Parallel Computation of Large-Scale Dynamic Optimal Power Flow Problems

In this paper, a novel approach is presented to compute large Dynamic Optimal Power Flow (DOPF) problems in a distributed computing architecture, referred to as a Smart Grid Communication Middleware (SGCM) system. The time horizon is split into shorter time intervals using the SGCM system to solve one iteration of the Primal Dual Interior Point Method (PDIPM) for each sub-problem in parallel. The sub-problems are therefore not solved for optimality, but exchange boundary variables after each iteration. The algorithm was improved by introducing three different stages where the variable exchange is handled differently. The sub-problems eventually converge to a solution near to the solution of the entire problem without solving the global Karush-Kuhn-Tucker (KKT)-conditions. The methodology was tested on the German transmission grid, where the computational effort was reduced significantly. With an entire horizon of 96 time steps and a decomposition into 8 sub-problems, the run-time was decreased from over 2 hours to below 10 minutes with an overall system cost increase of only 1%, which means the DOPF model is suitable for real-time applications. Keywords—Distributed computing, dynamic optimal power flow, energy storage, model predictive control, optimization, smart grids.