Distribution-aware multi-source domain adaptation for cross-domain lithium-ion battery state-of-health estimation

Accurate state-of-health (SOH) estimation is essential for the safe and reliable operation of lithium-ion battery management systems. However, conventional data-driven methods often rely on single-source aging data and suffer from poor generalization when applied to batteries with different operating conditions and degradation trajectories. Existing transfer learning methods may further suffer from negative transfer when heterogeneous source domains are treated without sufficient distinction. To address this issue, this paper proposes a distribution-aware multi-source domain adaptation framework for cross-domain SOH estimation. The framework decouples shared degradation representations from source-specific aging patterns through a common feature encoder, a shared estimator, and multiple domain-specific estimators. A maximum mean discrepancy-based weighting mechanism is introduced to measure the distribution similarity between the target domain and each source domain, enabling adaptive source contribution assignment during target-domain adaptation. For a new target battery, the pretrained modules are reused, and only limited labeled target-domain samples are required for lightweight fine-tuning. The proposed framework is validated using capacity-voltage curve segments extracted from partial discharge intervals as representative degradation-related inputs. Cross-domain experiments on four heterogeneous battery aging datasets demonstrate that the proposed method reduces the average RMSE by 65.2% compared with the representative multi-source baseline methods, indicating improved accuracy and robustness in multi-source SOH estimation. This work provides a practical multi-source adaptation strategy for transforming increasingly accumulated heterogeneous battery aging data into transferable battery health knowledge.