Airport passenger throughput is the main production indicator of civil airports, and its forecast data is an important reference indicator for making airport construction plans, which directly affects whether the project can be built and its construction scale. In order to improve the prediction accuracy of airport passenger throughput and better guide the planning, construction and development of airports, on the basis of comprehensive evaluation of grey GM (1,1) model, exponential smoothing and BP neural network, the entropy method is used to divide the The weight of a single model is used to construct a combined forecasting model with higher forecasting accuracy. Taking the passenger throughput data of Beijing Capital International Airport as an example, the above combined model is used to predict the passenger throughput of the airport. The average fitting error of the combined model is 2.94%, which has high prediction accuracy and can accurately predict the future passenger throughput of the capital airport.