Deep learning enhanced lithium-ion battery nonlinear fading prognosis

With the assistance of artificial intelligence,advanced health prognosis technique plays a critical role in the lithium-ion(Li-ion) batteries management system.However,conventional data-driven early aging prediction exhibited dramatic drawbacks,i.e.,volatile capacity nonlinear fading trajectories create obstacles to the accurate multistep ahead prediction due to the complex working conditions of batteries.Herein,a novel deep learning model is proposed to achieve a universal and accurate Li-ion battery aging prognosis.Two battery datasets with various electrode types and cycling conditions are developed to validate the proposed approaches.Knee-point probability(KPP),extracted from the capacity loss curve,is first proposed to detect knee points and improve state-of-health(SOH) predictive accuracy,especially during periods of rapid capacity decline.Using one-cycle data of partial raw voltage as the model input,the SOH and KPP can be simultaneously predicted at multistep ahead,whereas the conventional method showed worse accuracy.Furthermore,to explore the underlying characteristics among various degradation tendencies,an online model update strategy is developed by leveraging the adversarial adaptationinduced transfer learning technique.This work gains new sights into the comprehensive Li-ion battery management and prognosis framework through decomposing capacity degradation trajectories and adversarial learning on the unlabeled samples.

Nonlinear health evaluation for lithium-ion battery within full-lifespan

Lithium-ion batteries(LIBs), as the first choice for green batteries, have been widely used in energy storage, electric vehicles, 3C devices, and other related fields, and will have greater application prospects in the future. However, one of the obstacles hindering the future development of battery technology is how to accurately evaluate and monitor battery health, which affects the entire lifespan of battery use. It is not enough to assess battery health comprehensively through the state of health(SoH) alone, especially when nonlinear aging occurs in onboard applications. Here, for the first time, we propose a brand-new health evaluation indicator—state of nonlinear aging(SoNA) to explain the nonlinear aging phenomenon that occurs during the battery use, and also design a knee-point identification method and two SoNA quantitative methods. We apply our health evaluation indicator to build a complete LIB full-lifespan grading evaluation system and a ground-to-cloud service framework, which integrates multi-scenario data collection, multi-dimensional data-based grading evaluation, and cloud management functions. Our works fill the gap in the LIBs’ health evaluation of nonlinear aging, which is of great significance for the health and safety evaluation of LIBs in the field of echelon utilization such as vehicles and energy storage. In addition, this comprehensive evaluation system and service framework are expected to be extended to other battery material systems other than LIBs, yet guiding the design of new energy ecosystem.