Electric Vehicles Charging Time Prediction Based on Multimodel Fusion

Abstract With the rapid development of the global electric vehicle (EV) market, accurately predicting charging times is of significant importance for promoting the widespread adoption of EVs and enhancing the efficiency of charging infrastructure. Existing prediction methods often disregard battery aging and predominantly use single-model approaches, resulting in limited predictive accuracy. This article proposes a multimodel fusion-based method for predicting EV charging times. The approach utilizes data from ten EVs across various regions and operational conditions. Driving segment data are used to identify the ohmic internal resistance of the equivalent circuit model as a battery health indicator, employing the forgetting factor recursive least squares method. Key features such as state of charge, current, and ambient temperature are also extracted. Initial charging time predictions are generated using XGBoost, LightGBM, and CatBoost models and are subsequently fused using a random forest model to improve accuracy and robustness. Experimental results demonstrate that the proposed method achieves superior prediction performance under both fast and slow charging strategies, with a root mean square error of 0.130 h and a mean absolute percentage error of 5.676%. This research introduces a robust approach for enhancing the accuracy of EV charging time predictions.