A Distribution Network State Estimation Method With Non-Gaussian Noise Based on Parallel Particle Filter

The particle filter (PF) algorithm is a powerful method for tackling non-Gaussian noise interference in distribution network state measurement. However, this algorithm suffers from slow solving speed and lengthy calculation time. To overcome this, a state estimation method based on parallel particle filter (PPF) is proposed, which leverages the independent computation features of each particle in the PF model to improve computational efficiency. This study utilizes the parallel architecture of Compute Unified Device Architecture (CUDA) and General Purpose Graphics Processing Units (GPGPU) to establish a one-to-one correspondence between particles and computing threads. An improved rejecting-resampling method is introduced to solve the problem of low execution efficiency caused by unmerged access to GPGPU memory. In addition, according to the relationship between the particle number and estimation accuracy of state variable of the PPF, the optimal particle number suitable for parallel computation is solved. Ultimately, the simulation results indicate that the proposed method can be used to effectively filter the non-Gaussian-colored noises from the collected data, which meets the requirements of the distribution network state estimation for the accuracy and real-time performance.