Chapter 3

Probabilistic Power Flow Module

for PowerFactory DIgSILENT

Saeed Teimourzadeh and Behnam Mohammadi-Ivatloo

Abstract Electric power system could be known as the most complex, large-scale
human-built assembly. Power system engineers are deal with several challenging
problems such as power system operating, planning, and protection. Different types
of studies should be carried out in order to handle different problems of power
system. However, like many other engineering endeavors, future operating condi-
tions, and operating criteria could not be predicted with certainty. Load increment,
generation scheduling, and power system topology are most uncertain conditions in
which power system engineers are deal with them. Therefore, in order to drive a
reliable system, along with deterministic studies, probabilistic studies should be
carried out. Since power system includes multiple numbers of linear and nonlinear
elements, computer-based software packages are sufficiently helpful. In most of the
computer packages such as DIgSILENT PowerFactory and PSSE/PTI, determin-
istic analysis tools such as deterministic power flow (DLF) are available. However,
probabilistic characteristic of these studies is not included in the software. In this
chapter, a probabilistic power flow (PLF) analysis module for DIgSILENT Pow-
erFactory is proposed. The proposed module is implemented using DPL. Two
different PLF analysis methods are incorporated in the proposed module including
Monte Carlo simulation and point estimate method. The user can select the desired
algorithm for PLF analysis. A line-by-line explanation is provided for the written
DPL code. The application of the proposed module is investigated using illustrative
example incorporating IEEE 14-bus standard test system.

Electronic supplementary material The online version of this chapter (doi:10.1007/978-3-

319-12958-7_3) contains supplementary material, which is available to authorized users.

S. Teimourzadeh
B. Mohammadi-Ivatloo (&)

Faculty of Electrical and Computer Engineering, University of Tabriz,
29 Bahman Blvd, Tabriz, Iran
e-mail: bmohammadi@tabrizu.ac.ir
S. Teimourzadeh
e-mail: teimourzadeh91@ms.tabrizu.ac.ir

© Springer International Publishing Switzerland 2014 61

F.M. Gonzalez-Longatt and J. Luis Rueda (eds.), PowerFactory Applications
for Power System Analysis, Power Systems, DOI 10.1007/978-3-319-12958-7_3
62 S. Teimourzadeh and B. Mohammadi-Ivatloo

Keywords Deterministic power flow
Probabilistic power flow
Monte Carlo

simulation Point estimate method

3.1 Introduction

Most of the engineering problems are subject to uncertainties owing to either

inherent randomness or inaccurate considered assumptions in modeling approach
[1]. Considering probabilistic aspects of the engineering problems allows engineers
to have a better feeling about the future outputs of the problem. Electrical power
system, as the largest human-built system, is always subjected to the various types
of uncertainties, i.e., daily load variation, generation outage, faults, and failures.
Due to the presence of different decision makers, the level of the uncertainties will
be even more in the restructured electrical power systems. Effective uncertainty
modeling tools are required to control and minimize the risks associated with the
design and operation of power systems [2].
Power flow studies are the most efficient tools which help power system engi-
neers to investigate system state variables, i.e., magnitude and angle of the bus
voltages. However, deterministic power flow (DLF) studies provide system vari-
ables under a certain load and generation circumstances. Load and generation
uncertainties are not considered in DLF studies. As the results of power flow studies
are used as essential inputs for decision making of system planners, considering the
effective uncertainties is vital. Probabilistic power flow (PLF) can provide more
information about the possible future conditions of the power system [3, 4]. In the
probabilistic load flow studies, the input and system variables of power system are
considered as random variables. Hence, the possible ranges of power flow results
could be identified.
In order to handle the PLF problem, different approaches have been proposed in
the literature. The proposed methods could be categorized into three main cate-
gories as follows: Monte Carlo simulation (MCS)-based methods [5–7], analytical
methods [8–12], and approximate approaches [10, 13–16]. The MCS-based
methods do not require approximation of the power flow problem and solve the
problem via repetitive process. The main drawback of the MCS method is the
requirement of large number of simulations, which needs large amount of com-
putation time and storage. However, with the development of the efficient com-
putation tools in the recent years, the MCS is used frequently in PLF studies [5–7].
In the analytical approaches for PLF, the power flow equations are linearized to
make working with probability density functions easier. The approximation
methods did not require linearization of the power flow equations. Instead, in these
methods, the number of evaluation points is decreased using several techniques
such as point estimate method [14]. Latin hypercube sampling (LHS) is also used
for PLF studies with the limited number of simulations [7, 17, 18]. The LHS
method is combined with Nataf transformation in [17] and Cholesky decomposition
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 63

in [18] for PLF analysis. For more details on the efficiencies and deficiencies of the
mentioned approaches, refer to [1]. In this chapter, MCS method and point estimate
method are used for PLF analysis and the detailed descriptions of them are provided
in the following sections.

3.2 Monte Carlo Simulation

MCS is an iterative approach which utilizes cumulative density function (CDF) of

random variables to determine the final result. In this method, firstly, certain numbers
of samples should be generated corresponding to the density function of random
variables. Then, fitness function should be evaluated for all generated samples.
Afterward, mean value of the outputs is set as result of the iteration. Discussed
process should be derived until the termination criterion satisfies. Various termina-
tion criterions are proposed in the literature; the maximum numbers of iterations and
output variation-based criterions are the most popular criterions. Figure 3.1 depicts
the flowchart of the MCS. More details on MCS are available in [2].

3.3 Point Estimate Method

The main drawback of the MCS-based studies is the fact that a great number of
iterations are needed to reach the desired convergence [1]. However, approximate
approaches require a fewer number of iterations to attain the desired convergence.
Point estimate-based methods, as the approximate approaches, are the most popular
and efficient approximate methods. The 2m and 2m + 1 point estimate methods are
widely proposed and utilized in the literature [1, 13–15]. The point estimate-based
approaches offer to evaluate the fitness function by using some specially determined
points. Analytical investigations prove that, despite light computational burden,
desired convergence characteristic is accessed. The main steps of point estimate
method are discussed as follows.
In this method, k points should be estimated for each random variable. There-
fore, if there are M input variables, the fitness function should be evaluated k  M
times. During each evaluation process, input vector is calculated as follows:

ðlp1 ; lp2 ; . . .; Ppl ; . . .; lpM Þ ð3:1Þ

where lpi is the mean value of the ith input variable. Additionally, Pl;k is calculated
as follows:

Ppl ¼ lpl þ nl;k rpl ð3:2Þ

64 S. Teimourzadeh and B. Mohammadi-Ivatloo

Fig. 3.1 Flowchart of Monte

Start
Carlo simulation

Set parameters:
Itermax, Nsample, Density functions

Initialize iteration counter iter=0

Initialize Sample counter Sample=0

Update iteration counter iter=iter+1

Update sample counter Sample=Sample+1

Generate sample

Evaluate members

Save result

Yes
Sample < Nsample ?

No
Calculate Mean Value

Termination Criterion No
Satisfied?

Yes
End

where lpl and rpl are the mean and standard deviation of the ith input random
variable, respectively. The nl;k is the standard location which is related to the
number of estimated points. For instance, for 2m + 1 point estimate method,
standard locations could be calculated using (3.3–3.5) [19].
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kl;3 3
nl;1 ¼ þ kl;4  k2l;3 ð3:3Þ
2 4
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kl;3 3
nl;2 ¼  kl;4  k2l;3 ð3:4Þ
2 4
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 65

nl;3 ¼ 0 ð3:5Þ

where kl;3 and kl;4 are the skewness and kurtosis of lth input variable. After esti-
mating sample points, the fitness function should be evaluated for all estimated
points. Finally, expected values of outputs are reported as final results. In this
method, expected values are computed as follows:

m X
X K
E½Z ¼ xl;k Zðl; kÞ ð3:6Þ
l¼1 k¼1

where Z is the vector of output random variables and E½Z is the expected value of
vector Z. The xl;k are weighting coefficients and for 2m + 1 point estimate method,
these coefficients could be calculated as follows:

ð1Þ3k
xl;k ¼ k ¼ 1; 2 ð3:7Þ
fl;k ðfl;1  fl;2 Þ

1 1
xl;3 ¼  ð3:8Þ
m kl;4  k2l;3

where m is the number of input random variables. Figure 3.2 presents the flowchart
of the point estimate method.

3.4 Implementation of the Probabilistic Power Flow

Analysis

In this section, step-by-step implementation of the PLF analysis is presented. In the

following, implementation process of both Monte Carlo-based PLF analysis and
point estimate-based PLF analysis is discussed.

3.4.1 Monte Carlo Simulation

Step 1. First of all, CDF of all random variables should be prepared. In this
chapter, the loads are assumed as the input random variables. The loads are
assumed as normally distributed random variables with the mean value of
base load and optional standard deviation.
66 S. Teimourzadeh and B. Mohammadi-Ivatloo

Fig. 3.2 Flowchart of the Start

point estimate method
Set parameters:
Mean, Standard deviation
Skewness, kurtosis, Nvar, Npoint

Compute:
Standard Central Moments λl , j
Standard Locations ξl , k

Initialize counters l=0, i=1

Update point counter l=l+1

Compute:
Ppl = μ pl + ξl ,kσ pl
( μ p1 , μ p 2 ,..., Ppl ,..., μ pM )

i=i+1
Evaluate Fitness Function

Update Moments

Yes
i < Nvar ?

No
Yes
l < Npoint ?

No
End

Step 2. MCS is an iterative algorithm which needs excessive number of iterations to

access acceptable convergence. In this step, maximum number of iterations
should be set. Additionally, MCS method utilizes samples to evaluate the
fitness function. The maximum number of samples, also, is set in this step.
Step 3. In this step, samples should be determined. In order to determine the
samples in a specific iteration, firstly, two random numbers should be
generated. Afterward, samples are generated as follows [2]:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Si ¼ l þ r 2 lnðu1 Þ cosð2pu2 Þ ð3:9Þ

where Si is the ith sample. The li and ri are the mean value and the
standard deviation of the random variable, respectively. Additionally, u1
and u2 are uniformly distributed random number in the range of (0, 1). In
the probabilistic power calculations, the li is the base load and ri is
considered as an input data.
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 67

Step 4. In this step, the fitness function should be evaluated by using all of the
determined samples. In the PLF calculations, the fitness function is the
DLF analysis.
Step 5. Check sampling criterion: The sampling criterion is the maximum number
of samples. The algorithm will go to Step 6 if the maximum sample
number is reached; otherwise, it continues from Step 3.
Step 6. Expected values of the outputs are computed in this step. In the case of the
PLF analysis, bus voltages and transmission line loadings are considered
as the outputs. For a specific iteration, the computed expected values are
assigned as the results of the iteration.
Step 7. Check stopping criterion: The terminating criterion is the maximum
number of iterations. The algorithm will be terminated if the maximum
iteration number is reached; otherwise, it continues from Step 3.

3.4.2 Point Estimate Method

Step 1. Basic input data such as number of random variables (Nvar ), number of
estimated points (Npoint ), and probabilistic characteristics of random vari-
ables, i.e., the mean, standard deviation, skewness, and kurtosis, are
determined. In this chapter, 2m + 1 method is utilized. Therefore, three
points should generated using (3.1)–(3.5). In addition, system total load is
considered as the random variable and the Nvar equals to 1.
Step 2. In this step, standard central moments (kl;j ) and standard locations (nl;k )
should be determined. The kl;1 , kl;2 , kl;3 , and kl;4 are the mean, standard
deviation, skewness, and kurtosis of the lth random variable, respectively.
Moreover, the standard locations could be calculated using (3.3–3.5).
Step 3. The input vector of the power flow analysis should be computed in this
step. The input vector could be computed using (3.1).
Step 4. In this step, the fitness function should be evaluated. The DLF analysis is
considered as the fitness function. Therefore, the DLF is derived for the
input vector. The algorithm will go to Step 5, if the maximum variable
number is reached; otherwise, it continues from Step 3.
Step 5. The termination criterion is the maximum number of points. The algorithm
will be terminated if the maximum point number is reached; otherwise, it
continues from Step 3.
68 S. Teimourzadeh and B. Mohammadi-Ivatloo

3.5 DPL Implementation of Probabilistic Power Flow

Analysis

In this section, implementation of the PLF analysis by using DIgSILENT pro-

gramming language (DPL) is discussed. The prepared PLF module includes a
routine which has two subroutines. The main routine is an interface between users
and DIgSILENT to acquire some additional data. In the following, the main routine
is discussed.
The proposed main routine is provided as follows:

Lines 1 and 2 are used to define the variables which are utilized in the proposed
DPL script. In line 4, the method of PLF analysis is identified. The term
“Method = 1” provides MCS-based PLF analysis. On the other hand, in order to
derive point estimate-based PLF calculations, the term “Method = 2” should be
entered. In lines 5 and 6, a maximum number of iterations and number of samples
for MCS-based PLF analysis are identified, respectively. Different moments of the
random variable should be entered from lines 7 to 10. Additionally, DLF method is
set in line 10. The term “PF_Mode = 0” stands for balanced DLF analysis.
Alternatively, “PF_Mode = 1” provides unbalanced DLF calculations. The main
routine includes two subroutines entitled “MCS” and “PEM.” The MCS and PEM
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 69

provide MCS-based and point estimate-based PLF analysis, respectively. The set-
tings, which are acquired using lines 4–10, should be transferred to its corre-
sponding subroutine. Therefore, if Monte Carlo simulation method is chosen, lines
13.a–13.e transfer relevant data to the MCS. Similarly, if point estimate method is
chosen, lines 16.a–16.e transfer relevant data to the PEM. The subroutines are
executed by using either the term “MCS.Execute()” or “PEM.Execute()”. In the
following, details of both MCS and PEM subroutines are discussed. It is recom-
mended to use DIgSILENT user manual in order to get familiar with the proposed
DPL scripts.

3.5.1 MCS Subroutine

As previously mentioned, MCS subroutine provides MCS-based PLF calculations.

The proposed DPL script is as follows:
70 S. Teimourzadeh and B. Mohammadi-Ivatloo
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 71

Lines 1–6 are used to define necessary variables. In lines 8–17, essential objects
and sets are acquired. For instance, the term “AllRelevant(‘*.class-name’);” returns
a set with relevant objects, according to the given class name. Hence, the term
“S = AllRelevant(‘*.ElmLod’);” returns a set which is related to the load class and
72 S. Teimourzadeh and B. Mohammadi-Ivatloo

store it in S. The term “O = S.First();” is used to save the first object of the set S in
object O. In addition, in line 10, the term “PF = GetCaseObject(‘*.ComLdf’);” is
utilized to define object PF. Therefore, this object contains the class ‘*.ComLdf’
which presents load flow calculation. Afterward, line 11 is used to set the calcu-
lation method of load flow study. The term “PF_Mode” is defined in line 10 of
main routine. In line 18, the number of objects that are stored in the set S is acquired
using the term “n = S.Count();”. Since the set S represents the loads of the system,
n presents the number of system loads. The DIgSILENT user manual could be
efficient to investigate the functions and class names. Some vectors and matrices are
needed to store various outputs. In DPL script, the matrices and vectors should be
initialized. Lines 21–28 are defined to initialize the vital matrices and vectors. For
instance, the term “Vs.Init (Nsample, n_bus);” initializes the matrix entitled Vs, with
the dimensions of Nsample and n_bus. System total active and reactive loads are
calculated by lines 30–34. In line 33.a, for the nth load, the amount of active load is
acquired using the term “O:plini;”. Afterward, the acquired value is added to the
conventional value of Pt. Similar process is performed for reactive load using the
line 33.b. Moreover, the contribution of each load in forming the total load of the
system is computed using lines 37.a–37.c.
The sampling procedure and relevant calculations along with sample evaluation
and expected value computations are provided using lines 41–51. Lines 43.a–43.q
provide the sampling process. The sampling process is discussed in Sect. 4.1.
Moreover, the averaging process to compute the expected values of the output
variables is performed using lines 45–51. Lines 45–47 are used to compute the
expected value of bus voltage magnitudes. Similarly, lines 48–51 provide com-
puting the expected line loadings.
Finally, system total load is set to its initial value, and outputs are displayed
using lines 62.a–62.c and 65–81, respectively.

3.5.2 PEM Subroutine

The PEM subroutine of the main routine provides the point estimate-based PLF
calculations. In the following, proposed DPL script is presented.
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 73
74 S. Teimourzadeh and B. Mohammadi-Ivatloo
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 75

Most of the utilized DPL codes of this section are discussed in MCS subroutine.
Nevertheless, some different parts are discussed in the following.
In the proposed DPL script, lines 1–40 are similar to the MCS subroutine.
However, during lines 41–52, based on Eqs. (3.3–3.8), standard locations and
weighting coefficients are computed. The computed standard locations (lines
42–43) are stored in set eMAT using lines 44–46. Similarly, the computed
weighting coefficients (lines 47–49) are stored in set wMAT using lines 50–53.
Moreover, lines 55.a–55.f are utilized to compute the estimated points. After-
ward, the loads should be set to the estimated points. Discussed process is per-
formed using lines 55.h.I–55.h.V. Then, the power flow study is performed using
term “PF.Execute();” and the estimated points are evaluated. The voltage magni-
tudes and line loadings, as the output variables, are stored in sets Vs.set and Is.set
using lines 55.l–55.r.
Similar to the MCS subroutine, in lines 57–62, the expected values of output
variables are computed. Additionally, system loads are restarted using line 65.
Finally, the results are reported using lines 69–85.

3.6 Illustrative Examples

In this section, three illustrative examples are presented in order to verify accuracy
of the proposed methods. The IEEE 14-bus standard test system is simulated in the
power factory DIgSILENT and utilized to confirm the efficiency of the proposed
methods. The system data of the mentioned test system are available in [20].
However, some minor modifications are performed. In the modified form of the
IEEE 14-bus standard test system, the voltage magnitudes of all PV buses are set to
1.05 pu. In this section, firstly, DLF is derived on the mentioned test system.
Afterward, Monte Carlo and point estimate-based PLF calculations are carried out.
76 S. Teimourzadeh and B. Mohammadi-Ivatloo

For all cases, the system total load with normal density distribution is assumed as
the random variable and base load is considered as the mean value. However, the
standard deviation is an arbitrary parameter which is chosen by the user.

3.6.1 Case I: Deterministic Load Flow

As previously mentioned, in this example, the DLF analysis is derived. In this

example, balanced AC load flow analysis is implemented. Table 3.1 reports voltage
magnitude and voltage angles for different buses of the test system. According to
Table 3.1, all buses are working in their acceptable limits. Additionally, loading
percent of the various branches are presented in Table 3.2. According to Table 3.2,
all branches are operating in less than 100 % loading level.

3.6.2 Case II: Monte Carlo Probabilistic Load Flow

In this example, Monte Carlo-based PLF analysis is investigated. In order to derive

Monte Carlo-based PLF analysis, the method should be chosen in the main routine.
Therefore, the term “Method = 1;” should be set in 4th line of the main routine. In
addition, the value of standard deviation should be identified. As previously
mentioned, standard deviation is an arbitrary value which should be set by the user.
The mentioned parameter should be set in line 9 of the main routine. In this
example, PLF analysis is carried out by considering standard deviation of 15 % for

Table 3.1 Voltage

Bus number Voltage magnitude Voltage angle
magnitude and angle
(p.u) (deg)
(deterministic power flow
results) Bus 1 1.050 0.00
Bus 2 1.040 −5.13
Bus 3 1.010 −12.95
Bus 4 1.020 −10.58
Bus 5 1.024 −9.09
Bus 6 1.050 −15.00
Bus 7 1.050 −13.77
Bus 8 1.050 −13.77
Bus 9 1.020 −15.44
Bus 10 1.020 −15.65
Bus 11 1.032 −15.45
Bus 12 1.034 −15.87
Bus 13 1.029 −15.92
Bus 14 1.008 −16.70
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 77

Table 3.2 Branch loadings

Line number Loading percent
(deterministic power flow
results) Line(1) 25.36
Line(2) 60.74
Line(3) 33.54
Line(4) 58.79
Line(5) 19.91
Line(6) 45.23
Line(7) 51.22
Line(8) 73.90
Line(9) 23.56
Line(10) 21.57
Line(11) 36.95
Line(12) 33.16
Line(13) 78.14
Line(14) 7.942
Line(15) 26.37
Line(16) 39.56
Line(17) 31.57
Transformer(1) 43.53
Transformer(2) 15.60
Transformer(3) 31.04

system total load. Moreover, it should be mentioned that the MCS is done with 5
samples per iteration and 500 iterations. Similar to standard deviation, sample and
iteration numbers are arbitrary parameters which could be set using lines 6 and 8 of
the main routine, respectively. Voltage magnitudes of bus numbers 5 and 11 during
deriving each iteration of MCS are depicted in Fig. 3.3. According to Fig. 3.3,
owing to the load variation, voltage magnitudes of various buses are altering.
In the PLF analysis, the expected value and standard deviation of the outputs are
considered as the final results. Figure 3.4 presents the expected value of the voltage
magnitudes of bus numbers 5 and 11. In Fig. 3.4, mean value of each iteration is
computed with respect to the pervious iterations. According to Fig. 3.4, it could be
illustrated that the expected voltage magnitudes (EVMs) of bus numbers 5 and 11
are 1.025 and 1.032 (p.u), respectively. Moreover, final results of Monte Carlo-
based PLF analysis are reported in Tables 3.3 and 3.4.
According to Table 3.3, for PQ buses, the voltage magnitudes have small
variations in the vicinity of the mean values. In addition, comparison between the
results of Tables 3.1 and 3.3 illustrates that EVMs of some buses are different from
those calculated by DLF analysis. The EVM could be calculated by computing the
mean value of the output variables. Moreover, the loading percent of branches are
reported in Table 3.4. Second and fifth columns of Table 3.4 present the expected
loading percent (ELP) of transmission lines. Comparison between deterministic
results that are presented in Tables 3.2 and 3.4 confirms that lines loading are
78 S. Teimourzadeh and B. Mohammadi-Ivatloo

1.04
Bus 5
Bus 11

1.035
Voltage(p.u)

1.03

1.025

1.02

0 50 100 150 200 250 300 350 400 450 500

Iteration

Fig. 3.3 Voltage magnitude of bus numbers 5 and 11 during each iteration

(a)
1.029
Voltage (p.u)

1.028

1.027

1.026
X: 481

1.025 Y: 1.025

0 50 100 150 200 250 300 350 400 450 500

Iteration
(b)
1.037
Voltage(p.u)

1.036

1.035

1.034

1.033 X: 483
Y: 1.032

1.032
0 50 100 150 200 250 300 350 400 450 500
Iteration

Fig. 3.4 The expected voltage magnitudes of bus numbers 5 (a) and 11 (b)

somehow similar. However, some branches have considerable standard deviations.

For instance, Line (2) that is connected to the slack bus has considerable standard
deviation.
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 79

Table 3.3 Voltage

Bus number EVM Standard deviation
magnitude (probabilistic
power flow results) Bus 1 1.050000 6.45E−15
Bus 2 1.040000 8.22E−15
Bus 3 1.010000 6.22E−15
Bus 4 1.019891 0.001782
Bus 5 1.024819 0.001753
Bus 6 1.050000 6.45E−15
Bus 7 1.050000 6.45E−15
Bus 8 1.050000 6.45E−15
Bus 9 1.025501 0.001744
Bus 10 1.022224 0.00197
Bus 11 1.032395 0.001246
Bus 12 1.034099 0.001121
Bus 13 1.028521 0.001522
Bus 14 1.008346 0.002974

Table 3.4 Branch loadings

Line number ELP Standard deviation
(probabilistic power flow
results) Line(1) 25.3844 2.27
Line(2) 60.73101 4.67
Line(3) 33.54901 2.10
Line(4) 58.84013 4.15
Line(5) 20.10876 1.51
Line(6) 45.22532 2.98
Line(7) 51.21575 3.79
Line(8) 73.90754 5.18
Line(9) 23.53704 1.59
Line(10) 21.59663 1.55
Line(11) 36.95489 2.63
Line(12) 33.14841 2.32
Line(13) 78.11202 5.50
Line(14) 7.945386 0.57
Line(15) 26.38333 1.91
Line(16) 39.55121 2.76
Line(17) 31.66516 2.17

3.6.3 Case III: Point Estimate Probabilistic Load Flow

In this example, point estimate-based PLF analysis is studied. In order to derive

point estimate-based PLF analysis, the term “Method = 2;” should be set in line 4 of
the main routine. Similar to the Monte Carlo-based PLF analysis, some settings
80 S. Teimourzadeh and B. Mohammadi-Ivatloo

Table 3.5 The standard locations and weighting coefficients

Standard locations Value Weighting coefficients Value
nl;1 1.723623 xl;1 0.184584
nl;2 −1.419523 xl;2 0.224126
nl;3 0 xl;3 0.591290

should be identified beforehand. The standard deviation, skewness, and kurtosis are
the settings which should be identified by the user. In this study, the values of the
standard deviation, skewness, and kurtosis are set on 15 %, 0.3041, and 2.5392,
respectively. These values are considered based on the studies on the practical
power systems. Nevertheless, any other values could be adopted. In order to derive
point estimate-based PLF analysis, firstly, standard locations and weighting coef-
ficients should be computed using Eqs. (3.3–3.8). Table 3.5 presents the standard
locations and weighting coefficients for this example.
The estimated points could be computed by using Eq. (3.2) with respect to
Table 3.5. In this example, system total load is 259 MW and 81.4 MVAr. The
estimated points are presented in Table 3.6. According to Table 3.6, three points are
estimated, and for each point, the DLF analysis should be carried out and the results
should be saved. Afterward, the results are used to compute the expected value and
the standard deviation of outputs. According to Table 3.5, the standard location of
the first, second, and third points is positive, negative, and zero, respectively.
Hence, the first point should be greater than system’s base load and the second one
should be lower than system’s base load. In addition, the third point should be equal
to the base load of system. Table 3.6 confirms this claim.
Table 3.7 reports voltage magnitudes of the system buses using the estimated
points. According to Table 3.7, the results of point #3 are similar to Table 3.1.
However, owing to load increment that is estimated by point #1, the voltage
magnitudes corresponding to the point #1 are lower in comparison with those of
point #3. On the other hand, the voltage magnitudes of the second point are higher
than those of the third point. The results of Table 3.7 could be verified with respect
to the load increment and decrement which are estimated by the first and second
points, respectively. Moreover, expected value and standard deviation of the volt-
age magnitudes are reported in Table 3.8. The expected values and standard
deviations are computed using weighting coefficients of Table 3.5.

Table 3.6 The estimated

points Point Active power Reactive power
number (MW) (MVAR)
1 325.96 102.44
2 203.85 64.06
3 259.00 81.40
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 81

Table 3.7 Voltage

Bus number Point #1 Point #2 Point #3
magnitude related to the
estimated points Bus 1 1.05000 1.05000 1.05000
Bus 2 1.04000 1.04000 1.04000
Bus 3 1.01000 1.01000 1.01000
Bus 4 1.01281 1.02528 1.02000
Bus 5 1.01779 1.03009 1.02496
Bus 6 1.05000 1.05000 1.05000
Bus 7 1.05000 1.05000 1.05000
Bus 8 1.05000 1.05000 1.05000
Bus 9 1.01874 1.030869 1.025527
Bus 10 1.01461 1.028299 1.022242
Bus 11 1.02759 1.03624 1.032402
Bus 12 1.029795 1.037566 1.034095
Bus 13 1.022671 1.033227 1.028520
Bus 14 0.996875 1.017523 1.008363

Table 3.8 Expected voltage

Bus number EVM Standard deviation
magnitude (probabilistic
power flow results) Bus 1 1.05000 2.11E−08
Bus 2 1.04000 0
Bus 3 1.01000 0
Bus 4 1.01986 0.003971
Bus 5 1.02479 0.003919
Bus 6 1.05000 2.11E−08
Bus 7 1.05000 2.11E−08
Bus 8 1.05000 2.11E−08
Bus 9 1.02547 0.00386
Bus 10 1.02219 0.00435
Bus 11 1.03237 0.00275
Bus 12 1.03407 0.00247
Bus 13 1.02849 0.00335
Bus 14 1.00829 0.00657

According to the results of Table 3.8, there are some differences between the
results of Tables 3.8 and 3.3, which present the results of MCS. The MCS is an
iterative solution and takes large number of iterations to converge the result. In
Case II, MCS is derived only for 500 iterations. However, the point estimate
method finds the most promising solution without any iterative process. Finally,
Table 3.9 reports the expected branch loadings using point estimate-based PLF
analysis.
82 S. Teimourzadeh and B. Mohammadi-Ivatloo

Table 3.9 Expected branch

Line number ELP Standard deviation
loadings (probabilistic power
flow results) Line(1) 25.4224 5.02
Line(2) 60.8107 10.29
Line(3) 33.5847 4.64
Line(4) 58.9084 9.17
Line(5) 20.1253 3.40
Line(6) 45.2764 6.58
Line(7) 51.2797 8.36
Line(8) 73.9954 11.43
Line(9) 23.5647 5.02
Line(10) 23.5647 3.50
Line(11) 21.6225 3.44
Line(12) 36.9993 5.80
Line(13) 33.1881 5.11
Line(14) 78.2059 12.13
Line(15) 7.95503 1.26
Line(16) 26.4155 4.23
Line(17) 39.5985 6.10

3.7 Discussion

This chapter presents a PLF analysis for power factory DIgSILENT. Here are some
important features that should be considered about this chapter:
• This chapter deals with the PLF analysis.
• In this chapter, Monte Carlo-based and point estimate-based power flow anal-
yses are discussed.
• A DPL-based module is designed for DIgSILENT PowerFactory.
• The provided module is a flexible module, and for different cases and purposes,
all setting could be changed by users.
• The designed module is capable of handling both Monte Carlo-based and point
estimate-based PLF analysis.
• The Monte Carlo-based PLF analysis requires just standard deviation of total
load as input data. Nevertheless, owing to its iterative feature, it requires large
number of iterations to converge.
• The point estimate-based probabilistic power is the most promising analysis and
has low computational burden. However, it requires additional information of
the random variables, i.e., skewness and kurtosis.
• In the provided module, it is proposed to utilize point estimate-based analysis
when all information about random variable is available. On the other hand, for
those with little data, the Monte Carlo-based analysis is the best solution.
3 Probabilistic Power Flow Module for PowerFactory DIgSILENT 83

3.8 Conclusion

In this chapter, a new module is developed for PLF analysis using DIgSILENT
PowerFactory. The DPL is used for modeling the effect of load uncertainties in
power flow analysis. Two methods based on the Monte Carlo simulation and point
estimate method are incorporated in the developed module. The user can select the
used procedure for PLF calculation and perform the analysis with different con-
ditions. The written program in DPL is described in detail, and its implementation is
investigated using the illustrative IEEE 14-bus standard test system.

References

1. Morales JM, Perez-Ruiz J (2007) Point estimate schemes to solve the probabilistic power flow.
Power Syst IEEE Trans 22:1594–1601
2. Anders GJ (1989) Probability concepts in electric power systems
3. Borkowska B (1974) Probabilistic load flow. Power Apparatus Syst IEEE Trans 93
(3):752–759
4. Sexauer JM, Mohagheghi S (2013) Voltage quality assessment in a distribution system with
distributed generation—a probabilistic load flow approach. Power Deliv IEEE Trans
28:1652–1662
5. Rubinstein RY, Kroese DP (2011) Simulation and the monte carlo method, vol 707. Wiley,
New York (Wiley.com)
6. Jorgensen P, Christensen JS, Tande JO (1998) Probabilistic load flow calculation using Monte
Carlo techniques for distribution network with wind turbines. In: Proceedings of the 8th
international conference on harmonics and quality of power, vol 2. pp 1146–1151
7. Hajian M, Rosehart WD, Zareipour H (2013) Probabilistic power flow by Monte Carlo
simulation with latin supercube sampling. Power Syst IEEE Trans 28:1550–1559
8. Allan R, Leite da Silva A, Burchett R (1981) Evaluation methods and accuracy in probabilistic
load flow solutions. Power Apparatus Syst IEEE Trans 2539–2546
9. Meliopoulos AS, Cokkinides GJ, Chao XY (1990) A new probabilistic power flow analysis
method. Power Syst IEEE Trans 5:182–190
10. Zhang P, Lee ST (2004) Probabilistic load flow computation using the method of combined
cumulants and Gram-Charlier expansion. Power Syst IEEE Trans 19:676–682
11. Williams T, Crawford C (2013) Probabilistic load flow modeling comparing maximum
entropy and Gram-Charlier probability density function reconstructions. Power Syst IEEE
Trans 28:272–280
12. Yan C, Jinyu W, Shijie C (2013) Probabilistic load flow method based on Nataf transformation
and Latin hypercube sampling. Sustain Energy IEEE Transn 4:294–301
13. Su CL (2005) A new probabilistic load flow method. In: Power engineering society general
meeting, 2005. IEEE, pp 389–394
14. Su C-L (2005) Probabilistic load-flow computation using point estimate method. Power Syst
IEEE Trans 20:1843–1851
15. Verbic G, Cañizares CA (2006) Probabilistic optimal power flow in electricity markets based
on a two-point estimate method. Power Syst IEEE Trans 21:1883–1893
16. Soleimanpour N, Mohammadi M (2013) Probabilistic load flow by using nonparametric
density estimators. Power Syst IEEE Trans 28:3747–3755
17. Chen Y, Wen J, Cheng S (2013) Probabilistic load flow method based on Nataf transformation
and Latin hypercube sampling. Sustain Energy IEEE Trans 4:294–301
84 S. Teimourzadeh and B. Mohammadi-Ivatloo

18. Yu H, Chung CY, Wong KP, Lee HW, Zhang JH (2009) Probabilistic load flow evaluation
with hybrid latin hypercube sampling and Cholesky decomposition. Power Syst IEEE Trans
24:661–667
19. Hong H (1998) An efficient point estimate method for probabilistic analysis. Reliability Eng
Syst Saf 59:261–267
20. https://www.ee.washington.edu/research/pstca/