GPU-Based Accelerated Algorithms for the Power Flow Calculation

Power flow (PF) calculation is critical for power systems, as the development of multiple energy supplies. Power system modeling and analysis have been challenging on power engineers and leading to great pressure for the PF calculation. For the safety, stability and real-time response in grid operation, grid planning and analysis of the power system, it is urged to require designing high-performance computing methods, accelerating PF calculation, obtaining the voltage magnitude and phase angle of buses inside the power system, and coping with the increasingly complex large-scale power system. The PF algorithm is, generally, classified into the iterative and direct methods in the perspective of numerical methods. As for iterative method, a pre-conditioner is required to be designed to reduce the condition number of sparse Jacobian matrix toimprove convergency of the power system. Although the iterative method can save much memory and solve some large-scale sparse linear equations, the PF solver severely depends on the complicated pre-conditioner of the Jacobian matrix. Usually, the PF calculation cannot get a convergent solution without validating the pre-conditioner, repeatedly. For direct method, the traditional sequential, the Newton-Raphson (NR) algorithm will consume much of the computing resource and take a long time to converge on solving large-scale sparse linear equations of the power system. To address these issues, the GPU-based parallel computing architecture, singleGPU, and multi-GPUs, was proposed to take advantage of multi-thread, task parallelism and data parallelism, accelerating the PF calculation. Also, the utilization of GPUDirect technology enhances communication efficiency and significantly reduces data transmission overhead, leading to superior performance improvements compared to the traditional sequential methods.