Short-term Forecasting Method of Air Traffic Flow based Neural Network Ensemble

In this research, in order to address interferences of air traffic from complex factors like weather and local data abnormality of radar samples, fuzzy clustering and neural network ensemble were introduced into the short-term forecasting of air traffic flow. Firstly, with K-means cluster analysis, this research compared traffic volume at different time with that of each clustering center to identify the temporal clustering of traffic volume. Secondly, according to different data sets from clustering analysis, corresponding neural network models were established. On the basis of Bagging method, a neural network ensemble weight allocation algorithm of fuzzy subordinative degree was also built to identify weight of each neural network and to establish neural network ensembles model. Finally, according to 3. principle of normal distribution, abnormal data out of Section (mu-3 sigma, mu+3 sigma) was cleaned and short-term forecasting results were acquired. Our model showed superior results of short-term radar data forecasting for Shanghai Terminal Area, overmatching regression analysis and neural network forecasting. The experiment verified that the method is valid and feasible for short-term forecasting of air traffic flow.