Research on Model Optimization Method Based on Differential Evolution

Abstract

Nowadays, industrial upgrading and transformation continues to deepen, based on the deep neural network model of the integration of the advantages of the application is more and more prominent. Rolling bearings can not be ignored as the existence of mechanical equipment. A convolutional neural network parameter optimization algorithm based on differential evolution is proposed. The loss function of the fault diagnosis model is taken as the objective function of the optimization algorithm, and the optimization model is mainly aimed at optimizing the training parameters such as the number of convolutional kernels, convolutional kernel size, and step size of the 3D convolutional neural network, and the optimal parameter setting values are derived. Compared with the pre-optimization fault diagnostic model, the prediction accuracy of the fault diagnostic model is improved by 0.35%, and the loss rate is decreased by 1.13%, and the obtained fault diagnostic model has higher diagnostic accuracy.