A Comprehensive Exploration on Different Machine Learning Techniques for State of Charge Estimation of EV Battery

The State of Charge (SoC) is a measurement of the amount of energy available in a battery at a specific interval of time, mostly expressed as percentage. Proportional relationships between the electromotive force of a battery, current, terminal voltage and temperature determine the SoC. There can be a considerable error in the calculations due to a sharp drop of the terminal voltage at the end of discharge. This research has explored how important SoC is, as a factor in Battery Management Systems. The work focuses on using machine learning techniques to obtain an accurate and reliable status of battery charge, this includes Random Forest, Decision Tree, Gradient Boosting, Support Vector Regression, Polynomial Regression and Multilayer Perceptron. In this paper, these techniques are tested and compared with two real world captured datasets of Lithium-ion batteries which includes LG Battery and Unibo Powertools Battery. For supporting this study, statistical methods like K-fold cross validation and Grid Search cross validation techniques are used to estimate the skill of machine learning models. After implementing these techniques, it is found that Random Forest model returns the best Accuracy and Decision Tree returns the least Mean Absolute Error.