Data-driven Probabilistic Constraint Elimination for Accelerated Optimal Power Flow

Optimal power flow calculations are run by operators to determine the cheapest operation of electrical grids given a particular system loading. Considering the increasing uncertainty in demand and availability of renewables, there is a need to develop faster solutions which can be run quickly for a variety of loading scenarios. The non-linear constraints in the optimal power flow problem increase computational costs. However, in reality, many of the constraints are non-binding and do not affect the optimum of the problem. In this paper, we propose a data-driven method for determining constraints that may be excluded from the formulation. The framework combines a latin-hypercube sampling method for generating training scenarios, with maximum likelihood estimation of Gamma distributions. We demonstrate that the method can reduce the solution time of a 24k network from 85.4 seconds to 11.1 seconds.