Probabilistic Energy Consumption Estimation for Electric Buses

Abstract

Electric buses play a vital role in future sustainable transportation and estimating the energy consumption of electric buses is important for reducing range anxiety and optimizing charging schedules. However, due to the numerous unobserved factors in real-world driving, there are naturally significant uncertainties in the energy consumption of electric vehicles. Therefore, besides focusing on improving model accuracy, the estimating probabilistic distribution of the uncertainties can also help to better characterize electric bus energy consumption and increase the confidence in the estimation results. In this paper, two probabilistic models, based on Bayesian regression and quantile regression, are proposed to estimate the probabilistic distribution of electric bus energy consumption; the probabilistic models are trained and validated using real-world driving data from 10 electric buses over a year. The results show that the proposed methods both capture the probabilistic characteristics well; however, the variations of uncertainties are better adapted in quantile regression.