BOTTLENECKS IN PARALLEL ALGORITHMS FOR POWER-SYSTEM STABILITY ANALYSIS

Research in parallel algorithms for power system stability analysis started over a decade ago but has intensified recently. This experience with new parallel algorithms implemented on actual multiprocessors is starting to highlight the constraints and some clear choices are beginning to emerge. In this paper, Gauss, Newton and relaxed Newton type algorithms are compared showing the last to be the most promising. Different implementation strategies are discussed together with their advantages and disadvantages. One of the algorithms, the Maclaurin-Newton, and much of the implementation techniques, especially the Toroidal method, are presented here for the first time and are shown to be improvements. However, the thrust of the paper is to show the superiority of the relaxed Newton methods and the importance of implementation techniques, the new methods being mere examples to demonstrate this.