Interpretable machine learning models for predicting Ebus battery consumption rates in cold climates with and without diesel auxiliary heating

Abstract

The global shift towards sustainable and environmentally friendly transportation options has led to the increasing adoption of electric buses (Ebuses). To optimize the deployment and operational strategies of Ebuses, it is imperative to accurately predict their energy consumption under varying conditions, particularly in cold climates where battery life is typically degraded. The exploration of this aspect within the Canadian context has been limited. In addition, we have found that existing models in the literature perform poorly in the Canadian environment, giving rise to the need for new models using Canadian data. This paper focuses on the development, comparison, and evaluation of various data-driven models designed to predict the energy consumption of different Ebuses with different heating technologies under a wide range of climate conditions. We specifically use Canadian data as a good representative of cold climates in general. The results show that the performance of the different bus types varies substantially under the exact same conditions. In addition, tree-based family of models proves to be the most suitable approach for predicting the Ebus consumption rate. The results indicate that the Random Forest method emerges as the superior choice for predicting the energy consumption rate, with a resulting mean absolute error of 0.09–0.1 ​kWh/km observed across the different models. Furthermore, SHAP analysis shows that the main variables influencing the energy consumption rate depend on the type of heating system (using the battery for heating or using an auxiliary system that utilizes diesel for heating) adopted.