1. Use of Machine Learning in Complexity Calculations

2. Introduction

3. Capacity VS Complexity

4. Complexity
Dynamic Density (DD), A metric that includes both traffic density and trafic complexity, was first introduced in Laudeman, Shelden, Branstron, Brasil, Dynamic Density: An Air Traffic Management Metric. NASA/TM April 1998:
Where:
W is the factor weighting
TD is the Traffic Density
TC are traffic complexity factor
CI is Controller Intent term

5. Complexity
Hilburn B. Cognitive complexity in air traffic control - A literature review. EEC Note 2004; 4(04), Identified 108 factors that contributed to Controller Complexity
1. Aerodromes, number of airline hubs
Cognitive Complexity in Air Traffic Control – A Literature Review Page 65
75. RT, frequency congestion
2. Aerodromes, total number in airspace
EUROCONTROL
76. RT, frequency of hold messages sent to aircraft
3. Aircraft mix climbing and descending
39. Flight entries, number entering per unit time
77. RT, total number of Air-Ground communications
4. Airspace, number of sector sides
40. Flight exits, number aircraft exiting in climb
78. Separation standards (separation/spacing/standards)
5. Airspace, presence/proximity of restricted airspace
41. Flight exits, number aircraft exiting in cruise
79. Staffing
6. Airspace, proximity of sector boundary
42. Flight exits, number aircraft exiting in descent
80. Time, total climb
7. Airspace, sector area
43. Flight Levels, average FL per aircraft
81. Time, total cruise
8. Airspace, sector boundary proximity
44. Flight Levels, difference between upper and lower
82. Time, total descent
9. Airspace, sector shape
45. Flight Levels, number available within sector
83. Traffic density, aircraft per unit volume
10. Airspace, total number of navaids
46. Flight time, mean per aircraft
84. Traffic density, average instantaneous count
11. Conflicts, average flight path convergence angle
47. Flight time, total
85. Traffic density, average sector flight time
12. Conflicts, degree of flight path convergence
48. Flight time, total time in climb
86. Traffic density, localised traffic density / clustering
13. Conflicts, number of aircraft in conflict
49. Flight time, total time in cruise
87. Traffic density, mean distance traveled
14. Conflicts, number of along track
50. Flight time, total time in descent
88. Traffic density, number flights during busiest 3 hours
15. Conflicts, number of crossing
51. Flight type, emergency / special flight operations, number
89. Traffic density, number flights during busiest 30 minutes UROCONTROL Experimental Centre
16. Conflicts, number of opposite heading
52. Flow organisation, altitude, number of altitudes used
17. Conflicts, total time-to-go until conflict, across all aircraft
53. Flow organisation, average flight speed
Cognitive Complexity in Air Traffic Control – A Literature Review Page 66
18. Convergence, presence of small angle convergence routes
54. Flow organisation, complex routing required
19. Coordination, frequency of coordination with other controllers
55. Flow organisation, distribution of Closest Point of Approach
90. Traffic density, number flights per hour
20. Coordination, hand-off mean acceptance time
56. Flow organisation, flow entropy/structure
91. Traffic density, number of arrivals
21. Coordination, hand-offs inbound, total number
57. Flow organisation, geographical concentration of flights
92. Traffic density, number of current aircraft proportional to historical maximum
22. Coordination, hand-offs outbound, total number
58. Flow organisation, multiple crossing points
93. Traffic density, number of departures
23. Coordination, number aircraft requiring hand-off to tower/approach
59. Flow organisation, number of altitude transitions
94. Traffic density, total fuel burn per unit time
24. Coordination, number aircraft requiring vertical handoff
60. Flow organisation, number of current climbing aircraft proportional to historical maximum
95. Traffic density, total number aircraft
25. Coordination, number flights entering from another ATC unit
61. Flow organisation, number of current descending aircraft proportional to historical
96. Traffic distribution/dispersion
26. Coordination, number flights entering from same ATC unit
maximum
97. Traffic mix, aircraft type, jets vs props
27. Coordination, number flights exiting to another ATC unit
62. Flow organisation, number of current level aircraft proportional to historical maximum
98. Traffic mix, aircraft type, slow vs fast aircraft
28. Coordination, number flights exiting to same ATC unit
63. Flow organisation, number of intersecting airways
99. Traffic mix, climbing vs descending
29. Coordination, number of communications with other sectors
64. Flow organisation, number of path changes total
100. Traffic mix, military activity
30. Coordination, number of other ATC units acceptiing hand-offs
65. Flow organisation, routes through sector, total number
101. Traffic mix, number of special flights (med, local traffic)
31. Coordination, number of other ATC units handing off aircraft
66. Flow organisation, vertical concentration
102. Traffic mix, proportion of arrivals, departures and overflights
32. Coordination, total number LOAs
67. Other, controller experience
103. Traffic mix, proportion of VFR to IFR pop up aircraft
33. Coordination, total number of handofffs
68. Other, level of aircraft intent knowledge
104. Weather
34. Coordinations, total number required
69. Other, pilot language difficulties
105. Weather, at or below minimums (for aerodrome)
35. Equipment status
70. Other, radar coverage
106. Weather, inclement (winds, convective activity)
36. Flight entries, number aircraft entering in climb
71. Other, resolution degrees of freedom
107. Weather, proportion of airspace closed by weather
37. Flight entries, number aircraft entering in cruise
72. Procedural requirements, number of required procedures
108. Weather, reduced visibility
38. Flight entries, number aircraft entering in descent UROCONTROL Experimental Centre
73. RT, average duration of Air-Ground communications
Network Capacity and Demand management – NCD
74. RT, callsign confusion potential
One consistent theme throughout this review has been the notion that the transform between complexity and workload is always context specific, and will always contain some statistical noise. Individual differences in decision making and control strategies, or slight differences in interfaces, mean that a generalized complexity index will never reach 100% predictive accuracy in all settings. This should be seen as inevitable, and not necessarily a sign of weakness in the approach.

6. Complexity
Laudeman, Shelden, Branstron, Brasil, Dynamic Density: An Air Traffic Management Metric. NASA/TM April 1998
Prandini, Putta, Hu, A probabilistic measure of air traffic complexity in three-dimensional airspace. International Journal of Adaptive Control and Signal Processing, 2010 ; 00:1-16
Delahaye, Puechmorel, New Trends in Air Traffic Complexity. ENRI International Workshop on ATM/CNS. Tokyo, Japan (EIWAC2009)
Delahaye, Puechmorel, Air traffic complexity based on dynamical systems. 49th IEEE Conference on Decision and Control, Dec 2010, Atlanta, US. Pp
Lee, Feron, Pritchett, Air traffic complexity: An Input-Output Approach. The 7th USA/Europe Air Traffic Management R&D Seminar, 2007 (ATM2007)

7. Controller Differences

8. Workload Measurement

9. Machine Learning
Supervised learning: (also called inductive learning) Training data includes desired outputs. For example, this is spam this is not, learning is supervised.
Unsupervised learning: Training data does not include desired outputs. Clustering is an example.
Semi-supervised learning: Training data includes a few desired outputs.
Reinforcement learning: Rewards from a sequence of actions. Every action has some impact on the environment, and the environment provides feedback that guides the learning algorithm

10. Machine Learning Workload Assesment Complexity Model Prediction Model
Features Vector
Complexity Model
Prediction Model
Evaluation

11. Conclusion
Model has been Trialed with Intrusive work load sensors
Trials show promising results but limited due to the amount of data
Progress on non intrusive workload models are progressing
Generic non intrusive workload models is proving to be difficult
Next trials will commence once a proven non intrusive workload model has been found.

12. Author Biography
Mark Palmer is currently the innovation director for Thales Air Traffic Management business line, leading the innovation across 8 Sites and 7 Countries. Before this role Mark has served as the head of the joint avionics and air traffic management innovation lab and as the technical director for air traffic management in Australia for Thales. Early in his career Mark worked as a software engineer on many projects including the Hawk Leadin Fighter software for BAE Systems.
Franco Basti has a M.S. Degree in Electronic Engineering and Computer Science. He’s currently working for Thales Air Traffic Management in the US as chief architect and innovation lead. He has over 18 years of experience in the air traffic management domain and lead several projects with the FAA on Datalink, SWIM and FIXM. Early in his career Franco lead the tower automation team in Italy and worked on SESAR airports projects and A-CDM initiatives.
Yan Lu is currently working for Thales Air Traffic Management as a systems engineer based in Melbourne Australia. Yan has extensive background in telecommunications and machine learning. She has been working in the Air Traffic Management domain for more than 10 Years.

13. a data-driven solution, providing
decision support for improved aviation operations