Leveraging Multi-View Imputation Strategy for Robust Battery Lifetime Prediction under Missing-Data Scenarios

Abstract

While lifetime prediction of rechargeable batteries is crucial for ensuring the reliability and sustainability of electric devices, the accuracy and robustness of prediction models are often impacted by practical non-idealities in operational scenarios. In order to ensure the reliability of battery lifetime prediction, this work is dedicated to addressing a specific challenge posed by missing information in training data, which can be induced by multiple practical factors. To address this issue, this paper investigates multiple modeling strategies for handling missing data challenges, among which a novel multi-view imputation strategy is proposed that explores the diversity of underlying data patterns, thereby substantially improving the prediction accuracy. Experiments have been conducted to quantitatively evaluate the efficacy of the modelling techniques, where the proposed method is highlighted with substantial improvements in prediction accuracy and robustness, such that the root mean square error (RMSE) was reduced by up to 35.7% under intensive missing data conditions compared to conventional approaches. Through offering an innovative solution for accommodating missing data in predictive modeling, this study has advanced the development of efficient and reliable battery management systems.