The solution to an optimal power flow (OPF) problem provides a minimum cost operating point for an electric power system. The performance of OPF solution techniques strongly depends on the problem's feasible space. This paper presents an algorithm for provably computing the entire feasible spaces of small OPF problems to within a specified discretization tolerance. Specifically, the feasible space is computed by discretizing certain of the OPF problem's inequality constraints to obtain a set of power flow equations. All solutions to the power flow equations at each discretization point are obtained using the Numerical Polynomial Homotopy Continuation (NPHC) algorithm. To improve computational tractability, "bound tightening" and "grid pruning" algorithms use convex relaxations to eliminate the consideration of discretization points for which the power flow equations are provably infeasible. The proposed algorithm is used to generate the feasible spaces of two small test cases.
Index Terms-Optimal power flow, Feasible space, Convex optimization, Global solution
I. INTRODUCTIONO PTIMAL power flow (OPF) is one of the key problems in power system optimization. The OPF problem seeks an optimal operating point in terms of a specified objective function (e.g., minimizing generation cost, matching a desired voltage profile, etc.). Equality constraints are dictated by the network physics (i.e., the power flow equations) and inequality constraints are determined by engineering limits on, e.g., voltage magnitudes, line flows, and generator outputs.The OPF problem is non-convex due to the non-linear power flow equations, may have local optima [1], and is generally NP-Hard [2], [3], even for networks with tree topologies [4]. Since first being formulated by Carpentier in 1962 [5], a broad range of solution approaches have been applied to OPF problems, including successive quadratic programs, Lagrangian relaxation, heuristic optimization, and interior point methods [6], [7]. Many of these approaches are computationally tractable for large OPF problems. However, despite often finding global solutions [8], these approaches may fail to converge or converge to a local optimum [1], [9].Recently, there has been significant effort focused on convex relaxations of the OPF problem. These include relaxations based on semidefinite programming (SDP) [2], [10]-[16], second-order cone programming (SOCP) [17]-[20], and linear programming (LP) [21], [22]. In contrast to traditional approaches, convex relaxations provide a lower bound on the optimal objective value, can certify problem infeasibility, and, in many cases, provably yield the global optimum.The performance of both traditional algorithms and convex relaxations strongly depends on the OPF problem's feasible