Power metering system is a key component of power grid operation, and its failure may lead to inaccurate metering or even grid fault expansion. In this paper, a fault detection and diagnosis method for power metering system based on adaptive filtering algorithm and neural network is proposed. Firstly, the EMD-NLMS adaptive filtering model is constructed by combining the empirical mode decomposition (EMD) and normalized least mean square (NLMS) algorithms to extract the fault feature signals; then, the particle swarm optimization (PSO) is used to improve the BP neural network to construct the fault diagnostic model and improve the fault identification accuracy. Simulation results show that the EMD-NLMS algorithm can effectively decompose the fault signal, filter out the noise interference, and extract more detailed IMF components; detection experiments for nine common fault types show that the convergence error of the improved PSO-BP neural network can reach 0.001 within 4000 iterations, which is three times faster than the convergence speed of the traditional BP network, and the accuracy of fault judgment reaches 97.5%. The established 5-10-5 structural neural network can accurately identify primary side short circuit faults with a diagnostic error of only 8.47 × 10-6 in the power metering system fault diagnostic test. The results of the study prove that the proposed method combining adaptive filtering and improved neural network has high accuracy and practicability in fault detection and diagnosis of power metering system, which provides effective technical support for the safe operation of power system.
