A multi-stage genetic algorithm for travel time tomography of flat-layered media

Abstract

Genetic algorithms have long been employed in seismic tomographic inversion to obtain subsurface models from seismic traces, despite their relative lack of accuracy. While most such algorithms are basic in their design, I propose a multi-stage genetic algorithm for flat layer cellular seismic models which exploits the velocity similarities within individual layers. The algorithm starts coarse, with only one velocity value per layer, and gradually increases its granularity to 16 values, accordingly changing the algorithm parameters to reflect the different stages. By reducing the number of model parameters in early stages, the dimension of the search space is also made smaller leading to faster convergence. Although only approximations, the results of these early stages can then be used as improved initial guesses for the later phases of the algorithm. For a similar computational effort, this implementation yields more accurate models than the classic genetic approaches, thus rendering this type of inversion more practical.