Modified PSO-based Artificial Neural Network for Power Electronic Devices Fault Diagnosis Modeling

摘要

Particle swarm optimization (PSO) algorithm has been proven to be effective for neural network (NN) training. But in some cases, standard PSO converges prematurely without finding global optimum. In this paper a modified PSO (MPSO) is introduced to address the issue of premature. A novel search mechanism is proposed and the diversity of the population is controlled then. That is, the exploration and exploitation of search space are increased, resulting in avoiding premature convergence. 294 fault states of 12-pulse waveform controlled rectifier circuit are studied. A three-layer NN is employed to construct a fault mapping and MPSO is used as training algorithm. Simulation and experiment study demonstrate that the proposed technique is effective with high fault identification rate. It is suitable for the fault diagnosis of complex power electronic devices.