Applicability of Regression-Based Machine Learning Models in Energy Consumption Prediction for Electric Two-Wheelers Based on Micro-Trip Approach

Machine learning (ML) models are crucial in foreseeing the energy consumption (EC) of alternative powertrain vehicles, which is necessary for determining the driving range (DR) and eliminating the ‘range anxiety’ of commuters. This research aims to develop an appropriate ML model for predicting the EC of electric two-wheelers (E2Ws) using a dataset from real-world driving tests. Primarily, data on vehicle, geography, and atmosphere are compiled from 42 driving trips using a data acquisition system installed on an electric two-wheeler along a predetermined route within Vellore city. The data from these driving trips are segmented into 1815 micro-trips, and the input dataset is created with 20 feature variables and one target variable to estimate and validate the EC or DR of E2Ws. In addition, the seven regression-based ML models are trained and validated with the derived dataset. Furthermore, the selected ML models are compared in terms of their prediction ability using various assessment measures, including error matrices (mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE) and coefficient of determination (R2). The comparison assessment shows that the ExtraTree regression model outperformed other ML models in predicting the EC of E2Ws. Moreover, this article offers valuable insights for commutators, manufacturers, and other stakeholders to achieve sustainable, energy-efficient electric mobility.