In recent years, the rapid development of electric vehicles (EVs) has likewise led to the construction of EV charging infrastructure, so the research on charging reliability and safety of EV charging facilities has become a focus of attention. However, most of the data used in existing research are complete and sufficient simulation data, when faced with actual data, the prediction accuracy is often affected by insufficient or incomplete data. To solve these problems, a data-driven approach is used to achieve early warning of faults during the charging process of charging equipment. Firstly, feature selection is performed to select appropriate data features. Secondly, the order data is filtered, the dataset is constructed and normalized. Secondly, the dataset is divided into a training group and a test group, the training group is used for model training, and the test group is used to judge the advantages and disadvantages of model training. Then, the divided training group is augmented with generative adversarial networks (GAN) to expand the data size and form a sufficient amount of new data. Subsequently, the data are inputted into bi-directional long-short term memory (Bi-LSTM) and the initial parameters are optimized using particle swarm optimization (PSO). A number of trials are conducted to observe the results of the modeling tests. Finally, in comparison with other prediction models, it is verified that the GAN-PSO-Bi-LSTM model has higher prediction performance, which improves the fault prediction accuracy of charging piles.