A Novel Transfer Learning-Based Cell SOC Online Estimation Method for a Battery Pack in Complex Application Conditions
Pengliang Qin, Linhui Zhao
Abstract
The crucial function of a battery management system is to estimate the state of charge (SOC). However, complex application conditions, such as different temperatures, aging, and inconsistency between cells would cause a distribution difference between data domains and lead to a significant estimation error in the established SOC model. Transfer learning's domain adaptation offers a way to lessen the disparity in distribution across the data domains, and yet it also requires that the target space be the same across domains, which is challenging to do with the SOC online estimate. This article proposes a SOC differential processing method as well as designs a combined transfer learning method. As experiments of actual battery pack provided by China First Automobile Work-based confirmed, the proposed method can obtain precise and robust SOC estimation results with a mean absolute percentage error of less than 1.5%.