Lithium-ion battery prognostics with fusion model of uncertainty integration based on Bayesian Model Averaging

Abstract

Lithium-ion battery remaining useful life (RUL) estimation has become a critical issue of intelligent battery management system (BMS). Various models and algorithms have been developed to achieve the RUL prognostics for lithium-ion batteries, to obtain high estimation performance. Generally, a single model usually requires long train time and complex train progress to reach satisfactory precision, while complex data-driven approaches sometimes result in overfitting results. Therefore, many fused and integrated methods have been proposed to overcome the disadvantages of the single method. However, few fusion approaches deal with the uncertainty management or uncertainty integration at present. In order to solve the lack of uncertainty management of existing fusion RUL estimation methods, a fusion model based on Bayesian Model Averaging (BMA) of uncertainty integration for lithium-ion battery RUL estimation is presented in this paper. Firstly, sub-models applying Multiple Linear Regression (MLR) are created. Then, the posterior probability of these sub-models are calculated and integrated to implement the probability fusion of BMA process for RUL prediction. Experimental results with the battery data sets from Center for Advanced Life Cycle Engineering (CALCE) show that compared with a single model, the BMA fusion model achieves higher accuracy and stability. In addition, uncertainty management is taken into consideration, which can be referred as a new research direction for lithium-ion battery RUL estimation.