An Enhancement to the Linear Dynamic System Model for Air Traffic Forecasting

Abstract

traffic forecast is among the most important functionalities of air traffic controls. The Linear Dynamic System Model that predicts the traffic demand within the Air Route Traffic Control Centers serves well for this purpose. This model formulates inflows and outflows between Centers’ boundaries by assuming that the boundary crossings between Centers are in conformance with Poisson distribution and that the traffic patterns do not change too much over multiple days. As a result, the traffic in the near future can be predicted based on knowledge of historical traffic patterns and anticipated departures. As a predictive model, its prediction accuracy relies heavily on parameter estimation. In the earlier implementations of this model, the traffic patterns are obtained by averaging estimations of multiple days. However, given the uncertainties in the traffic system and the deficiencies inherent in the radar track data, using observed traffic data of a few days to train the parameters is likely to result in bias due to limited samples. A large training set on the other hand contains a lot of noise, to which the mean is susceptible. This paper introduces the Kernel Density Estimation into the Linear Dynamic System Model. This non-parametric method serves as an enhancement to the model in that it is able to capture the major transition patterns from a large data set in the presence of outliers and data deficiencies regardless of the actual distribution of data. Therefore, this statistical approach is useful in extracting normal traffic patterns that are representative of major behavior of the traffic flows. An one-month traffic simulation shows that, by incorporating with the Kernel Density Estimation, the Linear Dynamic System Model reduces the estimation errors by 20% on average.

Original document

The different versions of the original document can be found in:

• http://pdfs.semanticscholar.org/8ffd/e6f1dd5fa016cb32e97e167049b0d51e80d8.pdf

• http://arc.aiaa.org/doi/pdf/10.2514/6.2013-5030,

http://dx.doi.org/10.2514/6.2013-5030

• https://web.ics.purdue.edu/~dsun/pubs/gnc13a.pdf,

https://arc.aiaa.org/doi/10.2514/6.2013-5030,

https://academic.microsoft.com/#/detail/2319408836