An artificial neural network approach for the inversion of surface wave dispersion curves

Abstract

We describe a new algorithm for the inversion of one‐dimensional shear‐wave velocity profiles from dispersion curves of the fundamental mode of Rayleigh surface waves. The novelties of our approach are that the layer velocities and thicknesses are set as unknowns, and an artificial neural network is proposed to solve the inverse problem. We suggest that training data should be calculated for a set of random synthetic velocity layered models, while layer thicknesses and velocities should be set to fixed intervals, with ranges estimated based on the systematic application of empirical relations between Rayleigh and S‐wave velocities to the dispersion data. Our main challenge is a total overhaul of the artificial neural network, which includes selecting the optimal artificial neural network architecture and parameters by performing a large number of numerical experiments. Our synthetic results show that the accuracy of the proposed approach outperforms that of the Monte Carlo approach. We illustrate our proposed method with West Siberia data processing obtained from an area of approximately 800 km2. From a user perspective, the main strength of our method is the computationally efficient processing of large amounts of dispersion data, which make it well suited for four‐dimensional near‐surface monitoring.

Joint inversion of Rayleigh wave fundamental and higher order mode phase velocity dispersion curves using multi‐objective grey wolf optimization

“…In addition to this, several authors have successfully solved complex geophysical inverse problems by sampling the posterior distribution of model parameters using Markov chain Monte Carlo sampling and accepting the candidate solutions having high likelihood between the observed and computed data (Sambridge and Mosegaard, 2002;Lehujeur et al, 2018;Stuart et al, 2019;Figueiredo et al, 2019). Furthermore, the multi-dimensionality and non-linearity of geophysical inverse problems have also been dealt with a number of machine learning techniques, like the use of convolutional neural networks for seismic impedance inversion (Das et al, 2019), surface wave inversion (Hu et al, 2020), use of artificial neural networks for potential field inversion (Kaftan et al, 2014;Purohit et al, 2019), surface wave inversion (Yablokov et al, 2021), layered earth inversion (El-Qady and Ushijima, 2001;Neyamadpour et al, 2009) and the use of random forest regressor for layered earth inversion (Singh et al, 2019). Surface wave inversion is an ill-posed, nonlinear, mixdetermined and multimodal problem (Cox and Teague, 2016).…”

Surface Wave Inversion Method Based on Convolutional Neural Network and Support Vector Regression Machine

“…Combined with geophysics, the U-net network is used to invert the multi-mode dispersion curve to obtain the shear wave velocity Vs model (Fu et al, 2021). A four-layer stratigraphic model is established, a fully-connected neural network, and a one-dimensional Vs model is trained by using the training data set as the input through the dispersion data inversion method based on the arti cial neural network (ANN) algorithm (Yablokov et al, 2021). The 2D-CNN is applied to the surface wave inversion, and the phase velocity and group velocity dispersion curves of Rayleigh waves are inverted into a one-dimensional stratigraphic structure (Hu et al, 2020).…”

Uncertainty quantification of multi-modal surface wave inversion using artificial neural networks

“…Aleardi and Stucchi (2021) train a residual neural network (ResNet) to map the spectral dispersion image of the surface wave into S-wave velocity and layer thicknesses. The advantages of using artificial neural networks (ANN) are higher computational efficiency without a need to adjust optimization parameters and the lack of necessity to include any model constraint into the error function, unlike global optimization methods (Yablokov and Serdyukov, 2020;Aleardi and Stucchi, 2021;Yablokov et al, 2021).…”

“…In this contribution, we focus on selection, adopting and tune-up of several seismic surface waves data processing methods in order to combine them into a novel multimodal surface waves DCs ANN inversion and uncertainties quantification approach that is further tested on both synthetic and real data. We follow the ANN method for the surface wave fundamental mode DCs inversion for S-wave velocity and layer thicknesses suggested by Yablokov et al (2021) and further develop their approach for the multiple mode DCs inversion. First, we define the optimal parametrization (number of geological layers of restoring velocity model) and possible near-surface parameter ranges for the layered model based on the observed frequency-depended phase velocity.…”

“…Then we prepare a training data set by generating numerous random layered models by uniform distribution over the determined parameter intervals and computing DCs using fast parallel forward solver. Due to the verified data-driven approach offered by Yablokov et al (2021), the ANN is trained on a data set containing realistic subsurface scenarios and learns how to reproduce a similar velocity model to fit the input multiple mode DCs data. At current research, we use the rigorous adjusted ANN architecture and parameters proposed by Yablokov et al (2021).…”