Power System Transient Simulation Technology Based On Heterogeneous Parallel Computing Of Multi-core CPU And GPU

To ensure the safe and stable operation of the power system,electromechanical transient simulation is one of the essential tools.Restricted by software and hardware conditions, the existing transientsimulation programs usually adopt the off-line calculation mode based oncontingency analysis. With the scale enlargement of the power system andthe massive access of volatile new energy in the future, the need of on-linesimulation becomes more and more urgent, thus improving the speed oftransient simulation has become a top priority.This paper, based on the emerging multi-core CPU and GPU parallelcomputing technology, analyzes the calculation process of power systemtransient stability simulation, and researches the parallel solvers of thelarge scale linear systems which account for most of the calculationamount. Combined OpenMP and CUDA programing technology, thepower system transient simulation technology based on heterogeneouscomputing of multi-core CPU and GPU is achieved. Specific content isdescribed as follows:Research on the method and process of the power system transientstability simulation, and learn the programing tools for the parallelcomputing of multi-core CPU and GPU. On the basis of the above, adoptOpenMP and CUDA technology the write a simple but full-functionaltransient simulation program. Considering the solving process oflarge-scale linear systems accounts for most of the calculation amount ofthe transient stability simulation, the transient simulation programmodulated the solving process of linear system, in order to adopt variouslinear system solvers to evaluate and compare the parallelization effect.This paper also analyses the hardware and application characteristicsof multi-core CPU and GPU, and researches a direct solver based onnetwork partition and iterative solvers which are suitable for parallel computing. For the iterative solvers, preconditioners are adopted to reducethe condition number and improve convergence speed. According to thesparse characteristic of the linear systems in the process of transientsimulation, sparse technology is adopted to reduce the computation amountand memory occupation.This paper also tests the proposed power system transient stabilityprograms by using a series of cases to evaluate the correctness,parallelization effect and compare different linear system solvers. Testsprove that the simulation results of the proposed GPU algorithm aresimilar to the results of PSAT software. Tests also show that, when thesystem is complicated enough, the speedup ratio of GPU parallel algorithmis significant compared with the CPU serial algorithm, which is3.3in ourtests.