Comparative Assessment of Machine Learning Techniques for Modeling Energy Consumption of Heavy-Duty Battery Electric Trucks

Efforts to decarbonize the heavy-duty vehicle sector have generated vast interest in transitioning from conventional diesel trucks to battery electric trucks (BETs). As a result, understanding energy consumption characteristics of BETs has become important for a variety of applications, for instance, assessing the feasibility of deploying BETs in place of conventional diesel trucks, predicting the state-of-charge (SOC) of BETs after specific duty cycles, and managing BET charging needs at the home base or en-route. For these applications, mesoscopic energy consumption models offer a good balance between the amount and fidelity of the input data needed, such as average traffic speed and road grade on a link-by-link basis, and the model performance. As a common intelligent transportation system (ITS) application, this paper presents a comparative assessment of mesoscopic energy consumption models for BETs developed using three different machine learning techniques. The results show that the random forest (RF) regression outperforms the extreme gradient boosting (XGBoost), the light gradient-boosting machine (LightGBM), as well as the conventional linear regression as evidenced by the resulting model having a higher coefficient of determination (R2) value than that of its counterparts. When applied to the simulated dataset, the RF regression can capture the behaviors of BET energy consumption well where the R2 value of the resulting model is 0.94.