Magnetotelluric inversion method based on full convolutional neural network

Abstract

The invention discloses a magnetotelluric inversion method based on a full convolutional neural network, and the method comprises the steps: constructing a multi-dimensional geoelectric model, carrying out the forward calculation of the apparent resistivity of a corresponding dimension, forming a sample set, and dividing the sample set into a plurality of training sets and test sets; constructing a full convolutional neural network structure model to obtain an initialized full convolutional neural network model parameter; training and testing the model based on the training set and the test set to obtain optimized full convolutional neural network structure model parameters; determining whether the training of the full convolutional neural network structure model is completed or not according to the fitting error change corresponding to the training set and the test set; finally, inputting the actually measured apparent resistivity into the trained model for inversion, and further optimizing the model by analyzing the precision of an inversion result until an inversion fitting error meets a set error requirement. By using the nonlinear features of the full convolutional neural network, the problem of local extremum in conventional linear inversion is solved, the operation memory and time loss are effectively reduced, and the fitting precision is improved.

Description

technical field [0001] The invention relates to the technical field of magnetotelluric sounding, in particular to a magnetotelluric inversion method based on a fully convolutional neural network. Background technique [0002] The magnetotelluric method (Magnetotelluric Method, MT) is a geophysical exploration method that uses natural alternating electromagnetic fields to study the earth's electrical structure. -4 -10 4 Hz), using the skin effect principle of electromagnetic wave propagation, that is, the high-frequency electromagnetic field penetrates shallowly, and the low-frequency electromagnetic field penetrates deeply. Under the condition that the distance between the field source and the receiving point remains unchanged, the frequency of the electromagnetic field is changed to achieve the purpose of sounding, namely The mutually orthogonal electromagnetic field components are collected on the surface, and the vertical electrical structure information of the undergrou...

AI Technical Summary

Problems solved by technology

Although this kind of nonlinear global optimization inversion method can overcome the local extremum problem and obtain the global optimal solution, it requires a large amount of computing memory and a long computing time

In addition, the network convergence speed used by the artificial neural network method is slow, and the prediction accuracy decreases with the increase of the resistivity data volume and the increase of the model parameters. The position information will be lost during the network transmission process, and the phenomenon of over-fitting is prone to occur during the training process.

Images