Application of PSO-ELM in electronic system fault diagnosis

Abstract

Extreme Learning Machine (ELM) has many advantages, such as fast learning speed, good generalization performance and high diagnostic accuracy when it is applied in fault diagnosis, but its classification performance is affected by the two network random parameters-input weights and thresholds. Particle swarm optimization (PSO) algorithm has the characteristics of simple, easy to implement and found the local optimum quickly. This paper proposes the Particle Swarm Optimization algorithm (PSO) to optimize the two parameters and to obtain the electronics system fault diagnosis based on PSO-ELM. Two analog circuits, one is complex; the other is simple, are designed to obtain the original data in the essay. Then, wavelet transform and PCA are combined to extract the feature of samples information. Take the processed data into the PSO-ELM to get the diagnosis results, meanwhile, compared the optimal performance of PSO, glowworm swarm optimization (GSO)and bat algorithm (BA) to ELM, experiments show that PSO is the most efficient to improve the diagnostic accuracies of the two circuits, the results obtained with PSO-ELM reach to 98.89% and 99.46% respectively.