Fault Detection of Flexible DC Grid Based on Empirical Wavelet Transform and WOA-CNN

Abstract

Flexible DC grid solves the disadvantages of high line loss and small transmission capacity of traditional AC grid, but it still has the problems of difficult to extract characteristic signals and fault diagnosis. To solve this problem, a fault detection method based on empirical wavelet transform (EWT) with multiscale fuzzy entropy (MFE) and Whale algorithm optimization with convolutional neural network (WOA-CNN) is proposed. Firstly, EWT is used to decompose the fault line mode voltage signal and obtain the fault component. Then, the correlation coefficient of each component is calculated, and the components with more feature information are reconstructed. The MFE value of the reconstructed signal under different faults is calculated. Finally, the fault feature quantity is input into WOA-CNN for classification. A large number of experiments demonstrate that this method has strong anti-interference ability and high accuracy, and can reliably detect line fault under different fault types, fault positions and transition resistance conditions. Its accuracy is significantly improved comparing with CNN, PSO-CNN, K-means clustering, PSO-SVM and BP neural network, with an average of 99.5834%.