Interactive Evolutionary Computation in Geology

1. Interactive Evolutionary Computation in Geology Chris Wijns, Louis Moresi, Fabio Boschetti, Alison Ord, Brett Davies, Peter Sorjonen-Ward

2. Outline • IEC intro • Geological examples • Conclusion

3. IEC Scheme

4. Geological Examples • Extension of the earth’s crust • Folding of rock layers • Subduction of oceanic crust

5. Extension of the Earth’s Crust • We wish to model faults which arise as a consequence of extension Vein and breccia systems develop above advancing fold and thrust belt (B.Davies, Normandy Mining)

6. Goal • After much field work, a geologist draws a picture of what may have happened • Can this be realised by a numerical model following the laws of physics?

7. Problem Starting model of the earth’s crust Combination of material parameters? Behaviour observed (and sketched) by the geologist

8. Non-linear Physics • Many variables may affect the result: • Crustal strength • Strength dependence with depth • Strength dependence with stress • Maximum crustal weakening with increasing stress • etc.

9. Complication • No numerical target for the evaluation of a geological cross-section • Two numerically similar outputs may have qualitative differences which are unacceptable to the geologist

10. Complication • We rely on the expert evaluation of model results • We would like a computationally rigorous and effective way to optimise our search for parameters

11. Solution • IEC lets us capitalise on the knowledge of an expert user • A genetic algorithm lets us optimise our search in parameter space • The simulation outputs of each generation are ranked by the geologist

12. IEC Scheme

13. Simplified Visual Target

14. Folding of Rock Layers • We wish to model folding with a double wavelength

15. Problem Starting configuration Physical properties? Double-wavelength folding (i.e. a second wavelength different from the initial perturbation)

16. Illustration Initial layers with slight perturbation Two wavelengths of folding are present

17. Folding Variables • Layer viscosities • Layer yield strengths • Layer thicknesses

18. Results of 5th Generation • Ranking is simply according to presence or absence of double wavelength

19. Final Parameters • Analysis involves sifting through all generations to collect statistics of the parameters • Conclusion: at least one strong layer is needed (high viscosity and yield stress) and one substantially weaker layer

20. Subduction of Oceanic Crust • We wish to model oceanic crust subducting under a continent

21. Subduction Variables • Strength of continental/oceanic crust • Convergence rate • Depth of sedimentary wedge

22. Problem Parameter or parameter combination? Subducting slab: • remains attached to the continent • detaches at depth animations

23. Accumulating Models • 5 generations, population 10 • Rank both end-members highly in order to accumulate many models

24. Visualisation of Parameters All generations Slab behaviour Red=attached Black=detached

25. Visualisation of Parameters All generations Slab behaviour Red=attached Black=detached

26. Visualisation of Parameters All generations Slab behaviour Red=attached Black=detached

27. Final Parameters • Attached slab: • Parameter 4 = convergence rate = varied • Parameter 5 = continent viscosity = low • Detached slab: • Parameter 4 = convergence rate = high • Parameter 5 = continent viscosity = high

28. Conclusions • Using IEC-based inversion, we have recovered initial parameter combinations which lead to geological behaviour which is observed in the field

29. Conclusions The IEC technique (1) is time-effective, (2) produces high quality results, (3) is well-suited to geological problems

30. End

31. GA Coding • Pop size = 8 • Type of crossover = uniform • Crossover rate = 0.9 • Mutation rate = 0.1 • Coding type = real code • Maximum generations = 6