DDRE/DTU AUGUST 2007 1 Decision Support System for Fighter Pilots Lars Rosenberg Randleff Danish Defence Research Establishment / Technical University.

1DDRE/DTUAUGUST 2007

Decision Support SystemDecision Support Systemfor Fighter Pilotsfor Fighter Pilots

Lars Rosenberg RandleffLars Rosenberg Randleff

Danish Defence Research Establishment /Danish Defence Research Establishment /Technical University of DenmarkTechnical University of Denmark

2DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

3DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

4DDRE/DTUAUGUST 2007

Hostile Environment

5DDRE/DTUAUGUST 2007

Project Description

• To explorer a number of technologies within AI and OR that may be used in designing a DSS for fighter pilots.

• The DSS is to be used in finding a response to ground based threats.

6DDRE/DTUAUGUST 2007

Requirements

• Real-time

• Hardware

• Updateable

• Trustworthy

• Useful

• User Interface

7DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

8DDRE/DTUAUGUST 2007

A Prolog Example

%-----------------% Find appropriate countermeasures.%-----------------proper_cm(Angle, jammer) :- ( warning(rwr, (Angle, _)), warning(mws, (Angle, _)), jammer_mode(auto) );( warning(rwr, (Angle, _)), warning(mws, (DiffAngle, _)), Angle \== DiffAngle );( warning(rwr, (Angle, _)), not(warning(mws, (_, _))) ).

• ”When the jammer is jamming it may influence the RWR and jammers of other aircraft.”

• “When the jammer is in 'auto' mode, it will jam the RF sources detected.”

• If the MWS indicates a missile in a given direction, and the RWR does not, the missile is IR guided.

9DDRE/DTUAUGUST 2007

Some Results

”Use flares, chaff, jammer, or towed decoy. Do not use chaff to counter SA-5 and SA-10. Use jammer in auto mode.”

10DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

11DDRE/DTUAUGUST 2007

The Bayesian Net

12DDRE/DTUAUGUST 2007

Semi-automatic Population

ExcelMATLAB

desc.MATLAB

prog.

HUGIN

TextfileTextfileTextfile

13DDRE/DTUAUGUST 2007

Structural Learning

Fly-In

HUGIN(SL)

TextfileTextfileLogfile

Perl

Textfile

14DDRE/DTUAUGUST 2007

Some Results

15DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

16DDRE/DTUAUGUST 2007

Parameters

α

ρ

17DDRE/DTUAUGUST 2007

Lethality

-1 0 1 2 3 4 5

x 104

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

4x 10

4

-1 0 1 2 3 4 5

x 104

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

4x 10

4

0 50 100 150 200 250 300 350 400 450 5000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Scenario Lethality

18DDRE/DTUAUGUST 2007

Deployment Scheme

Jammer

Decoy

Chaff

19DDRE/DTUAUGUST 2007

Reduction of Lethality (α)

20DDRE/DTUAUGUST 2007

Reduction of Lethality (ρ)

21DDRE/DTUAUGUST 2007

Deployment Phases

I II III IV V

Time

Ot

At

22DDRE/DTUAUGUST 2007

Constraints I

Time

Ot

ttt

t

TtTtt

OOO

O

AAOstartstart

1

1

''

0'

'

At

Ot’

Ot’’

Tstart Tend

23DDRE/DTUAUGUST 2007

Constraints II

Time

At

Tstart

Ot

Ct

starttt TCAT max

0

0

1)1(

1

max

1max

1

t

tt

ttt

tt

C

OTC

COTC

CC

24DDRE/DTUAUGUST 2007

Constraints III

Time

Ot

At

Ot’

Ot’’

Tstart Tend

''1

'1

tTt

t

Tt

tjstartj

OA

OTO

end

start

)''1(

''

1

1

t

Tt

tjendj

t

Tt

tjendj

OTO

OTA

end

end

25DDRE/DTUAUGUST 2007

Scenarios

sc1sc2sc3sc4sc5sc6

26DDRE/DTUAUGUST 2007

Running Time

27DDRE/DTUAUGUST 2007

Results

28DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

29DDRE/DTUAUGUST 2007

Three Metaheuristics

• Steepest Ascent

• Local Search

• Simulated Annealing

30DDRE/DTUAUGUST 2007

Parameters in SA

• Algorithm• Cooling Schedule• Start Temperature• End Temperature• Acceptance• Stopping Criteria• Objective Function• Neighbourhood• Initial Solution

31DDRE/DTUAUGUST 2007

32DDRE/DTUAUGUST 2007

Results

33DDRE/DTUAUGUST 2007

Comparing Metaheuristics

34DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

35DDRE/DTUAUGUST 2007

Comparing Approaches I

• Prolog is easy to use, and easy to model

• BN is not categorical; it treats probabilities

• IP ensures the optimal solutions; but is too slow for a real-time DSS

• A metaheuristic will not necessarily give an optimal solution; but it will find a solution fast

36DDRE/DTUAUGUST 2007

Comparing Approaches II

Prolog BayesianNet

MathematicalModel

Metaheuristic

37DDRE/DTUAUGUST 2007

Agenda

• Project Description

• The Four Approaches• Prolog• Bayesian Net• Integer Programming• Metaheuristics

• Comparing Approaches

• Conclusion

38DDRE/DTUAUGUST 2007

Conclusion

• Four approaches to a DSS for fighter pilots have been implemented

• All approaches have pros and cons

• None of the approaches fulfil all requirements

• More research may find the best technology for developing a DSS for fighter pilots

39DDRE/DTUAUGUST 2007

Decision Support SystemDecision Support Systemfor Fighter Pilotsfor Fighter Pilots

Lars Rosenberg RandleffLars Rosenberg Randleff

Danish Defence Research Establishment /Danish Defence Research Establishment /Technical University of DenmarkTechnical University of Denmark