CN-121115144-B - Three-dimensional geoelectromagnetic forward method, device, medium and equipment

CN121115144BCN 121115144 BCN121115144 BCN 121115144BCN-121115144-B

Abstract

The invention discloses a three-dimensional geoelectromagnetic forward modeling method, a device, a medium and equipment, which relate to the technical field of geophysics and comprise the steps of generating a three-dimensional geoelectromagnetic model conforming to an actual field geological electrical structure based on a Gaussian random field, performing multiple downsampling operation on the three-dimensional geoelectromagnetic model to extract semantic features of different scales and semantic layers, acquiring a low-level feature map according to the semantic features, distributing different weights to different space regions in the low-level feature map through introducing a space attention fusion mechanism to generate a weighted feature map corresponding to a resistivity abnormal region, performing multiple upsampling operation on the low-level feature map, determining a high-level feature map, and respectively fusing the high-level feature map obtained by each upsampling operation with the weighted feature map of the same scale to determine the fused feature map, and acquiring a predicted value of forward modeling response data according to the fused feature map.

Inventors

ZHONG XIN
LING WEIWEI
PAN KEJIA
LIU CHAOFEI
ZHANG JIAJING
SUN YUAN
ZHAN ZHILIANG
XIAO WENBO

Assignees

江西理工大学
江西应用技术职业学院
中南大学

Dates

Publication Date: 20260508
Application Date: 20250919

Claims (7)

1. A three-dimensional geoelectromagnetic forward method, comprising: Based on the Gaussian random field, generating a three-dimensional magnetotelluric model conforming to an actual field geological electrical structure; performing multiple downsampling operations on the three-dimensional magnetotelluric model to extract semantic features of different scales and semantic layers, and acquiring a low-level feature map according to the semantic features; By introducing a spatial attention fusion mechanism, different weights are distributed to different spatial regions in the low-level feature map, and a weighted feature map corresponding to the resistivity abnormal region is generated; Performing multiple upsampling operations on the low-level feature map, determining a high-level feature map, and fusing the high-level feature map obtained through each upsampling operation with the weighted feature map with the same scale, determining a fused feature map, and acquiring a predicted value of forward modeling response data according to the fused feature map; The predicted value of the forward response data is obtained through a neural network DPGA-Net, and the neural network DPGA-Net specifically comprises: the system comprises three layers of symmetrical encoding modules, decoding modules and a spatial attention fusion module added in a jump connection layer of the encoding modules and the decoding modules; the encoding module is used for performing multiple downsampling operations on the three-dimensional magnetotelluric model so as to extract semantic features of different scales and semantic layers and obtain a low-level feature map; The spatial attention fusion module is used for distributing different weights to different spatial regions in the low-level feature map by introducing a spatial attention fusion mechanism to generate a weighted feature map corresponding to the resistivity abnormal region; the decoding module is used for carrying out up-sampling operation on the low-level feature map for a plurality of times, determining a high-level feature map, fusing the high-level feature map obtained by each up-sampling operation with the weighted feature map with the same scale, determining a fused feature map, and acquiring a predicted value of forward response data according to the fused feature map; the neural network DPGA-Net is trained by adopting a three-dimensional model data set, and the acquisition of the three-dimensional model data set specifically comprises the following steps: Forward modeling calculation is carried out on the three-dimensional magnetotelluric model by adopting a finite element method, forward modeling response data of any direction in the three-dimensional magnetotelluric model is determined, and a three-dimensional model data set is constructed according to the forward modeling response data.
2. The three-dimensional magnetotelluric forward method of claim 1, wherein the forward response data of any direction specifically comprises: Apparent resistivity and phase in the XY and YX directions.
3. The three-dimensional magnetotelluric forward method of claim 1, wherein the acquisition of the gaussian random field comprises: the underground region to be measured is discretized into a three-dimensional uniform grid, noise with a preset power spectrum is constructed in a frequency domain through Fourier transformation, and then inverse transformation is carried out to map the noise to a real space, so that a Gaussian random field is generated.
4. A three-dimensional magnetotelluric forward method as defined in claim 3 wherein said constructing noise with a preset power spectrum in the frequency domain by fourier transformation and then inverse transforming to map it to real space, generating gaussian random fields, comprises: The amplitude of each frequency point in the Fourier space is obtained by adopting the following formula : Wherein, the For the index of frequencies in the momentum space, Is to avoid The time period is divided by zero and, Is a smoothing factor; The complex gaussian noise is generated using the following formula: Wherein, the A standard normal distribution with a mean of 0 and a standard deviation of 1 is represented; For weighted noise An inverse fourier transform is performed to obtain the random field in space.
5. A three-dimensional earth electromagnetic forward modeling device, characterized in that the device implements the three-dimensional earth electromagnetic forward modeling method according to any one of the preceding claims 1-4.
6. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the three-dimensional magnetotelluric forward method of any of the preceding claims 1-4.
7. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the three-dimensional magnetotelluric forward method of any of the preceding claims 1-4 when the program is executed.

Description

Three-dimensional geoelectromagnetic forward method, device, medium and equipment Technical Field The invention relates to the technical field of geophysics, in particular to a three-dimensional geoelectromagnetic forward modeling method, a device, a medium and equipment. Background The magnetotelluric sounding (MT) is used as a geophysical exploration method, and is widely applied to the fields of mineral resource exploration, electrical structure research, oil and gas energy exploration, earth deep structure research and the like due to the characteristics of low cost, convenient construction, large exploration depth range and the like. However, MT forward modeling is a precondition and basis for inversion, and its computation speed and accuracy directly affect inversion performance. The traditional numerical forward modeling method, such as a finite difference method, a finite volume method and a finite element method, can simulate an underground electrical structure with high precision, but is limited by grid division and hardware resources, and particularly under a three-dimensional scene, the calculated amount increases sharply along with the increase of the complexity of a model, so that the calculation efficiency is low, and the requirement of forward modeling of a large-scale complex geological model cannot be met. In recent years, the problem of forward modeling and inversion in an electromagnetic field is being solved by deep learning, but most of researches remain in one-dimensional forward modeling and two-dimensional forward modeling, three-dimensional forward modeling is less, and a constructed data set is mostly an abnormal body with simple rules, so that it is difficult to simulate an actual complex geological environment, and further it is difficult to accurately calculate three-dimensional earth electromagnetic forward modeling response. Disclosure of Invention The invention provides a three-dimensional geoelectromagnetic forward modeling method, a device, a medium and equipment, which are used for solving the problems in the prior art, namely how to improve the calculation efficiency of the three-dimensional geoelectromagnetic forward modeling in the prior art, and the invention provides the three-dimensional geoelectromagnetic forward modeling method which comprises the following steps: Based on the Gaussian random field, generating a three-dimensional magnetotelluric model conforming to an actual field geological electrical structure; performing multiple downsampling operations on the three-dimensional magnetotelluric model to extract semantic features of different scales and semantic layers, and acquiring a low-level feature map according to the semantic features; By introducing a spatial attention fusion mechanism, different weights are distributed to different spatial regions in the low-level feature map, and a weighted feature map corresponding to the resistivity abnormal region is generated; and carrying out up-sampling operation on the low-level feature map for multiple times, determining a high-level feature map, and fusing the high-level feature map obtained by each up-sampling operation with the weighted feature map with the same scale, determining a fused feature map, and acquiring a predicted value of forward modeling response data according to the fused feature map. Optionally, the predicted value of the forward response data is obtained through a neural network DPGA-Net, where the neural network DPGA-Net specifically includes: the system comprises three layers of symmetrical encoding modules, decoding modules and a spatial attention fusion module added in a jump connection layer of the encoding modules and the decoding modules; the encoding module is used for performing multiple downsampling operations on the three-dimensional magnetotelluric model so as to extract semantic features of different scales and semantic layers and obtain a low-level feature map; The spatial attention fusion module is used for distributing different weights to different spatial regions in the low-level feature map by introducing a spatial attention fusion mechanism to generate a weighted feature map corresponding to the resistivity abnormal region; The decoding module is used for carrying out up-sampling operation on the low-level feature map for multiple times, determining the high-level feature map, fusing the high-level feature map obtained through each up-sampling operation with the weighted feature map with the same scale, determining the fused feature map, and obtaining the predicted value of forward response data according to the fused feature map. Optionally, the neural network model is trained by using a three-dimensional model data set, and the acquiring of the three-dimensional model data set specifically includes: Forward modeling calculation is carried out on the three-dimensional magnetotelluric model by adopting a finite element method, forward modeling response data of any direct