Sie sind auf Seite 1von 6

Three Phase Probabilistic Load Flow

in Radial Distribution Networks


A.C. Melhorn and A. Dimitrovski
Energy & Transportation Science Division
Oak Ridge National Laboratory
Oak Ridge, TN USA
Email: acmelhorn@gmail.com;
dimitrovskia@ornl.gov
K. Tomsovic
Department of Electrical Engineering
and Computer Science
The University of Tennessee
Knoxville, TN USA
Email: tomsovic@utk.edu
AbstractProbabilistic load ow is a helpful tool in accounting
for inconsistent or unknown loads and generation. This is
especially true with the push for renewable generation and
demand response. This paper proposes an improved version
of the probabilistic load ow solution for balanced distribution
systems and takes the next step by applying it to unbalanced
three phase systems. It is validated by comparing the solutions to
that obtained by Monte Carlo simulation. The proposed method
provides an accurate and practical way for nding the solution to
the stochastic problems occurring in power distribution system
analysis today.
Index TermsProbabilistic load ow, Power distribution sys-
tems, Three phase, Radial networks
I. INTRODUCTION
Probabilistic load ow (PLF), as the name suggests is an
extension of the load ow calculation. Load ow, also referred
to as power ow, is one of the most well known and used
calculations in power system analysis. Even though there are
a large variety of load ow techniques, most are very accurate
and allow for detailed modeling of the system. The downside
is the reliance on the input variables, which are generally
unknown. The accuracy of the solution depends heavily on
the input variables. In cases with statistical uncertainty, the
problem can be addressed using a statistical approach. In PLF
random variables (RV) are used as the input variables in the
load ow problem and probability theory methods are used to
nd the solution. [1]
PLF works as a tool to allow modeling of uncertainties in
generation and load. These uncertainties in the power system
are of particular interest today with the push for renewable
energy sources and dynamic loads. Two renewable sources
in particular, wind and solar, are very stochastic in nature.
Wind and solar both generate power based on surrounding
environmental factors and most wind turbines actually absorb
reactive power. Several different probability density functions
(PDF) have been proposed [2], [3] to represent their outputs
This manuscript has been authored by UT-Battelle, LLC, under Contract
No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United
States Government retains and the publisher, by accepting the article for
publication, acknowledges that the United States Government retains a non-
exclusive, paid-up, irrevocable, world-wide license to publish or reproduce
the published form of this manuscript, or allow others to do so, for United
States Government purposes.
and they show a wide range of uncertainty. Dynamic loads,
for demand response, can be placed strategically throughout
a system and the amount of load can be changed as needed
for system stability. Plug-in hybrid electric vehicles (PHEV)
are a mix of both generation and load. This makes them even
more challenging to model. PHEVs can also travel around
the system unpredictably and become an unexpected load or
possibly a distributed generator. There is very little data on the
actual performance and effects that the PHEVs will have on a
power system. By statistically representing the uncertainty of
the stochastic generation and load as RVs, they can be applied
as inputs in the system where PLF can give a valuable insight
[4][6].
The objective of this paper builds on the approach and
concepts presented in [1]. It extends a radial PLF to the general
case of three phase systems. This adds a level of complexity
to the original method but allows probabilistic analysis of
unbalanced systems, something that is of high importance in
power distribution systems.
II. THREE PHASE PLF DERIVATION
Some of the rst work in three phase PLF is presented in [7],
[8], where a linearized version of the Newton-Raphson method
is proposed for use in a meshed system. This paper proposes
a simpler approach that takes advantage of the topography of
a radial system. This allows the solution to be found without
explicitly using the Y-bus matrix.
The nonlinearity of the load ow equations is addressed by
linearizing around the expected operating points. The solution
of the PLF then becomes a sum of independent RVs weighted
by their sensitivity coefcients, (1), without complex relations
within the load ow equations.
f(x) =
n

i
1
c
i
f
i
_
y
i
c
i
_
(1)
where:
f
i
PDF of random variable y
i
(the i-th element)
c
i
sensitivity coefcient for y
i
The solution can then be obtained by convolution, (2), [9]
using a number of numerical techniques.
f(x) =
1
|c
1
|
f
1
_
y
1
c
1
_

1
|c
2
|
f
2
_
y
2
c
2
_

1
|c
n
|
f
n
_
y
n
c
n
_
(2)
The most common and efcient techniques are based on the
Fast Fourier Transform algorithm.
As mentioned before, the three phase PLF derivation pre-
sented here is an extension of the work in [1]. It follows
the same linearization techniques and uses most of the same
assumptions. The modied method can handle a variety of
load models, which can be either in a balanced or unbalanced
conguration. A more detailed derivation can be found in
Appendix A.
A. Powers
The load is dened as:
S

k
= P

nom
k
_
|V

k
0
|
Vnom
_
np
k
+ jQ

nom
k
_
|V

k
0
|
Vnom
_
nq
k
, (3)
where V
k0
is the voltage at node k obtained from the deter-
ministic load ow (DLF) solution performed with the mean
(expected) values of the input stochastic nodal powers, V
nom
is the systems rated voltage, and np
k
and nq
k
depend on
the corresponding load models (typically Z, I, PQ, or a
combination thereof).
P
k
and Q
k
are the real and imaginary parts of S
k
respec-
tively.
P
k
= R(S
k
) , (4)
Q
k
= I (S
k
) . (5)
If the power losses are neglected at the moment, the real and
reactive power ows in the element i j for a radial system
become:
P
ij
=

kj
P
k
, (6)
Q
ij
=

kj
Q
k
, (7)
where
j
denotes the set of all of the nodes supplied via node
j including node j.
If (6) is rearranged around the expected values,

P
k
, of the
input RVs then:
P
ij
=

kj

P
k
+

kj
_
P
k


P
k
_
. (8)
The rst term on the right hand side of the equation is
an approximation of P
ij
. When all of the inputs are at
their expected values, the approximation can be replaced with
P
ij0
. This is the value obtained from the DLF, where the
inputs are the respective expected values which implicitly takes
into account power losses. Equation (6) becomes:
P
ij
= P
ij0
+

kj
_
P
k


P
k
_
. (9)
The same can be applied for the reactive power:
Q
ij
= Q
ij0
+

kj
_
Q
k


Q
k
_
. (10)
The probabilistic three phase power ow through the ele-
ment i j is then:
S
ij
=
_
_
P
A
ij
+ jQ
A
ij
P
B
ij
+ jQ
B
ij
P
C
ij
+ jQ
C
ij
_
_
. (11)
B. Voltages
The voltage drop across the element i j is:
V
ij
= Z
ij

_
S
ij
V
j
_

, (12)
where:
Z
ij
=
_
_
Z
AA
ij
Z
AB
ij
Z
AC
ij
Z
BA
ij
Z
BB
ij
Z
BC
ij
Z
CA
ij
Z
CB
ij
Z
CC
ij
_
_
. (13)
Assuming the voltages in the distribution network operate
at a medium or low voltage level, two approximations can be
made for calculating (12):
Since the voltages in every node of the network do not
differentiate much from the the rated voltage, the nominal
voltage can be used instead of the actual voltage to
calculate the voltage drop.
Since the imaginary part of the voltage drop in any
element of the network comparing to the real one is much
smaller, it can be neglected. This is equivalent to saying
that the phase angle differences among the voltages at
different nodes are small.
Using the rst approximation, (12) becomes:
V
ij
Z
ij

1
V
nom

S
ij

, (14)
where:
1
V
nom
=
_
_
1
Vnom0
1
Vnom120
1
Vnom120
1
Vnom120
1
Vnom0
1
Vnom120
1
Vnom120
1
Vnom120
1
Vnom0
_
_
. (15)
Equation (14) is then expanded and the imaginary part is
ignored and for
A
it becomes:
V
A
ij

R
AA
ij
Vnom
P
A
ij
+
X
AA
ij
Vnom
Q
A
ij
+
R
AB
ij
+

3X
AB
ij
2Vnom
P
B
ij
+
X
AB
ij

3R
AB
ij
2Vnom
Q
B
ij
+
R
AC
ij

3X
AC
ij
2Vnom
P
C
ij
+
X
AC
ij
+

3R
AC
ij
2Vnom
Q
C
ij
.
(16)
The other two phases follow suit.
The total voltage drop from the substation (slack) to node
k, is the sum of all the voltage drops in the elements in the
supply path,
k
, starting from the slack up to node k. The
total voltage drop for node k is:
V
k
=

(ij)
k
V
ij
. (17)
Equation (16) contains the output RVs, P
ij
and Q
ij
,
which are not quite suitable for evaluating the resultant PDF
of the voltage for node k. These variables could be evaluated
with (9) and (10) rst; however, they are not just deterministic
values, but RVs dened with their respective PDFs which
need not be statistically dependent. This dependence is hard to
follow. Therefore, it is more appropriate to express the voltage
drops in terms of the input RVs. The real and reactive powers
in (16) are substituted with (6) and (7). The voltage drop for

A
at node k then becomes:
V
A
k

n

i=1
_
R
AA
ik
Vnom
P
A
i
+
X
AA
ik
Vnom
Q
A
i
+
R
AB
ik
+

3X
AB
ik
2Vnom
P
B
i
+
X
AB
ik

3R
AB
ik
2Vnom
Q
B
i
+
R
AC
ik

3X
AC
ik
2Vnom
P
C
i
+
X
AC
ik
+

3R
AC
ik
2Vnom
Q
C
i
_
,
(18)
where R
ik
and X
ik
are the total resistance and reactance of
the common elements found along the supplying paths of the
nodes i and k. As before, the other two phases follow suit.
Equation (18) can be linearized around the point of expected
values

P
i
,

Q
i
of the input RVs: P
i
and Q
i
, same as for the
power ows. The value of V
k0
, the point obtained from the
DLF solution is substituted in to complete the linearization of
the voltage drop. Equation (18) linearized and expanded out
for
A
becomes:
V
A
k
V
A
k0
+
n

i=1
_
R
AA
ik
V
nom

_
P
A
i


P
A
i
_
+
X
AA
ik
V
nom

_
Q
A
i


Q
A
i
_
+
R
AB
ik
+

3X
AB
ik
2V
nom

_
P
B
i


P
B
i
_
+
X
AB
ik

3R
AB
ik
2V
nom

_
Q
B
i


Q
B
i
_
+
R
AC
ik

3X
AC
ik
2V
nom

_
P
C
i


P
C
i
_
+
X
AC
ik
+

3R
AC
ik
2V
nom

_
Q
C
i


Q
C
i
_
_
.
(19)
The voltage for node k is then:
V
k
= V
0
V
k
, (20)
where V
0
denotes the current phases voltage at the feeding
point, in most cases the slack bus.
C. Power Losses
Complex power loss in the element i j is dened as:
S
ij
= P
ij
+ jQ
ij
= (R
ij
+ jX
ij
) |I
ij
|
2
= (R
ij
+ jX
ij
)
P
2
ij
+ Q
2
ij
|V
ij
|
2
,
(21)
which can then be converted to three phase:
S
ij
= Z
ij

_

_
P
A
2
ij
+Q
A
2
ij
|V
A
ij
|
2
P
B
2
ij
+Q
B
2
ij
|V
B
ij
|
2
P
C
2
ij
+Q
C
2
ij
|V
C
ij
|
2
_

_
, (22)
where Z
ij
is given with (13).
Again, (22) is not quite suitable for evaluating the resultant
PDFs. Similar to the voltage drop, the power ows (6) and
(7) are substituted into (22). Then, it is linearized around the
expected values of the input variables, and the deterministic
value P
ij0
is substituted in to dene the real and reactive
power loss as:
P
A
ij
P
A
ij
0
+
2R
AA
ij
Vnom
2


k
j

P
A
k


k
j
P
A
k
+

k
j

Q
A
k


k
j
Q
A
k

2R
AA
ij
Vnom
2


k
j

P
A
k

2
+


k
j

Q
A
k

+
2R
AB
ij
Vnom
2


k
j

P
B
k


k
j
P
B
k
+

k
j

Q
B
k


k
j
Q
B
k

2R
AB
ij
Vnom
2


k
j

P
B
k

2
+


k
j

Q
B
k

+
2R
AC
ij
Vnom
2


k
j

P
C
k


k
j
P
C
k
+

k
j

Q
C
k


k
j
Q
C
k

2R
AC
ij
Vnom
2


k
j

P
C
k

2
+


k
j

Q
C
k

,
(23)
Reactive power loss is the same as (23), except P

ij0
is
replaced with Q

ij0
and the resistance, R, is replaced with
the reactance, X.
The total real and reactive power loss in the network, P
tot
and Q
tot
, are the sum of the losses given by (23) and the
equivalent reactive power losses of all of the elements in the
network:
P
tot
=

ij
P
ij
, (24)
Q
tot
=

ij
Q
ij
, (25)
where denotes the set of all the elements in the network.
III. EXAMPLE
The described technique is applied to the IEEE 13 Node
Test Feeder [10]. The IEEE Test Feeder is chosen because it
is a standard test system, with a full solution, that is designed
for proposed load ow techniques to be evaluated on. The test
system also provides a variety of load types and variations. A
one line diagram for the test feeder is given in Fig. 1.
A modied version of the system is shown in Fig. 2. The
nodes have been renumbered, the voltage regulator has been
removed and the switch is considered closed. The voltage
regulator is removed from the system in this paper for sim-
plicity. Control devices in PLF will be the subject of a future
!"#
!$%
!&' !&! !"" !"&
!()
!*# !$' !(& !%%
!'#
!')
Fig. 1. Original IEEE 13 Node Test Feeder Conguration
!
"
# $ %# %$
%&
%% %! ' (
)
%
*
Fig. 2. Modied IEEE 13 Node Test Feeder Conguration
report. The slack bus voltages are then assigned as the voltage
regulators output voltages. An extra node has also been added
to factor in the distributed load between nodes 632 and 671.
The distributed load is modeled as a constant current load.
Following the exact load model [11], node 5 is placed a quarter
of the way between nodes 2 and 6. Two-thirds of the load is
then placed at the newly made node and the rest of the load
is added to end of the line at node 6.
For simplicity, all of the loads PDFs are modeled as a
normal distribution with the mean values, , corresponding to
the respective loads given in [10]. The standard deviation for
the loads is assumed to be 10% of the mean value. The loads
are represented by either a constant PQ, a constant current or
a constant impedance model. They can also be structured in
either a wye or a delta conguration.
For this example, all of the loads are independent, but the
phases for each of the loads are correlated, dependent. Tables
I - III show the results for the bus voltages and the real and
reactive power ows obtained from the proposed PLF. The
results are given in parameters of their respective PDFs, the
mean value, , and the standard deviation, .
Due to the linearization of the problem and because all
TABLE I
VOLTAGE RESULTS FROM THREE PHASE PROBABILISTIC LOAD FLOW
bus
A
V

A
V
%
B
V

B
V
%
C
V

C
V
%
1 1.0625 0.00 1.0500 0.00 1.0687 0.00
2 1.0210 0.24 1.0418 0.22 1.0176 0.31
3 1.0326 0.26 1.0156 0.32
4 1.0309 0.27 1.0135 0.33
5 1.0133 0.29 1.0440 0.25 1.0069 0.38
6 0.9899 0.47 1.0526 0.37 0.9780 0.61
7 0.9880 0.48 0.9760 0.62
8 0.9740 0.63
9 0.9824 0.51
10 0.9899 0.47 1.0526 0.37 0.9780 0.61
11 0.9899 0.47 1.0526 0.37 0.9780 0.61
12 0.9834 0.52 1.0550 0.37 0.9761 0.64
13 1.0180 0.25 1.0399 0.22 1.0150 0.31
14 0.9940 0.41 1.0215 0.32 0.9961 0.41
TABLE II
REAL POWER RESULTS FROM THREE PHASE PROBABILISTIC LOAD FLOW
bus
A
P

A
P
%
B
P

B
P
%
C
P

C
P
%
1 3.7548 5.21 2.9421 4.95 4.0440 4.08
2 3.6898 5.30 2.9507 4.94 3.9202 4.21
3 0.9951 7.07 0.2383 9.97
4 0.4843 10.00 0.2375 10.00
5 3.1940 5.94 1.5876 7.71 3.2899 4.84
6 3.1350 6.05 1.4641 8.26 2.9859 5.21
7 0.3730 9.93 0.4979 9.98
8 0.4968 10.00
9 0.3706 10.00
10
11 1.5931 9.17 0.2050 9.995 1.2484 7.59
12 1.4550 10.00 0.2040 10.00 0.8700 10.00
13 0.4876 9.85 0.3643 9.88 0.3645 9.88
14 0.4800 10.00 0.3600 10.00 0.3600 10.00
TABLE III
REACTIVE POWER RESULTS FROM THREE PHASE PROBABILISTIC LOAD
FLOW
bus
A
Q

A
Q
%
B
Q

B
Q
%
C
Q

C
Q
%
1 2.0453 4.93 1.1272 7.90 2.0078 5.46
2 1.8072 5.58 1.0297 8.65 1.7645 6.21
3 0.3772 9.94 0.4133 9.98
4 0.0016 10.00 0.4127 10.00
5 1.4129 6.75 0.3637 20.94 1.0325 9.47
6 1.2452 7.65 0.2747 27.07 0.8014 11.69
7 0.2502 9.95 -0.0497 47.04
8 -0.0509 45.97
9 0.2491 10.00
10
11 0.3643 18.63 -0.4859 3.70 0.1449 44.24
12 -0.0102 557.66 -0.4877 3.69 0.0644 98.79
13 0.3437 9.60 0.2778 9.72 0.2782 9.71
14 0.3300 10.00 0.2700 10.00 0.2700 10.00
of the loads are assumed to be normal, the resultant PDFs
will also follow a normal distribution. The mean values of the
input RVs are used as the input variables for the deterministic
solution. In turn, the resultant PDFs are linearized around
the deterministic solution points making the mean values of
the PLF results to be the deterministic solution points. As
explained earlier, this improves the accuracy of the solution
impacted by the linearization of the problem. In order to check
the solution a Monte Carlo simulation (MCS) is performed
with 100,000 experiments of an accurate DLF.
0.96 0.98 1 1.02 1.04 1.06
0
10
20
30
40
50
60
70
80
90
100
110
d
e
n
s
i
t
y
voltage (p.u)


Phase A ! Voltage Magnitude PLF
Phase B ! Voltage Magnitude PLF
Phase C ! Voltage Magnitude PLF
Phase A ! Voltage Magnitude MCS
Phase B ! Voltage Magnitude MCS
Phase C ! Voltage Magnitude MCS
Fig. 3. PDFs of all three phases voltage magnitude at node 12.
1.03 1.035 1.04 1.045 1.05 1.055 1.06 1.065 1.07 1.075 1.08
0
10
20
30
40
50
60
70
80
90
100
110
voltage (p.u)
d
e
n
s
i
t
y


Dependent Phases PLF
Dependent Phases MCS
Independent Phases PLF
Independent Phases MCS
Fig. 4. PDFs of phase B voltage magnitude at node 12.
The PDFs of all three phases voltages at node 12 are shown
in Fig. 3. At node 12, the drop in voltage is the summation
of all the voltage drops in the path from the slack to node
12. The PLF results for all three phases are shown along
with the results from the respective MCSs. There is a small
error between the PLF and the accurate MCS due to the
assumptions made in the formulation of the proposed PLF
method, (16). However, this error is insignicant as conrmed
by the matching of the resultant PDF shapes of the PLF and
MCS solutions, especially in the tail regions which are of the
most interest to the analyst.
The voltage magnitude of phase B of node 12 is shown in
more detail in Fig. 4. Here the voltage magnitude is plotted
for two cases when the phases, with each other, are correlated
and independent. The standard deviation for the PLF is smaller
than that of the MCS when the phases are dependent as before,
2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 4.4
0
0.5
1
1.5
2
2.5
3
power (p.u)
d
e
n
s
i
t
y


Phase A ! Real Power PLF
Phase B ! Real Power PLF
Phase C ! Real Power PLF
Phase A ! Real Power MCS
Phase B ! Real Power MCS
Phase C ! Real Power MCS
Fig. 5. PDFs of all three phases real power at branch 1-2.
0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
5
10
15
20
25
30
35
power (p.u)
d
e
n
s
i
t
y


Phase A ! Real Power Loss PLF
Phase B ! Real Power Loss PLF
Phase C ! Real Power Loss PLF
Phase A ! Real Power Loss MCS
Phase B ! Real Power Loss MCS
Phase C ! Real Power Loss MCS
Fig. 6. PDFs of total real power losses for all three phases.
but larger when the phases are independent of each other. This
difference occurs because the phases are to a certain extent
always negatively correlated with each other: I
B
+I
C
I
A
even for an unbalanced system. This is overridden if the phases
are assumed to be independent of each other.
Fig. 5 shows the real power ows for all three phases
through the rst branch of the system. Both the PLF and MCS
solution are again given for comparison. The branch between
node 1 and 2 is chosen because it is in the path of supply for all
of the other nodes in the system and is the most complicated
power calculation in the system. The dependence between the
phases within a branch does not effect the resultant power ow
PDFs and the results shown are valid for both cases. It can be
seen in Fig. 5 that the PLF solution closely matches that from
the MCS. The small error is introduced due to linearizations
in the power ow calculations. Furthermore, the real power
loss is relatively small. The reactive power loss is larger and
causes the PLF to have slightly larger inaccuracy compared
to the MCS. The inaccuracies in the losses add up for the
entire system, as evident from the total real power loss for the
system plotted in Fig. 6.
IV. CONCLUSION
The use of PLF in power system analysis gives qualitatively
more information about the system when compared to the
standard deterministic approach. This allows for the model-
ing of statistical uncertainties that are natural in the system
along with renewable generation and other technologies like
PHEVs that have stochastic characteristics. It is possible to
analyze these uncertainties and their effects more accurately
and efciently by being able to treat various system conditions
in a unied manner.
This paper proposes a modication of the PLF method rst
proposed in [1]. It makes the leap from the single phase
representation of a radial distribution network to a general
three phase representation. It is designed to work for a variety
of load models and congurations as well as unbalanced
systems.
APPENDIX A
GENERAL PROBABILISTIC LOAD FLOW FORMULATION
The load ow calculations can be represented by two sets
of nonlinear equations [1]:
Y = g [X] , (26)
Z = h[X] , (27)
where:
Y input random vector (real and reactive nodal power
injections);
X state random vector (voltage magnitudes and angles);
Z output state vector (power ows);
g, h load ow functions.
Once Y is specied, X must be evaluated in order to
determine Z. It is not possible to explicitly express X in
terms of Y . Therefore, (26) is linearized around specied
values Y
0
. Since the input variables are RVs and given in
terms of their respective PDFs, the most appropriate values to
linearize around are the expected values. So, Y
0
denotes the
expected value of Y and, X
0
and Z
0
are such so they satisfy
the following equations:
Y
0
= g [X
0
] , (28)
Z
0
= h[X
0
] . (29)
X
0
and Z
0
are only approximations for the EVs of X and
Z due to the nonlinear load ow functions.
Linearizing (26) and (27) around the points (Y
0
, X
0
) and
(Z
0
, X
0
), produces the following:
X X
0
+A Y = X

0
+A Y , (30)
Z Z
0
+B Y = Z

0
+B Y , (31)
where:
A =
_
g
X

x=x0
_
1
,
B =
_
h
X

x=x0
_
A,
Y = Y Y
0
, X

0
= X
0
A Y
0
, Z

0
= Z
0
B Y
0
.
Equations (30) and (31) express each element of X and Z
as a linear combination of the RVs of the input vector, Y .
These RVs are weighted by sensitivity coefcients, elements
of the matrices A and B. Therefore, if all the elements of Y
are satistically independent, the resultant RV of X or X, has
a PDF given by (1).
The solution can then be obtained by convolution, (2), [9]
using a number of numerical techniques. The most common
and efcient techniques are based on the Fast Fourier Trans-
form algorithm.
REFERENCES
[1] A. Dimitrovski and R. A ckovski, Probabilistic load ow in radial
distribution networks, in Transmission and Distribution Conference,
1996. Proceedings., 1996 IEEE, sep 1996, pp. 102 107.
[2] M. Mohanpurkar and R. Ramakumar, Probability density functions for
power output of wind electric conversion systems, in Power and Energy
Society General Meeting, 2010 IEEE, july 2010, pp. 1 7.
[3] X. Li, J. Pei, and S. Zhang, A probabilistic wind farm model for
probabilistic load ow calculation, in Power and Energy Engineering
Conference (APPEEC), 2010 Asia-Pacic, march 2010, pp. 1 4.
[4] L. Dong, W. Cheng, H. Bao, and Y. Yang, Probabilistic load ow
analysis for power system containing wind farms, in Power and Energy
Engineering Conference (APPEEC), 2010 Asia-Pacic, march 2010, pp.
1 4.
[5] F. Rui-xiang, W. Su-nong, W. Yan, and H. Min-xiao, The analysis
of distribution system with photovoltaic system generation based on
probabilistic power ow, in Power and Energy Engineering Conference
(APPEEC), 2011 Asia-Pacic, march 2011, pp. 1 5.
[6] P. Caramia, G. Carpinelli, M. Pagano, and P. Varilone, Probabilistic
three-phase load ow for unbalanced electrical distribution systems with
wind farms, Renewable Power Generation, IET, vol. 1, no. 2, pp. 115
122, june 2007.
[7] P. Caramia, G. Carpinelli, P. Varilone, and P. Verde, Probabilistic three-
phase load ow, International Journal of Electrical Power and Energy
Systems, vol. 21, no. 1, pp. 55 69, 1999.
[8] P. Caramia and P. Varilone, Probabilistic ac/dc 3-phase load ow, in
Harmonics And Quality of Power, 1998. Proceedings. 8th International
Conference on, vol. 2, oct 1998, pp. 1018 1028 vol.2.
[9] A. Papoulis, Probability, Random Variables and Stochastic Processes.
McGraw-Hill Inc.,New York, 1965.
[10] W. Kersting, Radial distribution test feeders, in Power Engineering
Society Winter Meeting, 2001. IEEE, vol. 2, 2001, pp. 908 912 vol.2.
[11] , Distribution System Modeling and Analysis, 2nd ed. CRC Press,
2007.

Das könnte Ihnen auch gefallen