A novel ranking method based on semi-SPO for battery swapping allocation optimization in a hybrid electric transit system

Abstract

The allocation of batteries in hybrid charging stations has consistently played a significant role in the decision-making process for plug-in charging and battery swapping. Predicting the state of charge (SOC) for each electric bus (EB) in advance is crucial to assist in making future battery allocation decisions. This paper proposes a semi-Smart “Predict, then Optimize” (semi-SPO) framework for the battery allocation scheduling problem. The battery allocation optimization problem is reduced to a ranking problem with respect to SOC and integrated into the prediction model. The rank of SOC is determined from both pairwise and listwise perspectives. Considering the inherent characteristics of rankwise regression, such as missing parameters and infinite number of optimal solutions, a geometric analysis is conducted. A model enhancement approach is then proposed to ensure the accuracy of both prediction and optimization models. This enhancement facilitates optimal decision-making while preserving the interpretability of predicted values. A case study is conducted using the real-world data of Nanjing, China. The result shows that the proposed semi-SPO framework offers superior decision-making outcomes in the battery allocation pro.