This research provides a prognostic framework for off-line SOH estimation of Li-ion battery. With a CNN-Transformer architecture, this program is capable of modeling the temporal correlations of battery signals from both local and global views. In this way, the learning ability of both local features and long period contexts will be enhanced.