Safety accidents that occur during the charging process of new energy vehicles are the key cause of new energy vehicles fires. Charging fault diagnosis research is crucial to improving the safety level of new energy vehicles. Based on the actual operation data of new energy vehicles and charging piles, this article studies the charging voltage fault diagnosis method of vehicle-pile data fusion to address the charging safety issues of new energy vehicles. In view of the difficulty of vehicle-pile data fusion caused by the large data structure differences and poor quality of the data sources of new energy vehicles and charging piles, preprocessing methods such as data decoding, data cleaning, and fragment division are designed, and a method for matching and fusing charging fragments with charging orders is proposed; in view of the difficulty of power battery voltage fault diagnosis and the problem of delayed diagnosis under charging conditions, the random forest algorithm is used to screen the parameters related to charging voltage from the vehicle-pile charging data, and a convolutional neural network model with an attention mechanism is constructed. The sliding window method is used to predict the charging voltage to achieve real-time monitoring of the charging status. The superiority of the model in this article is verified by comparing the prediction accuracy results of different models.