Offline and Online Blended Machine Learning for Lithium-Ion Battery Health State Estimation

This article proposes an adaptive state-of-health (SOH) estimation method for lithium-ion (Li-ion) batteries using machine learning. Practical problems with feature extraction, cell inconsistency, and online implementability are specifically solved a proposed individualized scheme blending offline model migration ensemble First, based on the data of pseudo-open-circuit voltage measured over battery lifespan, systematic comparison different incremental capacity features is conducted to identify suitable SOH indicator. Next, pool candidate models, composed slope-bias correction (SBC) radial basis function neural networks (RBFNNs), trained offline. For operation, prediction errors due inconsistency in target new then mitigated by modified random forest regression (mRFR)-based learning process high adaptability. The results show that compared prevailing methods, SBC-RBFNN-mRFR-based can achieve considerably improved accuracy (15%) while only small amount early-age measurements needed practical operation. Furthermore, applicability SBC-RBFNN-mRFR algorithms real-world operation validated from electric vehicles, it shown 38% improvement be achieved.