1: \begin{abstract}

2: 

3: Air Traffic Control (ATC) is a complex safety critical

4: environment. A tower controller would be making many decisions in

5: real-time to sequence aircraft. While some optimization tools

6: exist to help the controller in some airports, even in these

7: situations, the real sequence of the aircraft adopted by the

8: controller is significantly different from the one proposed by the

9: optimization algorithm. This is due to the very dynamic nature of

10: the environment.

11: 

12: The objective of this paper is to test the hypothesis that one can

13: learn from the sequence adopted by the controller some strategies

14: that can act as heuristics in decision support tools for aircraft

15: sequencing. This aim is tested in this paper by attempting to

16: learn sequences generated from a well-known sequencing method that

17: is being used in the real world.

18: 

19: The approach relies on a genetic algorithm (GA)  to learn these

20: sequences using a society Probabilistic Finite-state Machines

21: (PFSMs). Each PFSM learns a different sub-space; thus, decomposing

22: the learning problem into a group of agents that need to work

23: together to learn the overall problem. Three sequence metrics

24: (Levenshtein, Hamming and Position distances) are compared as the

25: fitness functions in GA. As the results suggest, it is possible to

26: learn the behavior of the algorithm/heuristic that generated the

27: original sequence from very limited information.

28: 

29: \end{abstract}

30: