Large-scale branch contingency analysis through master/slave parallel computing

Abstract

Contingency analysis (CA) requires fast execution time for real-time power system operations. Because CA problems can naturally be divided into separate subtasks, parallel computing helps to speed up the computation time. This paper proposes a master/slave parallel computing architecture and studies the computation of CA in a large-scale power system through high performance computing, adopting a message passing interface for implementation. In particular, although the execution time of CA varies, there is a tradeoff between having an imbalanced workload and “paying” a synchronization penalty for parallel computing: either factor blocks the progress of scalability. The proposed layered dynamic scheduling method is effective to tackle the challenge of high synchronization cost and workload imbalance and have the potential to further scale for the N − 2 contingency analysis.