A Data-Driven Framework for Lithium-Ion Battery SOH Estimation Using VMD-GRU Hybrid Approach with Multi-Scale Feature Analysis

The accurate state of health (SOH) estimation in lithium-ion batteries represents a critical technological challenge with profound implications for electric vehicle performance and user experience. Precise SOH assessment not only enables reliable mileage prediction but also ensures operational safety. However, the complex and non-linear capacity fading process during battery cycling poses a challenge to obtaining accurate SOH. To address this issue, this study proposes an effective health factor derived from the local voltage range during the battery charging phase. First, the battery charging phase is divided evenly with reference to voltage intervals, and an importance analysis is conducted on each voltage interval. From these, the voltage interval with the strongest correlation to State of Health (SOH) is extracted as the feature interval. Then, a data-driven framework integrating variational mode decomposition (VMD) with gated recurrent unit (GRU) neural networks enables comprehensive multi-scale temporal feature analysis for enhanced SOH estimation. The methodology begins with rigorous feature engineering to identify and extract optimal health indicators demonstrating superior correlation. Subsequently, the VMD algorithm performs sophisticated signal processing to decompose both the measured capacity and derived health indicators into their constituent intrinsic mode functions and residual components. Finally, a GRU-based neural network is implemented to establish a robust SOH estimation model. Experimental validation using cycling data from different datasets shows that the root mean square error of the estimation results is consistently below 3%, demonstrating the good accuracy and generalisation of the proposed method, using only local data from the charging phase.