It is of great significance to make fault diagnosis for charging equipment of electric vehicle (EV) accurately and timely. The methods based on deep learning are promising because they can extract fault features and conduct fault diagnosis effectively. However, deep learning networks require a large number of labeled samples for model training, and in practical conditions, the labeled samples available of charging equipment are quite limited. To address these problems, we, inspired by semi-supervised learning, proposed a semi-supervised charging equipment fault diagnosis method combining Tri-training and deep belief network (DBN). The proposed method adopts Tri-training to enable full utilization of unlabeled data of charging equipment, and obtain a large amount of valid pseudo-labeled data, which will be used for the training of Tri-DBN model. Subsequently, the fault features of charging equipment are input into Tri-DBN, which is used for the classification and identification of charging equipment faults. The experimental results show that the proposed method effectively improves the faults classification accuracy of charging equipment and show more than 90 accuracy at all fault types. In addition, this method still performs well in the presence of less labeled data.
展开▼
