Documente Academic
Documente Profesional
Documente Cultură
A. J. Conejo E.T.S. de Ingenieros Industriales Universidad de Castilla-La Mancha Ciudad Real, Spain.
Problem Formulation
Introduction
An Optimal Power Flow (OPF) problem determines the optimal steady-state operation of a power system that minimizes a chosen objective function and satisfies certain physical and operating constraints. This problem is a typical nonlinear programming problem. The Multi-area Optimal Power Flow (MOPF) problem arises when the operation of several interconnected systems is involved. One way to solve this problem is to assume the interconnected systems as a global-single system, and a single OPF is solved by, say, an InteriorPoint (IP) algorithm [l]. On the other hand, several approaches have been proposed to solve the MOPF problem so that each interconnected system is independently operated and still a global optimum is achived. In this type of solution,
*Visiting Scholar at the University of Waterloo
In this paper we solve a nonlinear programming problem - the OPF problem - by a sucessive linear aproxis considered in this paper is the mation. The OPF that i Multi-area OPF problem. In its formulation we assume that there exists a bus at each tie-line that interconnect different areas (see Fig. (4); for a detailed discussion, see [3]). The MOPF problem can be mathematically formulated as follows:
min subject to
682
f ( x ) : is a scalar function that models the global operating cost ; Z(x) : is a vector with the load flow equations a t buses in the tie-lines ; g ( x ) : is a vector with the load flow equations at the rest of the buses;
h ( z ) : is a vector with the operating limits on the system, with lower bound hmin, and upper bound
hmax
.
7
this model is limited to a small region around a given operating point. Therefore, an iterative procedure is s well required to update the system operating point a as the linear model used in the problem formulation. Once the problem (2) is linearized, an efficient onestep Interior-Point algortihm can be implemented to solve the problem [l]. In the next secction we present a decompositon approach in which a Lagrangian Relaxation and an Interior Point algorithms are combined to solve problem (5).
x : is a vector of the unkown variables with lower and upper limits xmln and xmaz, respectively.
It should be noted that all voltages magnitudes and all voltages angles, except for the swing bus, are modeled as bounded variables. Assuming that a current point ~k satisfies load-fiow equations g ( x k ) a d I(.k), the linearized problem wn be written in the following form: min subject to
The Lagrangian Relaxation technique Permits the decomposition of the OPF multi-area problem into several one-area OPF problems. Once the original problem is decomposed, an infeasible primal-dual interior-point algorithm is used to solve each independent problem.
3 . 1
t2)
Since in problem (5) all variables are continuous, the objective function is strictly convex and the constraints are linear, it can be solved the dual problem of ( 5 ) instead of solving the primal problem directly. The dual problem of (5) can be written as
where J g ( z k ) , J l ( x k ) and J h ( z k ) are the Jacobians (at the kth iteration) of the vector functions g ( x ) , Z(z) and h ( x ) , respectively; and Ahmin - hmin - h ( x k ) , Ah = h - h ( x k ) , A x m i n - xmin xk, &ax = %ax - x k * (3)
Maximizex
where the dual function is
q(X)
q(A)
(5)
(6)
It is worth noting that Jl is a matrix that models equations in which variables from different systems are involved. Using some algebra, the problem can be formulated
V F x subject to J ~ = x bl Ax = b 220
min
(4)
A dual maximization algorithm is better suited to this particular problem because it can exploit the separation structure of the dual objective. It is neccesary to implement an iterative algorithm to update the multipliers As. A subgradient-based dual maximization algorithm has been implemented [SI,
A(+1) = A() + p C 4
Ji
I1 JI I1
where z vector includes the slacks variables and A is a matrix with the system restrictions. The linear formulation presented above is only an approximation to the original problem; the validity of
(7)
Given a A(), the problem (7) can be decomposed into N subproblems, where N is the number of the different areas interconnected (for the sake of clarity, the
683
[Vp + X()TJi]
The LR algorithm works is as follows: Step 1: Initidizate de dual vector A() Step 2: Solve each independent subproblem ; Step 3: Update the dual vector A(); Step 4: If JIX(+l) - X()II 2.
I54
1
3 4 Nwdm01 SLP-OPF
5 E,
3.2
This subsection-deals with the solution of the problem (9). It is solved through an infeasible primal-dual interior-point algorithm. The dual problem of (9) can be formulated as
(12) can be obtained by applying a one-step Newtons Method. Starting with any initial point that satisfies x > 0, z > 0 and p > 0, equations (12) are solved. After updating variables, a new problem of is solved with a reduced value of p.
where z is the dual slack vector. Eliminating the above nonnegativity constraints by incorporating them in a logarithmic barrier term that is appended to the objective function, the Lagrangian for (10) is defined as
where the barrier parameter p > 0 is decreased to zero as the algorithm iterations progress. An optimal solution of (10) is expressed in terms of a stationary point of the Lagrange function. Therefore, an optimal solution must satisfy the KKT first-order necessary conditions
where X = diug(x1,...,xn), 2 = diag(z1, ...,2 , ) and e = (1,1, ...,1). An aproximate solution to equations
We have tested the two approaches presented above, using a small system (ll-bus system) . To obtain the numerical results, a prototype has been coded in MATLAB [9]. the evolution of the objective function through the iterations of the SLP algortihm is shown in Fig. (1). The same cost evolution is obtained using both approaches; however,the number of flops and CPU time requierements are substantially differents. Notice that, once the problem (2) is linearized, the One-step IP approach solves a problem that has a matrix restriction with the structure shown in figure (2). On the other hand, once the problem (2) is linearized, the LR-IP approach decompose the original problem into smaller systems, one for each interconnected area (2 areas in this case). Thus, smaller systems are solved; however, these are solved more times. In our tests, for the small systems, the number of iterations of the Lagrangian-Relaxation loop ranges from 7 to 12. In this case, the structure of the matrix restriction of one of the interconnected systems is shown in Fig. (3). The number of flops requiered by the LR-IP approach is more than 10 times the ones requiered by the one-step IP approach.
684
Conclusions
10
20
30
40
50
80
70
...e
5
.. . . . . . . . ..,, . . . . . . . . } ..... .
..
We have presented a computational comparison between two approaches that solve the Mult-area Optimal Power-Flow (MOPF) problem. At the time of writting this paper, the implementation served the purpose of testing the overall performance of the "Lagrangian-Relaxation" algorithm under two new features: an AC model for the network and the combination of the LR technique with an Interior-Point algorithm to solve each independent-area subproblem. Results derived from tests, using small systems, show that the "one-step IP" approach requieres less number of flops to obtain the solution (that is, less execution time); however, the "LR-IP" approach, although requires more CPU time, seems to be an appropiate solution to solve the MOPF problem where a decentralized implementation is requiered. Preparations for testing large scale power systems are under way and distributed implementation of the "LRIP" approach is a task for future work.
. e . . .
-..
e . . .
e .
e . .e..
..... . .. .
References
[l] Yan X and Quintana V. Improving an interiorPoint Algorithm for Security-Constrained economic Dispatch. IEEE Trans. on Power Systems, Vol. 12, NO. 2, August 1997, pp. 803-810.
3 5
0 5 1 0 1 5
..
2 0 2 5 3 0 3 5
.
*
4 0
[2] B. Kim, R. Baldick. Coarse-grained distributed optimal power flow. IEEE Trans. on Power Systems, Vol. 12, No. 2, pp. 932-939, May 1997. [3] Conejo A., Aguado J. Multiarea Coordinated Decentralized DC Optimal Power Flow. Accepted IEEE Trans. on Power Systems.
[4] B. Sttot,O. Alsac, and A.J. Monticelli,
[5] I. J. Lustig, R.E. Marsten and D.F. Shanno. Interior Points for Linear Programming: Computational State of the Art. ORSA J. on Computing G, 1994, pp. 1-14. [6] R.J. Vanderbei. Interior Points Methods: Algorithms and Formulations ORSA J. on Computing 6, V O 6 ~, NO 1, 1994, pp. 32-34. [7] D. Bertsekaa. Nonlinear programming. Athhenea Scientific, Belmont, Mass, 1995
AREA 1
Figure 4 11 Bus System
AREA2
[8] B.T. Polyak. Introduction to Optimization. Optimization Softtware Inc. New York, 1987. [9] The Math Works Inc; Natick, Massachusetts MATLAB 1993.