Three-dimensional finite difference forward modeling method and system for controlled source audio frequency magnetotelluric method

Abstract

The present invention relates to a three-dimensional finite difference forward modeling method and system of a controlled source audio frequency magnetotelluric method, wherein the method obtains the control equations of electric field and magnetic field in a medium; solves the control equations to obtain a linear equation group; the linear equation group includes a sparse matrix; stores and solves the linear equation group based on the SQMR method to obtain an analytical result; obtains characteristic information data of the analytical result; the characteristic information data includes apparent resistivity and phase response; calculates the weight value corresponding to the characteristic information data, and obtains a characteristic vector according to the weight value and the characteristic information data; inputs the characteristic vector into a typical model selection network to obtain a typical model; and verifies and simulates the analytical result according to the typical model. Thus, the simulation effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of controlled source audio frequency magnetotelluric detection, and in particular to a three-dimensional finite difference forward modeling method and system of controlled source audio frequency magnetotelluric detection. Background Art

[0002] Controlled Source Audio Frequency Magnetotelluric Method (CSAMT) is a new geophysical technology developed since the 1980s. It has controllable artificial sources and strong anti-interference capabilities. This method has been applied to water resource exploration, metal mine exploration, geological survey and other fields. Inversion is an important means of CSAMT data processing, and its effect determines the accuracy of the final interpretation results. Traditional inversion methods include Bostick inversion, OCCAM inversion, least squares inversion, fast relaxation inversion, nonlinear conjugate gradient inversion, etc. They are linear or local linear methods. These methods rely on the initial model and are prone to fall into local optimal solutions. In addition, the inversion may fail due to the appearance of ill-conditioned matrices during the inversion process.

[0003] Since the intensity of the transmitting source is significantly enhanced compared to the signal intensity of the natural field source, the exploration resolution capability for the target body is improved, and the time required to observe the electromagnetic field value is greatly reduced, so this method has high production efficiency. However, due to the introduction of the field source, compared with the magnetotelluric sounding method, it is more difficult to simulate numerically, especially the simulation of three-dimensional problems. At the same time, the introduction of the source also brings some other problems, such as near-field effect, shadow effect and copy effect. How to improve the application effect of the CSAMT method and overcome the interpretation error caused by the field source effect is the current research hotspot in this field and also the research difficulty. Therefore, it is very necessary to develop two-dimensional and three-dimensional controlled source electromagnetic method numerical simulation for three-dimensional field sources.

[0004] The prior art discloses the theory and formula derivation of the three-dimensional staggered grid finite difference method starting from Maxwell's equations. In order to eliminate the singularity of the field value at the source point, the electric field is divided into a primary field and a secondary field and calculated separately. The primary field is solved by an analytical method, and the second-order partial differential equation of the secondary field is discretized by finite difference using Yee's staggered grid. In order to save memory space, the sparse symmetric complex coefficient linear equations are stored in a row compression format (CSR) and solved using the symmetric quasi-minimum residual method (SQMR). Finally, the correctness of the method is verified by a typical three-layer model. However, the typical three-layer model selection process has not been accurately discussed and studied, resulting in unsatisfactory simulation results of the method. Summary of the invention

[0005] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a three-dimensional finite difference forward modeling method and system for a controlled source audio frequency magnetotelluric method to solve the problem of poor simulation effect in the prior art.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A three-dimensional finite difference forward modeling method for a controlled source audio frequency magnetotelluric method, comprising:

[0008] Obtain the governing equations for electric and magnetic fields in a medium;

[0009] Solving the control equation to obtain a linear equation system; the linear equation system includes a sparse matrix;

[0010] The linear equations are stored and solved based on the SQMR method to obtain analytical results;

[0011] Acquiring characteristic information data of the analysis result; the characteristic information data includes apparent resistivity and phase response;

[0012] Calculating a weight value corresponding to the feature information data, and obtaining a feature vector according to the weight value and the feature information data;

[0013] Inputting the feature vector into a typical model selection network to obtain a typical model;

[0014] The analytical results are verified and simulated according to the typical model.

[0015] Preferably, the control equations for obtaining the electric field and the magnetic field in the medium include:

[0016] Collect preset angular frequency, magnetic permeability and formation conductivity;

[0017] Constructing an electric field control equation according to the angular frequency and the magnetic permeability;

[0018] A magnetic field control equation is constructed according to the conductivity; the control equation includes the electric field control equation and the magnetic field control equation.

[0019] Preferably, the method for determining the typical model selection network is:

[0020] Obtain feature information data of the analysis results to be trained;

[0021] Performing category labeling on the feature information data of the to-be-trained parsing result to obtain labeling information;

[0022] Determine a first training set and a second training set; the first training set includes the apparent resistivity and corresponding annotation information in the characteristic information data of the mineral to be trained; the second training set includes the phase response and corresponding annotation information in the characteristic information data of the mineral to be trained;

[0023] Calculating a weight value of the apparent resistivity in the first training set, and multiplying the apparent resistivity in the first training set by the weight value of the apparent resistivity to obtain a first eigenvector;

[0024] Calculating a weight value of a phase response in the second training set, and multiplying the phase response by a corresponding weight value to obtain a second eigenvector; the eigenvector includes the first eigenvector and the second eigenvector;

[0025] Construct convolutional neural network modules and fully connected network modules;

[0026] Inputting the first feature vector and the annotation information in the first training set into the convolutional neural network module for training;

[0027] The second feature vector and the annotation information in the second training set are input into the fully connected network module for training, and the trained convolutional neural network module and the fully connected network module are merged and input into a fully connected neural network to obtain a trained typical model selection network.

[0028] Preferably, the step of inputting the first feature vector and the label information in the first training set into the convolutional neural network module for training includes:

[0029] Determine a loss function according to the first feature vector and the annotation information in the first training set;

[0030] With the goal of minimizing the loss function, the convolutional neural network module is trained using a gradient descent optimization algorithm to obtain a trained convolutional neural network.

[0031] A three-dimensional finite difference forward modeling system of a controlled source audio frequency magnetotelluric method, comprising:

[0032] An equation acquisition module is used to obtain the governing equations of the electric and magnetic fields in the medium;

[0033] A first solving module, used for solving the control equation to obtain a linear equation system; the linear equation system includes a sparse matrix;

[0034] A second solving module is used to store and solve the linear equations based on the SQMR method to obtain an analytical result;

[0035] A feature acquisition module, used to acquire feature information data of the analysis result; the feature information data includes apparent resistivity and phase response;

[0036] A vector determination module, used to calculate a weight value corresponding to the feature information data, and obtain a feature vector according to the weight value and the feature information data;

[0037] A model acquisition module, used for inputting the feature vector into a typical model selection network to obtain a typical model;

[0038] A simulation module is used to verify and simulate the analytical results according to the typical model.

[0039] Preferably, the equation acquisition module comprises:

[0040] A collection unit, used for collecting preset angular frequency, magnetic permeability and conductivity of the formation;

[0041] A first equation building module, used for building an electric field control equation according to the angular frequency and the magnetic permeability;

[0042] The second equation building module is used to build a magnetic field control equation according to the conductivity; the control equation includes the electric field control equation and the magnetic field control equation.

[0043] Preferably, the model acquisition module includes:

[0044] A feature acquisition unit, used to acquire feature information data of the analysis result to be trained;

[0045] A labeling unit, used for labeling the feature information data of the to-be-trained parsing result by category to obtain labeling information;

[0046] A training set determination unit, used to determine a first training set and a second training set; the first training set includes the apparent resistivity and corresponding annotation information in the characteristic information data of the mineral to be trained; the second training set includes the phase response and corresponding annotation information in the characteristic information data of the mineral to be trained;

[0047] a first calculation unit, configured to calculate a weight value of the apparent resistivity in the first training set, and multiply the apparent resistivity in the first training set by the weight value of the apparent resistivity to obtain a first eigenvector;

[0048] A second calculation unit, configured to calculate a weight value of a phase response in the second training set, and multiply the phase response by a corresponding weight value to obtain a second eigenvector; the eigenvector includes the first eigenvector and the second eigenvector;

[0049] Construction unit, used to construct convolutional neural network modules and fully connected network modules;

[0050] A first training unit, configured to input the first feature vector and the annotation information in the first training set into the convolutional neural network module for training;

[0051] The second training unit is used to input the second feature vector and the annotation information in the second training set into the fully connected network module for training, and merge the trained convolutional neural network module and the fully connected network module into a fully connected neural network to obtain a trained typical model selection network.

[0052] Preferably, the first training unit includes:

[0053] A loss function determination subunit, used to determine a loss function according to the first feature vector and the annotation information in the first training set;

[0054] The gradient training subunit is used to train the convolutional neural network module using a gradient descent optimization algorithm with the goal of minimizing the loss function to obtain a trained convolutional neural network.

[0055] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0056] The present invention obtains the control equations of the electric field and magnetic field in the medium; solves the control equations to obtain a linear equation group; the linear equation group includes a sparse matrix; stores and solves the linear equation group based on the SQMR method to obtain an analytical result; obtains characteristic information data of the analytical result; the characteristic information data includes apparent resistivity and phase response; calculates the weight value corresponding to the characteristic information data, and obtains a characteristic vector according to the weight value and the characteristic information data; inputs the characteristic vector into a typical model selection network to obtain a typical model; and verifies and simulates the analytical result according to the typical model. Thus, the simulation effect is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0058] Figure 1 A flow chart of a method in an embodiment of the present invention;

[0059] Figure 2This is a module connection diagram in the embodiment provided by the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0062] The terms "first", "second", "third" and "fourth" in the specification and claims of the present application and the drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a series of steps, processes, methods, etc. are not limited to the listed steps, but may optionally include steps that are not listed, or may optionally include other step elements inherent to these processes, methods, products or devices.

[0063] The purpose of the present invention is to provide a three-dimensional finite difference forward modeling method and system for a controlled source audio frequency magnetotelluric method, so as to solve the problem of poor simulation effect in the prior art.

[0064] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] See also Figure 1 The present invention provides a three-dimensional finite difference forward modeling method of a controlled source audio frequency magnetotelluric method, comprising:

[0066] Step 100: Obtaining the governing equations of the electric field and magnetic field in the medium;

[0067] Step 200: Solving the control equation to obtain a linear equation system; the linear equation system includes a sparse matrix;

[0068] Step 300: storing and solving the linear equations based on the SQMR method to obtain analytical results;

[0069] Step 400: Acquire characteristic information data of the analysis result; the characteristic information data includes apparent resistivity and phase response;

[0070] Step 500: Calculate the weight value corresponding to the feature information data, and obtain a feature vector according to the weight value and the feature information data;

[0071] Step 600: inputting the feature vector into a typical model selection network to obtain a typical model;

[0072] Step 700: Verify and simulate the analysis results according to the typical model.

[0073] Preferably, the control equations for obtaining the electric field and the magnetic field in the medium include:

[0074] Collect preset angular frequency, magnetic permeability and formation conductivity;

[0075] Constructing an electric field control equation according to the angular frequency and the magnetic permeability;

[0076] A magnetic field control equation is constructed according to the conductivity; the control equation includes the electric field control equation and the magnetic field control equation.

[0077] Specifically, the control equations in the present invention are: ▽×E=iωμH;▽×H=J S +σE. Where ω is the angular frequency, μ is the magnetic permeability in vacuum, and σ is the electrical conductivity of the formation. In actual exploration, the signal transmission frequency generally does not exceed 10000Hz, so the quasi-steady field assumption is applied and the displacement current is ignored.

[0078] Preferably, the method for determining the typical model selection network is:

[0079] Obtain feature information data of the analysis results to be trained;

[0080] Performing category labeling on the feature information data of the to-be-trained parsing result to obtain labeling information;

[0081] Determine a first training set and a second training set; the first training set includes the apparent resistivity and corresponding annotation information in the characteristic information data of the mineral to be trained; the second training set includes the phase response and corresponding annotation information in the characteristic information data of the mineral to be trained;

[0082] Calculating a weight value of the apparent resistivity in the first training set, and multiplying the apparent resistivity in the first training set by the weight value of the apparent resistivity to obtain a first eigenvector;

[0083] Calculating a weight value of a phase response in the second training set, and multiplying the phase response by a corresponding weight value to obtain a second eigenvector; the eigenvector includes the first eigenvector and the second eigenvector;

[0084] Construct convolutional neural network modules and fully connected network modules;

[0085] Inputting the first feature vector and the annotation information in the first training set into the convolutional neural network module for training;

[0086] The second feature vector and the annotation information in the second training set are input into the fully connected network module for training, and the trained convolutional neural network module and the fully connected network module are merged and input into a fully connected neural network to obtain a trained typical model selection network.

[0087] As an optional implementation, the classifier pre-training unit before the convolutional neural network module includes a neural network training data set generation module, which pre-processes the optical characteristics in the characteristic information data of the mineral to be trained for training the neural network and creates a first training set, and the convolutional neural network training module uses the first training set and annotation information output by the neural network training data set generation module as input to perform neural network calculations.

[0088] Preferably, the step of inputting the first feature vector and the label information in the first training set into the convolutional neural network module for training includes:

[0089] Determine a loss function according to the first feature vector and the annotation information in the first training set;

[0090] With the goal of minimizing the loss function, the convolutional neural network module is trained using a gradient descent optimization algorithm to obtain a trained convolutional neural network.

[0091] Figure 2 The module connection diagram in the embodiment provided by the present invention is as follows: Figure 2 As shown, this embodiment also provides a three-dimensional finite difference forward modeling system of a controllable source audio frequency magnetotelluric method, comprising:

[0092] An equation acquisition module is used to obtain the governing equations of the electric and magnetic fields in the medium;

[0093] A first solving module, used for solving the control equation to obtain a linear equation system; the linear equation system includes a sparse matrix;

[0094] A second solving module is used to store and solve the linear equations based on the SQMR method to obtain an analytical result;

[0095] A feature acquisition module, used to acquire feature information data of the analysis result; the feature information data includes apparent resistivity and phase response;

[0096] A vector determination module, used to calculate a weight value corresponding to the feature information data, and obtain a feature vector according to the weight value and the feature information data;

[0097] A model acquisition module, used for inputting the feature vector into a typical model selection network to obtain a typical model;

[0098] A simulation module is used to verify and simulate the analytical results according to the typical model.

[0099] Preferably, the equation acquisition module comprises:

[0100] A collection unit, used for collecting preset angular frequency, magnetic permeability and conductivity of the formation;

[0101] A first equation building module, used for building an electric field control equation according to the angular frequency and the magnetic permeability;

[0102] The second equation building module is used to build a magnetic field control equation according to the conductivity; the control equation includes the electric field control equation and the magnetic field control equation.

[0103] Preferably, the model acquisition module includes:

[0104] A feature acquisition unit, used to acquire feature information data of the analysis result to be trained;

[0105] A labeling unit, used for labeling the feature information data of the to-be-trained parsing result by category to obtain labeling information;

[0106] A training set determination unit, used to determine a first training set and a second training set; the first training set includes the apparent resistivity and corresponding annotation information in the characteristic information data of the mineral to be trained; the second training set includes the phase response and corresponding annotation information in the characteristic information data of the mineral to be trained;

[0107] a first calculation unit, configured to calculate a weight value of the apparent resistivity in the first training set, and multiply the apparent resistivity in the first training set by the weight value of the apparent resistivity to obtain a first eigenvector;

[0108] A second calculation unit, configured to calculate a weight value of a phase response in the second training set, and multiply the phase response by a corresponding weight value to obtain a second eigenvector; the eigenvector includes the first eigenvector and the second eigenvector;

[0109] Construction unit, used to construct convolutional neural network modules and fully connected network modules;

[0110] A first training unit, configured to input the first feature vector and the annotation information in the first training set into the convolutional neural network module for training;

[0111] The second training unit is used to input the second feature vector and the annotation information in the second training set into the fully connected network module for training, and merge the trained convolutional neural network module and the fully connected network module into a fully connected neural network to obtain a trained typical model selection network.

[0112] Preferably, the first training unit includes:

[0113] A loss function determination subunit, used to determine a loss function according to the first feature vector and the annotation information in the first training set;

[0114] The gradient training subunit is used to train the convolutional neural network module using a gradient descent optimization algorithm with the goal of minimizing the loss function to obtain a trained convolutional neural network.

[0115] The beneficial effects of the present invention are as follows:

[0116] The present invention obtains the control equations of the electric field and magnetic field in the medium; solves the control equations to obtain a linear equation group; the linear equation group includes a sparse matrix; stores and solves the linear equation group based on the SQMR method to obtain an analytical result; obtains characteristic information data of the analytical result; the characteristic information data includes apparent resistivity and phase response; calculates the weight value corresponding to the characteristic information data, and obtains a characteristic vector according to the weight value and the characteristic information data; inputs the characteristic vector into a typical model selection network to obtain a typical model; and verifies and simulates the analytical result according to the typical model. Thus, the simulation effect is improved.

[0117] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0118] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A three-dimensional finite difference forward modeling method for controlled source audio frequency magnetotelluric method, characterized in that: include: Obtain the governing equations for electric and magnetic fields in a medium; Solving the control equation to obtain a linear equation system; the linear equation system includes a sparse matrix; The linear equations are stored and solved based on the SQMR method to obtain analytical results; Acquire characteristic information data of the analysis result; The characteristic information data includes apparent resistivity and phase response; Calculating a weight value corresponding to the feature information data, and obtaining a feature vector according to the weight value and the feature information data; Inputting the feature vector into a typical model selection network to obtain a typical model; Verifying and simulating the analytical results according to the typical model; The method for determining the typical model selection network is: Obtain feature information data of the analysis results to be trained; Performing category labeling on the feature information data of the to-be-trained parsing result to obtain labeling information; Determine a first training set and a second training set; the first training set includes the apparent resistivity and corresponding annotation information in the feature information data of the analytical result to be trained; the second training set includes the phase response and corresponding annotation information in the feature information data of the analytical result to be trained; Calculating a weight value of the apparent resistivity in the first training set, and multiplying the apparent resistivity in the first training set by the weight value of the apparent resistivity to obtain a first eigenvector; Calculating a weight value of a phase response in the second training set, and multiplying the phase response by a corresponding weight value to obtain a second eigenvector; the eigenvector includes the first eigenvector and the second eigenvector; Construct convolutional neural network modules and fully connected network modules; Inputting the first feature vector and the annotation information in the first training set into the convolutional neural network module for training; The second feature vector and the annotation information in the second training set are input into the fully connected network module for training, and the trained convolutional neural network module and the fully connected network module are merged and input into a fully connected neural network to obtain a trained typical model selection network.

2. The three-dimensional finite difference forward modeling method of controlled source audio frequency magnetotelluric method according to claim 1 is characterized in that: The control equations for obtaining the electric field and magnetic field in the medium include: Collect preset angular frequency, magnetic permeability and formation conductivity; Constructing an electric field control equation according to the angular frequency and the magnetic permeability; A magnetic field control equation is constructed according to the conductivity; the control equation includes the electric field control equation and the magnetic field control equation.

3. The three-dimensional finite difference forward modeling method of controlled source audio frequency magnetotelluric method according to claim 1 is characterized in that: The step of inputting the first feature vector and the label information in the first training set into the convolutional neural network module for training includes: Determine a loss function according to the first feature vector and the annotation information in the first training set; With the goal of minimizing the loss function, the convolutional neural network module is trained using a gradient descent optimization algorithm to obtain a trained convolutional neural network.

4. A three-dimensional finite difference forward modeling system for controlled source audio frequency magnetotelluric method, characterized in that: include: An equation acquisition module is used to obtain the governing equations of the electric and magnetic fields in the medium; A first solving module, used for solving the control equation to obtain a linear equation system; the linear equation system includes a sparse matrix; A second solving module is used to store and solve the linear equations based on the SQMR method to obtain an analytical result; A feature acquisition module, used to acquire feature information data of the analysis result; The characteristic information data includes apparent resistivity and phase response; A vector determination module, used to calculate a weight value corresponding to the feature information data, and obtain a feature vector according to the weight value and the feature information data; A model acquisition module, used for inputting the feature vector into a typical model selection network to obtain a typical model; A simulation module, used for verifying and simulating the analytical results according to the typical model; The model acquisition module includes: A feature acquisition unit, used to acquire feature information data of the analysis result to be trained; A labeling unit, used for labeling the feature information data of the to-be-trained parsing result by category to obtain labeling information; A training set determination unit, used to determine a first training set and a second training set; the first training set includes the apparent resistivity and corresponding annotation information in the feature information data of the analytical result to be trained; the second training set includes the phase response and corresponding annotation information in the feature information data of the analytical result to be trained; a first calculation unit, configured to calculate a weight value of the apparent resistivity in the first training set, and multiply the apparent resistivity in the first training set by the weight value of the apparent resistivity to obtain a first eigenvector; A second calculation unit, configured to calculate a weight value of a phase response in the second training set, and multiply the phase response by a corresponding weight value to obtain a second eigenvector; the eigenvector includes the first eigenvector and the second eigenvector; Construction unit, used to construct convolutional neural network modules and fully connected network modules; A first training unit, configured to input the first feature vector and the annotation information in the first training set into the convolutional neural network module for training; The second training unit is used to input the second feature vector and the annotation information in the second training set into the fully connected network module for training, and merge the trained convolutional neural network module and the fully connected network module into a fully connected neural network to obtain a trained typical model selection network.

5. The three-dimensional finite difference forward modeling system of controlled source audio frequency magnetotelluric method according to claim 4 is characterized in that: The equation acquisition module comprises: A collection unit, used for collecting preset angular frequency, magnetic permeability and conductivity of the formation; A first equation building module, used for building an electric field control equation according to the angular frequency and the magnetic permeability; The second equation building module is used to build a magnetic field control equation according to the conductivity; the control equation includes the electric field control equation and the magnetic field control equation.

6. The three-dimensional finite difference forward modeling system of controlled source audio frequency magnetotelluric method according to claim 4 is characterized in that: The first training unit comprises: A loss function determination subunit, used to determine a loss function according to the first feature vector and the annotation information in the first training set; The gradient training subunit is used to train the convolutional neural network module using a gradient descent optimization algorithm with the goal of minimizing the loss function to obtain a trained convolutional neural network.

Images