Battery SOH estimation via an optimized CNN–BiLSTM–Attention network using ICA-Based ageing features

Abstract

Accurate estimation of lithium-ion battery State of Health (SOH) remains challenging because most existing methods rely on full charging cycles, are sensitive to noise and capacity regeneration, or require manual hyperparameter tuning that limits generalization across cells and datasets. To address these issues, this study proposes a hybrid CNN–BiLSTM–Attention framework optimized by a Genetic Grey Wolf Optimizer (GGWO) for SOH estimation using only two informative features extracted from partial charging data via Incremental Capacity Analysis. The CNN extracts local degradation patterns, the BiLSTM captures long-range temporal dependencies, and the attention mechanism adaptively emphasizes salient temporal information, while the GGWO automatically searches for optimal hyperparameters to improve robustness and accuracy. Extensive experiments on both public CALCE datasets and a private multi-cell LBP dataset demonstrate that the proposed model achieves superior estimation performance across varying temperatures and loading conditions. The GGWO-optimized model attains a minimum MAE of 0.42% and RMSE of 0.51%, consistently outperforming conventional machine learning baselines as well as the non-optimized CNN–BiLSTM–Attention model. These results confirm the model’s strong generalization capability and its suitability for real-time implementation in battery management systems.