A parallel genetic algorithm for seismic velocity optimization.

摘要

I investigate a parallel Genetic Algorithm applied to the geophysical inverse problem of seismic travel-time inversion. GAs are search algorithms that mimic the biological processes that control natural selection. The method uses first arrival-times from seismic refraction surveys to invert for the 2-dimensional subsurface velocity field. The parallel implementation follows the “island model,” in which several large subpopulations evolve semi-independently, with periodic migration of individuals between the population “islands.” I design and use modified genetic operators and test our implementation on real data collected from actual field surveys. The parallel GA performs well in terms of both parallel performance and the quality of the optimized velocity models. Comparing with a Simulated Annealing implementation on the same problem, my results qualitatively match the simulated annealing results while taking less time to run.