What is it about?

The new algorithm for inverting seismic surface waves data using an artificial neural network is developed. An approach to the selection of hyperparameters and an architecture of a neural network is described in detail. Examples of the inversion of a large number of synthetic dispersion curves of surface waves (1,250,000 curves) for media with a different number of layers are given, as well as examples of the inversion of synthetic dispersion curves for various geological media using an artificial neuron network, the Monte Carlo method and GWO. Finally, an example of field data validation is provided for a field in western Siberia.

Featured Image

Why is it important?

The use of artificial neural networks makes it possible to obtain accurate inversion results in real-time. It expands the limits of applicability of the MASW method. But there is still no work on creating a complete inversion algorithm, which would include: determining the number of layers, the ranges of possible parameters of the velocity model from the observed dispersion curve, an approach to calculating a representative set of training data, a way to configure the architecture of an artificial neural network.

Perspectives

The accuracy of the ANN is similar to that of the GWO algorithm and outperforms that of Monte Carlo inversion. Moreover, the ANN inversion is more robust in the presence of noisy data and is dramatically faster than the two global search methods when a large number of similar dispersion curves must be processed. The proposed method is especially promising for MASW time-lapse 4D near-surface monitoring. We believe that the considered ANN inversion method should be adapted to include higher-order modes or the whole f-k spectrum of surface waves to improve the accuracy of inversion. Another promising approach is to apply the ANN method to the skeletonized inversion of surface-wave dispersion curves, which can invert lateral velocity variations.

Alexandr Yablokov
Trofimuk Institute of Petroleum Geology and Geophysics of SB RAS

Read the Original

This page is a summary of: An artificial neural network approach for the inversion of surface wave dispersion curves, Geophysical Prospecting, June 2021, Wiley,
DOI: 10.1111/1365-2478.13107.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page