Implementation of the fast decoupled load flow on a vector computer

Abstract

The authors present a method which improves the application of supercomputer technology to power systems problems. The approach explored compensates for the short vectors that occur in power systems calculations by modifying the technique used for partitioning a matrix. The concept of indistinguishable nodes is extended to the factorization path graph and is employed in the partition of a matrix with no fill-ins. The general idea is to increase the size of the vectors without increasing the number of operations. The authors demonstrate the performance of the proposed method by incorporating it into a fast decoupled load flow program running on a CRAY X/MP-48 supercomputer. Significant improvements over results using conventional computers were obtained. With the more than 22 MFLOPS (million floating-point operations per second) for the entire process using standard FORTRAN language. Claims are substantiated by experiments using test data from several configurations of actual power systems. >