Deep Koopman operator-based remaining useful life prediction of Lithium-ion batteries under multi-condition scenarios

Abstract

Addressing the challenges of extensive training data requirements and limited generalization in lithium-ion battery remaining useful life (RUL) prediction, this paper proposes a novel Koopman-inspired degradation model. The model captures dynamic linear features during battery degradation by integrating operational parameters into the Koopman operator framework, enhancing RUL prediction accuracy and adaptability to varying conditions. Two experiments using voltage data from charging and relaxation phases demonstrate the model's superiority over traditional methods, highlighting its real-world applicability. Key innovations include: (1) a Koopman-based approach for extracting dynamic linear features, improving trendability; (2) embedding operational parameters into the model to enhance generalization; and (3) effective RUL prediction under small sample conditions. This work advances battery RUL prediction, offering a robust solution for multi-condition scenarios.