Identifying Redundant Constraints for AC OPF: The Challenges of Local Solutions, Relaxation Tightness, and Approximation Inaccuracy

To compute reliable and low-cost operating points, electric power system operators solve optimization problems that enforce inequality constraints such as limits on line flows, voltage magnitudes, and generator outputs. A common empirical observation regarding these constraints is that only a small fraction of them are binding (satisfied with equality) during operation. Furthermore, the same constraints tend to be binding across time periods. Recent research efforts have developed constraint screening algorithms that formalize this observation and allow for screening across operational conditions that are representative of longer time periods. These algorithms identify redundant constraints, i.e., constraints that can never be violated if other constraints are satisfied, by solving optimization problems for each constraint separately. This paper investigates how the choice of power flow formulation, represented either by the non-convex AC power flow, convex relaxations, or a linear DC approximation, impacts the results and the computational time of the screening method. This allows us to characterize the conservativeness of convex relaxations in constraint screening and assess the efficacy of the DC approximation in this context.