Electric Vehicle Remaining Driving Range Prediction via Gated Recurrent Units and Kolmogorov–Arnold Network Model

When driving electric vehicles, the remaining driving range shows different decreasing trends as the battery status and driving condition change. Accurately predicting the remaining driving range of the vehicle can alleviate the driver’s range anxiety and improve the driving experience. This study designs a hybrid model combining Gated Recurrent Units (GRU) and Kolmogorov-Arnold Network (KAN) to predict the remaining driving range. The model obtains 30 sets of sampling data as input from the recent driving process during which the battery state of charge decreases by 1%. Each set includes two types of data: battery status and driving environment. The GRU network is utilized to extract relevant temporal information, followed by the KAN network performing regression prediction to output the remaining driving range online. Comparing the prediction results of using GRU, Long Short-Term Memory Network (LSTM), LightGBM, XGBoost and KAN alone under the same conditions, the designed GRU-KAN model has the lowest mean absolute error. The results of this research can be applied to the battery management system to improve the intelligent management of electric vehicles.