REST Network: An Ensemble Deep Learning Approach for EV Charging Load Forecasting in Artificial Port Supply Chains

Electric vehicles (EVs) are increasingly used in logistics and port operations, making charging requirements prediction difficult. Energy supply management, infrastructural readiness, and daily operations depend on these estimates. Traditional forecasting algorithms can fail to address logistics’ complicated, variable charge patterns and uneven data. The ResNet-SENet-Transformer Ensemble (REST) Network, a novel forecasting model for EV charging demand prediction in artificial port supply chains, addresses these issues. Our model improves feature selection and corrects data imbalances using Texas data from January 2020 to June 2024 and sophisticated approaches including PSS and DSRS. REST outperforms standard methods with a MAPE of 6.5%, an RMSE of 9.8 kW, and strong results across several custom metrics, including Energy Efficiency Error (3.2%), Load Variability Deviation (4.1%), Cost of Prediction Error ($12.50), and Time-Aware Accuracy (92.3%). These findings demonstrate that REST may improve port EV fleet energy management and operational planning. This study prepares next-generation EV forecasting systems that are more flexible, dependable, and sustainable logistics-focused by reducing energy waste and enhancing responsiveness.