Pregnancy pulse diagnosis is an important part of traditional Chinese medicine. In this literature, we propose a hybrid deep learning framework by combining the network advantages of one-dimensional convolutional neural network(1DCNN), gated recurrent unit(GRU) and attention mechanism(ATT), which realized automatic identification and classification of pregnancy period through wrist pulse data. First, the wrist pulse signals of 120 healthy pregnant women in different stages of pregnancy were collected. Second, a hybrid networks for classification in three stages of pregnancy pulse was designed, which is comprised of six 1DCNN blocks and GRU module and attention module. Specifically, the proposed 1DCNN layer are adopted to extract local features from the input wrist pulse data and GRU unit was utilized for overall pulse sequence learning. In addition, the attention mechanism assigned different weights to feature parts according to pulse trends to focus and highlight effective features. Finally, the designed network was used to realize the discrimination of pregnancy pulse. On our local research dataset evaluation, the accuracy of pulse classification of this network for the three stages achieves encouraging results of 93.7%, 93.6%, and 94%, respectively. Furthermore, the model reaches 90.4%, 90.7%, 90.6% and 98% in terms of average precision, sensitivity and F1 score and AUC, respectively. The experiment shows that the designed network has a good reference value for the auxiliary diagnosis of TCM pregnancy pulse diagnosis, which is conducive to promoting the development of modernization of TCM.