Acceleration of Sparse Matrix Solution Based on GLU

Optimal Power Flow (OPF) is an important analytical tool that considers both the safety and economy of the power system. Developing an efficient and accurate optimal power flow calculation scheme has become an important research topic in order to meet the demand for computing speed in real-time scenarios such as power system regulation and operation. In recent years, GPUs have become a research hotspot in high-performance computing due to their advantages in floating-point computing speed and storage bandwidth. They have also been successfully applied in the field of power systems, making larger scale and more efficient power system analysis and computing applications possible. This article optimizes the GPU based sparse matrix solver GLU to improve its speed in solving sparse matrices. Our contributions are as follows: 1) We optimized the symbol decomposition method in GLU and determined the symbol pattern of each column by depth first searching for non-zero data in each column, significantly improving the symbol decomposition speed of GLU; 2) We have utilized a large number of threads for parallel execution, improving the process of reassembling non-zero data in matrices based on GLU symbol decomposition, and accelerating the speed of data reassembly; 3) We have validated the effectiveness of our proposed optimization method on multiple publicly available power flow calculation data. The experimental results show that the proposed improvement method can significantly improve the speed of GLU solving.