Predicting Lithium-Ion Battery State of Health using Linear Regression

Abstract: Knowledge of battery health is very important. It provides insight into the capacity of a given system and allows the operators to plan ahead more efficiently. But measuring state of health (SoH) of a battery is difficult, and takes time. More importantly, the battery needs to be taken out of operation to be analysed correctly. This paper aims to evaluate a proposed linear regression method for predicting battery health, based on easily acquired operational data. The main predictor being voltage deviation, a characteristic of battery voltages during charge/discharge cycles. Using this method, the only time a battery would need to be extracted is to gather training data. Then, the model could be used for similar batteries to predict their SoH. Meaning those systems would never need to be halted, increasing productivity. The results of this paper is that the data used was not suitable for linear regression. There were problems with heteroskedasticity and non-normality of the residuals, but mainly the estimated parameter for the relationship between voltage deviation and SoH ran contrary to established theory. Which could not be overlooked. Therefore, the estimated models should not be used to predict SoH. To accomplish the goal of accurate SoH prediction, more research should be conducted and a better sample used.