Forward Modeling Method, Forward Modeling Device and Electronic Device for Rock Physics Elastic Parameters

Abstract

The present invention provides a forward modeling method for rock physical elastic parameters, a forward modeling device and an electronic device. The forward modeling method for rock physical elastic parameters includes: obtaining depth-domain logging data of a study area to form a training sample set of a depth feedforward neural network; constructing a rock physics forward modeling model based on the depth feedforward neural network, and training by using the training sample set to obtain a non-linear mapping relationship between rock parameters and elastic parameters; and performing rock physics forward modeling of elastic parameters for a target well section based on the rock physics forward modeling model. The method of the present invention uses a depth feedforward neural network model to replace the conventional Xu-White rock physics forward modeling model, which can fully explore the internal relationship between rock parameters and elastic parameters, thereby establishing a non-linear mapping relationship between rock parameters and elastic parameters and improving the forward modeling accuracy of elastic parameters.

Description

Technical Field

[0001] The present invention belongs to the field of oil and gas geophysical exploration, and particularly relates to a rock physics forward modeling method, a forward modeling device, and an electronic device for improving the prediction accuracy of elastic parameters. Background Art

[0002] The three elastic parameters of longitudinal wave velocity, transverse wave velocity, and density are the bridges connecting various physical properties of rocks and seismic wave exploration. Using the three elastic parameters, physical quantities reflecting fluid properties can be obtained, thereby reducing the non-uniqueness of seismic amplitude interpretation, and having important applications in aspects such as AVO analysis of seismic exploration data, prestack inversion, and lithology, physical property, and fluid identification of reservoirs. However, in actual production, due to various reasons, the lack and incompleteness of the three elastic parameters seriously affect the subsequent exploration work.

[0003] In order to obtain accurate three elastic parameters, many domestic and foreign geophysicists have proposed empirical formulas and rock physics forward modeling models. Among them, the Xu-White model combines the Gassmann equation and the model and the differential equivalent medium theory, and simultaneously considers the influences of factors such as rock matrix, shale content, porosity size, pore shape, and pore fluid, and is widely used in the prediction of the three elastic parameters of shaly sandstone. However, this model requires the pore space to be divided small enough, and the computational amount is very large during iterative operations. In addition, the underground medium is quite complex, and conventional rock physics models cannot describe the physical parameter relationships of all types of rocks and must be simplified. Therefore, there will be certain uncertain factors in the prediction results of rock physics parameters, resulting in a reduction in the accuracy of rock physics forward modeling.

[0004] Aiming at the above deficiencies, the present invention is based on the rock parameter input data of the conventional Xu-White model, and fully explores the internal relationships between data through a typical deep learning model - a deep feedforward neural network, establishes a non-linear mapping relationship between rock parameters and elastic parameters, and obtains a rock physics forward modeling model based on the deep feedforward neural network, thereby improving the forward modeling accuracy of elastic parameters. Summary of the Invention

[0005] Aiming at the deficiencies of the conventional Xu-White rock physics model, which cannot meet the current high-precision seismic interpretation requirements, the present invention proposes a method for forward modeling of rock physics elastic parameters based on deep learning. Based on the rock parameter input data of the conventional Xu-White model, the present invention fully explores the internal relationships between rock physics parameters through a deep feedforward neural network, establishes a non-linear mapping relationship between rock parameters and elastic parameters, and obtains a rock physics forward modeling model based on the deep feedforward neural network, thereby improving the forward modeling accuracy of elastic parameters.

[0006] According to one aspect of the present invention, there is provided a forward modeling method for rock physical elastic parameters based on deep learning, including:

[0007] Obtain the depth-domain logging data of the study area and form a training sample set for the depth feedforward neural network;

[0008] Construct a rock physics forward model based on the depth feedforward neural network, and use the training sample set for training to obtain the non-linear mapping relationship between rock parameters and elastic parameters;

[0009] Based on the rock physics forward model, perform rock physics forward modeling of elastic parameters for the target well section.

[0010] Further, the depth-domain logging data of the study area is obtained through actual drilling logging and logging interpretation, and the depth-domain logging data is preprocessed by unit conversion and linear normalization to form a training sample set for the depth feedforward neural network, where the rock parameters and elastic parameters are used as the input and output of the depth feedforward neural network respectively.

[0011] Further, the depth-domain logging data Well(z) includes the following logging curves:

[0012] Elastic parameters Elastic(z) obtained from conventional logging and full waveform logging, including longitudinal wave velocity VP(z), shear wave velocity VS(z), and density DEN(z), as the output data y(z) of the depth feedforward neural network;

[0013] Rock parameters Rock(z) obtained by processing and interpreting logging data, including total porosity POR, shale content VCL, quartz content VQUA, and water saturation SW, as the input data x(z) of the depth feedforward neural network;

[0014] The input data x(z) after linear normalization processing and the output data y(z) after unit conversion form the training sample set Set(z).

[0015] Further, outliers in the depth-domain logging data Well(z) are removed, and the unit of the output data is converted.

[0016] Further, the input data is linearly normalized according to the following formula:

[0017]

[0018] In the formula, b(z) and a(z) are the logging values before and after normalization respectively; b(z)max and b(z)min are the maximum and minimum values of this parameter respectively.

[0019] Further, the topological structure of the deep feedforward neural network is: multiple hidden layers, fully connected and directed acyclic, where neurons in each hidden layer of the feedforward neural network are not connected to each other, neurons in separated hidden layers are not connected to each other, and neurons between adjacent layers are fully connected to each other.

[0020] Further, the input-output relationship of the deep feedforward neural network is:

[0021]

[0022] Among them, the output of the hidden layer of the deep feedforward neural network is:

[0023]

[0024] Excluding the input layer h (0) and the output layer h (L) , the number of hidden layers of the deep feedforward neural network is L - 1 layers, and the corresponding hyperparameters: the number of network layers, the number of neurons in each layer, and the activation function are:

[0025]

[0026] Among them, n0 = m, n L = s, the activation function of the input layer selects the ReLU function, and the activation functions of the remaining layers select the Sigmoid function. The parameters to be learned by the deep feedforward neural network are:

[0027]

[0028] Further, the rock physics forward modeling of elastic parameters for the target well section includes:

[0029] Select the rock parameters in the depth domain of the target well section as the input data x(z) of the deep feedforward neural network, perform linear normalization preprocessing on the input data x(z), and form the prediction sample set of the deep feedforward neural network;

[0030] Use the trained rock physics forward modeling model based on the deep feedforward neural network to process the prediction sample set to obtain three elastic parameters: the longitudinal wave velocity VP, the transverse wave velocity VS, and the density DEN.

[0031] According to another aspect of the present invention, there is provided a device for forward modeling of rock physics elastic parameters based on deep learning, including:

[0032] An acquisition unit that acquires the logging data in the depth domain of the study area and forms the training sample set of the deep feedforward neural network;

[0033] A training unit that constructs a rock physics forward modeling model based on a deep feedforward neural network and trains it using the training sample set to obtain a non-linear mapping relationship between rock parameters and elastic parameters;

[0034] A forward modeling unit that performs rock physics forward modeling of elastic parameters for a target well section based on the rock physics forward modeling model.

[0035] According to another aspect of the present invention, there is provided an electronic device, which includes:

[0036] A memory that stores executable instructions;

[0037] A processor that runs the executable instructions in the memory to implement the above-mentioned method for forward modeling of rock physics elastic parameters based on deep learning.

[0038] The rock physics forward modeling method based on a deep feedforward neural network of the present invention has the following characteristics:

[0039] Using a deep feedforward neural network model to replace the conventional Xu-White rock physics forward modeling model can fully explore the internal relationship between rock parameters and elastic parameters, thereby establishing a non-linear mapping relationship between rock parameters and elastic parameters and improving the forward modeling accuracy of elastic parameters. The deep neural network model only needs to be trained once, and then it can perform rock physics forward modeling on other target well sections in the study area to obtain high-precision elastic parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By describing the exemplary embodiments of the present disclosure in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present disclosure will become more apparent. Among them, in the exemplary embodiments of the present disclosure, the same reference numerals generally represent the same components.

[0041] Figure 1 It is a flow chart of the method for forward modeling of rock physics elastic parameters based on deep learning of the present invention.

[0042] Figure 2 It is a flow chart of the rock physics forward modeling method based on a deep feedforward neural network according to an embodiment of the present invention.

[0043] Figure 3 It is the structure of a deep feedforward neural network according to an embodiment of the present invention.

[0044] Figure 4 It is the well logging curve in the depth domain of the target well according to an embodiment of the present invention.

[0045] Figure 5 It is the training sample set 1 according to an embodiment of the present invention.

[0046] Figure 6 The training sample set 2 of the embodiment according to the embodiment of the present invention.

[0047] Figure 7 The training result of the training sample sample set 1 of the embodiment according to the embodiment of the present invention.

[0048] Figure 8 The training result of the training sample sample set 2 of the embodiment according to the embodiment of the present invention.

[0049] Figure 9 The input data of the target well section of the embodiment according to the embodiment of the present invention.

[0050] Figure 10 The forward modeling result of petrophysics of the target well section of the embodiment according to the embodiment of the present invention.

[0051] Figure 11 The schematic diagram of the device for forward modeling of petrophysical elastic parameters based on deep learning according to the embodiment of the present invention. Detailed implementation manners

[0052] The preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0053] In view of the deficiencies of the conventional Xu-White petrophysical forward modeling model, which cannot effectively meet the current high-precision seismic interpretation requirements, the present invention proposes a method for forward modeling of petrophysical elastic parameters based on deep learning. This method uses a deep feedforward neural network model to replace the conventional Xu-White petrophysical forward modeling model, combines the rock parameters of the input data of the conventional Xu-White model and the corresponding elastic three parameters as the training sample set, fully explores the internal relationship between the data, establishes a non-linear mapping relationship between the input data and the elastic parameters, and obtains a petrophysical forward modeling model based on the deep feedforward neural network, thereby improving the forward modeling accuracy of the elastic parameters.

[0054] As Figure 1 shown, the present disclosure proposes a method for forward modeling of petrophysical elastic parameters based on deep learning, including:

[0055] Obtain the depth-domain logging data of the study area and form the training sample set of the deep feedforward neural network;

[0056] Construct a rock physics forward model based on a deep feedforward neural network, and use the training sample set for training to obtain a non-linear mapping relationship between rock parameters and elastic parameters;

[0057] Based on the rock physics forward model, perform rock physics forward of elastic parameters for the target well section.

[0058] Specifically, the rock physics elastic parameter forward method based on deep learning of the present invention realizes the prediction of elastic parameters Elastic(z), including longitudinal wave velocity VP(z), transverse wave velocity VS(z), and density DEN(z), through a deep feedforward neural network and deep domain logging data Well(z), where z represents the deep domain, and the unit is m or ft. The method of the present invention includes two processes: a network training process and a network application process. The network training process is used to train the deep feedforward neural network to realize the rock physics forward function.

[0059] First, obtain the deep domain logging data of the study area obtained through actual drilling logging and logging interpretation, perform unit conversion and linear normalization preprocessing on the logging data, so as to form a training sample set of the deep feedforward neural network, and use the rock parameters and elastic parameters as the input and output of the deep feedforward neural network respectively.

[0060] Preferably, the deep domain logging data Well(z) includes the following logging curves: ① Obtain the elastic parameters Elastic(z) from conventional logging and full waveform logging, including longitudinal wave velocity VP(z), transverse wave velocity VS(z), and density DEN(z), and use this as the output data y(z) of the deep feedforward neural network; ② After processing and interpreting the logging data, obtain the rock parameters Rock(z), including total porosity POR, shale content VCL, quartz content VQUA, water saturation SW, etc., as the input data x(z) of the deep feedforward neural network.

[0061] Outliers in the deep domain logging data Well(z) can be removed. The deep domain logging curve Well(z) represents that each depth point corresponds to a data sample point; then convert the unit of the output data y(z), where the unit of the longitudinal wave velocity VP(z) after conversion is km / s, the unit of the transverse wave velocity VS(z) is km / s, and the unit of the density DEN(z) is g / cm 3 。

[0062] Preferably, the input data can be linearly normalized according to the following formula:

[0063]

[0064] where \(b(z)\) and \(a(z)\) are the logging values before and after normalization, respectively; \(b(z)_{max}\) and \(b(z)_{min}\) are the maximum and minimum values of this parameter, respectively. The input data \(x(z)\) after linear normalization and the output data \(y(z)\) after unit conversion form the training sample set \(Set(z)\).

[0065] Next, a rock physics forward modeling model based on a deep feedforward neural network is constructed, and it is trained using the training sample set to obtain the non-linear mapping relationship between rock parameters and elastic parameters.

[0066] Specifically, the topological structure of the deep feedforward neural network is: multiple hidden layers, fully connected and directed acyclic, that is, each neuron between each hidden layer of the feedforward neural network is not connected to each other, each neuron between the separated hidden layers is not connected to each other, and the neurons between adjacent layers are fully connected to each other. When the number of hidden layers of the network is 2 or more, its structure is as Figure 3 shown.

[0067] For the given input data of rock parameters \(x(z)\in R\) m , and the output data of elastic parameters \(y(z)\in R\) s , then the output of the hidden layer of the deep feedforward neural network is:

[0068]

[0069] Excluding the input layer \(h\) (0) and the output layer \(h\) (L) , the number of hidden layers of the deep feedforward neural network is \(L - 1\) layers, and the corresponding hyperparameters: the number of network layers, the number of neurons in each layer, and the activation function are:

[0070]

[0071] where \(n_0 = m\), \(n\) L \(= s\), the activation function of the input layer selects the ReLU function, and the activation function of the remaining layers selects the Sigmoid function. The parameters to be learned by the deep feedforward neural network are:

[0072]

[0073] Then the relationship between the input and output of the deep feedforward neural network is:

[0074]

[0075] According to the above formula, the rock physics forward modeling model based on deep learning can be obtained:

[0076] \(y = f(x,\theta)\)

[0077] Using the obtained training sample set, the rock physics forward model is trained to establish a non - linear mapping relationship between the input rock parameters and the elastic parameters, so as to obtain a rock physics forward model based on a deep feed - forward neural network. And this model only needs to be trained once, and then it can perform rock physics forward modeling on other target well sections in the study area to obtain high - precision elastic parameters.

[0078] Finally, based on this rock physics forward model, rock physics forward modeling of elastic parameters is carried out for the target well section. For the target well section that needs to carry out rock physics forward modeling, first, rock parameters such as porosity POR, shale content VCL, quartz content VQUA, and water saturation SW in the depth domain of the target well section are also selected as the input data x(z) of the deep feed - forward neural network; then the input data is pre - processed by linear normalization to form a prediction sample set of the deep feed - forward neural network; finally, using the trained rock physics forward model based on the deep feed - forward neural network, the prediction sample set to be predicted is processed to obtain three elastic parameters, namely the longitudinal wave velocity VP, the shear wave velocity VS, and the density DEN.

[0079] To facilitate the understanding of the solution and its effects of the embodiments of the present invention, a specific application example is given below. Those skilled in the art should understand that this example is only for facilitating the understanding of the present invention, and any specific details are not intended to limit the present invention in any way.

[0080] Example 1

[0081] In this embodiment, a deep - water drilling well in a sandstone block in southern China is taken as an example, and the method provided by the present invention is used for rock physics forward modeling.

[0082] Figure 2 The technical flow chart of rock physics forward modeling according to the embodiment of the present invention is shown.

[0083] Figure 3 The structure diagram of the deep feed - forward neural network is shown. The deep feed - forward neural network includes an input layer, an output layer, and two or more hidden layers.

[0084] Figure 4 The depth - domain logging data of the drilling well in the embodiment is shown. It can be seen from the figure that the hole enlargement phenomenon occurs in the deeper well section of this well. This is because the formation horizontal pressure difference is small, the clay content of the formation rock is high, showing strong hydration characteristics, resulting in serious wellbore instability during deep - water drilling. Therefore, in the embodiment, the data of the well section with stable hole diameter is used for training the depth neural network to obtain a rock physics forward model, and then the elastic parameters of the hole - enlarged well section are predicted by forward modeling.

[0085] Such as Figure 2As shown in the figure, elastic parameters are first obtained from conventional logging and full-wave logging. After processing and interpreting the logging data, rock parameters are obtained, and the rock parameters and elastic parameters are used as the input and output of the depth feedforward neural network respectively.

[0086] Specifically, two well sections with stable hole diameters in the well, namely 3806m - 4128m and 4167m - 4497m, are selected to construct two pseudo-wells. Five rock parameters, namely total porosity POR, shale content VCL, coal content VCOA, quartz content VQUA, and water saturation SW, are used as the input data x(z) of the depth feedforward neural network, and three elastic parameters, namely longitudinal wave velocity VP, transverse wave velocity VS, and density DEN, are used as the output data y(z) of the depth feedforward neural network.

[0087] Next, the units of the elastic parameters are converted, and the rock parameters are preprocessed by linear normalization. Specifically, the unit of the longitudinal wave velocity VP in the output data of the two pseudo-wells is converted to km / s, the unit of the transverse wave velocity VS is converted to km / s, and the unit of the density DEN is converted to g / cm 3 ; The input data is linearly normalized according to the following formula. The input data x(z) and output data y(z) of the processed pseudo-wells are respectively as Figure 5 and Figure 6 shown, thus forming the training sample set of the depth feedforward neural network.

[0088]

[0089] Next, a depth feedforward neural network structure for rock physics forward modeling is designed. Specifically, a 10-layer depth feedforward neural network is constructed. The neural network contains 8 hidden layers, and the number of neurons in each hidden layer is 10. The number of iterations for neural network training is set to 1000, and the conjugate gradient method is selected as the optimization algorithm. Using the training sample set composed of the two pseudo-wells, the depth feedforward neural network is trained to obtain a rock physics forward modeling model based on the depth feedforward neural network. The training results of the depth feedforward neural network of the training sample set are as Figure 7 and Figure 8 shown, and the correlation coefficient and mean error between the training results and the original elastic parameters are shown in Table 1.

[0090] Table 1 Rock physics forward modeling training results based on depth feedforward neural network

[0091] VP VS DEN Correlation coefficient 92.55％ 93.93％ 94.68％ Mean error 0.16 km / s 0.12 km / s <![CDATA[0.04g / cm 3 >

[0092] Finally, using the trained rock physics forward modeling model based on the deep feedforward neural network, the corresponding elastic parameters are forward modeled from the rock parameters of the well section to be measured (the rock parameters are obtained by processing and interpreting conventional logging data). Specifically, for the target well section of the embodiment, first, five rock parameters, namely the total porosity POR, shale content VCL, coal content VCOA, quartz content VQUA, and water saturation SW in the depth domain of the target well section, are selected as the input data of the deep feedforward neural network, and linear normalization preprocessing is performed to form the prediction sample set of the deep feedforward neural network, as shown in Figure 9 shown. Using the trained rock physics forward modeling model based on the deep feedforward neural network, the prediction sample set is processed to obtain three elastic parameters, namely the longitudinal wave velocity VP, the transverse wave velocity VS, and the density DEN, and the results are as shown in Figure 10 shown.

[0093] From Figures 7 - 10 the training and prediction results, it can be seen that the method proposed by the invention can fully explore the internal relationship between the input rock parameters and the output elastic parameters of the conventional Xu-White rock physics forward modeling model. As can be seen from Table 1, the method proposed by the invention can obtain highly accurate elastic parameters. The present invention can use the well section with stable hole diameter for training and then predict the enlarged diameter section, thus effectively avoiding the influence of the enlarged diameter on the elastic parameters and providing strong data support for the subsequent oil and gas geophysical exploration in the study area.

[0094] Example 2

[0095] As shown in Figure 11 the present embodiment provides a rock physics elastic parameter forward modeling device based on deep learning, including:

[0096] An acquisition unit that acquires the logging data in the depth domain of the study area and forms the training sample set of the deep feedforward neural network;

[0097] A training unit that constructs a rock physics forward modeling model based on the deep feedforward neural network and uses the training sample set for training to obtain the non-linear mapping relationship between the rock parameters and the elastic parameters;

[0098] A forward modeling unit that performs rock physics forward modeling of the elastic parameters for the target well section based on the rock physics forward modeling model.

[0099] An acquisition unit, a training unit, and a forward modeling unit are communicatively connected in sequence. The acquisition unit provides a training sample set to the training unit, and a rock physics forward model is trained. Based on the rock physics forward model, the forward modeling unit performs rock physics forward modeling of elastic parameters for a target well section. In addition, the acquisition unit is communicatively connected to the forward modeling unit and is configured to provide rock parameters of a well section to be measured to the forward modeling unit so that the forward modeling unit performs forward modeling to obtain corresponding elastic parameters.

[0100] Example 3

[0101] This embodiment provides an electronic device, which includes: a memory storing executable instructions; and a processor that runs the executable instructions in the memory to implement the above-mentioned method for forward modeling of rock physics elastic parameters based on deep learning.

[0102] An electronic device according to an embodiment of the present disclosure includes a memory and a processor.

[0103] The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0104] The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In an embodiment of the present disclosure, the processor is configured to run the computer-readable instructions stored in the memory.

[0105] Those skilled in the art should understand that, in order to solve the technical problem of how to obtain a good user experience effect, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included in the protection scope of the present disclosure.

[0106] For a detailed description of this embodiment, reference may be made to the corresponding descriptions in the foregoing embodiments, and details are not repeated here.

[0107] Example 4

[0108] An embodiment of the present disclosure provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the above-mentioned method for forward modeling of rock physics elastic parameters based on deep learning is implemented.

[0109] A computer-readable storage medium according to an embodiment of the present disclosure stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are run by a processor, all or part of the steps of the methods of the various embodiments of the present disclosure described above are executed.

[0110] The above computer-readable storage medium includes but is not limited to: optical storage media (such as CD-ROMs and DVDs), magneto-optical storage media (such as MOs), magnetic storage media (such as magnetic tapes or external hard drives), media with built-in rewritable non-volatile memories (such as memory cards), and media with built-in ROMs (such as ROM cartridges).

[0111] Those skilled in the art should understand that the purpose of the above description of the embodiments of the present invention is only to exemplarily illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any of the examples given.

[0112] The various embodiments of the present invention have been described above. The above description is exemplary and not exhaustive, and is also not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A forward modeling method for rock physical elastic parameters based on deep learning, characterized in that, Including: Obtain the depth-domain logging data of the study area to form a training sample set for the depth feedforward neural network; Construct a rock physics forward modeling model based on the depth feedforward neural network, and use the training sample set for training to obtain the non-linear mapping relationship between rock parameters and elastic parameters; Based on the rock physics forward modeling model, perform rock physics forward modeling of elastic parameters for the target well section; Among them, use the well section data with stable well diameter for depth neural network training to obtain the rock physics forward modeling model, and then perform forward prediction of the elastic parameters of the enlarged well section; Among them, obtain the depth-domain logging data of the study area through actual drilling logging and logging interpretation, perform unit conversion and linear normalization preprocessing on the depth-domain logging data to form a training sample set for the depth feedforward neural network, where the rock parameters and elastic parameters are used as the input and output of the depth feedforward neural network respectively; Select a well section with stable well diameter to construct a pseudo-well. The elastic parameters Elastic(z) obtained from conventional logging and full-wave logging, including the longitudinal wave velocity VP(z), the shear wave velocity VS(z), and the density DEN(z), are used as the output data y(z) of the depth feedforward neural network; the rock parameters Rock(z) obtained after processing and interpreting the logging data, including the total porosity POR, the shale content VCL, the quartz content VQUA, and the water saturation SW, are used as the input data x(z) of the depth feedforward neural network; the input data x(z) after linear normalization processing and the output data y(z) after unit conversion form a training sample set Set(z).

2. The forward modeling method for rock physical elastic parameters based on deep learning according to claim 1, wherein Eliminate the outliers in the depth-domain logging data Well(z), and convert the unit of the output data.

3. The forward modeling method for rock physical elastic parameters based on deep learning according to claim 1 or 2, characterized in that, Perform linear normalization processing on the input data according to the following formula: In the formula, b(z) and a(z) are the logging values before and after normalization respectively; b(z)max and b(z)min are the maximum and minimum values of this parameter respectively.

4. The forward modeling method for rock physical elastic parameters based on deep learning according to claim 1, wherein, The topological structure of the depth feedforward neural network is: multi-hidden layers, fully connected and directed acyclic. Among them, the neurons in each hidden layer of the feedforward neural network are not connected to each other, the neurons in the separated hidden layers are not connected to each other, and the neurons between adjacent layers are fully connected to each other.

5. The forward modeling method for rock physics elastic parameters based on deep learning according to claim 4, characterized in that The input-output relationship of the depth feedforward neural network is: Among them, the output of the hidden layer of the depth feedforward neural network is: Remove the input layer h (0) and the output layer h (L) , the number of hidden layers of the deep feedforward neural network is L - 1 layers, and the corresponding hyperparameters: the number of network layers, the number of neurons in each layer, and the activation function are: where n0 = m, n L = s, the activation function of the input layer selects the ReLU function, and the activation functions of the remaining layers select the Sigmoid function. The parameters to be learned by the deep feedforward neural network are as follows:

6. The forward modeling method for rock physical elastic parameters based on deep learning according to claim 1, characterized in that Performing rock physics forward modeling of elastic parameters for the target well section includes: Select the rock parameters in the depth domain of the target well section as the input data x(z) of the depth feedforward neural network, and perform linear normalization preprocessing on the input data x(z) to form a prediction sample set for the depth feedforward neural network; Use the trained rock physics forward modeling model based on the depth feedforward neural network to process the prediction sample set to obtain the three elastic parameters of the corresponding longitudinal wave velocity VP, shear wave velocity VS, and density DEN.

7. A forward modeling device for rock physical elastic parameters based on deep learning, characterized in that, Including: An acquisition unit that acquires the depth-domain logging data of the study area to form a training sample set for the depth feedforward neural network; A training unit that constructs a rock physics forward model based on a deep feedforward neural network and trains it using the training sample set to obtain a non-linear mapping relationship between rock parameters and elastic parameters; A forward unit that performs rock physics forward calculation of elastic parameters for a target well section based on the rock physics forward model; Among them, the well section data with stable well diameter is used for deep neural network training to obtain a rock physics forward model, and then the elastic parameters of the enlarged well section are forward predicted; Among them, the depth-domain logging data of the study area is obtained through actual drilling logging and logging interpretation, and the depth-domain logging data is preprocessed by unit conversion and linear normalization to form a training sample set for the deep feedforward neural network, where the rock parameters and elastic parameters are used as the input and output of the deep feedforward neural network respectively; Select a well section with stable well diameter to construct a pseudo-well. The elastic parameters Elastic(z) obtained from conventional logging and full-wave logging, including longitudinal wave velocity VP(z), transverse wave velocity VS(z), and density DEN(z), are used as the output data y(z) of the deep feedforward neural network; the rock parameters Rock(z) obtained after processing and interpreting logging data, including total porosity POR, shale content VCL, quartz content VQUA, and water saturation SW, are used as the input data x(z) of the deep feedforward neural network; the input data x(z) after linear normalization processing and the output data y(z) after unit conversion form a training sample set Set(z).

8. An electronic device, characterized in that, The electronic device includes: A memory that stores executable instructions; A processor that runs the executable instructions in the memory to implement the rock physics elastic parameter forward calculation method based on deep learning according to any one of claims 1-6.

Images