Li-ion Battery Health Estimation Based on Multi-layer Characteristic Fusion and Deep Learning

Existing methods for estimating lithium-ion (Li-ion) battery capacity fading mostly depend on a priori knowledge on aging mechanism and on the use of chemical or physical formulation and analytical battery models. This dependence is usually difficult to determine in practice, which restricts the application of these methods. In this study, we propose a new method for Li-ion battery health assessment based on multi-layer characteristic integration and deep learning. The proposed method realizes multi-layer characteristic fusion (charge current, discharge voltage, charge time under constant voltage etc.) of relative input parameters by unsupervised self-learning both in the data-layer and the feature-layer, in which the effect of multi-feature fusion can be improved by refactoring learning. Based on the feature extracted by the fusion of multi-layer self-learning, the health assessment of Li-ion battery under multi-feature fusion can be realized using the logistic regression model. Experiments are implemented based on data provided by the NASA Ames Prognostics Center of Excellence. Results confirm that the proposed estimation method performs well by comparing the actual trend results of battery capacity fading.