Performance Studies for a Diakoptics-based Partitioning Approach for Power System Simulation

Performance studies of partitioned power system simulations on multicore computer are presented. A partitioning approach developed by the authors was used to speed up the simulation of a shipboard power system. The system was torn from its capacitor loops and simulated as subsystems on a quad-core computer. Several performance metrics are presented to assess the performance of this simulation approach: speed gain, accuracy, and core impact. Results from several case studies show the general trend of each performance metric and show that the run-time of large-scale simulations can reduce three orders of magnitude if all cores are used.