Fault Diagnosis of Gas Turbine Based on Improved BP Neural Network with the Combination of N-W and L-M Algorithm

Abstract

In order to solve the problem that the training speed of traditional BP neural network is slow in the process of gas turbine fault diagnosis, a new fault diagnosis method based on a combination of Nguyen-Widrow method and L-M optimized BP algorithm was proposed. The Nguyen-Widrow method is used to initialize the weights and thresholds of neurons in the BP neural network, and the L-M algorithm is used to improve the search space of the BP neural network, which reduces the times of network training and accelerates the learning speed of the network. The gradient descent method, the conjugate gradient method and the N-W and L-M combination optimization methods are respectively applied to the fault diagnosis of gas turbine. The results show that the BP neural network model optimized by combining N-W and L-M has faster learning speed and higher diagnostic efficiency for gas turbine fault diagnosis.