Machine Learning-Based Online MPC for Large-Scale Charging Infrastructure Management

The rapid expansion of Electric Vehicle (EV) charging capacities, driven by the surge in transportation electrification, has a direct impact on the electric power system, particularly the Low Voltage (LV) grid. Consequently, there is a pressing need to redesign energy management systems, placing a greater emphasis on mitigating the effects of EV charging. In this article, we crafted an online grid-level Model Predictive Control (MPC) incorporating Machine Learning (ML) prediction enhanced with Singular Spectrum Analysis (SSA). ML models predicts the future EV charging demands, which are the key factors for the MPC for managing efficiently a large-scale charging infrastructure, aiming to mitigate EV charging impacts, specifically phase unbalance and overloading. The MPC results, especially on phase unbalance metrics, mirrored in a predictive manner those of offline/no forecast optimization from our previous work. A week-long MPC simulation comparison with uncontrolled case showcased a 63% reduction in peak demand duration, a 68% decrease in Phasing Unbalance Index (PUI), and a 45% reduction in Voltage Unbalance Factor (VUF). The minimum voltage unbalance improved from 0.897 pu to 0.913 pu, reducing out-of-standard drops duration by 21%. Lastly, we observed a 20% cut in lines’ losses, equating to a 50 kgCO₂ reduction within a week.