s The clock error prediction of atomic clock plays an important role in the time and frequency control of atomic clock，which is related to the accuracy of time scale calculation and the stability of the punctual system. In order to further improve the long-term stability of clock group time and improve the accuracy of clock error prediction of atomic clocks，a deep learning network prediction model suitable for clock error prediction is constructed. The influence of the network model’s hyperparameters on clock error data is analyzed. And the best parameter index is given. That is，the number of hidden layers is 3，and the number of hidden layer neurons is 256. The results show that in terms of long-term prediction，the prediction effect of the optimized deep network model is 86.93% higher than that of the least squares model，and the deep network model is more suitable for high-precision time and frequency control，and has application prospects in the processing of clock error data in punctual systems.