Professional Documents
Culture Documents
Abstract
This paper develops an improved real and reactive power control technique using linear programming (LP)
for an integrated power system. The problem is decomposed into two subproblems comprising real (P) and
reactive power (Q) modules, and, using a unified approach, the real power generation, voltage magnitude,
and transformer tap settings are optimized. The objective function is the fuel cost which is minimized in
both the P and Q modules, subject to the operating constraints. The P - Q decomposition combined with the
LP formulation improve the computation speed. The paper has another advantage of using the same cost
objective function for both modules, unlike other conventional methods which use the power loss function
for the Q module. The LP formulation is used for both the P and Q optimization modules, utilizing the
revised simplex method which is normally available in a mainframe computer.
[Aal
AP~ = a12J22-' LAp~j+(J,3- J~Je~-'J~) AV
S u b s t i t u t i n g for A6gt from (13), (21) becomes
AQ~,= JG AP~ + J a AV + Ji AN (22)
+ (J~ - J,~J~-~J~4) AN (14)
where
Since AP~ --0 as noted in (10),
JG ~ 1st (m - 1) columns of
AP~ = J~ AP~ + J~ AV + J c AN (15)
J42J2,-l: l' × (m + I - l')
where
JH ~ J 4 s - - J42J22-1J2~ : l' × (m + l)
J~ ~ 1st (m - 1) elements of
J, =~ J44 - J42J29-1J24: l~ × Nt
Jl~Je~-l: 1 × (m + l - 1)
This sensitivity equation reflects the effect of
J~ ~ J13 - J ~ J ~ - ~ J ~ : 1 x (m + l)
the control vectors, APg, AV, and AN, on the
J c ~ Jl4 - J12J22 ~J24:1 × Nt reactive power in the load buses.
3.3. P optimization module by the incremental changes in voltage magnitude
In this module the P optimization problem is and tap settings. A constraint for this module is,
solved by minimizing the incremental cost func- from (27),
tion (7) as a function of the incremental change
in real power generation. Linear programming is [1: - JA] APsg = JR AV + Jc AN (29)
used to obtain the optimal values of the real
Additional constraints are obtained from (9),
power generation subject to the operating con-
by using the decoupling conditions A~sg~= 0 and
straints. For the P module using the decoupling
AQr = 0 for the Q module:
condition
AP~ = A~s = 0 (23) AQsgc = J33 AV + J34 AN (30a)
the linearized constraints (16) and (20) are sim- 0 : J43 AV + J44 AN (30b)
plified as
The constraint (29) implies that when AV and
[1: - JA] AP.,g = SS (24) AN are adjusted in the Q optimization modules,
APsg will also be changed, which will then
and
change the incremental cost (26). Therefore, the
AQ~gc = JD APg (25) cost function for the Q module can be obtained
by augmenting the real cost function (26) with
The P optimization module then becomes
constraint (27):
minimize AC.(AP~g) = Cp T APsg (26)
ACq = ACp + ~([1: - JA] APsg - JR AV - ac AN)
subject to
(31)
[1: - JA] AP~g = SS (27)
where ~ is the Lagrange multiplier. Note that
AP~g ~ Psg <~AP~g has a physical meaning in that it represents
where the first-order sensitivity of the objective func-
tion (ACp) with respect to active constraint on
AV.~g = Vsg - Vsg APsg. Thus, its inclusion in the Q module objec-
tive function is to show the effects of the P
AP~g = Psg - P~g module.
Note in (31) that the variation
SS -~ JR AV + Jc AN
This is a linear programming (LP) optimiza- AC, + ~([1: - JA] APsg)
tion problem which can be solved by the revised
is assumed to be already optimized in the P
simplex method [10] to find the optimal solution
optimization module, and is thus a fixed constant
APsg. Another benefit of this P optimization mod-
in this Q optimization module. Additional mini-
ule is that, as a byproduct, it computes the
mization of (31) can be achieved by the changes
reactive power changes AQ~gc caused by the P
in A V and AN.
optimization as given by the sensitivity equation
Thus, from the constraints (29) and (30) and
(25).
the cost function (31), the Q optimization module
Finally, the solution of the P optimization
becomes
module is used to update P and Q as follows:
P~+~=P~+AP~ minimize ACq(AV, A N ) = -A(JR A V + Jc AN)
Qk÷~
SgC = Qs~c + A Q ~ (28) (32)
where k is the iteration number. subject to
[ A: A~] 1
Solve the LP problem to find optimal
g'= / g~ adjustment to control variables
[_-i, -A2J 1
I'e o thelo o I
b' = [0, AQ ~ g ~ T, --AQ~gcT] T -- A'X
~= I x - x ]
then the Q module becomes
minimize Z = CTy (44)
subject to Fig. 2. F l o w c h a r t for the linear p r o g r a m m i n g for the P or Q
module.
A'Y~b' (45a)
O<Y<Y (45b) vised simplex method. The constraint matrices
are constructed and updated for each iteration. If
These are the standard forms for the LP, appli- the limits of the control variables are exceeded,
cable in finding the minimal incremental cost, then the nearest limits are used. Also, to limit
and the optimal incremental real power genera- the oscillations in the variables, step values be-
tion, bus voltage, and transformer tap settings. tween 5% and 10% are used as the upper or
Conventional LP packages are easily available lower limits. This is necessary as it is undesirable
and can be used. In this paper the IMSL DLPRS to allow large swings of real power and bus
subroutines have been used, where the revised voltage magnitudes which will affect the perfor-
simplex method is implemented [7]. The con- mance of the generators. The results obtained
straint matrices have to be constructed for each are compared with those of the GPM method
P and Q optimization to conform with the stan- used in refs. 8 and 9.
dard input format for DLPRS subroutines and
are updated after every P and Q optimization 5.1. The 6-bus system
module. The complete computation procedure is The operational problem for the 6-bus system
shown in Fig. 2. was solved by the LP method. Figure 3 shows the
single-line diagram for the 6-bus system. The line
and generator data are given in ref. 6. Two shunt
5. S i m u l a t i o n r e s u l t s capacitors are used to control the bus voltage
magnitudes at load buses 4 and 6.
The optimal operation problem is solved for The initial load flow is run when no reactive
sample systems using the linear programming compensation has been applied at the load buses
(LP) formulations. For comparison, the power apart from the reactive generation at buses 1 and
systems used in the paper are the 6-bus and the 2. It can be seen from Table 1 that load buses 3-6
modified IEEE 30-bus, which were also used in all have undervoltages below the lower operat-
the earlier work [6]. ing limit of 0.90p.u. The optimal power flow
The problem is decomposed into two subprob- (OPF) results for both methods are tabulated and
lems, the P and Q modules, and each is solved compared in Table 1. Column 4 shows the initial
using LP. The LP code used is implemented in load flow, column 5 the GPM results, and column
the IMSL DLPRS subroutine that uses the re- 6 the LP results.
bus 2. The corresponding reactive power genera-
~ _ ~ S'C'I T2 _3
tions for the LP are more evenly distributed than
for the GPM. At the reactive compensated load
buses 4 and 6, the reactive power increase is
about 2.4% and 2.7%, respectively, being higher
for the LP.
The operating security constraints, that is,
voltage magnitudes, tap settings, and real power
5 . 2 generations, are all within the operating limits
6- I I I
for both methods. All undervoltages at the load
buses have been compensated by the reactive
Fig. 3. Single-line diagram of the 6-bus system: 1 , . . . , 6, bus
shunt capacitors at buses 4 and 6 with 15.00 and
numbers; G, generators; S.C., shunt capacitors; T, tap setting 15.50 MVAR, respectively. Also, the transformer
transformers. tap settings on lines 4 and 7 are adjusted to 0.94
and 1.10 ratios, respectively, to improve the
From an initial generation cost of 619.53 voltages at uncompensated load buses 5 and 3.
S/hour, the GPM gives a lower optimal operation The operation cost convergence is shown in
cost of 392.11 S/hour, compared with 395.78 Fig. 4, where the LP method takes more itera-
S/hour for the LP method, with net savings of tions, about 12, to reach the optimal value, com-
58.0% and 56.5%, respectively. The difference pared with 9 for the GPM; however, the average
between the two methods is about 0.94%; this is computation time per iteration is shorter for the
due to the linear approximation of the nonlinear LP than for the GPM.
cost function in the LP. The real power loss by
LP is about 3.3% higher than that of the GPM, 5.2. The 30-bus system
both being within an acceptable range. Note that The operational problem for the IEEE 30-bus
both methods result in higher system loss com- system was solved by the LP method, and the
pared to the initial state, but the operation cost results are compared with those of the GPM [8].
is reduced since the objective of this methodol- The line and generator data are given in ref. 6.
ogy is to minimize it, while other conventional Figure 5 shows the single-line diagram of the
methods minimize the loss function. modified IEEE 30-bus system.
The real power generation by LP at swing bus The initial load flow from Fig. 6 shows that the
1 is 1.1% more than that by the GPM, and both voltage magnitudes at load buses 14-30 are all
methods have the same generation of 100 MW at below the lower operating limit of 0.95 p.u. This
~First s t u d y - - w i t h both P and Q optimization using GPM; second study with both P and Q optimization using LP.
650.00
600.00 27 28
550.00 i 25 I 26
23
~ 500.00
g
o "1~ 14 ----L- I ~ ~I$
450.00 "~-
0 " ~ " "-~,. " ' B
12-
300.00 . . . . . . . . . ~. . . . . . . . . ~. . . . . . . . . I . . . . . . . . . I . . . . . . . . . I . . . . . . . . . I
0.00 2.00 4.00 6.00 8.00 10.00 12.00
No. of iterotions
1.10
1.00 ~ 7 - -,~- . . . . . . . . . . . . .
&
o
e ~ . o ~ Before compensation
: : : : ~ After compensation
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0.80
Fig. 6. voltage magnitude distribution for the modified IEEE 30-bus system before and after optimal power flow for peak load using
linear programming.
10
TABLE 2. Summary of results for modified IEEE 30-bus system ~' the optimization improved them to smaller val-
ues, thus enhancing the power system stability.
Variable Limits Initial First Second It was noted t h a t the new formulation using
state study study
Lower Upper
a unified J a c o b i a n matrix to compute sensitiv-
ity matrices improved the c o m p u t a t i o n time
P1 (MW) 50.0 200.00 99.905 177.870 235.615 compared with that of the earlier w o r k [6].
P2 (MW) 80.0 70.00 80.000 48.596 20.000 Also, comparing the LP with the G P M [8, 9]
P5 (MW) 15.0 50.00 50.000 21.373 10.000 with a similar formulation, it was found that
Ps (MW) 10.0 35.00 20.000 21.038 10.000
PlI (MW) 10.0 30.00 20.000 11.783 10.000
the LP has a faster c o m p u t a t i o n speed than
P~:3 (MW) 12.0 40.00 20.000 12.000 12.000 the GPM, making it a t t r a c t i v e for real-time
Q1 (MVAR) -20.0 200.0 0.000 5.788 6.046 application in economic dispatch, and reac-
Qz (MVAR) - 20.0 100.0 15.973 41.752 100.000 tive p o w e r and voltage control. The use of
Q5 (MVAR) - 15.0 80.0 44.132 22.639 35.609 the voltage m a g n i t u d e instead of the reac-
Qs (MVAR) -15.0 60.0 56.127 30.425 33.935
QI~ (MVAR) - 10.0 50.0 30.238 17.475 - 10.000
tive power as the control variable in the Q
Q13 (MVAR) -15.0 60.0 32.499 2 0 . 7 9 4 -13.704 module resulted in a linear formulation and
NI~ 0.90 1.10 1.00 1.053 0.903 r e d u c e d oscillations in the convergence charac-
N~2 0.90 1.10 1.00 0.969 1.094 teristics.
NI~ 0.90 1.10 1.00 1.045 1.0320
/~ 0.90 1.10 1.00 1.006 1.031
Qc~.~ (MVAR) 0.0 5.0 0.0 3.941 5.793 h
Qdv (MVAR) 0.0 5.0 0.0 4.652 15.749 b
Acknowledgement
Qc20 (MVAR) 0.0 5.0 0.0 3.389 5.994b
Qc2~ (MVAR) 0.0 5.0 0.0 4.118 16.626 b
Qc2:~ (MVAR) 0.0 5.0 0.0 3.037 0.491 h This work has been supported in part by the
Qc24 (MVAR) 0.0 5.0 0.0 4.582 6.283 b Allegheny Power System, Greensburg, PA, USA,
Qc29 (MVAR) 0.0 5.0 0.0 3.287 2.699 b and by the Korean Electric Power Corporation,
Generation cost (S/h) 903.309 799.5601 840.315 Seoul, South Korea.
System loss (MW) 6.505 9.260 14.215