Electric vehicle charging demand forecasting: A data-driven integrated learning approach

Amid the global climate change and energy crisis, the rapid development of EVs offers a viable solution to achieving zero emissions during the vehicle usage phase in the transportation sector. Accurate forecasting of EV charging demand is essential for optimizing grid management and enhancing the efficient utilization of renewable energy. This study introduces a data-driven integrated learning model based on high-dimensional anonymized data from EV charging stations in Jiangsu Province, China. The model incorporates tree-based ensemble methods, including RF, GBDT, and LightGBM, alongside the BiLSTM deep learning model. The model's hyperparameters are optimized using the DBO algorithm. Key factors considered in the prediction include the operational characteristics of the charging stations and associated time-series features. Experimental results demonstrate the superior performance of the proposed DBO-Stacking model in EV charging demand forecasting. On the Nanjing dataset, the model achieved improvements of 6.06 % and 12.57 % in MAE and RMSE, respectively, compared to the best single model, LightGBM. On the Wuxi dataset, the model outperformed the best single model, RF, by 8.27 % and 9.89 %, respectively. On the Suzhou dataset, it improved by 4.85 % and 1.34 %, respectively, compared to RF. In Nantong, the DBO-Stacking model demonstrated substantial improvements of 19.75 % and 19.62 % in comparison to the top single model, GBDT. Additionally, a detailed feature importance analysis using SHAP revealed that factors such as DC pile power and service fees play a critical role in predicting charging demand. This study provides valuable methodological support and decision-making insights for optimizing the layout and management of charging infrastructure.