Abstract: It's worth to predict available short-term air traffic flow and reduce ATCO workload. An air traffic flow model is built based on K-nearest neighbor. First, relative errors from different K values are compared to determine the appropriate K values. After that, space parameter is introduced to improve the model. Then these three kinds of state vectors are combined and new K-nearest neighbor models are proposed including time dimension model, to route-time dimension model and time-space parameter model. Radar data within a certain sector is used to test K-neighbor model, showing out that K-nearest neighbor model with time-space parameter has minimum error,whose average error equals to 14.6%. Distance measuring method based on weight index can attain the goal of prediction accuracy optimization. Gaussian function can produce a better result under time parameter model while it is weaker when space parameter is taken into consideration. Statistics show prediction's error is only 13.94% under the index weight distance method of inverse function model with time-space parameter. The improved K -nearest neighbor model has applicability for different traffic situations and strong portability for complicated air traffic situation of China.

Key words: short-term air traffic flow prediction, K-nearest neighbor model, state vector, space parameter, Gaussian
function