Multi-Parameter Fault Prediction in Electric Vehicle Charging Using a CNN-BiGRU Hybrid Deep Learning Model with Sliding Window Residual Analysis

Abstract This study proposes a hybrid deep learning model that combines Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU) to address the problem of local abnormal features being easily ignored and high false alarm rates caused by noise interference during the charging process of electric vehicles. Sliding window residual analysis is introduced to achieve adaptive alarms. The model utilizes CNN to extract local temporal features such as voltage drops and current spikes, captures temporal dependencies bidirectionally through BiGRU, and suppresses noise interference based on a joint threshold of residual mean and standard deviation. The experiment is validated using a self built self-constructed dataset and a public CALCE dataset. The results show that the root mean square errors of voltage, current, and temperature on the self-constructed dataset are 0.0083 V, 0.0216 A, and 0.0771℃, respectively, with a Mean Absolute Percentage Error (MAPE) of less than 0.9%. On the CALCE dataset, the corresponding error is 0.0094 V, 0.0257 A, 0.0892℃, and MAPE is below 0.9%. The sliding window warning mechanism has a false alarm rate of 1.2% under normal conditions and 2.3% under low signal-to-noise ratio conditions, and can identify anomalies up to 1.3 seconds in advance. This method effectively improves the timeliness and reliability of early warning by integrating local feature extraction and bidirectional temporal modeling, providing a robust solution for charging safety monitoring that can generalize across datasets.