Professional Documents
Culture Documents
Q-Problem Sequential: (1) (Or Cost)
Q-Problem Sequential: (1) (Or Cost)
K.Y. Lee, Member IEEE Y.M. Park J.L. Ortiz, Member IEEE
Abstract - This paper presents a unified method production cost under the assumption that system
for optimal real and reactive power dispatch for the voltages are held constant, and the Q-problem is to
economic operation of power systems. As in other minimize the transmission loss under the assumption
methods, the problem is decomposed into a P- that real power generation is held constant. Due to
optimization module and a Q-optimization module, but in the loose coupling between two problems, the sequential
this method both modules use the same generation cost optimization of these provided a considerable advantage
objective function. The control variables are over the simultaneous optimization of all control
generator real power outputs for the real power module; variables. It should also be noted that a number of
and generator reactive power outputs, shunt articles are devoted to the Q-problem alone for the
capacitors/reactors, and transformer tap settings for optimum control of reactive power flow [8-10].
the reactive power module. The constraints are the The optimization problem is a general nonlinear
operating limits of the control variables, power line programming problem with nonlinear objective functions
flows, and bus voltages. and nonlinear functional equality and inequality
The optimization problem is solved using the constraints. A typical approach is to augment the
gradient projection method (GPM) which is used for the constraints into objective function by using the
first time in the power systems study. The GPM allows Lagrange multipliers [11] and/or penalty functions, and
the use of functional constraints without the need of to minimize the augmented objective function by using
penalty functions or Lagrange multipliers among other one of the optimization schemes, such as the steepest
advantages. descent algorithm [1,7], or the sequential
Mathematical models are developed to represent the unconstrained minimization technique (SUMT) [61 . Other
sensitivity relationships between dependent and control approaches are the use of linear programming
variables for both, real and reactive power, approximation to the nonlinear problem [8,10J, or the
optimization modules, and thus eliminate the use of B- use of the quadratic approximation to the objective
coefficients. Results of two test systems are function in order to apply the quadratic programming
presented and compared with conventional methods. technique [12]. Due to the size of the problem as well
as the large number of functional inequality
I. INTRODUCTION constraints, improvement on computational efficiency
has been the thrust of most works.
The optimal operation of a power system is This paper presents a new method for optimal real
required to precede the optimal planning of facilities and reactive power dispatch for economic operation of a
or devices for the system. In general, these power system. The method is based upon three modules
facilities consist of generating plants, reacti-ve power coupled to each other. First, the P-optimization
compensation, and transmission networks. Since the module, which is equivalent to the conventional
optimal power flow method was first introduced by economic load dispatch, optimally allocates the real
Dommel and Tinney [11, many articles have appeared in power generation among generators. The second module,
the literature on this subject [2-4]. Q-optimization module, optimally determines the
The optimal power flow problem is to minimize the reactive power output of generators and other var
fuel costs, system losses, or some other appropriate sources as well as transformer tap settings. Finally,
objective function, while maintaining an acceptable the load-flow module is used to make fine adjustmentr
system performance in terms of limits on generator real on the results of P- and Q-optimization modules. The
and reactive power outputs, outputs of compensating algorithm developed and presented here has the
devices, transformer tap settings, or bus voltage following distinctive features and advantages:
levels, etc. When total fuel cost is minimized, the (1) Unlike the usual approach of minimizing the
optimal power flow results in an economic dispatch, and transmission loss [6,7,101, the total production
in addition to optimally allocating real power .cost (or fuel cost) is the objective function to
generation to units, determines the reactive power be minimized by both, P and Q, optimization
output of generators and other var sources as well as modules. This approach unifies two decoupled
transformer tap settings. optimization problems into one reference framework
In order to handle the large scale problems of and avoids the switching of objective functions
this nature, the idea of P-Q decomposition was applied from one to another.
to the optimal power flow [5-71 , where the problem is (2) Like the P-optimization module, the Q-optimization
decomposed into the real power optimization problem (P- module also allows an economic and efficient
problem) and the reactive power optimization problem division of real power generation between
r(obrblem). The P-problem is to minimize the generators. This additional benefit is, of
course, due to our choice of the objective
function mentioned above.
84 SM 572-4 A paper recommended and approved (3) The optimization problem is solved by the use of
by the IEEE Power System Engineering Committee of the gradient projection method (GPM) [131, which
the IEEE Power Engineering Society for presentation introduces several advantages over traditional
at the IEEE/PES 1984 Summer Meeting, Seattle, approaches. This method avoids the use of penalty
Washington, July 15 - 20, 1984. Manuscript submit- functions and/or calculation of Lagrange
ted September 1, 1983; made available for printing multipliers [1,6,7], additionally, the step length
May 2, 1984. is calculated without the need of a linear search
algorithm as in other methods.
(4) Mathematical models are developed to represent the
sensitivity relationships between dependent and
0018-9510/85/0005-1 147$01.00© 1985 IEEE
1148
control variables for both optimization modules, Minimize
and thus eliminate the use of B-coefficients.
(5) Mathematical models are developed with a great C = f (P ) (3)
care in order to preserve the sparsity of
Jacobians, and thus the sparsity is fully utilized Subject to
in computing the sensitivity matrices.
(6) Since the Q-optimization module determines the P < P P
optimum reactive power allocation, all buses,
except the swing bus, are made to be the Q-type
sg sg sg
1I (4)
bus in the load flow module, where reactive power sg)
g(P9(sg) 0
(output from the Q-optimization module)
specified instead of voltage magnitude. The swing
bus voltage is not fixed, but is determined
optimally by the Q-optimization module.
is
H(6) . H
where
I
11. FORMULATION AND DECOMPOSITION OF THE PROBLEM
f ("): the total summation of generator fuel
The optimal real and reactive power dispatch costs of Eq. (l)expressed as a function
problem is mathematically defined as: of Psg
P .P .P Subject to
sg sg sg
Q sgc .<Q sgc . Q
Q .Q .Q sgc sgc
sqc sgc
n < n <n (6)
(2)
n 5! n -, n
AP AS6
or, in matrix form
AC(AP ) = 3 AP +APT YpAP (10) 9
sg P sg sg Ps APi -L. - - -- -
where AQS
AQg
p - [bI+2cIPj,b2+2c2P2 *...bm+2cmPm] AQc 12..3.. AV
AQ , AV,
=p- diag.Ac1,c2,... ,cmI (12)
m: total number of generator buses where
APsg: vector of changes in real power
generations Psg. s,g,c: indices for swing bus, other generator
buses, and reactive-power compensating
An important point associated with the cost device buses, respectively.
function is the fact that Eq. (10) is directly usable 9,V: indices for all load buses, and the load
in the P-optimization module since AP itself is the buses which do not have reactive power
decision variable in that module. In the Q- compensating devices, respectively.
optimization module, however, AP should be expressed
as a function of the Q-optimizaivon control variables Although more exact decomposition can be realized
AQs and An. This will be shown later. by setting
AP = AV = A65 = AQ = O , (13)
Minimize
AC P = a P Ap Sg +APTSg y p Ap Sg (15)
Subject to
[1 -jiAp =0° 1
Ap . AP < TP Sg (16)
sg sg
Figure 1. Computational sequence of three modules.
Another benefit of this P-optimization module is
1150
that as a byproduct, it also computes the reactive Minimize
power changes AQ caused by the P-optimization. This
relationship is g-ven by a sensitivity equation AQ sgc 1AQ sgc
sgcl
AQ
gc
=j
B
Ap
t(17)
g
AC = L.QAn An
An
YQ [n
(23)
Subject to
where J is defined in Appendix A.
Afver the optimum solution to the P-optimization AQ . AQ sgc .7ibf
subproblem is found by using the gradient projection sgc sgc
method (GPM), both the real and reactive powers are
updated from P and Qg to P +AP and Q +AQ An . An . An (24)
respectively. g gc g g gc gc
IV. Q-OPTIMIZATION MODULE A QsgcI
D
hV re D --. AV
For the Q-optimization module, the Jacobian matrix
J in Eq. (11) is augmented to include the sensitivity
coefficients representing the changes in real and where
reactive power with respect to the changes in off-
nominal tap settings of the LTC as
[API - QA SPjC y Ty
Q=-A i CPC
i
TABLE i. Data for the 6-Bus System (100 M?4A Base) presented in this paper represents a significant
(A) Load Data (C) Generator, Data improvement over traditional methods.
Bus Loads
Another important advantage is in the
Number P(p.u.) I Q(p.u. computational aspect. These simulations were performed
Bus Cost Coefficients on the NEC AS/900ON system in the University of
1 0.000 0.000 No. a b c
Houston. The time per single iteration for the first
2 0.000 0.000
3 0.550 0.130 1 0 1. 0.05 study was about 0.318 seconds. The relatively short
4 0.000 0.000 2 0 1. 0.01 execution time for this method suggests a bright
5 0.300 0.180 promise of using the method for on-line application to
6 0.500 0.050 the optimal load flow calculations. The time per
(B) Line Data single iteration for the second and third studies was
about 0.173 seconds which is much smaller than in the
Line From Bus To Bus Line Impedance Tap first study. The reason for this shorter time is due
Number Number Number R X Setting to the use of the P-module which requires less
1 1 6 0.123 0.518 _ computations than the Q-module. For all three studies,
2 1 4 0.080 0.370 - the short computation time per iteration and the fast
3 4 6 0.097 0.407 -
convergence are not only due to our particular
4 6 5 0.000 0.300 1.025 mathematical formulation, but also due to, in large
5 5 2 0.282 0.640 -
VIl. CONCLUSIONS TABLE IV. Summary of Results for the 30-Bus System
TABLE III. Data for the 30-Bus System (100 MVA Base) P13(MW) 12. 40. 20.000 13.355 12.000
Bus Load Bus Load Q2 (MVAR) -20. 100. 27.657 27.423 27.439
No. P(p.u.) Q(p.u.) Mo. P(p.u.) Q(p.u.)
Q5 (MVAR) -15. 80. 21.544 21.205 21.223
1 0.000 0.000 16 0.035 0.018
2 0.217 0.127 17 0.090 0.058 Q8 (MVAR) -15. 60. 22.933 23.086 23.104
3 0.024 0.012 18 0.032 0.009 Qll (HVAR) -10. 50. 38.583 38.438 38.453
4 0.076 0.016 19 0.095 0.034
5 0.942 0.190 20 0.022 0.007 Q3 (HVAR) -15. 60. 40.345 40.423 40.355
6 0.000 0.000 21 0.175 0.112
7 0.228 0.109 22 0.000 0.000 13
8 0.300 0.300 23 0.032 0.016 VI (p.u.) 0.95 1.10 1.05 1.10 1.10
9 0.000 0.000 24 0.087 0.067
10 0.058 0.020 25 0.000 0.000 V2(p.u.) 0.95 1.10 1.04 1.08 1.08
11 0.000 0.000 26 0.035 0.023 V5(p.u.) 0.95 1.10 1.01 1.03 1.03
12 0.112 0.075 27 0.000 0.000
13 0.000 0.000 28 0.000 0.000 V8(p.u.) 0.95 1.10 1.01 1.04 1.04
14 0.062 0.016 29 0.024 0.009
15 0.082 0.025 30 0.106 0.019 V11(P.u.) 0.95 1.10 1.05 1.08 1.09
(B) Line Data V13(p.u.) 0.95 1.10 1.05 1.08 1.08
Line From Bus To Bus Line Impedance Tap n11 0.900 1.100 1.078 1.072 1.072
Number Number Number R(p.u.) X(p.u.) Setting
n12 0.900 1.100 1.069 1.070 1.069
1 1 2 0.0192 0.0575 -
2 1 3 0.0452 0.1852 -
n15 0.900 1.100 1.032 1.032 1.032
3 2 4 0.0570 0.1737 -
4 3 4 0.0132 0.0379 -
n36 0.900 1.100 1.068 1.068 1.068
5 2 5 0.0472 0.1983 -
6 2 6 0.0581 0.1763 -
Qcio(MVAR) 0.0 5.0 0.0 0.692 0.711
7 4 6 0.0119 0.0414 -
8 5 7 0.0460 0.1160 -
9 6 7 0.0267 0.0820 - Qc12(MVAR) 0.0 5.0 0.0 0.046 0.045
10 6 8 0.0120 0.0420 -
11 6 9 0.0000 0.2080 1.078 Qc15 (MVAR) 0.0 5.0 0.0 0.285 0.304
12 6 10 0.0000 0.5560 1.069
13 9 11 0.0000 0.2080 - Qcl7(MVAR) 0.0 5.0 0.0 0.287 0.306
14 9 10 0.0000 0.1100 -
15 4 12 0.0000 0.2560 1.032 Qc2O(}VAR) 0.0 5.0 0.0 0.208 0.129
16 12 13 0.0000 0.1400 -
17 12 14 0.1231 0.2559 - Qc2l(MVAR) 0.0 5.0 0.0 0.000 0.000
18 12 15 0.0662 0.1304 -
19 12 16 0.0945 0.1987 - QC23 (MVAR) 0.0 5.0 0.0 0.330 0.350
20 14 15 0.2210 0.1997 -
21 16 17 0.0824 0.1932 - Qc24(MVAR) 0.0 5.0 0.0 0.938 0.957
22 15 18 0,1070 0.2185 -
QC9(MzAR) 0.0 5.0 0.0 0.269 0.290
23 18 19 0.0639 0.1292 -
24 19 20 0.0340 0.0680 -
25 10 20 0.0936 0.2090 - Generation Cost [$/hr] 901.918 804.853 823.629
26 10 17 0.0324 0.0845 -
27 10 21 0.0348 0.0749 - System Losses [MW] 5.812 10.486 10.154
28 10 22 0.0727 0.1499 -
29 21 22 0.0116 0.0236 -
30 15 23 0.1000 0.2020 - here utilizes the same performance measure (the
31 22 24 0.1150 0.1790 - total fuel costs) for both optimizations, which is
32 23 24 0.1320 0.2700 -
more realistic and reasonable performance.
33 24 25 0.1885 0.3292 -
34 25 26 0.2544 0.3800 - (2) More economical reallocation of real powers in the
35 25 27 0.1093 0.2087 - case of reactive power optimization - unlike the
36 28 27 0.0000 0.3960 1.068 conventional methods which, for reactive power
37 27 29 0.2198 0.4153 -
optimization, fix all non-swing bus real powers
38 27 30 0.3202 0.6027 -
39 29 30 0.2399 0.4533 - and change only the swing-bus real power to absorb
40 8 28 0.6360 0.2000 - the reduction in transmission losses, the Q-
41 6 28 0.0169 0.0599 - optimization method presented here optimally
reallocates all generator real powers because of
(C) Generator Data the performance measure being defined as the total
Bus Cost Coefficients fuel .cost.
No. a b | c (3) Non-fixed swing-bus voltage - In some of the
conventional algorithms the swing-bus voltage
1 0.0 2.00 0.00375 remains fixed in the reactive power optimization
2 0.0 1.75 0.01750 procedure, but the method presented here optimally
5 0.0 1.00 0.06250 determines the swing-bus voltage, along with any
8 0.0 3.25 0.00834 other bus voltages.
11 (4) More accurate algorithms - The conventional B-
0.0 3.00 0.02500 coefficients for transmission loss formula are
1 13- 0.0 3.00 0.02500 functionally replaced by more accurate sensitivity
1153
matrices which are successively updated during the APPENDIX A
iteration.
(5) Fast computation - The method presented here, P-Optimization Model
inspite of a lot of matrix operations, still
preserves the highly-sparse characteristics of Using the conditions AV = A'S = 0
system Jacobian matrices, and thus makes it
possible to use the optimally-ordered triangular of Eq. (13), the real powers in Eq. (12) can be
factorization technique which allows for handling expressed in terms of power angles as
of a large power system with faster computation.
(6) Application of the gradient projection method -The AP = 9ll (Al)
computer program developed for this study is shown
to be efficient and highly reliable for finding an
optimum in the optimization process. Also, this
method provides faster convergence in the optimal g 1
power dispatch problem than other conventional
methods.
= L (A2)
rAPJ
A wt12
A t
oj
(7) Possibility for on-line applications - The fast Accordingly, with the use of Eqs. (A2) and J (A3) we
and reliable characteristics in computation get
mentioned above present the future possibility for AP
its on-line applications, such as the economic AP
5
= j- 21-
i I I1 9 A
JA 9 (A3)
load dispatch, reactive power-voltage control, on-
line optimal load flow, etc. or w
APt
I..
tAP!
VIII. REFERENCES
[I JA] fAPgf (A4)
[1] H.W. Dommel and W.F. Tinney, "Optimal Power Flow
Solutions", IEEE Transactions on Power Apparatus since AP =- from Eq. (13). Here JA is the vector of
and Systems, Vol. PAS-87, pp. 1866-1876, October first (m-\) elements of the matrix product in Eq. (A3).
1968. The dependent variable AQ c for the P-optimization
[2] H.H. Happ, "Optimal Power Dispatch-A -Comprehensive module also can be derived as
Survey", IEEE Transactions on Power Apparatus and
[3]
Systems, Vol. PAS-96, pp. &41-854, May/June 1977.
J. Carpentier, "Optimal Power Flows", Int. J. of
Q =gi
A
LAQc
Q 13 12. gAPJ
LApJ B g(A5)
Electrical Power and Energy Systems, Vol. 1, pp.
3-15, April 1979. where iB is the vector of first (m-k.) elements of the
[4] 0. Alsac and B. Scott, "Optimal Load Flow with matrix product in Eq. (A5).
Steady-State Security", IEEE Transactions on Power
App tUs and , Vol. PAS-93, pp. 745-751, Q-Optimization Model
May/June 1974. Using the conditions = .A'S A6' 0
[5] F.F. Wu, G. Gross, J.F. Luini, and P.M. Look, "A of Eq. (20), the real and reactive power generations in
As