Abstract: Aiming at the problem of low accuracy in identifying inrush current in transformer excitation, a method for identifying inrush current in transformer excitation based on the CTHM model is proposed. Firstly, this model adopts CNN to automatically extract the local sudden change features of the excitation inrush current. Meanwhile, it combines the position encoding of the Transformer to add timing information and the multi-head attention mechanism to focus on the key features. Secondly, no-load closing and internal fault models of transformers under different working conditions were built on the Simulink simulation platform, and MATLAB script automatic simulation was developed to provide a large number of samples for the CTHM model. Finally, new samples with CT saturation conditions added were used as the test set to test the trained model. The results show that the recognition accuracy of the CTHM model can reach 98.05%, and the excitation inrush current recall rate can reach 96%. Under the CT saturation condition, the accuracy and recall rates can still reach 97.05% and 95% respectively.