Vs Profiling by the Inversion of Rayleigh Wave Ellipticity Curve Using a Hybrid Artificial Intelligence Method

Adequate estimation of S-wave velocity (Vs) structure is a significant parameter in the seismic micro zonation studies. To this purpose, different techniques, such as down-hole measurements and inversion of surface wave’s dispersion curves are proposed for modeling V_S profile. In the last decade, modeling Vs profile from the Rayleigh wave’s ellipticity curve (H/V) is more applicable owing to its rapid and simple data gathering procedure. However, regarding the ambiguities in the inversion of H/V curves, obtaining the reliable results priori information, such as down-hole measurement, to constrain the final Vs model is vital. This study addressed this challenge, and based on a hybrid artificial intelligence method introduced a new technique to invert the Rayleigh wave ellipticity curve with acceptable performance. To do that, first model parameters (i.e. number of layers and corresponding thicknesses and shear wave velocities) by the ensemble of neural networks (ENN) were predicted, and then further inversion by jellyfish searching (JS) algorithm (named ENN-JS inversion method) was carried out to obtain a more reasonable Vs model. To build the ensemble system, ten base networks were arranged. To train the neural networks, synthetic Rayleigh wave ellipticity data by forward modeling approach were generated. The combination of the outputs of based networks was performed using the averaging method. Then, JS inversion algorithm was applied to estimate the final adequate Vs model. ENNs provide essential information to the JS searching algorithm on the number of layers and proper search spaces for model parameters. Synthetic and actual datasets tested the ENN-JS inversion technique. Findings show the proposed method provides a robust method for the inversion of Rayleigh wave ellipticity data.