Research Based on Improved CNN-SVM Fault Diagnosis of V2G Charging Pile

With the increasing number of electric vehicles, V2G (vehicle to grid) charging piles which can realize two-way flow vehicle and electricity have been put into market on a large scale, fault maintenance has gradually become problem. Aiming at problems that convolutional neural networks (CNN) are easy overfit low localization accuracy in diagnosis piles, an improved classification model based (CNN-SVM) is proposed. Firstly, hardware adaptation optimization carried out for CNN structure, wavelet packet transformation used extract current signal feature information CNN, CNN-SVM constructed by SVM (Support Vector Machine) instead SoftMax classifier CNN. The PSO (particle swarm algorithm) optimize parameters obtain optimal model. Finally, superiority proposed method verified multi-working cases. experimental results show far higher than traditional deep learning network practical significance switch module pile.