Battery State of Health and Charge Estimation Using Machine Learning Methods

Abstract

In this study, state of health (SOH) and state of charge (SOC) estimation of series connected batteries were evaluated for their charge and discharge durations. For this purpose, an ARM-based electronics card module was developed for observing instantaneous batteries voltage, current and temperature values during the charge and discharge process. The implemented microcontroller based card module gathers data from the current, voltage, and temperature sensors and it transfers to the computer environment via serial communication port. A specific human machine interface is designed via app-designer. The obtained variables were used for estimating regression models of the machine-learning toolbox. Random forest, decision tree, polynomial, extreme gradient boosting, linear and gradient boosting regression models were used for instantaneous SOH and SOC estimation for batteries during the charge-discharge period. Root Mean Square Error (RMSE) and í µí± 2 score results were used for performance evaluation of regression models. When the RMSE and í µí± 2 score results were compared, the decision tree regression model was the regression model that made the most accurate SOH and SOC estimation and the results were presented.