High Performance State Estimation Algorithm Of Power System Based On GPU Parallel Acceleration

Nowadays,with the development of integration of transmission-distribution network and large-scale access to distributed energy resourse in power systems,the dimensions of power system analysis are getting higher and higher.Traditional calculation methods are difficult to meet future real-time calculation requirements.As the most basic and important part in modern energy management system(EMS),state estimation must guarantee its real-time performance.Therefore,based on the parallel acceleration technology of modern Graphic Processing Unit(GPU),this paper focuses on the high performance state estimation method of large-scale power systems.First,this paper briefly introduces the research background and current research status of bad data identification and state estimation technology.Also,it introduces the characteristics of GPU general computing architecture and explains the basic outline of the research.Subsequently,based on the theory of Maximum Normal Measurement Rate(MNMR),a bad data identification method using GPU parallel acceleration was proposed.The algorithm's coarse and fine granularity parallel acceleration strategy was designed to provide good data support for subsequent state estimation.Then,according to the characteristics of large-scale expression evaluation in power flow and state estimation,an expression parallel evaluation method based on Optimal Parallel Expression Forest(OPEF)is proposed.It demonstrates the definition,generation and heterogeneous computing of Expression Forest(EF)and a complete introduction to the entire theory.Finally,based on the method of weighted least squares(WLS)state estimation using Newton's method,the main time-consuming components of the iteration were extracted from the solution steps.The GPU parallel acceleration strategies were designed for each step to achieve the Multi-stage parallel accelerated WLS state estimation.The case analysis shows that the GPU can be accelerated in parallel at all steps of the proposed algorithm.The accelerated algorithm has a short calculation time and high acceleration efficiency,which can meet the real-time operation requirements.The memory usage of the algorithm have a smaller peak value and can be calculated on an ordinary PC.