RUL prediction for lithium-ion batteries based on variational mode decomposition and hybrid network model

作者：Hu, Liwen; Wang, Wenbo^*; Ding, Guorong

来源：Signal Image and Video Processing, 2023, 17(6): 3109-3117.

DOI：10.1007/s11760-023-02532-z

• Lithium-ion battery RUL prediction
• GRU
• LSTM
• SOH estimation
• VMD

Lithium-ion batteries are widely used in the field of electric vehicles and energy storage due to their superior performance. However, with increased use time, lithium-ion battery performance declines significantly, which can indirectly lead to the decline of device performance or failure. Therefore, accurate prediction of the remaining useful life (RUL) of lithium-ion batteries enables timely maintenance and replacement of batteries to ensure safe and reliable operation. In this paper, a hybrid network model is constructed by combining a long and short-term memory network with a gated unit recurrent network, and a RUL prediction method for lithium-ion batteries based on variational modal decomposition and GRU/LSTM hybrid network model is proposed. The method is divided into two stages: the estimation of SOH and the prediction of RUL. In the SOH estimation stage, four easily collected lithium-ion battery health indices are used to estimate the SOH of lithium-ion battery through a hybrid GRU/LSTM network. In the RUL prediction stage, the SOH data obtained in the previous stage are firstly decomposed by the VMD method to reduce the effects of capacity regeneration and other noises. Then, on the basis of the SOH data processed by VMD, a hybrid GRU/LSTM network is used for accurate prediction of the remaining lifetime of the lithium-ion battery. The proposed method is validated by NASA battery dataset and CALCE battery dataset, and compared with GRU method, LSTM method and Bi-LSTM method. The experimental results show that the VMD-GRU/LSTM two-stage prediction method can predict RUL of lithium battery more accurately, and has good robustness and generalization.