A Dynamic Warning Method for Electric Vehicle Charging Safety Based on CNN-BiGRU Hybrid Model

Electric vehicles (EVs) are prone to spontaneous combustion during charging, which can lead to safety accidents. Therefore, it is critical to accurately obtain the charging crises of EVs for timely fault identification and early warning. This paper proposes a hybrid convolutional neural networks (CNN) and bi-directional gated recurrent unit (BiGRU) dynamic early warning method for EV charging safety. The method combines CNN and BiGRU features to rapidly extract deep characteristics of EV charging data, establish charging safety prediction models, and train it with historical normal charging data. After training, real-time EV charging data is input for prediction to identify whether EV charging processes are irregular. Sliding windows are used with the residual analysis of the historical data forecast outcomes to generate the safety dynamic warning threshoThe energy rules. The experimental results demonstrated that the CNN-BiGRU model has a superior prediction effect and accuracy. With e_RMSE and e_MAPE as the evaluation criteria, the charging current is 0.2393 A and 0.1888%, the charging voltage is 0.3859 V and 0.084%, and the temperature is 0.0543°C and 0.1658%; The charging current, voltage and temperature data can be used for early fault warning, which can be advance by 20.7 s, 20.2 s and 17.7 5 s, respectively.