Remaining Useful Life Prediction of Lithium-Ion Battery With Adaptive Noise Estimation and Capacity Regeneration Detection

As an indispensable energy device, 18650 lithium-ion battery has widespread applications in electric vehicles. Remaining useful life (RUL) prediction of lithium-ion battery is critical for the normal operation of electric vehicles. In conventional approaches, the adaptive estimation of model parameters and the detection of capacity regeneration await further research. To adaptively estimate the noise variables in the degradation model and to accurately detect the battery capacity regeneration, this article proposes a novel expectation maximization-unscented particle filter-Wilcoxon rank sum test (EM-UPF-W) approach. In detail, in the case of unlabeled small samples, this article constructs a dynamic degradation model on the basis of UPF for a single battery, which adaptively estimates the noise variables with the aid of EM algorithm. Furthermore, the Wilcoxon rank sum test is introduced to determine the capacity regeneration point, so as to decrease the prediction error. A 18650 lithium-ion battery dataset produced by NASA is used to demonstrate the approach. Experimental results show that the proposed EM-UPF-W outperforms some existing data-driven techniques.