Multi-objectives ML-based online MPC for EV charging: A case study of a grid-connected large-scale charging infrastructure supported by PV and battery storage system

All-Electric Vehicles produce zero emissions during operation, but achieving net-zero transportation requires recharging infrastructure powered exclusively by Renewable Energies (REs). Given financial constraints, a common compromise is to integrate Photovoltaic (PV) systems and battery storage into large-scale, grid-connected charging infrastructures which require proper control strategies. This paper presents a multi-objective Machine Learning (ML)-based Model Predictive Control (MPC) system to manage energy flow in such infrastructures, addressing conflicting goals of maximizing PV use and reducing grid impacts. To enhance MPC performance, we incorporated ML predictions to handle uncertainties in EV charging and PV production. For PV power forecasting, we demonstrate that combining Singular Spectrum Analysis with Light Gradient-Boosting Machine achieves the highest accuracy. This hybrid approach reduces the Mean Squared Error by 92% compared to XGBoost. The MPC results show that an Achievement Scalarization Function (ASF) compromise provides the best trade-off among four solutions: minimizing grid impacts, maximizing PV utilization, and two compromise approaches. Compared to Rule-Based Control, the ASF strategy reduces peak overloading, Voltage Unbalance Factor, Total Demand Distortion of the current, and line losses by 66%, 48%, 56% and 35%. The model’s performance was validated through one-day and one-week simulations, confirming its effectiveness.