GPU-Based Sparse Power Flow Studies With Modified Newton’s Method

Abstract

The Power system is getting larger and more complicated due to development of multiple energy supplies. Solving large-scale power flow equations efficiently plays an essential role in analysis of power system and optimizing their performance during normal or contingencies operation. The traditional Newton-Raphson (NR) algorithm used for power flow calculations is computationally expensive due to updating Jacobian matrix in each iteration. As alternative to update the Jacobian matrix repeatedly, this paper presents a GPU-based sparse modified Newton’s method by the introduction of a fixed Jacobian matrix, which integrates vectorization and parallelization technique to accelerate power flow calculations. Moreover, this research in the paper also investigates the performance of the corresponding CPU versions and a MATLAB-based library package, MATPOWER. The comparison of the results on several power system and power distribution systems demonstrate that the GPU variant is more reliable and faster for power flow calculation in large-scale power systems.

Keywords

• GPU
• CUDA
• modified Newton’s method
• compressed row storage (CRS)
• Jacobian matrix
• vectorization