Objective A modified SEIR-A model based on the SEIR model was proposed to describe the transmission mechanism of infectious diseases more accurately in view of the increasing complexity of infectious diseases and the gradual prevalence of a kind of diseases with a large number of latent infections in the process of transmission. Methods In terms of the kinetic model there were two main improvements. One was to assume that the incubation period population and the dominant infected population had the same infectious factors the other was to introduce a recessive infection population A with different infectiousness and add a constant factor ω for one-way conversion from recessive infection to dominant infection so as to construct an SEIR-A model of a special recessive infection population. In addition the improved SEIR-A model was linearly combined with the TCN model of time convolution network and a hybrid model with the fusion of dynamic model and deep learning model was obtained. Results Through the fitting of real data the results showed that the SEIR-A model could simulate the general trend of infectious diseases and could accurately fit the existing recessive infected population and the accumulated recovered population in the disease and the determination coefficient R2 reached 0. 987 0 and 0. 989 9 respectively proving that the model was reasonable. The hybrid model of SEIR-A and TCN can fit the complex existing dominant infected population. Compared with the single SEIR-A model TCN model and LSTM model the determination coefficient R2 of the hybrid model reached 0. 9611 obtaining the best fitting accuracy among the five comparative models. Conclusion The combination of traditional dynamics and deep learning can effectively solve the problem of low fitting accuracy of traditional models while embodying the mechanism of disease transmission which has practical significance for the study of infectious diseases.