Cycle-efficient modeling for degradation staging and early life prediction of lithium batteries

An effective and time-saving early life prediction model is crucial for rapid battery assessment. However, existing models face a dilemma: they either rely heavily on extensive historical data or provide limited predictive insights into battery degradation. To address this, this study proposes a cycle-efficient battery life assessment framework integrating data-driven and empirical models. The framework consists of two components: degradation stage detection relying solely on data from one cycle and early life prediction using five-cycle data. The early life prediction model is capable of achieving joint prediction of the battery's remaining useful life and the cycle to knee point. Experimental results demonstrate that the degradation staging model achieves an accuracy of 0.9776 for lithium iron phosphate batteries. Meanwhile, the early life prediction model yields mean absolute percentage errors of 10.5% for remaining useful life and 12.8% for the cycle to knee predictions. The model's accuracy and generalizability have been validated across diverse battery types, health states, and operating conditions. This proposed framework exhibits excellent generalizability capability under all evaluated scenarios, establishing a robust foundation for rapid battery design assessment and retirement decisions. • A degradation staging model relying solely on one-cycle data is proposed. • An early life prediction model using five-cycle data is presented. • The few-shot transfer learning capability is interpretably verified. • Generalization across battery types, health, and operating conditions is validated.