A novel hybrid model based on ensemble strategy for lithium-ion battery residual life prediction

Abstract

As a power source widely used in electronic systems, the stability and safety of lithium-ion battery operation is of paramount importance. Generally, the decline in lithium-ion battery capacity can be used as a health indicator to track its degradation. This paper establishes a hybrid model to describe the capacity degradation of lithium-ion batteries, and proposes a novel ensemble strategy to improve the prediction accuracy of the remaining useful life (RUL) of lithium-ion batteries simultaneously. An empirical exponential model is selected to track the degradation trend of the battery, and the model parameters are updated by the unscented Kalman filter (UKF) algorithm. A homogeneous ensemble model based on the Bagging algorithm is generated to predict the residual evolution. The exponential model modified by residuals is introduced to predict the RUL of lithium-ion batteries. Simulation experiments on two kinds of batteries indicate that the proposed ensemble method can obtain accurate and stable predictions of RUL of lithium-ion batteries with strong robustness and less generalization errors.