Relative Significance of Trajectory Prediction Errors on an Automated Separation Assurance Algorithm

Transcription

1 Relative Significance of Trajectory Prediction Errors on an Automated Separation Assurance Algorithm Todd Lauderdale Andrew Cone Aisha Bowe NASA Ames Research Center

2 Separation Assurance Automation Should detect all conflicts with sufficient time to resolve them Should not suggest resolutions which result in near-term losses of separation Should not resolve false conflicts 2

3 Separation Assurance Automation Should detect all conflicts with sufficient time to resolve them Should not suggest resolutions which result in near-term losses of separation Should not resolve false conflicts If we could perfectly predict the future positions of all aircraft this would be fairly easy 2

4 Prediction Errors Actual Trajectory Predicted Trajectory Any trajectory prediction will have some error Different error sources impact trajectories in different ways 3

5 Impact of Errors Error Correlation Wind errors affect all aircraft in a certain area Cruise speed errors are independent of each other Type of impact Cruise speed errors result in along-track errors Descent profile errors result in altitude errors 4

6 Outline Automated separation assurance Simulation environment Error sources studied Conflict detection results Conflict resolution results 5

7 Automation Objectives To be robust to trajectory prediction errors To be as efficient as possible given a certain amount of prediction error 6

8 Layered Approach TCAS/ACAS (Avoid collisions) Tactical System (Maintain legal sep n) Strategic System (Maintain legal separation) Time to Loss of Separation, minutes 7

9 Layered Approach This Study TCAS/ACAS (Avoid collisions) Tactical System (Maintain legal sep n) Strategic System (Maintain legal separation) Time to Loss of Separation, minutes 7

10 Study Objectives Understand how different sources of trajectory prediction errors affect conflict detection and resolution Compare the relative effects across error sources Highlight algorithmic improvements to mitigate these errors 8

11 Airspace Simulation Airspace Concept Evaluation System (ACES) United States airspace fast-time, gate-to-gate simulation 4D trajectory from departure fix to arrival fix 9

12 Key Simulation Feature Truth Perturbed Every time conflict detection is performed, both a truth and a perturbed prediction are created 10

13 Separation Assurance Algorithm Advanced Airspace Concept (AAC) Autoresolver A strategic separation assurance and problemsolving tool Many recent zero-prediction-error studies 11

14 Error Studies Perform two separate studies: Detection study Resolution study Use 3 hours of United States traffic with over 10,000 flights Study a single source of error at a time 12

15 Error Sources Studied Wind prediction Cruise speed prediction Weight Top of descent Descent speed Resolution maneuver initiation time 13

16 Error Amounts Error Type: Applied: Values: Wind Simulation-Wide -10%,10%, 25% Cruise Speed Per Aircraft ±2%, ±5% Weight Per Aircraft ±10%, ±20% Maneuver Timing Per Maneuver ±20 sec, ±40 sec Top of Descent Per Aircraft ±5 nmi, ±10 nmi Descent Speed Per Aircraft ±5%, ±10% Generally slightly larger than values found in previous studies 14

17 Trajectory Error Examples 15

18 Wind Errors Predicted Trajectory Along-Track Distance (nmi) Actual Time (minutes) 16

19 Cruise-Speed Errors Predicted Trajectory Along-Track Distance (nmi) Actual Time (minutes) 17

20 Weight Errors Altitude (feet)

21 Top-of-Descent Errors Altitude (feet)

22 Descent-Speed Errors Indicated Airspeed (knots)

23 Descent-Speed Errors IAS Time Altitude (feet)

24 Resolution-Maneuver-Initiation-Time Errors Latitude (Degrees) Longitude (Degrees) 21

25 Detection Study 22

26 Detection Study Separation Requirement Ran simulation with no resolutions performed Performed deterministic conflict detection Over 1800 losses of separation 23

27 Conflict Detection Algorithm Predicted trajectories are composed of discrete points 24

28 Conflict Detection Algorithm Predicted trajectories are composed of discrete points Compare points of all trajectories for violations 24

29 Conflict Detection Algorithm Predicted trajectories are composed of discrete points Compare points of all trajectories for violations 24

30 Conflict Detection Algorithm Predicted trajectories are composed of discrete points Compare points of all trajectories for violations 24

31 Missed and False Alerts Missed Alert False Alert Missed alert: Perfect trajectories include a loss and perturbed trajectories do not False alert: Perfect trajectories do not have a loss and perturbed trajectories do 25

32 Detection Results 26

33 Missed Alerts Maximum Error Missed Alerts (%) 10 Minutes to Loss 27

34 Missed Alerts Maximum Error Missed Alerts (%) 10 Minutes to Loss 5 Minutes to Loss 27

35 Wind Errors 25% Missed Alerts (%) 10% -10% 28

36 Weight Error ±20% Missed Alerts (%) ±10% 29

37 Weight Error ±20% Missed Alerts (%) ±10% Late Inflection 29

38 Descent-Speed Errors ±10% Missed Alerts (%) ±5% 30

39 Missed Alerts and False Alerts Maximum Error; 5 Minutes to Loss False Alerts (%) Weight Descent Speed Wind TOD Cruise Speed Missed Alerts (%) 31

40 Missed Alert Reduction Strategy Detection Buffer Separation Requirement Same no-resolution simulations Increased horizontal detection area 32

41 Missed Alerts and Buffer No Buffer With no buffer cruise-speed errors result in most missed 33

42 Missed Alerts and Buffer No Buffer 2 nmi Buffer With no buffer cruise-speed errors result in most missed With buffer TOD and descent-speed errors result in the most missed 33

43 Missed Alerts and False Alerts Maximum Error; 5 Minutes to Loss False Alerts (%) No Buffer 2 nmi Buffer Missed Alerts (%) 34

44 Resolution Study and Results 35

45 Resolution Study Resolution Buffer Detection Requirement 1 nmi detection buffer and 8-minute look-ahead 12-minute look-ahead for successful resolutions Over 4000 conflicts resolved No metering 36

46 Losses of Separation Maximum Error Losses of Separation 37

47 Wind Errors Losses of Separation 38

48 Cruise-Speed Errors Losses of Separation 39

49 Weight Errors Losses of Separation 40

50 Top-of-Descent Errors Losses of Separation ±5 nmi ±10 nmi 41

51 Descent-Speed Errors Losses of Separation 42

52 Maneuver-Initiation-Time Errors Losses of Separation 43

53 Delay Prediction errors result in re-planning of resolutions and resolving non-conflicts This results in additional system-wide delay 44

54 Delay Summary Total Delay (minutes) Baseline 45

55 Conclusions Over 95% of all losses were resolved for all cases Prediction errors result in increased losses and delay for all error types Descent prediction errors result in many late predictions and the largest number of losses The algorithms need to handle descent uncertainty better 46

56 Future Work Modify detection and resolution algorithms for zero losses Study other error sources such as horizontal intent errors Experiment with combinations of errors Optimize algorithms to maximize throughput for a set level of prediction error 47