State of Health Estimation of Lithium-ion Battery Based On Improved Transfer Learning

Many existing methods of forecasting the state- of-health(SOH) assume that training and testing data follow the same distribution. The model based on dataset under one working condition may be ineffective for the dataset under another working condition due to the distribution discrepancy. In order to meet this challenge, this paper proposes an improved method Mutual Information Domain-Adversial Neural Networks (DANN) based on domain adaptation, which improves the domain discriminator to better extract domain invariant features. In addition, to avoid the loss of target information, the mutual information among target features, source features, and original target data is calculated to fix the features on the target site during the migration process. Different from the traditional methods, we only use 40% of the data sets for training, and the rest are used for prediction, so we can complete the prediction of more scenes. Experimental results show that this method can accurately and stably predict SOH.