| dc.description.abstract | Growing air traffic has resulted in congestion and flight delays. Delays not only inconvenience passengers but also have negative environmental impacts and cause monetary losses for airlines. Reducing delays is therefore crucial for operating a sustainable, efficient, and robust aviation infrastructure. Network analysis has been a popular tool to study large-scale interconnected systems due to its analytical and computational tractability. However, time-varying topologies, multilayered interactions, and the inability to model the high variability in flight delays have limited the utility of traditional network models in aviation applications. In this dissertation, we use ideas from switched-systems theory, graph signal processing, and machine learning to develop tools that overcome some of these limitations.
The key idea behind our modeling approach is to (i) simplify the complex network interactions by identifying a small, finite set of representative network topologies, (ii) use data to learn the delay dynamics for each individual topology, and (iii) identify an appropriate topology transition policy to model the dynamics on time-varying networks. We call this the Markov jump linear system (MJLS) model for airport delays. We develop this delay model for the US, validate it, and demonstrate its superior predictive performance in comparison to other benchmarks. Next, we use this model to identify appropriate interventions that can minimize delays, and develop novel controllers for regulating delays. Our findings suggest that (i) optimal interventions that target highly-connected airports at Atlanta, San Francisco and Chicago can provide maximum systemwide benefits, and (ii) the most effective time for such interventions is between 11 and 2 pm ET. The methods developed in this dissertation can help airlines and air traffic managers improve the efficiency and robustness of the air transportation system. | |
