| Popis: |
Two methods for objective function minimization based on the idea of estimating the optimum gradient and on global search were compared in performing numerical and real data inversions. The required calculation of the gradient is based here on the numerical estimation of the reflection coefficient derivatives with respect to the model parameters. This results in a cheap computation method usable in specialized applications where quantitative EM analysis is required. It is shown that local optimization techniques allow for solving the inverse problem involving many layers that is intractable with global optimization techniques. It is remarkable that in the implemented smooth model, where the function G xx ↑(b, ω) and the reflection coefficients depend on smoothly varying model parameters, the gradient technique performs as well as the global minimization, with a considerable advantage in time inversion. The same advantage is observed in real data inversion. |
