Joint intra- and inter-cycle modeling for co-estimating state of health and remaining useful life of lithium-ion batteries

Accurate estimation of the state of health (SOH) and remaining useful life (RUL) is essential for the safe and reliable operation of lithium-ion batteries. However, existing data-driven methods do not fully integrate degradation information across intra- and inter-cycle scales, and SOH evolution is not fully used for RUL prediction. To address these issues, this study proposes a hierarchical model for SOH and RUL co-estimation using partial charging data. The model describes battery degradation at two temporal scales: (1) an Intra-cycle Encoder extracts variable-specific temporal features and inter-variable interactions within each charging cycle to characterize short-term degradation patterns; and (2) the resulting cycle embeddings are treated as nodes in a temporal directed graph, where graph attention network v2 (GATv2) aggregates information from historical cycles to model long-term inter-cycle degradation dependencies. The resulting multi-scale representations are used for SOH and RUL co-estimation, while a dual-branch RUL prediction head integrates implicit degradation representations with explicit local SOH evolution information to improve RUL prediction. To improve cross-chemistry adaptation with limited labeled target-domain data, a low-rank adaptation (LoRA)-based fine-tuning strategy is introduced. The proposed method is evaluated on a mixed-condition dataset containing 99 batteries of NCA, NCM, and NCM+NCA chemistries. Experimental results show that the model outperforms representative baselines, achieving root mean square error (RMSE) values of 0.42% for SOH estimation and 28.46 cycles for RUL prediction. Cross-domain transfer experiments further demonstrate its adaptability to unseen battery types, highlighting its potential for advanced battery management systems.