Transforming waste to value: Enhancing battery lifetime prediction using incomplete data samples

Abstract

The widespread usage of rechargeable batteries in portable devices, electric vehicles, and energy storage systems has underscored the importance for accurately predicting their lifetimes. However, data scarcity often limits the accuracy of prediction models, which is escalated by the incompletion of data induced by the issues such as sensor failures. To address these challenges, we propose a novel approach to accommodate data insufficiency through achieving external information from incomplete data samples, which are usually discarded in existing studies. In order to fully unleash the prediction power of incomplete data, we have investigated the Multiple Imputation by Chained Equations (MICE) method that diversifies the training data through exploring the potential data patterns. The experimental results demonstrate that the proposed method significantly outperforms the baselines in the most considered scenarios while reducing the prediction root mean square error (RMSE) by up to 18.9%. Furthermore, we have also observed that the penetration of incomplete data benefits the explainability of the prediction model through facilitating the feature selection.