Machine Learning-based Electric Vehicle Battery State of Charge Prediction and Driving Range Estimation for Rural Applications
Magdy Abdullah Eissa, Pingen Chen
Abstract
This paper aims to address the gap in the literature by proposing a machine learning-based approach to predict the battery state of charge (SOC) and the driving range of electric vehicles (EVs) in rural applications. While the literature revealed a lack of a comprehensive model that considers all major factors influencing EV range prediction, the proposed approach considers the three major classes of factors that influence EV range and SOC: vehicle parameters, driver behavior, and exploitation environment. The real-world driving cycle (RDC) data used in this study were collected from an On-Board Diagnostics (OBD) device connected to the driver's vehicle. The results of our study showed that the proposed machine learning approach was able to achieve an average accuracy of 95% in predicting the SOC of lithium-ion batteries. This high level of accuracy was achieved despite the large variations in the RDC data and the presence of noise in the measurements. The proposed machine learning approach was also able to accurately predict the Remaining Driving Range (RDR) of EVs using the predicted SOC values, with an average error of less than 2%.