Combining direct and inverse factors for solving sparse network equations in parallel

摘要

A new parallel algorithm for solving the forward and back substitution part of the solution of sparse network equations for power systems is proposed. The approach assumes that the network equations are solved by LDU factorization and sparse vector techniques. The parallelization approach is based on the factorization path tree and its main feature is the switching from direct factor to inverse factors to best exploit the parallelism during the solution. The algorithm can be easily mapped on a multiprocessor architecture that exhibits both global and local memories. An illustrative example and validation results showing the effectiveness of the algorithm are also presented.