Indirect prediction of remaining useful life for lithium-ionion batteries based on multiple health indicators and multi-scale decomposition

This study addresses the challenge of low prediction accuracy in estimating the remaining useful life (RUL) of Lithium-ion batteries due to high-dimensional health indicators (HIs). We propose a novel method that integrates multiple HIs and multi-scale decomposition. By applying Spearman's correlation coefficient and adaptive chirp mode decomposition, we extract HIs strongly correlated with capacity and decompose them into global trends, local fluctuations, and noise. Validation on NASA and CALCE datasets shows our method outperforms Gaussian Mixture Regression (GMR), achieving RUL prediction errors within 0–2 cycles and mean absolute and root mean square errors below 1.05 % and 1.61 %, respectively. This approach enhances RUL prediction accuracy, offering significant benefits for battery management systems.