A hybrid prognostic strategy with unscented particle filter and optimized multiple kernel relevance vector machine for lithium-ion battery

Abstract

To make up the deficiencies of single methods in lithium-ion battery state of health (SOH) and remaining useful life (RUL) estimation, this paper presents a novel hybrid method using unscented particle filter (UPF) with optimized multiple kernel relevance vector machine (OMKRVM). Firstly, the errors between the initial estimation by UPF and the actual capacity are obtained. After that, the residuals are reconstructed by complementary ensemble empirical mode decomposition (CEEMD) to reduce interference. In addition, OMKRVM is adopted to provide multiple predictive abilities, and kernel parameters and weights of OMKRVM are yielded by the grid search. Finally, the initial estimation is corrected by the predicted residuals using OMKRVM to further improve prediction performance. The new method (UPF-OMKRVM) is compared with existing methods in predicting the degradation process of lithium-ion battery. The experimental results show that the UPF-OMKRVM has high prediction accuracy in lithium-ion battery SOH and RUL estimation.