Research on Fault Diagnosis Method of DC Charging Pile Based on Deep Learning

EasyChair Preprint 6783

Abstract

Aiming at the fault diagnosis of the charging module of the electric vehicle DC charging pile, a fault diagnosis method of the DC charging pile based on deep learning is proposed. First, through circuit simulation, the DC charging pile model is simulated under different faults and different working conditions, and the three input current signals are obtained as fault characteristic parameters. Perform three-layer wavelet packet decomposition and reconstruction of the fault characteristic parameters, calculate the frequency band energy spectrum data through the reconstruction coefficients, and normalize it. Finally, the fault characteristic set is composed of the fault data and the fault result, which is used as a deep neural network (DNN) Model input and verification. After the output layer of the constructed DNN model, a Softmax classifier is added to fine-tune the output fault characteristics and realize fault type recognition. Through the analysis of different types of faults of the charging module of the DC charging pile, the accuracy and effectiveness of the fault diagnosis method is verified, and its accuracy rate can reach more than 95.56%.

Keyphrases: DC Charging pile, deep learning, fault diagnosis, wavelet analysis

@booklet{EasyChair:6783,
  author    = {Jinfeng Du and An Wen and Miaoxian Zhou and Wenlei Mao and Guangshan Huang and Shangyun Deng},
  title     = {Research on Fault Diagnosis Method of DC Charging Pile Based on Deep Learning},
  howpublished = {EasyChair Preprint 6783},
  year      = {EasyChair, 2021}}