The thesis presents the prediction of battery ageing and remaining battery life using machine learning methods. Features were calculated from battery cycling data and used to train classical machine learning models to predict battery ageing using a leave one battery out approach. The predictions were evaluated using mean absolute error, root mean square error and the coefficient of determination. A hidden Markov model was applied to the final predictions. The results showed that successful prediction requires a sufficiently large set of training batteries, similar ageing behaviour among the batteries, and an appropriate selection of features and prediction model.
|
