Online EV Fast Charging Scheduling With Energy Storage System Using a Profit-Predictable MILP

Abstract

Fast charging stations aim to efficiently allocate power to electric vehicles (EVs) through optimized charging scheduling, providing rapid services while meeting increasing demand. However, due to the limited capacity of the grid infrastructure, an energy storage system (ESS) is often integrated to buffer power in advance. Coordinating the charging and discharging schedules of both EVs and ESS remains a challenging task. Many existing EV charging scheduling approaches overlook ESS power management, while others incorporate ESS considerations but neglect the dynamic nature of EV charging. For instance, the maximum charging power of an EV typically decreases as its battery nears full capacity. Moreover, uncertainties such as unknown EV arrival and departure times, unpredictable waiting behaviors, and fluctuating time-of-use (TOU) pricing further complicate the problem. To address these challenges, this paper formulates an online optimization problem and introduces a novel algorithm that predicts future profits to assist decision-making in power allocation. The algorithm leverages carefully designed mixed-integer linear programming (MILP) to emulate a neural network (NN). Simulation results using generated optimal charging data demonstrate that the proposed algorithm significantly outperforms existing baseline methods by up to 50%.