Remaining Useful Life Prediction of Lithium-ion Batteries with Fused Features and Multi-kernel Gaussian Process Regression

In this paper, both State of Health estimation and Remaining Useful Lifetime prediction of lithium-ion battery are investigated. In order to accurately predict the RUL of battery, a multi-kernel Gaussian Process Regression (GPR) model which combines an adaptive feature fusion method is proposed. Firstly, the five raw features are extracted from charging and discharging curves. Secondly, an adaptive feature fusion method is used to combine different features and integrate their advantages of the features. In the same time, the optimized multi-kernel GPR model with Fruit-fly Optimization Algorithm is established to solve the GPR kernel function selection problem. The effectiveness of the proposed method is verified with simulation experiments derived from battery dataset of NASA. Simulation results show that the proposed method achieves better prediction accuracy and reliability than the GPR model using single feature or single kernel function.