Utilizing Machine Learning Algorithms in a Data-Driven Approach to the Prediction of Vehicle Battery State of Charge with BMW i3 Datasets

The state of charge (SoC) in an electric vehicle must be assessed and projected for any scenario, using the array of data points that can be extracted from a vehicle. In this paper, we explored the utility of data-driven approaches to SoC prediction that do not rely upon any internal or equation-based understanding of the device operation. We leveraged three unique machine learning algorithms to predict the battery SoC using data from other features of electric vehicles. We used a publicly available dataset describing vehicle parameters and trip details for 70 trips in EV BMW™ i3 (60 ah) vehicles and evaluated aforementioned machine learning algorithms for predicting SoC percentage. We utilized a data processing technique (delta and stagger) to extract different perspectives from each trip record and demonstrated that machine learning techniques can be effectively used to predict battery SoC for a wide range of driving conditions and trip parameters.