Data-Driven SOH Estimation for Lithium-Ion Battery Based on CEEMDAN-BiLSTM and Bayesian Optimization

Accurate estimation of State of Health (SOH) in lithium-ion batteries is critical for ensuring the safety, reliability, and operational efficiency of energy storage systems in electric vehicles and grid applications. However, battery degradation exhibits complex nonlinear behaviors, while capacity regeneration and noise introduce significant uncertainties into traditional SOH estimation methods. This paper introduces a novel hybrid framework, CEEMDAN-BiLSTM-BO, designed to address these challenges through three core components: Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for robust signal decomposition, Bidirectional Long Short-Term Memory (BiLSTM) networks for bidirectional temporal feature learning, and Bayesian Optimization (BO) for optimal weight tuning. CEEMDAN decomposes raw capacity data into Intrinsic Mode Functions (IMFs), effectively isolating global degradation trends from noise and capacity regeneration components. BiLSTM networks capture bidirectional temporal dependencies in each IMF, while BO dynamically optimizes IMF weights to mitigate interference from non-stationary signals. Experimental validation on multiple battery datasets demonstrates that CEEMDAN-BiLSTM-BO outperforms existing methods by leveraging adaptive noise suppression, data-driven optimization, and bidirectional sequence learning.