Implicit full waveform inversion method, device and electronic equipment

Abstract

The present application provides an implicit full waveform inversion method, device and electronic device, the method comprising: constructing a deep learning model; inputting a training coordinate system into the deep learning model for training until the objective function of the deep learning model meets a preset condition, and obtaining a trained deep learning model; inputting the coordinate system to be inverted into the trained deep learning model to obtain the inversion result of the physical property parameters of the underground medium. The method has strong robustness and generalization, allowing accurate inversion of the physical property parameters of the underground medium from a randomly initialized model, getting rid of the strong dependence of the full waveform inversion on the initial model, and being applicable to complex structural media; directly by randomly closing some neurons of the implicit representation deep neural network, uncertainty analysis of the inversion results is achieved without additional computational overhead.

Description

Technical Field

[0001] The present application relates to the field of geophysical exploration technology, and in particular to an implicit full waveform inversion method, device and electronic equipment. Background Art

[0002] Full waveform inversion is a method for imaging underground media that can maximize the use of full wavefield information. Contrary to the strong nonlinearity and pathological nature of full waveform inversion, the traditional solution process is usually based on the convex optimization assumption and the use of the adjoint state method. Therefore, a good initial model is a prerequisite for avoiding extreme value traps and obtaining accurate inversion results. In other words, full waveform inversion has a strong dependence on the initial model. In existing full waveform inversion, other methods such as tomographic inversion based on travel time, envelope inversion, velocity analysis, etc. are usually used to construct an underground medium model that is closer to the true solution as the initial model of full waveform inversion, thereby avoiding convergence to local extreme values. Although the above methods can solve the dependence of full waveform inversion on the initial model to a certain extent, when faced with complex underground structural media and low-frequency loss, the inversion accuracy of such methods is difficult to guarantee, and they cannot provide the required initial model for full waveform inversion.

[0003] The Chinese patent (CN 115184986 A) discloses a global envelope cross-correlation full waveform inversion method that is independent of the source. This method combines the advantages of envelope inversion, global cross-correlation inversion method and convolution type independent source inversion method. The objective function constructed by using envelope operator can restore the effective low-frequency information of the model when the original seismic data lacks low-frequency components. The global cross-correlation method can overcome the influence of inaccurate amplitude of seismic records on the inversion results. The application of convolution type independent source algorithm can effectively alleviate the influence of inaccurate source wavelet estimation on full waveform inversion. However, it still has a certain dependence on the initial model.

[0004] In addition, due to the ill-posedness and multiple solutions of geophysical inversion, how to evaluate the uncertainty of the final inversion result is an inevitable challenge for full waveform inversion in practical applications. The existing full waveform inversion uncertainty analysis method requires the calculation of the posterior covariance matrix related to the Hessian to achieve this. This method requires a huge amount of computation and memory consumption, and is not feasible in practical applications. Summary of the invention

[0005] The embodiments of the present application provide an implicit full waveform inversion method, device and electronic device to solve the technical problems in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method comprising:

[0007] Constructing a deep learning model, the deep learning model comprising: an implicit representation deep neural network based on a coordinate system and a recurrent neural network based on theoretical guidance;

[0008] Inputting a training coordinate system into the deep learning model for training until an objective function of the deep learning model meets a preset condition, thereby obtaining a trained deep learning model;

[0009] The coordinate system to be inverted is input into the trained deep learning model, and some neurons of the implicit representation deep neural network of the trained deep learning model are randomly turned off to obtain the inversion results of the physical property parameters of the underground medium.

[0010] Furthermore, the step of inputting the training coordinate system into the deep learning model for training until the objective function of the deep learning model satisfies a preset condition includes:

[0011] a. Inputting the training coordinate system into the implicit representation deep neural network to obtain simulation data;

[0012] b. Obtain observation data;

[0013] c. Calculating the data error between the observed data and the simulated data according to the objective function;

[0014] d. Determine whether the objective function meets the preset conditions. If not, execute step e. If so, complete the training of the deep learning model to obtain the trained deep learning model.

[0015] e. Back-propagate the data error to the implicit representation deep neural network, update the weight value of the implicit representation deep neural network, and continue to execute steps a to d.

[0016] Furthermore, the training coordinate system is input into the implicit representation deep neural network to obtain simulation data, including:

[0017] Inputting the training coordinate system into the implicit representation deep neural network, randomly shutting down some neurons of the implicit representation deep neural network, and obtaining an initial model of underground medium physical property parameters for training;

[0018] Performing a denormalization process on the initial model of underground medium physical property parameters for training to obtain the underground medium physical property parameter model for training;

[0019] The training underground medium physical property parameter model is input into the theory-guided recurrent neural network to obtain the forward simulation results of the training underground medium physical property parameter model, and the forward simulation results of the training underground medium physical property parameter model are used as simulation data.

[0020] Furthermore, the obtaining of observation data includes:

[0021] The real underground medium physical property parameter model is input into the theory-guided recurrent neural network to obtain the forward simulation results of the real underground medium physical property parameter model, and the forward simulation results of the real underground medium physical property parameter model are used as observation data.

[0022] Furthermore, the mathematical expression of the denormalization process is:

[0023] ,

[0024] in, is the weight value of the implicit representation deep neural network, To train the underground medium physical parameter model, is the initial model of underground medium physical property parameters for training, μ is the mean value, is the standard deviation.

[0025] Furthermore, the objective function is:

[0026] ,

[0027] Among them, d is the observed data, is a recurrent neural network based on theoretical guidance, s is the input source, is the underground medium physical property parameter model for training, and t is the time.

[0028] Furthermore, the weight value of the implicit representation deep neural network is updated according to the following formula:

[0029] ,

[0030] in, is the weight value of the implicit deep neural network, n is the number of iterations, α is the learning rate, is the gradient, The expression is:

[0031] ,

[0032] in, is the objective function right The partial derivative of for Implicitly represent the weight values ​​of deep neural networks The partial derivative of .

[0033] Furthermore, after inputting the coordinate system to be inverted into the trained deep learning model, randomly shutting down some neurons of the implicit representation deep neural network of the trained deep learning model, and obtaining the inversion result of the physical property parameters of the underground medium, it also includes:

[0034] The part of neurons of the implicit deep neural network representing the trained deep learning model is randomly closed repeatedly for multiple times to obtain multiple inversion results of underground medium physical property parameters, and perform uncertainty analysis on the inversion results of the underground medium physical property parameters.

[0035] In a second aspect, an embodiment of the present application provides an implicit full waveform inversion device, the device comprising:

[0036] A model building module, used to build a deep learning model based on a coordinate system, wherein the deep learning model includes: an implicit representation deep neural network and a recurrent neural network based on theoretical guidance;

[0037] A model training module, used to input a training coordinate system into the deep learning model for training until an objective function of the deep learning model meets a preset condition, thereby obtaining a trained deep learning model;

[0038] The inversion module is used to input the coordinate system to be inverted into the trained deep learning model, randomly close some neurons of the implicit representation deep neural network of the trained deep learning model, and obtain the inversion results of the physical property parameters of the underground medium.

[0039] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0040] processor;

[0041] Memory;

[0042] And a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the electronic device executes any one of the methods described in the first aspect.

[0043] The present application uses an implicit full waveform inversion method, and further proposes a device and an electronic device for implementing the method. The method and the device and the electronic device based on the method construct a deep learning model including an implicit representation deep neural network based on a coordinate system and a recurrent neural network based on theory guidance, wherein the implicit representation deep neural network based on the coordinate system can learn high-frequency features and utilize the spectral bias phenomenon of the deep learning model; the constructed deep learning model has strong robustness and generalization, allowing accurate inversion of underground medium physical parameters from a randomly initialized model, getting rid of the strong dependence of full waveform inversion on the initial model, and is suitable for complex structural media; the uncertainty analysis of the inversion results is achieved by directly randomly closing some neurons of the implicit representation deep neural network without the need for additional computational overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] Figure 1 It is a flowchart of an implicit full waveform inversion method provided according to an embodiment of the present application;

[0046] Figure 2 is a real underground medium physical property parameter model in the embodiment of the present application;

[0047] Figure 3 It is a comparison diagram of the traditional full waveform inversion result and the implicit full waveform inversion result under the random initial model condition in the embodiment of the present application;

[0048] Figure 4 is a schematic diagram of uncertainty analysis results of implicit full waveform inversion results in an embodiment of the present application;

[0049] Figure 5 It is a structural block diagram of an implicit full waveform inversion device provided in an embodiment of the present application;

[0050] Figure 6 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0052] It should be clear that the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0053] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0054] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0055] Embodiment 1

[0056] The present application embodiment provides an implicit full waveform inversion method. For details, please refer to Figure 1 The flowchart of an implicit full waveform inversion method provided in accordance with an embodiment of the present application is shown. The implicit full waveform inversion method provided in an embodiment of the present application may include the following steps during implementation.

[0057] S1. Construct a deep learning model, wherein the deep learning model includes: an implicit representation deep neural network based on a coordinate system and a recurrent neural network based on theoretical guidance.

[0058] In this embodiment, the implicit characterization deep neural network is constructed based on the coordinate system, that is, the input parameters of the implicit characterization deep neural network are the coordinate system parameters; the coordinate system refers to the spatial coordinate system of the initial model of the physical property parameters of the underground medium. By using a periodic activation function or using the Fourier characteristic matrix of the input coordinate system, the implicit characterization deep neural network can learn high-frequency features. The implicit characterization deep neural network can therefore be inverted from any initial model, solving the initial model dependency problem of full waveform inversion. The framework of the implicit characterization deep neural network is any one or more combinations of convolutional neural networks, recurrent neural networks, multi-layer perceptrons, graph neural networks, etc.

[0059] In this embodiment, a recurrent neural network based on theory guidance, specifically a recurrent neural network based on geophysical propagation equations, is constructed. The recurrent neural network constructed based on geophysical propagation equations will be used as a geophysical forward modeling framework.

[0060] S2. Inputting the training coordinate system into the deep learning model for training until the objective function of the deep learning model meets a preset condition, thereby obtaining a trained deep learning model;

[0061] In one embodiment, step S2 specifically includes:

[0062] a. Input the training coordinate system into the implicit representation deep neural network to obtain simulation data.

[0063] Among them, step a specifically includes inputting the training coordinate system into the implicit representation deep neural network, randomly shutting down some neurons of the implicit representation deep neural network, and obtaining the initial model of underground medium physical property parameters for training. Taking two-dimensional implicit full waveform inversion as an example, the coordinates x and z of the coordinate system are input to generate a normalized initial model of underground medium physical property parameters for training based on implicit neuron representation. .

[0064] The initial model of underground medium physical property parameters for training Perform inverse normalization processing to obtain the underground medium physical property parameter model for training. The mathematical expression of inverse normalization processing is:

[0065] ,

[0066] in, is the weight value of the implicit representation deep neural network, To train the underground medium physical parameter model, is the initial model of underground medium physical property parameters for training, μ is the mean value, is the standard deviation. The above mean and standard deviation can be calculated based on the logging data to obtain the mean and standard deviation of the logging data; they can also be the global mean and global standard deviation. The global mean and global standard deviation are specific values ​​within a large range. For example, the propagation speed of seismic waves in underground media is usually within 1000-6000. The global mean can be set to 4000 and the global standard deviation to 3000. In the actual inversion process, the values ​​of the global mean and global standard deviation can be set in combination with specific circumstances, and this application does not make specific limitations.

[0067] The theory-guided recurrent neural network constructed in step S1 is used as a geophysical forward simulation framework, the training underground medium physical property parameter model is input into the theory-guided recurrent neural network, the forward simulation results of the training underground medium physical property parameter model are obtained, and the forward simulation results of the training underground medium physical property parameter model are used as simulation data.

[0068] b. Obtain observation data.

[0069] Among them, step b specifically includes inputting the real underground medium physical property parameter model into the theory-guided recurrent neural network, obtaining the forward simulation results of the real underground medium physical property parameter model, and using the forward simulation results of the real underground medium physical property parameter model as observation data. Figure 2 is the real underground medium physical parameter model in the embodiment of the present application, and the model is input into the said theory-guided recurrent neural network for forward simulation to obtain the observed data d. Its theoretical expression is:

[0070] ,

[0071] Among them, d is the observed data, Based on the theory-guided recurrent neural network, is the real underground medium physical parameter model, s is the input source, and t is the time.

[0072] c. Calculate the data error between the observed data and the simulated data based on the objective function.

[0073] Wherein, the objective function is:

[0074] ,

[0075] Among them, d is the observed data, is a recurrent neural network based on theoretical guidance, s is the input source, t is the time, It is a model of underground medium physical property parameters for training.

[0076] d. Determine whether the objective function meets the preset conditions. If not, execute step e. If so, complete the training of the deep learning model and obtain the trained deep learning model.

[0077] Among them, the preset conditions include that the data error is less than a preset error value, the error reduction rate is less than a preset error reduction rate, or the number of iterations exceeds a preset value.

[0078] e. Back-propagate the data error to the implicit representation deep neural network, update the weight value of the implicit representation deep neural network, and continue to execute steps a to d.

[0079] The data error is back-propagated to the implicit representation deep neural network, and the weight value of the implicit representation deep neural network is updated according to the following formula:

[0080] ,

[0081] in, is the weight value of the implicit deep neural network, n is the number of iterations, α is the learning rate, is the gradient, The expression is:

[0082] ,

[0083] in, is the objective function right The partial derivative of for Implicitly represent the weight values ​​of deep neural networks The partial derivative of .

[0084] S3. Input the coordinate system to be inverted into the trained deep learning model, randomly close some neurons of the implicit representation deep neural network of the trained deep learning model, and obtain the inversion results of the physical property parameters of the underground medium.

[0085] Specifically, the coordinate system to be inverted is input into the implicit representation deep neural network in the trained deep learning model, and some neurons of the implicit representation deep neural network are randomly turned off to obtain the inversion results of the physical property parameters of the underground medium.

[0086] Figure 3 This is a comparison chart of the traditional full waveform inversion results and the implicit full waveform inversion results under the random initial model conditions in the embodiment of this application. Figure 3 a) is the random initial model; b) is the traditional full waveform inversion result based on the random initial model; c) is the implicit full waveform inversion result based on the random initial model. Figure 3 The comparison results show that based on the random initial model, the traditional full waveform inversion is basically unable to invert the physical parameters of the underground medium; while the implicit full waveform inversion can obtain accurate inversion results based on the random initial model. It can be seen that the traditional full waveform inversion is extremely sensitive and dependent on the initial model, while the implicit full waveform inversion successfully reduces the dependence on the initial model and can invert the accurate physical parameters of the underground medium based on the random initial model.

[0087] In one embodiment, after inputting the coordinate system to be inverted into the trained deep learning model, randomly shutting down some neurons of the implicit representation deep neural network of the trained deep learning model, and obtaining the inversion result of the physical property parameters of the underground medium, it also includes:

[0088] S4. Repeatedly and randomly shutting down some neurons of the implicit deep neural network of the trained deep learning model, obtaining multiple inversion results of underground medium physical property parameters, and performing uncertainty analysis on the inversion results of the underground medium physical property parameters.

[0089] Specifically, by repeatedly randomly closing some neurons of the implicit characterization deep neural network, the implicit characterization deep neural network is run multiple times to obtain multiple inversion results of underground medium physical property parameters, and Bayesian approximation is achieved. By calculating the mean and standard deviation of the above multiple inversion results of underground medium physical property parameters, uncertainty analysis of the inversion results of the underground medium physical property parameters is performed.

[0090] Figure 4 Schematic diagram of uncertainty analysis results of implicit full waveform inversion results in an embodiment of the present application. Figure 4 a) is the result of an inversion of underground medium physical property parameters; b) is the mean of 1000 inversion results of underground medium physical property parameters obtained after randomly shutting down 40% of the neurons of the implicit representation deep neural network; c) is the standard deviation of 1000 inversion results of underground medium physical property parameters obtained after randomly shutting down 40% of the neurons of the implicit representation deep neural network. The results show that by randomly shutting down some neurons of the implicit representation deep neural network, the uncertainty analysis of the inversion results can be effectively realized.

[0091] An implicit full waveform inversion method provided in an embodiment of the present application is constructed by constructing a deep learning model including an implicit representation deep neural network based on a coordinate system and a recurrent neural network based on theory guidance, wherein the implicit representation deep neural network based on the coordinate system can learn high-frequency features and utilize the spectral bias phenomenon of the deep learning model; the constructed deep learning model has strong robustness and generalization, allowing accurate inversion of underground medium physical properties from a randomly initialized model, getting rid of the strong dependence of full waveform inversion on the initial model, and being suitable for complex structural media; uncertainty analysis of the inversion results is achieved by directly randomly closing some neurons of the implicit representation deep neural network without the need for additional computational overhead.

[0092] Embodiment 2

[0093] Corresponding to the above-mentioned embodiments, the present application also provides an implicit full waveform inversion device.

[0094] See also Figure 5 , is a structural block diagram of an implicit full waveform inversion device provided in an embodiment of the present application. Figure 5 As shown, it mainly includes the following modules.

[0095] A model building module 501 builds a deep learning model based on a coordinate system, wherein the deep learning model includes: an implicit representation deep neural network and a recurrent neural network based on theoretical guidance;

[0096] A model training module 502 inputs a training coordinate system into the deep learning model for training until an objective function of the deep learning model meets a preset condition, thereby obtaining a trained deep learning model;

[0097] The inversion module 503 inputs the coordinate system to be inverted into the trained deep learning model, randomly turns off some neurons of the implicit representation deep neural network of the trained deep learning model, and obtains the inversion results of the physical property parameters of the underground medium.

[0098] It should be pointed out that the specific contents involved in the embodiments of the present application can be found in the description of the above method embodiments, and for the sake of brevity, they will not be repeated here.

[0099] Embodiment 3

[0100] Corresponding to the above embodiment, an embodiment of the present application further provides an electronic device.

[0101] See also Figure 6 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device 600 may include: a processor 601, a memory 602 and a communication unit 603. These components communicate via one or more buses. Those skilled in the art will appreciate that the electronic device structure shown in the figure does not limit the embodiments of the present application, and it may be a bus structure or a star structure, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0102] The communication unit 603 is used to establish a communication channel so that the electronic device can communicate with other devices.

[0103] The processor 601 is the control center of the electronic device. It uses various interfaces and lines to connect various parts of the entire electronic device. It runs or executes software programs and / or modules stored in the memory 602, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 2001 can only include a central processing unit (CPU). In the embodiment of the present application, the CPU can be a single computing core or multiple computing cores.

[0104] The memory 602 is used to store the execution instructions of the processor 601. The memory 602 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0105] When the execution instructions in the memory 602 are executed by the processor 601 , the electronic device 600 is enabled to execute part or all of the steps in the above method embodiment.

[0106] Corresponding to the above embodiment, the embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium may store a program, wherein when the program is running, the device where the computer-readable storage medium is located may be controlled to execute some or all of the steps in the above method embodiment. In a specific implementation, the computer-readable storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0107] Corresponding to the above-mentioned embodiment, the embodiment of the present application further provides a computer program product, which includes executable instructions. When the executable instructions are executed on a computer, the computer executes some or all of the steps in the above-mentioned method embodiment.

[0108] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0109] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0110] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0111] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0112] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An implicit full waveform inversion method, characterized in that: include: Constructing a deep learning model, the deep learning model comprising: an implicit representation deep neural network based on a coordinate system and a recurrent neural network based on theoretical guidance; Inputting a training coordinate system into the deep learning model for training until an objective function of the deep learning model meets a preset condition, thereby obtaining a trained deep learning model; Inputting the coordinate system to be inverted into the trained deep learning model, randomly shutting down some neurons of the implicit representation deep neural network of the trained deep learning model, and obtaining the inversion results of the physical property parameters of the underground medium; The step of inputting the training coordinate system into the deep learning model for training until the objective function of the deep learning model meets a preset condition includes: a. Inputting the training coordinate system into the implicit representation deep neural network to obtain simulation data; b. Obtain observation data; c. Calculating the data error between the observed data and the simulated data according to the objective function; d. Determine whether the objective function meets the preset conditions. If not, execute step e. If so, complete the training of the deep learning model to obtain the trained deep learning model. e. back-propagating the data error to the implicit representation deep neural network, updating the weight value of the implicit representation deep neural network, and continuing to perform steps a to d; Inputting the training coordinate system into the implicit representation deep neural network to obtain simulation data includes: Inputting the training coordinate system into the implicit representation deep neural network, randomly shutting down some neurons of the implicit representation deep neural network, and obtaining an initial model of underground medium physical property parameters for training; Performing a denormalization process on the initial model of underground medium physical property parameters for training to obtain the underground medium physical property parameter model for training; Inputting the training underground medium physical property parameter model into the theory-guided recurrent neural network, obtaining the forward simulation results of the training underground medium physical property parameter model, and using the forward simulation results of the training underground medium physical property parameter model as simulation data; The objective function is: ， Among them, d is the observed data, is a recurrent neural network based on theoretical guidance, s is the input source, t is the time, It is a model of underground medium physical property parameters for training.

2. The method according to claim 1, characterized in that The obtaining of observation data comprises: The real underground medium physical property parameter model is input into the theory-guided recurrent neural network to obtain the forward simulation results of the real underground medium physical property parameter model, and the forward simulation results of the real underground medium physical property parameter model are used as observation data.

3. The method according to claim 1, characterized in that The mathematical expression of the denormalization process is: ， in, is the weight value of the implicit representation deep neural network, To train the underground medium physical parameter model, is the initial model of underground medium physical property parameters for training, μ is the mean value, is the standard deviation.

4. The method according to claim 1, characterized in that: The weight value of the implicit representation deep neural network is updated according to the following formula: ， in, is the weight value of the implicit deep neural network, n is the number of iterations, α is the learning rate, is the gradient, The expression is: ， in, is the objective function right The partial derivative of for Implicitly represent the weight values ​​of deep neural networks The partial derivative of .

5. The method according to claim 1, characterized in that After inputting the coordinate system to be inverted into the trained deep learning model, randomly shutting down some neurons of the implicit representation deep neural network of the trained deep learning model, and obtaining the inversion result of the underground medium physical property parameters, the method further includes: The part of neurons of the implicit deep neural network representing the trained deep learning model is randomly closed repeatedly for multiple times to obtain multiple inversion results of underground medium physical property parameters, and perform uncertainty analysis on the inversion results of the underground medium physical property parameters.

6. An implicit full waveform inversion device, characterized in that: include: A model building module, used to build a deep learning model based on a coordinate system, wherein the deep learning model includes: an implicit representation deep neural network and a recurrent neural network based on theoretical guidance; A model training module, used to input a training coordinate system into the deep learning model for training until an objective function of the deep learning model meets a preset condition, thereby obtaining a trained deep learning model; An inversion module, used for inputting the coordinate system to be inverted into the trained deep learning model, randomly shutting down some neurons of the implicit representation deep neural network of the trained deep learning model, and obtaining the inversion results of the physical property parameters of the underground medium; The step of inputting the training coordinate system into the deep learning model for training until the objective function of the deep learning model meets a preset condition includes: a. Inputting the training coordinate system into the implicit representation deep neural network to obtain simulation data; b. Obtain observation data; c. Calculating the data error between the observed data and the simulated data according to the objective function; d. Determine whether the objective function meets the preset conditions. If not, execute step e. If so, complete the training of the deep learning model to obtain the trained deep learning model. e. back-propagating the data error to the implicit representation deep neural network, updating the weight value of the implicit representation deep neural network, and continuing to perform steps a to d; Inputting the training coordinate system into the implicit representation deep neural network to obtain simulation data includes: Inputting the training coordinate system into the implicit representation deep neural network, randomly shutting down some neurons of the implicit representation deep neural network, and obtaining an initial model of underground medium physical property parameters for training; Performing a denormalization process on the initial model of underground medium physical property parameters for training to obtain the underground medium physical property parameter model for training; Inputting the training underground medium physical property parameter model into the theory-guided recurrent neural network, obtaining the forward simulation results of the training underground medium physical property parameter model, and using the forward simulation results of the training underground medium physical property parameter model as simulation data; The objective function is: ， Among them, d is the observed data, is a recurrent neural network based on theoretical guidance, s is the input source, t is the time, It is a model of underground medium physical property parameters for training.

7. An electronic device, characterized in that: include: processor; Memory; And a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, which, when executed by the processor, enable the electronic device to perform the steps of the method described in any one of claims 1 to 5.

Images