Documente Academic
Documente Profesional
Documente Cultură
Received 31 August 2004; received in revised form 25 November 2004; accepted 24 January 2005
Available online 7 April 2005
Abstract
This paper applies the concept of distributed processing to the problem of measurement placement for power system state estimation. The
proposed method uses the minimum condition number of the measurement matrix as a criterion in conjunction with sequential elimination to
reach the near optimal placement positions. Firstly, the entire network of the power system is decomposed into smaller sub-networks. Then,
in each sub-network, the optimal positions for measurement placement are determined by using the minimum condition number criteria. A
heuristic algorithm for reducing the number of placement sites is also presented in order to minimize the communication costs. The numerical
experimental results on the IEEE 14, 30, and 118 bus systems indicate that the proposed technique will provide a measurement matrix with
smaller condition number and the computation time is much shorter than the non-decomposition approach.
2005 Elsevier B.V. All rights reserved.
Keywords: Measurement placement; Phasor measurement units; Decomposition technique; State estimation
0378-7796/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2005.01.002
42 C. Rakpenthai et al. / Electric Power Systems Research 75 (2005) 4149
center (or individual company). The information of each sub- currents, and currents injected into the busbar to be deter-
network may be sent to the central control center that controls mined. The measured values are the busbar voltage phasors,
these sub-networks [5]. An algorithm based on combining the injection current phasors, and the line current phasors.
the recursive quadratic programming with the dual method Based on these measured values, all possible relations of the
[6] and the parallel processing algorithm [7] is developed to estimated state variable vector v and the measurement vector
solve WLS problem. The distributed state estimator based on z in the presence of measurement error can be expressed as,
reduced model is presented in [8]. The distributed state esti-
mator based on Lagrange theory to solve large power system zV [I] [0] V
SE problem is presented in [9]. These papers were intended to vM
z =
I IM ]
[Y [Y IC ] + I (4)
show the improved robustness and computational efficiency vC
of the distributed state estimator. However, these papers do zL [Y LM ] [Y LC ] L
not address the problem of where to install the measuring de-
vices. In this paper, the idea of decomposing the large power where [I] is an identity matrix, zV , zI , and zL are the mea-
system into small sub-networks is employed to help finding sured values of the busbar voltage, the injection current, and
a measurement placement set of a large power system SE the line current sub-vectors, respectively, [YIM ] and [YIC ] are
problem. By properly assigning the appropriate measuring the bus admittance sub-matrix of the measured and the cal-
devices, the measurement matrix of each sub-network is de- culated busbar voltage related to zI , respectively, [YLM ] and
coupled. Hence, the placement algorithm can be applied to [YLC ] are the bus admittance sub-matrix of the measured and
each sub-network independently. the calculated busbar voltage related to zL , respectively, vM
This paper is organized as follows. Firstly, a WLS state and vC the measured and the calculated busbar voltage sub-
estimation problem is briefly explained. Next, the proposed vectors, respectively, and V , I , and L are the corresponding
measurement placement algorithm is described. The numeri- measurement noise sub-vectors.
cal experimental of the proposed algorithm is applied to find Eq. (4) indicates all possible relationships between the
the positions of measurement placement in three IEEE test measured values and the estimated state variables. In practi-
systems. The performances of the proposed algorithm are cal, the measuring devices will not be placed at all possible
illustrated and compared with the measurement placement positions because of economical reasons. Thus, it is neces-
method based on minimum condition number [10]. Finally, sary to find an appropriate measurement placement position
concluding comments are given. set such that all state variables under considerate conditions
can still be estimated with minimum number of measuring
devices.
2. Power system state estimation
Fig. 1. Decomposition of a network into four non-overlapping sub- Practically, the measuring devices will be installed at the
networks. buses representing power system substations. The measured
data are exchanged and synchronized to the central control
written as, center through GPS satellite or optical fiber channels. This
process requires extensive communication equipment. Thus,
z1 [H 1 ] [0] v1 1 it is important to minimize the number of placement sites in
z v
2 [H 2 ] 2 2 order to reduce the communication costs. It is noticed that
= + (5)
.. .. .. .. although the line current measuring device installed close to
. . . . one site is moved to the other site of the branch, the condition
zr [0] [H r ] vr r number of the measurement matrix did not change. However,
the bus voltage and the injection current measuring devices
where zi , vi , and Hi denote the measurement vector, the state cannot be moved from its original position. Thus, the mea-
variable vector, and the measurement matrix of the ith sub- surement placement obtained from the minimum condition
network, respectively. i represents the measurement error number criteria can be rearranged to minimize the number of
vector related to zi . These zi , vi , Hi , and i are formulated placement sites by the following heuristic algorithm:
according to Eq. (4) for the ith sub-network.
Step 1. Based on the placement position list obtained from
In Eq. (5), it can be noticed that there is no coupling be-
the measurement placement algorithm, determine the bus
tween sub-networks. Thus, there is no need to interchange
where either an injection current or a busbar voltage measur-
the measurement or state vector between them. Moreover,
ing device is installed. These buses are called major buses.
the state variables of each sub-network can be estimated inde-
Other buses are called minor buses.
pendently. Since the size of the sub-network is smaller, deter-
mining an optimal measurement placement can be performed Step 2. If there is a line current measuring device on the
with less computation efforts using either the minimum con- branch connected to the major buses, the device is moved
dition number criteria [10] or other approaches [1416]. In close to the major buses.
Fig. 2. Six-bus power system (a) before reduce site and (b) after reduce site.
44 C. Rakpenthai et al. / Electric Power Systems Research 75 (2005) 4149
Step 3. From the branches with line current measuring de- 4. Results and discussions
vices which are not connected to major buses, determine the
minor bus with the maximum connection number of those In this section, the proposed algorithm is applied to the
branches which have line current measurements. Then, the IEEE test systems under balanced conditions in which single-
line current measuring device on the connected branches is phase representation is adequate. The results are compared
moved close to the selected bus. Note that this minor bus will with the proposed method in [10] called Madtharads algo-
not be considered again in the next iteration. rithm. The true state of the network is computed by using
power-flow calculation. It is assumed that the suitable mea-
Step 4. Repeat Step 3 until the maximum connection num- surement equipment capable taking synchronized measure-
ber of line current measuring branches equals to one. ments are available. The covariance of the measurements is
assumed to be an identity matrix. The program has been de-
Example. The six-bus power system is shown in Fig. 2(a) veloped using MATLAB and the tests have been performed
in which there are five placement sites. on Intel Celeron 2.2 GHz 256 MB PC.
Table 1
Internal buses, boundary buses, and slack bus of each sub-network in the IEEE test systems
System Sub-network Buses in sub-network Internal buses Boundary buses Slack bus
IEEE 14 bus 1 13, 5 1 2, 3, 5 1
2 4, 79 7, 8 4, 9 8
3 6, 1014 1113 6, 10, 14 6
IEEE 30 bus 1 19, 11, 28 13, 5, 7, 8, 11 4, 6, 9, 28 1
2 1219, 23 1316, 18 12, 17, 19, 23 13
3 10, 2022, 2427, 29, 30 21, 22, 25, 26, 29, 30 10, 20, 24, 27 22
IEEE 118 bus 1 118, 31, 33, 3537, 113, 117 17, 914, 16, 18, 33, 35, 113, 8, 15, 17, 18, 31, 36, 37, 113 1
117
2 6871, 7390, 97, 116, 118 7379, 81, 8388, 116, 118 6871, 80, 82, 89, 90, 97 116
3 1930, 32, 34, 38, 43, 44, 65, 2023, 2528, 43, 114, 115 19, 24, 29, 30, 32, 34, 38, 44, 32
66, 72, 114, 115 65, 66, 72
4 3942, 4564, 67 41, 42, 46, 48, 5061, 63 39, 40, 45, 47, 49, 62, 64, 67 59
5 9196, 98112 9395, 100112 91, 92, 96, 98, 99 110
Table 2
The numerical experimental results of Madtharads algorithm
System Sub-network Number of Results
possible positions
Number of Number of placement sites Number of CPU time (s) Condition
measurement measurement number
positions Before reduce After reduce typesa
IEEE 14 bus None 57 14 9 8 3, 0, 11 1.14 22.00
IEEE 30 bus None 115 30 23 16 3, 8, 19 14.48 146.05
IEEE 118 bus None 481 118 98 88 4, 79, 35 b 1, 113.19
a For example, 3, 8, 19 indicate that the number of busbar voltage, injection current, and line current measurement types are 3, 8, and 19, respectively.
b Denotes 11,805.29 s or 3.28 h.
For the proposed algorithm, the entire network is appropri- the number of placement sites is reduced using the outlined
ately decomposed into many sub-networks and the slack bus heuristic algorithm.
of each sub-network is chosen. The bus no. 1 is defined as the For the Madtharads algorithm, which considers the entire
global reference bus of the system; the slack bus of each sub- network, the busbar voltage measurements are the same as
network will be calculated by referencing to this reference in the proposed algorithm. And the reduction of the place-
bus. After the measurement placement of each sub-network is ment sites is also performed after the placement positions are
performed based on the minimum condition number criteria, determined.
Table 3
The numerical experimental results of the proposed algorithm
System Sub-network Number of Results
possible positions
Number of Number of placement sites Number of CPU time (s) Condition
measurement measurement number
positions Before reduce After reduce typesa
IEEE 14 bus 1 10 4 4 2 1, 0, 3 0.03 19.23
2 11 4 2 2 1, 0, 3 0.02 18.96
3 16 6 4 3 1, 0, 5 0.04 20.72
IEEE 30 bus 1 34 11 8 6 1, 0, 10 0.34 82.58
2 24 9 6 5 1, 0, 8 0.12 34.83
3 29 10 8 5 1, 1, 8 0.20 59.21
IEEE 118 bus 1 78 25 18 12 1, 6, 18 4.37 681.46
2 81 25 18 14 1, 8, 16 4.94 654.98
3 58 22 16 12 1, 2, 19 1.66 196.40
4 90 25 17 13 1, 8, 16 6.11 548.84
5 75 21 13 9 1, 1, 19 3.19 127.10
a For example, 1, 1, 8 indicate that the number of busbar voltage, injection current, and line current measurement types are 1, 1, and 8, respectively.
46 C. Rakpenthai et al. / Electric Power Systems Research 75 (2005) 4149
Table 4
Measurement placement of Madtharads algorithm
System Busbar voltage at bus Injection current at bus Line currenta
IEEE 14 bus 1, 6, 8 None (1, 5), (3, 2), (4, 7), (4, 9), (5, 2), (6, 12), (8, 7), (9, 14), (11, 6), (11,
10), (13, 12)
IEEE 30 bus 1, 13, 22 1, 5, 12, 14, 15, 16, 27, 30 (1, 2), (5, 2), (6, 2), (6, 8), (6, 9), (6, 28), (10, 17), (10, 20), (10, 21),
(11, 9), (12, 4), (13, 12), (15, 12), (19, 18), (19, 20), (24, 23), (26,
25), (27, 28), (27, 29)
IEEE 118 bus 1, 59, 110, 116 2, 3, 6, 7, 9, 10, 12, 13, (8, 5), (9, 8), (11, 4), (17, 15), (17, 30), (17, 31), (17, 113), (18, 17),
1519, 2125, 2729, 32, 33, (19, 20), (25, 26), (32, 114), (34, 36), (35, 37), (38, 37), (38, 65), (49,
38, 4052, 63, 66, 6972, 54), (50, 49), (51, 49), (51, 58), (59, 55), (59, 56), (59, 63), (61, 60),
7476, 7984, 8697, 99107, (61, 64), (65, 68), (66, 49), (66, 62), (66, 67), (69, 68), (71, 70), (80,
110112, 115, 117, 118 77), (80, 79), (80, 98), (81, 68), (105, 108)
a For example, (1, 5) represents the line current measuring device on the branch connecting bus nos. 1 and 5 (I15 ).
The IEEE 14 bus system is a small system and can be power system. The condition numbers of the measurement
decomposed to three observable sub-networks. The IEEE 30 matrices of all sub-networks are less than that obtained from
and 118 bus systems are decomposed based on the method Madtharads algorithm. The number of placement sites is re-
in [18] to three and five sub-networks, respectively. Table 1 duced when the heuristic reduction algorithm is applied for
shows the internal buses, the boundary buses and the slack most cases. In addition, by this decomposition technique,
bus of each sub-network. The numerical experimental results the state estimation can be performed in parallel. The mea-
of the Madtharads algorithm and the proposed algorithm are surement placements of the Madtharads algorithm and the
summarized in Tables 2 and 3, respectively. In these tables, proposed algorithm for all IEEE test systems are shown in
the CPU time indicates the computational time for solving Tables 4 and 5, respectively. Figs. 3 and 4 illustrate the IEEE
the measurement placement algorithm, not including the re- test systems and the associated measurement set from the pro-
duction of placement sites in order to compare both algo- posed algorithm in case of the IEEE 14 and 30 bus systems,
rithms. Notice that the CPU time of the proposed algorithm is respectively. The figure for the case of IEEE 118 bus can be
shorter than Madtharads algorithm, especially for the large likewise obtained with both cases using results in Table 5.
Table 5
Measurement placement of the proposed algorithm
System Sub-network Busbar voltage at bus Injection current at bus Line currenta
IEEE 14 bus 1 1 None (1, 5), (2, 3), (2, 5)
2 8 None (4, 7), (4, 9), (8, 7)
3 6 None (6, 11), (6, 12), (11, 10), (13, 12), (13, 14)
IEEE 30 bus 1 1 None (1, 2), (1, 3), (2, 4), (2, 5), (2, 6), (6, 9), (6, 28), (7, 5), (8, 28),
(11, 9)
2 13 None (12, 14), (12, 16), (13, 12), (15, 14), (15, 23), (15, 18), (17,
16), (19, 18)
3 22 30 (10, 20), (10, 21), (22, 10), (22, 24), (25, 24), (25, 26), (25,
27), (29, 27)
IEEE 118 bus 1 1 1, 2, 9, 10, 11, 35 (5, 6), (5, 8), (7, 6), (9, 8), (11, 5), (11, 13), (12, 7), (12, 14),
(12, 16), (12, 117), (15, 14), (15, 33), (17, 16), (17, 18), (17,
31), (17, 113), (35, 37), (37, 33)
2 116 74, 75, 81, 8386, 118 (71, 70), (71, 73), (75, 69), (75, 74), (77, 69), (77, 80), (77,
82), (79, 78), (79, 80), (80, 97), (81, 68), (81, 80),
3 32 20, 43 (83, 82), (85, 88), (89, 88), (89, 90), (19, 34), (20, 21), (22,
21), (23, 22), (23, 24), (23, 25), (24, 72), (26, 25), (26, 30),
(27, 28), (27, 115), (29, 28), (32, 23), (32, 27), (32, 114), (38,
30), (38, 65), (43, 34), (66, 65)
4 59 41, 42, 46, 48, 51, 57, 59, 63 (40, 39), (42, 49), (46, 47), (48, 49), (51, 58), (54, 49), (54,
53), (57, 56), (58, 56), (59, 54), (59, 55), (59, 56), (59, 63),
(62, 60), (62, 67), (64, 61)
5 110 107 (92, 91), (92, 93), (94, 93), (96, 95), (94, 96), (100, 92), (100,
98), (100, 99), (100, 101), (100, 104), (102, 101), (103, 104),
(103, 105), (107, 105), (108, 105), (110, 103), (110, 109),
(110, 111), (110, 112)
a For example, (2, 3) represents the line current measuring device on the branch connecting bus nos. 2 and 3 (I23 ).
C. Rakpenthai et al. / Electric Power Systems Research 75 (2005) 4149 47
Fig. 4. IEEE 30 bus system and associated measurement set obtained from the proposed algorithm.
Each sub-network is connected together through the hand, for the Madtharads algorithm the error may affect the
tie-lines. With the placement position obtained from the system globally.
proposed algorithm, the power system can still be observ- In order to demonstrate the effectiveness of the proposed
able although any tie-line is outage. On the contrary, with algorithm, the proposed approach is compared with one of the
the placement obtained from Madtharads algorithm, the en- well-known methods [19]. In [19], the concept of spanning
tire network may become unobservable if some measuring tree has been applied to obtain a fully observable measure-
devices are installed on these tie-lines since these measur- ments set. Then, the number of placement sites is minimized
ing devices are lost when they are out of service. For ex- by using the modified bisection search combined with the
ample, according to the placement positions obtained from simulated annealing-based method. The result of IEEE 14
Madtharads algorithm shown in Fig. 3(a), if the line (914) is bus system by this method has three placement sites, at bus
outage the network becomes completely unobservable since nos. 2, 6, and 9. Each placement site consists of one injec-
the line current measurement (I914 ) is lost. However, for the tion measurement and four flow measurements. A current
placement positions obtained from the proposed algorithm as pseudo-measurement is installed at bus no. 7. Thus, network
shown in Fig. 3(b), the network can still be observable under observability is obtained by 15 measurements and 1 pseudo-
the same outage condition. measurement. The condition number of measurement matrix
The effect of gross error condition is performed for the of this result is 7.5 1016 . Comparison of the placement re-
proposed algorithm and the results are compared to the sults obtained by both methods is summarized in Table 7.
Madtharads algorithm. Table 6 illustrates the error in the Note that although the method in [19] gives smaller num-
magnitude and the angle of bus voltages when the gross error ber of placement sites, the condition number of the mea-
is occurred. For the proposed algorithm, since the error is not surement matrix is much higher than the proposed method.
propagated into other sub-networks, the estimated values of Thus, the precision of the estimated states tends to be less
those sub-networks are still the true solution. On the other accurate.
48 C. Rakpenthai et al. / Electric Power Systems Research 75 (2005) 4149
[17] C. Madtharad, S. Premrudeepreechacharn, N.R. Watson, Power sys- [19] T.L. Baldwin, L. Mili, M.B. Boisen Jr., R. Adapa, Power system
tem state estimation using singular value decomposition, Electr. observability with minimal phasor measurement placement, IEEE
Power Syst. Res. 67 (2) (2003) 99107. Trans. Power Syst. 18 (2) (1993) 707715.
[18] I.O. Habiballah, M.R. Irving, Multipartitioning of power system state [20] J.J. Grainger, W.D. Stevenson Jr., Power System Analysis, McGraw-
estimation networks using simulated annealing, Electr. Power Syst. Hill, Singapore, 1994.
Res. 34 (2) (1995) 117120.