An enhanced bilayer long short-term memory method for energy consumption estimation of electric buses with real-time passenger load

Abstract

The accuracy in estimating energy consumption of electric buses is of significant importance for formulating electric bus route planning and charging schedules. Current approaches for estimating energy consumption of electric buses can be categorized into three major types: physics-driven models, statistical models, and deep learning methods. This study develops an Enhanced Bilayer Long Short-Term Memory (EBLSTM) method for energy consumption estimation of electric buses considering real-time passenger load, along with the Powertrain-based Physical Model (PPM) and Scale Tractive Power-based Model (STPM). A linear interpolation model is first implemented to reconstruct the bus trajectory (i.e., position, speed, and acceleration) from 0.1 Hz to 1 Hz for model calibration and verification. A tanh activation function is designed to mitigate fluctuations in the estimation results of the traditional LSTM method. The genetic algorithm, least mean square method and grid search approach were conducted respectively to calibrate the above three different models. Numerical results indicate that the EBLSTM method achieves the best estimation performance, with a verification Root Mean Square Percentage Error (RMSPE) of 0.68 %. In contrast, the RMSPEs of the PPM and STPM models are 0.90 % and 1.13 %, respectively. Furthermore, both qualitative and quantitative analysis were conducted to examine the impacts of initial SOC, travel time, and the heterogeneous characteristic of different bus datasets on the accuracy of the three models.