RUL prediction for lithium-ion batteries via adaptive modeling and improved particle filter

Abstract:

To improve the safety and reliability of lithium-ion battery operation and maintain the stable operation of the system, we proposed a prediction method for remaining useful life (RUL) of lithium-ion batteries based on adaptive hybrid model and improved particle filter (PF) algorithm. An adaptive hybrid model was established by combining empirical model and neural network model to describe the degradation trend of battery capacity, and beetle antennae search (BAS) algorithm was used to optimize the PF resampling step to solve the problem of loss of particle diversity, so as to improve the estimation accuracy and achieve accurate RUL prediction. Two groups of different types of lithium-ion batteries published by NASA and CALCE were selected as the research objects to verify the validity of the model and the accuracy of RUL prediction via comparing PF and improved PF algorithms. Experimental results show that the adaptive hybrid model was more expressive in terms of battery capacity change, which can reflect the variation of the internal parameters as well as the external environment of the battery. Compared with the traditional PF algorithm, the BAS-based improved PF (BAS-PF) method had higher estimation accuracy and more accurate RUL prediction results with the prediction errors of 5.88%, 3.92%, 1.96%, and 3.75%, 1.25%, 0%, respectively, for the two test batteries from different prediction points. The adaptive hybrid model can describe the characteristics of battery capacity more effectively, and the BAS-PF algorithm based on adaptive hybrid model has better prediction ability and greater reliability for battery RUL, which is helpful to improve the prediction accuracy and performance for RUL.