Predicting Charging Duration for On-Demand Electric Vehicle Mobile Charging Service

Electric vehicles (EVs) are one key to more sustain-able transportation. However, the widespread use of EVs faces challenges, such as the limited public charging infrastructure. One solution is a mobile charging service (MCS) providing on-demand EV charging. This paper focuses on predicting charging durations for EVs and MCS, addressing the challenge of deriving predictions from small datasets. By applying data-driven machine learning algorithms, we show that effective predictions are achievable even with limited data, employing techniques like data augmentation, feature engineering, and data transformations. A simple reference model called a combined model is proposed, exploiting the average charging speeds. Our finding is that accurate predictions can be achieved through tailored approaches for both EVs and MCSs, where the prediction errors are from 4 to 7 minutes for MCSs and below 6 minutes for EVs depending on the prediction scenario. The novel combined model and ensemble methods, such as Random Forest and Gradient Boosting Regression Trees, produced the best results.