A Data-Driven Air Traffic Network Delay Modeling Approach Considering En-Route Congestion


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date10 Dec 2021


Flight delay is of great concern for civil aviation industries worldwide. Due to the imbalance between the fast increasing air travel demand and the stagnant capacity improvement of aviation infrastructures, resultant flight delay has become more severe. Excessive flight delay not only wastes travelers' time and fuel consumption of the aircraft, but also takes a toll on economic activities and the environment. Therefore, to alleviate the current operational pressure and enhance flight safety and efficiency, it is necessary to learn the flight delay characteristics and model the air traffic network flight delay accordingly.

Traditional methods focusing on air traffic network flight delay modeling normally assume that air traffic networks are constrained at airports, i.e., flight delays are only generated at airports. This is true when the en-route airspace is not congested several decades ago. Recently, as the en-route is getting more crowded, en-route delay has become more prominent, which is regarded as a significant factor contributing to total flight delays. However, most existing studies on flight delay modeling do not consider en-route congestion explicitly, and would underestimate flight delays if applied directly to current air traffic systems. Hence, it is crucial to devise a new network delay modeling framework which takes en-route congestion into account.

To fill the research gap, a data-driven network delay modeling framework, named Multi-layer Air Traffic Network Delay (MATND), is proposed. The MATND learns the en-route bottlenecks automatically from massive flight operational data, and then uses a queuing network model, which incorporates these en-route bottlenecks, to give a more precise system-wide delay at a macroscopic level. The proposed MATND is enabled by the availability of flight operational data, which includes aircraft tracking data and flight schedules published by many commercial or noncommercial websites.

The first part of the proposed framework employs data-mining algorithms to identify the operational flight routes and en-route constraint. Powered by the public available datasets, i.e., aircraft tracking data and flight schedules, a clustering algorithm is employed to generate the operational flight routes. Next, in order to find the en-route bottlenecks, the whole airspace is discretized into cells by Cartesian grid. A score function is proposed to evaluate the congestion level of this cell using the structure of the airspace only. The proposed score function composes of three metrics, i.e., the traffic load, the number of operational routes, and the entropy of route directions. A cell with a higher score is more likely to reach its capacity, therefore, aircraft that flies through this region has a high chance of experiencing delay or diversion. The identified congestion cells are further clustered as en-route congestion points, which are further validated using busy waypoint published by CAAC. Later, the en-route congestion points are incorporated into the flight delay model to capture the en-route characteristics.

The second part of the proposed framework involves a stochastic and dynamic queuing network model that treats each airport or en-route congestion point as a node of the queuing network to model the flight delays and their propagation. The en-route congestion points learned from the previous step are taken into the queuing network, and flight delays would be generated when the aircraft passes an airport or an en-route congestion point. In this way, the network bottlenecks are incorporated into the flight delay model, which is more similar to the current characteristics of air traffic systems.

In order to demonstrate the importance of including en-route congestion points, both theoretical proofs and simulation tests are conducted, and the results show that including en-route congestion point is necessary for the air traffic system with excessive en-route congestion.

Test on real operational data of China's air traffic system shows the superiority of MATND compared to classical queuing network models without en-route constraint. Several "what-if" scenarios are also conducted to evaluate the effectiveness of air traffic network improvement strategies, where the manipulation of reality at such a scale is impossible. Results show that en-route congestion is prominent in China's aviation system, and delay propagation effects are more severe at en-route congestion points than at airport. Moreover, most airports in China aviation system would benefit more from en-route capacity improvement, which confirms that the bottleneck in current air traffic network of China is en-route airspace constraint. Besides, MATND model is computationally efficient, well suited for evaluating the impact of policy alternatives on system-wide delay at a macroscopic level.

To sum up, this thesis presents a novel data-driven method for air transport network abstraction and simulation, referred as Multi-layer Air Traffic Network Delay (MATND) model. Unlike existing researches which did not model the en-route constraint explicitly, this model constructs a multi-layer air traffic network consisting of airports, en-route congestion points, and the air routes liking them, which are learned automatically from flight operational data. Then, flight delays generated at each airport and en-route congestion point are further propagate through the entire network. Both theoretical proofs and simulation tests are provided to show the superiority of the proposed MATND model. Finally, scenario tests are conducted to show how MATND can be used to evaluate a broad range of alterative policies and strategies in air traffic management improvement.