Sei sulla pagina 1di 8

Electrical Power and Energy Systems 32 (2010) 551558

Contents lists available at ScienceDirect

Electrical Power and Energy Systems


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

Simplied NewtonRaphson power-ow solution method


Thanatchai Kulworawanichpong
Power System Research Unit, School of Electrical Engineering, Institute of Engineering, 111 University Avenue, Suranaree University of Technology,
Nakhon Ratchasima 30000, Thailand

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

Article history: This paper presents a simplied version of the well-known NewtonRaphson power-ow solution
Received 5 November 2007 method, which is based on the current balance principle to formulate a set of nonlinear equations.
Received in revised form 22 October 2009 Although there exist several powerful power ow solvers based on the standard NewtonRaphson
Accepted 6 November 2009
(NR) method, their corresponding problem formulation is not simple due to the need for calculation of
derivatives in their Jacobian matrix. The proposed method employs nonlinear current mismatch equa-
tions instead of the commonly-used power mismatches to simplify overall equation complexity. Deriva-
Keywords:
tion of Jacobian matrixs updating formulae is illustrated in comparison with those of the standard
Power-ow solution
NewtonRaphson iterative method
NewtonRaphson method. To demonstrate its use, a simple 3-bus power system was selected as a
Floating-point operation counting numerical example. The effectiveness of the proposed method was examined by computer simulations
Quadratic convergence through ve test systems: (1) 5-bus test system, (2) 6-bus test system, (3) 24-bus IEEE test system, (4)
30-bus IEEE test system and (5) 57-bus IEEE test system. Its convergence and calculation time were
observed carefully and compared with solutions obtained by the standard NR power ow method. The
results show that the proposed NR method spends less execution time than the standard method does
with similar convergence characteristics.
2009 Elsevier Ltd. All rights reserved.

1. Introduction Having a long history of development gives power ow algo-


rithms a vast number and a various kind of applications. Enhancing
The main function of electric power systems is to deliver elec- the algorithm efciency of power ow calculation has been carried
tric energy to its loads sufciently, efciently and economically. out in many different approaches. Network partition technique can
The steady-state performances of an interconnected power system separate a whole power system into subsystems, therefore power-
during normal operation can be analyzed based on nonlinear nodal ow solution of the complete system can be obtained by direct
analysis to form power ow equations and must be solved by some coupling of solutions from separate subsystems [15] based on the
efcient iterative methods [19]. Power ow analysis is commonly GS method. This concept is very useful to parallelize power ow
used as a part of power system operation and planning. Since AC algorithms in order to implement a parallel and sequential power
power-ow solution methods were rst developed over half a cen- ow performing on a computer cluster whether the GS method or
tury ago, there exist two widely-used numerical methods (the some other numerical methods such as successive over-relaxation
GaussSeidel: GS and the NewtonRaphson: NR) to solve this (SOR) method is used as the main solver [16,17]. In some point of
problem and therefore referred to as the GS and the NR power-ow view, an initial guest solution of power ow calculation is one of
solution methods, respectively. As broadly known, the NR method key factors that cause slow computation. In [18,19], an initial linear
has been successfully developed and accepted as the most power- solution based on the decoupling principle of real and reactive
ful algorithm for the power ow analysis in electric power systems. power decomposition was utilized as the starting point to the
In large-scale power systems containing several hundred or up to power ow calculation. In addition, there are some modied ver-
thousand buses, the standard NR method gives a slow execution sions of the NR power ow method to handle ill-conditioned
time due to a large updated Jacobian matrix that needs to be recal- power systems [20,21]. The calculation algorithm has been contin-
culated and factorized at each iteration [10,11]. Consequently, the ually developed by several researchers across the world. A complex
de-coupled and fast de-coupled power ow versions [12,13] were form of the power ow calculation was introduced for a three-
released. Hence, the power-ow solution can be obtained faster. phase unsymmetrical power-ow solution [22,23]. Power-ow
This method is very useful in practical power system analyses, solutions based on a local search method was claimed [24] to be
e.g. contingency analysis, on-line power ow control, etc. [4,14]. robust and be applicable to those cases in which conventional
power ow method failed. Due to advancement of FACTS technol-
E-mail address: thanatchai@gmail.com ogy, the power ow equations were modied and rewritten into

0142-0615/$ - see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijepes.2009.11.011
552 T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558

current-injected forms for the incorporation of FACTS devices and Substitute the above expressions into (2), thus
any kind of control strategy [25,26]. Moreover, the study of   Xn
Ssch;k 
power-ow solution methods for particular applications, e.g. eco- F k  \u dk 
 k jY ki V i j\hki di 0 3
nomic dispatch [27], optimal power ow [28], FACTS devices Vk i1
 
[29,30], and AC/DC power systems [31], was reported. Ssch;k  Xn

Over several decades, electrical power systems have been char- Gk   cosu dk 
 k jY ki V i j coshki di 0 4
Vk i1
acterized using the nodal analysis to solve for a set of voltage solu-  
Ssch;k  Xn
tions. In general, electrical demands are dened in constant power. Hk   sinu dk  jY ki V i j sinhki di 0 5
This leads to non-linearity of nodal voltage equations. To date, the Vk  k
i1
standard NR power ow method is one of the most powerful algo-
(4) and (5) are real and imaginary parts of the current mismatch at
rithms, which has long history of development, and is widely used
bus k. The mismatches will be zero when all unknown bus voltages
to develop commercial power-ow solution software. Although the
are successfully solved. To nd a set of voltage solutions by using
standard NR power ow method is very efcient and commonly
the NR method [34,35], these two equations must be expanded by
used for the power ow calculation in several power system text-
Taylor series as follows.
books [19], to formulate iterative Jacobian updating matrix equa-
tions requires complicated formulae and long expressions. In this X
n
@Gk X
n
@Gk
paper, the iterative NR method is still employed as the main solu- Gk Ddi DjV i j 6
@di @jV i j
tion framework. The essential difference is that the proposed algo- i1 i1
rithm is to nd roots of the current mismatch equations instead of is is
those of the power mismatch equations. This approach can sim- X
n Xn
@Hk @Hk
plify a very long and complicated mathematical formula to a very Hk Ddi DjV i j 7
@di @jV i j
simplistic and short mathematical expression. With this simplica- i1 i1
tion, reduction of the overall execution time is expected. To is is
achieve this goal, expressions to obtain elements of Jacobian
updating matrix formulae must be derived. where s denotes the slack bus.
This paper consists of six sections. Section 2 appeared next pre- With n  1 unknowns of complex variables and n  1 complex
sents the formulation of the proposed NR power ow problem. current mismatch equations, a compact matrix form used to up-
Derivation of the Jacobian updating matrix elements is included date the voltage solution can be expressed as follows.
in this section. Section 3 presents the oating-point operation   " @G #    
@G
counting to evaluate its computational effort. A simple 3-bus G @d @jVj Dd J1 J2 Dd
8
power system [1] was selected as a numerical example in Section H @H
@d
@H
@jVj
DjVj J3 J4 DjVj
4. Section 5 presents the simulation results based on 5-bus [2],
6-bus [4], 24-bus [32], 30-bus [1] and 57-bus [33] IEEE standard Hence, elements of the Jacobian matrices can be derived in the
test systems for benchmarking. The last section, Section 6, provides similar manner as those of the standard NR method and are sum-
conclusion of the work. marized in (9)(16).
Sub-matrix J1
2. Simplied NewtonRaphson power ow method @Gk
jV i Y ki j sinhki di for k i 9
@di
The power ow problem is a zero-nding problem to determine @Gk Ssch;k
voltage solutions of nonlinear power mismatch equations [1]. If jV k Y kk j sinhkk dk j j sinuk dk 10
@dk Vk
alternative nonlinear current mismatch equations are selected
and used as functions of estimating roots. Given that an n-bus Sub-matrix J2
power system, which bus number 1 is assigned to be a slack bus @Gk
of constant voltage magnitude and zero phase angle. Considering jY ki j coshki di for k i 11
@jV i j
the kth bus, current balance equations characterizing this bus  
@Gk S 
can be expressed as follows.  sch;k 
jY kk j coshkk dk  2  cosuk dk 12
@jV k j  V  k
X
n
Igen;k  Idem;k  Y ki V i 0 1 Sub-matrix J3
i1
@Hk
jV i Y ki j coshki di for k i 13
where Igen,k denotes generator current at bus k, Idem,k denotes load @di
 
current at bus k, Vk denotes phasor voltage at bus k, Yki denotes @Hk Ssch;k 
jV k Y kk j coshkk dk    cosu dk 14
the kth-row and ith-column element of the system bus admittance @dk Vk  k
matrix.In practice, loads in electrical power systems are in form of
powers, therefore it is convenient to rewrite (1) into a function of Sub-matrix J4
powers as follows. @Hk
jY ki j sinhki di for k i 15
  @jV i j
Sgen;k  Sdem;k  X n  
Fk  Y ki V i 0 2 @Hk S 
Vk  sch;k 
i1
jY kk j sinhkk dk  2  sinuk dk 16
@jV k j  V  k

Dene Fk = Gk + jHk be the current mismatch at bus k,


In the NR method, (8) is iteratively solved for Dd and D|V|. If a spec-
V k jV k j\dk ied norm of the current mismatches G and H is smaller than max-
Y ki jY ki j\hki is imum mismatch allowance, the voltage solution is successfully
obtained. Otherwise, the current voltage solution at iteration h
Sgen;k  Sdem;k Ssch;k jSsch;k j\uk
must be updated for the next iteration h + 1, as shown in (17).
T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558 553

 h1  h  h X
n
d d Dd @P k
17 jV k V i Y kk j sinhki di  dk
jVj jVj DjVj @dk 20
i1
To compare the effectiveness of the proposed NR method against ik
the standard NR method, expressions of the Jacobian matrix ele-
ments of J1, J2, J3 and J4, the calculated real and imaginary current Each diagonal element of J1 for either the standard or the proposed
matrix elements of G and H, and the calculated real and reactive NR method can be determined its operation counting. (10) requires
power matrix elements of Pcal and Qcal need to be evaluated as de- three multiplications and one division. For convenience, it is equiv-
scribed in the next section. alent to four multiplications. Whereas, (20) requires 3  (n  1)
multiplications. In the similar manner, the amount of FLOPs re-
3. Evaluation of oating-point operation counting quired by each method to formulate Jacobian matrices is summa-
rized in Table 1.
The execution time of the power ow calculation depends on As a total number of buses n gets larger, the number of FLOPs
the amount of oating-point operations (FLOPs) [34,35]. Assume grows quadratically in the standard NR method. Interestingly, the
that other steps of the two NR methods are exactly the same, FLOP number required by the proposed NR method is linearly pro-
therefore the Jacobian updating step dominates the overall execu- portional to the total number of buses n. Fig. 1 shows the amount
tion time. In general, the time consumed to perform multiplication of FLOPs required by the two methods.
and division is about the same, but is larger than addition and At each iteration, the real and imaginary current vectors G and
subtraction. Hence, the operation counting of addition FLOPs is H are updated using (4) and (5), respectively, to evaluate the cur-
negligible. Throughout this paper, FLOPs always means the multi- rent mismatches. In the standard NR method, the real and reactive
plication FLOPs for short and it is employed to evaluate the compu- power vector Pcal and Qcal are also recalculated iteratively. (21) and
tational effort of the proposed algorithm. (22) are the expression to perform these calculated powers.
As commonly written in most power system analysis textbooks
[19], the Jacobian updating matrix formula for the standard NR 9
10
method is described in (18).
Standard NR
  " @P @P
#     8
Proposed NR
P @d @jVj Dd J1 J2 Dd 10
@Q @Q
18
Q @d @jVj
DjVj J3 J4 DjVj 7
10
Recall (9), it represents the updating formula for off-diagonal 6
elements of J1 in the proposed NR method, while (19) describes 10
the updating formula for off-diagonal elements of J1 in the stan-
FLOPs

5
dard NR method [1] for comparison. 10

4
@Pk 10
jV k V i Y ki j sinhki di  dk for k i 19
@di
3
10
Each off-diagonal element of J1 for either the standard or the
proposed NR methods can be expressed in a single term. (9) 2
10
requires two multiplications, while three multiplications are
needed to perform (19). There are (n  1)  (n  2) off-diagonal ele- 1
10
ments in J1 sub-matrix. Thus, the standard NR method requires 10
0
10
1
10
2
10
3
10
4

3  (n  1)  (n  2) FLOPs per one iteration or 3n2 + O(n). O(n) Total number of buses
means terms of order n [34]. Remarkably, the proposed NR meth-
Fig. 1. Number of FLOPs per iteration to update the Jacobian matrix.
od requires 2  (n  2) FLOPs.
For the diagonal elements of J1, (10) is the expression used for
the proposed NR method, while (20) is used for the standard NR
method [1]. 10
5

Standard NR
Proposed NR
Table 1 4
Number of FLOPs. 10

Sub-matrix Number of FLOPs


Standard NR Proposed NR 3
10
J1 Diagonal 3  (n  1) 4  (n  2)
FLOPs

Off-diagonal 3n2 + O(n) 2  (n  2)


Total 3n2 + O(n) 6  (n  2) 2
10
J2 Diagonal 2  (n  1) + 3 4  (n  2)
Off-diagonal 2n2 + O(n) (n  2)
Total 2n2 + O(n) 5  (n  2)
1
J3 Diagonal 3  (n  1) 4  (n  2) 10
Off-diagonal 3n2 + O(n) 2  (n  2)
Total 3n2 + O(n) 6  (n  2)
0
J4 Diagonal 2  (n  1) + 3 4  (n  2) 10 0 1 2 3 4
Off-diagonal 2n2 + O(n) (n  2) 10 10 10 10 10
Total 2n2 + O(n) 5  (n  2) Total number of buses
Overall 10n2 + O(n) 22  (n  2)
Fig. 2. Number of FLOPs per iteration to update the mismatch vectors.
554 T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558

X
n
Pcal;k jV k V i Y ki j coshki di  dk 21
i1
X
n
Q cal;k  jV k V i Y ki j sinhki di  dk 22
i1

(4) and (5) require 4  n + 4 FLOPs per iteration for both. In contrast,
formulating Pcal and Qcal costs 6  n FLOPs. The difference can be
represented graphically in Fig. 2.
As can be seen, the proposed NR method has required a smaller
number of FLOPs than the standard NR method when the system Fig. 3. Single-line diagram of a simple 3-bus power system [1].
size is increased. This can guarantee that shorter calculation time
to re-calculate all Jacobian sub-matrices plus all mismatch matri-
ces of the proposed NR method can be expected if the convergence The slack bus voltage V1 is 1.050 p.u., starting with initial guess
rate (number of iteration used) of both methods is the same. The voltages V2 = 1.000 p.u. and V3 = 1.040 p.u., thus
next two sections will give detail to verify this assumption. jV 3 V 1 Y 31 j sinh31 d1  d3
Q cal;3 jV 3 V 2 Y 32 j sinh32 d2  d3
4. Numerical example
jV 23 Y 33 j sinh33 1:0192 p:u:
The power-ow solution by the proposed NR method is demon- Ssch;3 2:0 j1:0192 p:u:
strated in a simple 3-bus power system as shown in Fig. 3. The ob- The current mismatch of the initial state can be obtained as follows.
tained result can be compared with the solution given in example
6.10 in [1].
The bus admittance matrix can be described by, Compute the Jacobian matrix with the initial guess voltage solu-
2    3
tion, the updating equation of the rst iteration becomes,
53:8517\  68:20 22:3607\116:57 31:6228\108:43
6
Y bus 4 22:3607\116:57
   7 2 3 2 30 32
58:1378\  63:43 35:7771\116:57 5 2:8600 54:50 33:28 Dd2 22:00
  
31:6228\108:43 35:7771\116:57 67:2310\  67:25 6 7 6 76 7
4 1:3831 5 4 32:00 63:50 16:00 54 Dd3 5
From (4) and (5), the expressions for real current mismatches at 0:2200 30:00 16:64 49:50 DjV 2 j
bus 2 and 3, and imaginary current mismatch at bus 2 are Thus, the updated bus voltages for the rst iteration are:
  X3 2 30 2 3 2 31 2 30 2 30
Ssch;2  Dd 2 0:0460 d2 d2 Dd2
G2   cosu d2 
 jY 2i V i j cosh2i di
V2 2 6 7 6 7 6 7 6 7 6 7
i1 4 Dd3 5 4 0:0088 5 ! 4 d3 5 4 d3 5 4 Dd3 5
  X3
Ssch;3  DjV 2 j 0:0294 jV 2 j jV 2 j DjV 2 j
G3   cosu d3 
 3 jY 3i V i j cosh3i di 2 31 2 3 2 3 2 3
V3 i1 d2 0:0 0:046 0:046 rad
  X 6 7 6 7 6 7 6 7
Ssch;2  3 4 d3 5 4 0:0 5 4 0:0088 5 4 0:0088 rad 5
H2   sinu d2 
 2 jY 2i V i j sinh2i di
V2 jV 2 j 1:00 0:0294 0:9706 p:u:
i1

Elements of the Jacobian matrix are obtained by the following For the second iteration, the updating equation becomes,
equations. 2 3 2 31 32
  0:0408 54:3425 33:4251
Dd2 19:4624
@G2 Ssch;2  6 7 6 76 7
jV 2 Y 22 j sinh22 d2   sinu d2 4 0:0241 5 4 31:7416 63:2414 14:5117 54 Dd3 5
@d2 V2  2
0:0826 26:8874 16:3466 50:2945 DjV 2 j
@G2
jV 3 Y 23 j sinh23 d3 and
@d3
  2 31 2 3 2 32 2 31 2 31
@G2 S  Dd 2 0:0011 d2 d2 Dd 2
 sch;2 
jY 22 j cosh22 d2  2  cosu2 d2 6 7 6 7 6 7 6 7 6 7
@jV 2 j  V2  4 Dd3 5 4 0:0001 5 ! 4 d3 5 4 d3 5 4 Dd3 5
@G3 DjV 2 j 0:0010 jV 2 j jV 2 j DjV 2 j
jV 2 Y 32 j sinh32 d2 2 32 2 3 2 3 2 3
@d2 d2 0:046 0:0011 0:0471 rad
 
@G3 Ssch;3  6 7 6 7 6 7 6 7
jV 3 Y 33 j sinh33 d3   sinu d3 4 d3 5 4 0:0088 5 4 0:0001 5 4 0:0087 rad 5
@d3 V3  3
jV 2 j 0:9706 0:0010 0:9716 p:u:
@G3
jY 32 j cosh32 d2
@jV 2 j For the third iteration, the updating equation becomes,
  2 3 2 32 32
@H2 Ssch;2 
jV 2 Y 22 j cosh22 d2    cosu d2 6:71  105 54:4238 33:4238 Dd 2 19:4172
@d 2 V  2
2
6 7 6 76 7
4 4:03  104 5 4 31:7908 63:2908 14:4766 54 Dd3 5
@H2 6
jV 3 Y 23 j cosh23 d3 6:3  10 26:8492 16:3493 50:3215 DjV 2 j
@d3
@H2
jY 23 j sinh23 d3 therefore,
@jV 3 j 2 32 2 3 2 33 2 32 2 32
With 100 MVA base, the scheduled power in per-unit can be
Dd 2 7:23  106 d2 d2 Dd2
6 7 6 7 6 7 6 7 6 7
written as, 4 Dd3 5 4 1:02  105 5 ! 4 d3 5 4 d3 5 4 Dd3 5
7
DjV 2 j 6:80  10 jV 2 j jV 2 j DjV 2 j
Ssch;2 4:0  j2:5 p:u:
T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558 555

Table 2 Table 3
Power ow solution for the example. Power ow solutions of the modied test systems.

Item Standard NR [1] Proposed NR Bus TC1 TC2 TC3 TC4 TC5
V2 (p.u.) 0.97168\2.696 0.97168\2.698 1 1.06 + j0.00 1.05 + j0.00 1.05 + j0.00 1.06 + j0.00 1.07 + j0.00
V3 (p.u.) 1.04\0.4988 1.04\0.4979 2 0.92  j0.09 0.87  j0.62 0.46  j0.91 1.04  j0.12 1.04  j0.03
Q3 (p.u.) 1.4617 1.4618 3 0.82  j0.13 0.93  j0.50 0.67  j0.79 1.02  j0.31 1.06  j0.09
Max. mismatch 2.2  104 p.u. power 4.0  104 p.u. current 4 0.81  j0.14 0.79  j0.62 0.52  j0.88 0.97  j0.27 1.03  j0.12
5 0.78  j0.17 0.82  j0.51 0.70  j0.74 0.98  j0.23 1.05  j0.04
6 0.88  j0.42 0.79  j0.44 1.00  j0.22 1.05  j0.13
7 0.58  j0.77 0.98  j0.25 1.03  j0.13
1
8 0.52  j0.87 0.90  j0.30 0.96  j0.16
10 9 0.46  j0.90 0.98  j0.28 0.98  j0.15
Standard NR 10 0.73  j0.55 0.96  j0.31 1.02  j0.15
Proposed NR 11 0.62  j0.71 0.92  j0.30 1.03  j0.13
0
10 12 0.46  j0.81 0.97  j0.30 1.01  j0.17
13 0.46  j0.76 0.93  j0.31 1.03  j0.12
Maximum Mismatch

-1 14 0.57  j0.77 0.95  j0.31 1.03  j0.10


10 15 0.52  j0.83 0.94  j0.31 1.05  j0.06
16 0.52  j0.84 0.96  j0.30 1.03  j0.12
-2 17 0.52  j0.86 0.95  j0.31 1.05  j0.08
10 18 0.52  j0.87 0.93  j0.32 1.00  j0.17
19 0.47  j0.87 0.93  j0.32 0.97  j0.18
-3 20 0.89  j0.27 0.94  j0.32 0.97  j0.18
10 21 0.73  j0.45 0.94  j0.31 0.97  j0.16
22 0.72  j0.6 0.94  j0.31 0.97  j0.16
-4 23 1.00  j0.08 0.93  j0.31 0.97  j0.16
10
0 1 2 3 4 24 0.60  j0.75 1.01  j0.19 0.95  j0.16
Iteration 25 1.04  j0.30 0.89  j0.23
26 0.93  j0.29 0.96  j0.16
Fig. 4. Solution convergence of the standard and the proposed NR methods. 27 0.99  j0.23 0.98  j0.15
28 1.02  j0.16 1.00  j0.14
29 0.90  j0.31 1.01  j0.13
2 32 2 3 2 3 30 0.88  j0.32 0.87  j0.24
d2 0:0471 7:23  106 31 0.85  j0.24
6 7 6 7 6 7
4 d3 5 4 0:0087 5 4 1:02  105 5 32 0.88  j0.24
jV 2 j 0:9716 7 33 0.88  j0.24
6:80  10
2 3 2  3 34 0.94  j0.16
0:04709 rad 2:698 35 0.95  j0.16
6 7 6  7 36 0.96  j0.16
4 0:00869 rad 5 4 0:4979 5
37 0.97  j0.16
0:97168 p:u: 0:97168 p:u: 38 0.98  j0.15
39 0.96  j0.16
To assess the effectiveness of the proposed NR method, comparison 40 0.97  j0.15
41 0.98  j0.20
with the third-iteration bus voltages obtained by the standard NR
42 0.94  j0.21
method is given in Table 2. Fig. 4 gives solution convergence of both 43 1.02  j0.15
methods. 44 0.99  j0.14
The numerical example shows that the standard NR and the 45 1.03  j0.10
proposed NR methods show similar solution convergence for the 46 1.01  j0.13
47 0.99  j0.15
simple 3-bus test system. Some might say that this can imply the
48 0.99  j0.15
quadratic convergence of the simplied NR method as well as 49 0.99  j0.16
the standard NR method does. As can be seen, there is no signi- 50 0.97  j0.17
cant difference between their convergences, the proposed NR 51 1.03  j0.09
52 0.99  j0.16
method is expected to perform faster in obtaining the power-ow
53 1.03  j0.12
solution due to its simplication. 54 1.04  j0.11
55 1.03  j0.15
56 0.93  j0.22
5. Simulation results 57 0.93  j0.22

The effectiveness of the simplied NewtonRaphson power


ow method was tested against 5-bus [2], 6-bus [4], 24-bus [32],
30-bus [1] and 57-bus [33] IEEE test systems. Table 3 shows
Table 4
power-ow solutions of each modied test system. Each individual Specication of three processors used to perform the tests.
test was performed by using three different processors (Pentium,
Machine ID Machine specication
AMD and Duron processors) as shown in Table 4 in which the
power ow programs were coded in MATLAB [35]. MC1 Intel pentium IV 1.5 GHz, 256 DDR-RAM
MC2 AMD Athlon (TM) XP 2000plus 1.67 GHz, 256 DDR-RAM
From the computer simulation, the voltage solution of each test
MC3 Duron 1.2 GHz, 384 DDR-RAM
case was calculated. Both NR power ow methods used here took
1  106 per-unit as the termination criteria for the maximum
allowable voltage tolerance. Table 5 is the summary of the effec-
tiveness of the proposed method by giving the required iteration Fig. 5 shows the calculation time ratio for the four test cases.
and calculation time in comparison with those of the standard From the gure, the PNR method spends shorter calculation times
NR method. It notes that, in Table 5, SNR and PNR denote the stan- for all test cases. TC1 and TC5 show that the PNR method requires
dard NR method and the proposed NR method, respectively. four and three iterations, respectively, while the SNR method uses
556 T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558

Table 5
Simulation results of all test cases.

Case Required iteration Calculation time (s) Calculation time ratio


MC1 MC2 MC3 MC1 MC2 MC3 MC1 MC2 MC3
TC1 SNR 5 5 5 0.025 0.016 0.016 1.563 1.600 1.455
PNR 4 4 4 0.016 0.010 0.011 Average = 1.539
TC2 SNR 5 5 5 0.047 0.038 0.042 1.516 1.462 1.400
PNR 5 5 5 0.031 0.026 0.030 Average = 1.459
TC3 SNR 7 7 7 0.616 0.305 0.359 1.124 1.151 1.122
PNR 8 8 8 0.548 0.265 0.320 Average = 1.132
TC4 SNR 5 5 5 0.711 0.338 0.410 1.344 1.325 1.318
PNR 5 5 5 0.529 0.255 0.311 Average = 1.329
TC5 SNR 4 4 4 2.438 1.856 1.991 1.794 1.681 1.709
PNR 3 3 3 1.359 1.104 1.165 Average = 1.728

Fig. 5. Comparison of the calculation time between the two methods for all test cases.

ve and four iterations, respectively. Undoubtedly, the PNR meth- the voltage solution for the PNR method and the SNR method are
od is faster for these two test cases. It is different for the second eight and seven iterations, respectively. In this case, the PNR meth-
and the forth test cases. Both methods spends equally the same re- od still consumes less computational effort than the SNR method
quired iterations (ve iterations) to obtain the voltage solutions. requires. All these strongly conrm the high efciency of the PNR
Since the PNR method takes less requirement of re-calculation in method for obtaining the power-ow solutions. In addition, to de-
its Jacobian matrix per iteration, the calculation time ratios for scribe the convergence property of each test case, Figs. 610 illus-
these two test cases are remarkably larger with a factor of 1.459 trate the maximum voltage error at each iteration during the
and 1.329, respectively. For TC3, the required iterations to reach power-ow solution process for all test cases.

0
10 0
10
SNR
SNR
-1 PNR -1
10 10 PNR

-2 -2
10 10
Maximum voltage error (p.u.)

-3
Maximum voltage error (p.u.)

-3 10
10
-4
-4 10
10
-5
10
-5
10
-6
10
-6
10 -7
10
-7
10 -8
10
-8
-9
10 10

-9 -10
10 10
1 1.5 2 2.5 3 3.5 4 4.5 5 1 1.5 2 2.5 3 3.5 4 4.5 5
Iteration Iteration

Fig. 6. Solution convergence of TC1. Fig. 7. Solution convergence of TC2.


T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558 557

10
-1
6. Conclusions
SNR
-2 PNR
10 Power ow calculation is one of the most essential parts in elec-
-3 tric power system operation in order to analyze, simulate, design
10
and control the steady-state system performances properly.
-4
Although there exist several powerful power ow solvers based
Maximum voltage error (p.u.)

10
on the standard NR method, their problem formulation gives com-
-5
10 plication due to the need to calculate derivatives in the Jacobian
-6 matrix. The proposed method uses nonlinear current mismatch
10
equations instead of the commonly-used power mismatches to
10
-7
simplify overall equation complexity. With performance evalua-
tion found in Section 3, a total number of operations required by
-8
10 the proposed NR method is linearly proportional to the size of
-9 the Jacobian matrix, while that of the standard NR method is qua-
10
dratic. This means that the calculation time of the standard NR
10
-10
method increases more rapidly as a total bus number increases
than that of the proposed NR method does. From this advantage,
-11
10 the calculation time consumed by the proposed NR method is ex-
1 2 3 4 5 6 7 8
Iteration pected to be less than that of the standard one. This can leads to
improvement of power-ow software development in fast compu-
Fig. 8. Solution convergence of TC3. tational speed and less memory usage.

10
0 References
SNR
-1 PNR [1] Saadat H. Power system analysis. McGraw-Hill; 2004.
10 [2] Stagg GW, El-Abiad AH. Computer methods in power system
analysis. McGraw-Hill; 1968.
-2
10 [3] Grainger JJ, Grainger JJ, Stevenson WD. Power system analysis. McGraw-Hill;
1994.
[4] Wood AJ, Wollenberg BF. Power generation, operation, and control. John Wiley
Maximum voltage error (p.u.)

-3
10
& Sons; 1996.
[5] Bergen AR, Vittal V. Power systems analysis. Prentice-Hall; 2000.
-4
10 [6] Glover JD, Sarma M. Power system analysis and design. PWS Publishing; 1994.
[7] Kothari DP, Nagrath IJ. Modern power system analysis. McGraw-Hill; 2004.
-5 [8] Natarajan R. Computer-aided power system analysis. Marcel Dekker; 2002.
10
[9] Weedy BM, Cory BJ. Electric power systems. Johns Wiley & Sons; 1999.
-6
[10] Sambarapu KM, Halpin SM. Sparse matrix techniques in power systems. In:
10 The 39th southeastern symposium on system theory, Macon, Georgia; 46
March 2007. p. 2126.
-7 [11] Chan KW. Parallel algorithms for direct solution of large sparse power system
10
matrix equation. IEE Proc Gener Transm Distr 2001;148:61522.
-8 [12] Stott B. Decoupled Newton load ow. IEEE Trans Power Apparatus Syst
10 1972;91:19559.
[13] Stott B. Fast decoupled load ow. IEEE Trans Power Apparatus Syst
-9
10 1974;93:85969.
1 1.5 2 2.5 3 3.5 4 4.5 5 [14] Mori H, Tanaka H, Kanno J. A preconditioned fast decoupled power ow
Iteration method for contingency screening. IEEE Trans Power Syst 1996;11:35763.
[15] Carre BA. Solution of load ow problems by partitioning systems into trees.
Fig. 9. Solution convergence of TC4. IEEE Trans Power Apparatus Syst 1968;87:19318.
[16] Huang G, Ongsakul W. Managing the bottlenecks in parallel GaussSeidel type
algorithms for power ow analysis. IEEE Trans Power Syst 1994;9:67784.
-1
10 [17] Huang G, Ongsakul W. An adaptive SOR algorithm and its parallel
SNR implementation for power system applications. In: The 6th IEEE symposium
10
-2 PNR on parallel and distributed processing, Dallas, USA; 2629 October 1994. p.
8491.
-3 [18] Leoniopoulos G. Efcient starting point of load-ow equations. Int J Electric
10
Power Energy Syst 1994;16:41922.
-4 [19] Hubbi W, Refsum A. Starting algorithm and modication for NewtonRaphson
10
Maximum voltage error (p.u.)

load-ow method. Int J Electric Power Energy Syst 1983;5:16672.


-5
[20] Prasad GD, Jana AK, Tripathy SC. Modications to NewtonRaphson load ow
10 for ill-conditioned power systems. Int J Electric Power Energy Syst
1990;12:1926.
-6
10 [21] Slochanal SMR, Mohanram KR. A novel approach to large scale system load
ows NewtonRaphson method using hybrid bus. Electric Power Syst Res
10
-7 1997;41:21923.
[22] Nguyen HL. NewtonRaphson method in complex form [power system load
-8 ow analysis]. IEEE Trans Power Syst 1997;12:13559.
10
[23] Strezoski VC, Trpezanovski LD. Three-phase asymmetrical load-ow. Int J
-9 Electric Power Energy Syst 2000;22:51120.
10 [24] Acharjee A, Goswami SK. Robust load ow based on local search. Expert Syst
Appl 2008;35:14007.
-10
10 [25] Vinkovic A, Mihalic R. A current-based model of an IPFC for NewtonRaphson
power ow. Electric Power Syst Res 2009;79:124754.
-11
10 [26] da Costa VM, Pereira JLR, Martins N. An augmented NewtonRaphson power
1 1.5 2 2.5 3 3.5 4 ow formulation based on current injections. Int J Electric Power Energy Syst
Iteration 2001;23:30512.
[27] Chen S, Chen J. A direct NewtonRaphson economic emission dispatch. Int J
Fig. 10. Solution convergence of TC5. Electric Power Energy Syst 2003;25:4117.
558 T. Kulworawanichpong / Electrical Power and Energy Systems 32 (2010) 551558

[28] Hur D, Park J, Kim BH. On the convergence rate improvement of mathematical [31] Tylavski DJ, Trutt FC. The NewtonRaphson load ow applied to AC/DC
decomposition technique on distributed optimal power ow. Int J Electric systems with commutation impedance. IEEE Trans Ind Appl 1983;19:
Power Energy Syst 2003;25:319. 9408.
[29] Fuerte-Esquivel CR, Acha E. NewtonRaphson algorithm for the reliable [32] El-Hawary ME. Electric power applications of fuzzy systems. IEEE Press; 1998.
solution of large power networks with embedded FACTS devices. IEE Proc [33] Wan HB, Song YH. Hybrid supervised and unsupervised neural network
Gener Transm Distr 1996;143:44754. approach to voltage stability analysis. Electric Power Syst Res
[30] Wanliang F, Ngan HW. Extension of Newton Raphson load ow techniques to 1998;47:11522.
cover multi unied power ow controllers. In: The 4th international [34] Chapra SC, Canale RP. Numerical methods for engineers with software and
conference on advances in power system control, operation and programming applications. McGraw-Hill; 2003.
management (APSCOM-97); 1114 November 1997. p. 3838. [35] Fausett LV. Applied numerical analysis using MATLAB. Prentice-Hall; 1999.

Potrebbero piacerti anche