Study On Sensitivity Algorithm Of Power System Based On GPU Acceleration

In recent years,with the rapid development of modern power systems,how to ensure the security and stability of power grids has become an urgent technical problem to be solved.Power system sensitivity analysis can obtain weak points in the grid to improve system safety performance.However,the traditional sensitivity calculation method is slow and cannot realize the function of rapid analysis of the power grid.In this paper,the parallel processing of batch sensitivity is realized by the Graphics Processing Unit(GPU),which effectively improves the calculation speed.The specific research contents are as follows:Firstly,the hardware architecture of GPU and CUDA computing model are introduced.The parallel computing mode of GPU is described according to its characteristics.The GPU kernel functions such as sparse matrix preprocessing,sparse linear systems(SLS)up-looking LU decomposition,and Back/ forward sweep are designed in detail.Then,according to the principle and flow of sensitivity analysis,the overall parallelization scheme of sensitivity calculation on CPU+GPU heterogeneous platform is studied.The specific steps of perturbation calculation principle,GPU batch disturbance and admittance matrix batch correction are studied.Explain in detail.Then,the batch sensitivity calculation algorithm and process are optimized.According to the work of CPU and GPU and the hardware and software architecture characteristics,the batch calculation principle of multi-type sensitivity is calculated by using multiple GPU cards,and different types of disturbances and power flows are calculated in parallel..Finally,the performance of the above GPU-accelerated sensitivity analysis algorithm was tested by five power system examples.The results show that the optimization of the batch sparse isomorphic equations on the single-block K40 C GPU is more than 72 times faster than Matpower algorithm by optimizing the combined memory access and fine-grained parallelism of the 9421 node grid matrix power flow calculation.At the same time,the power flow calculation time is controlled within 2.75 ms,which is 149 times faster than Matpower.