PHASE VELOCITY INVERSION USING GENETIC ALGORITHMS

Abstract

The applicability of genetic algorithms in an inversion of surface-wave phase velocities to infer an S-wave velocity profile was examined. S-wave velocities and thicknesses are coded to binary strings, and an individual model is generated by connecting all strings for a subsurface structure model. The fitness is defined by squared summation of differences between observed and calculated phase velocities. In addition to three basic genetic processes of selection, crossover, and mutation, the algorithm was improved by the introduction of elite selection rule and dynamic mutation in which the mutation probability varied according to the variety of individuals in a generation. We applied the method to the inversions of synthetic phase velocities and phase velocities from an actual array observation of long-period microtremors. These results indicate that the GA is highly applicable in phase velocity inversion.