A message-passing distributed-memory Newton-GMRES parallel power flow algorithm

This paper presents a parallel Newton-GMRES (generalised minimal residual) power flow solution algorithm based on a message-passing distributed-memory multiprocessor architecture such as a cluster of workstations. The results show that the new algorithm can achieve good performance for two large-scale power system cases on a small cluster of GNU/Linux single-processor workstations. The workstations are connected via 1.0 Gbit/s Ethernet. The comparison between iterative methods and direct methods for the parallel solution of large sparse linear systems is also presented. It is shown that the parallel GMRES method is more scalable than the parallel direct method in a low latency network. Based on the presented parallel iterative linear solver, it is possible to exploit any parallelism in the Newton power flow solution process. For a workstation cluster on 1.0 Gbit/s Ethernet, which has high latency, the speedup appears to saturate. Nevertheless, when a low latency network, such as Myrinet, is available, the power flow algorithm with an iterative linear solver is expected to be more scalable than that with a direct linear solver.