]
Xlnλcap is the reactance of the feeder to the point of the capacitor/DG connection [Ω]
Vnom is the nominal voltage of the network [kV]
Vcap is the change in voltage due to the capacitor/DG at the point λ [kV]

As mentioned PVAT can perform point calculations of voltage which is beneficial if the voltage at a specific point is
required. Generally power system engineers would require visualising a voltage profile. In order to create a voltage profile
using PVAT, the voltage along points/location of the network would need to be calculated and plotted against the distance
of the feeder.
Fig. 6 illustrates the process of using PVAT in a flow diagram, i.e. sequentially calculating the voltage at each point on the
feeder to visually form a voltage profile. Initially the feeder is screened to check which formulae are required for the voltage
calculation. When the flow chart reaches “Calculate the voltage @ λ”, the algorithm loops to calculate the voltage at the
next location along the feeder. The loop will continue to calculate the voltage until the last point on the feeder is reached.
PVAT can calculate the voltage at any point on the feeder without sequentially building the voltage profile, however, to
visually represent the voltage profile it is best to calculate the voltage at each point and then plot the results as a function
of distance.

5.3. Calculation using SVAT. For SVAT, equations in [4] are used to formulate Equation (8) where Equation (8) is used to
calculate the voltage and results are used to display the feeder voltage profile. Equation (8) is an expansion of Equation (2).
To derive Equation (8), the use of power flow theory is used i.e. active/reactive current flowing through a resistive/reactive
element causes the voltage to sag. Power flowing from a capacitor bank or DG unit will cause the voltage to rise locally
on the network and alters the current flow along the network. This injection of active/reactive power also causes power to
flow both up the network and down the network. In order to cater for the voltage rise and depreciation on the network
the direction of power flowing in each segment is noted, where power flowing from the source is positive (+) and power
flowing towards the source is negative (-). This can be a challenge to estimate and was done manually in this approach. The
voltage calculation at each segment is calculated using (V = (±RP±XQ
Vr
)
in V olts). This result is added to the result calculated
in the previous segment, hence the approach being sequential. The approach is sequential as each voltage drop along the
segments needs to be calculated and summated before subtracting from the source voltage to obtain the voltage at any
point along the feeder.
n m  
1 (P − 1 PSm )Rn Dn + n1 (Q − m1 QSm )Xn Dn
Vx = Vs − (8)
Vp−p
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 7

Fig. 6. The process for voltage profile visualization using PVAT

Fig. 7 shows the process to create the voltage profile using SVAT.
Using SVAT simulations were conducted using the same scenarios as those used for the DIgSILENT Powerfactory simula-
tions so that comparisons could be made.
5.4. Impact of DG using Load Flow Sensitivities (LFS). Sensitivity analysis performed by DIgSILENT Powerfactory produces
results for parameters δθδ P , δ P , δ Q and δ Q . These results represent the impact of a 1 MW or 1 MVAr system disturbance. The
δV δθ δV
results of interest in this study are dV/dP and dV/dQ.
Detailed analysis of the LFS method can be found in the study in [19] by Molver and Chowdhury, where the key results
of that study indicate that the results of LFS is valid for varying loading conditions; the result can be used to estimate, using
linear extrapolation, power disturbances that are not equal to 1 MW or 1 MVAr. LFS result can also be used to evaluate the
hosting capacity at any location on the network and also it is shown that LFS can be used to complete voltage analysis faster
when compared to conventional PSA techniques. Additionally, LFS can be used to in planning tools to predict the impact on
voltage caused by power injection by capacitors and DG units.

6. Results

6.1. Simulation using DIgSILENT Powerfactory

The results tabulated in Table 1 and Table 2 show the results of the capacitor output (in kVAr), placed at segments 20
and 29 under NO DG, DG connected at bus 23 and DG connected at bus 28. The impact on the voltage at the location of
the capacitor is also shown.
8 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

Fig. 7. The process to calculate the voltage in each segment to produce a voltage profile using SVAT

Table 1
High Load – High Generation Scenario

Cap Position DG Position Rated Power [kVAr] Cap Output [kVAr] Voltage (at cap) [kV]

20 NO DG 300 297 21.889

20 23 300 315 22.545
20 28 300 315.2 22.552
29 NO DG 300 296.5 21.870
29 23 300 316.1 22.579
29 28 300 318.7 22.671

Table 2
Low Load – High Generation Scenario

Cap Position DG Position Rated Power [kVAr] Cap Output [kVAr] Voltage (at cap) [kV]

20 NO DG 300 316.8 22.607

20 23 300 334.2 23.219
20 28 300 334.4 23.226
29 NO DG 300 317.8 22.643
29 23 300 336.7 23.306
29 28 300 339.2 23.393

The results listed in Table 1 and 2 show that the capacitor kVAr output is impacted by the voltage of the network. Using
the relation Q∝V2 and linear extrapolation it can be shown that the capacitor output can be estimated from Equation (9).
That means Equation (9) will yield similar results for the capacitor output knowing the voltage at the capacitor terminal.
This shows that it is adequate to use linear extrapolation to estimate the impact of DG on shunt capacitors at any point on
the network.
Vcap 2
Qcap_out = Qcap_rated × (9)
Vnom 2
where,

Qcap_out is the actual reactive power output from capacitor [kVAr]

Qcap_rated is the rated reactive power output from the capacitor at nominal voltage [kVAr]
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 9

Table 3
High Load – High Generation

Cap Position DG Position Cap Rating [kVAr] Voltage [kV] Cap, DG Voltage [kV] NO Cap, DG Voltage Difference [V]

20 23 300 22.545 22.437 107.182

20 28 300 22.552 22.444 107.398
29 23 300 22.582 22.429 153.589
29 28 300 22.671 22.519 151.384

Table 4
Low Load – High Generation

Cap Position DG Position Cap Rating [kVAr] Voltage [kV] Cap, DG Voltage [kV] NO Cap, DG Voltage Difference [V]

20 23 300 23.219 23.110 109.495

20 28 300 23.226 23.116 109.698
29 23 300 23.306 23.148 157.402
29 28 300 23.393 23.235 158.217

Fig. 8. Feeder voltage profile with DG and capacitors – comparison between PVAT and Powerfactory (PF).

Vcap is the voltage at the point of capacitor connection [kV]

Vnom is the nominal voltage of the network [kV]

It is noted that the voltage variation on the first half of the feeder may not be significant to impact capacitor operation
on a radial network. The results in Tables 1 and 2 have established that the rise in voltage caused by DG connected to a
network increases the reactive power output of a capacitor and this can be estimated using linear extrapolation knowing
the rated power of the capacitor, the nominal voltage of the feeder and the voltage at the capacitor terminals. The increase
in voltage at the terminals caused by the DG injection increased the reactive power output of the capacitor. It is expected
that the increase in reactive power increases the voltage support provided by the capacitor. Tables 3 and 4 show the voltage
at the terminals of the capacitor with and without the capacitor connected. The difference is that the voltage is calculated
to show the voltage support provided by the capacitor. Simulations show that the voltage support produced by the capacitor
is between 0.47 % and 0.68 % in the study. The additional voltage support provided by the additional reactive power is small
and can be considered negligible for an MV network.

6.2. Calculation using PVAT

Feeder voltage profiles for each scenario i.e. No Cap-No DG, Cap-No DG, Cap-DG can be seen in Fig. 8, where the differ-
ence between the simulated Powerfactory (PF) voltage profiles (dashed lines) and the calculated voltage profile (solid lines)
is shown. It can be seen from the voltage profile that the percentage error produced at the initial calculation is carried
through to subsequent calculations.
Tables 5 and 6 show the calculated results where the feeder voltage is calculated using PVAT up to the location of the
capacitor. Using this calculated feeder voltage at the terminals of the capacitor (Vcap ), the capacitor Q output is calculated
using Equation (9). While comparing the results in Tables 5 and 6 to Tables 1 and 2, it is seen that the same effect of voltage
10 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

Table 5
High Load – High Generation Scenario

Cap Position DG Position Rated Power [kVAr] Cap Output [kVAr] Voltage (at cap) [kV]

20 NO DG 300 300 21.976

20 23 300 318.2 22.631
20 28 300 318.2 22.631
29 NO DG 300 300 22.043
29 23 300 319.5 22.750
29 28 300 321.9 22.836

Table 6
Low Load – High Generation Scenario

Cap Position DG Position Rated Power [kVAr] Cap Output [kVAr] Voltage (at cap) [kV]

20 NO DG 300 300 22.623

20 23 300 317.6 23.278
20 28 300 317.6 23.278
29 NO DG 300 300 22.674
29 23 300 319 23.381
29 28 300 321.4 23.467

Table 7
High Load – High Generation

Cap Position DG Position Cap Rating [kVAr] Simulated Cap Output [kVAr] Calculated Cap Output [kVAr] Percentage Variance [%]

20 NO DG 300 297 299.3 0.788

20 23 300 315 317.5 0.779
20 28 300 315.2 317.5 0.715
29 NO DG 300 296.5 301.2 1.577
29 23 300 316.1 320.8 1.489
29 28 300 318.7 323.2 1.427

Table 8
Low Load – High Generation

Cap Position DG Position Cap Rating [kVAr] Simulated Cap Output [kVAr] Calculated Cap Output [kVAr] Percentage Variance [%]

20 NO DG 300 316.8 317.2 0.136

20 23 300 334.2 335.9 0.500
20 28 300 334.4 335.9 0.440
29 NO DG 300 317.8 318.7 0.268
29 23 300 336.7 338.8 0.634
29 28 300 339.2 341.3 0.633

rise on a network impacts the capacitor output. A comparison of the DIgSILENT Powerfactory simulated and PVAT calculated
results are shown in Tables 7 and 8.
Tables 7 and 8 compares the Powerfactory simulated results (Simulated Cap Output) and the calculated results using
PVAT (Calculated Cap Output). While calculating the voltage rise produced by the capacitor using PVAT, the output of the
capacitor was assumed to be 300 kVAr, which can be seen in the NODG case for the calculated results. This initial assump-
tion introduces an additional error between the simulated and calculated results. The percentage error calculated is between
0.9 % and 5.6%. This corresponds to a difference in capacitor output of 3 kVAr and 17 kVAr. To reduce the error of the cal-
culated value, the initial capacitor output was calculated to be proportional to the square of the voltage at the point of the
capacitor using Equation (9) instead of assuming an output of 300 kVAr. It is seen that the maximum percentage error drops
to 1.577 %. This results in a 4.03 % improvement. It can be concluded that PVAT can produce results close to PSA analysis
under uniformly distributed feeders. PVAT can be used easily to perform voltage calculations at any point on the network.
The studies were conducted on a feeder with uniformly distributed loads. The study by Molver and Chowdhury in [16]
exposed PVAT to a non-uniformly distributed feeder. The percentage error between simulated and calculated voltage profiles
(using PVAT) increased for non-uniform load distributions. Fig. 9 shows the calculated voltage profile against the simulated
voltage profiles for different load distributions. Two extreme cases for the load distribution are presented here, keeping the
location of DG at λ=0.93 (called Scenario 2 in [16]). The first case of load distribution is where the 90 % of the load is
concentrated at the end of the feeder (10-90). The second case is where the 90 % of load is concentrated at the start of the
feeder (90-10). Fig. 9 illustrates that PVAT does not have the ability to cater for varying load distributions. In cases where
the feeder has extreme non-uniformly distributed loads, PVAT can still be used for planning purposes where it should be
expected that results have lower accuracies. It is not recommended to use PVAT in scenarios where operational voltage
limits are being reached and high accuracy to make engineering decisions are required.
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 11

Fig. 9. Approximated voltage profile and actual voltages at different load distributions for a feeder with non-uniform network impedance

Fig. 10. Graphical results of the voltage profile produced by SVAT and superimposed with the simulated Powerfactory results

Fig. 11. Feeder voltage profile with DG and capacitors – comparison between SVAT and Powerfactory (PF) for varying load distributions. [16]

6.3. Calculation using SVAT

Fig. 10 and 11 show the calculated voltage profile (solid lines) and the Powerfactory simulated profile (dashed lines). It
is seen that SVAT and PSA techniques produce similar results for voltage. Fig. 10 shows the scenarios where the capacitor
is placed at bus 20. With the capacitor in service a No DG scenario and DG scenario are shown, where the DG location is
changed from bus 23 to bus 28. Also, a No Cap-No DG Scenario is shown. Fig. 11 shows that SVAT is capable of following
PSA results for varying load distributions as well. This was confirmed in a study by Molver and Chowdhury in [16]. Using
the calculated voltage from SVAT, the reactive power output is calculated and displayed in Table 9 and 10.
12 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

Table 9
High Load – High Generation Scenario

Cap Position DG Position Rated Power [kVAr] Cap Output [kVAr] Voltage (at cap) [kV]

20 NO DG 300 296.64 21.877

20 23 300 316.13 22.583
20 28 300 318.18 22.657
29 NO DG 300 297.86 21.921
29 23 300 317.31 22.626
29 28 300 319.41 22.701

Table 10
Low Load – High Generation Scenario

Cap Position DG Position Rated Power [kVAr] Cap Output [kVAr] Voltage (at cap) [kV]

20 NO DG 300 316.99 22.614

20 23 300 334.62 23.235
20 28 300 334.22 23.221
29 NO DG 300 316.98 22.614
29 23 300 334.30 23.224
29 28 300 334.55 23.232

6.4. LFS technique

The study in [19] by Molver and Chowdhury shows that the LFS technique is not load dependent. This means that
the results obtained for DG injection at a point in a network remain valid for all loading conditions. This is an advantage
over conventional PSA techniques when calculating the voltage impact of DG injection. The steps taken to perform voltage
analysis on an MV feeder include obtaining a network base case where the existing network status is assessed for both
techniques. A case study in [20] provides an example of how conventional PSA analysis is performed where the steady state
and voltage variation ratio case studies relate to this study. The difference in the two techniques i.e. conventional PSA and
LFS analysis, is that firstly studies at peak and minimum loading for conventional PSA methods needs to be performed and
the results recorded. Secondly, subtracting the voltage obtained before DG injection and after DG injection to obtain the
voltage variation in a network needs to be performed, which shows the voltage impact to other customers when connect-
ing/disconnecting the DG unit. The LFS technique is shown to be faster as the LFS results just need to be added to the
initial voltage results for the base case and the LFS result can be used as the result for voltage variation i.e. no calculation
is needed. For DG units not injecting 1 MW or 1 MVAr, the injected load just needs to be multiplied by the LFS results as
shown in [19].
In addition to the study conducted in [19], steps were taken in completing a quick rudimentary analysis of 1 MW DG
injection on a test network as indicated in Fig. 5. The difference in voltage results between conventional PSA and LFS analysis
on the network vary by 0.162 % and 0.238 % at peak and minimum load respectively. These results are taken at the point
of DG injection and are acceptable for a high level voltage study and shows that LFS can be used for steady state voltage
analysis.
Additionally, planning tools to supplement PSA analysis can be developed using LFS to easily visualise the impact of
placing capacitors, DG units or a combination of network devices without performing a PSA study, where conventional PSA
methods can be iterative in nature to find optimal planning solutions. In order to obtain a voltage, the new voltage at a
point on a network the Equation (10) can be used. To plot a voltage profile, sequentially calculating the voltage at each
point and plotting the voltage with respect to the feeder distance will provide a visual representation of the voltage of the
network.

dV dV
Vnew = Vold ± P ±Q (10)
dP dQ

LFS analysis is calculated from the busbar where the DG is injected. The LFS results provide the voltage impact on all
busbars or point terminals in the casefile. This allows for the assessment of the impact of DG injection on voltage at a
location anywhere on the feeder. Hence, the voltage impact (V) caused by DG injection on capacitors and other network
devices can easily be identified. The LFS result needs to be converted from p.u. to volts by multiplying the LFS result by
the nominal voltage. This is then added to the voltage in the NO DG scenario using Equation (10) to obtain the expected
voltage rise at the capacitor terminals. In the case of the impact on capacitors, Equation (9) can be used to calculate the
impact of DG injection on capacitor output at steady state. Tables 11 and 12 show the results of the impact of DG on a
shunt connected capacitor at two different locations under two network loading conditions.
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 13

Table 11
High Load – High Generation Scenario

Cap Position DG Position Cap Output [kVAr] Voltage (at cap) [kV] LFS Results (at cap) p.u./MW

20 NO DG 297 21.889 -
20 23 316.23 22.448 0.02541
20 28 316.31 22.587 0.02551
29 NO DG 296 21.873 -
29 23 317.37 22.628 0.02746
29 28 319.94 22.719 0.03078

Table 12
Low Load – High Generation Scenario

Cap Position DG Position Cap Output [kVAr] Voltage (at cap) [kV] LFS Results (at cap) p.u./MW

20 NO DG 316.8 22.607 -
20 23 336.68 23.306 0.02364
20 28 336.76 23.309 0.02366
29 NO DG 317.8 22.643 -
29 23 337.84 23.346 0.02556
29 28 340.32 23.432 0.02868

Table 13
Summary of results between conventional PSA analysis and three techniques PVAT, SVAT and LFS

Cap Position DG Position DIgSILENT Powerfactory PVAT (non-PSA) SVAT (non-PSA) LFS (Alternative PSA)
(Conventional PSA)

High Load Low Load High Load Low Load High Load Low Load High Load Low Load

20 NO DG 297 316.8 300 300 296.64 316.99 297 316.8

20 23 315 334.2 318.2 317.6 316.13 334.62 316.23 336.68
20 28 315.2 334.4 318.2 317.6 318.18 334.22 316.31 336.76
29 NO DG 296.5 317.8 300 300 297.86 316.98 296 317.8
29 23 316.1 336.7 319.5 319 317.31 334.30 317.37 337.84
29 28 318.7 339.2 321.9 321.4 319.41 334.55 319.94 340.32

7. Discussion

The impact of DG injection on capacitor output is assessed in this paper where Tables 1 and 2 show the results using PSA
software DIgSILENT Powerfactory. The study then changes the method of obtaining the voltage profile i.e. using two non-PSA
techniques PVAT and SVAT [16] to calculate the voltage profile. SVAT results are shown match those of the PSA analysis. It
is shown that linear extrapolation is also adequate to estimate the impact of voltage rise on capacitor output. Additionally,
a third PSA technique using sensitivity analysis is investigated, where the voltage results for all three techniques were used
to calculate the impact of reactive power output at the capacitor terminals. The results are summarised for comparison in
Table 13. Table 13 shows that between the two non-PSA techniques, SVAT is more accurate in calculating the voltage and
hence results on the impact of the capacitor output is closer to the DIgSILENT Powerfactory simulated result. The results
of the LFS technique closely match the DIgSILENT Powerfactory result and shows that this technique is sufficient for steady
state voltage analysis.
The need for non-PSA techniques and its benefit to developing countries was outlined in the introduction; however,
advantages and limitations are present in both techniques and are summarised also highlighting some of the characteristics
of each technique:

PVAT SVAT

340- PVAT has the ability to calculate the voltage at any point in the 340- The voltage at any point on the network needs to be sequentially
network. This can increase computation times as the number of calculated from the start of the feeder to any point on the
points can be optimised. network where the number of points is dependent on the
340- PVAT equations do not cater for the load distribution of a number of segments.
network and hence the accuracy of PVAT decreases with the 340- Due to the sequential nature of SVAT, the load distribution is
extreme conditions of load distribution. The highest accuracy inherently incorporated in the calculations, hence retaining the
observed is on network with uniformly distributed loads. accuracy of SVAT under varying load distributions. Higher
340- PVAT has multiple equations and options catering to network accuracy when compared to PVAT.
composition but is simple to execute. 340- SVAT is more complex in the sense that power flow in each
segment needs to be calculated and the direction of the power
flow needs to be known. Estimations can reduce the accuracy of
this technique.
14 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

Fig. 12. Change in voltage varies with DG location

An advantage of using PVAT over PSA analysis is that this method can be used to quickly calculate the impact of DG
injection on capacitor output. PVAT and the use of Equation (9) can be incorporated in non-PSA analysis tools to estimate
the impact of DG injection on capacitor reactive power output. These tools can be programmed in Microsoft Excel where
PVAT can be used as a basis for the voltage analysis. The completeness of this tool still needs to be assessed further but
has the potential to provide a cost saving for small scale companies. Analysing the cost for a basic PSA tool, costs for three
PSA packages were attained i.e. ReticMaster 18, PSS®E Version 33.12.1, Nayak PSCAD V4 Professional. The cost for a single
use license is R88 250 with an optional annual maintenance fee of R13 240 pa for RecticMaster 18, a monthly cost of
R47 685.60 for PSS®E Version 33.12.1 and a once of cost of R249 624 for Nayak PSCAD V4 Professional which includes a
6 month maintenance fee. These costs are based on 2019 rand value and dependent on the relevant exchange rate, and
represent a potential cost saving to small scale entities.
The third technique LFS is shown to produce results close to those simulated using DIgSILENT Powerfactory. It is also
shown that LFS can be used in steady state voltage analysis reducing some steps in conventional PSA analysis. Equation
(10) shows that LFS can be used to calculate changes in voltage caused by a change in power on a network. Exporting net-
work parameters and sensitivity results from PSA tools into programs like Microsoft Excel allows for inexpensive, predictive
planning and optimisation tools to be developed.
A comparison of the two PSA techniques is shown highlighting the characteristics showcasing the advantages and limi-
tation.

DIgSILENT Powerfactory (PSA) LFS (PSA)

340- PSA technique 340- PSA technique

340- Results are valid for a specific loading snapshot of the network 340- Results are presented as a voltage change. These results are
in which the simulation was conducted. Simulations will have to independent of load change and hence valid for all loading
be re-conducted when the loading condition changes. conditions.
340- Feeder voltage is readily available. 340- Feeder voltage need to be calculated using Equation (10). This
340- Predictive/scenario planning needs to be manually assessed, can be programmed into external tools.
graphed and recorded. 340- Predictive/scenario planning tools can be developed with the use
of LFS, Equation (10) and PSA results. Results can be
automatically plotted saving time.

In Table 13, at cap position 29 the simulated and calculated cap output show higher values when the DG is placed at
position 28. This is due to the impact of DG injection on voltage being higher when the DG unit is located closer to the tail
end of the feeder. The impact of this higher voltage is seen in the capacitor output when the capacitor is located beyond
the DG unit on a radial feeder. When the capacitor is placed before the DG unit there is no visible impact on the capacitor
output when the DG location is changed. This can be seen in the results where the capacitor is placed at position 20.
This is due to the change in voltage following the same gradient before the DG location. The gradient is dependent in the
impedance of the line when considering a constant DG output. The impedance will only change with infrastructure change
where a conductor change will impact the impedance the greatest. Fig. 12 shows how the change in voltage is impacted by
the change in location of the DG unit and visually shows why a capacitor placed at the end of the feeder will experience
different voltage changes and hence the Q output is affected by DG location.
A shunt capacitor placed at the end of the feeder is more likely to experience a variance in voltage produced by DG
location. The change in voltage produced by the DG unit increases as the location is moved closer to the tail end. The
higher the change in voltage experienced by the capacitor, the higher the impact on reactive power output. Capacitor placed
before a DG unit is not impacted by the change in location of the DG as the voltage at the capacitor terminals does not
change.
 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676 15

The impact of DG on shunt capacitor output could affect the capacitor sizing should the network be weak, and the
voltage variation caused by the DG injection be high. A capacitor placed at the tail end of the network is more sensitive
to voltage fluctuations caused by the DG injection as the network voltage fluctuates more. The impact of DG injection on
the capacitor output may vary and depend on the network operating conditions. DG injection may impact the sizing of the
capacitor depending on the DG injection where the sizing may need to be reduced. The introduction of DG injection to a
network may support the capacitor on the network with growing reactive power needs. In turn, the increased capacitor
output increases the network voltage slightly, limiting the DG injection into the network. The study conducted in this paper
shows that there is an impact of DG on shunt connected capacitors which is not significant on an average stiff grid.
It is noted that there are varying PSA techniques available for DG simulation. These techniques vary in level of details
and are dependent on the available information and the purpose of the study i.e. protection studies, performance stud-
ies, time based studies etc. The level of PSA studies can vary from being rudimentary to comprehensive and dynamic or
quasi-dynamic. Dynamic simulations take large computational effort and involve higher complexities and are not consid-
ered. Comparing rudimentary PSA techniques and LFS analysis is considered and it is found that LSF analysis can be used to
quickly ascertain the impact of DG inject on voltage.

8. Conclusion

Impact of distributed generation on medium voltage shunt-connected capacitor reactive power output is analysed in this
work. Simulation results confirmed that Q∝V2 can be used to determine the impact of distributed generation on shunt
compensation using linear extrapolation. Simulations confirmed that capacitors placed closer to the feeder end experience
a higher voltage change and hence higher reactive power output with distributed generators placed towards the tail end.
The actual voltage change on the network is small. That capacitor reactive power output is affected implies that the sizing
of the capacitor may be affected depending on network loading, strength, distributed generation injection and location and
capacitor location.
The study compares the results calculated from two proposed techniques, viz. (A) Point Voltage Analysis Technique and
(B) Sequential Voltage Analysis Technique to those obtained from the power system analysis tool DIgSILENT Powerfactory. It
is seen that (B) follows the DIgSILENT Powerfactory results. Also (B) requires knowledge of the direction of power flow which
can be difficult to estimate; this is a manual estimation. By knowing the limitations of both techniques, future improvements
can be made to each of them, such as developing and testing an algorithm to estimate the direction of active and reactive
power flow in a radial network to optimise (B) and modifying equations for (A) to better handle varying load distributions.
Once these are improved, they can be programmed into tools and tested further using actual networks.
It is noted that Load Flow Sensitivities analysis technique (C) can be used to quickly ascertain the impact of DG injection
on voltage and voltage variation caused by the connection/disconnection of the unit. The impact of voltage can be assessed
along the entire feeder when using (C), and hence the impact of voltage on other network devices such as shunt connected
capacitors. This paper provides results showing that (C) can adequately perform steady state voltage analysis where Equation
(10) is provided for voltage calculation using Load Flow Sensitivities extending its use for planning and optimisation tools.
Future ideas include using Load Flow Sensitivities in external tools to assist with predictive/scenario planning and incorpo-
rating its use in analysis methodologies to improve simulation efficiencies. Thus, this technique becomes beneficial to all
users of commercial power system analysis software such as DIgSILENT Powerfactory as used in this study.

Acknowledgement

The authors gratefully acknowledge the support and infrastructure provided by Electrical Engineering Department, Uni-
versity of Cape Town and Eskom Holdings Ltd., South Africa for carrying out this research.

References

[1] T. Manglani, Y. S. Shishodia, “A Survey of Optimal Capacitor Placement Techniques on Distribution Lines to Reduce
Losses”. International Journal of Recent Research and Review, vol. 1, pp 1-6, March 2012.
[2] D. Chetty, M.M. Bello and I.E. Davidson, “The Application of Volt/VAr Optimization on Eskom South Africa Distribution
Feeders” Proceedings of the 2016 CIGRE Canada Conference, Vancouver, BC, Oct 17-19, 2016.
[3] S.RamaIyer, K.Ramachandran, S.Hariharan, “Optimal capacitor allocation in power systems”. Computers & Electrical
Engineering Journal, vol. 10, issue 4, pp 247-258, 1983 (In Issue)
[4] N. Jenkins, J.B. Ekanayake, G. Strbac, (2010), “Distributed Generation” (Series 1), London: The Institute of Engineering
and Technology, 2010, Chapter 1.
[5] M. Bollen, F. Hassan. (2011, October 5). Integration of Distributed Generation in the Power System (1st edition). New
Jersey: John Wiley & Sons, 2011, Chapter 5.
[6] J. Nye, “Increasing distributed generation penetration when limited by voltage regulation”, MSc of Engineering, Uni-
versity of Stellenbosch University, South Africa (Western Cape), April 2014.
[7] “Making Power Affordable for Africa and Viable for Its Utilities”, The World Bank, https://www.worldbank.org/en/
topic/energy/publication/making- power- work- for- africa, 2019, [Nov. 2019].
16 I. Molver and S. Chowdhury / Computers and Electrical Engineering 85 (2020) 106676

[8] “Africa’s shortage of engineering skills ‘will stunt its growth”, The Guardian, https://www.theguardian.com/
global- development/2016/sep/14/africa- shortage- of- engineering- skills- and- female- students- will- stunt- its- growth, 14
September 2016, [Nov. 2019]
[9] S. Conti, S. Raiti, S, G. Vagliasindi, “Voltage sensitivity analysis in radial MV distribution networks using constant
current models”. IEEE International Symposium on Industrial Electronics. pp 2548 – 2554, August 2010.
[10] J. S. Bhadoriya, C. K. Daheriya, “An Analysis of Different Methodology for Evaluating Voltage Sensitivity”, International
Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, vol. 3, Issue 9, pp 12239-12246,
September 2014.
[11] S. Malapermal, “Network Planning Guideline for Shunt Capacitors”, Eskom South Africa, DST34-2201, October 2010.
[12] T. Sarkar, A. K. Dan, A. Ghosh. “Effect of X/R ratio on low voltage distribution system connected with constant speed
wind turbine”, in 2016 2nd International Conference on Control, Instrumentation, Energy & Communication (CIEC), 2016.
[13] A. Golieva, “Low Short Circuit Ratio Connection of Wind Power Plants”, Master Thesis, Norwegian University of
Science and Technology, Trondheim, Norway, 2015.
[14] T Gönen (2014): Electric Power Distribution Engineering "Application of capacitors to distribution systems": CRC
Press, Boca Raton, FL,3rd Edition, pp 299-306.
[15] G.Carter-Brown , C.T. Gaunt, “Voltage Management by the Apportionment of Total Voltage Drop in the Planning
and Operation of Combined Medium and Low Voltage Distribution Systems”, South African Institute of Electrical Engineers
Journal, vol. 97, issue 1, pp 66-73.
[16] I. Molver, S. Chowdhury, “Comparison of Voltage Calculation Techniques for Varying Load Distribution in DG-
integrated Networks”, Proc. of the IEEE Power Africa Conference, 2018, Cape Town, South Africa.
[17] S. W. Blume (2007, April 11). Electrical Power Systems Basics. (1st edition) New Jersey: John Wiley & Sons, 2011,
Chapter 5.
[18] I. Molver , S. Chowdhury, “Impact of Distributed Generation Penetration on Radial networks with Shunt Compensa-
tion”, Proc. of the 10th International Renewable Energy Congress (IREC 2019), March 2019, Sousse, Tunisia.
[19] I. Molver, S. Chowdhury, “Investigation of the impact of Grid-integrated Distributed Generation on MV Network
Voltage using Load Flow Sensitivities”, Proc of the IEEE Power Africa Conference, 2019, Abuja, Nigeria.
[20] Cheng-Ting Hsu, Roman Korimara, Tsun-Jen Cheng, “Power quality analysis for the distribution systems with a wind
power generation system”, Computers & electrical Engineering Journal, vol. 54, pp 131-136, August 2016 (In Issue).
11. Vitae
Ivani Molver received her BSc. Electrical Engineering Degree from the University of KwaZulu Natal, Durban, KZN, South
Africa in 2012. She was an Engineer in Training from January 2013 to September 2014. She is currently a Network Develop-
ment Planning Engineer in Eskom Holdings Ltd. Registered as a Professional Engineer with ECSA. Research interests include
the integration of renewable energy in power systems
Sunetra Chowdhury received her PhD from Jadavpur University, Kolkata, India in 1998. She is currently Associate Pro-
fessor in Electrical Engineering Department and Deputy Dean (Undergraduate Studies) in the Faculty of Engineering and the
Built Environment in University of Cape Town, South Africa.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.compeleceng.
2020.106676.