Vous êtes sur la page 1sur 9

Voltage unbalance assessment in secondary radial distribution networks

with single-phase photovoltaic systems


F.J. Ruiz-Rodriguez
a
, J.C. Hernndez
b,
, F. Jurado
a
a
Dept. of Electrical Engineering, University of Jan, 23700 EPS Linares, Jan, Spain
b
Dept. of Electrical Engineering, University of Jan, 23071 EPS Jan, Spain
a r t i c l e i n f o
Article history:
Received 10 September 2013
Received in revised form 21 July 2014
Accepted 23 July 2014
Keywords:
Secondary area networks
Voltage unbalance
Single-phase photovoltaic systems
Stochastic assessment
Probabilistic load ow
a b s t r a c t
This research describes a way to analyze voltage unbalance sensitivity for different maximum sizes of a
single-phase photovoltaic system (SPPVS) with multiple PV penetration levels in a typical secondary
radial distribution network (SRDN) in Spain. This analysis effectively assesses current requirements as
specied in regulations concerning maximum size to be connected. It thus helps distribution network
operators to dene optimal limits, depending on their context. A stochastic assessment method is pro-
posed to account for any random combination of SPPVSs in an SRDN. In addition, this method evaluates
weekly voltage unbalance during a one-year time period, on the basis of 10-min intervals. More specif-
ically, the voltage unbalance in SRDNs with SPPVSs is assessed for each 10-min interval by means of a
probabilistic radial three-phase load ow (RTPLF). The results obtained show the maximum sizes of
the SPPVS to be connected as a function of the PV penetration level in the SRDN, where high PV penetra-
tions can produce voltage unbalance problems.
2014 Elsevier Ltd. All rights reserved.
Introduction
Nowadays, the scientic community is involved into an intense
investigation to nd new photovoltaic devices (third generation PV
technology) which provide low-cost solar electricity. These devices
are composed of materials characterized by low preparation cost,
minimum environmental impact, light weight and wide availabil-
ity [16]. The performances of many of these devices are lower
than those of silicon cells, however the trade-off between energy
generated and invested capital can be protable.
Several countries have reached the PVgrid parity as the levelised
cost of electricity for the PV technology in these countries can be
compared with their local retail electricity prices in a competitive
way. In most of the cases, this parity has been reached without
any current subsidy or feed-in tariff incentive, but it has also been
necessary a favourable regulatory framework, through net-meter-
ing or self-consumption laws [7,8]. The grid parity paradigm is
possible in the rest of the world, but future research works are
required in the smart grid research to mature this outcome [9,10].
The integration of high PV penetration levels in low voltage
radial distribution networks can cause inadmissible voltage unbal-
ances [1114]. In order to minimize this problem as well as other
adverse impacts, there are various technical regulations for the
interconnection of PV systems [15]. These connection require-
ments are based on a deterministic analysis. However, this type
of analysis has the disadvantage of not being able to objectively
specify the location where the voltage unbalance in a secondary
radial distribution network (SRDN) could surpass the standard
limit during a given time interval. Furthermore, it is also incapable
of determining the frequency of such an event. The reason for this
lies in the fact that the variables in an SRDN with PV systems are
subject to uncertainties stemming from the inherent randomness
of PV power outputs and loads. Therefore, probabilistic techniques
are the tools that can best assess the impact of uncertainty on
SRDN variables. Among all these techniques, the most frequently
used are probabilistic load ow based on Monte Carlo simulation,
analytical methods, and approximation methods.
Although analytical methods are computationally more effec-
tive than Monte Carlo simulations, they must be based on certain
mathematical assumptions. Firstly, it is necessary to dene linear
models to handle the nonlinearities of the balanced power system,
either for meshed [16] or radial [17,18] congurations. Then, con-
volution techniques must be implemented to obtain the uncer-
tainty of random outputs [1619]. Finally, GramCharlier [17,19]
or CornishFisher [16,18] expansions estimate the probability
functions of random outputs.
An example of an approximation method is the point-estimate
method (PEM), which directly provides the rst statistical moments
http://dx.doi.org/10.1016/j.ijepes.2014.07.071
0142-0615/ 2014 Elsevier Ltd. All rights reserved.

Corresponding author. Tel.: +34 953 212463; fax: +34 953 212478.
E-mail addresses: fjruiz@ujaen.es (F.J. Ruiz-Rodriguez), jcasa@ujaen.es
(J.C. Hernndez), fjurado@ujaen.es (F. Jurado).
Electrical Power and Energy Systems 64 (2015) 646654
Contents lists available at ScienceDirect
Electrical Power and Energy Systems
j our nal homepage: www. el sevi er . com/ l ocat e/ i j epes
of random outputs based on only a few deterministic load ows.
GramCharlier or CornishFisher expansions are then used to char-
acterize random outputs. The rst PEM in [20] was subsequently
followed by newand improved versions. In fact, this research study
used one of the schemes of Hongs PEM [21]. The number of deter-
ministic simulations performed (e.g. K
*
m or K
*
m + 1) grows linearly
with the number of input random variables (m) and the scheme
selected (K = 2, 3, 4). According to [22,23], the 2m + 1 scheme pro-
vides the best performance for a high number of random inputs.
Although reference [22] focuses on deterministic balanced load
ows and reference [23] focuses on deterministic three-phase load
ows, both studies only consider meshed congurations [24].
In the context of four-wire SRDNs, phase domain analysis is nec-
essary, not only for incorporating the load unbalance and line
impedance asymmetries, but also to assess the impact of single-
phase photovoltaic systems (SPPVSs). The contribution of this paper
is the use of Hongs PEM (2m + 1 scheme) [22,23] to solve the prob-
abilistic three-phase power ow problem in radial congurations.
The use of a specic radial load ow(e.g. [25]) is mandatory in radial
congurations since the conventional load ow [24] gives conver-
gence problems. This PEMfor radial congurations is thencombined
with the CornishFisher expansion. The advantages of this tool for
the assessment of the impact of SPPVSs on SRDNs are the following:
This PEM uses deterministic routines, but has a lower simula-
tion cost than that of the Monte Carlo simulation.
The simulation cost of this PEM is only slightly higher than the
cost of the analytical method based on convolution techniques.
However, when the analytical method is applied to unbalanced
power systems and radial congurations, its computational cost
is much higher than that of the PEM.
The CornishFisher expansion used in our study performs better
for non-Gaussian PV random variables [18].
Aim of the study
Voltage unbalance [26] is a growing power quality concern in
SRDNs with SPPVSs because of their variable size and location
[11,12,2731]. Such PV systems are currently being connected on
the basis of the t and inform principle, which mainly depends
on the perspectives and nancial conditions of homeowners.
Within this context, even if the voltage unbalance in an SRDN
without PV is within standard limits, there is no guarantee that it
will remain so. Therefore, the number of admissible SPPVSs or their
maximum size in SRDNs must be analyzed in such a way as to keep
the voltage unbalance within standard limits.
Currently, the connection of SPPVSs in SRDNs is subject to
requirements regarding maximum size (power). The objective of
such regulations is to limit voltage unbalance. In most national reg-
ulations, this limit is approximately 5 kVA: 3 kW in the Endesa
Utility Company [32]; 3.4 kVA in the UK [33]; 4.6 kVA in Austria
and Germany [33]; 5 kW in Spain [34]; and 6.6 kW in Italy [35].
However, in Norway and France [33], the limit is much higher
(15 kVA and 18 kVA, respectively).
This difference in standards indicates that the impact of vari-
ables such as the type of SRDN and PV penetration level is in urgent
need of clarication. This knowledge will help distribution net-
work operators to dene optimal limits for specic scenarios. In
our study, this information was obtained by means of a voltage
unbalance sensitivity analysis in an SRDN of the maximum size
of SPPVS and of the PV penetration level. PV penetration is dened
as follows:
PVpenetration
Annual PVcapacity factor Installed PVpower
MaximumSRDNpower
1
where the annual PV capacity factor is the ratio of annually pro-
duced energy to the energy that could have been produced if the
PV had operated continuously at full power.
A stochastic assessment method was used to account for any
random combination of SPPVSs in an SRDN. In addition, this
method can be used to evaluate the voltage unbalance in a one-
year time period, for 10-min intervals, according to regulations
[36,37]. More specically, the voltage unbalance in SRDNs with
SPPVSs is assessed by a probabilistic radial three-phase load ow
(RTPLF).
Voltage unbalance
Voltage unbalance can be characterized by different variables
[38]. Generally addressed is the ratio of the fundamental nega-
tive-sequence component U
!
2
to the fundamental positive-
sequence component U
!
1
. Therefore, the voltage unbalance factor
(VUF) is dened as n j U
!
2
=U
!
1
j:
Voltage unbalance is time-variant in power systems. In this sce-
nario, the use of indices is the most useful way of reducing the
voltage unbalance to a single number [38]. Although the PV sys-
tems can inuence the unbalance levels of all the nodes of a radial
distribution network [11,12], the use of a site index is preferred
[39,40]. This site index is obtained from a certain percentile (e.g.
95th, 99th) of the statistical characterization (cumulative distribu-
tion function or CDF) of the measurements of voltage unbalance
over a long observation period (e.g. a one-day or one-week period)
with a given time average (e.g. 3-s or 10-min interval). The most
commonly used voltage unbalance index is the 95th weekly per-
centile of the variable n
,10-min
(VUF based on a 10-min mean), i.e.,
n
,10-min,95w
[38].
As voltage unbalance can have various adverse effects [41], the
allowable compatibility level in low voltage supply systems is usu-
ally limited to 2% [36]. In the same context, different regulations,
utility guidelines, and international standards suggest allowable
planning/compatibility levels in the 15% interval: 1.3% in the UK
[33]; 2% in France, Germany [33], and the EU [37]; and 22.5% in
IEEE standard [42]. In this paper, a 2% limit is assumed for the site
index 95th weekly percentile of the variable 10-min VUF at any kth
SRDN node (n
k,10-min
), i.e., n
k;10- min;95w
.
Probabilistic PV and load model
Probabilistic PV system model
The probabilistic PV system model in [18] is specied for a
10-min interval. It provides information regarding the marginal
distribution (probability density function [PDF] and cumulative
distribution function [CDF]) of the PV random power for each ith
10-min interval, mth month, and jth SRDN node u
j;10-min
i
m. Thus,
the random variables hourly diffuse fractions and daily clearness
index are used to build u
j;10-min
i
m. Furthermore, this model
accounts for the stochastic interdependence of the PV distributions
corresponding to close locations (nodes) due to similar meteorolog-
ical conditions [43]. This dependence is modelled separately from
marginal distributions with a specic rank correlation matrix [18].
The model generates dependent PV power outputs based on multi-
variate dependent random numbers [18].
Probabilistic load model
Currently, certain distribution network operators are involved
in the massive deployment of smart meters in the SRDNs to mea-
sure electrical load. This makes it possible to statistically character-
ize the load at each node by using measurements obtained and
F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654 647
recorded over various years. Accordingly, the stochastic load model
used in this research is directly based on smart meter measure-
ments. The load model, based on a 10-min mean, obtains a typical
daily prole for each dth day of the week and kth SRDN node for
each month. Indeed, measurements for a given month over a per-
iod of several years were used to dene typical daily proles. This
approach takes into account the real (reactive) power consumed by
the kth SRDN node p
d
k;10-min
i
mq
d
k;10-min
i
m as a Gaussian random
variable, which changes each mth month, dth day of the week, and
ith 10-min interval.
Probabilistic radial three-phase load ow
Deterministic radial three-phase load ow
The conventional method [24] applied to the resolution of
three-phase load ows is not valid for SRDNs because of its poor
convergence. Therefore, a specic method is needed for radial con-
gurations. The method in this research [25] consists of an itera-
tive process, which arrives at the solution by covering all the
SRDN nodes by rst moving forward and backward, and then back-
ward and forward. The iterative process ends when convergence is
reached. For example, in a 3-node SRDN (Fig. 1) the steps taken to
solve the deterministic RTPLF are the following:
Step 1: Forward sweep
The unbalance current at the extreme load node of the SRDN,
for example, node 3, I
abc

load
3
is determined by:
I
abc

3
I
abc

load
3
S
abc

3
=V
abc

2
in which:
I
load
a3
S
a3
=V
a3

; I
load
b3
S
b3
=V
b3

; I
load
c3
S
c3
=V
c3

3
where S
3
(V
3
) is the complex power (voltage) of load node 3, and a,
b, and c are the phases of the system.
For the rst iteration, the extreme load node voltage equals that
of the source node (i.e., node 1). From the current at node 3, the
voltage and current at node 2 in matrix form can be determined
as follows:
V
abc

2
a V
abc

3
b I
abc

3
I
abc

line
2
c V
abc

3
d I
abc

3
4
where the coefcient values of matrices [a], [b], [c] and [d] are sta-
ted in [25].
From the voltage at node 2, the current produced by the load
conditions at this node I
abc

load
2
is obtained by Eq. (3). Thus, the total
current per phase at node 2 is the following:
I
abc

2
I
abc

load
2
I
abc

line
2
5
Voltages and currents per phase at source node [V
abc
]
1
,
I
abc

1
I
abc

line
1
are obtained by Eq. (4).
Step 2: Verication of the convergence criterion
When the convergence criterion is veried, the following result
is obtained:
Mismatches V
abc

Source
V
abc

V
base
= 6
where V
base
is the voltage base in the SRDN. When the highest value
in the matrix [Mismatches] is less than 0.001 p.u., the iterative pro-
cess is nished.
Step 3: Backward sweep
The reference voltage [V
abc
]
Source
is considered at node 1. The
voltages at the successive nodes are calculated with the currents
in step 1. Voltages of nodes 2 and 3 are given by:
V
abc

2
A
1
V
abc

Source
B
1
I
abc

2
V
abc

3
A
2
V
abc

2
B
2
I
abc

3
7
where matrices [A
1
], [A
2
], [B
1
], [B
2
] are calculated according to [25].
With the preceding voltages, the iterative process begins again at
the rst step.
Probabilistic radial three-phase load ow
The formulation of a deterministic RTPLF is based on parame-
ters with xed values. However, certain input parameters of a
SRDN with PV systems are subject to uncertainties (e.g. loads, PV
power outputs). One good way to characterize the sources of
uncertainty of this type of SRDN is to represent the input data as
random variables. In this respect, the Monte Carlo simulation and
PEM are simulation techniques that make it possible to continue
using the deterministic load ow routine.
Monte Carlo simulation
The Monte Carlo simulation generates random values of ran-
dom inputs from their distribution functions. With these values,
the steady state variables are evaluated by solving a deterministic
RTPLF. Random outputs are then reconstructed from these deter-
ministic data.
Hongs PEM: 2m + 1 scheme
The PEM uses only a few deterministic RTPLFs to solve the prob-
abilistic RTPLF. Thus, the PEM concentrates the statistical informa-
tion provided by the rst few moments of a problem input random
variable x
r
at K deterministic locations ^x
r;K
(a specied value of the
random input). By using these locations and the function h, which
relates input and output variables, the random outputs can be
obtained. Obviously, a weighting factor x
r,K
, which accounts for
the relative importance of each evaluation in the random outputs,
must be included. Pair ^x
r;K
; x
r;K
is known as the Kth concentration.
There are different schemes in Hongs PEM, which are charac-
terized by different K values in the deterministic evaluation of
function h (K = 2m, 2m + 1, 3m + 1, etc., where m is the random
input number). This study uses the 2m + 1 scheme since it provides
the best performance for a high number of random inputs [22,23].
The procedure used to compute the raw statistical moments of
random outputs for the RTPLF can be summarized as follows:
The rotational transformation method is used to transform a set
of correlated random inputs (u
1
, . . ., u
m
) into a set of uncorre-
lated randominputs (x
1
, . . ., x
m
) [44]. This is a necessary rst step
in the 2m + 1 scheme in the presence of the correlation between
randominputs. The method is based on the eigenvalues and cor-
responding eigenvectors of the covariance matrix.
The Kth concentration (^x
r;K
; x
r;K
) of each random input x
r
is
obtained from the statistical input data (e.g. PDF of x
r
, i.e., f
xr
).
The two deterministic locations are determined by:
1 2
3
11 22
33
[V
abc
]
3
[V
abc
]
3
[V
abc
]
2
[V
abc
]
2
[V
abc
]
1
[V
abc
]
1
[I
abc
]
1
line
[I
abc
]
1
line
[I
abc
]
1
line
[I
abc
]
2
line
[I
abc
]
2
line
[I
abc
]
2
line
[[[[ , [S
abc
]
2
I
abc
]
2
load
, [S
abc
]
2
I
abc
]
2
load
, [S
abc
]
2
I
abc
]
2
load
I
abc
]
2
load
, [S
abc
]
3
[I
abc
]
3
load
, [S
abc
]
3
[I
abc
]
3
load
, [S
abc
]
3
[I
abc
]
3
load
[I
abc
]
3
load
Source Source
Fig. 1. Example of SRDN with 3 nodes.
648 F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654
^x
r;K
l
xr
d
r;K
r
xr
K 1; 2 8
where d
r,K
is the standard location; and l
xr
and r
xr
are the mean
and standard deviation of the random input x
r
. The two standard
locations and weights are the following [21]:
d
r;K
k
r;3
=2 1
3K

k
r;4
3=4 k
2
r;3
q
K 1; 2 9
x
r;K
1
3K
=d
r;K
d
r;1
d
r;2
K 1; 2 10
where k
r;3
and k
r;4
are the skewness and kurtosis of the random
input x
r
.
The function h must be evaluated only two times for each ran-
dom input x
r
at the points specied for the Kth location ^x
r;K
of
the random input x
r
and the mean of the m 1 remaining
inputs, namely (l
x1
, l
x2
, . . ., ^x
r;K
, . . ., l
xm
).
One additional evaluation of function h at the point specied for
the m random input means (l
x1
, l
x2
, . . ., l
xr
, . . ., l
xm
) is required
with the following specic weighting factor x
o
:
x
o
1
X
m
r1
1 k
r;4
k
2
r;3
.
11
The solution for each of the preceding deterministic RTPLFs is:
zr; K h l
x
1
; l
x
2
; . . . ; ^x
r;K
; . . . ; l
xm

12
where z(r, K) is a vector of output variables associated with the
Kth concentration of random input x
r
.
The vector z(r, K) is used to estimate the statistical moments
around zero of the output random variables z (z
1
, . . ., z
ov
, where
ov is the random output number):
Ez
j

X
ov
r1
X
2
K1
x
r;K
zr; K
j
x
o
h. . . ; l
xr
; . . .

j
13
Stochastic assessment method to evaluate the weekly voltage
unbalance in an SRDN
Voltage unbalance assessment due to the uncertainty regarding the PV
location
This research proposes a 3000-trial stochastic assessment of the
voltage unbalance in an SRDN for any random combination of loca-
tions for SPPVSs of a given size and PV penetration level (left side of
Fig. 2, dashed-line shapes). The number of trials, I = 3000, has the
advantage of providing an acceptable coverage of different sto-
chastic PV locations in the SRDN (see section Voltage unbalance
assessment due to the uncertainty in the PV location).
The two random inputs of this stochastic assessment determine
the random location of SPPVSs in the SRDN, in other words, the
node/s and phase where each SPPVS is connected. The uncertainty
of PV locations is modelled by drawing random numbers distrib-
uted uniformly in the interval (1, n
B
), which represent the nodes
with PV. Another random number, distributed uniformly in the
interval (1), (3), represents the phase in which the SPPVS is con-
nected at each node. Previously, a PV penetration level (0%, 5%,
10%, and 15%) and an SPPVS size (5, 10, 15 kW) are specied in
order to perform a sensitivity analysis of these variables. Obvi-
ously, the number of SPPVSs to be connected in the SRDN is deter-
mined by both variables.
For each Ith PV location (I = 3000 trials), the random variable
weekly VUF is calculated based on the 10-min mean in the mth
month for each kth SRDN node n
week;I
k;10- min
m. Then, a nite mixture
distribution [45] is applied to all of the PV locations because each
location has the same probability of occurrence. Thus, given a nite
set I of PDFs for each kth random variable n
week;I
k;10- min
m, (f
n
week;1
k;10- min
m
;
. . . ; f
n
week;I
k;10- min
m
), or the corresponding CDFs (F
n
week;1
k;10- min
m
; . . . ;
F
n
week;I
k;10- min
m
), and weights -
1
, . . ., -
I
, such that -
i
P0 and
P
-
i
= 1,
the nite mixture distribution for all the PV locations of every kth
random variable n
week
k;10- min
m can be represented by writing either
the PDF, f
n
week
k;10- min
m
, or the CDF, F
n
week
k;10- min
m
, as a sum:
f
n
week
k;10- min
m

X
I
p1
-
p
f
n
week;p
k;10- min
m
k 1. . . n
B
14
F
n
week
k;10- min
m

X
I
p1
-
p
F
n
week;p
k;10- min
m
k 1. . . n
B
15
As all -
i
are equal, -
i
is 1/n
B
.
Once the mixture distribution for all of the PV locations is built
(f
n
week
1;10- min
m
; . . . ; f
n
week
n
B
;10- min
m
), (F
n
week
1;10- min
m
; . . . ; F
n
week
n
B
;10- min
m
), the mixture
distribution is selected at month m
*
with the worst weekly VUF,
i.e., (f
n
week
1;10- min
m

; . . . ; f
n
week
n
B
;10- min
m

), (F
n
week
1;10- min
m

; . . . ; F
n
week
n
B
;10- min
m

).
Voltage unbalance assessment due to time varying load and PV proles
The node loads in the SRDN have a daily, weekly, and monthly
variation. PV power output also has a daily and monthly variation.
Hence, the voltage unbalance assessment of both series at each Ith
PV location should extend throughout a one-year time period
(right side of Fig. 2, continuous-line shapes). It is thus possible to
ascertain the worst weekly VUF for the twelve months of the year,
according to regulations [37].
Given an Ith PV location, the probabilistic PV system model in
section Probabilistic PV system model determines correlated PV
random power for the ith 10-min interval, mth month, and jth
SRDN node u
I
j;10-min
i
m. The set of correlated PV random power
outputs is transformed into a set of uncorrelated PV random out-
puts pv
I
j;10-min
i
m. Taking PV power and load concentrations as
inputs, the probabilistic RTPLF (see section Probabilistic radial
three-phase load ow) provides the raw statistical moments of
the VUF for each SRDN node at an Ith PV location, ith 10-min inter-
val, dth day of the week, and mth month (e.g. at a kth SRDN node
n
d;I
k;10-min
i
m. Then, the CornishFisher expansion reconstructs
solutions for random variable VUFs in terms of PDFs and CDFs,
(e.g. PDFs -f
n
d;I
1;10-min
i
m
; . . . ; f
n
d;I
n
B
;10-min
i
m
-). This process is repeated
12,096 times, which corresponds to an annual evaluation, based
on 10-min intervals in which seven days of each month are consid-
ered. When all simulations are run, a nite mixture distribution is
applied to the seven days of each mth month, i.e., only one week
per month (e.g. PDFs -f
n
week;I
1;10- min
m
; . . . ; f
n
week;I
n
B
;10- min
m
-).
Case studies: rural SRDN
A representative SRDN for a rural and urban environment
should reect all potential load proles and design characteristics.
For this reason, the characteristics of the SRDNs selected in this
study are very close to the average of the wide range of SRDNs in
Andalusia (Spain).
Initial simulations showed that voltage proles and voltage
unbalance were not a source of problems in the urban SRDN. For
example, its worst voltage drop (voltage unbalance) based on
the10-min mean, (i.e., the worst weekly 95th percentile through-
out the year) was 0.69% (0.11%) as compared to 6.58% (1.64%),
which was the value in the rural SRDN. Therefore, the voltage
F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654 649
unbalance sensitivity analysis, implemented in MATLAB, was only
performed on the rural SRDN.
Test SRDN data and PV input data
The rural SRDN is a 30-branch, 31-node, 0.4-kV SRDN located in
a rural area of Andalusia. It is a four-wire SRDN which has a multi-
grounded neutral system. Its total area is 0.0478 km
2
, and it sup-
plies a 114.36-kVA annual mean load distributed in an unbalanced
pattern. The most distant nodes (131) are 478 m away from each
other. Maximum SRDN power is 211 kVA. The single-phase dia-
gram is shown in Fig. 3 and the SRDN data are given in Appendix
A. It is important to highlight that this rural SRDN has unbalanced
lines and line sections carrying a mixture of single-phase, double-
phase, and three-phase loads. Data for real and reactive load pro-
les were collected at each node in 10-min intervals by using
smart meters over a two-years time period.
Global irradiation data for the hourly diffuse fraction and the
daily clearness index was obtained from [43]. The spatial PV
dependence structure was modelled with a rank correlation matrix
based on the distance between nodes [43]. Annual PV capacity fac-
tor is approximately 0.15 at the SRDN site.
Base case
To study the effect of different loading conditions on voltage
unbalance, two scenarios were considered. Load scenario 1 repre-
sented the lightly loaded SRDN. This was a 101.15-kVA load in
the 10-min interval at 4:20 a.m. on a Sunday morning in April.
Load scenario 2 represented the heavily loaded SRDN. This was a
149.47-kVA load at 9:10 p.m. on a Monday night in July.
Fig. 4 shows the VUF variation (mean, 95th percentile, and stan-
dard deviation) for every kth SRDN node in the load scenario 1 and
load scenario 2. In general, the VUF increased with the distance
from the source node. It was assumed that the source node was
a balanced three-phase source. As we moved along the SRDN,
unbalanced load currents caused unequal voltage drops, which
produced unbalanced phase voltages. The voltage unbalance mag-
nitude for the test SRDN with a low X/R ratio was mostly due to
unequal voltage magnitude.
As can be observed, during the light load condition, the VUF at
node 31 was intermediate and reached a 1.04% (1.09%) value for
the mean (95th percentile). There was no appreciable voltage drop
along the SRDN, and as a result the phase voltages remained nearly
balanced. However, during the heavy load condition, the VUF
reached a higher mean (95th percentile), i.e., 1.59% (1.69%).
Accuracy of the PEM to assess the voltage unbalance in the SRDN for a
specic 10-min interval and PV location
To verify the accuracy of Hongs PEM (2m + 1 scheme), this
method and the Monte Carlo simulation were used to obtain the
results of VUFs in the test SRDN at a specic I
*
th PV location and
for a specic i
*
th 10-min interval. In particular, the 10-min interval
at 12:00 a.m. on a Monday in July was analyzed. A 10% PV penetra-
tion level with 10-kWp SPPVSs was considered. This amounted to
Fig. 2. Flowchart of the stochastic assessment method for weekly voltage unbalance in a SRDN.
650 F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654
15 SPPVSs, which were located at the following nodes and phases:
2a, 3c, 4b, 7a, 7c, 10c, 12a, 16b, 18c, 20b, 20a, 25b, 28c, 30a, and
30c. A Monte Carlo simulation with 10,000 trials was used as a
reference.
Table 1 shows the individual maximum relative errors of the
rst seven moments about zero for the VUFs in the SRDN. The indi-
vidual maximum relative error e for the r-order moment about
zero (l
r
) of the VUF (n) in the PEM for the i
*
th 10-min interval,
d
*
th day, m
*
th month, and I
*
th PV location, i.e., e
l
r
n
d

;I

max;10min
i

is
given as follows [18]:
e
l
r
n
d

;I

k;10min
i

100 l
r;PEM
n
d

;I

k;10min
i

l
r;MCS
n
d

;I

k;10min
i

l
r;MCS
n
d

;I

k;10min
i

16
e
l
r
n
d

;I

max;10min
i

max
k
e
l
r
n
d

;I

k;10min
i

8
<
:
9
=
;
k 1. . . n
B
17
The values in Table 1 shows the high accuracy level of the PEM for
all moments. The individual relative error always achieved its max-
imum value for each moment in an SRDN node with an SPPVS.
Fig. 3. Single-line diagram of the rural SRDN and relevant phase impedance matrixes.
0.01
0.03
0.05
0.07
0.0
0.5
1.0
1.5
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
95th baja 95th alta
Mean Standard desviation 95
th
percentil
V
U
F

-
m
e
a
n
,

9
5
t
h

p
e
r
c
e
n
t
i
l
e
-


(
%
)
SRDN node number
Load scenario 1
VUF
Load scenario 2
V
U
F

-
s
t
a
n
d
a
r
d

d
e
s
v
i
a
t
i
o
n
-

(
%
)
Fig. 4. Variation of VUF (mean, 95th percentile, and standard deviation) along the SRDN for two load scenarios without PV.
Table 1
Individual maximum relative errors of the rst seven moments about zero for the VUFs in the SRDN.
r-Order moment about zero 1st 2nd 3rd 4th 5th 6th 7th
Individual maximum relative error (%) 0.265 0.510 0.725 0.864 0.854 0.624 0.667
F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654 651
Voltage unbalance assessment due to time varying load and PV proles
for a specic PV location
Firstly, this section analyzes the voltage unbalance variation in
the test SRDN during a full day (Monday) in July for a specic PV
location. The weekly evaluation for the same month is then
presented.
Let us assume three scenarios for a specic PV location, given a
10% PV penetration level and three potential SPPVS sizes: (i) 5-kW
SPPVS (PV scenario 1 composed of 30 SPPVSs); (ii) 10-kW SPPVS
(PV scenario 2 composed of 15 SPPVSs); (iii) 15-kW SPPVS (PV sce-
nario 3 composed of 10 SPPVSs). Another PV scenario without PV is
also included.
Fig. 5 shows the daily VUF variation (mean, 95th percentile)
based on a 10-min mean for node 31, which has the highest VUF
in the four PV scenarios. The unbalanced location of the SPPVSs
in the phases further increased the already existing unbalance in
the SRDN. For example at midday, the mean (95th percentile) of
the random variable VUF increased from 1.18% (1.25%) without
PV to 1.58% (1.85%) in the worst PV scenario. This was PV scenario
3 with the largest SPPVS size (15 kW). Thus, many small SPPVSs
distributed throughout the SRDN helped to maintain its balanced
condition. However, as the SPPVSs became larger in size, their
number also decreased, which heightened the probability of unbal-
ance conditions.
Fig. 6 shows the CDF of weekly VUF based on a 10-min mean for
node 31 in July for the four PV scenarios. PV scenario 1 had a worse
weekly 95th percentile than PV scenario 2 in spite of its better per-
formance at midday on Monday (Fig. 5). This occurred because of
the inuence of the other days of the week. However, the higher
probability of a lower weekly VUF is evident in PV scenario 1. Addi-
tionally, the distribution functions moved away from the normal
distribution (see skewness and kurtosis) as a result of non-Gaussian
distributions for each 10-min PV power output throughout the
week. The SPPVSs led to a shift in the distribution towards higher
voltage unbalances only in PV scenery 3. The dispersion of resulting
distributions became greater as the number of SPPVSs increased. In
any case, the voltage unbalance at node 31, evaluated by the weekly
95th percentile of the VUF, in PV scenario 1 (1.59%), PV scenario 2
(1.50%), and PV scenario 3 (1.72%), was lower than the 2% voltage
unbalance limit.
Voltage unbalance assessment due to the uncertainty in the PV
location
The voltage unbalance assessment in the test SRDN, given a cer-
tain SPPVS size and PV penetration level, requires a high number of
different stochastic PV locations, which represent the widely
diverging values that voltage unbalance can reach at each node.
It is a known fact that connecting an SPPVS at the beginning or
end of an SRDN often leads to a signicant difference in voltage
unbalance. Since this number of PV locations may be very high, it
is important to determine the lowest number of different PV loca-
tions that originate different values of voltage unbalance.
Based on the assumption of a 10% PV penetration level and a
10-kW SPPVS size, Fig. 7 shows the weekly VUF (mean, 95th per-
centile) for node 31, i.e., n
week
31;10- min;l
7, as a function of the number
of different random PV locations. Even though the number of
potential combinations is very high, only 1000 random PV loca-
tions generated widely diverging values for voltage unbalances at
this node. In fact, mean or higher-order moments varied very little
after 1000 trials.
0.5
1.0
1.5
2.0
"mean PV sin pv" percentil sin PV
"mean PV scenario 1" 95th pv scenario 1
"mean PV scenario 2" 95th pv scenario 2
"mean PV scenario 3" 95th pv scenario 3
V
U
F

-
m
e
a
n
,

9
5
t
h

p
e
r
c
e
n
t
i
l
e
-


(
%
)
Time (hours)
Mean 95
th
percentil
Without PV
PV scenario 1
VUF
PV scenario 2
PV scenario 3
6 9 12 15 18 21 0 24 3
Fig. 5. Daily variation of VUF (mean, 95th percentile) based on a 10-min mean for
node 31 of the SRDN on a Monday in July for four PV scenarios.
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
"cdf sin pv"
"CDF PV scenario 1"
"CDF PV scenario 2"
"CDF PV scenario 3"
C
D
F

Weekly VUF
Weekly VUF
Skewness Kurtosis
Without PV 0.16 2.10
PV scenario 1 -0.34 1.96
PV scenario 2 -0.13 2.49
PV scenario 3 0.51 3.01
Fig. 6. CDF of the weekly VUF based on a 10-min mean for node 31 in July for four
PV scenarios.
1.2
1.4
1.6
1.8
2.0
2.2
1 10 100 1,000
Percentile 95th
Mean
W
e
e
k
l
y

V
U
F

-
m
e
a
n
,

9
5
t
h

p
e
r
c
e
n
t
i
l
e
-


(
%
)

Number of PV locations
10,000
Mean
95
th
percentil
Fig. 7. Weekly VUF (mean, 95th percentile) for node 31 in July for a varying number
of PV locations (xed SPPVS size and PV penetration).
0.0
0.2
0.4
0.6
0.8
1.0
0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 2.1 2.3
CDF sin PV
CDF 5%-PV
penetration, 5kW
CDF 5%-PV
penetration, 10k
CDF 5%-PV
penetration, 15k
CDF 10%-PV
penetration, 5kW
CDF 10%-PV
penetration, 10k
CDF 10%-PV
penetration, 15k
CDF 15%-PV
penetration, 5kW
CDF 15%-PV
penetration, 10k
CDF 15%-PV
penetration, 15k
C
D
F

Weekly VUF
5%-PV penetration, 5-kW SPPVS
Without PV
15%-PV penetration, 5-kW SPPVS
10%-PV penetration, 15-kW SPPVS
10%-PV penetration, 10-kW SPPVS
10%-PV penetration, 5-kW SPPVS
5%-PV penetration, 15-kW SPPVS
5%-PV penetration, 10-kW SPPVS
15%-PV penetration, 15-kW SPPVS
15%-PV penetration, 10-kW SPPVS
Month of July, weekly VUF is the worst
SRDN node #31
Fig. 8. Stochastic assessment of weekly VUF based on a 10-min mean for node 31 in
the worst month (July) with different PV penetration levels and SPPVS sizes.
652 F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654
Given the different PV penetration levels and SPPVS sizes in this
paper, 3000 trials were considered.
Stochastic assessment of voltage unbalance for varying SPPVS sizes
and PV penetration levels
This section discusses the impact of SPPVS size on voltage
unbalance in the test SRDN for different PV penetrations. Fig. 8
shows the results of stochastic assessment of the weekly VUF
based on a 10-min mean for node 31 in the test SRDN with 5%,
10%, and 15% PV penetration levels as well as with SPPVS sizes of
5-kW, 10-kW, and 15-kW.
As can be observed, whenever the PV penetration level was
below 5%, the connection of large SPPVSs (up to 15 kW) did not
deteriorate the voltage unbalance in the SRDN. Thus, for the worst
node (node 31) and month (July, 7
*
), the weekly 95th percentile of
the VUF for a 5-kW SPPVS (1.67%), a 10-kW SPPVS (1.78%) and a
15-kW SPPVS (1.90%) was lower than the 2% voltage unbalance
limit. When the PV penetration level increased up to 10%, the con-
nection of large SPPVSs (15-kW) led to an inadmissible voltage
unbalance in the SRDN. Thus, index n
week
31;10- min;95w
7

was 2.08%
higher than 2%. However, the connection of smaller SPPVSs (e.g.
5-kW or 10-kW) was permitted since the indices of voltage unbal-
ance (1.73% and 1.98%) met the 2% requirement. For the 15% PV
penetration level, only the connection of small SPPVSs (5-kW)
was allowed since voltage unbalance reached an admissible level,
i.e., index n
week
31;10- min;95w
7

was 1.69%.
Conclusion
A voltage unbalance sensitivity analysis for different maximum
sizes of SPPVSs and PV penetration levels was performed on a typ-
ical SRDN in Spain. The objective was to verify current require-
ments as specied in the regulations regarding the maximum
size to be connected. For this purpose, a stochastic assessment
method was used, which was able to account for any random com-
bination of SPPVSs in the SRDN. In addition, this method evaluated
any weekly voltage unbalance during a one-year time period,
assuming 10-min intervals. More specically, the voltage unbal-
ance in SRDNs with SPPVSs for each 10-min interval was assessed
by using a probabilistic RTPLF.
The results obtained in this research study show that the con-
nection of large SPPVSs (of a size up to 15 kW) is permissible for
a 5% PV penetration level. However, current maximum size levels
in regulations (approximately 5 kW) only apply to high PV pene-
tration levels, such as a 15% PV penetration.
This tool should also be applied in sensitivity analyzes for other
SRDN variables (e.g. voltage regulation, line losses, line ow, etc.).
This would be particularly valuable for the analysis of line ow in
urban SRDNs, which is the limiting factor.
Acknowledgment
The authors would like to thank J. de la Cruz of ENDESA Distrib-
ucin Elctrica S.L.U., who provided the practical information and
data regarding SRDNs.
Appendix A
The loads in Table 2 refer to annual means for 10-min intervals.
References
[1] Lee JW, Lee TY, Yoo PJ, Graatzel M, Mhaisalkard S, Park NG. Rutile TiO
2
-based
perovskite solar cells. J Mater Chem A 2014;2:92519.
[2] Bella F, Imperiyka M, Ahmadda A. Photochemically produced quasi-linear
copolymers for stable and efcient electrolytes in dye-sensitized solar cells. J
Photoch Photobio A: Chem 2014;289:7380.
Table 2
SRDN data.
Nodes Branch parameters Node loads
Sending node Receiving node Length Conguration (see Fig. 3) Phase a Phase b Phase c
m kW kvar kW kvar kW kvar
1 2 38 3 150/95 mm
2
0 0 0 0 0 0
2 3 90 4 150 mm
2
Al 5.42 1.99 0 0 4.45 2.03
3 4 50 4 50 mm
2
Al 0 0 0 0 0 0
4 5 6 4 50 mm
2
Al 0 0 0 0 2.67 0.98
5 6 13 4 25 mm
2
Al 0 0 3.58 1.40 0 0
3 7 17 4 150 mm
2
Al 3.42 1.42 0 0 0 0
7 8 29 4 150 mm
2
Al 0 0 0 0 1.74 0.60
8 9 11 4 25 mm
2
Al 0 0 4.45 1.25 0 0
8 10 65 2 25 mm
2
Al 0 0 0 0 2.59 1.01
10 11 28 2 25 mm
2
Al 0 0 0 0 0 0
10 12 11 2 25 mm
2
Al 0 0 2.59 0.77 2.52 1.32
8 13 114 4 150 mm
2
Al 0.77 0.29 2.63 1.03 0 0
13 14 113 4 25 mm
2
Al 0 0 0 0 0 0
13 15 36 4 150 mm
2
Al 0 0 0 0 1.77 0.43
15 16 6 4 150 mm
2
Al 5.09 1.59 0 0 0 0
16 17 17 4 25 mm
2
Al 0 0 7.12 2.62 0 0
16 18 16 2 25 mm
2
Al 0 0 0 0 2.79 0.95
18 19 50 2 25 mm
2
Al 7.39 1.53 0 0 3.41 1.80
19 20 5 4 25 mm
2
Al 0 0 0 0 0 0
19 21 5 4 25 mm
2
Al 0 0 1.98 0.67 0 0
15 22 26 2 25 mm
2
Al 3.76 1.28 0 0 0 0
22 23 18 4 25 mm
2
Al 0 0 0 0 0 0
22 24 104 2 25 mm
2
Al 2.63 1.03 0 0 0 0
24 25 54 2 25 mm
2
Al 0 0 0 0 0 0
25 26 6 4 25 mm
2
Al 1.55 0.43 0 0 5.8 3.06
24 27 40 2 25 mm
2
Al 0 0 3.41 1.49 0 0
27 28 16 2 25 mm
2
Al 0 0 0 0 4.30 2.26
28 29 10 4 25 mm
2
Al 0 0 2.55 0.87 0 0
28 30 17 2 25 mm
2
Al 0 0 0 0 2.59 1.01
30 31 4 4 25 mm
2
Al 0 0 7.39 1.80 6.62 3.48
F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654 653
[3] Pianezzi F, Reinhard P, Chirila A, Bissig B, Nishiwaki S, Buecheler S, et al.
Unveiling the effects of post-deposition treatment with different alkaline
elements on the electronic properties of CIGS thin lm solar cells. Phys Chem
Chem Phys 2014;16:884351.
[4] Mola GT, Abera N. Correlation between LUMO offset of donor/acceptor
molecules to an open circuit voltage in bulk heterojunction solar cell.
Physica B 2014;445:569.
[5] Saber EM, Lee SE, Manthapuri S, Yi W, Deb C. PV (photovoltaics) performance
evaluation and simulation-based energy yield prediction for tropical buildings.
Energy 2014;71:58895.
[6] Solanki A, Wu B, Salim T, Yeow EKL, Lam YM, Sum TC. Performance
improvements in polymer nanober/fullerene solar cells with external
electric eld treatment. J Phys Chem C 2014;118:1128591.
[7] Talavera DL, Casa J, Muoz-Ceron E, Almonacid G. Grid parity and self-
consumption with photovoltaic systems under the present regulatory
framework in Spain: the case of the University of Jaen Campus. Renew Sust
Energy Rev 2014;33:75271.
[8] Spertino F, Paolo DL, Valeria C. Economic analysis of investment in the rooftop
photovoltaic systems: a long-term research in the two main markets. Renew
Sust Energy Rev 2013;28:53140.
[9] Cdric C. Smart grids: another step towards competition, energy security and
climate change objectives. Energy Policy 2011;39:5399408.
[10] Giordano V, Meletiou A, Covrig CF, Mengolini A, Ardelean M, Fulli G, et al.
Smart grid projects in Europe: lessons learned and current development (2012
update). JRC scientic and policy reports; 2013.
[11] Shahnia F, Majumder R, Ghosh A, Ledwich G, Zare F. Voltage imbalance
analysis in residential LV distribution networks with rooftop PVs. Electr Power
Syst Res 2011;81:180514.
[12] Alam MJE, Muttaqi KM, Sutanto D. A three-phase power ow approach for
integrated 3-wire MV and 4-wire multigrounded LV networks with rooftop
solar PV. IEEE Trans Power Syst 2013;28(2):172837.
[13] Shahnia F, Ghosh A, Ledwich G, Zare F. Voltage unbalance improvement in low
voltage residential feeders with rooftop PVs using custom power devices. Int J
Electr Power Energy Syst 2014;55:36277.
[14] Omar R, Rahim NA. Voltage unbalanced compensation using dynamic voltage
restorer based on supercapacitor. Int J Electr Power Energy Syst
2012;43(1):57381.
[15] Hernandez JC, De la Cruz J, Ogayar B. Electrical protection for the grid-
interconnection of photovoltaic-distributed generation. Electr Power Syst Res
2012;89:8599.
[16] Usaola J. Probabilistic load ow with correlated wind power injections. Electr
Power Syst Res 2010;80:52836.
[17] Ruiz-Rodriguez FJ, Hernandez JC, Jurado F. Probabilistic load ow for radial
distribution networks with photovoltaic generators. IET Renew Power Gener
2012;6(2):11021.
[18] Ruiz-Rodriguez FJ, Hernandez JC, Jurado F. Probabilistic load ow for
photovoltaic distributed generation using the CornishFisher expansion.
Electr Power Syst Res 2012;89:12938.
[19] Zhang P, Lee ST. Probabilistic load ow computation using the method of
combined cumulants and GramCharlier expansion. IEEE Trans Power Syst
2004;19(1):67682.
[20] Rosenblueth E. Point estimates for probability moments. Proc Nat Acad Sci
1975;72:38124.
[21] Hong HP. An efcient point estimate method for probabilistic analysis. Reliab
Eng Syst Saf 1998;59:2617.
[22] Morales JM, Prez-Ruiz J. Point estimate schemes to solve the probabilistic
power ow. IEEE Trans Power Syst 2007;22(4):1594961.
[23] Caramia P, Carpinelli G, Varilonec P. Point estimate schemes for probabilistic
three-phase load ow. Electr Power Syst Res 2010;80:16875.
[24] Arrillaga PJ, Harnold CP, Harker BJ. Computer analysis of power systems. New
York: Wiley; 1990.
[25] Kersting WH. Distribution system modeling and analysis. New Mexico: CRC
Press; 2002.
[26] Chen TH, Yang CH, Yang NC. Examination of the denitions of voltage
unbalance. Int J Elect Power Energy Syst 2013;49:3805.
[27] Briceno Vicente WC, Caire R, Hadjsaid N. Probabilistic load ow for voltage
assessment in radial systems with wind power. Int J Elect Power Energy Syst
2012;35:936.
[28] Oshiro M, Tanaka K, Senjyu T, Toma S, Yona A, Saber AY, et al. Optimal voltage
control in distribution systems using PV generators. Int J Elect Power Energy
Syst 2011;33:48592.
[29] Patsalides M, Stavrou A, Efthymiou V, Georghiou GE. Towards the
establishment of maximum PV generation limits due to power quality
constraints. Int J Elect Power Energy Syst 2012;42:28598.
[30] Parmar D. Yao L. Impact of unbalanced penetration of single phase grid
connected photovoltaic generators on distribution network. In: 46th
International Universities Power Engineering Conference (UPEC); 2011. p. 18.
[31] Yan R, Saha TK. Voltage variation sensitivity analysis for unbalanced
distribution networks due to photovoltaic power uctuations. IEEE Trans
Power Syst 2012;27(2):107889.
[32] Endesa Distribution Company. Guidelines regarding the conditions for the
connection to the ENDESA LV distribution network of PV resources of power
lower than 100 kVA, Spain; 2009.
[33] Arsenal research. DGFACTS WP1, D1: evaluation of the quality of supply
requirements specied by existing standards, national legislation and relevant
technical reports inside and outside EU; 2004.
[34] Royal Decree 1663/2000. Connection of photovoltaic installations to the LV
network, Spain; 2000.
[35] CEI Std. 0-21. Reference technical rules for the connection of active and passive
users to the LV electrical utilities; 2011.
[36] IEC Std. 61000-2-2. EMC: environment compatibility levels for low-
frequency conducted disturbances and signalling in public low-voltage
power supply systems; 2002.
[37] EN Std. 50160. Voltage characteristics of electricity supplied by public
distribution systems; 2000.
[38] Ortega MJ, Hernndez JC, Garca OG. Measurement and assessment of power
quality characteristics for photovoltaic systems: harmonics, icker, unbalance,
and slow voltage variations. Electr Power Syst Res 2012;96:2335.
[39] IEC/TR Std. 61000-3-14. EMC: assessment of emission limits for the
connection of disturbing installations to lv power systems; 2011.
[40] IEC/TR Std. 61000-3-15. EMC: limits: assessment of low frequency
electromagnetic immunity and emission requirements for DG systems in LV
network; 2011.
[41] Fan L, Miao Z, Domijan A. Impact of unbalanced grid conditions on PV systems.
In: 2010 IEEE Power and Energy Society General Meeting; 2010. p. 16.
[42] IEEE Std. 1159. Recommended practice for monitoring electric power quality;
2009.
[43] Hernndez JC, Ruiz-Rodriguez FJ, Jurado F. Technical impact of photovoltaic-
distributed generation on radial distribution systems: stochastic simulations
for a feeder in Spain. Int J Electr Power Energy Syst 2013;50:2532.
[44] Mardia KV, Kent JT, Bibby JM. Multivariante analysis. London: Academic Press;
1997.
[45] Ray S, Lindsay BG. The topography of multivariate normal mixtures. Ann Stat
2005;33(5):204265.
654 F.J. Ruiz-Rodriguez et al. / Electrical Power and Energy Systems 64 (2015) 646654

Vous aimerez peut-être aussi