Answers are grounded in this paper and cite their sources.
Prediction model demonstrates accurate RUL assessment in lithium-ion batteries, suggesting robustness against noise interference.
The nonlinearity of capacity degradation and the noise interference induced by capacity regeneration phenomena significantly constrain the accuracy of Remaining Useful Life (RUL) prediction for Lithium‐ion batteries. To mitigate these issues, this paper establishes a prediction model integrating SHapley Additive exPlanations (SHAP) feature enhancement with an attention‐based dual‐channel network. In the data preprocessing phase, the SHAP algorithm is employed to quantify the contribution of historical capacity sequences. A high‐discrimination weighting strategy is designed to selectively amplify degradation trend signals while suppressing random fluctuation noise, thereby reducing interference from local regeneration phenomena at the source. For modeling, a parallel dual‐channel structure comprising Long Short‐Term Memory (LSTM) and Gated Recurrent Unit (GRU) is constructed to capture long‐term memory dependencies and short‐term local variations in the capacity time series, respectively. Furthermore, an attention mechanism is introduced to dynamically adjust channel weights for adaptive feature fusion. Validation on the Center for Advanced Life Cycle Engineering (CALCE) dataset demonstrates that the proposed model achieves precise prediction with limited historical data. Notably, under complex conditions with high noise, the model exhibits superior noise robustness, effectively avoiding misleading local capacity jumps while accurately fitting the global degradation trajectory.
No takes yet. Share an insight, caveat, or question.
Zheng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: