A multi-task neural network with hybrid feature selection for predicting electric vehicle fast charging durations

This study addresses the limitations of existing electric vehicle (EV) charging duration prediction models, which suffer from unclear influencing factors and fail to account for nonlinear charging behavior, leading to inaccurate estimates and inefficiencies. We propose a Multi-Task Neural Network (MT-NN) model designed to predict EV fast charging durations across two distinct state of charge (SOC) ranges, leveraging both time-series and static features. To identify the key features affecting fast charging, we combine Recursive Feature Elimination with Cross-Validation (RFECV) and a Genetic Algorithm (GA) to obtain the optimal feature combination. The MT-NN model integrated with RFECV-GA, outperforms traditional machine learning and advanced deep learning models, achieving lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Robustness evaluations and inference-time analysis further demonstrate that the model maintains stable performance under varying conditions while providing efficient runtimes suitable for real-world deployment. Moreover, Shapley Additive Explanations (SHAP) analysis provides insights into the contribution of key features, revealing how they influence charging duration.