Available online at www.sciencedirect.

com

Solar Energy 81 (2007) 14731481

www.elsevier.com/locate/solener

Probabilistic load ow using Monte Carlo techniques

for distribution networks with photovoltaic generators
S. Conti *, S. Raiti
Dipartimento di Ingegneria Elettrica, Elettronica e dei Sistemi, Universita` degli Studi di Catania, Viale A. Doria, 6, 95125 Catania, Italy
Received 26 September 2006; accepted 28 February 2007
Available online 22 March 2007
Communicated by: Associate Editor H. Gabler

Abstract
Connections of distributed generation (DG) systems to distribution networks are increasing in number, though they may often be
associated with the need of costly grid reinforcements or new control issues to maintain optimal operation. Appropriate analysis tools
are required to check distribution networks operating conditions in the evolving scenario. Load ow (LF) calculations are typically
needed to assess the allowed DG penetration level for a given network in order to ensure, for example, that voltage and current limits
are not exceeded.
The present paper deals with the solution of the LF problem in distribution networks with photovoltaic (PV) DG. Suitable models for
prediction of the active power produced by PV DG units and the power absorbed by the loads are to be used to represent the uncertainty
of solar energy availability and loads variation. The proposed models have been incorporated in a radial distribution probabilistic load
ow (PLF) program that has been developed by using Monte Carlo techniques. The developed program allows probabilistic predictions
of power ows at the various sections of distribution feeders and voltage proles at all nodes of a network.
After presenting theoretical concepts and software implementation, a practical case is also discussed to show the application of the
study in order to assess the maximum PV peak power that can be installed into a distribution network without violating voltage and
current constraints. A comparison between Deterministic Load Flow (DLF) and PLF analyses is also performed.
 2007 Elsevier Ltd. All rights reserved.
Keywords: Electrical distribution networks; Photovoltaic generators; Probabilistic load ow

1. Introduction
Distributed generation (DG) can be dened as electric
power generation within distribution networks or on the
customer side of the network. Network integration of
DG is a very complex issue that is signicantly dierent
from traditional network integration of power generation
into transmission networks, since actual distribution networks are designed as radially operated, passive systems.
For this reason connection of DG to the distribution system may be associated with costly grid reinforcements or

Corresponding author. Tel.: +39 095 7382604; fax: +39 095 330793.
E-mail address: stefania.conti@diees.unict.it (S. Conti).

0038-092X/$ - see front matter  2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.solener.2007.02.007

new control issues to maintain optimal operating conditions (e.g. acceptable voltage quality, correct protections
operation, etc.). However, the general view is that DG is
expected to play an important role in future electrical
energy systems. Two major reasons for an increased utilization of DG are liberalized markets and the global trend of
reducing greenhouse gas emissions, which leads to more
renewable energy based power sources.
With regard to LV distribution networks, an example of
grid-connected renewable energy generators has been provided in many European Countries by the implementation
of the photovoltaic (PV) roof-tops programs, supported
by national governments.
At the LV level, one of the most relevant issues is the
quality of the voltage supplied to customers, according to

1474

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

Nomenclature
network incidence matrix, dimension is (n n)
vector of the node currents, dimension (n 1)
vector of the branch currents, dimension (n 1)
network complex impedance matrix, dimension
(n n)
Z b 
diagonal matrix whose elements are the complex
impedances of the corresponding branches,
dimension (n n)
Si
complex power at node i
Vi
voltage phasor at node i
Ii
current phasor at node i
I i
complex conjugate of I i
Pi
net real power at node i
Qi
net reactive power at node i
PLi and QLi real and reactive power absorbed by the
load at node i
PGi and QGi real and reactive power delivered by the
generators at node i
Rij and Xij real and imaginary part of Z ij , which is the
generic element of matrix Z
r and x line resistance and reactance per kilometre
Vmin and Vmax minimum and maximum voltage values
allowed by EN 50160
Imax
line current carrying capacity
PL
random variable load demand
PL
mean value of PL
r
standard deviation of PL
kt
random variable hourly clearness index
ktu
upper bound for kt
C, k
parameters of the probability density function
for kt
It
random variable irradiance on an horizontal
plane
I0
random variable extraterrestrial total solar
irradiance
[A]
I
J 
Z

the European Standard EN 50160 (European Standard EN

50160, 1999), especially in terms of possible overvoltage
caused by connection of generators (Conti et al., 2003a,b).
Appropriate load-ow (LF) calculations are needed to
assess the allowed DG penetration level for a given network in order to ensure that the maximum voltage at the
point of common coupling (PCC) and lines current carrying capacity are not exceeded (Conti et al., 2003c).
However, when connection of generators based on
renewable energy (such as solar energy) is considered, it
is not possible to achieve a realistic evaluation of where
and when overvoltages may occur by simply using traditional deterministic LF (DLF) analysis. In fact, this analysis (recalled in Section 2) is normally based on some
selected combinations of consumer loads and PV power
productions. Consequently, it does not take into account
the statistical variation of loads and solar radiation.

random variable irradiance on a surface with

inclination b to the horizontal plane
Ppv
random variable power output of the PV system
AC
PV array surface area
g
eciency of the PV system in realistic reporting
conditions (RRC)
d
declination angle
q
reectance of the ground
/
latitude
x
hour angle
xs
sunset hour angle
PPVpeak total photovoltaic peak power installed in the
network
PPVpeak(max)DLF maximum value of PPVpeak that can be
installed in the network without violating voltage and current constraints calculated by means
of a DLF
PPVpeak(max)PLF maximum value of PPVpeak that can be
installed in the network without violating voltage and current constraints calculated by means
of a PLF
PLTOT network overall contractual load
lOV
yearly average number of hours with overvoltage
rOV
yearly average standard deviation of the number
of hours with overvoltage
lUV
yearly average number of hours with undervoltage
rUV
yearly average standard deviation of the number
of hours with undervoltage
lOC
yearly average number of hours with overcurrent
rOC
yearly average standard deviation of the number
of hours with overcurrent

Ib

Then, to solve this problem, the paper presents a probabilistic load ow (PLF), based on Monte Carlo techniques, for radial distribution networks with PV DG
(Section 3). The procedure incorporates suitable models
for the active power produced by PV DG units and the
power absorbed by the loads in order to represent the
uncertainty for solar energy availability and load variation.
In Section 4 the development of a software tool to
implement the method proposed will be presented in order
to assess voltages and currents in a distribution network,
calculated hour by hour throughout a year. Then, the
application to a practical case study will be discussed to
show the results of the proposed method. In particular,
the results obtained by the DLF (Section 6) and the PLF
(Section 7) applied to a realistic LV distribution network
with PV DG (described in Section 5) will be compared.
The two LF analyses will be used to assess the maximum

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

PV peak power that can be installed into the network without violating voltage and current constraints.
It will be shown that the use of a traditional DLF leads
to an overestimation of the maximum PV peak power that
can be installed according to the evaluation performed by
the PLF.

Let us consider a three-phase symmetrical, radial distribution network with n nodes and n branches, where we
dene as nodes the points of load connections, the points
of change in the line characteristics and the junctions, and
as branches the conductor segments between two nodes.
The nodes can be numbered according to the following
rule: the origin of the network takes the number 0 and it
is not considered in the subsequent calculation as we
assume that the voltage at this node, V 0 , is known exactly.
The other nodes are numbered sequentially imposing that a
receiving node takes a number higher than the sending
node nearer to it. The terms receiving and sending are
used under the assumption that in a traditional radial network, i.e. without distributed generators, the power ow is
directed from a lower to a higher number. The branches are
identied by the same number as their receiving node, as
shown in Fig. 1.
This numbering method allows a simple storage of the
network structure in a single square matrix (called incidence matrix, [A]) whose dimension is (n n). In particular,
the rows corresponds to the n branches and the columns to
the nodes. The elements of [A] describe the network topology and are equal to 1 if the node corresponding to column
j is fed through the branch corresponding to row i; 0
otherwise.
The calculation of the branch ows is easily obtained
applying the mesh method for network analysis. It can
be easily shown that
J  AI

where
I is the vector of the load currents, dimension (n 1);
J  is the vector of the branch currents, dimension
(n 1).
The network complex impedance is equal to
4

I4
4

1
1
7

I7

Z At  Z b   A

where Z b  is the diagonal matrix (n n) whose elements are

the complex impedances of the corresponding branches.
For the ith node the complex power S i is dened as
S i V i I i P i jQi

where

2. Deterministic load-ow (DLF)

1475

5
5

3
6

I3

I6

Fig. 1. One-line diagram of a three-phase symmetrical, radial distribution

network.

V i is the voltage phasor at node i;

I i is the current phasor at node i;
I i is the complex conjugate of I i ;
Pi is the net real power at node i;
Qi is the net reactive power at node i.
The real and reactive powers Pi and Qi can be regarded
as the dierence between the powers absorbed by the load
(PLi, QLi) and the powers delivered by the generators
(PGi, QGi) connected at node i:
P i P Li  P Gi

Qi QLi  QGi

Then, the voltage at the ith node can be evaluated by

means of the following expression:
Vi V0

n
X
j1

zij 

P Lj  P Gj  jQLj  QGj
V j

Assuming phasor V 0 on the real axis, neglecting the phase

dierence between the node voltages (which is commonly
done in MV and LV distribution networks analysis) and
considering the real components of expression (5) we
obtain

n 
X
1
Vi V0
Rij P Lj  P Gj X ij QLj  QGj 
6
Vj
j1
for i = 1 . . . n, where Rij and Xij are, respectively, the real
and imaginary part of Z ij , generic element of matrix Z.
A possible use of these equations is calculation of the
allowed PV DG penetration level for a given network in
order to ensure that the voltage limits at the point of common coupling (PCC) and the lines current carrying capacity are not exceeded (Conti et al., 2003c).
To do this the values of loads and generations at each
node, e.g. for each hour of the day, must be known and
provided as inputs to the solving program. Typically, the
available inputs are the loads and generations mean values
(or expected values) for each hour, i.e. the possible stochastic variations that can be introduced in the absorbed and
produced powers by customers behaviour and meteorological conditions are not taken into account.
As a consequence, especially in the presence of DG
based on renewable energy sources, such as solar energy,
it is not possible to achieve a realistic evaluation of where
and when overvoltage and/or overcurrent occurs in a distribution network during a year by simply using a DLF

1476

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

analysis, which is based on some selected combinations of

consumer loads and PV power productions.
3. Probabilistic load-ow (PLF)
As outlined in the previous Section, a realistic investigation of node voltages is possible if the statistical variations
of loads and productions are taken into account. This is
done by means of appropriate statistical models for loads
and PV generators productions, which will be presented
in the following subsections. The consumer loads and PV
generator productions are assumed to be mutually
independent.
Once the statistical models are dened in terms of probability density functions (pdf), Monte Carlo simulations
(MCS) involves repeating the simulation process using in
each simulation a particular set of values of the random
variables (loads and PV productions at each node of the
considered distribution network, hour by hour) generated
in accordance with the corresponding pdf, as explained in
Section 4.
A sample from a MCS is similar to a sample of experimental observations. Therefore, the results of a PLF based
on MCS in terms of power ows at the various sections of
the feeders and voltage proles at all nodes of the network,
may be treated statistically, and the methods of statistical
estimation and inference are applicable.

Many studies have proved that cloudiness is the main factor aecting the dierence between the values of solar radiation measured outside the atmosphere and on earthly
surface. To account for the dierence between these two
values, a daily or a hourly clearness index is used. The
hourly clearness index, kt, is dened as
kt

It
I0

where
It is the irradiance on an horizontal plane [kW/m2],
I0 is the extraterrestrial total solar irradiance [kW/m2].
Known kt it is possible to determine the irradiance on a
surface with inclination b to the horizontal plane, Ib [kW/
m2] (Conti et al., 2002; Gagliano et al., 2006).
Since the PV system is usually equipped with a Maximum Power Point Tracker (MPPT) and the relationship
between the maximum power per unit area of array surface
available from the PV system and Ib is linear, (Kroposki
et al., 1994), the power output of the PV system (Ppv) is
given by
Ppv AC  g  I b AC  g  T  kt  T 0  k2t

where
3.1. Probabilistic load model
The load is assumed to be a random variable (PL) normally distributed within each hour of a given month (Hatziargyriou et al., 1993). Then, the probability density
function of PL is given by the following expression:
 2
PL P
L
1
fPL PL p  e 2r2
r 2p

where
 L is the mean value;
P
r is the standard deviation.
This means that an expected value and a standard deviation specify the load at each load-point and for each hour.
Note that the variables that depend on the hour of the
day and on the month of the year are written in boldface.
All loads are fully correlated and follow the same pdf.
Hence a single pdf can be used for the total load of the network. The feeder load is then distributed among the nodes
in proportion to their contract rated capacity.
3.2. Probabilistic PV generator model
The amount of solar radiation that reaches the ground,
besides on the daily and yearly apparent motion of the sun,
depends on the geographical location (latitude and altitude) and on the climatic conditions (e.g. cloud cover).

AC is the array surface area [m2];

g is the eciency of the PV system in realistic reporting
conditions (RRC) (Gay et al., 1982);
T and T 0 are parameters that depend on inclination b,
declination d, reectance of the ground q, latitude /,
hour angle x, sunset hour angle xs, day of the year y
(Conti et al., 2002; Gagliano et al., 2006).
From (9), if the probability density function fkt kt for
the random variable kt is known, it is possible to obtain
the probability density function fPpv Ppv for Ppv by applying the fundamental theorem for the function of a random
variable (Papoulis, 1982).
In the present work the expression used for the probability density function of kt is the one proposed by Hollands
and Huget (1983).
Depending on the sign of the parameters T and T 0 , the
probability density function fPpv Ppv has four dierent
expressions but only two have a physical meaning (Conti
et al., 2002; Gagliano et al., 2006).
In particular, if T > 0 and T 0 < 0 we have
8
Ck tu 12aa0
k
0
>
>
< ktu Ac gT 0 a0  e2aa
10
fPpv Ppv
if Ppv 2 0; Ppv k tu 
>
>
:
0; otherwise
while, if T > 0 and T 0 > 0 we have

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

8
1
0
k2aa0
>
< Cktu 2aa
e
k tu Ac gT 0 a0
fPpv Ppv
if
P
2
0;
Ppv k tu 
pv
>
:
0; otherwise

1477

Read input data

11
yY

where
a

T
T0

and

s
Ppv
a0 a2  4 
g  T 0  Ac

STOP

Statistic Data
Generation

12

Hourly LF:
nodal voltage (Vh,i) and
branch current (Ih,b)
calculation

ktu is the upper bound for kt;

C, k are parameters of the probability density function
for kt (Hollands and Huget, 1983).
In the software implementation, presented in the next
section, the solar radiation at all nodes are fully correlated
and follow the same pdf.

Vh,i < Vmin

Vh,i > Vmax
Ih,b > Imax

4. Software implementation of the PLF algorithm

F

h = h+1

The authors developed a software tool on MATLAB

platform to implement the PLF calculations. This program
performs a PLF based on MCS, in which from the generated load and the PV power production time series (according to the pdfs previously described) the corresponding
node voltages and branch currents time series are obtained,
performing subsequent LF calculations.
Eventually, the calculated voltage and current time series are used for deriving the probability distribution of the
voltage at each node and the current at each branch. The
application of the European Standard EN 50160 is also
possible, since it recognizes the statistical nature of the
voltage variations.
One of the most important application of the proposed
method is the evaluation of the maximum value of the total
PV peak power that can be installed in the nodes of the network without violating voltage and current constraints
(Conti et al., 2003c).
To do this it is preliminary necessary to assess the average number of hours in the year with under/overvoltage
and overcurrent.
Fig. 2 shows the ow chart explaining the PLF algorithm used to perform this calculation.
The input data are provided to the program in terms of:
parameters of the pdfs for power required by the loads
and produced by the PV generators for each hour;
electrical and geometrical parameters of the distribution
network;
initial nodal voltage values;
minimum and maximum voltage values (Vmin and Vmax)
allowed by EN 50160;
values of maximum current allowed in each branch
(Imax) given by the line current carrying capacity.

Write Results:
under/overvoltage
and overcurrent

y = y+1
T

hH
F

Fig. 2. Flow chart of the PLF algorithm.

h = 1 . . . H with H = 8760, number of hours in a year) of

the year (y, y = 1 . . . Y) loads and productions statistic values (to ensure reasonable accuracy of the calculation from
the Monte Carlo simulations, an opportune number of repetitions Y are to be performed).
For each hour the following steps are performed:
a LF calculation is carried out using as inputs the sample values of loads and PV productions. In this way it is
possible to obtain for each hour the sample value of the
voltage at each node (Vh,i, i = 1 . . . n) and the current in
each branch (Ih,b, b = 1 . . . n);
a test is done in order to detect undervoltage (Vh,i <
Vmin), overvoltage (Vh,i > Vmax), and overcurrent
(Ih,b > Imax)
the results are saved and made available for subsequent
analysis;
if the counter h is less than H, it is increased by one,
otherwise the counter y is increased by one and the
new statistic data are generated for the current year.
To reduce the calculation time it is necessary to choose
an appropriate initial value for the nodal voltages. The initial value is calculated carrying out a fast DLF (with linear
equations) which assumes constant current model for loads
and generators.
5. Radial distribution network with PV DG

Based on these detailed input specications and on the

probability density functions described in Sections 3.1
and 3.2, the program generates for each hour (h,

In this section the application of the proposed algorithm

to the study of a realistic three-phase LV distribution

1478

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

network will be shown. The structure of the considered network is represented in Fig. 3.
The network supplies 36 residential loads connected to
nine load points (LPs), labelled sequentially from 3 to 11
in Fig. 3. Each load can absorb a maximum contractual
power of 3.3 kW and it is connected to its LP by means
of a short line (here called customer cable).
The load demand used for DLF calculations is specied
in terms of mean value at each hour and it is derived by
representative proles of residential customers.
As explained in Section 3.1, in order to implement the
PLF calculations, load demand in a given hour, PL, will
be represented by a normal probability density function
of a random variable (Hatziargyriou et al., 1993), characterised by an expected value PL and a standard deviation
r. The load expected values for each hour are the same
as the mean values used in DLF calculations.
The standard deviation will be set to 15% of the
expected value.
The electrical and geometrical parameters of the network are shown in Table 1.
Three dierent cases will be identied and studied with
reference to the position of the PV generators connected
to the LPs:
Case 1. PV generators installed at the terminal LPs (5, 8
and 11).
Case 2. PV generators installed at the initial LPs (3, 6 and
9).
Case 3. PV generators installed at all the LPs.

In all the above cases, we assume to install in each feeder

the same per cent PV peak power with respect to the feeder
overall contractual load.
6. DLF calculations
Let us indicate the total PV peak power installed in the
network by PPVpeak.
Firstly, the maximum value of the total PV peak power
(PPVpeak(max)DLF) that can be installed in the network without violating voltage and current constraints (Conti et al.,
2003c) has been calculated by means of a DLF in all the
three cases.
To do this the implemented algorithm increases sequentially the installed PV peak power until under/overvoltage
and/or overcurrent occur in the distribution network during a year.
The results are expressed in terms of per cent value of
PPVpeak(max)DLF with respect to the network overall contractual load (PLTOT = 118.8 kW):
P PVpeakmaxDLF %

P PVpeakmaxDLF
 100
P LTOT

13

as summarized in Tables 24.

From these Tables, it can be noted that
PPVpeak(max)DLF % is a function of the LV busbars voltage
(V0) (Conti et al., 2003a) and the maximum voltage variation (m.v.v.) allowed in the customer cable.
Table 2
PPVpeak(max)DLF % as a function of V0 without accounting for the customer
cable eect
PPVpeak(max)DLF %

4
3

4
3

MV/LV
substation

440

430

420

Case 1
Case 2
Case 3

29
33+
32+

55
66+
62+

82
100+
92+

410
+

109
114
121+

400

390

382

113
111
150+

110
108
150

107
105
148

Table 3
PPVpeak(max)DLF % as a function of V0 assuming a customer cable m.v.v. of
1.5%

PPVpeak(max)DLF %

11

10
9

11

10

Fig. 3. Radial LV distribution network supplying residential loads.

Table 1
Electrical and geometrical parameters of the network
Branches Phase + neutral r
x
Total
Line current
line section
(X/km) (X/km) feeder
carrying
(mm2)
length (m) capacity (A)
2
345
91011
678

V0 [V]

3 50 + 25Cu
3 50 + 25Cu

0.391
0.391

0.078
0.078

100
300

166
166

3 50 + 25Cu

0.391

0.078

240

166

V0 [V]

434

430

420

410

Case 1
Case 2
Case 3

29
33+
32+

39
46+
44+

66
79+
73+

92
112+
102+

400

390

388

113
111
131

110
108
150

109
106
149

Table 4
PPVpeak(max)DLF % as a function of V0 assuming a customer cable m.v.v. of
3%
PPVpeak(max)DLF %
V0 [V]

428

420

410

Case 1
Case 2
Case 3

29
33+
32+

49
59+
55+

75
91+
84+

400

393
+

101
111+
113+

110
109
133

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

In order to explain the meaning of parameter m.v.v.,

consider that overvoltage and undervoltage power limits
are evaluated at the LPs, which are not the actual connection point of the residential loads (points of common coupling PCC with the generators). The residential loads
are supplied through a customer cable line, which has the
eect to increase or decrease the voltage at the PCC
depending on the sign of its voltage variation (i.e. on the
power ow direction). As a consequence, voltage limits
that are not exceeded at the LPs may be exceeded at the
PCCs.
Note that the section of the cable line is chosen by
imposing the desired maximum voltage variation (in the
following called m.v.v., which is a design parameter and
whose limit value is imposed by the standards as per cent
value of the nominal voltage). Consequently, it is possible
to establish how great the voltage variation can be at the
connection points with regard to the corresponding LP
voltage.
Then, in order to have a better evaluation of
PPVpeak(max)DLF, we have considered three examples to
account for the presence of the customer cable: m.v.v. =
0% (Table 2), m.v.v. = 1.5% (Table 3), m.v.v. = 3%
(Table 4).
Then, the value of PPVpeak(max)DLF is calculated assuming that in the LPs the maximum voltage value reached
(Vsup) is lower/equal than/to the maximum voltage value
allowed by the European Standard EN 50160 (Vmax):
V sup V max  jm:v:v:j

14

Similarly, the minimum voltage value reached (Vinf) has to

be higher/equal than/to the minimum voltage value allowed by the standard (Vmin):
V inf V min jm:v:v:j

15

As for the meaning of the superscripts used in Tables 24,

the symbol + refers to the values of PPVpeak(max)DLF %
for which the higher voltage limit (Vsup) is reached at least
in one LP;  refers to the values of PPVpeak(max)DLF % for
which the current limit (thermal limit) is reached at least in
one network branch;  refers to the power values for
which the thermal limit is reached, but it is not possible
to further reduce voltage V0 or reduce DG penetration
without causing undervoltages.

1479

Table 5
Average number of hours with overvoltage, undervoltage and overcurrent
per year obtained by the PLF in Case 3 with reference to the PV DG
penetration given by PPVpeak(max)DLF %
V0 [V]

PPVpeak(max)DLF %

lOV

rOV

lUV

rUV

lOC

rOC

434
430
420
410
400
390
388

32
44
73
102
131
150
149

430.3
166.7
32.5
18.8
13.7
0
0

33.97
21.00
7.28
3.27
2.24
0
0

0
0
0
0
0
11.7
9.3

0
0
0
0
0
2.4
2.5

0
0
0
0
0
1.9
91.4

0
0
0
0
0
1.6
13.97

lOV yearly average number of hours with overvoltage.

rOV yearly average standard deviation of the number of hours with
overvoltage.
lUV yearly average number of hours with undervoltage.
rUV yearly average standard deviation of the number of hours with
undervoltage.
lOC yearly average number of hours with overcurrent.
rOC yearly average standard deviation of the number of hours with
overcurrent.

voltage and thermal constraints according to the results

of the DLF (PPVpeak(max)DLF %).
Table 5 shows the number of hours with undervoltage
(UV), overvoltage (OV) and overcurrent (OC) occurring
in the network in terms of yearly average number (lOV,
lUV, lOC) and yearly average standard deviations
(rOV, rUV, rOC).
From the analysis of Table 5 we can remark that the
probabilistic method highlights the possibility to have voltage and thermal constraints violation due to PV peak
power values that according to the DLF procedure would
not cause it (this is especially true for high values of V0).
We have reported the third case only for the sake of
brevity. However, it can be concluded that in all the cases
considered the penetration limits calculated by means of
the DLF are higher than the ones calculated by the PLF,
which provides a more restrictive and realistic evaluation.
Fig. 4 shows the graphs of the expected value lOV and
standard deviation rOV obtained by the PLF in Case 3
for V0 = 430 V and m.v.v. = 1.5%.

200
Mean Value

OV
150

7. PLF calculations
In this section, the PLF is applied to the three cases
described in Section 5. The expected values of the probability density functions used for loads and generators power
are the same as the mean values used in the DLF calculations. This allows a correct comparison between the results
of the DLF and PLF.
Table 5 refers to Case 3 accounting for a customer cable
m.v.v. of 1.5%. The calculation is performed assuming the
PV DG penetration level that do not cause violations of

100

50
STD

OV

0
1

12

23

34

45

56

67

78

89

100

Repetitions
Fig. 4. Graphs of expected value and standard deviation obtained by the
PLF in Case 3 with V0 = 430 V.

1480

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

For all the simulations performed in the present study

the number of repetitions used to achieve a good accuracy
is 100.
Then, the maximum PV peak power that can be
installed without violating voltage and current constraints
(PPVpeak(max)PLF %) has been calculated by means of the
PLF.
The PLF algorithm has been developed in accordance to
the European Standard EN 50160, which states that:
under normal operating conditions, excluding situations
arising from faults or voltage interruptions
during each period of one week 95% of the 10 min mean
rms values of the supply voltage shall be within the
range of Un 10% (where Un is the nominal voltage);
all 10 min mean rms values of the supply voltage shall be
within the range of Un + 10% 15%.
In cases of electricity supplies in remote areas with long
lines the voltage could be outside the range of Un + 10%
15%.
The aforesaid conditions are equivalent to the following
expressions:


P U n  10% 6 v 6 U n 10% F v U n 10%  F v U n  10% P 0:95

P U n  15% 6 v 6 U n 10% F v U n 10%  F v U n  15% 1

16

where
v is the random variable voltage;
P is the probability;
Fv is the cumulative distribution function of random
variable voltage, v.
The results are shown in Tables 68, which provide the
quantity PPVpeak(max)PLF % as a function of V0.

Table 6
PPVpeak(max)PLF % as a function of V0 without accounting for the customer
cable eect
PPVpeak(max)PLF %
V0 [V]

440

430

420

Case 1
Case 2
Case 3

24
27+
27+

50
61+
57+

77
95+
87+

410
+

104
109
116+

400

390

382

367

108
106
145+

105
103
145

102
100
143

99
97
140

Table 7
PPVpeak(max)PLF % as a function of V0 assuming a customer cable m.v.v. of
1.5%
PPVpeak(max)PLF %
V0 [V]

434

430

420

410

Case 1
Case 2
Case 3

24
27+
27+

34
41+
39+

61
74+
67+

87
107+
97+

400

390

388

373

108
106
126+

105
103
145

104
101
144

100
97
140

Table 8
PPVpeak(max)PLF % as a function of V0 assuming a customer cable m.v.v. of
3%
PPVpeak(max)PLF %
V0 [V]

428

420

410

400

Case 1
Case 2
Case 3

24
27+
27+

44
54+
50+

70
86+
79+

96
106+
108+

393

379

105
104
127

102
101
124

From these Tables, it can be noted that

PPVpeak(max)PLF % is lower than PPVpeak(max)DLF % in all
the studied cases.
By comparing Tables 24 with, respectively, Tables 68,
we can conclude that the use of a traditional DLF leads to
an overestimation of the maximum PV peak power that
can be installed according to the evaluation performed by
the PLF.
8. Conclusion
In this paper theoretical concepts, models and their
application have been presented to develop a probabilistic
load ow (PLF) using Monte Carlo techniques for radial
distribution networks with grid-connected photovoltaic
generators.
A software tool has been developed to implement the
proposed algorithm, which allows to evaluate the eect of
connecting PV DG units to the distribution network, with
specic regard to slow voltage variations and line current
carrying capacity.
In practice, the procedure allows to evaluate the penetration limits of PV DG with respect to voltage and thermal constraints.
The proposed PLF recognizes the statistical nature of
the voltage variations and it is in keeping with the
approach of the European Norm EN50160.
For example, it is possible to calculate the average number of hours with under/overvoltage and overcurrent per
year for a distribution network with several PV generators.
Also, this makes it possible to calculate the yearly energy
not supplied to the network owing to overvoltage protections tripping and consequent PV generators disconnection
to avoid unacceptable overvoltage.
A case study has been presented to show the application
of the proposed method to solve the PLF. The results
obtained by a traditional deterministic load ow (DLF)
and the proposed PLF, applied to a realistic LV distribution network with PV distributed generation (DG), have
been compared. The two LF analyses have been used to
assess the maximum PV peak power that can be installed
into a network without violating voltage and current constraints. It has been shown that the use of a traditional
DLF leads to overestimation of the maximum PV peak
power that can be installed according to the evaluation performed by the PLF. Then, the PLF method using Monte
Carlo techniques gives a better knowledge of the inuence
PV DG has on correct network operation.

S. Conti, S. Raiti / Solar Energy 81 (2007) 14731481

Further, the method can give information on many

other aspects of distribution network operation, such as
on how many hours per year overvoltage, undervoltage
and overcurrent occur if the PV peak power is increased
beyond a given threshold.
Note that, so far, the problem of possible overvoltage at
DG connection nodes has been solved by network operators by simply using maximum voltage protections embedded in the generation systems, which operate shutting
down the generation unit (on/o control) when the node
voltage is close or at the statutory limit. This has a potential negative impact on a correct exploitation of energy
made available by the installed PV arrays. Then, applying
voltage relays for stopping PV generators should, only
after careful analysis, be used as an alternative to reinforcement of the network, including the possibility of modifying
the network voltage regulation criteria. So, at a planning
stage, a useful application of the proposed PLF is the analysis of the cost-benet ratio relative to installation of voltage relaying vs. grid reinforcement or new control systems.
References
Conti, S., Crimi, T., Raiti, S., Tina, G., Vagliasindi, U., 2002. Probabilistic
approach to assess the performance of grid-connected PV systems. In:
Proceedings of 7th International Conference on Probabilistic Methods
Applied to Power Systems, Naples, Italy.

1481

Conti, S., Raiti, S., Tina, G., 2003a. Small-scale embedded generation
eect on voltage prole: an analytical method. IEE Proceedings on
Generation, Transmission and Distribution 150 (1), 7886.
Conti, S., Raiti, S., Tina, G., Vagliasindi, U., 2003b. Integration of
multiple PV units in urban power distribution systems. Journal of
Solar Energy 75 (2), 8794.
Conti, S., Raiti, S., Tina, G., Vagliasindi, U., 2003c. Distributed
generation in LV distribution networks: voltage and thermal constraints. In: Proceedings of the 2003 IEEE PowerTech Conference,
Bologna, Italy.
European Standard EN 50160, 1999. Voltage characteristics of electricity
supplied by public distribution systems.
Gagliano, S., Raiti, S., Tina, G., 2006. Hybrid solar/wind power system
probabilistic modelling for long-term performance assessment. Journal
of Solar Energy 80 (5), 578588.
Gay, C.F., Rumberg, J.E., Wilson, J.H., 1982. AM/PM-All-Day module
performance measurements. In: Proceedings of the 16th IEEE Photovoltaic Specialists Conference, San Diego, CA, USA.
Hatziargyriou, N.D., Karakatsanis, T.S., Papadopoulos, M., 1993.
Probabilistic load ow in distribution systems containing dispersed wind power generation. IEEE Transactions on Power Systems
8 (1).
Hollands, K.G.T., Huget, R.G., 1983. A probability density function for
the clearness index, with applications. Journal of Solar Energy 30, 195
209.
Kroposki, B., Emery, K., Myers, D., Mrig, L., 1994. A comparison of
photovoltaic module performance evaluation methodologies for
energy ratings. In: Proceedings of the 1st WCPEC, Hawaii, USA.
Papoulis, A., 1982. Probability, Random Variables and Stochastic
Processes. McGraw-Hill, New York.