Abstract:

Inverse problems are relatively challenging to solve due to inherent illposedness and computational intractability. In this paper we adopt the use of a simulationoptimization approach that couples a numerical simulation model with evolutionary algorithms for solution of the inverse problem. In this approach, the simulation model is solved iteratively during the evolutionary search, which in general can be computationally intensive since several hundreds to thousands of forward model evaluations are typically required for solution. Numerical search methods such as parallel hybrid methods and noisy genetic algorithms are investigated for optimization algorithm improvement. Given the potential computational intractability of such a simulationoptimization approach, grid computing and surrogate models are explored as a means to facilitate computationally tractable solution of such problems. In this paper, the solution of a groundwater inverse problem is explored to test and illustrate the methods. The computational experiments were performed on the National Scientific Foundation's TeraGrid. The results demonstrate the performance of the gridenabled simulationoptimization approach in terms of solution quality and computational performance. A set of preliminary results from ongoing research is discussed.
