State-of-Health Estimation for Lithium-ion Batteries Based on An Improved Gated Dual Attention Unit Neural Networks

State of health (SoH) as a crucial parameter plays a vital role in battery management system, which can assess the degree of aging in lithium-ion batteries. However, the SoH values cannot be directly obtained through sensors. To solve the issue and obtain precision SoH values, an improved gated dual attention unit (IGDAU) neural network is proposed. Firstly, the moving average filter is applied to smooth the differential thermal voltammetry curve and health indicators with six are derived. Subsequently, the pearson correlation coefficient is utilized to analysis the relationship between health features and SoH. Then, the whale optimization algorithm method is employed to optimize the hyperparameters of the gated dual attention unit model. Finally, the proposed methods are validated by using open sources dataset provided by the University of Oxford. The results indicate that the IGDAU model can precisely estimate the SoH values. In comparison to the SVM, GPR and IGRU methods, the values of the four evaluation indexes are reduced, showing good estimation ability.