A dual-stage framework for lithium-ion battery state of health prediction using multi-strategy optimization for decomposition and neuro-fuzzy networks

With lithium-ion batteries (LIBs) serving as a cornerstone in modern energy storage systems, accurate state of health (SOH) prediction is vital for ensuring operational safety, optimizing performance, and extending service life. However, the highly nonlinear, time-varying, and noise-contaminated nature of battery degradation trajectories poses substantial challenges to achieving both accurate and robust SOH prediction. To address these issues, this study proposes a dual-stage synergistic enhancement framework for LIBs SOH prediction. A multi-strategy sand cat swarm optimization (MSCSO) algorithm is developed to enhance exploration–exploitation balance, which simultaneously drives entropy-guided variational mode decomposition for noise suppression and fine-tunes a neuro-fuzzy random vector functional link (NF-RVFL) network. The NF-RVFL network takes the extracted features from the denoised IMF components and concatenates them along the feature dimension with the health index (HI) to form the input vector for each sample, and outputs the predicted SOH. Finally, the proposed framework was rigorously evaluated on capacity data from two public LIBs datasets and benchmarked against state-of-the-art algorithms. The proposed algorithm achieves an average RMSE reduction of over 90 % compared with NF-RVFL on the NASA and CALCE datasets. The results demonstrate that our approach achieves superior accuracy, robustness, and generalizability in SOH prediction, confirming its significant promise for practical battery health monitoring systems.