ADAPTIVE ENERGY CONSUMPTION FORECASTING FOR ELECTRIC BUSES USING MACHINE LEARNING

Abstract

Electric city buses have become a key solution for reducing carbon emissions and improving air quality in urban areas, with over 670,000 electric buses in operation globally as of 2023 over 95% of which are deployed in China. Public transport systems account for around 25% of total urban energy consumption, and improving the energy efficiency of electric buses can reduce fuel costs by up to 30%. However, traditional methods for estimating energy consumption often depend on historical averages, static models, and driverspecific behavior, making them time-consuming, error-prone, and poorly suited for adapting to dynamic traffic and route conditions. To overcome these limitations, this study presents a machine learning-based framework to predict the energy economy (measured in kWh/km) of electric city buses. The proposed approach leverages real-world operational data, including speed profiles, passenger loads, elevation variations, and ambient temperatures, collected via a smart fleet monitoring system across various urban routes and thousands of recorded trips. The data undergoes preprocessing steps such as imputation, normalization, outlier removal, and feature engineering to enhance model performance. While benchmark models like Linear Regression and XGBoost Regressor are used for comparison, they exhibit shortcomings in modeling complex nonlinear relationships. In contrast, the proposed CatBoost Regressor model demonstrates superior performance due to its efficient handling of categorical data, robustness in high-dimensional feature spaces, and lower dependency on manual tuning. The model is trained using an 80-20 train-test split and delivers high predictive accuracy. Test results confirm that the CatBoost model achieves lower Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), making it a valuable tool for energy consumption forecasting and intelligent scheduling in electric bus operations