In order to forecast air traffic flow in short term, a combined forecasting model based on decomposition and integration method is proposed. Firstly, the EEMD method is used to decompose the traffic time series data into several components; secondly, the permutation entropy is used to calculate the complexity of each component. The high-frequency component is classified as high-frequency component and the rest is classified as low-frequency component; then, BP neural network algorithm is used to predict the high-frequency component, and the least square method is used to predict the low-frequency component; then, the prediction results of the components are added Finally, the final prediction value is obtained. Finally, the actual operation data are collected for example analysis. By comparing the prediction results of 0 ~ 6 h and 6 ~ 12 h, the EC Value of the model in 0 ~ 6 h is 0.905, and the accuracy is higher. Compared with EMD-BP-OLS model and BP model, the evaluation index of this model is better than other models. By comparing the prediction results of 60 min, 30 min and 15 min time scale data, the EC Value of 60 min time scale is 0.924, with the highest accuracy. The results show that the model proposed in this paper is feasible and effective, and more suitable for short-term flow forecasting.