Inversion of acoustic data using a combination of genetic algorithms and the Gauss-Newton approach

An inversion procedure for obtaining speeds, attenuation, densities, and thicknesses for a layered medium is described. The inversion is carried out using the least-squares technique and the forward modeling is based on SAFARI. The optimization is a hybrid method combining the global genetic algorithms and the local Gauss-Newton method. This is done by taking several gradient steps between each update of the object function for each "individual" in the population. The gradients for the Gauss-Newton method are computed analytically; this makes the computation faster and more stable than computing the gradients by numerical differentiation. The combination of a global and a local method makes the hybrid method faster and it gets closer to the global minimum than a pure global method. Examples based on both real and synthetic data in wave-number-frequency and range-frequency domains show that the method works well.