Power system dynamic simulation using system partitioning and multiple stepsize techniques

The main objective of the research was to study the feasibility of using time domain techniques for on-line Dynamic Security Assessment (DSA) for power system. The computation time for simulating the system dynamic process after a disturbance, even for a few seconds, is quite large. This has been the main impediment to the use of such computer programs for on-line purposes. This research concentrated on speeding up the dynamic simulation by investigating the inherent dynamic characteristic of the power system together with parallel algorithms for a specific hardware architecture.; Both the sequential speedup and parallel speedup of different algorithms were investigated in this research. A partitioned Very Dishonest Newton (VDHN) algorithm that was implemented on a DEC VAX computer can produce a speedup of about 2 due to the use of system partitioning and variable time steps. A parallel algorithm, the Waveform Relaxation (WR) Gauss-Jacobi, was implemented on the Sequent Symmetry S81 shared memory machine. Four power systems with different sizes were tested. An overall speedup of over 27 was achieved for the 2500-bus system by combining the multiple stepsize scheme and concurrent processing, on 20 processors. The actual parallel efficiency factors were 35.43% and 43.52% for the 662-bus and 970-bus systems on 12 CPUs, 47.95% for the 2500-bus system on 20 CPUs, and the overall speedups of more than the numbers of CPUs are a result of the sequential speedup from partitioning and multiple stepsizes.; This is the first time that the WR algorithms has been implemented on a real parallel computer for power system analysis. Its performance is presented and analyzed with plenty of results from numerous simulations on the symmetry parallel computer. It is concluded that, by exploiting the physical properties of power systems combined with algorithm features and implementation techniques, stability computation can be made significantly faster in both sequential and parallel environments. This will assist its on-line application.