RUL Prediction of Lithium-Ion Batteries based on Combined Network Model Considering Partial Charge and Discharge Data

Abstract

Lithium-ion batteries are widely used in new energy vehicles, but capacity regeneration and fluctuations during aging affect the accuracy of remaining useful life (RUL) prediction. Complete charge/discharge data are often unavailable during actual usage. To address these issues, this paper proposes a combined model for RUL prediction using partial charge/discharge data. Five health indicators are extracted from the voltage vs time curve and processed using variational mode decomposition to remove outliers and noise, improving the correlation between HIs and battery capacity. Spearman’s correlation coefficient verifies the relationship between HIs and capacity. The Kolmogorov-Arnold Networks-Structured State Space model (KAN-S4) is then developed, capturing spatial correlations and long-term degradation patterns. Experimental validation using data from our laboratory and the University of Maryland's CALCE center shows that the KAN-S4 model achieves accurate RUL predictions, even under complex conditions like capacity regeneration and rapid decline. The model demonstrates strong robustness and generalization across varying usage scenarios.