State‐of‐Health Prediction for Lithium‐Ion Batteries Based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise‐Gate Recurrent Unit Fusion Model

Abstract

State‐of‐health (SOH) estimation is one of the most critical battery management system (BMS) tasks. A challenge remains for the SOH prediction due to the complicated battery aging mechanism. The most common health indicator is the capacity of the lithium‐ion battery. The fluctuation of capacity caused by the capacity regeneration phenomenon can seriously affect the prediction performance. A new complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and gate recurrent unit (GRU) based fusion prediction model for SOH estimation is proposed to solve the problem effectively. First, the CEEMDAN algorithm decomposes the original SOH into local fluctuations and global degradation trends. Then, the GRU network and autoregressive integrated moving average model are used to predict the above trends, respectively. Next, a sliding window is designed to calculate an average value of the global degradation trend prediction residuals. Then, the second GRU algorithm can be used to correct prediction residuals. Finally, the prediction results of the aforementioned parts are combined to obtain the final SOH estimation. The proposed method is verified by experimental battery data from NASA and CALCE datasets. The results show that the fusion method has both higher estimation accuracy and stronger robustness than other methods.