Health status estimation of Lithium-ion battery under arbitrary charging voltage information using ensemble learning framework

The safe and reliable operation of lithium-ion (Li-ion) batteries is crucial for electric vehicles (EVs). As a result, the state of health (SOH) of Li-ion batteries has always been a critical factor in the energy management of EVs. Since the charging process of Li-ion batteries is often stable and controllable, researchers can extract health characteristics from the charging voltage, and then use data-driven methods to assess the Li-ion batteries’ SOH. Although existing research has utilized health characteristics in charging voltage for SOH assessment, most methods have failed to effectively address the issue of EV users charging at various state of charge (SOC) region arbitrarily. To address this gap, an advanced ensemble learning framework technique is proposed, which estimates SOH of Li-ion batteries by utilizing arbitrary charging voltage information. The uniqueness of this method lies in the introduction of relevance vector machines (RVM) to construct the base models, and enhancing estimation accuracy by extracting features of local incremental capacity during a short charging period. Subsequently, a stacking-based ensemble method is proposed to integrate the base models for flexible SOH estimation. Finally, we conduct experiments on three datasets to validate the effectiveness of our technique, and the results demonstrate that the average RMSE is <1.5 % for any partial charging behavior.