Research on hybrid data-driven method for predicting the remaining useful life of lithium-ion batteries

The instability and inconsistency of lithium-ion batteries (LIBs) may lead to sudden battery failures that cause serious accidents, hence the safety and reliability of the battery can be ordinarily effectively improved via improving the accuracy and uncertainty of the remaining useful life (RUL). Nevertheless, capacity data of LIBs display significant nonlinearity and are plagued by problems such as capacity regeneration (CR) and difficult to precise uncertainty. In order to address this issue, the improved northern goshawk optimization (INGO) algorithm and the variational mode decomposition (VMD) algorithm are combined in this article to present a unique hybrid driven by data prediction technique that adaptively breaks down the nonlinear, non-smooth initial battery capacity sequence into several trend subsequences and fluctuating subsequences. Its goal is to make the battery capacity sequence less complicated. Additionally, the deconstructed fluctuation subsequence is summed into a reconstructed sequence to optimize the computational process. Ordered neurons-long short-term memory attention mechanism (ONLSTM-AM) architectures and Tensor transfer learning-deep neural network (TTL-DNN) are employed to forecast the trending subsequence and rebuilt sequences, respectively. By doing this, the quantity of data that needs to be predicted is decreased and the training process is expedited. In this paper, the method is experimentally validated using the NASA dataset and the CALCE dataset, and the accuracy is compared with several common machine learning algorithms. The experiment's findings show that the proposed strategy produces the lowest RMSE values of 0.0055 Ah in the NASA dataset and 0.0061 Ah in the CALCE dataset, displaying high prediction accuracy, strong long-term prediction ability and high generalization ability. Our source code is available at https://github.com/Mmabc333/A-hybrid-method.