Neural network training method and device, model construction method and device, equipment and medium

Abstract

The invention provides a neural network training method, a model construction method, devices, equipment and a medium, when a two-phase medium fluctuation feature learning network is trained, reservoir physical property parameters are used as input, a plurality of prediction wave equation coefficients of a sample two-phase pore medium elastic wave propagation model are predicted, and the two-phase medium fluctuation feature learning network is trained by combining a physical constraint mechanism based on plane wave analysis. According to error loss between predicted frequency dispersion attenuation characteristics and real frequency dispersion attenuation characteristics of a sample biphase pore medium elastic wave propagation model, parameters of a biphase medium fluctuation characteristic learning network are adjusted, and the network is made to have higher prediction stability and physical consistency through back propagation; the prediction wave equation coefficient output by the two-phase medium fluctuation feature learning network is ensured, and meanwhile, the data precision and physical consistency requirements are considered, so that the physical consistency, the expression accuracy and the generalization ability of the constructed target two-phase pore medium elastic wave propagation model are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geophysical exploration, in particular to a neural network training method, a model construction method, a device, equipment and a medium. BACKGROUND

[0002] Seismic wave propagation modeling is a key technical means in oil and gas exploration and underground medium imaging. Especially in complex media such as low porosity and permeability or multiphase flow, the propagation of waves is affected by the coupling of solid skeleton, pore fluid and viscoelastic effect, etc., and shows significant dispersion and attenuation characteristics. Accurate description of these complex wave behaviors is the core problem of constructing high-precision seismic simulation and reservoir inversion models. For example, the Biot model and other two-phase porous medium elastic wave propagation models are widely used to describe the propagation rules of P waves and S waves in fluid-saturated porous media.

[0003] However, the two-phase porous medium elastic wave propagation model has many parameters and complex physical meanings, and some parameters are difficult to obtain directly through measurement. How to effectively construct the two-phase porous medium elastic wave propagation model has become a problem to be solved. SUMMARY

[0004] Therefore, the present application provides a neural network training method, a model construction method, a device, equipment and a medium, which can effectively predict the wave equation coefficients of the two-phase porous medium elastic wave propagation model and improve the physical consistency, expression accuracy and generalization ability of the model on the basis of neural network deep learning and the physical constraint mechanism of plane wave analysis.

[0005] Specifically, the present application is implemented by the following technical solutions: According to a first aspect of the present application, a neural network training method is provided, which comprises: obtaining a sample data set, the sample data set comprising sample reservoir physical parameters associated with a sample two-phase porous medium elastic wave propagation model; inputting the sample data set into a two-phase medium wave characteristic learning network to output a plurality of predicted wave equation coefficients of the sample two-phase porous medium elastic wave propagation model; performing plane wave analysis based on the plurality of predicted wave equation coefficients to determine the predicted dispersion and attenuation characteristics of the sample two-phase porous medium elastic wave propagation model; adjusting the parameters of the two-phase medium wave characteristic learning network according to the error loss between the predicted dispersion and attenuation characteristics and the true dispersion and attenuation characteristics of the sample two-phase porous medium elastic wave propagation model until a cutoff condition is met.

[0006] In an alternative implementation, the plane wave analysis based on the plurality of predicted wave equation coefficients determines predicted dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model, including: obtaining a pore fluid density and a fluid viscosity coefficient corresponding to the sample dual-phase porous medium elastic wave propagation model; performing plane wave analysis based on the plurality of predicted wave equation coefficients, the pore fluid density, and the fluid viscosity coefficient to determine predicted dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model.

[0007] In an alternative implementation, the predicted dispersion attenuation characteristics include predicted P-wave velocity and predicted S-wave velocity, and the true dispersion attenuation characteristics include true P-wave velocity and true S-wave velocity. The error loss between the predicted dispersion attenuation characteristics and the true dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model is determined by: determining the error loss between the predicted dispersion attenuation characteristics and the true dispersion attenuation characteristics according to a deviation between the predicted P-wave velocity and the true P-wave velocity, and a deviation between the predicted S-wave velocity and the true S-wave velocity; Alternatively, the predicted dispersion attenuation characteristics include predicted P-wave velocity, predicted S-wave velocity, and predicted P-wave inverse quality factor, and the true dispersion attenuation characteristics include true P-wave velocity, true S-wave velocity, and true P-wave inverse quality factor. The error loss between the predicted dispersion attenuation characteristics and the true dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model is determined by: determining the error loss between the predicted dispersion attenuation characteristics and the true dispersion attenuation characteristics according to a deviation between the predicted P-wave velocity and the true P-wave velocity, a deviation between the predicted S-wave velocity and the true S-wave velocity, and a deviation between the predicted P-wave inverse quality factor and the true P-wave inverse quality factor.

[0008] In an alternative implementation, the sample dual-phase porous medium elastic wave propagation model includes a viscoelastic wave propagation model, the dual-phase medium wave feature learning network includes a viscoelastic wave feature learning network, and at least part of the plurality of predicted wave equation coefficients is a complex number. Alternatively, the sample dual-phase porous medium elastic wave propagation model includes an elastic wave propagation model, the dual-phase medium wave feature learning network includes an elastic wave feature learning network, and all of the plurality of predicted wave equation coefficients are real numbers.

[0009] In an alternative embodiment, the dual-phase medium wave characteristic learning network comprises an input module, a plurality of wave characteristic learning modules, and an output module, the input module is configured to receive an input sample dataset, each of the wave characteristic learning modules is configured to determine a predicted coefficient intermediate variable based on the sample dataset, and the output module is configured to generate the predicted wave equation coefficient based on the predicted coefficient intermediate variable, the predicted coefficient intermediate variable being a real part parameter and an imaginary part parameter that constitute the predicted wave equation coefficient.

[0010] In an alternative embodiment, the number of the wave characteristic learning modules is consistent with the number of the predicted coefficient intermediate variables.

[0011] In an alternative embodiment, the sample dataset comprises at least one of: angular frequency, solid matrix bulk modulus, Lame coefficient, porosity, permeability, and solid matrix density.

[0012] According to a second aspect of the present application, a model construction method is provided, the method comprising: obtaining a target dataset, the target dataset comprising target reservoir physical property parameters associated with a target dual-phase porous medium elastic wave propagation model; inputting the target dataset into a trained dual-phase medium wave characteristic learning network to output a plurality of wave equation coefficients of the target dual-phase porous medium elastic wave propagation model; the trained dual-phase medium wave characteristic learning network is obtained by training the neural network according to the above method; constructing the target dual-phase porous medium elastic wave propagation model based on the plurality of wave equation coefficients.

[0013] In an alternative embodiment, the method further comprises: performing plane wave analysis based on the plurality of wave equation coefficients to determine the dispersion attenuation characteristics of the target dual-phase porous medium elastic wave propagation model; determining the propagation information of the P-wave and the S-wave based on the dispersion attenuation characteristics.

[0014] According to a third aspect of the present application, a neural network training device is provided, the device comprising: a sample acquisition module configured to acquire a sample dataset, the sample dataset comprising sample reservoir physical property parameters associated with a sample dual-phase porous medium elastic wave propagation model; a network prediction module configured to input the sample dataset into a dual-phase medium wave characteristic learning network to output a plurality of predicted wave equation coefficients of the sample dual-phase porous medium elastic wave propagation model; a characteristic analysis module configured to perform plane wave analysis based on the plurality of predicted wave equation coefficients to determine predicted dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model; a parameter adjustment module configured to adjust parameters of the dual-phase medium wave characteristic learning network according to an error loss between the predicted dispersion attenuation characteristics and real dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model until a stop condition is met.

[0015] According to a fourth aspect of the present application, a model construction device is provided, the device comprising: a data acquisition module configured to acquire a target data set comprising target reservoir physical property parameters associated with a target dual-phase porous medium elastic wave propagation model; a network processing module configured to input the target data set into the trained dual-phase medium wave characteristic learning network to output a plurality of wave equation coefficients of the target dual-phase porous medium elastic wave propagation model; the trained dual-phase medium wave characteristic learning network is obtained by training the neural network training method described above; a model construction module configured to construct the target dual-phase porous medium elastic wave propagation model based on the plurality of wave equation coefficients.

[0016] According to a fifth aspect of the present application, a computer device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the steps of the neural network training method of the first aspect or the steps of the model construction method of the second aspect.

[0017] According to a sixth aspect of the present application, a computer readable storage medium is provided, having a computer program stored thereon, wherein the program is executed by a processor to implement the steps of the neural network training method of the first aspect or the steps of the model construction method of the second aspect.

[0018] The neural network training method, the model construction method, the device, the equipment and the medium provided by the embodiment of the application are used to input the reservoir physical property parameters when training the two-phase medium wave characteristic learning network, predict a plurality of predicted wave equation coefficients of a sample two-phase porous medium elastic wave propagation model, combine a physical constraint mechanism based on plane wave analysis, adjust the parameters of the two-phase medium wave characteristic learning network according to the error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model, make the network have stronger prediction stability and physical consistency through back propagation, ensure that the predicted wave equation coefficients output by the two-phase medium wave characteristic learning network take into account the data accuracy and the physical consistency requirement, and thus improve the physical consistency, the expression accuracy and the generalization ability of the constructed target two-phase porous medium elastic wave propagation model.

[0019] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the technical solutions of the present disclosure.

[0020] In order to make the above-mentioned purposes, features and advantages of the present disclosure more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are described in detail as follows. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart of a neural network training method according to an example embodiment of the present application; Figure 2 is a structural schematic diagram of a neural network according to an example embodiment of the present application; Figure 3 is a process schematic diagram of neural network training according to an example embodiment of the present application; Figure 4 is a flowchart of a model construction method according to an example embodiment of the present application; Figure 5 is a schematic diagram of a P-wave dispersion characteristic according to an example embodiment of the present application; Figure 6 is a schematic diagram of a S-wave dispersion characteristic according to an example embodiment of the present application; Figure 7 is a schematic diagram of a neural network training device according to an example embodiment of the present application; Figure 8 is a schematic diagram of a model construction device according to an example embodiment of the present application; Figure 9 is a structural schematic diagram of a computer device according to an example embodiment of the present application. DETAILED DESCRIPTION

[0022] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, like reference numerals refer to like elements throughout the description. The following exemplary embodiments described herein represent illustrations of apparatuses and methods that are in accordance with some aspects of the present application as detailed in the appended claims. These examples are, of course, merely illustrative and do not limit the application as claimed.

[0023] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0024] It is to be understood that the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. It is to be further understood that the terms "comprise", "comprising", "comprises", "including", "includes" or "contain" or "containing" when used in this specification, specify the presence of stated features, integers, steps, operations, elements, or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or groups thereof. It is to be understood that the terms "if' and "when" as used herein are interpreted as "if the probability is greater than zero" or "when the probability is greater than zero".

[0025] It is found through research that the dual-phase porous medium elastic wave propagation model has numerous parameters and complex physical meanings, and some parameters are difficult to be directly obtained through measurement or experiment, which limits the wide application of the dual-phase porous medium elastic wave propagation model. Therefore, how to effectively construct the dual-phase porous medium elastic wave propagation model becomes a problem to be solved.

[0026] Based on the above research, the present application provides a neural network training method, a model construction method, a device, equipment and a medium. In the training of the dual-phase medium wave characteristic learning network, the reservoir physical property parameters are taken as the input, the multiple prediction wave equation coefficients of the sample dual-phase porous medium elastic wave propagation model are predicted, the physical constraint mechanism based on the plane wave analysis is combined, the network has stronger prediction stability and physical consistency through the back propagation, the prediction wave equation coefficients output by the dual-phase medium wave characteristic learning network are ensured to simultaneously consider the data accuracy and the physical consistency requirement, so as to improve the physical consistency, expression accuracy and generalization ability of the constructed target dual-phase porous medium elastic wave propagation model.

[0027] To facilitate the understanding of the present embodiment, first, a neural network training method and a model construction method disclosed by the present application are introduced in detail. The execution subject of the neural network training method and the model construction method provided by the present application is generally an electronic device with certain computing power. The electronic device can be a server, which can be a physical server, a server cluster composed of multiple physical servers, or a distributed system. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, big data, and artificial intelligence platforms. In some possible implementation manners, the neural network training method and the model construction method can be realized by calling computer readable instructions stored in the memory through the processor.

[0028] Next, a neural network training method provided by the present application is described in conjunction with the accompanying drawings.

[0029] Referring to Figure 1 FIG. 1 is a flowchart of a neural network training method according to an example embodiment of the present application. Figure 1 As shown in FIG. 1, the neural network training method provided by the present disclosure includes steps S101-S104, wherein: S101: Obtain a sample data set, which includes sample reservoir physical property parameters associated with a sample dual-porosity medium elastic wave propagation model.

[0030] To better understand the present embodiment, first, a dual-porosity medium elastic wave propagation model is introduced. The dual-porosity medium elastic wave propagation model includes an elastic wave propagation model and a viscoelastic wave propagation model. The elastic wave propagation model has elastic wave propagation characteristics, and the viscoelastic wave propagation model has viscoelastic wave propagation characteristics.

[0031] The Biot model is an elastic wave propagation model and is widely used to describe the propagation rules of longitudinal and transverse waves in fluid-saturated porous media. However, the Biot model cannot effectively explain the high-frequency dispersion and strong attenuation of waves in the low-frequency band (especially the seismic wave frequency band).

[0032] Therefore, the Biot-Squirt (BISQ) model is proposed. The BISQ model is also an elastic wave propagation model that enhances the description of microscale energy loss by introducing a squirt mechanism, such as the frictional loss that occurs when fluid in a pore throat or other pore flows into the pore.

[0033] In most practical problems, the reservoir medium not only behaves as a multiphase medium, but also has obvious viscoelastic behavior. Accordingly, the viscoelastic BISQ model is proposed, which is a viscoelastic wave propagation model.

[0034] Specifically, the two-phase porous medium elastic wave propagation model can be represented by the following formula (1): (1) wherein, ( , , , ) represents a wave equation coefficient, and respectively represent a solid phase and a flow phase displacement tensor, represents a solid phase strain tensor, represents a flow phase strain tensor, represents a fluid viscosity coefficient, represents a porosity, and K represents a permeability; , , wherein represents a solid matrix density, represents a pore fluid density, represents a solid-fluid coupling density.

[0035] In practical applications, the wave equation coefficient is usually difficult to obtain directly, and the embodiments of the present disclosure mainly solve the problem that the wave equation coefficient of the two-phase porous medium elastic wave propagation model is difficult to obtain directly.

[0036] In the embodiments of the present disclosure, considering the deep learning technology, especially the physical information neural network (PINNs), the partial differential equation information or prior knowledge is supported to be introduced into the neural network training process, the physical rationality and generalization ability of the network output are improved, and a new idea is provided for modeling of complex physical systems. Therefore, the two-phase porous medium elastic wave propagation model is embedded into the deep neural network structure, and the existing modeling bottleneck is broken through.

[0037] In practical applications, the wave equation coefficient is often related to the viscoelastic modulus of the solid skeleton , , the Biot-William coefficient , the Biot fluid storage coefficient F, the jet flow coefficient S, the angular frequency ω, the solid matrix bulk modulus , , , the porosity , the permeability K, the fluid viscosity coefficient , and the solid matrix density and the like. The complex physical coefficients that are difficult to measure can be reflected through the wave equation coefficient .

[0038] In this step, in order to train the network, multiple data can be collected in different environments to form sample data sets corresponding to different environment complexities.

[0039] In the network training, the physical information is often guided. In the embodiments of the present disclosure, the dual-phase porous medium elastic wave propagation model is used as the guided physical information. Specifically, according to the dual-phase porous medium elastic wave propagation model as the physical information guide, the reservoir physical parameters that can improve the network learning accuracy are selected. The sample data set is constructed.

[0040] Optionally, the dual-phase porous medium elastic wave propagation model and the multiple reservoir physical parameters can be subjected to correlation analysis, and the reservoir physical parameters with a correlation degree higher than a threshold with the dual-phase porous medium elastic wave propagation model are selected from the multiple reservoir physical parameters as the sample reservoir physical parameters.

[0041] In some possible implementations, the sample data set includes at least one of the following: angular frequency, solid matrix bulk modulus, Lame coefficient, porosity, permeability, and solid matrix density.

[0042] The sample reservoir physical parameters can include parameters included in the dual-phase porous medium elastic wave propagation model, such as porosity , permeability K, and solid matrix density , and can also include parameters related to the dual-phase porous medium elastic wave propagation model, such as angular frequency ω, solid matrix bulk modulus , and Lame coefficient . .

[0043] In this way, the above parameters are all key reservoir physical parameters of the dual-phase porous medium elastic wave propagation model. By using these parameters to construct the sample data set, the physical characteristics of the dual-phase porous medium elastic wave propagation model can be more comprehensively reflected, so that the dual-phase medium wave characteristic learning network can learn more abundant and more accurate characteristics, which helps to further improve the prediction accuracy and physical consistency of the network for the dual-phase porous medium elastic wave propagation model, enhances the generalization ability of the model, and makes the application of the model under different geological conditions more reliable and adaptable.

[0044] S102: inputting the sample data set into the dual-phase medium wave characteristic learning network, and outputting multiple predicted wave equation coefficients of the sample dual-phase porous medium elastic wave propagation model.

[0045] In this step, the sample data set can be input into the dual-phase medium wave characteristic learning network, and the dual-phase medium wave characteristic learning network can output multiple predicted wave equation coefficients of the sample dual-phase porous medium elastic wave propagation model. , , , ).

[0046] In some possible embodiments, the sample dual-phase porous medium elastic wave propagation model comprises a visco-elastic wave propagation model, the dual-phase medium wave motion feature learning network comprises a visco-elastic wave motion feature learning network, and at least part of the plurality of predicted wave equation coefficients are complex numbers; or, the sample dual-phase porous medium elastic wave propagation model comprises an elastic wave propagation model, the dual-phase medium wave motion feature learning network comprises an elastic wave motion feature learning network, and the plurality of predicted wave equation coefficients are all real numbers.

[0047] Here, in the case where the sample dual-phase porous medium elastic wave propagation model comprises an elastic wave propagation model, correspondingly, the dual-phase medium wave motion feature learning network comprises an elastic wave motion feature learning network, and at this time, the plurality of predicted wave equation coefficients are all real numbers, that is, , , , all only include real parts and do not include imaginary parts.

[0048] In the case where the sample dual-phase porous medium elastic wave propagation model comprises a visco-elastic wave propagation model, correspondingly, the dual-phase medium wave motion feature learning network comprises a visco-elastic wave motion feature learning network. For a visco-elastic wave propagation model, this means that in the process of seismic wave propagation, the medium response contains phase lag and energy dissipation, which needs to be described by physical quantities such as complex modulus. At least part of the plurality of predicted wave equation coefficients of the visco-elastic wave motion equation are often frequency-dependent complex numbers, and the imaginary part directly reflects the absorption characteristics of the medium. The modeling of the visco-elastic wave propagation model inevitably involves complex number domain mathematical problems in theory.

[0049] Alternatively, the plurality of predicted wave equation coefficients can all be complex numbers, that is, , , , all include real part parameters and imaginary part parameters.

[0050] In actual applications, and the imaginary part parameters of and are small, if the imaginary part parameters of and are introduced into network calculation, the network may output the values of the imaginary part parameters of and to be large, resulting in introducing deviation. Alternatively, the plurality of predicted wave equation coefficients can be partly complex numbers and partly real numbers, so as to improve the accuracy of the dual-phase medium wave motion feature learning network.

[0051] Exemplarily, the sample can be , is a complex number, , is a real number.

[0052] In this way, the specific types of the sample dual-phase porous medium elastic wave propagation model and the dual-phase medium wave characteristic learning network and the corresponding predicted wave equation coefficient forms are further clarified, so that learning can be performed on the two different types of models, i.e., the viscoelastic wave propagation model and the elastic wave propagation model, respectively, to ensure better adaptability and accuracy when processing different types of dual-phase porous medium elastic wave propagation models, further improve the physical consistency and generalization ability of the constructed target dual-phase porous medium elastic wave propagation model, and thus better meet the modeling needs of different types of dual-phase porous medium elastic wave propagation models in actual applications.

[0053] In some possible implementations, the dual-phase medium wave characteristic learning network includes an input module, a plurality of wave characteristic learning modules, and an output module, the input module is configured to receive an input sample data set, each wave characteristic learning module is configured to determine a predicted coefficient intermediate variable according to the sample data set, and the output module is configured to generate the predicted wave equation coefficient based on the predicted coefficient intermediate variable, the predicted coefficient intermediate variable being a real part parameter and an imaginary part parameter that constitute the predicted wave equation coefficient.

[0054] Here, the structures of the wave characteristic learning modules are the same. Exemplarily, each wave characteristic learning module includes 50 neural units, and the optimizer adopts an adaptive moment estimation (Adam) optimization algorithm. When the wave characteristic learning module is used to predict a predicted coefficient intermediate variable corresponding to a wave equation coefficient in a complex number form, a linear rectification function (ReLU) is used as an activation function, and when the wave characteristic learning module is used to predict a predicted coefficient intermediate variable corresponding to a wave equation coefficient in a real number form, a sigmoid function is used as an activation function.

[0055] Exemplarily, when the sample dual-phase porous medium elastic wave propagation model includes an elastic wave propagation model and the dual-phase medium wave characteristic learning network includes an elastic wave characteristic learning network, the predicted wave equation coefficient can be represented by the following formula (2):

[0056]

[0057]

[0058] (2) wherein, , , , denote wave equation coefficients, , , , denote predicted coefficient intermediate variables output by respective wave feature learning modules, denote the input sample dataset, , , , denote parameters of respective wave feature learning modules.

[0059] In the case where the sample biphasic porous medium elastic wave propagation model comprises a viscoelastic wave propagation model, and the biphasic medium wave feature learning network comprises a viscoelastic wave feature learning network, if the plurality of predicted wave equation coefficient parts are complex numbers, and parts are real numbers, the predicted wave equation coefficients can be shown by the following formula (3):

[0060]

[0061]

[0062] (3) wherein, , , , denote wave equation coefficients, , , , , , denote predicted coefficient intermediate variables output by respective wave feature learning modules, denote the input sample dataset, , , , , , denote parameters of respective wave feature learning modules.

[0063] In a case where the sample dual-phase porous medium elastic wave propagation model comprises a viscoelastic wave propagation model, and the dual-phase medium wave characteristic learning network comprises a viscoelastic wave characteristic learning network, if the plurality of predicted wave equation coefficients are complex numbers, the predicted wave equation coefficients can be represented by the following formula (4):

[0064]

[0065]

[0066] (4) wherein, , , , represents a wave equation coefficient, , , , , , , , represents a predicted coefficient intermediate variable output by each wave characteristic learning module, represents the input sample data set, , , , , , , , represents a parameter of each wave characteristic learning module.

[0067] In this way, by refining the dual-phase medium wave characteristic learning network into an input module, a plurality of wave characteristic learning modules, and an output module, the structure and functional division of the network are further optimized. This modular design not only improves the learning efficiency and accuracy of the network, but also enhances the adaptability and generalization ability of the network to complex dual-phase porous medium elastic wave propagation models, further improving the performance and reliability of network training.

[0068] In some possible implementations, the number of wave characteristic learning modules is consistent with the number of predicted coefficient intermediate variables.

[0069] For example, in combination with formula (2), in a case where the sample dual-phase porous medium elastic wave propagation model comprises an elastic wave propagation model, and the dual-phase medium wave characteristic learning network comprises an elastic wave characteristic learning network, the plurality of predicted wave equation coefficients are real numbers, and at this time , , Each has 1 real part parameter, so there are 4 real part parameters in total, and correspondingly, 4 prediction coefficient intermediate variables need to be determined, so the number of wave characteristic learning modules is 4.

[0070] For example, as can be seen from formula (3), in the case that the sample dual-phase porous medium elastic wave propagation model includes a viscoelastic wave propagation model and the dual-phase medium wave characteristic learning network includes a viscoelastic wave characteristic learning network, the prediction wave equation coefficients are partly complex numbers and partly real numbers, and at this time Each has 1 real part parameter and 1 imaginary part parameter, Each has 1 real part parameter, so there are 4 real part parameters and 2 imaginary part parameters in total, and correspondingly, 6 prediction coefficient intermediate variables need to be determined, so the number of wave characteristic learning modules is 6.

[0071] For example, as can be seen from formula (4), in the case that the sample dual-phase porous medium elastic wave propagation model includes a viscoelastic wave propagation model and the dual-phase medium wave characteristic learning network includes a viscoelastic wave characteristic learning network, the prediction wave equation coefficients are all complex numbers, and at this time Each has 1 real part parameter and 1 imaginary part parameter, so there are 4 real part parameters and 4 imaginary part parameters in total, and correspondingly, 8 prediction coefficient intermediate variables need to be determined, so the number of wave characteristic learning modules is 8.

[0072] In this way, by making the number of wave characteristic learning modules consistent with the number of prediction coefficient intermediate variables, the network structure is precisely matched with the task requirements, each wave characteristic learning module is specially responsible for extracting and processing features related to one prediction coefficient intermediate variable, ensuring the pertinence and efficiency of feature extraction, and this one-to-one module and variable correspondence relationship enables the network to learn and optimize each prediction coefficient intermediate variable more meticulously, thereby further improving the accuracy and reliability of the prediction wave equation coefficients. At the same time, this structural design also enhances the scalability and flexibility of the network, facilitating the adjustment of the number of modules according to different task requirements, and further improving the performance and adaptability of the dual-phase medium wave characteristic learning network.

[0073] For example, refer to Figure 2 Figure 2 A structural diagram of a neural network according to an exemplary embodiment of the present application is shown. As shown in Figure 2 ​​​​​​​As shown in the middle, the present example illustrates the case of the viscoelastic wave propagation model in which part of the predicted wave equation coefficients shown in formula (3) are complex numbers, and the two-phase medium wave characteristic learning network includes an input module, 6 wave characteristic learning modules, and an output module. The input module can receive a sample data set to provide a basis for subsequent processing; the wave characteristic learning module can extract and process key information in the sample data set to determine the predicted coefficient intermediate variables, which can help to capture complex wave characteristics more carefully; and the output module can generate predicted wave equation coefficients based on these predicted coefficient intermediate variables, so that the prediction results are more accurate and complete.

[0074] In this way, for the viscoelastic wave propagation model, to more accurately depict its dispersion and attenuation characteristics, the present embodiment introduces a partial complex neural network structure, models the predicted coefficient intermediate variables involved in the wave characteristic learning module and the output module as complex numbers, specifically uses a double-channel real / imaginary parallel representation, and the network training process supports the back propagation of complex parameters, supports the expression of predicted coefficient intermediate variables in the complex domain during the network training process, so that the two-phase medium wave characteristic learning network has the expression ability to handle viscoelastic characteristics such as frequency-dependent energy dissipation and phase lag, thereby enabling the two-phase medium wave characteristic learning network to directly learn and output wave equation coefficients in complex form, thereby improving the expression ability and physical consistency of viscoelastic medium characteristics. The trained two-phase medium wave characteristic learning network outputs can effectively reflect the physical mechanisms such as energy dissipation and phase lag in the medium under the complex constraint condition, and are closer to the real data.

[0075] For the elastic wave propagation model shown in formula (2) or the viscoelastic wave propagation model in which all predicted wave equation coefficients shown in formula (4) are complex numbers, the corresponding two-phase medium wave characteristic learning network structure is similar to the two-phase medium wave characteristic learning network structure corresponding to the viscoelastic wave propagation model in which part of the predicted wave equation coefficients shown in formula (3) are complex numbers, and only the number of wave characteristic learning modules needs to be adjusted accordingly.

[0076] S103: Perform plane wave analysis based on the plurality of predicted wave equation coefficients to determine the predicted dispersion and attenuation characteristics of the sample two-phase porous medium elastic wave propagation model.

[0077] Here, in traditional network training, if the two-phase medium wave characteristic learning network outputs predicted wave equation coefficients, the real value of the wave equation coefficient is often used as a network training label for supervised learning. Considering that the real value of the wave equation coefficient is difficult to obtain directly in practice, the present embodiment adopts a training method driven by physical consistency, and uses the easily measured dispersion and attenuation characteristics as the network training label.

[0078] Optionally, the predicted dispersion attenuation characteristics include a predicted P-wave velocity and a predicted S-wave velocity. Further optionally, the predicted dispersion attenuation characteristics include a predicted P-wave velocity, a predicted S-wave velocity, and a predicted P-wave inverse quality factor.

[0079] Here, the S-wave inverse quality factor has a small value, and thus is not used as a network training label in the embodiments of the present disclosure.

[0080] In some possible implementations, the determining, based on the plurality of predicted wave equation coefficients, the predicted dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model includes: obtaining a pore fluid density and a fluid viscosity coefficient corresponding to the sample dual-phase porous medium elastic wave propagation model; performing plane wave analysis based on the plurality of predicted wave equation coefficients, the pore fluid density, and the fluid viscosity coefficient to determine the predicted dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model.

[0081] In the above steps, the pore fluid density and the fluid viscosity coefficient corresponding to the sample dual-phase porous medium elastic wave propagation model can be obtained, and a reference angular frequency can be obtained. Based on the plurality of predicted wave equation coefficients, the pore fluid density, the fluid viscosity coefficient, and the reference angular frequency, plane wave analysis is performed on an expression (such as formula (1)) of the dual-phase porous medium elastic wave propagation model by using Helmholtz decomposition and Fourier transform to obtain a correlation between wave equation coefficients and dispersion attenuation characteristics. Based on the correlation between the wave equation coefficients and the dispersion attenuation characteristics, the plurality of predicted wave equation coefficients, the sample data set, the pore fluid density, the fluid viscosity coefficient, and the reference angular frequency, the predicted dispersion attenuation characteristics of the sample dual-phase porous medium elastic wave propagation model are determined.

[0082] Specifically, the correlation between the wave equation coefficients and the dispersion attenuation characteristics can be represented by the following formula (5):

[0083]

[0084]

[0085] wherein, represents a P-wave velocity, represents an S-wave velocity, represents a P-wave inverse quality factor, represents a P-wave calculation intermediate variable, represents an S-wave calculation intermediate variable.

[0086] Here, the longitudinal wave velocity Specifically, the phase velocity of the longitudinal wave, the transverse wave velocity Specifically, the phase velocity of the transverse wave.

[0087] wherein the longitudinal wave calculation intermediate variable and the transverse wave calculation intermediate variable can be determined by the following formula (6): (6) wherein, represents the longitudinal wave calculation intermediate variable, represents the transverse wave calculation intermediate variable, ( , , , ) represents the wave equation coefficient, , , , , , , , represents the solid matrix density, represents the pore fluid density, represents the solid-fluid coupling density, represents the reference angular frequency, represents the imaginary unit, represents the angular frequency, represents the correlation relationship calculation intermediate variable, represents the porosity, represents the fluid viscosity coefficient, and K represents the permeability.

[0088] Here, it can be seen from formula (6) that the first expression therein is a quadratic equation about , and two roots can be solved through the expression, and two longitudinal wave velocities can be obtained, one is the fast P-wave velocity, and the other is the slow P-wave velocity, and the fast P-wave velocity is greater than the slow P-wave velocity. Since the value of the slow P-wave velocity is small, the present embodiment only uses the fast P-wave velocity as the network training label, that is, the longitudinal wave velocity in the present embodiment represents the fast P-wave velocity.

[0089] Thus, by introducing the pore fluid density and fluid viscosity coefficient in the plane wave analysis process, the determination process of the predicted dispersion attenuation characteristics is further improved, thereby more comprehensively reflecting the physical phenomena in the two-phase porous medium, making the predicted dispersion attenuation characteristics more accurately close to the real situation, and helping to more accurately evaluate the error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics, thereby more effectively adjusting the parameters of the two-phase medium wave characteristic learning network, and further improving the accuracy and reliability of the predicted wave equation coefficients output by the network, and helping to enhance the physical consistency and generalization ability of the constructed target two-phase porous medium elastic wave propagation model.

[0090] S104: According to the error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model, the parameters of the two-phase medium wave characteristic learning network are adjusted until the stopping condition is met.

[0091] In this step, after determining the predicted dispersion attenuation characteristics, a loss function can be constructed in combination with the physical constraints of the sample two-phase porous medium elastic wave propagation model to determine the error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model, so as to adjust the parameters of the two-phase medium wave characteristic learning network according to the error loss until the stopping condition is met.

[0092] In this way, the relationship between the predicted wave equation coefficients and the predicted dispersion attenuation characteristics can be established in the backpropagation process of network training, so as to explicitly constrain the physical model information in the network structure during the training process. By adding physical constraints in network training, the physical consistency of network output can be effectively improved, and the shortcomings of traditional end-to-end neural networks such as poor interpretability and "black box" can be overcome.

[0093] Optionally, the training stopping condition can be that the error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics is less than a preset error, or the number of training iterations of the two-phase medium wave characteristic learning network reaches a preset number, etc.

[0094] In some possible implementations, the predicted dispersion attenuation characteristics include a predicted P-wave velocity and a predicted S-wave velocity, and the real dispersion attenuation characteristics include a real P-wave velocity and a real S-wave velocity. The error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model is determined by the following steps: According to the deviation between the predicted P-wave velocity and the real P-wave velocity and the deviation between the predicted S-wave velocity and the real S-wave velocity, the error loss between the predicted dispersion attenuation characteristics and the real dispersion attenuation characteristics is determined. Alternatively, the predicted dispersion attenuation features include a predicted P-wave velocity, a predicted S-wave velocity, and a predicted P-wave inverse quality factor, and the real dispersion attenuation features include a real P-wave velocity, a real S-wave velocity, and a real P-wave inverse quality factor; The error loss between the predicted dispersion attenuation features and the real dispersion attenuation features of the sample dual-phase porous medium elastic wave propagation model is determined by the following steps: The error loss between the predicted dispersion attenuation features and the real dispersion attenuation features is determined according to the deviation between the predicted P-wave velocity and the real P-wave velocity, the deviation between the predicted S-wave velocity and the real S-wave velocity, and the deviation between the predicted P-wave inverse quality factor and the real P-wave inverse quality factor.

[0095] For example, the error loss can be determined by the following formula (7): (7) Wherein, Loss represents the error loss, represents the predicted P-wave velocity, represents the real P-wave velocity, represents the predicted S-wave velocity, represents the real S-wave velocity, represents the predicted P-wave inverse quality factor, represents the real P-wave inverse quality factor, and N is the number of samples, and are the weights corresponding to the deviation between the predicted P-wave velocity and the real P-wave velocity, the deviation between the predicted S-wave velocity and the real S-wave velocity, and the deviation between the predicted P-wave inverse quality factor and the real P-wave inverse quality factor, respectively.

[0096] Here, if some data is missing, the corresponding weight can be set to 0.

[0097] In this way, the dispersion attenuation features are selected as the network training labels, and the correlation between the dispersion attenuation features and the wave equation coefficients is derived through the plane wave analysis method, so that the loss function contains the velocity deviation, the inverse quality factor deviation, and the physical constraint information of the sample dual-phase porous medium elastic wave propagation model, thereby guiding the network output to meet the data accuracy and physical consistency requirements, and the correlation is not embedded in the forward propagation path of the network training, but is used in the backward propagation stage as a physical constraint term to constrain the error loss in the determination, so that the network training process has stronger prediction stability and physical consistency.

[0098] In some possible implementation manners, in the training of the dual-phase medium wave motion feature learning network, a supervised learning manner is adopted, a sample data set and a real dispersion attenuation feature can be constructed by using synthetic data or laboratory measurement data, a batch gradient descent strategy can be adopted in the training process, and learning rate decay, regularization, early stopping and other strategies can be combined to improve the training efficiency and accuracy, so that the network can stably converge under different parameter conditions, the predicted dispersion attenuation feature determined according to the predicted wave equation coefficients has a high fitting degree with the actual data, and has good generalization ability.

[0099] In order to more clearly show the training process of the neural network, please refer to Figure 3 , Figure 3 A process diagram of neural network training is shown for an exemplary embodiment of the present application. As shown in Figure 3 , a sample data set is obtained, the sample data set is input into the dual-phase medium wave motion feature learning network, a plurality of predicted wave equation coefficients of the sample dual-phase porous medium elastic wave propagation model are output, plane wave analysis is performed based on the plurality of predicted wave equation coefficients, a predicted dispersion attenuation feature of the sample dual-phase porous medium elastic wave propagation model is determined, parameters of the dual-phase medium wave motion feature learning network are adjusted according to an error loss between the predicted dispersion attenuation feature and a real dispersion attenuation feature of the sample dual-phase porous medium elastic wave propagation model, until a cutoff condition is met. In this way, on the basis of following the wave motion law of the dual-phase porous medium elastic wave propagation model, the dependence of the dual-phase porous medium elastic wave propagation model on a large number of physical parameters is avoided, so that the trained dual-phase medium wave motion feature learning network has good physical consistency, interpretability, expression accuracy and generalization ability at the same time, and is suitable for a variety of complex reservoir environments. For specific descriptions, please refer to the foregoing embodiments, which will not be described here.

[0100] The neural network training method provided in the embodiments of the present application, in the training of the dual-phase medium wave motion feature learning network, takes the reservoir physical parameters as input, predicts a plurality of predicted wave equation coefficients of a sample dual-phase porous medium elastic wave propagation model, combines a physical constraint mechanism based on plane wave analysis, adjusts parameters of the dual-phase medium wave motion feature learning network according to an error loss between a predicted dispersion attenuation feature and a real dispersion attenuation feature of the sample dual-phase porous medium elastic wave propagation model, and makes the network have stronger prediction stability and physical consistency through back propagation, so as to ensure that the predicted wave equation coefficients output by the dual-phase medium wave motion feature learning network take into account both data accuracy and physical consistency requirements, thereby improving the physical consistency, expression accuracy and generalization ability of the constructed target dual-phase porous medium elastic wave propagation model.

[0101] It can be understood that after the training of the dual-phase medium wave characteristic learning network is completed, the trained dual-phase medium wave characteristic learning network can be used to generate wave equation coefficients for constructing a dual-phase porous medium elastic wave propagation model. Therefore, the present disclosure also provides a model construction method, please refer to Figure 4 , Figure 4 The flowchart of a model construction method provided by the present disclosure is shown in FIG. 4. As shown in FIG. 4, the model construction method provided by the present disclosure includes steps S401-S403, wherein: Figure 4 S401: Obtain a target data set, wherein the target data set includes target reservoir physical parameters associated with a target dual-phase porous medium elastic wave propagation model.

[0102] In this step, when it is necessary to construct a dual-phase porous medium elastic wave propagation model, the target data set and the dual-phase medium wave characteristic learning network trained according to the neural network training method described above can be obtained.

[0103] Here, the way of determining the target data set is similar to that of determining the sample data set, and the specific steps are described in the foregoing embodiments, which will not be repeated here.

[0104] S402: Input the target data set into the trained dual-phase medium wave characteristic learning network, and output a plurality of wave equation coefficients of the target dual-phase porous medium elastic wave propagation model; the trained dual-phase medium wave characteristic learning network is trained by the neural network training method described above.

[0105] In this step, the target data set can be input into the trained dual-phase medium wave characteristic learning network, and the dual-phase medium wave characteristic learning network can output a plurality of predicted wave equation coefficients of the target dual-phase porous medium elastic wave propagation model.

[0106] Here, the application process of the dual-phase medium wave characteristic learning network is similar to the training process, and the specific steps are described in the foregoing embodiments, which will not be repeated here.

[0107] S403: Construct the target dual-phase porous medium elastic wave propagation model based on the plurality of wave equation coefficients.

[0108] Here, after obtaining the plurality of wave equation coefficients, the target dual-phase porous medium elastic wave propagation model can be constructed, and an expression of the target dual-phase porous medium elastic wave propagation model as shown in formula (1) is obtained.

[0109] In some possible implementations, the method further includes: Perform plane wave analysis based on the plurality of wave equation coefficients to determine the dispersion attenuation characteristics of the target dual-phase porous medium elastic wave propagation model.​ determine propagation information of the longitudinal wave and the transverse wave based on the dispersion attenuation characteristics.

[0110] Here, the dispersion attenuation characteristics of the target dual-porosity medium elastic wave propagation model can also be directly predicted by means of the plane wave analysis mechanism adopted during network training, so as to determine the propagation information of the longitudinal wave and the transverse wave based on the dispersion attenuation characteristics.

[0111] The propagation information of the longitudinal wave and the transverse wave can include longitudinal wave velocity, transverse wave velocity, etc.

[0112] In this way, not only the wave equation coefficients can be obtained, but also the wave propagation characteristics closely related to actual applications can be obtained. This conversion from theoretical coefficients to actual wave propagation information enhances the practicality and interpretability of the model, making it more intuitive to serve geological exploration, reservoir evaluation and other practical application scenarios, and further improving the application value and guiding significance of the target dual-porosity medium elastic wave propagation model.

[0113] The dual-porosity medium elastic wave propagation model constructed by the embodiments of the present disclosure can accurately depict the dispersion and attenuation behavior of the medium under multiple frequency bands. Compared with other traditional models, the physical consistency, expression accuracy and generalization ability are significantly improved, and significant advantages are shown in wave velocity prediction accuracy, wave propagation characteristic description ability and adaptability to complex reservoirs. For example, refer to Figure 5 and Figure 6 , Figure 5 a schematic diagram of longitudinal wave dispersion characteristics shown by an example embodiment of the present application, Figure 6 a schematic diagram of transverse wave dispersion characteristics shown by an example embodiment of the present application. As shown in Figure 5 and Figure 6 , where the dashed line represents the longitudinal wave dispersion characteristics and the transverse wave dispersion characteristics predicted by the Biot model, the solid line represents the longitudinal wave dispersion characteristics and the transverse wave dispersion characteristics obtained by using the embodiments of the present application, and the circle point represents the real data collected. It can be seen that, compared with the Biot model, the longitudinal wave dispersion characteristics and the transverse wave dispersion characteristics obtained by using the embodiments of the present application are more consistent with the real data, and can effectively reflect the physical mechanisms such as energy dissipation and phase lag in the medium. Especially in low porosity and permeability, fracture type or strong dissipation shale and other dissipation complex media, the embodiments of the present application can still accurately model without the need to measure all model parameters, significantly reduce the modeling threshold, improve the engineering applicability and promotion value of the method, expand the application boundary of complex neural networks in the field of geophysical modeling, and have good engineering applicability and realizability.

[0114] The model construction method provided in the embodiments of the present application can efficiently output a plurality of wave equation coefficients of a target two-phase porous medium elastic wave propagation model, so as to construct the target two-phase porous medium elastic wave propagation model, and since the two-phase medium wave characteristic learning network has strong prediction stability and physical consistency, the output wave equation coefficients meet the data accuracy requirement and satisfy the physical consistency constraint, thereby improving the physical consistency, expression accuracy and generalization ability of the constructed target two-phase porous medium elastic wave propagation model.

[0115] Those skilled in the art can understand that, in the above method of the specific implementation, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process, and the specific execution order of each step should be determined according to its function and possible internal logic.

[0116] Corresponding to the above-mentioned embodiments of the neural network training method, the present application also provides an embodiment of a neural network training device.

[0117] Please refer to Figure 7 A schematic diagram of a neural network training device according to an example embodiment of the present application is shown. As shown in Figure 7 The neural network training device 700 provided in the embodiments of the present application includes: A sample acquisition module 701 is configured to acquire a sample data set, wherein the sample data set includes sample reservoir physical parameters associated with a sample two-phase porous medium elastic wave propagation model.

[0118] A network prediction module 702 is configured to input the sample data set into a two-phase medium wave characteristic learning network, and output a plurality of predicted wave equation coefficients of the sample two-phase porous medium elastic wave propagation model.

[0119] A feature analysis module 703 is configured to perform plane wave analysis based on the plurality of predicted wave equation coefficients, and determine a predicted dispersion attenuation characteristic of the sample two-phase porous medium elastic wave propagation model.

[0120] A parameter adjustment module 704 is configured to adjust parameters of the two-phase medium wave characteristic learning network according to an error loss between the predicted dispersion attenuation characteristic and a true dispersion attenuation characteristic of the sample two-phase porous medium elastic wave propagation model until a cutoff condition is met.

[0121] In some possible implementations, the feature analysis module 703 is specifically configured to: acquire a pore fluid density and a fluid viscosity coefficient corresponding to the sample two-phase porous medium elastic wave propagation model; performing plane wave analysis based on the plurality of predicted wave equation coefficients, the pore fluid density, and the fluid viscosity coefficient to determine a predicted dispersion attenuation characteristic of the sample dual-phase porous medium elastic wave propagation model.

[0122] In some possible implementation, the predicted dispersion attenuation characteristic includes a predicted P-wave velocity and a predicted S-wave velocity, and the true dispersion attenuation characteristic includes a true P-wave velocity and a true S-wave velocity. The parameter adjustment module 704 determines an error loss between the predicted dispersion attenuation characteristic and a true dispersion attenuation characteristic of the sample dual-phase porous medium elastic wave propagation model by the following steps: determining the error loss between the predicted dispersion attenuation characteristic and the true dispersion attenuation characteristic according to a deviation between the predicted P-wave velocity and the true P-wave velocity and a deviation between the predicted S-wave velocity and the true S-wave velocity; Alternatively, the predicted dispersion attenuation characteristic includes a predicted P-wave velocity, a predicted S-wave velocity, and a predicted P-wave inverse quality factor, and the true dispersion attenuation characteristic includes a true P-wave velocity, a true S-wave velocity, and a true P-wave inverse quality factor. The parameter adjustment module 704 determines an error loss between the predicted dispersion attenuation characteristic and a true dispersion attenuation characteristic of the sample dual-phase porous medium elastic wave propagation model by the following steps: determining the error loss between the predicted dispersion attenuation characteristic and the true dispersion attenuation characteristic according to a deviation between the predicted P-wave velocity and the true P-wave velocity, a deviation between the predicted S-wave velocity and the true S-wave velocity, and a deviation between the predicted P-wave inverse quality factor and the true P-wave inverse quality factor.

[0123] In some possible implementation, the sample dual-phase porous medium elastic wave propagation model includes a visco-elastic wave propagation model, the dual-phase medium wave characteristic learning network includes a visco-elastic wave characteristic learning network, and at least part of the plurality of predicted wave equation coefficients is a complex number. Alternatively, the sample dual-phase porous medium elastic wave propagation model includes an elastic wave propagation model, the dual-phase medium wave characteristic learning network includes an elastic wave characteristic learning network, and the plurality of predicted wave equation coefficients are all real numbers.

[0124] In some possible implementation manners, the two-phase medium wave characteristic learning network comprises an input module, a plurality of wave characteristic learning modules, and an output module, the input module is configured to receive an input sample data set, each wave characteristic learning module is configured to determine a predicted coefficient intermediate variable according to the sample data set, and the output module is configured to generate the predicted wave equation coefficient based on the predicted coefficient intermediate variable, the predicted coefficient intermediate variable being a real part parameter and an imaginary part parameter that constitute the predicted wave equation coefficient.

[0125] In some possible implementation manners, the number of the wave characteristic learning modules is consistent with the number of the predicted coefficient intermediate variables.

[0126] In some possible implementation manners, the sample data set comprises at least one of the following: an angular frequency, a solid matrix volume modulus, a Lame coefficient, a porosity, a permeability, and a solid matrix density.

[0127] Corresponding to the foregoing model construction method embodiments, the present application further provides model construction device embodiments.

[0128] Please refer to Figure 8 , a schematic diagram of a model construction device according to an exemplary embodiment of the present application. As shown in Figure 8 , the model construction device 800 provided by the embodiment of the present application comprises: a data acquisition module 801 configured to acquire a target data set, the target data set comprising a target reservoir physical property parameter associated with a target two-phase porous medium elastic wave propagation model.

[0129] a network processing module 802 configured to input the target data set into a trained two-phase medium wave characteristic learning network and output a plurality of wave equation coefficients of the target two-phase porous medium elastic wave propagation model; the trained two-phase medium wave characteristic learning network is obtained by training the neural network according to the neural network training method described above.

[0130] a model construction module 803 configured to construct the target two-phase porous medium elastic wave propagation model based on the plurality of wave equation coefficients.

[0131] In some possible implementation manners, the model construction device 800 further comprises an information determination module 804, the information determination module 804 is configured to: determine a dispersion attenuation characteristic of the target two-phase porous medium elastic wave propagation model based on the plane wave analysis of the plurality of wave equation coefficients; determine propagation information of a P wave and an S wave based on the dispersion attenuation characteristic.

[0132] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0133] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0134] Based on the same technical concept, the embodiment of the present application further provides a computer device 900, referring to Figure 9 FIG. 1 is a schematic diagram of a computer device according to an exemplary embodiment of the present application, comprising: Processor 910, memory 920, and bus 930. Memory 920 is used to store execution instructions and includes internal memory 921 and external memory 922. Memory 921, also referred to as internal memory, is used to temporarily store computational data from processor 910 and data exchanged with external memory 922, such as a hard disk. Processor 910 exchanges data with external memory 922 via internal memory 921.

[0135] In the embodiment of the present application, the memory 920 is specifically used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 910. That is, when the electronic device 900 is running, the processor 910 communicates with the memory 920 via the bus 930, or the processor 910 communicates with the memory 920 through other means, so that the processor 910 executes the application code stored in the memory 920, and then performs the steps of the neural network training method or model building method described in any of the aforementioned embodiments.

[0136] The memory 920 can be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only memory (PROM), an Erasable Programmable Read-Only memory (EPROM), an Electric Erasable Programmable Read-Only memory (EEPROM), etc.

[0137] The processor 910 can be an integrated circuit chip that has the processing capability of signals. The processor described above can be a general processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor.

[0138] It can be understood that the structure illustrated in the embodiments of the present application does not constitute a specific limitation on the electronic device 900. In some other embodiments of the present application, the electronic device 900 can include more or fewer components than those illustrated, or combine certain components, or split certain components, or different component arrangements. The illustrated components can be implemented in hardware, software or a combination of software and hardware.

[0139] The embodiments of the present disclosure also provide a computer readable storage medium, and the computer readable storage medium stores a computer program. When the computer program is run by a processor, the steps of the neural network training method or the model construction method described in the above method embodiments are executed. The storage medium can be a volatile or non-volatile computer readable storage medium.

[0140] The embodiments of the present disclosure further provide a computer program product, and the computer program product has a computer program stored thereon. The computer program, when executed by a processor, performs the steps of the neural network training method or the model construction method provided by any of the embodiments of the present disclosure. For details, refer to the method embodiments described above, which will not be repeated here.

[0141] The computer program product can be implemented by hardware, software or a combination thereof. In an optional embodiment, the computer program product is embodied as a computer storage medium, which can be a volatile or non-volatile computer readable storage medium. In another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (SDK) and the like.

[0142] In addition, the embodiments of the subject matter and the functional operations described in this specification can be implemented in: digital electronic circuitry, a tangibly embodied computer software or firmware, computer hardware including the structural means disclosed in this specification and structural equivalents thereof, or a combination of one or more of them. The embodiments of the subject matter described in this specification can be implemented as one or more computer programs, that is, one or more modules of computer program instructions encoded on a tangible non-transitory program carrier to be executed by, or to control the operation of, data processing apparatus. Alternatively or additionally, the program instructions can be encoded on artificially generated propagated signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0143] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform the functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuitry, for example, an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit), and the apparatus can also be implemented as special purpose logic circuitry.

[0144] Computers suitable for the execution of a computer program include, by way of example, general and / or special purpose microprocessors, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory and / or a random access memory. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device (e.g., a universal serial bus (USB) flash drive), to name just a few.

[0145] Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0146] While this specification contains many specifics, these should not be construed as limitations on the scope of any invention or on the required scope of patent protection. Certain features outside the scope of the claimed invention are described in order to provide a clearer understanding of the features of the particular inventions. Some features described in multiple embodiments can be combined in a single embodiment. Conversely, various features described in a single embodiment can be divided among several embodiments. Moreover, no component or structure of the described embodiment is intended to be essential to the practice of the claimed invention unless the component or structure is directly numbered and described as an essential element of the invention in the claims. It is intended that additional modifications and variations to these specific implementation be considered as coming within the scope of the claimed invention. It is intended that only such limitations as two-fully described and clearly induced the patent and / or industrial property office be placed upon the invention so that the patent rights and interests in the invention are guarded.

[0147] Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring such an order, nor that all illustrated operations be performed, to implement and / or benefit from the present invention. In certain circumstances, multitasking and parallel processing can be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated in a single software product or packaged into multiple software products.

[0148] Thus, particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, actions recited in the claims can be performed in a different order and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0149] The above-described embodiments are merely possible implementations of the present application, and do not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the present application.

Claims

1. A neural network training method, characterized in that: The method comprises: Acquiring a sample data set, the sample data set including sample reservoir physical property parameters associated with a sample two-phase porous medium elastic wave propagation model; Inputting the sample data set into a two-phase medium wave characteristic learning network, and outputting a plurality of predicted wave equation coefficients of the elastic wave propagation model of the sample two-phase porous medium; Performing plane wave analysis based on the plurality of predicted wave equation coefficients to determine predicted dispersion attenuation characteristics of the elastic wave propagation model of the sample two-phase porous medium; According to the error loss between the predicted dispersion attenuation characteristics and the actual dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model, the parameters of the two-phase medium wave characteristic learning network are adjusted until a cutoff condition is met.

2. The method according to claim 1, characterized in that The performing of plane wave analysis based on the plurality of predicted wave equation coefficients to determine the predicted dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model includes: Obtaining the pore fluid density and fluid viscosity coefficient corresponding to the elastic wave propagation model of the sample two-phase porous medium; A plane wave analysis is performed based on the multiple predicted wave equation coefficients, the pore fluid density, and the fluid viscosity coefficient to determine the predicted dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model.

3. The method according to claim 1, characterized in that The predicted dispersion attenuation characteristics include predicted longitudinal wave velocity and predicted shear wave velocity, and the real dispersion attenuation characteristics include real longitudinal wave velocity and real shear wave velocity; The error loss between the predicted dispersion attenuation characteristics and the actual dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model is determined by the following steps: determining an error loss between the predicted dispersion attenuation characteristic and the true dispersion attenuation characteristic based on a deviation between the predicted longitudinal wave velocity and the true longitudinal wave velocity and a deviation between the predicted shear wave velocity and the true shear wave velocity; Alternatively, the predicted dispersion attenuation characteristics include predicted longitudinal wave velocity, predicted shear wave velocity, and predicted longitudinal wave inverse quality factor, and the true dispersion attenuation characteristics include true longitudinal wave velocity, true shear wave velocity, and true longitudinal wave inverse quality factor; The error loss between the predicted dispersion attenuation characteristics and the actual dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model is determined by the following steps: The error loss between the predicted dispersion attenuation characteristic and the true dispersion attenuation characteristic is determined based on the deviation between the predicted longitudinal wave velocity and the true longitudinal wave velocity, the deviation between the predicted shear wave velocity and the true shear wave velocity, and the deviation between the predicted longitudinal wave inverse quality factor and the true longitudinal wave inverse quality factor.

4. The method according to claim 1, wherein The sample two-phase porous medium elastic wave propagation model includes a viscoelastic wave propagation model, the two-phase medium wave characteristic learning network includes a viscoelastic wave characteristic learning network, and at least some of the multiple prediction wave equation coefficients are complex numbers; Alternatively, the sample two-phase porous medium elastic wave propagation model includes an elastic wave propagation model, the two-phase medium wave characteristic learning network includes an elastic wave characteristic learning network, and the multiple prediction wave equation coefficients are all real numbers.

5. The method according to any one of claims 1 to 4, characterized in that: The two-phase medium fluctuation characteristic learning network includes an input module, multiple fluctuation characteristic learning modules and an output module. The input module is used to receive the input sample data set, each of the fluctuation characteristic learning modules is used to determine the prediction coefficient intermediate variable according to the sample data set, and the output module is used to generate the prediction wave equation coefficient based on the prediction coefficient intermediate variable. The prediction coefficient intermediate variable is the real part parameter and imaginary part parameter that constitute the prediction wave equation coefficient.

6. The method according to claim 5, characterized in that The number of the fluctuation feature learning modules is consistent with the number of the prediction coefficient intermediate variables.

7. The method according to claim 1, characterized in that The sample data set includes at least one of the following: angular frequency, solid matrix bulk modulus, Lame coefficient, porosity, permeability, and solid matrix density.

8. A model building method, characterized in that: The method comprises: Acquiring a target data set, the target data set including target reservoir physical property parameters associated with a target two-phase porous medium elastic wave propagation model; Inputting the target data set into a trained two-phase medium wave characteristic learning network, and outputting a plurality of wave equation coefficients of the target two-phase porous medium elastic wave propagation model; the trained two-phase medium wave characteristic learning network is obtained by training the neural network training method according to any one of claims 1 to 7; Based on the multiple wave equation coefficients, an elastic wave propagation model of the target two-phase porous medium is constructed.

9. The method according to claim 8, characterized in that The method further comprises: Performing plane wave analysis based on the multiple wave equation coefficients to determine the dispersion attenuation characteristics of the target two-phase porous medium elastic wave propagation model; Based on the dispersion attenuation characteristics, the propagation information of the longitudinal wave and the shear wave is determined.

10. A neural network training device, characterized in that: The device comprises: A sample acquisition module is used to acquire a sample data set, wherein the sample data set includes sample reservoir physical property parameters associated with a sample two-phase porous medium elastic wave propagation model; a network prediction module, configured to input the sample data set into a two-phase medium wave characteristic learning network and output a plurality of predicted wave equation coefficients of the elastic wave propagation model of the sample two-phase porous medium; a characteristic analysis module, configured to perform plane wave analysis based on the plurality of predicted wave equation coefficients to determine predicted dispersion attenuation characteristics of the elastic wave propagation model of the sample two-phase porous medium; A parameter adjustment module is used to adjust the parameters of the two-phase medium wave characteristic learning network according to the error loss between the predicted dispersion attenuation characteristics and the actual dispersion attenuation characteristics of the sample two-phase porous medium elastic wave propagation model until a cutoff condition is met.

11. A model building device, characterized in that: The device comprises: a data acquisition module, configured to acquire a target data set, wherein the target data set includes target reservoir physical property parameters associated with a target two-phase porous medium elastic wave propagation model; a network processing module, configured to input the target data set into a trained two-phase medium wave characteristic learning network and output a plurality of wave equation coefficients of the target two-phase porous medium elastic wave propagation model; the trained two-phase medium wave characteristic learning network is obtained by training the neural network training method according to any one of claims 1 to 7; A model building module is used to build the target two-phase porous medium elastic wave propagation model based on the multiple wave equation coefficients.

12. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the neural network training method according to any one of claims 1 to 7 or the model building method according to any one of claims 8 to 9 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the neural network training method according to any one of claims 1 to 7 or the model building method according to any one of claims 8 to 9 are implemented.

Images