Synchronous state of health estimation and remaining useful lifetime prediction of Li-Ion battery through optimized relevance vector machine framework

Abstract

This study proposes a hybrid kernel function relevance vector machine (HKRVM) optimized model for battery prognostics and health management. To monitor battery state of health (SOH), two ageing features (AFs) are extracted from the incremental capacity curve to quantify capacity degradation. To further predict remaining useful life (RUL), the AFs are treated with the BOXCOX transformation to enhance the linearity between AFs and cycles. Then, a metabolic extreme learning machine is developed to successionally predict the degradation trends of AFs quickly and accurately. The HKRVM is proposed to capture the underlying relationship between AFs and capacity. To determine the optimal weights and kernel parameters in HKRVM, the biological evolution in the genetic algorithm (GA) is integrated into the grey wolf optimizer (GWO) to further improve the population diversity and optimization performance of the basic GWO.Furthermore, 23 benchmark functions are employed to illustrate the effectiveness of the genetic grey wolf optimizer (GGWO). Finally, the extracted and predicted AFs are fed into the optimized HKRVM for validation. For four battery packs, the mean error of estimated SOH is 2.7773% and the predicted RUL is 21.3333 cycles, which is better than the Gauss process regression, support vector regression, and some particle-filter-based methods.