3, August 1991
1069
Jestls Lumbreras
During the last decade there has been a growing concem in the
electric utility industry about reactive power issues at the
operation planning and capacity expansion levels, as it can be
easily perceived by scanning the pertinent technical literature, see
[l-151. Among other reasons, this interest stems from the fact that
in many power systems there is a trend towards operating them
closer to their technical limits. This is particularly true at the
transmission network level. Therefore, issues such as voltage
control, voltage instability, reactive power security dispatch and
reactive power planning have attracted much attention.
The objective of reactive power planning is the determination
of the new VAR sources (in terms of type, size and location in the
network) that will result in an adequate voltage control capability
by achieving a correct balance between security and economic
concems. Reactive power planning typically has a time horizon
of one to three years, and it is performed in coordination with
transmission capacity studies, that have a longer time horizon and
a higher priority.
In reactive power planning it is sought to find the right
tradeoff between the investment cost of the new VAR sources and
the benefits in system operation derived from the presence of
these additional means of reactive compensation.
It is therefore clear that, at the heart of any reactive power
planning automated tool, there should be a good system operation
module that is able to evaluate the impact of the reactive power
control elements (generators, regulating transformers and shunt
capacitors and reactors) on the economy and security of the system
operation.
1070
VAR sources are modelled as discrete elements, each one having a
total associated cost. that includes purchasing and installation
costs. The formulation of the master problem results in a mixed
integer linear programming problem, where the linear constraints
are the Benders cuts provided by the operation subproblems.
The main features of the proposed technique that differ from the
methodology in [13] are:
- A two-level decomposition scheme, instead of the
proposed three-level, is always used between each
operation case (base or contingency) and the master
problem. This is possible because the dispatch of
preventive control variables has been included in this
master problem, in addition to the VAR investment
problem. The derived advantages are a simpler formulation
and better computational efficiency.
- The operation problem is simpler and faster than a
complete OPF. This is critical for the overall
computational efficiency of the algorithm, because the
operation subproblems have to be solved many times
during the global process.
- When an operation subproblem is not feasible, the
objective function is defined as a linear penalty function of
the existing feasibility violations. Benders cuts now
represent the feasibility constraints.
- The constraint matrix in the formulation of the operation
problem is sparse. The adopted formulation exploits this
feature and also avoids having to invert the matrix,
regardless of the control variable selection made by the
user.
The installation of new VAR sources is modeled by discrete
(0-1) variables, instead of a fixed cost plus a variable cost,
therefore improving and accelerating the convergence
process of the Benders method.
After this introductory section, the complete mathematical
formulation is presented in section 2, where the operation
subproblem is formulated first, and then the decomposition
strategy for the global process is explained. In section 3, the
results of applying the developed tool to real electric power
systems are presented.
F(z)
subject to:
S(z) 5 0
G(z)
00
where:
WZ)
se)
G(z)-og
OD-
- S
1071
T b e o/v o a b i z a t i o n p r o b h
As it was mentioned before, several situations requiring
different formulations have to be considered: the user must specify
a reduced number of base cases that represent the whole range of
normal operating conditions and also a number of associated
contingencies which could produce voltage or reactive power
supply problems in the system. As the adopted formulation is Q/V
decoupled, the proposed operating cases should have been
previously checked for compliance with economic dispatch
conditions and security limits conceming active power controls.
The user may also specify whether preventive or corrective control
actions are desired for each one of the contingency cases. When a
contingency occurs, the control variables in preventive mode will
remain fixed on their previously specified optimal positions,
while the corrective variables will be able to change in order to
prevent violations of the operating limits. The control variables
in preventive or corrective mode can be chosen among the
following altematives:
- voltage magnitude at generation bus V c
- voltage magnitude at a load bus which is controlled by a
regulating transformer or by a VAR shunt element V
- regulating transformer tap NT
- susceptance of a VAR shunt element BC
Fomiulation of a base case
Here the objective function to be minimized is the total
amount of active power losses. Among several altematives, the
loss formulation in [21] was chosen:
where:
Pijo
;
: buses of line i-j
: voltage magnitudes at buses i and j.
(3)
<
The list of preventive control variables that can be selected by
the user was given before. The procedure that has been chosen here
to make sure that these variables stay at their pre-contingency
values is the following: The precontingency values are optimized
in a separate problem, which is called the master problem and is
the one that finds the optimal investments in VAR sources (see
section 2.2). The trial values for the preventive control variables
that are provided by the master problem, are used in equations (4)
and ( 5 ) of the subproblem as both upper and lower bounds,
therefore forcing the subproblem variables to take these precise
values.
1072
-
2.2
eroblem
Once the representation of the system operation for a given
installed VAR capacity has been described in the preceding
section, the global formulation of the entire reactive expansion
problem is presented here. Then the decomposition procedure is
explained, leading to the formulation of the investment
subproblem. Note that, for the sake of simplicity, only one base
case and its associated contingencies have been formulated.
Minimize
0.Zk.y
....
Z
0
this is a subset of
the operation variables, whose values have to be
the same for cases k=1, L.
C(o)
: cost function of the new reactive power
compensation equipment. This cost corresponds
to the annual charge calculated using a prescribed
discount rate.
Fk(Zk) : operation cost function for the k-th operating
subproblem, as described in (3).
Sk(zk) : set of constraints for the k-th operating
subproblem, as described in (4) and (5).
Gk(zk)-o: subset of the constraints (4) whose limits
( B c m a x , B c m i n ) are imposed by the VAR
investment variables. The reactive compensation
capacity to be used at a bus cannot exceed the
limits set by the installed capacity.
: set of constraints on VAR investments, i.e.,
R(o)
maximum capacity allowed at each candidate bus
or maximum number of reactive power elements to
be installed at each candidate bus.
T(y)
: bounds for the preventive control variables. Note
that if the control variable is the reactive power
delivered by a new VAR source in a candidate bus,
then the corresponding constraint will take the
form T(y) + R(o) 5; 0.
A two-level Benders decomposition procedure for solving the
global VAR expansion problem is presented. In the Benders
terminology, the decomposition will be achieved by
simultaneously making a projection of the complete problem on
the VAR investment variables CO and on the preventive control
variables y, i.e., an iterative process between the master or
investment subproblem and the operation subproblems is
established. The master problem determines CO* and y*, using only
incomplete information about the operating subproblems. On the
other hand, each operation subproblem optimizes the operating
variables z for the given values of U)* and y* and retums to the
master problem information about the optimal objective function
and its sensitivities with respect to o and y in (co*,y*). See figure
2.
TO start the algorithm, an initial guess is needed for a feasible
set of the new installed VAR sources WO and for the values of the
preventive control variables yo, so that the Q/V operation
problems can be solved. The base case (k=l) and its associated
preventive contingencies (k=2, ...,L) are sequentially formulated
and solved, as it was explained previously. That is:
: preventive control variables;
....
Minimize
Zk
Fk(zk)
subject to:
(9)
k=1, ...,L
where:
k
zk = YO
where 00 is a constant term that represents the VAR shunt
equipment available for dispatch.
The same process takes place with the rest of the corrective
contingencies (k=L+l, ...,M). That is:
1073
Minimize
Zk
not necessarily the optimal one, because the master problem did
not have exact information about vk to find the optimum of 0 and
Fk(Zk)
4
MASTER PROBLEM
Obtain the new V A R sources
and the values of the preventive
control variables
I@
. Y*
I I
of the
I M A L STATE
FERATIONSUBPROBLEMS
-RFADTHE~IAL
INSTALLED Q
-PROPOSE I
"
VALUES
FOR PREVENTIVE
COMROL VARIABLES
QjV OPERATING
PROBLEMS
W R
W W PROBLEM
OFVARINYBSTMENTAND
PREVPNTIVECONIROL
DISPATCH.
EQ (8.9)
EQ (12.13)
[ C(o) +
vk ]
k= 1
C v;
to the optimal
0;
and the
JRESULTS
1074
the CPU solution times that have been obtained with the three
different approaches on a SIEMENS 7590F computer (12 MIPS). It
should be noted that the 1 0 s minimization is formulated as a
linear maximization of voltage magnitudes in the SLP approach.
Table I
NUMBER OF BUSES
OPF (MW)
350.4
492.5
SLP (Mw)
SLP s
BD (Mw)
351.3
492.6
13.3
492.9
25.6
350.6
10.6
686.9
176.2
689.8
45.5
689.6
83.2
Table I1
METHOD
OPF (Mvar)
OPF (s)
I
SLP(Mvar) I
SLP (s)
BD (Mvar)
.
. I
BD (s)
NUMBER OF BUSES
102
I 213 I
259.0 I
313.41
9.5 I
35.4 I
258.1.1
310.1 I
5.5 I
13.5
272.1 I
319.1 I
7.1
13.7
I
I
44 3
230.7
116.7
235.0
52.9
254.8
46.7
1075
Table IILa
nn
q/v
IT.
IT,
BASE
14473.
11475.
14179.
14471.
6
6
I44TI.
(F4)
(F3)
2867.
568.
245.
12.
186.
70.
28.
15.
I.l462.
7
6
7
5
(i
7
8
(Fz)
14509.
~~~~~18.
OP ER AT IN G C ( ) S TS
CUNT2
CONT3
c.ONTI
(FI)
12
I
2
1
I
537.
996.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
TOTAL.
.\PROS.
OPT.
VAL.
CON\'.
C'KI'I.
CI'IJ
Fk)
(X''k)
(1
(5-5)
18909.7
15302.9
1499,i.z
14773.7
11773.7
14773.7
14773.7
14773.7
I8909.7
602.9
259.1
25.4
20.0
18.1
11.7
(x
1.~10.
15016.
14707.
I448G.
14662.
11549.
14499.
14486.
0.0
0.0
, T .AI,LEI
ITER 2
EEi
0.00
0.0
l44IZ.R
1wa.2
14460.3
I1462.8
11468.7
14466.2
14471.2
0.00
0.00
0.00
100.00
0.00
0.00
50.01
0.00
100.00
0.00
11.13
19.84
100.00
40.92
0.00
98.78
0.00
48.51
89.50
0.00
53.95
87.25
0.00
100.00
0.00
0.00
0.00
0.00
0.00
2a7.25
90.30
0.00
96.54
ioo.00
0.00
0.00
100.00
0.00
~86.84
14.20
287.99
290.93
r.4
'r
,I I6
563
R!)fi
8I5
nfio
IODI
1222
1353
0.00
75.57
49.60
ioo.no
Table IV.a
Table I V.b
I79
Acknowledgements
I65
IO8
mrr,\L
0.00
0.00
300.00
100.00
0.00
300.00
0.00
0.00
350.00
0.00
0.00
350.00
4. CONCLUSIONS
[ 11
[2]
[3]
[4]
1076
".