Joint inversion method of time-frequency electromagnetic and magnetotelluric based on deep learning

Abstract

The invention discloses a time frequency electromagnetic and magnetotelluric joint inversion method based on deep learning. The method comprises the following steps: (a) establishing a neural networkN which comprises three deep convolutional sub-networks NMT, NF and NT, combining data of m channels output by the NMT, NF and NT to form joint data, connecting with an intermediate transition layer FA, FA is further connected with a deep convolutional network NL which processes the joint data, and NL is further connected with an output layer; (b) acquiring magnetotelluric and time frequency electromagnetic training and verification data bodies; (c) inputting a magnetotelluric response data body DMT into the NMT, inputting a frequency domain response data body DF into the NF, and inputting a time domain response data body DT into the NT; (d) performing convolution and pooling calculation on the data by the deep convolutional sub-network NMT, the NF and the NT, and then outputting the datato the intermediate transition layer FA; (e) performing full connection or convolution-pooling calculation on the data by the intermediate transition layer FA, and then outputting the data to the deepconvolutional network NL; and (f) performing calculation processing on the data by the deep convolutional network NL, and then outputting m values by the output layer.

Description

technical field [0001] The invention relates to the field of new joint inversion methods of geophysical electromagnetic methods, in particular to a joint inversion method of time-frequency electromagnetic and magnetotelluric based on deep learning. Background technique [0002] The magnetotelluric method is an electromagnetic exploration method based on natural field sources. Its theory is relatively simple, its exploration depth is large, and its cost is low, so it is widely used. The disadvantages are that the resolution is not high, the data inversion is multi-solution, and the quality of the observation data is also affected by the signal change of the natural field, the signal is weak, and the signal-to-noise ratio is low. In order to overcome the problems of the magnetotelluric method, active electromagnetic methods can be used as supplements, such as the controlled source electromagnetic method, which can be regarded as a plane wave in the far field region, and the pr...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problems existing in the prior art, the present invention provides a time-frequency electromagnetic and magnetotelluric joint inversion method based on deep learning, which establishes an adaptive neural network for time-frequency electromagnetic and magnetotelluric, and solves the problem of time-frequency electromagnetic The problem of joint inversion with magnetotellurics, and can input various types of data used in geophysical joint inversion, can directly apply deep learning method, solves the processing of various types of massive input data and the depth of multi-type data Convolutional Network Construction Design

Images