Surrogate-Assisted Combinatorial Optimization of EV Fast Charging Stations

In this work, an artificial neural network (ANN) surrogate model-based method is proposed to assist the optimal design of the modular electric vehicle (EV) fast dc charging station. This is a typical combinatorial optimization problem, which is hard to solve analytically as the key design parameters are discrete, e.g., the numbers of charging poles, power electronic converter modules, and switching contactors. In the first part of the article, the details on how to generate the expected charging power demand of the charging station are presented, where characteristic EV charging curves are considered. The charging station is designed to operate under a modified first-come-first-serve (FCFS) policy to maximize the quality of service (QoS) to the customers. The system time ratio, energy efficiency, and capital expenditure (CAPEX) are then taken as the performance indicators to evaluate different designs in correlation to expected charging demand. By varying the design parameters, we generate a group of datasets from an adjustable charging station simulation model, which is then used for supervised training of an ANN. As a surrogate model of the charging station, the trained neural network is finally used to quickly map the design parameters into the performance indicators, where optimal design parameters are found by evaluating the proposed cost function.