Currently, prediction of trip-based electricity consumption of electric buses (EBs) has become an important prerequisite for the deployment of large-scale electric bus fleets and the location of the charging infrastructures. Previous state-of-the-art approaches to estimate the electricity consumption focus on making rough electricity consumption assumptions or building physics-based electricity consumption model. This paper constructed a neural network model to predict the trip-based electricity consumption of EBs and six influencing factors were taken as input variables. Further, sensitivity analyses were performed to investigate how these factors influence the consumption results. This model was implemented and validated on real-world electric bus data from a five-month consecutive collection in Shenzhen, China, comprising 1024 EBs. The experiment demonstrated the predictive effectiveness of this model and the results from sensitivity analyses show that trip length is the key factor to determine the consumption, but other factors average travel speed, the number of bus stops and traffic lights, direction, time parameters also have different level of impacts on the results.
Keywords Electric bus, electricity consumption, neural network, big data