The delay diffusion phenomenon caused by large-scale flight delays has a significant impact on air traffic networks. In order to better predict and control delay diffusion, a delay diffusion prediction model combining convolutional neural network (CNN) and traditional epidemiological SEIR（susceptible-exposed-infectious-recovered） model is proposed for the nonlinear and complex characteristics of flight delay diffusion, as well as for the difficulty of combining real-time and accuracy. Based on the flight delay diffusion propagation mechanism and SEIR model, a flight delay diffusion dynamics model was established, and the key parameters in the model are optimized by convolutional neural network. The optimized model was simulated using MATLAB, and the accuracy was improved by 17.95% compared with the traditional SEIR model after combining the CNN model.