Multiple neural network fusion method for fault diagnosis based on GA-DS theory
Le Zhang · Electronic Design Engineering · 2015
A improved method for neural network evidence formalism process based on information entropy was presented solve the shortage that the parameter definition relied on expert experience, where the parameter in evidence formalism process was defined automatically by GA method among training samples, with a high efficiency of parallel optimization and a low sensitivity of initial population, and the fusion efficiency of DS method was promoted. The improved method was verified in the multi-class bearing fault data processing, the evidence conflict in the fusion process was eliminated effectively, and the rate of fault diagnosis was improved remarkably in stability by adjusting the neural network evidence formalism process automatically.