Implementation of the waveform relaxation algorithm on a shared memory computer for the transient stability problem

Parallel processing to obtain time domain simulation of power system dynamic behavior has been studied for over a decade. In this paper, a parallel across space implementation of a Waveform Relaxation (WR) based algorithm is presented. The algorithm can take full advantage of the inherent properties of power systems, such as coherency grouping and localized fault response, thus significantly speeding up the simulations. In the proposed implementation, a system is broken up into several subsystems based on its coherency, and then each processor is devoted to the computation of the waveform of each subsystem. This is the first time the WR algorithm has been implemented on a parallel machine (Sequent Symmetry S81) and tested on large power systems. The numerical results show very promising performance compared with the commonly used Very Dishonest Newton algorithm, with overall speedups of up to 27 obtained for a 2500-bus system using 20 processors. Keywords Waveform relaxation, parallel processing, dynamic simulation, network partitioning, multiple stepsizes.