Fault Early Warning of Wind Turbine Generator based on Residual Autoencoder Network

Abstract

The condition monitoring and fault Early Warning of wind turbine can find its faults early and reduce its failure rate and maintenance cost. This paper presents a fault diagnosis method of wind turbine generator based on residual autoencoder network (RAE). The proposed RAE has an autoencoder network structure. The encoding network is responsible for extracting the feature vector reflecting the distribution law of supervisory control and data acquisition (SCADA) data, the decoding network is responsible for reconstructing SCADA data according to the feature vector, and training the RAE network according to the reconstruction error of input data and reconstructed data. There are several shortcut connections between the corresponding layers of the encoder and decoder of the RAE. Shortcut connections introduce the shallow features in the encoder into the decoder and combines them with the deep semantic features in the decoder. Moreover, the shortcut connections allow the network to get additional supervision during back propagation process, avoiding the problem of gradient disappearance. Through the simulation analysis of the recorded data before and after generator fault, the effectiveness of the proposed RAE network for wind turbine generator fault diagnosis is verified.