A cross-material lithium-ion battery state of health estimation method based on three-stage domain adaptation

With the rapid development of electric vehicles and energy storage systems, the accurate estimation of the battery state of health (SOH) is crucial for system reliability. However, the cross-material system transferability and the randomness of charging segments result in insufficient generalization capability of traditional methods. This paper proposes a three-stage transfer learning framework that operates on arbitrary charging segments to enable an accurate SOH estimation across different battery materials. In this method, through correlation analysis, six types of health features are extracted, covering the State of Charge (SOC) range of 20 %–85 % to address the issue of charging segment randomness. On this basis, a three-stage transfer learning framework is constructed. The global statistical distribution is explicitly aligned using the Maximum Mean Discrepancy (MMD). The proposed Simplified Mixed Domain Adaptation (SMDA) method synthesizes a Mixup-based interpolated domain to enable a smooth transition between the source and target distributions. Fine-tuning (FT) with 1 % and 5 % of the labeled data from the target domain is used to optimize regression accuracy. Furthermore, a weighted fusion strategy is adopted to adaptively integrate the estimation results from multiple SOC ranges, enhancing the robustness of the SOH estimation method. Experimental results indicate that when 5 % of the target-domain data is used for fine-tuning, the average root mean square error (RMSE) of single-interval transfer across four battery materials (LCO, NCA, NMC, and Hybrid) is 0.75 %. For full-range transfer from LCO to other materials, the average RMSE is 0.84 %. The method significantly improves the accuracy and engineering applicability of battery SOH estimation in cross-material and fragmented charging scenarios.