Online Learning-based State of Health Estimation and Consistency Diagnosis for Lithium-ion Battery Pack in Embedded Battery Management System

Accurate state-of-health (SOH) estimation and consistency diagnosis are essential for the safe operation of lithiumion battery packs. However, existing approaches often rely on large-scale data training or complex models, which introduce significant computational and storage challenges for embedded battery management systems (BMS). This paper proposes a lightweight online learning framework for pack-level consistency diagnosis and SOH estimation. A dual-path recursive least squares algorithm is first developed to extract open-circuit voltage (OCV) curves and diagnose cell consistency in real time, thereby constructing health feature (HF) curves for SOH estimation. These HF curves are then used to train an electrode-level OCV model using an online gradient descent (OGD) algorithm, enabling SOH determination based on learned electrode model and consistency parameters. Experimental verification on an inconsistent battery pack demonstrates that the proposed approach can identify individual HF curves across various temperatures and discharge rates. Based on the HF data, the method achieves a maximum SOC consistency diagnosis error of 0.9%, while the SOH estimation error is 1.75% for individual cells and 3.2% for the battery pack. Furthermore, execution analysis in an embedded BMS indicates that the proposed algorithm offers reduced memory usage and faster computation, making it well-suited for embedded BMS applications.