ECO-TrainNet: An Integrated Framework for Carbon Emission Prediction and Energy-Efficient Control in Urban Rail Transit

Abstract

Urban rail transit systems are a key component of sustainable urban mobility, yet their increasing operational intensity leads to growing energy consumption and associated carbon emissions. This study proposes ECO-TrainNet, an integrated framework for carbon emission prediction and energy-efficient control in urban rail transit. The framework combines machine learning-based prediction with reinforcement learning-based control in a closed-loop structure. A simulation-based dataset is constructed to capture train dynamics, infrastructure characteristics, and time-varying carbon emission factors. Ensemble learning models, particularly XGBoost, are employed to estimate energy consumption and carbon emissions, while a reinforcement learning agent is trained to optimize acceleration strategies under operational constraints. Experimental results show that the proposed framework achieves high prediction accuracy and enables effective control optimization. Compared with baseline operation, the reinforcement learning-based strategy reduces energy consumption by 11.7% and carbon emissions by 11.0%, while maintaining acceptable schedule deviation. The results demonstrate that integrating prediction and control provides a practical approach for carbon-aware and energy-efficient rail transit operation.