关键词: 锂电池, 荷电状态, 卷积神经网络, 长短期记忆网络, 科尔莫戈洛夫-阿诺德网络

Abstract: During vehicle operation, the estimation of the State of Charge (SOC) heavily relies on current measurements. However, failures in current sensors can lead to missing data, thereby reducing the accuracy of SOC estimation. Therefore, an improved method is urgently needed to accurately estimate SOC even when current data is abnormal or missing. To address this issue, a data-driven approach based on a Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) networks-Kolmogorov-Arnold Network (KAN) architecture is proposed. This method does not rely on current data and can serve as an alternative SOC estimation scheme in the event of current sensor failure. The CNN-LSTM networks-KAN model leverages the feature extraction capabilities of CNNs, the time-series modeling strength of LSTMs, and the nonlinear decomposition ability of KANs to estimate SOC during vehicle operation. The model is validated using real-world driving datasets. The results show that the proposed method achieves a Mean Absolute Error (MAE) of 0.381, a Root Mean Square Error (RMSE) of 0.467, and a coefficient of determination R² of 0.980 between the predicted and actual SOC values. These findings indicate that the proposed method can effectively estimate vehicle SOC even in the case of current sensor failure.

Key words: lithium-ionbattery, State of Charge(SOC), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, Kolmogorov-Arnold Networks (KAN)