A Dual-Axis Enhanced State-Space Model for Remaining Useful Life Prediction of Lithium-ion Batteries

Abstract Accurate remaining useful life (RUL) prediction is essential for the safe operation and health management of lithium-ion batteries (LIBs). However, capturing stable long-term degradation trends remains challenging due to local noise, capacity regeneration, and redundant correlations among multi-channel sensor features. To address these issues, this paper proposes CAMamba2, a dual-axis enhanced architecture. The feature axis denotes the input-channel dimension, while the temporal axis represents the ordered battery-cycle dimension. Specifically, cross-feature interaction gating (CFIG) enhances cross-channel interactions before embedding, and a context-aware (CA) module combines instantaneous features with locally smoothed temporal context to mitigate short-term fluctuations. A Mamba2 state-space backbone models long-range dependencies. The method is evaluated within a two-stage framework comprising first prediction cycle (FPC) identification and RUL prediction. On the MIT dataset, CAMamba2 achieves a mean absolute error (MAE) of 0.0082 and a mean squared error (MSE) of 0.0001, outperforming representative baselines. Additional experiments on the HUST dataset, with CAMamba2 trained independently, demonstrate favorable performance under varying operating conditions. These results support dual-axis enhancement as an effective strategy for LIB RUL prediction.