Study On GPU-accelerated Power System Optimal Power Flow Algorithm

As the modern power system tends to develop in the direction of large-scale,diversification and complexity,the demand for safe and stable operation of the power system is also increasing.Optimal power flow(OPF)is an important analysis tool that considers power system safety and economy at the same time.It has been widely applied in economic dispatch,safe operation,power grid planning,reliability analysis,etc.Under the background of unified analysis for the entire grid,in order to meet the real-time computing requirements of power system dispatch,the development of an efficient and accurate optimal power flow calculation scheme has become an important research topic.In recent years,the graphics processing unit(GPU)has become important issues in high-performance computing problems due to its superior performance on floating-point computing and memory bandwidth.GPU has been successfully applied to scientific computing fields such as power system,and it makes largerscale and faster power system analysis and calculation applications become possible.In this thesis,optimal power flow algorithm is accelerated in parallel by GPU.The specific research contents are as follows:Firstly,this thesis studies optimal power flow parallel algorithm based on the primal dual interior point method.Considering that the solution of KKT system is the main calculation bottleneck,the powerful parallel computing capability of GPU is used to improve the efficiency of solving sparse linear equations.With the cross-layer parallel strategy for elimination tree,the GPU-accelerated sparse direct method is implemented.Compared with KLU solver and cu SOLVER library,the well-designed GPU-based algorithm shows better computing performance.Compared with CPU-based solution,GPU-based solution achieves the acceleration of 2.86 times in the classical optimal power flow calculation of 9241-bus system.Next,security constrained optimal power flow(SCOPF),that is expanded on the basic optimal power flow model,is discussed.This thesis explores the potential special structure of the KKT system through decoupling and reconstruction of the problem.A parallel SCOPF algorithm based on block bordered diagonal form(BBDF)model is studied.According to the respective computing characteristics of CPU and GPU,the overall flow of SCOPF algorithm is studied on heterogeneous computing platform.The task allocation strategies in hybrid architecture,such as process control,data transmission,are specifically optimized.The scheduling mechanism of multi-GPUs is elaborated.Furthermore,based on the process parallelization of the SCOPF algorithm,the GPU-accelerated block elimination method is studied,which optimizes the design of GPU kernel functions to obtain the fine-grained parallelism of computing processes.In the time-consuming local Schur Complement calculation,batch sparse triangular equations solver is designed by regularization modeling method,sparse matrix-matrix multiplication algorithm is designed by thread allocation strategy with efficient access to shared memory.In the solution of dense linear system,the GPU-based algorithm further reduces the computing time of this serial component,which improves the efficiency of the whole algorithm.Finally,the performance tests of the GPU-accelerated parallel SCOPF algorithm are implemented on several power systems of different scales.The test results show that,when solving the optimization problems of large-scale power system,the GPU-accelerated solution based on block elimination method performs higher computational efficiency.Taking the SCOPF problem of the 2869-bus system as an example,compared with the traditional CPUbased solution,the GPU-based solution achieves 21.77 times speed-up in solving the KKT system and 8.52 times speed-up in the whole process overall.Through the test results,the efficiency and applicability of the proposed method are verified.