Predictive Modeling of Li-Ion Battery State of Charge in Electric Vehicles: Study of Modern Machine Learning Regressors

One of the crucial factors that affects range anxiety is the estimated state of charge of the battery in an electric car. Having a higher estimation accuracy is a tough job since the traditional methods are inefficient and inaccurate. The limits of conventional approaches are examined in this study, followed by modern machine learning regression models such support vector regression, random forest regression, K-nearest neighbor regression, and linear regression. In the EV domain, estimation of SoC of an EV using these ML models and subsequent comparative examination of the performance metrics remain largely untapped. There has been almost no research using the K-nearest neighbor regression model. A full vehicle model that included the powertrain and the heating circuit was validated using real driving trips with a BMW i3 (60 Ah) to obtain the dataset for the research. Thus, the dataset had to be extracted, cleaned, and analyzed using several parameters done during the exploratory data analysis.