A Remaining Useful Life Prediction Method for Lithium Battery Based on Dynamic Regression Selection Algorithm

ABSTRACT Accurate prediction of remaining useful life (RUL) is a critical technology for ensuring the safe and efficient utilisation of lithium batteries. Effective feature extraction from data is pivotal in achieving precise RUL predictions. This study introduces a novel approach by defining the input as uniformly sized greyscale images, which not only facilitates the training of convolutional neural networks (CNN) and long short‐term memory (LSTM) models but also minimises information loss during the preprocessing stage. Furthermore, a dynamic regression selection (DRS) algorithm based on CNN and LSTM is proposed, which emphasises the local performance of sub‐models and intelligently selects the most suitable predictive models based on the inherent characteristics of the data. Extensive validation conducted on the NASA battery dataset demonstrates that the proposed CNN‐LSTM‐DRS algorithm significantly outperforms standalone CNN and LSTM models in terms of accuracy, predictive capability, and robustness across various data sizes (60%, 70% and 80%). The results of this research highlight the proposed methodology's effectiveness and contribute to advancements in energy storage and battery management systems.