Interpretable bus energy consumption model with minimal input variables considering powertrain types

This study aims to build an interpretable energy model for urban buses considering powertrain types to serve bus operators with minimal variables and simple structure, in contrast to existing literature which pursues high accuracy through complex machine learning models and engine-related parameters. Three different model types, the power-based model, the Long-Short-Term-Memory model and the XGBoost model, are applied for electric buses (e-buses) and diesel buses. The models are calibrated using empirical driving records and energy consumption rates. A novel state classifier is developed and integrated into the conventional power-based model, significantly improving the accuracy of the conventional one and showing comparable performance to the other two machine-learning models. For e-buses, the modified power-based model is more interpretable and simpler, showing superiority over other models. All three models cannot achieve high goodness-of-fit for diesel buses, illustrating the need to include more vehicle operational variables in the diesel bus energy model.