Accelerated sparse matrix-based computation of electromagnetic transients

Abstract

This paper is related to research on parallelization methods for the simulation of electromagnetic transients (EMTs). It presents an automatic parallelization approach based on the solution of sparse matrices resulting from the formulation of network equations. Modified-augmented-nodal analysis is used to formulate network equations. The selected sparse matrix solver is parallelized and adapted to improve performance by pivot validity testing and partial refactorization. Refactorization is needed when dealing with varying topology networks and nonlinear models. The EMT solver employs a fully iterative method for nonlinear functions. Conventional computer CPU-based parallelization is achieved and does not require any user intervention for given arbitrary network topologies. The presented approach is tested on real networks with complex models, including nonlinearities and power-electronics converters for wind generator applications.