End-to-End State-of-Health Prediction of Lithium-Ion Battery via Vision Transformer Enhanced by a Hybrid Scale Information Fusion Network

Accurately predicting the state-of-health (SOH) of lithium-ion batteries is crucial for achieving an efficient battery management system. Deep neural networks (DNNs), which can automatically extract features, are promoting the rapid development of data-driven SOH prediction for lithium-ion batteries. The completeness of the extracted features largely determines the performance of DNNs, so designing a network with excellent feature extraction ability has become an important issue. An end-to-end model, which can extract features from multiple channels and scales, is established to predict the SOH of lithium-ion batteries. In this model, a hybrid scale information fusion (HSIF) network is designed to extract multiscale hidden features and fuse multi-channel information, and a vision transformer network is used to capture long-range dependencies in feature maps. Tests are conducted on two public datasets with training samples not exceeding 38% of the total samples. The test results show that the root mean square errors are 0.00843 and 0.00671 for these two datasets, which are at least 16% greater than that of the comparison models. The comparative results show that the designed HSIF network has excellent feature extraction ability and that the proposed model has advantages in accurately predicting the SOH of batteries.