Study On Electromechanical Transient Simulation Of Power System Based On GPU Acceleration

With the rapid construction of China's ultra-high voltage backbone network,how to deal with the stability of the ultra-large interconnected grid is a technical problem that we urgently need to solve.Electromechanical transient time simulation(ETTS)is an effective method to analyze the stability of power system,and this method can directly reflect the stability of the system,but the computation speeds of the traditional method is slow.When dealing with batch simulation in multiple scenarios,the power consumption of multi-machine cluster system is large and the acceleration effect is limited by the number of processors.This thesis accelerates batch ETTS by Graphics Processing Unit(GPU)to accelerate the speed.The specific research contents are as follows:Firstly,according to the principle and process of implicit integration alternating solution method,this thesis studied the overall parallelization scheme of batch ETTS on CPU+GPU Hybrid platform.The solving process of dynamic element and the concrete steps of the complex network equation are extended into real equation are stated in detail.Next,the detailed design of the GPU kernel functions for solution of dynamic element injection current,batch sparse linear systems(SLS)up-looking LU factorization and batch sparse triangular equations solver is presented.Through data-level fine-grained parallelism and increasing the efficiency of memory access by combed design and other optimization strategies,for a system with more than 4000 generators,the average time of injection current of single ETTS is reduced to 0.05 ms.Aim at solving the large number of SLS with similar sparse structure in batch ETTS,a domino strategy is adopted excavate the cross-layer parallelism in the elimination tree.On a single K40 C GPU,the average time of single LU and sparse triangular solver can achieve about 200 and 35 times speedup respectively compared with KLU in serial.Then,the data transmission and algorithm flow are optimized and the detailed task allocation strategy on CPU+GPU Hybrid platform was designed according to the characteristics of software and hardware architecture,and the parallel mode on multi-GPU was proposed.Finally,the performance test of the above GPU algorithm is carried out with 5 power system examples.The results show that the algorithm of this thesis can limit the time in one alternate iteration of single ETTS within 0.5 ms for one system with over 10,000 nodes.