Multi-Objective Optimization of Hybrid Battery-Supercapacitor Energy Storage Systems: Machine Learning-Enhanced Performance for Fast-Charging Electric Vehicle Infrastructure

With the rapid growth of the global electric vehicle (EV) stock, the large-scale deployment of fast-charging infrastructure is facing an increasingly urgent need for compatible energy storage adaptation. All traditional single-type energy storage technologies have inherent, irreconcilable flaws: the power density of pure battery energy storage is insufficient to withstand the peak load shocks of fast-charging stations, while pure supercapacitor energy storage has low energy density and high full-lifecycle costs, which makes it unable to independently meet the operational requirements of long-duration fast-charging services. To address this issue, this study proposes a new multi-objective optimized battery-supercapacitor hybrid energy storage system (HESS) tailored for EV fast-charging stations. The study adopts the non-dominated sorting genetic algorithm II (NSGA-II) to complete multi-objective parameter optimization, introduces deep reinforcement learning (DRL) to realize real-time energy scheduling, and conducts simulation verification across multiple operational scenarios using measured EV charging data from 12 real fast-charging stations. The results show that compared with pure battery energy storage systems, the optimized HESS achieves a 23.7% increase in energy efficiency, a 41.2% reduction in grid peak load, and a 34.6% extension in system service life. The DRL module attains a 94.3% charging demand prediction accuracy, cuts operational costs by 18.9%, reduces grid harmonics by 27.4%, and can help fast-charging stations lower capital infrastructure investment by 15%-25%. This study still has two limitations: the training of the machine learning model relies on a large volume of historical data, and the computational complexity of the real-time optimization algorithm is relatively high, which requires iteration and optimization in follow-up research.