You are on page 1of 6

IEEE Transactions on Power Systems, Vol. PWRS-2, No.

1, February 1987 175

SECURITY-CONSTRAINED OPTIMAL POWER FLOW WITH


POST-CONTINGENCY CORRECTIVE RESCHEDULING

A. Monticelli M.V.F. Pereira S. Granville


Unicamp Cepel Stanford University
Campinas, Brazil Rio de Janeiro, Brazil Stanford, CA

Abstract - This paper describes a mathematical frame- the outage has occurred. These capabilities include
work for the solution of the economic dispatch problem generation rescheduling, switching, overload rotation,
with security constraints, which can take into account etc., and may be very significant for the elimination
the system corrective capabilities after the outage has of constrained violations. One example can be found in
occurred. The solution algorithm is based on mathemat- predominantly hydro systems, where the fast response
ical programming decomposition techniques that allow rate of the hydroelectric generators ensures that a
the iterative solution of a "base-case" economic dis- substantial rescheduling capability will be available.
patch and separate contingency analysis with generation
rescheduling to eliminate constraint violations. The This paper describes a mathematical framework for
resulting dispatch has the same security level as the the solution of the security-constrained dispatch that
usual security-constrained dispatch, but with lower can take into account the system rescheduling capabili-
operating costs. Case studies with the IEEE 118-bus ties. The methodology is based on mathematical pro-
are presented. The extension of the methodology to the gramming decomposition techniques that allow the itera-
probabilistic case will be discussed in a forthcoming tive solution of a "base-case" economic dispatch and
paper. separate contingency analysis with generation resche-
duling to eliminate constraint violations. The preven-
INTRODUCTION
tive control actions are implemented from linear con-
straints, known as feasibility cuts, that are generated
The objective of the economic dispatch of an on the contingency analysis and sent to the economic
electric energy system is to determine a generation dispatch problem. The resulting dispatch has the same
schedule that minimizes the system operating cost security level as the usual security-constrained
(measured, for example, in terms of fuel cost for the dispatch, but with lower operating costs.
thermal plants and of "shadow" costs for the energy- Development of the framework for the linearized power
limited hydro plants) and does not violate any of the flow model was based on Benders decomposition and is de-
system operating constraints such as maximum line scribed in [2]. This paper extends the methodology to
flows, bus voltage levels, etc... the AC power flow model by using generalized Benders decom-
position [3]. Convergence of the GBD algorithm can only be
It has beem recognized for some years that the guaranteed under some convexity assumptions, which cannot
economic dispatch may be unsafe, that is, it may not be be proven for the AC power flow model. Our computational
possible to keep the system in a normal state after a experience so far has been encouraging, and is illustrated
major disturbance (line and/or generator outages). by case studies with the IEEE 118-bus system.
This led to the concept of system security, and to the
view that the objective of system operation is to keep FORMULATION OF THE DISPATCHI PROBLEMS
the system in a normal state during the relatively long
periods between disturbances and to insure that, on the Example
occurrence of a major disturbance, the system does not
depart from the normal state [1]. The following simple example will be used to char-
acterize the economic dispatch problem with security
This security-constrained dispatch is usually constraints Generators #1 and #2 in the 2-bus sys tem
,

implemented by adding other cons-traints, known as of Figure 1 are used to supply a 200-MW load in bus #2,
security constraints, to the economic dispatch prob- The system data are as follows:
lem. These constraints impose additional limits on
branch flows and nodal voltages for the post-disturbance #1
configurations resulting from a given set of contingen- Generator #2
cies. In other words, the security-constrained dispatch
leads to the implementation of preventive control Min. Generation (MW) 50 0
actions in the system, and thus to a higher level of Max. Generation (MW) 200 120
system security. Incr. Cost ($/MW) 1 2

However, the security-constrained dispatch as Line #1 #2


defined above is conservative, because it does not take Max. Flow (MW) 100 120
into account the system corrective capabilities after
1 ( 1) 2

(2)

86 WM 085-5 A paper recommended and approved


by the IEEE Power System Engineering Commitee of Figure 1: Two-Bus System
the IEEE Power Engineering Society for presentation
at the IEEE/PES 1986 Winter Meeting, New York, New
York, February 2 - 7 1986. Manuscript submitted Since generator #1 has the lowest operation cost,
September 3, 1985; made available for printing the "pure" economic dispatch solution would be as in
January 27, 1986. Figure 2; the total operating cost would be equal to
200.
0885-8950/87/0100-0175$01 .00(1987 IEEE
176
100_- In this problem, the variables xo represent
operating decisions (e.g., generation level in each
200 cost = 200 bus). The set ao(xo) > bo represents the system
operating constraints (power flow equations, bounds on
100_ generation and flows, etc.). The objective is to
200 minimize the system operating cost, c(xo).
Figure 2: "Pure"l Economic Dispatch The Security-Constrained Dispatch
Suppose now that a security-constrained dispatch
is required, that is, the system should have no over- Suppose now that we have a list of M possible
loads if either line #1 or #2 is outaged. In this disturbances, corresponding for example to line and/or
case, it is easy to see that the output of generator#l generator outages.
will be constrained by the capacity of the remaining
line. Since line #1 has the lowest capacity (100 MW), Let the set of operating constraints for the i-th
the most severe contingency will correspond to the post-disturbance configuration be represented as
outage of line #2. The resulting security-constrained
dispatch will be as shown in Figure 3. hi(fi(x0)) >
bi for all i=l,22...,M (2)

100 where hi(.) and fi(.) are vector functions. fj(xo)


50 _w represents automatic control actions in case the dis-
1 00 ~ cost = 300 turbance disconnects any controllable device. For exam-
50 -w ple, a generator outage will cause other generators to
200 automatically adjust their outputs to restore the gen-
eration-demand balance. In this case,
Figure 3: Security-Constrained Dispatch fi (xO x0 6x0 (3)
It can also be seen in Figure 3 that the security-
constrained dispatch has a higher operating cost than where 6x0 is a vector with a few nonzero elements,
the "pure" economic dispatch of Figure 2, because corresponding to the generators that will be adjusted.
preventive generation at bus # 2 is now required.
For notational simplicity, we define
However, the above dispatch did not take into
account the system corrective capabilities after the
outage has occurred. Suppose, for example, that
ai (x0) =
hi(fi (x0)) (4)

generators #1l and #2 can be redispatched (increase or The security-constrained dispatch is then written as
decrease generation) by 40 MW and 35 MW, respectively,
before the protection system is actuated (ramping con- z2 Mmn c(xo)
straints). In this case, it is possible to increase
the output of generator #1 and obtain the dispatch of
Figure 4.
s /to ao (xo) >,- b 0 (5)
ai (x0) >, bi For all i =1,2,...,M
67.5 _ _
65 The constraints ai(xo) > bi impose additional
135 Cost =
265 limits on power flows for the post-disturbance configu-
_-200
67. 5_ rations. As mentioned in the Introduction, the
security-constrained dispatch (5) leads to the imple-
mentation of preventive control actions on the system,
Figure 4: Security Constrained Dispatch and thus to ahigher level of system security.
with Corrective Rescheduling
Security-Constrained Dispatch With Rescheduling
Note that this dispatch will not lead to overloads As discussed in the Example, the security-
in case any of the lines is outaged, because the output constrained dispatch defined in (5) does not take into
of generator #1 can be safely reduced to 100 MW in the account the possibility of post-disturbance corrective
most severe case (a reduction of 35 MW, less than the
ramping constraint of 40 MW), and compensated by an in- actions, such as generation reschedul'ing, to eliminate
creased generation at bus #2, within its ramping limit the constraint violations. The possible range of
of 35 MW. It is thus possible to obtainl the same level rescheduling actions depends on time constraints and
of security as the previous dispatch, but with a lower equipment limits, and will be represented by coupling
operating cost of 265 in this case. This kind of constraints of the type Jxo - xil < Ai (fxo - xl is a
vector whose elements are the absolute values of the
savings can be particularly significant during the elements of xo - xi). The security constrained dispatch
peak-hours. with rescheduling can then be represented as
The dispatch problems of Figures 2, 3 and 4 can be
mathematically formulated as follows. Z3 = Min c(xo)
s/to ao(xo) >b (6)
The Economic Dispatch Problem
The "pure" economic dispatch of an electric energy ai(xi) > bi
For all il,12,...,M
system is formulated as the following optimization
prob lem: xiI <
Ai
z1 - Min c(xo) The coupling constraints also serve as a "bridge"
(1) between the "pure" economic dispatch (1) and the
s/to a0(x0) > bo security-constrained dispatch (5). It is possible
177
to see that if we make A - 0 (no corrective actions Representation of Infeasibilities
allowed), Problem (6) becomes identical to the
security-constrained dispatch (5). Conversely, if we The infeasible post-contingency operating subprob-
make A - X (the full range of rescheduling actions lems will be represented as follows.
allowed in the period), the post-disturbance operating
points xi, i -1,2,...,M become independent of the Suppose that penalty variables have been added to
operating point xo, and Problem (6) becomes identical the constraints of each post-contingency subproblem.
to the "pure" economic dispatch (1). Their function is to ensure the mathematical feasibil-
ity of the subproblem for any given x0, and they can be
It is also possible to see that the operating interpreted as the amount of constraint violation asso-
costs of (6) have as lower and upper bounds respec- ciated to the post-contingency operating point xi. If a
tively the costs of economic dispatch (1) and the costs subproblem is feasible, these variables are naturally
of security-constrained dispatch (5), that is, equal to zero. Therefore, we can define the minimiza-
tion of constraint violations as an objective function.
The operating subproblem is then formulated as in (8),
z1 <z3 <
z2 (7)
where the post-contingency index i has been dropped for
notational convenience.
As discussed in the Example, the security-
constrained dispatch with rescheduling has the same w = Min d.r + d!s
security level as the usual security-constrained
dispatch, but with lower operating costs. The solution s/to a(x)
methodology will be described next. + r > b (8)
-xo -
Xl - s < A
OUTLINE OF THE SOLUTION METHODOLOGY where r > 0 is the vector of penalty variables for
set of operating constraints
Introduction
Problem (6) can be interpreted as a two-stage s > 0 is the vector of penalty variables for
set of coupling constraints
decision process:
In the first stage, find an operating point
dr and ds are positive cost vectors
o xo
for the "base-case" problem ao(xo) > bo From Problem (8), we can conclude that
o In the second stage, given the operating point w - 0 <-> Subproblem is feasible
xo, find new operating points xi that meet the w > 0 <-> Subproblem is not feasible
constraints ai(xi) bi and the coupling con-
>
straints |xo - xij |Ahi for each post-distur- These properties will be used in the decomposition
bance configuration i 1l,2, ...,M. methodology described next.
The objective is to minimize the operating cost Outline of the Decomposition Methodology
c(xo) while ensuring that the second-stage problems are
feasible. Figure 5 illustrates the decision process. The decomposition methodology is based on the
following observations:
The solution methodology is based on mathematical
programming decomposition techniques that allow us to 0 The objective function w in the operating
handle separately the "base-case" economic dispatch and subproblem (8) can be seen as a function of
each of the M post-contingency operation subproblems. the operating point xo, that is,
The representation of infeasibilities will be discussed
next. w(xo) - Min drr + d s
s/to a(x) + r > b (9)
STAGE I
Ixo - XI - s < A
0 The security-constrained dispatch with
rescheduling (6) can then be rewritten in
terms of the xo variables as

Min c(xo)
s/to ao(xo) > bo (10)

STAGE II
wi(x ) < 0 for all i -1,2,...,M

where the scalar functions wi(xo) are the solutions of


Problem (9) for any given operating point xo. The
constraint wi(xo) < 0 is equivalent to requiring the
XM problem feasibility.
aM(xM ) > bM Feasible
Each function wi(xo) supplies information about
IX0 - xml < -AM the "consequences" of the operating decision xo if the
---
- i-th contingency occurs. If these functions were
Figure 5. Decision Process for Economic Dispatch
available, the security-constrained dispatch with re-
scheduling could be solved as Problem (10), without an
with Security-Constrained Rescheduling explicit representation of the M operating subproblems.
178

The Benders decomposition scheme [3j is a technique found in [9].


for building an approximation to w(xo) based on the it-
erative solution of a "base-case" economic dispatch and Although our computational experience with the AC
M separate operating subproblems. The decomposition al- power flow model has been encouraging so far, and posi-
gorith can be summarized in the flowchart of Figure 6. tive results are also reported in a related VAR planning
application [8], great care should be used when applying
It can be seen in Figure 6 that the approximate this technique.
dispatch problem is only written in terms of the "base-
case" variables x0 and can be solved by any specialized CASE STUDY
algorithm. Similarly, the M post-contingency operation
subproblems can be solved independently. In this section, a specific dispatch problem will
be used to illustrate some of the concepts discussed
The critical point in the decomposition scheme is the previously.
improvement of each w.(x ) from the solutions of the
operating subproblems. Aossociated with each solution , Network Modelling
there is a set of Lagrange multipliers which measure the
change in system infeasibility caused by marginal changes The system will be represented by an ac-power flow
in the base-case operating point xo. These multipliers model:
are used to form a linear constraint, known as Benders
cut, written only in terms of the base-case variables a. active power mismatch: AP = 0
x.. The modification of wi(xo) corresponds to adding the b. reactive power mismatch: AQ = 0
associated Benders cut to the base-case dispatch prob-
lem.
c. bounds on flows ((P.. + Q.2) ): S < S (11)
In summary, the decomposition scheme summarized in
Figure 6 can be seen as a systematic way of constraining d. bounds on active generation: P < P P
eg g g
the solution space of the base-case dispatch problem in
order to avoid infeasibilities in the post-contingency e. bounds on reactive generation: Q ( Q ( Q
problems.
f. bounds on taps: t < t , t
A derivation of Benders decomposition can be found
in [3]. Other applications of this technique to power g. bounds on voltage: V ( V S V
system problems are described in [4-8]. An example in
which the system is represented by an ac-power flow model Objective Function/Base-Case Problem
will be discussed next.
The objective function of the base-case problem is
the minimization of the system operating costs.

z = Min c
E Z(P0 ) (12)
k kg

subject to (11-a) through (ll-g)


where k indexes the set of generating buses
ck is the active generating cost function
P0 is the base-case operating point
g
Objective Function Post-Disturbance Subproblem
As discussed previously, the objective of the
operation subproblem is to minimize some measure of the
system infeasibility after the contingency has occurred.
In the present study, we have chosen to minimize the
amount of violation of the coupling constraints. The
post-disturbance operation subproblem associated to a
given base-case operating point Pg is formulated as

w(P0) = Min E s
k
(13)
Figure 6. Outline of the Decomposition Algorithm g k

Convergence of the Benders decomposition algorithm subject to (11-a) through (ll-g)


can only be guaranteed under some convexity assumptions
[3]. In contrast with the linearized power flow model and
described in [2], these assumptions cannotbe proven for
the AC power flow model described in the next section. P
g
pl
g
-s4 A
If w(x) is not convex, the tangential approximations
produced by the decomposition method may not be valid where s > 0 is a vector of slack variables which
lower bounds. As a consequence, Benders decomposition correspond to violations of coupling constraints
may fail or may produce a sequence that does not converge
either to a global or local optimum when applied to some
problems. The effect of monconvexities of w(x) was il-
lustrated in the discussion/closure of [7]. Numerical
examples for a transmission planning application can be
179
Note that there are coupling constraints only for which can take into account the system corrective cap-
active power generation. In the present study, we are abilities after the outage has occurred, has been
assuming that reactive generation can be freely described. The solution methodology is based on decom-
rescheduled. position techniques that allow the iterative solution
of a "base-case" economic dispatch and separate contin-
Calculation of the Benders Cuts gency analysis with generation rescheduling to
eliminate constraint violations.
As discussed in the Outline of the Decomposition
Methodology, a linear approximation to the function One of the main advantages of the decomposition
w(xo) is generated after the solution of each post-dis- approach is its flexibility:
turbance operation subproblem. It can be shown (see
[9]) that the coefficients of this linear approximation o It is possible to use existing OPF models and
are the Lagrange multipliers associated to the con- solution procedures for the base-case and con-
straints of the operation subproblem. For the problem tingency subprob lems.
defined in (12), the specific form of the Benders cut
is o It is possible to extend the modelling of correc-
tive actions to include, for example, multi-period
w
0
-
(14) rescheduling, overload rotation, line switching,
+
XA(Pg Pg) < 0
etc.
o The M subproblems can be solved independently of
where w the optimal solution value of (14)
is each other, and can benefit directly from parallel
*0
Pg is the base-case trial operating point processing technologies [11].
x is the vector of Lagrange multipliers o It is possible to model different line limits and
associated to the coupling constraints ramping constraints corresponding to the different
time frames of emergency control for a given
*o
Note that w*, Pg and AA are known values. Expres- contingency (for example, no rescheduling allowed
sion (14) represents the change in the post-disturbance for the short-term emergency rating of lines; some
infeasibility as a linear functi2n of the changes in rescheduling allowed for the longer-term ratings).
the base-case operating point g. Therefore, the
constraint coefficients can be interpreted as "gradient o It is possible to take into account deviations in
directions" to enhanced system security. load forecast and changes in system configuration
(unit commitment, circuit outages, etc.).
Numerical Results o Although the solution scheme of Figure 6 assumes
that the security-constrained dispatch with re-
The methodology was applied to the IEEE 118-bus scheduling (3) is always feasible, it is possible
system, represented in Figure 7. Quadratic cost func- to reformulate the problem slightly to reflect the
tions were defined for the system generations. Five tradeoff between operating cost and system reli-
contingencies were selected: four circuit outages (5/8; ability, represented, for example, by the expected
26/30; 65/68; 89/92) and one shunt capacitor on bus 5. load curtailment in the post-disturbance subprob-
The ramping constraints A were defined as 1/9 of the lems. It is also possible to use the multipliers
generating capacity at each generating bus. As men- associated to the feasibility cuts in the "base-
tioned previously, no coupling constraints were imposed case" dispatch to determine what is the implicit
on reactive power rescheduling. cost of reliability in the system.

The operating subproblems were solved with a modi-


fied version of the optimal power flow model developed ACKNOWLEDGMENTS
by [8].
This work receives financial support from the Task
The first interaction corresponded to a "pure" Force responsible for the development of the National
economic dispatch problem (all security constraints Control Center (SINSC/Eletrobras), from EPRI under
relaxed). Given the resulting base case operating Project RP-2473-6, with Stanford University, and from
point and coupling constraints, contingency analysis the Brazilian National Research Council (CNPq). The
was then carried out. Only one contingency (#2) led to second author would like to thank Yr. Robert Iveson of
infeasibilities that could not be alleviated by resche- EPRI for his support of our technical activities. The
duling. One Benders Cut (14) was accordingly generated helpful comments and suggestions of Dr. Jacques
and returned to the base-case dispatch problem. Carpentier of EDF, Dr. Ongum Alsac of PCA, and
Professor George Dantzig and Dr. Michael Saunders of
A new base-case operating point was then calcu- Stanford University, are greatly appreciated. Special
lated and sent to the subproblems. Two contingencies thanks go to Word Processing Center 3 of EPRI and Elza
(#2 and#4) led to infeasibilities and returned Benders Aoki of Unicamp for the expert preparation of the
cuts. Optimality was reached after five iterations. manuscript. Figure 7 was taken from EPRI Report-3729,
prepared by Scientific Systems, Inc.
A post-optimality analysis was then carried out to
determine the tradeoff between operating cost and REFERENCES
system infeasibility, and to identity the contingencies
that most affect the system operating cost. In this 1. Dy Liacco, T.E., "Mathematical Challenges in Elec-
case, contingency #2 was dominant. Further details can tric Power System Operation: An Overview," in Elec-
be found in lO]. tric Power Problems-The Mathematical Challiiie
SIAM, 1980.
DISCUSSION AND CONCLUSIONS 2. Pereira, M.V.F, Monticelli, A. and Pinto, L.M.V.G.,
"Security-Constrained Dispatch with Corrective
A mathematical framework for the solution of the Rescheduling", IFAC Symposium on Planning and
economic dispatch problem with security constraints Operation of Electric Energy Systems, Rio de
180
Janeiro, Brazil, July 22-25, 1985. 8. Lebow, W.M., Rouhani, R., Nadira, R., Usoro, P.B.,
Mehra, R.K., Sobieski, D.W., Pal, M.K. and
3. Geoffrion, A.M., "Generalized Benders Decowposi- Bhavaraju, M.P., "A Hierarchical Approach to Reac-
tion", JOTA, Vol. 10, No. 4, 1972, pp. 237-261. tive Volt Ampere (VAR) Optimization in System Plan-
ning", IEEE Transaction on PAS, Vol. PAS-104, No. 8,
4. Bloom, J., "Long-Range Generation Planning Using pp. 2051-2057, August 1985.
Decomposition and Probabilistic Simulations", IEEE
Transactions on PAS, Vol. PAS-101, pp. 797-802 9. Granville, S., Pereira, M., "Analysis of the Lin-
1982. earized Power Flow Model in Benders Decomposition",
Technical Report SOL 85-04, Systems Optimization
5. Pereira, M. and Pinto, L., "Application of Decom- Laboratory, Stanford University, February 1985.
position Techniques to the Mid- and Short-Term
Scheduling of Hydrothermal Systems", IEEE Transac- 10. Granville, S., Pereira, M.V.F. and Monticelli, A.,
tios on PAS, Vol. PAS-102, pp. 3611-3618, 1983. "A Mathematical Framework for the Security-
Constrained Economic Dispatch with Corrective
6. Pereira, M. and Pinto, L., "Stochastic Optimization Rescheduling", EPRI Final Report for Project
of a Multi-reservoir Hydroelectric System - A RP2473-6, in preparation.
Decomposition Approach", to be published in Water
Resources Research, 1985. 11. Pereira, M. and Pinto, L., "Distributed Processors
and Decomposition Techniques in Power System Plan-
7. Pereira, M., Pinto, L., Cunha, S. and Oliveira, G., ning and Operation", CEPEL Technical Report, 1984.
"A Decomposition Approach to Automated Generation/
Transmission Expansion Planning", IEEE Transactions
on PAS, Vol. PAS-104, No. 11, November, 1985.

Figure 7 : IEEE118-bus System


(Taken from EPRI Report EL-3729,
prepared by Scientific Systems Inc.)

You might also like