Growing electric mobility makes it difficult for electric vehicles (EVs) to charge adequately while charging infrastructure capacities are limited. Due to the prolonged charging times, precise planning is needed, which necessitates knowing the availability of charging stations. In addition, inconsistencies in charging facilities and illogical charging arrangements cause partial queuing and idling of charging stations. To tackle these issues, it is necessary to first understand EV charging station choice behavior and its influence. This study examines EV charging station choice behavior and aims to find the best prediction method. This study implements a novel interpretable machine learning (ML) framework to predict EVs’ charging station choice behavior. The experiment was based on two years of real-world normal and fast charging event data from 500 EVs in Japan. The results revealed that the XGBoost model achieved the highest accuracy compared to the other ML classifiers in predicting charging station choice behavior. Furthermore, this study employed the newly developed SHAP approach to identify feature importance and the complex nonlinear and interactive effects of various attributes on charging station choice behavior. This study suggests that combining ML models with SHAP has the potential to develop an interpretable ML model for predicting EV charging station choice behavior.

日益增长的电动交通使得电动汽车 (EV) 在充电基础设施容量有限的情况下难以充分充电。由于充电时间延长，需要进行精确规划，这需要了解充电站的可用性。此外，充电设施不统一、充电安排不合理，造成充电站局部排队、闲置。为了解决这些问题，有必要首先了解电动汽车充电站选择行为及其影响。本研究考察了电动汽车充电站的选择行为，旨在找到最佳的预测方法。本研究采用一种新颖的可解释机器学习 (ML) 框架来预测电动汽车的充电站选择行为。该实验基于来自日本 500 辆电动汽车的两年真实世界正常和快速充电事件数据。结果表明，与其他 ML 分类器相比，XGBoost 模型在预测充电站选择行为方面的准确性最高。此外，本研究采用新开发的 SHAP 方法来识别特征重要性以及各种属性对充电站选择行为的复杂非线性和交互影响。这项研究表明，将 ML 模型与 SHAP 相结合有可能开发出一种可解释的 ML 模型，用于预测 EV 充电站的选择行为。本研究采用新开发的 SHAP 方法来识别特征重要性以及各种属性对充电站选择行为的复杂非线性和交互影响。这项研究表明，将 ML 模型与 SHAP 相结合有可能开发出一种可解释的 ML 模型，用于预测 EV 充电站的选择行为。本研究采用新开发的 SHAP 方法来识别特征重要性以及各种属性对充电站选择行为的复杂非线性和交互影响。这项研究表明，将 ML 模型与 SHAP 相结合有可能开发出一种可解释的 ML 模型，用于预测 EV 充电站的选择行为。