Lithium-ion battery pack on-line health diagnosis based on multi uncertainty model fusion.

Lithium-ion battery pack has become one of the most widely utilized energy storage component in renewable energy system under "Carbon neutralization and carbon peak" of China. The irreversible performance degradation of the lithium-ion battery directly impacts the working reliability and safety of the host system. However, the nonlinear degradation process, battery cell inconsistency and dynamic operating conditions brings barriers of lithium-ion battery pack health state estimation. To overcome these challenging issues, this paper proposed an on-line heterogeneous statistical fusion method to estimate the lithium-ion battery health state. Firstly, the data-driven relevance vector machine (RVM) is utilized to establish the direct, nonlinear mapping between the monitored degradation feature and battery capacity. Both the estimation results and the corresponding uncertainties are quantitatively represented. Then the unscented particle filter takes this datadriven health diagnosis results as temporal measurements and regards the uncertainty as the measurement noise. A filtered on-line state of health estimation result with corresponding ensembled uncertainty is generated based on this state-space model. The experiments based on the real tested lithium-ion batteries illustrate the potential of this fused statistical model to achieve accurate and robustness degradation estimation with a more accurate confidence interval.