Toward Multiperiod AC-Based Contingency Constrained Optimal Power Flow at Large Scale

This work presents a scaling study of a parallel nonlinear, nonconvex optimization approach applied to a multiperiod contingency constrained alternating current optimal power flow. We propose a “variable duplication” model for efficient parallelization of the numerical optimization on massively parallel HPC hardware. The model is expressed as a two-stage nonlinear programming problem, where the first stage captures time-dependent constraints and the second stage reflects the system changes in response to contingencies. The parallel interior-point optimization solver for nonlinear programming (PIPS-NLP) enables us to leverage the dual-block angular structure specific to the formulation by applying the Schur complement for efficient parallelization of the linear solves. The Julia modelling package StructJuMP, allows us to compactly and conveniently express the model's algebraic components. StructJuMP uses automatic differentiation to provide the first-and second-order derivatives to PIPS-NLP. Aiming at a strategy for computations at petascale, numerical experiments conducted on Theta, an Intel KNL-based system at the Argonne Leadership Computing Facility are presented.