Voltage Prediction of Electric Vehicle Based on Attention Mechanism CNN-GRU Model

Haiyang Feng, Weizhen Han, Haikuo Zhang, Lisha Yang, Meng Xu

Abstract: Accurate and timely voltage prediction can help battery management systems detect potential voltage faults in advance and take corresponding measures to ensure the safety of vehicles and passengers. This paper proposes a short-term voltage prediction method based on an attention mechanism CNN-GRU (AT-CNN-GRU). First, the voltage data undergoes convolution to capture spatial features, and then a gated recurrent unit with attention is used for temporal prediction. This method does not rely on auxiliary features and automatically assigns corresponding weights to the voltage at different time points. Experimental analysis shows that our model exhibits high accuracy and low error in voltage prediction. Compared to traditional LSTM, CNN-LSTM, and CNN-GRU models, our model significantly improves prediction accuracy. This indicates that the proposed model can effectively predict battery voltage, providing a solid foundation for subsequent fault diagnosis. In the future, we will continue to explore the integration of voltage prediction and fault diagnosis to take timely preventive measures before faults occur, ensuring the safety of electric vehicles and their passengers.