Inverter Fault Diagnosis Based on Optimized BP Neural Network

Abstract

The inverter fault diagnosis based on BP neural network can fall into local minimum and overfitting. To solve these problems, we propose a fault diagnosis method based on BP neural network optimized by cross entropy and L2 regularization. In this proposed method, the quadratic cost function is replaced by the cross entropy cost function, which avoids the influence of the partial derivative of the activation function. L2 regularization is used to adjust network toward the small weight distribution. This method reduces the possibility of falling into local minimum and overfitting. The experimental results show that the optimized neural network can improve the accuracy of inverter fault diagnosis.