Efficient Improvements of Domain Decomposition Method for Power Grid Transient Simulation

Qian Chen,Meng Li,Hao Yan,Longxing Shi

DOI: https://doi.org/10.1109/ISEDA59274.2023.10218650

2023-01-01

Abstract:The efficiency of large-scale power grid transient simulation has been of great interest. Combined with distributed computing, the domain decomposition method can well tackle this problem. However, the large and dense Schur complement matrix constrains the scalability. Although previous studies reduce the size of the Schur complement matrix by using iterative methods, they become insufficiently stable when encountering asymmetric positive definite problems. In this paper, we propose three techniques to improve the domain decomposition method, including order reduction, pruning, and adaptive number of processes. Order reduction and pruning reduce the computation. Adaptive number of processes increases resource utilization. Combined with the LU factorization, our work can well solve the ill-conditioned problem. Experiments on IBM benchmarks with millions of nodes show that the proposed techniques are all effective. Compared with the traditional domain decomposition method, the performance is improved by 44.16% - 55.34%, and the consumed memory is reduced by 16.66% - 51.35%.