Parallel computation of power system dynamics using multistep methods

Abstract

There is considerable interest in speeding up the dynamic simulation of power systems for on-line security assessment purposes. While software is being developed for automatically detecting parallelism in an otherwise serial code, research in developing parallel numerical algorithms is also useful. In this paper we develop a sequential algorithm that can be automatically converted (autotasked) to execute in parallel. We have also adapted an explicitly parallel algorithm for simulating the dynamics of a multimachine power system. Both were multistep predictor corrector algorithms and were implemented on a Cray 2 machine with four processors. Simulations were performed on four test systems (3, 10, 17 and 50 machines). While the speedups by both techniques are comparable, it is believed that continued research in numerical algorithms and their practical implementation is warranted. Autotasking may not always ensure the desirable spedups, since it may run into difficulties in certain instances.