Fundamental to seismic analysis is an accurate model of the geologic composition of the Earth's crust and upper mantle and the speed with which seismic waves travel through these layers. Each layer has its own velocity parameter, correlated with the velocity rate in the adjacent geologic layers and varying with layer depth. Velocity modeling is a multi-objective optimisation problem with competing objectives typically approached via a single or joint inversion of objective(s). Stochastic modeling of the velocity structure is a relatively new approach in seismology. Benefits include the ability to increase the number of objectives at relatively low cost and to incorporate multiple types of data including magnetotelluric and gravity data. Binary genetic algorithms limit the resolution by forcing the parameters into 5-bit (or similar) chromosomes. Given it is not uncommon to have 30 or more geologic layers, this increases the number of parameters to be estimated and computation time. However, the binary approach provides a more consistent search avoiding many extreme solutions on the Pareto front without additional constraints. On the other hand, real-valued genetic algorithms converge faster and have more consistent models on the Pareto front but require additional objective constraints to avoid producing extreme solutions. Given the potentially many-layered and heterogeneous geologic structure, we explore the choice of binary versus real-valued algorithm in this application.
Keywords: NSGA-II; Genetic algorithms; Geophysical processes
Biography: K.B. Boomer and Richard Brazier are joint collaborators working on statistical modeling of geophysical processes.