A Transfer Learning-Based Data-Driven Method for State-of-Health Estimation of Lithium-Ion Batteries

The health status of lithium-ion battery is a critical state to evaluate the degree of battery degradation. However, it cannot be measured directly, but needs to be estimated. Although significant progress has been made in accurate health state estimation, the time and resource consumption of generating target battery labels hinder the development of health state estimation methods. Firstly, A feature defined as the duration of the same charging voltage range (DSCVR) is extracted as the key health indicator for the LIB in this article, which is used to reflect the nonlinear degradation process of the battery. Secondly, we design a transfer learning framework to enable the estimation of battery state of health in the absence of target battery labels. This framework integrates a swarm of RVFL neural networks equipped with domain adaptation to produce accurate estimates. Cross-validation using two open-source battery data sets (NASA, CALCE). The verification results show that the proposed framework can ensure that the root mean square error (RMSE) is less than 2% in the absence of target labels.