SOFTWARE
| REFR-REFL_SeisTOMO | A nonlinear
method is used to compute first seismic arrivals and reflection travel
times for velocity models with complex reflector geometry. The method
uses a combination of refraction and reflection travel-times for
simultaneous determination of velocity and interface depth of the model.
The elastic waves assumed to be transmitted or reflected at interfaces
at which the raypaths satisfy Snell’s law. The travel times of
critically refracted waves and derivative matrix are computed by
applying a revised ray bending method, supplemented by an approximate
computation of the first Fresnel zone at each point of the ray.
Travel-times and raypaths of reflected waves are computed by applying a
fast finite-difference scheme based on a solution to the eikonal
equation and following the travel-time gradient backward from the
receiver to the source, respectively. A damped least squares inversion
scheme is used to reconstruct the velocity model above the reflector and
the geometry of the interface by minimizing the difference between
observed and calculated (direct, refracted and reflected) travel-times.
In order to reduce inversion artifacts both damping and smoothing
regularization factors are applied. The applicability of the proposed
method is tested using synthetic data. This simultaneous inversion
scheme is appropriate for static seismic corrections, as well as
determination of lateral velocity variations and reflector’s geometry. This algorithm presented in detail at the followings publications,
|
| HybridGenetic_SeisTOMO |
The application of hybrid genetic algorithms in seismic
tomography is examined and the efficiency of least-squares and genetic
methods as representative of the local and global optimization,
respectively, is presented and evaluated. The robustness of both
optimization methods has been tested and compared for the same
source-receiver geometry and characteristics of the model structure
(anomalies, etc.). In order to solve the forward modeling and estimate
the traveltimes, revisited ray bending method was used supplemented by
an approximate computation of the first Fresnel volume. The root mean
square (RMS) error as the misfit function was used and calculated for
the entire random velocity model for each generation. After the end of
each generation and based on the misfit of the individuals (velocity
models), the selection, crossover and mutation (typical process steps of
genetic algorithms) operators take place continuing the evolution theory
and coding the new generation. To optimize the computation time, since
the whole procedure, the
MATLAB Distributed
Computing Engine (MDCE) was used in a multicore engine. During the
tests, the initially fast convergence of the algorithm (typically first
5 generations) is followed by progressively slower improvements of the
reconstructed velocity models. Therefore, to improve the final
tomographic models, a hybrid genetic algorithm (GA) approach was adopted
by combining the GAs with a local optimization method after several
generations, on the basis of the convergence of the resulting models.
This approach is shown to be efficient, as it directs the solution
search towards a model region close to the global minimum. This algorithm presented in detail at the followings publications,
|
I expect questions in this email (soupios@chania.teicrete.gr).