Seismic electromagnetic logging joint inversion method and device based on physical information neural network, electronic equipment and storage medium

Abstract

The invention provides a seismic electromagnetic logging joint inversion method and device based on a physical information neural network, electronic equipment and a storage medium, and relates to the technical field of geophysical inversion, the method comprises the steps that to-be-solved data is acquired, and the to-be-solved data comprises pre-stack seismic data and apparent resistivity data; and inputting the data to be solved into a trained physical information neural network to obtain target geological data, the target geological data including a target longitudinal wave velocity, a target transverse wave velocity, a target density and a target resistivity, the trained physical information neural network comprises a trained earthquake closed-loop network and a trained electromagnetic closed-loop network, and each of the trained earthquake closed-loop network and the trained electromagnetic closed-loop network comprises a forward modeling network and an inversion network. The target geological data is obtained by forming reciprocal constraint through two groups of forward modeling networks and inversion networks. The inversion precision and resolution are improved, and a refined evaluation result of underground resources is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of geophysical inversion, and particularly to a seismic electromagnetic logging joint inversion method, device, electronic device and storage medium based on a physics-informed neural network. Background Art

[0002] Geophysical inversion is an important technology in the field of geophysical exploration. Geophysical inversion uses geophysical observation data to infer physical parameters of underground media. It involves determining the occurrence state (such as shape, attitude, spatial position) of geological bodies and physical property parameters (such as density, magnetism, electricity, elasticity, velocity, etc.) from the distribution of geophysical anomalies. This technology has wide application value in fields such as geological exploration, resource exploration, and engineering survey. Existing technologies generally invert physical parameters of underground media based on a certain type of geophysical data.

[0003] The existing geophysical inversion method, namely the seismic electromagnetic logging joint inversion method based on a physics-informed neural network, will lead to non-uniqueness of inversion, resulting in low inversion accuracy and low resolution. Summary of the Invention

[0004] The present invention provides a seismic electromagnetic logging joint inversion method, device, electronic device and storage medium based on a physics-informed neural network, so as to solve the defects of low inversion accuracy and low resolution in the prior art, and achieve improving inversion accuracy and resolution, and obtaining a refined evaluation result of underground resources.

[0005] The present invention provides a seismic electromagnetic logging joint inversion method based on a physics-informed neural network, including the following steps: Obtain data to be solved, where the data to be solved includes prestack seismic data and apparent resistivity data; Input the data to be solved into a trained physics-informed neural network to obtain target geological data, where the target geological data includes a target P-wave velocity, a target S-wave velocity, a target density, and a target resistivity. The trained physics-informed neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward modeling network and an inversion network. The target geological data is obtained through mutual inverse constraints formed by two groups of forward modeling networks and inversion networks.

[0006] A seismic electromagnetic logging joint inversion method based on a physics-informed neural network according to the present invention. The trained seismic closed-loop network includes an elastic inversion network and a seismic forward network. The elastic inversion network is used to realize the mapping from pre-stack seismic data to logging elastic parameters. The seismic forward network is used to realize the mapping from logging elastic parameters to pre-stack seismic data. The trained electromagnetic closed-loop network includes a resistivity inversion network and an electromagnetic forward network. The resistivity inversion network is used to realize the mapping from apparent resistivity data to logging resistivity data. The electromagnetic forward network is used to realize the mapping from logging resistivity data to apparent resistivity data.

[0007] A seismic electromagnetic logging joint inversion method based on a physics-informed neural network according to the present invention. The data to be solved further includes the initially inverted P-wave velocity and the initially inverted resistivity. The initially inverted P-wave velocity is the propagation velocity of P-waves in the underground medium obtained by processing seismic data through an inversion algorithm. The initially inverted resistivity is the distribution of resistivity in the underground medium obtained by processing electromagnetic data through an inversion algorithm.

[0008] A seismic electromagnetic logging joint inversion method based on a physics-informed neural network according to the present invention. Inputting the data to be solved into the trained physics-informed neural network to obtain target geological data includes: Inputting the initially inverted P-wave velocity into the elastic inversion network in the trained seismic closed-loop network, and inputting the initially inverted resistivity into the resistivity inversion network in the trained electromagnetic closed-loop network, as prior low-frequency data constraints for the trained physics-informed neural network to output the target P-wave velocity and the target S-wave velocity.

[0009] A seismic electromagnetic logging joint inversion method based on a physics-informed neural network according to the present invention. Before obtaining the data to be solved, the method further includes: Determining an initial seismic closed-loop network and an initial electromagnetic closed-loop network. The initial seismic closed-loop network is determined according to the forward and inverse rules of pre-stack seismic data. The electromagnetic closed-loop network is determined according to the electromagnetic forward and inverse rules. Determine the loss function, which includes a logging constraint term, a physical model constraint term, a closed-loop constraint term, an uncertainty constraint term, and a cross-gradient constraint term. The logging constraint term is used to constrain the inversion result to be consistent with the logging parameters. The physical model constraint term is used to optimize the angle wavelet to introduce the information of amplitude varying with offset into the seismic closed-loop network. The closed-loop constraint term is used to simultaneously model with the forward network and the inversion network, introducing a large amount of unlabeled data to participate in the training. The uncertainty constraint term is used to quantify the uncertainty of the network prediction result. The cross-gradient constraint term is used to constrain the similarity between the gradients of multiple inversion results; Train the initial seismic closed-loop network and the initial electromagnetic closed-loop network according to the loss function to obtain the trained physics-informed neural network.

[0010] According to a seismic and electromagnetic logging joint inversion method of a physics-informed neural network provided by the present invention, the training of the initial seismic closed-loop network and the initial electromagnetic closed-loop network according to the loss function to obtain the trained physics-informed neural network includes: Optimize the elastic inversion network and the seismic forward network in the initial seismic closed-loop network, and the resistivity inversion network and the electromagnetic forward network in the initial electromagnetic closed-loop network through the logging constraint term to obtain a first seismic closed-loop network and a first electromagnetic closed-loop network; Optimize the seismic forward network through the physical model constraint term to obtain a second seismic closed-loop network; Introduce unlabeled data, and optimize the elastic inversion network in the second seismic closed-loop network and the resistivity inversion network in the first electromagnetic closed-loop network by using the closed-loop constraint term to obtain a third seismic closed-loop network and a second electromagnetic closed-loop network; Optimize the elastic inversion network in the third seismic closed-loop network and the resistivity inversion network in the second electromagnetic closed-loop network by using the uncertainty constraint term, and quantify the uncertainty of the inversion result to obtain a fourth seismic closed-loop network and a third electromagnetic closed-loop network; Optimize the elastic inversion network in the fourth seismic closed-loop network and the resistivity inversion network in the third electromagnetic closed-loop network by using the cross-gradient constraint term to obtain the trained physics-informed neural network.

[0011] The present invention also provides a seismic and electromagnetic logging joint inversion device of a physics-informed neural network, including the following modules: The first acquisition module is used to acquire the data to be solved, and the data to be solved includes pre-stack seismic data and apparent resistivity data; A second acquisition module, configured to input the data to be solved into a trained physics-informed neural network to obtain target geological data, where the target geological data includes a target P-wave velocity, a target S-wave velocity, a target density, and a target resistivity. The trained physics-informed neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward network and an inverse network. The target geological data is obtained by constructing a reciprocal constraint through two sets of forward networks and inverse networks.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, the seismic and electromagnetic logging joint inversion method of the physics-informed neural network as described in any one of the above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the seismic and electromagnetic logging joint inversion method of the physics-informed neural network as described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the seismic and electromagnetic logging joint inversion method of the physics-informed neural network as described in any one of the above is implemented.

[0015] The seismic and electromagnetic logging joint inversion method, device, electronic device, and storage medium based on a physics-informed neural network provided by the present invention obtain data to be solved, input the data to be solved into a trained physics-informed neural network to obtain target geological data. The trained physics-informed neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward network and an inverse network. The target geological data is obtained by constructing a reciprocal constraint through two sets of forward networks and inverse networks, improving the inversion accuracy and resolution, and obtaining a refined evaluation result of underground resources. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a flowchart of the seismic and electromagnetic logging joint inversion method of the physics-informed neural network provided by the present invention.

[0018] Figure 2 It is a schematic flow chart of the method for obtaining a trained physics-informed neural network provided by the present invention.

[0019] Figure 3 It is a schematic flow chart of the method for training a network according to a loss function provided by the present invention.

[0020] Figure 4 It is a network block diagram of the seismic electromagnetic logging joint inversion method of the physics-informed neural network provided by the present invention.

[0021] Figure 5 It is one of the diagrams showing the inversion effect provided by the present invention.

[0022] Figure 6 It is another diagram showing the inversion effect provided by the present invention.

[0023] Figure 7 It is yet another diagram showing the inversion effect provided by the present invention.

[0024] Figure 8 It is still another diagram showing the inversion effect provided by the present invention.

[0025] Figure 9 It is yet again a diagram showing the inversion effect provided by the present invention.

[0026] Figure 10 It is another diagram showing the inversion effect provided by the present invention.

[0027] Figure 11 It is yet another diagram showing the inversion effect provided by the present invention.

[0028] Figure 12 It is still another diagram showing the inversion effect provided by the present invention.

[0029] Figure 13 It is yet again a diagram showing the inversion effect provided by the present invention.

[0030] Figure 14 It is a schematic structural diagram of the seismic electromagnetic logging joint inversion device of the physics-informed neural network provided by the present invention.

[0031] Figure 15 It is a schematic physical structure diagram of the electronic device provided by the present invention. Detailed implementation manners

[0032] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0033] Geophysical inversion is an important technology in the field of geophysical exploration. Geophysical inversion uses geophysical observation data to infer the physical parameters of underground media. It involves determining the occurrence state (such as shape, occurrence, and spatial position) and physical property parameters (such as density, magnetism, electricity, elasticity, velocity, etc.) of geological bodies from the distribution of geophysical anomalies. This technology has extensive application value in the fields of geological exploration, resource exploration, and engineering survey. Existing technologies generally invert the physical parameters of underground media based on a certain type of geophysical data.

[0034] The existing geophysical inversion method, namely the seismic electromagnetic logging joint inversion method of the physical information neural network, will lead to the multi-solution problem of inversion, resulting in low inversion accuracy and low resolution.

[0035] In view of this, the embodiments of the present invention provide a seismic electromagnetic logging joint inversion method of the physical information neural network. By obtaining the data to be solved, the data to be solved includes pre-stack seismic data and apparent resistivity data; inputting the data to be solved into the trained physical information neural network to obtain target geological data, the target geological data includes target P-wave velocity, target S-wave velocity, target density, and target resistivity. The trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward modeling network and an inversion network. The target geological data is obtained by constructing a reciprocal constraint through two sets of forward modeling networks and inversion networks. This method can improve the inversion accuracy and resolution and obtain a refined evaluation result of underground resources.

[0036] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings in the embodiments of the present invention.

[0037] Figure 1It is a schematic flowchart of the seismic electromagnetic logging joint inversion method of the physical information neural network provided by the present invention. The seismic electromagnetic logging joint inversion method of the physical information neural network can be applied to an electronic device, which can be various types of devices with information processing capabilities during implementation. For example, the electronic device can include a personal computer, a laptop computer, a handheld computer, or a server, etc.; the electronic device can also be a mobile terminal, for example, the mobile terminal can include a mobile phone, an in-vehicle computer, a tablet computer, or a projector, etc. As Figure 1 shown, the method can include the following steps 101 to step 102: Step 101: Obtain the data to be solved, where the data to be solved includes pre-stack seismic data and apparent resistivity data.

[0038] It should be noted that the method for obtaining the data to be solved can be obtained through a collection device or through transmission from other devices. The present invention does not limit the method for obtaining the data to be solved. Exemplarily, the pre-stack seismic data can be obtained through seismic exploration technology, and the apparent resistivity data can be obtained through geological electromagnetic exploration.

[0039] Step 102: Input the data to be solved into the trained physical information neural network to obtain target geological data, where the target geological data includes target P-wave velocity, target S-wave velocity, target density, and target resistivity. The trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward modeling network and an inversion network. The target geological data is obtained through the mutual inverse constraints formed by two sets of forward modeling networks and inversion networks.

[0040] It should be noted that by inputting the data to be solved into the trained physical information neural network, target geological data can be obtained. Among them, the trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward modeling network and an inversion network. The forward modeling network and the inversion network are not independently constructed and trained successively, but are modeled and trained simultaneously in the same framework. During the training process, they influence and interact with each other. Through this mutual inverse constraint between the forward / inversion networks, the entire trained physical information neural network can find a balance between physical laws and actual observations. This balance helps to improve the accuracy and reliability of the model, avoid the inversion results from being inconsistent with physical laws, and at the same time make the forward modeling results more consistent with actual observations.

[0041] It can be understood that the seismic electromagnetic logging joint inversion method of the physics-informed neural network proposed by the present invention realizes a seismic electromagnetic joint inversion method based on the physics-informed neural network. By acquiring seismic data and apparent resistivity data and inputting them into the trained physics-informed neural network, target geological data is obtained. Among them, the trained physics-informed neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network, and both closed-loop networks include a forward network and an inversion network. Through the reciprocal constraints between the forward network and the inversion network, the parameters of the two networks are continuously optimized, so that they converge together under the reciprocal constraints, and more accurate forward simulation and inversion inference results can be obtained. The inversion accuracy and resolution are improved, and a refined evaluation result of underground resources is obtained.

[0042] In some embodiments, the trained physics-informed neural network includes an elastic inversion network and a seismic forward network. The elastic inversion network is used to realize the mapping from pre-stack seismic data to well logging elastic parameters, and the seismic forward network is used to realize the mapping from well logging elastic parameters to pre-stack seismic data. The trained electromagnetic closed-loop network includes a resistivity inversion network and an electromagnetic forward network. The resistivity inversion network is used to realize the mapping from apparent resistivity data to well logging resistivity data, and the electromagnetic forward network is used to realize the mapping from well logging resistivity data to apparent resistivity data.

[0043] It should be noted that the trained seismic closed-loop network includes a seismic forward / inversion part and an electromagnetic forward / inversion part. Among them, the seismic forward / inversion part includes two networks, namely a seismic forward physical model and an elastic parameter inversion model. The seismic forward network is based on the Aki-Richard equation and is used to establish the theoretical physical relationship between the inversion result and the pre-stack angle seismic data. The elastic parameter inversion model is a neural convolutional network and is used to learn the non-linear mapping relationship between the pre-stack angle seismic data and the elastic parameters. The electromagnetic forward / inversion part also includes two networks, namely a resistivity inversion network and an electromagnetic forward network. The resistivity inversion network is used to learn the non-linear mapping relationship from apparent resistivity to resistivity, while the electromagnetic forward network is used to learn the non-linear mapping relationship from resistivity to apparent resistivity data. Among them, the elastic parameter data and the resistivity data are well logging data, and the well logging data is used to label the training data during the model training stage, that is, as label data.

[0044] It can be understood that through the reciprocal constraints formed by the elastic inversion network and the seismic forward network, and the reciprocal constraints of the resistivity inversion network and the electromagnetic forward network, the inversion accuracy and resolution are improved, and a refined evaluation result of underground resources is obtained. In addition, the seismic electromagnetic logging joint inversion method of the physics-informed neural network provided by the present invention integrates geophysical information in the deep network, which can improve the physical interpretability of the inversion result.

[0045] In some embodiments, the data to be solved may further include the initially inverted P-wave velocity and the initially inverted resistivity. The initially inverted P-wave velocity is the propagation velocity of P-waves in the subsurface medium obtained by processing seismic data through an inversion algorithm, and the initially inverted resistivity is the distribution of resistivity in the subsurface medium obtained by processing electromagnetic data through an inversion algorithm.

[0046] It should be noted that using the initially inverted result as the prior low-frequency data constraint can make the inverted geological data more accurate.

[0047] Furthermore, the step of inputting the data to be solved into the trained physics-informed neural network to obtain the target geological data may include: inputting the initially inverted P-wave velocity into the elastic inversion network in the trained seismic closed-loop network, and inputting the initially inverted resistivity into the resistivity inversion network in the trained electromagnetic closed-loop network, so as to serve as the prior low-frequency data constraint for the trained physics-informed neural network to output the target P-wave velocity and the target S-wave velocity.

[0048] It can be understood that for the problems of lack of high-frequency data and small amount of labeled data in seismic-electromagnetic joint inversion, using the forward network model and the inversion network model to jointly form two closed-loop networks can introduce a large amount of unlabeled data to participate in the training, and using the initially inverted result as the prior low-frequency data input can make the inversion result more accurate.

[0049] In some embodiments, before obtaining the data to be solved, the method may further include the step of obtaining a trained physics-informed neural network.

[0050] Figure 2 is a schematic flowchart of the method for obtaining a trained physics-informed neural network provided by the present invention. As Figure 2 shown, before obtaining the data to be solved, the method may further include: Step 201: Determine an initial seismic closed-loop network and an initial electromagnetic closed-loop network. The initial seismic closed-loop network is determined according to the forward and inverse modeling laws of prestack seismic data, and the electromagnetic closed-loop network is determined according to the electromagnetic forward and inverse modeling laws.

[0051] It should be noted that first, a seismic forward / inverse network and an electromagnetic forward / inverse network can be built. Among them, the seismic forward network is based on the Aki-Richard equation.

[0052] Step 202: Determine the loss function, which includes a logging constraint term, a physical model constraint term, a closed-loop constraint term, an uncertainty constraint term, and a cross-gradient constraint term. The logging constraint term is used to constrain the inversion result to be consistent with the logging parameters. The physical model constraint term is used to optimize the angular wavelet to introduce the information of amplitude varying with offset into the seismic closed-loop network. The closed-loop constraint term is used to simultaneously model with the forward network and the inversion network, introducing a large amount of unlabeled data to participate in the training. The uncertainty constraint term is used to quantify the uncertainty of the network prediction result. The cross-gradient constraint term is used to constrain the similarity between the gradients of multiple inversion results.

[0053] It should be noted that the designed loss function may include: logging constraint, physical model constraint, closed-loop constraint, uncertainty constraint, and cross-gradient constraint.

[0054] Exemplarily, the total loss function may include: a logging constraint term, a physical model constraint term, a closed-loop constraint term, an uncertainty constraint term, and a cross-gradient constraint term.

[0055] Among them, the elastic parameter logging constraint term can be expressed as: (1); Among them, respectively represent the logging P-wave velocity, logging S-wave velocity, and logging density, respectively represent the initially inverted P-wave velocity, initially inverted S-wave velocity, and initially inverted density, represents the prestack seismic data, represents the incident angle, represents the seismic inversion network, represents the weight parameter of the seismic inversion network.

[0056] Among them, the resistivity logging constraint term can be expressed as: (2); Among them, represents the logging resistivity, respectively represent the initially inverted resistivity, represents the prestack seismic data, represents the apparent resistivity, represents the electromagnetic inversion network, represents the weight parameter of the electromagnetic inversion network.

[0057] Among them, the physical model constraint term is expressed as: (3); Among them, represents the seismic forward physical model, Represent the weight parameters of the seismic forward network. The physical model of seismic forward is composed of the Aki-Richards approximate equation and three convolutional layers. Assume that , the normalized elastic impedance expression obtained from well logging elastic parameter data is: (4); Among them, represents the normalized elastic impedance at the t-th position of the i-th sample in the first sample set, i represents the sample number, and t represents the position point number. represents the incident angle; , and respectively represent the mean values of the well logging P-wave velocity, well logging S-wave velocity, and well logging density of the i-th sample. , and respectively represent the well logging P-wave velocity, well logging S-wave velocity, and well logging density at the t-th position of the i-th sample.

[0058] The reflection coefficient is obtained according to the elastic impedance data: (5); The forward pre-stack seismic data can be expressed as: (6); Among them, represents the angle wavelet.

[0059] It should be noted that the functional relationships established in equations (4) and (6) are equivalent to , and the number of wavelets is the same as the number of one-dimensional convolutional layers.

[0060] Among them, the closed-loop constraint terms include: seismic forward / backward closed-loop constraint terms and electromagnetic forward / backward closed-loop constraint terms. Among them, the seismic forward / backward closed-loop constraint term is expressed as: (7); Among them, the electromagnetic forward / backward closed-loop constraint term is expressed as: (8); Among them, represents the electromagnetic forward physical model, represents the weight parameters of the electromagnetic forward model. represents the pre-stack angle seismic data without well logging labels, represents the apparent resistivity data without well logging labels.

[0061] Among them, the uncertainty constraint term in seismic inversion is: (9); Among them, is the total number of virtual observations, which is proportional to the inversion error and is used to control the magnitude of uncertainty.

[0062] Correspondingly, the uncertainty constraint term in electromagnetic inversion is: (10); The cross-gradient constraint term can be expressed as: (11); (12); (13); (14); Among them, and respectively represent the parameters of two inversions. There are four inversion parameters, and the cross-gradient can be calculated between every two of them and then summed to obtain the total cross-gradient loss term.

[0063] Step 203: Train the initial seismic closed-loop network and the initial electromagnetic closed-loop network according to the loss function to obtain the trained physics-informed neural network.

[0064] It should be noted that after determining the loss function, the parameters of the two networks can be continuously optimized according to the loss function to make them converge jointly under multiple constraints, so as to obtain more accurate forward simulation and inversion inference results.

[0065] It can be understood that the well logging constraint term can be used to constrain the consistency between the inversion result and the well logging parameters, ensuring the distribution consistency between the inversion result and the well logging value. The seismic forward physical model constraint term uses the Aki-Richards forward equation to construct a seismic forward network, and introduces AVO (Amplitude Versus Offset) information into the network by optimizing the angular wavelet, improving the rationality and physical interpretability of the seismic inversion result. The closed-loop constraint term uses forward and inverse simultaneous modeling to form a reciprocal constraint, introduces unlabeled data to participate in training, and improves the generalization performance of the model. The cross-gradient constraint term constrains the similarity between the gradients of multiple inversion results, ensuring the consistency of the change trend in the spatial structure. The uncertainty constraint term quantifies the uncertainty of the network prediction result, reducing the uncertainty of the model. Among them, AVO (Amplitude Versus Offset) information is used to indicate that the input of the elastic parameter inversion network is prestack angle gather seismic data. The prestack data of these three input channels reflects the information of amplitude changing with offset. Labeled data is used to indicate the data at the well logging location, and unlabeled data is used to indicate the data at non-well logging locations.

[0066] Figure 3It is a schematic flowchart of the method for training a network according to a loss function provided by the present invention. As Figure 3 shown, training the initial seismic closed-loop network and the initial electromagnetic closed-loop network according to the loss function to obtain the trained physics-informed neural network may include: Step 301: Optimize the elastic inversion network and the seismic forward network in the initial seismic closed-loop network, and the resistivity inversion network and the electromagnetic forward network in the initial electromagnetic closed-loop network through the well logging constraint term to obtain a first seismic closed-loop network and a first electromagnetic closed-loop network; Step 302: Optimize the seismic forward network through the physical model constraint term to obtain a second seismic closed-loop network; Step 303: Introduce unlabeled data, and optimize the elastic inversion network in the second seismic closed-loop network and the resistivity inversion network in the first electromagnetic closed-loop network by using the closed-loop constraint term to obtain a third seismic closed-loop network and a second electromagnetic closed-loop network; Step 304: Optimize the elastic inversion network in the third seismic closed-loop network and the resistivity inversion network in the second electromagnetic closed-loop network by using the uncertainty constraint term, and perform uncertainty quantification on the inversion result to obtain a fourth seismic closed-loop network and a third electromagnetic closed-loop network; Step 305: Optimize the elastic inversion network in the fourth seismic closed-loop network and the resistivity inversion network in the third electromagnetic closed-loop network by using the cross-gradient constraint term to obtain the trained physics-informed neural network.

[0067] It should be noted that the present invention optimizes the network model in five steps. The optimization process of the network is divided into five steps. The first step is to simultaneously optimize two forward / inversion network models through the well logging constraint term using formulas (1) and (2). The second step is to optimize the seismic forward network using formula (3). The third step is to introduce unlabeled data and simultaneously optimize two inversion network models using two closed-loop constraint terms (7) and (8). The third step uses the uncertainty constraints of formulas (9) and (10) to further optimize the two inversion networks, perform uncertainty quantification on the network inversion result, and reduce the uncertainty of the model. The fourth step uses the cross-gradient constraint term of formula (11) to further optimize the two inversion networks and ensure the consistency of various inversion results in the spatial structure.

[0068] Among them, the physical model constraint term, that is, formula (3), is a physical constraint, which is used to constrain the seismic inversion network. Formula (7) is a closed-loop constraint. The electromagnetic inversion network has no physical constraint. The third step is carried out after the second step to perform a new round of adjustment on the adjusted model using training data. The fourth step is also carried out after the third step for adjustment, and so on.

[0069] Exemplarily, to improve the stability of the training process, the present invention proposes to train the physics-informed neural network in five stages.

[0070] 1. In the first stage, two logging constraint terms (1) and (2) are used to optimize the elastic inversion network weight parameters and the resistivity inversion network weight parameters , and introduce the high-frequency details of logging.

[0071] 2. In the second stage, the seismic forward physical model constraint (3) is used to optimize the seismic forward network .

[0072] 3. Unlabeled data is introduced, and the closed-loop constraint terms (7) and (8) are used to further optimize the elastic inversion network weight parameters and the resistivity inversion network weight parameters , and improve the generalization performance of the model.

[0073] 4. In the fourth stage, the uncertainty constraint terms (9) and (10) are used to further optimize the elastic inversion network weight parameters and the resistivity prediction network weight parameters , and reduce the uncertainty of the model.

[0074] 5. In the fifth stage, the cross-gradient constraint (11) is used to fine-tune the elastic inversion network weight parameters and the resistivity inversion network weight parameters , and ensure the consistency of the variation trends of multiple inversion parameters in the spatial structure.

[0075] It can be understood that the present invention uses five constraint terms and trains the initial neural network in five stages based on the five constraint terms to obtain the trained physics-informed neural network, which can improve the inversion accuracy and resolution, the generalization performance of the network model, and the physical interpretability of the model.

[0076] In addition, the initially inverted P-wave velocity can be used as one of the input channels of the seismic inversion network, and the initially inverted resistivity can be used as one of the input channels of the resistivity inversion network. Both of them are used as prior low-frequency data constraints, making the inversion modeling process easier to converge.

[0077] The seismic electromagnetic logging joint inversion method of the physics-informed neural network provided by the present invention, by obtaining the trained network model, after obtaining the trained network model, inputs the collected seismic data, apparent resistivity data, initially inverted P-wave velocity, and initially inverted resistivity data into the trained network model, outputs the inversion result, and the resolution in the obtained inversion result is extremely high.

[0078] The exemplary application of the embodiment of the present invention in an actual application scenario will be described below.

[0079] Figure 4 is the network block diagram of the seismic electromagnetic logging joint inversion method of the physics-informed neural network provided by the present invention. As Figure 4 shown, the physics-informed neural network includes a seismic forward / backward inversion part and an electromagnetic forward / backward inversion part. Among them, the seismic forward / backward inversion part, i.e., the seismic closed-loop network, includes two networks, namely the seismic forward physical model and the elastic parameter inversion model. The seismic forward network is based on the Aki-Richard equation and is used to establish the theoretical physical relationship between the inversion result and the pre-stack angle seismic data. The elastic parameter inversion model is a neural convolutional network and is used to learn the non-linear mapping relationship between the pre-stack angle seismic data and the elastic parameters. The electromagnetic forward / backward inversion part, i.e., the electromagnetic closed-loop network, also includes two networks, namely the resistivity inversion network and the electromagnetic forward network. The resistivity inversion network is used to learn the non-linear mapping relationship between the apparent resistivity and the resistivity, while the electromagnetic forward network is used to learn the non-linear mapping relationship between the resistivity and the apparent resistivity data.

[0080] To verify the effectiveness and superiority of the present invention, the seismic electromagnetic logging joint inversion method of the physics-informed neural network proposed by the present invention is applied to an actual scenario.

[0081] Figure 5 is one of the inversion effect display diagrams provided by the present invention. As Figure 5 shown, what is displayed is the 3D volume of the pre-stack small-angle seismic data. Figure 6 is the second inversion effect display diagram provided by the present invention. Figure 6 What is displayed is the 3D volume of the apparent resistivity. Figure 7 is the third inversion effect display diagram provided by the present invention. Figure 7 What is displayed is the 3D volume of the initial P-wave velocity inversion result. Figure 9 is the fifth inversion effect display diagram provided by the present invention. Figure 9 What is displayed is the 3D volume of the initial resistivity inversion result. The three well logging data are used as well logging labels respectively, and the corresponding seismic data, apparent resistivity data, initial P-wave velocity and initial resistivity data of the corresponding channels are used as labeled data, and the data at other positions are unlabeled data. Figure 8 is the fourth inversion effect display diagram provided by the present invention. Figure 8 The P-wave velocity inverted by this method is shown. Figure 10 is the sixth inversion effect display diagram provided by the present invention. Figure 10 The S-wave velocity inverted by this method is shown. Figure 11 is the seventh inversion effect display diagram provided by the present invention. Figure 11 The density inverted by this method is shown. Figure 12 is the eighth inversion effect display diagram provided by the present invention.Figure 12 The resistivity results inverted by the present method are shown. Figure 13 It is the ninth diagram showing the inversion effect provided by the present invention. In order to further verify the lithology of the inversion results, Figure 13 the lithology prediction results are shown. Due to the incorporation of logging information, the resolution of the inversion results is significantly improved, and the thinnest coal seam thickness that can be distinguished can reach 2 m. Due to the addition of cross-gradient constraints, the in-phase axis trends of the inversion results of the four parameters are relatively consistent. The coal seams are concentrated between 800 m and 900 m underground and are distributed in layers or lenticular shapes along the in-phase axis trend. The longitudinal wave velocity, transverse wave velocity and density values at the coal seams are relatively low, while the resistivity value is relatively high. The coal seams are surrounded by sandy mudstone, which is mainly composed of clay minerals and sandy particles and has a higher density than the coal seams.

[0082] The present invention proposes a seismic electromagnetic logging joint inversion method based on a physics-informed neural network, namely, a seismic electromagnetic logging joint inversion method of a physics-informed neural network. This method introduces geophysical information from three aspects: network structure design, prior low-frequency data constraint, and design of a multi-objective loss function. A first closed-loop network is constructed by using the pre-stack seismic forward / backward inversion rules. The forward model is based on the Aki-Richard equation, and a second closed-loop network is constructed by using the electromagnetic forward / backward inversion rules. The forward / backward inversion networks are modeled simultaneously to jointly form a reciprocal constraint. In addition, the initial inversion result is used as a prior low-frequency data constraint to make the inversion modeling process converge more easily. In the design of the multi-objective loss function, the logging constraint term can be used to constrain the consistency between the inversion result and the logging parameters, ensuring the distribution consistency between the inversion result and the logging values. The seismic forward physical model constraint term uses the Aki-Richards forward equation to construct a seismic forward network, and introduces AVO (amplitude versus offset) information into the network by optimizing the angular wavelet, improving the rationality and physical interpretability of the seismic inversion results. The closed-loop constraint term uses forward / backward inversion modeling simultaneously, introduces a large amount of unlabeled data to participate in the training, and improves the generalization performance of the model. The cross-gradient constraint term constrains the similarity between the gradients of multiple inversion results, ensuring the consistency of the change trend in the spatial structure. The uncertainty constraint term quantifies the uncertainty of the network prediction results and reduces the uncertainty of the model. Thereby, the inversion accuracy and resolution are improved, and a refined evaluation result of underground resources is obtained.

[0083] Based on the foregoing embodiments, the embodiments of the present invention provide a seismic electromagnetic logging joint inversion device of a physics-informed neural network. Each module included in the device and each unit included in each module can be implemented by a processor; of course, it can also be implemented by specific logic circuits; during the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0084] The seismic and electromagnetic logging joint inversion device of the physics-informed neural network provided by the present invention will be described below. The seismic and electromagnetic logging joint inversion device of the physics-informed neural network described below can be correspondingly referred to the seismic and electromagnetic logging joint inversion method of the physics-informed neural network described above.

[0085] Figure 14 It is a schematic structural diagram of the seismic and electromagnetic logging joint inversion device of the physics-informed neural network provided by the present invention. As Figure 14 shown, the device 400 includes a first acquisition module 401 and a second acquisition module 402, wherein: The first acquisition module 401 is configured to acquire data to be solved, and the data to be solved includes prestack seismic data and apparent resistivity data; The second acquisition module 402 is configured to input the data to be solved into a trained physics-informed neural network to obtain target geological data, where the target geological data includes a target P-wave velocity, a target S-wave velocity, a target density, and a target resistivity. The trained physics-informed neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward network and an inverse network. The target geological data is obtained by constructing a reciprocal constraint through two sets of forward networks and inverse networks.

[0086] In some embodiments, the trained seismic closed-loop network includes an elastic inverse network and a seismic forward network. The elastic inverse network is configured to realize the mapping from prestack seismic data to well logging elastic parameters, and the seismic forward network is configured to realize the mapping from well logging elastic parameters to prestack seismic data. The trained electromagnetic closed-loop network includes a resistivity inverse network and an electromagnetic forward network. The resistivity inverse network is configured to realize the mapping from apparent resistivity data to well logging resistivity data, and the electromagnetic forward network is configured to realize the mapping from well logging resistivity data to apparent resistivity data.

[0087] In some embodiments, the data to be solved further includes an initially inverted P-wave velocity and an initially inverted resistivity. The initially inverted P-wave velocity is the propagation velocity of P-waves in the underground medium obtained by processing seismic data through an inversion algorithm, and the initially inverted resistivity is the distribution of resistivity in the underground medium obtained by processing electromagnetic data through an inversion algorithm.

[0088] In some embodiments, the second acquisition module 402 is specifically configured to: Input the initially inverted P-wave velocity into the elastic inversion network in the trained seismic closed-loop network, and input the initially inverted resistivity into the resistivity inversion network in the trained electromagnetic closed-loop network, as the prior low-frequency data constraint for the trained physics-informed neural network to output the target P-wave velocity and the target S-wave velocity.

[0089] In some embodiments, the apparatus further includes a network acquisition module, and the network acquisition module is configured to: Determine an initial seismic closed-loop network and an initial electromagnetic closed-loop network. The initial seismic closed-loop network is determined according to the forward and inverse modeling laws of pre-stack seismic data, and the electromagnetic closed-loop network is determined according to the electromagnetic forward and inverse modeling laws; Determine a loss function, which includes a logging constraint term, a physical model constraint term, a closed-loop constraint term, an uncertainty constraint term, and a cross-gradient constraint term. The logging constraint term is used to constrain the inversion result to be consistent with the logging parameters. The physical model constraint term is used to optimize the angle wavelet to introduce the information of amplitude varying with offset into the seismic closed-loop network. The closed-loop constraint term is used to simultaneously model with the forward network and the inverse network, introducing a large amount of unlabeled data to participate in the training. The uncertainty constraint term is used to quantify the uncertainty of the network prediction result. The cross-gradient constraint term is used to constrain the similarity between the gradients of multiple inversion results; Train the initial seismic closed-loop network and the initial electromagnetic closed-loop network according to the loss function to obtain the trained physics-informed neural network.

[0090] In some embodiments, the network acquisition module is further specifically configured to: Optimize the elastic inversion network and the seismic forward network in the initial seismic closed-loop network, and the resistivity inversion network and the electromagnetic forward network in the initial electromagnetic closed-loop network through the logging constraint term to obtain a first seismic closed-loop network and a first electromagnetic closed-loop network; Optimize the seismic forward network through the physical model constraint term to obtain a second seismic closed-loop network; Introduce unlabeled data, and optimize the elastic inversion network in the second seismic closed-loop network and the resistivity inversion network in the first electromagnetic closed-loop network by using the closed-loop constraint term to obtain a third seismic closed-loop network and a second electromagnetic closed-loop network; Optimize the elastic inversion network in the third seismic closed-loop network and the resistivity inversion network in the second electromagnetic closed-loop network by using the uncertainty constraint term, and quantify the uncertainty of the inversion result to obtain a fourth seismic closed-loop network and a third electromagnetic closed-loop network; Optimize the elastic inversion network in the fourth seismic closed-loop network and the resistivity inversion network in the third electromagnetic closed-loop network by using cross-gradient constraint terms to obtain the trained physics-informed neural network.

[0091] In the embodiments of the present invention, the inversion accuracy and resolution can be improved to obtain a refined evaluation result of underground resources.

[0092] Figure 15 It is a schematic physical structure diagram of the electronic device provided by the present invention. As Figure 15 shown, the electronic device 500 may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540. Among them, the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call the logical instructions in the memory 530 to execute the seismic and electromagnetic logging joint inversion method of the physics-informed neural network. The method includes: obtaining data to be solved, where the data to be solved includes prestack seismic data and apparent resistivity data; inputting the data to be solved into the trained physics-informed neural network to obtain target geological data, where the target geological data includes target P-wave velocity, target S-wave velocity, target density, and target resistivity. The trained physics-informed neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward modeling network and an inversion network. The target geological data is obtained through mutual inverse constraints formed by two sets of forward modeling networks and inversion networks.

[0093] In addition, when the logical instructions in the above-mentioned memory 530 are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical disks, and other various media that can store program codes.

[0094] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the seismic electromagnetic logging joint inversion method of the physical information neural network provided by the above-mentioned various methods. The method includes: obtaining data to be solved, where the data to be solved includes prestack seismic data and apparent resistivity data; inputting the data to be solved into a trained physical information neural network to obtain target geological data, where the target geological data includes a target P-wave velocity, a target S-wave velocity, a target density, and a target resistivity. The trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward network and an inverse network. The target geological data is obtained by forming a reciprocal constraint through two sets of forward networks and inverse networks.

[0095] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are fully or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wirelessly (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0096] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the seismic electromagnetic logging joint inversion method of the physical information neural network provided by the above-mentioned various methods. The method includes: obtaining data to be solved, where the data to be solved includes prestack seismic data and apparent resistivity data; inputting the data to be solved into a trained physical information neural network to obtain target geological data, where the target geological data includes target P-wave velocity, target S-wave velocity, target density, and target resistivity. The trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network. Both the trained seismic closed-loop network and the trained electromagnetic closed-loop network include a forward modeling network and an inversion network. The target geological data is obtained by forming a reciprocal constraint through two sets of forward modeling networks and inversion networks.

[0097] The above computer-readable storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM) or flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device.

[0098] The computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal can take various forms, including - but not limited to - electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device.

[0099] The program code contained on a computer-readable medium can be transmitted using any appropriate medium, including—but not limited to—wireless, wire, optical fiber cable, radio frequency (RF), and the like, or any suitable combination of the foregoing.

[0100] The computer program code for performing the operations of this specification can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., by connecting through the Internet using an Internet service provider).

[0101] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative effort.

[0102] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A physical information neural network seismic electromagnetic logging joint inversion method, characterized in that: include: Acquiring data to be solved, wherein the data to be solved includes pre-stack seismic data and apparent resistivity data; The data to be solved are input into a trained physical information neural network to obtain target geological data, wherein the target geological data include target P-wave velocity, target S-wave velocity, target density and target resistivity, wherein the trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network, wherein the trained seismic closed-loop network and the trained electromagnetic closed-loop network both include a forward modeling network and an inversion network, and wherein the target geological data are obtained by constraining mutually inversely two sets of forward modeling networks and inversion networks.

2. The physical information neural network seismic electromagnetic logging joint inversion method according to claim 1 is characterized in that: The trained seismic closed-loop network includes an elastic inversion network and a seismic forward modeling network. The elastic inversion network is used to realize the mapping from pre-stack seismic data to well logging elastic parameters, and the seismic forward modeling network is used to realize the mapping from well logging elastic parameters to pre-stack seismic data. The trained electromagnetic closed-loop network includes a resistivity inversion network and an electromagnetic forward modeling network. The resistivity inversion network is used to realize the mapping from apparent resistivity data to well logging resistivity data, and the electromagnetic forward modeling network is used to realize the mapping from well logging resistivity data to apparent resistivity data.

3. The physical information neural network seismic electromagnetic logging joint inversion method according to claim 2 is characterized in that: The data to be solved also include the initial inversion of P-wave velocity and the initial inversion of resistivity. The initial inversion of P-wave velocity is the propagation velocity of P-wave in the underground medium obtained by processing seismic data with an inversion algorithm. The initial inversion of resistivity is the distribution of resistivity in the underground medium obtained by processing electromagnetic data with an inversion algorithm.

4. The physical information neural network seismic electromagnetic logging joint inversion method according to claim 3 is characterized in that: The step of inputting the data to be solved into a trained physical information neural network to obtain target geological data includes: The initial inverted P-wave velocity is input into the elastic inversion network in the trained seismic closed-loop network, and the initial inverted resistivity is input into the resistivity inversion network in the trained electromagnetic closed-loop network, so as to serve as a priori low-frequency data constraints for the trained physical information neural network to output the target P-wave velocity and the target S-wave velocity.

5. The physical information neural network seismic electromagnetic logging joint inversion method according to claim 1 is characterized in that: Before obtaining the data to be solved, the method further includes: Determine an initial seismic closed-loop network and an initial electromagnetic closed-loop network, wherein the initial seismic closed-loop network is determined according to the forward and inversion law of pre-stack seismic data, and the electromagnetic closed-loop network is determined according to the electromagnetic forward and inversion law; Determine a loss function, wherein the loss function includes a logging constraint term, a physical model constraint term, a closed-loop constraint term, an uncertainty constraint term, and a cross-gradient constraint term. The logging constraint term is used to constrain the inversion result to be consistent with the logging parameter. The physical model constraint term is used to optimize the angle wavelet to introduce information about the amplitude change with the offset distance into the seismic closed-loop network. The closed-loop constraint term is used to simultaneously model the forward network and the inversion network, and introduce a large amount of unlabeled data to participate in the training. The uncertainty constraint term is used to quantify the uncertainty of the network prediction result. The cross-gradient constraint term is used to constrain the similarity between the gradients of multiple inversion results. The initial seismic closed-loop network and the initial electromagnetic closed-loop network are trained according to the loss function to obtain the trained physical information neural network.

6. The physical information neural network seismic electromagnetic logging joint inversion method according to claim 5 is characterized in that: The step of training the initial seismic closed-loop network and the initial electromagnetic closed-loop network according to the loss function to obtain the trained physical information neural network comprises: Optimizing the elastic inversion network and the seismic forward modeling network in the initial seismic closed-loop network, and the resistivity inversion network and the electromagnetic forward modeling network in the initial electromagnetic closed-loop network through the logging constraint term, to obtain a first seismic closed-loop network and a first electromagnetic closed-loop network; Optimizing the seismic forward modeling network through the physical model constraint terms to obtain a second seismic closed-loop network; Introducing unlabeled data, and using the closed-loop constraint term to optimize the elastic inversion network in the second seismic closed-loop network and the resistivity inversion network in the first electromagnetic closed-loop network, to obtain a third seismic closed-loop network and a second electromagnetic closed-loop network; Optimizing the elastic inversion network in the third seismic closed-loop network and the resistivity inversion network in the second electromagnetic closed-loop network by using uncertainty constraints, and quantifying the uncertainty of the inversion results to obtain a fourth seismic closed-loop network and a third electromagnetic closed-loop network; The elastic inversion network in the fourth seismic closed-loop network and the resistivity inversion network in the third electromagnetic closed-loop network are optimized by using cross-gradient constraints to obtain the trained physical information neural network.

7. A physical information neural network seismic electromagnetic logging joint inversion device, characterized in that: include: A first acquisition module is used to acquire data to be solved, wherein the data to be solved includes pre-stack seismic data and apparent resistivity data; The second acquisition module is used to input the data to be solved into a trained physical information neural network to obtain target geological data, wherein the target geological data includes target longitudinal wave velocity, target shear wave velocity, target density and target resistivity, and the trained physical information neural network includes a trained seismic closed-loop network and a trained electromagnetic closed-loop network, and the trained seismic closed-loop network and the trained electromagnetic closed-loop network both include a forward modeling network and an inversion network, and the target geological data is obtained by forming a mutually inverse constraint between two sets of forward modeling networks and inversion networks.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the seismic and electromagnetic logging joint inversion method of the physical information neural network as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the seismic and electromagnetic logging joint inversion method of the physical information neural network as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the seismic and electromagnetic logging joint inversion method of the physical information neural network as described in any one of claims 1 to 6 is implemented.

Images