GPU Based Parallel Algorithm Of Exponential Integration Method For Electromagnetic Transient Simulation

The electromagnetic transient simulation with detailed dynamic modeling can accurately describe the fast dynamics of the studied system in microsecond level.Therefore,it is with great significance in the analysis and operation of power system.Meanwhile,the traditional electromagnetic transient simulation is greatly challenged in speed and accuracy considering the continuous expansion of the system scale,the largescale integration of renewable energy and the trend of the power electronic integration.Thus,a CPU/GPU hybrid parallel algorithm of exponential integration method was proposed in this thesis for electromagnetic transient simulation,which made full use of the parallel computing resources of GPU.It is applied to dynamical systems with continuous nonlinear features.A novel accelerating strategy for the simulation of power electronic converters were proposed.The main contributions are as follows:1)The high degree of data parallelism of matrix exponential integration algorithm was analyzed with GPU parallel computing resources.A CPU/GPU based parallel algorithm of exponential integration method for electromagnetic transient simulation was proposed.Large-scale matrix operations required by exponential integration method are proceeded by GPU while the complex part such as system status calculation and update remain in CPU.Simulation efficiency has been improved significantly.2)The accelerated strategy based on exponential integration algorithm was proposed for power system with power electronic converters.It includes a matrix storage method without resolving ? function,adaptive interpolation algorithms considering multiple switching events,and GPU hybrid parallel algorithms.The proposed strategy provides new ideas for solving the new demand of power electronic power system.3)The parallel algorithm was tested on a linear distribution network and a high dimensional nonlinear wind farm respectively.The correctness and efficiency were verified.It showed that the acceleration of the parallel algorithm of GPU algorithm will be more obvious with increasing system scale.In addition,testing on a power electronic power system showed that the proposed acceleration strategies for power electronic circuits can improve the simulation efficiency.