Comparative Study of Data Driven Methods for State of Charge Estimation of Li-ion Battery

The battery storage system is an inevitable component in an electric vehicle. A precise state of charge is critically notable to evaluate the level of battery aging and ensure electric vehicle reliability and security. Recently, data driven methods have procured much popularity in many research fields. Data-driven methods are found to be a promising approach to provide a high accuracy solution to battery state of charge estimation problem. Research works focus on the state of charge estimation under a fixed operating temperature. Apparently, the maximum deliverable charge of a battery degrades with charge-discharge cycle and temperature. Thus, it is important to consider the operating temperature as one of the input features. This paper identifies the best data driven model for state of charge estimation among linear regression, random forest, CatBoost and XGBoost models. The data driven models are developed via training, validating, and testing stages deploying Li-ion (LG 18650HG2) battery data set under -10°C, 0°C, 10°C and 25°C. The best model is identified using performance indices. The results show the superior performance of XGBoost model under all temperatures and best performance at 25°C with mean absolute error of 0.68%, mean square error of 0.01% and root mean square error of 1.10%, mean absolute percentage error of 1.78%, and R2 value of 99.86%, compared to linear regression, random forest, CatBoost models.