A Closed-Loop Prediction and Optimization Approach for Battery Swapping Station of Electric Vehicles under Demand Uncertainty

This work investigates the prediction of the quantity of fully charged batteries required at battery swapping stations for electric vehicles under demand uncertainty. Decision-making commonly follows a sequential model involving forecasting steps without feedback, referred to as an open-loop approach. For instance, during the operation of a battery swapping station, the operator uses predictive decision-making to determine which vehicles to swap in the upcoming time slot. This process helps to prepare the necessary number of fully charged batteries. A discrepancy, such as an insufficient number of charged batteries for swapping, leads to increased costs for the station to meet current demand. To address this, we adopt a closed-loop framework. Within this framework, the optimal prediction model is determined based on an application-specifc cost function. We structure the estimation method as a bilevel optimization problem. This problem is solved using an exact method that relies on the KKT (Karush–Kuhn–Tucker) conditions of the second-level problem. Finally, the case study demonstrates that the closed-loop prediction approach for the battery swapping station operation system outperforms the traditional open-loop method, which is least squares.