Batch Computing Method For Static Security Analysis And Sensitivity Analysis Of Power Grids Based On GPU

With the growing installed capacity of power system and the expanding scale of transmission network in China,the scale of network analysis and application calculation for the unified analysis of the entire network continues to increase.What’s more,the fault prediction is expected to be more comprehensive,which puts forward higher requirements for the real-time and accurate dispatching and control ability of the power grid.Online static security analysis and sensitivity analysis are methods essential for guaranteeing stable operation of power system.However,traditional calculation methods,such as those based on Central Processing Unit(CPU),cannot meet the requirements of these analyses for large-scale systems due to their low efficiency and poor applicability.Graphics processing unit(GPU)improves throughput through thousands of computing units and high-speed memory bandwidth,and has super floating-point computing power.Therefore,it is necessary to study the parallel computing methods of static security analysis and sensitivity analysis,and accelerate the methods according to the CPU-GPU hardware and software architecture and programming model,and realize the rapid and accurate static security analysis and sensitivity analysis of large-scale systems.The main research contents of this paper are as follows:First,the static security analysis method based on GPU acceleration is designed.Combined with the CPU-GPU hardware and software architecture and programming model,the computing resources are allocated rationally and the DC power flow method is adopted to scan the entire network for faults.According to the similar characteristics of the post-fault network topologies,a parallel computing strategy for the real-time fault scanning based on the data collection is proposed,and a fine-grained parallel algorithm for each level is designed.The method can generate line power in various scenarios in batches,and realize fast and real-time static security analysis of large power grids.Second,a batch calculation method for sensitivity analysis oriented to GPU acceleration is designed.Under the framework of CPU-GPU heterogeneous computing,the calculation procedure and characteristics of line outage distribution factor,generation shift distribution factor,and sensitivity matrix are analyzed.Through mining the finegrained parallelism among them,a GPU acceleration calculation method for commonly used sensitivity factors is proposed.The method contains the kernel function design of key steps,such as the generation of self-impedance and mutual impedance in batches,and the acceleration effect is verified through test of an actual power grid.Finally,the design and customization methods of network analysis application service software are proposed.In the multi-level hybrid parallel computing mode,the key steps of multi-GPU parallel are designed.Virtual container technology is used to achieve coarse-grained parallelism between multiple GPU boards for static security analysis and sensitivity analysis applications,and application service software with static security analysis and sensitivity analysis is designed.