Data-Driven Approach for Estimating Energy Consumption of Electric Buses under On-Road Operation Conditions

Abstract

The transformation of diesel buses into battery-powered electric buses for public transportation has become a global trend. The ability to evaluate the energy consumption of electric buses is critical in bus scheduling for alleviating range anxiety. In this study, an energy consumption estimation model for electric buses was proposed based on actual bus operation data. The operating states of an electric bus were categorized into four types: depressed accelerator pedal, depressed brake pedal, vehicle sliding, and vehicle idle states. Based on the bus state, two models were constructed to estimate the energy consumption. A multivariate linear model based on vehicle speed, accelerator pedal position, and instantaneous power was constructed to estimate the energy consumption of buses in the depressed accelerator pedal state. Combining that model with a long short-term memory (LSTM) algorithm, machine learning algorithms were calibrated to estimate bus energy consumption in the other three states over the four seasons. A comparative analysis was conducted for the different algorithms. The root-mean-square errors of the estimation results based on LSTM for vehicles in the depressed brake pedal, vehicle sliding, and vehicle idle states were 0.12%, 0.03%, and 27.27% lower than those of the artificial neural network, respectively. Accurate estimations of bus energy consumption during the four seasons allow bus operation companies to adjust the bus charging schedule to reduce the operating costs.