A mixed integer quadratically constrained programming model for optimal deployment of capacitated fast-charging stations for electric vehicles: a Bayesian optimization approach

Despite the global surge in electric vehicle (EV) adoption and the recognized environmental, economic, and technical advantages of EVs, limited research addresses the location-allocation problem for EV charging stations (EVCSs) on directed and separated highway networks, especially considering both driving directions and single- or dual-access facilities. This study bridges that gap by addressing the capacitated fast-charging EVCS location-allocation problem on directed highways, incorporating path-based demands and the limited EV ranges. Unlike previous studies that assume a uniform number of chargers at each station, we allow the number of chargers at each EVCS to vary. In this study, we propose a novel model to optimize both the location of EVCSs and the number of chargers at each station. To overcome the computational challenges of identifying the optimal EVCS deployment, we develop a modified Bayesian optimization algorithm that leverages the expected improvement acquisition function. This approach balances demand maximization with the minimization of epistemic uncertainty in the surrogate model. The applicability of the proposed framework is validated through extensive experiments on the Pennsylvania Turnpike highway. Furthermore, a payback period analysis is conducted to enhance the practical relevance of the proposed model, providing insights into the economic feasibility of EVCS investments under different deployment scenarios. Sensitivity analyses are conducted to evaluate the effects of key factors, including the EV driving range, battery charge levels, and average traffic flow rates. Lastly, our solution approach achieves these results with significantly reduced computation times compared to baseline solutions, assuming a fixed number of chargers per EVCS.