Rock physics and deep learning dual-driven reservoir physical property parameter intelligent prediction method

Abstract

The invention discloses a rock physics and deep learning dual-driven reservoir physical property parameter intelligent prediction method, and belongs to the technical field of unconventional reservoir seismic exploration, and the method comprises the following steps: building a rock physics model based on physical property parameter and elastic parameter observation data of calibration logging; respectively estimating probability distribution of the physical parameters and the elastic parameters by using a Gaussian mixture model, and forming a label data set by physical parameter label-free data and elastic parameter label-free data; building a residual convolutional neural network ResNet model, and training an initial ResNet model by using the labeled data; inputting elastic parameter label-free data into the ResNet model to form a full pseudo label data set; iteratively training the initial ResNet model to obtain a final ResNet model; and inputting the observation data to obtain predicted reservoir physical property parameters. Through rock physics and deep learning dual driving, the advantages of geological prior and data driving are fused, and the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unconventional reservoir seismic exploration, and specifically to an intelligent prediction method for reservoir physical property parameters driven by both rock physics and deep learning. Background Art

[0002] Accurate prediction of reservoir physical properties (such as porosity, water saturation, and clay content) is crucial to oil and gas exploration. These parameters not only help predict and identify potential reservoirs, but also play an important role in reservoir production prediction, fluid identification, and oil recovery analysis in the petroleum industry.

[0003] However, for unconventional reservoirs such as tight sandstone, their complex pore structure poses a huge challenge to the accurate prediction of reservoir physical parameters. Unconventional reservoirs such as tight sandstone usually develop complex pore structures, which leads to a complex mapping relationship between reservoir physical properties and elastic characteristics. Conventional rock physics models have difficulty accurately describing the nonlinear relationship between physical properties and elastic parameters, hindering the accurate prediction of physical parameters through classical rock physics modeling and inversion methods. In addition, although deep learning methods have made significant progress in many fields, they can automatically extract complex features in data and have strong nonlinear fitting capabilities, providing new methods for predicting reservoir physical parameters. However, conventional deep learning methods lack the use of geological prior information when processing geological data, and the logging observation data of unconventional reservoirs is limited. It is difficult to provide sufficient physical property-elastic parameter label samples, resulting in limited physical interpretability and generalization ability of neural network models, which restricts the applicability of conventional deep learning methods to the prediction of unconventional reservoir physical parameters. Summary of the Invention

[0004] Technical problem solved: In response to the problems faced in predicting reservoir physical property parameters, the present invention provides an intelligent prediction method for reservoir physical property parameters driven by both rock physics and deep learning. By dual-driving rock physics and deep learning, it integrates the advantages of geological priors and data-driven to improve prediction accuracy.

[0005] Technical solution: The present invention provides a method for intelligently predicting reservoir physical parameters driven by both rock physics and deep learning, comprising the following steps: Step 1: Based on the physical properties and elastic parameter observation data of the calibration logging, a rock physics model is constructed to predict the elastic parameters and match them with the elastic parameter observation data of the logging; Step 2: Based on the physical and elastic parameter observation data of the calibration logging, the probability distribution of the physical and elastic parameters are estimated using the Gaussian mixture model. Step 3: Based on the probability distribution of physical property parameters and elastic parameters, unlabeled data of physical property parameters and elastic parameters are generated through Monte Carlo random simulation, and then combined with the well logging observation data to form a labeled data set; Step 4: Build a Residual Convolutional Neural Network (ResNet) model consisting of convolutional layers, pooling layers, fully connected layers, and residual modules, and determine its hyperparameters. Step 5: Use labeled data to train the ResNet model and build the initial ResNet model; Step 6: Input the elastic parameter unlabeled data into the initial ResNet model to generate physical parameter data, forming the first part of pseudo-labeled data; Step 7: For the unlabeled data of physical parameters, elastic parameter data is predicted by rock physics model to form the second part of pseudo-labeled data, which is combined with the first part of pseudo-labeled data to form a full pseudo-labeled data set; Step 8: Fuse the labeled dataset and the fully pseudo-labeled dataset, iteratively train the initial ResNet model, and obtain the final ResNet model; Step 9: Input pre-stack seismic inversion or logging elastic parameter observation data into the final ResNet model to predict reservoir physical properties.

[0006] Preferably, step 1 includes the following sub-steps: Step 11: Calculate the bulk modulus K of the rock matrix using the VRH average model m and shear modulus μ m , and its calculation formula is: ; ; Where: K mi 、μ mi 、v mi are the bulk modulus, shear modulus and percentage content of the i-th mineral in the rock matrix, respectively, and n is the total number of mineral types; Step 12: Calculate the bulk modulus K of the pore fluid using the Voigt model fl , and its calculation formula is: ; Where: S w is water saturation; K w , K g are the bulk moduli of the liquid and gas in the pores, respectively; Step 13: Calculate the bulk modulus K of the dry rock skeleton using the Kuster-Toksoz model dry and shear modulus μ dry , and its calculation formula is ; ; Where: K * and μ * are the bulk modulus and shear modulus of the containing material, respectively; v is the pore volume fraction, α is the pore aspect ratio, T ijij and T iijj are the pore structure factors; Step 14: Calculate the bulk modulus K of fluid-saturated rock using the Biot-Gassmann equation sat and shear modulus μ sat , and its calculation formula is: ; ; Where: is the porosity; Step 15: Based on K sat 、μ sat and density ρ to calculate the longitudinal wave velocity V p , shear wave velocity V s , and its calculation formula is: ; ; ; Where: ρ fl and ρ m is the density of the fluid and rock matrix.

[0007] Preferably, the probability distribution function of the Gaussian mixture model in step 2 is: ; Where: m is the physical parameter or elastic parameter; N c is the number of Gaussian components; ω k is the weight; is the probability density function of the Gaussian component, μ k and are the mean and covariance matrices.

[0008] Preferably, the unlabeled data Y of the physical property parameters generated by the Monte Carlo random simulation in step 3 is * and elastic parameters unlabeled data X * A probability distribution that satisfies a Gaussian mixture model.

[0009] Preferably, the process of building the ResNet model in step 4 is: ; Among them, Conv is the convolution layer, which is used to extract features; ReLU is the activation function, which is used to insert nonlinearity in the calculation; Pool is the pooling layer, which is used to reduce the spatial dimension of the features and reduce the amount of calculation; ResidualBlock is a combination of residual blocks, which is used to solve the gradient disappearance problem in deep networks; FC is the fully connected layer, which is used to perform classification or regression tasks; Wherein: the structure of the residual block satisfies: ; Where: is the output of the convolution layer and activation function in the residual block, x is the input, and y is the output.

[0010] Preferably, in step 5, an initial ResNet model is constructed, and its expression is: ; Where: are the parameters of the network structure; is the training process of the residual convolutional neural network model, for Labeled data.

[0011] Preferably, the first part of pseudo-label data in step 6 The expression is: ; Where: is the initial ResNet model; is the unlabeled data of i elastic parameters, is the prediction result of the initial ResNet model.

[0012] Preferably, the second part of pseudo-label data in step 7 The expression is: ; Where: RPM is the rock physics model; is the unlabeled data of k physical property parameters; are rock physics model parameters; predicting results for rock physics models; Fully pseudo-labeled dataset The expression is: ; Where: is the first part of pseudo-label data; is the second part of pseudo-label data; the addition operation represents concatenation.

[0013] Preferably, the expression for fusion data set in step 8 is: ; Where: θ * is the parameter of the final network structure; ResNet train is the training process of the residual convolutional neural network model, and (X, Y) is a labeled dataset.

[0014] Preferably, the expression of the reservoir physical property parameter predicted in step 9 is: ; Where: is the final residual convolutional neural network model; x is the elastic parameter observation data including longitudinal wave velocity, shear wave velocity and density; These are the predicted reservoir physical property parameters including porosity, water saturation, and mud content.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: 1. This invention combines rock physics models with deep learning technology to form a dual-driven approach based on rock physics and deep learning. This method fully utilizes geological prior knowledge to improve physical interpretability, supplements labeled data to enhance generalization capabilities, and leverages the nonlinear fitting advantages of deep learning to achieve more accurate and reliable predictions of reservoir physical properties, providing technical support for oil and gas exploration and development. 2. By combining rock physics models with seismic reflectivity equations, the present invention establishes a forward model that links reservoir physical parameters with seismic response. This allows the physical parameters of tight sandstone reservoirs to be estimated directly from seismic data, more accurately describing the nonlinear relationship between physical parameters and elastic parameters. 3. When processing geological data, the present invention adopts a Gaussian mixture model as a prior model to handle the complex prior distribution of reservoir physical parameters; at the same time, a pseudo-label data set is added to train the model to make up for the limited label data of physical property-elasticity parameters, thereby improving the physical interpretability and generalization ability of the deep learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of intelligent prediction of reservoir physical property parameters of the present invention; Figure 2 Graphs showing data from well logging observations in an embodiment of the present invention ((a) is porosity; (b) is water saturation; (c) is shale content; (d) is P-wave velocity; (e) is S-wave velocity; and (f) is density); Figure 3 Graph showing the physical property parameter prediction results of the initial residual convolutional neural network model in the embodiment of the present application ((a) is porosity; (b) is water saturation; (c) is mud content); Figure 4This is a diagram of the physical parameter prediction results of the final residual convolutional neural network model in the embodiment of the present application ((a) is porosity; (b) is water saturation; (c) is mud content). DETAILED DESCRIPTION

[0017] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the following Figures 1 to 4 The technical solutions of the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, the present invention discloses a method for intelligent prediction of reservoir physical parameters driven by both rock physics and deep learning, comprising the following steps: 1. Based on the physical properties and elastic parameter observation data of the calibration logging, a rock physics model is constructed to predict the elastic parameters and match them with the elastic parameter observation data of the logging.

[0019] (1) Set basic rock physical parameters, including bulk modulus, shear modulus, density of minerals, bulk modulus and density of fluids; use VRH average model to calculate the bulk modulus K of rock matrix m and shear modulus μ m , and its calculation formula is: ; ; Where: K mi 、μ mi 、v mi are the bulk modulus, shear modulus and percentage content of the i-th mineral in the rock matrix, respectively, and n is the total number of mineral types.

[0020] (2) Calculate the bulk modulus K of pore fluid using the Voigt model fl , and its calculation formula is: ; Where: S w is water saturation; K w , K g are the bulk moduli of the liquid and gas in the pores, respectively.

[0021] (3) Calculate the bulk modulus K of the dry rock skeleton using the Kuster-Toksoz model dry and shear modulus μ dry , and its calculation formula is ; ; Where: K * and μ * are the bulk modulus and shear modulus of the containing material, respectively; v is the pore volume fraction, α is the pore aspect ratio, T ijij and T iijj are the pore structure factors, respectively.

[0022] (4) Calculate the bulk modulus K of fluid-saturated rock using the Biot-Gassmann equation sat and shear modulus μ sat , and its calculation formula is: ; ; Where: is the porosity.

[0023] (5) Based on K sat 、μ sat and density ρ to calculate the longitudinal wave velocity V p , shear wave velocity V s , and its calculation formula is: ; ; ; Where: ρ fl and ρ m is the density of the fluid and rock matrix.

[0024] In the embodiment of the present invention, the rock physical parameters used are as follows: , , , , , , , , , .

[0025] 2. Based on the physical and elastic parameter observation data of calibration logging, the Gaussian mixture model is used to estimate the probability distribution of physical and elastic parameters respectively.

[0026] The probability distribution function of the Gaussian mixture model is: ; Where: m is the physical parameter or elastic parameter; N c is the number of Gaussian components; ω k is the weight; is the probability density function of the Gaussian component, μ k and are the mean and covariance matrices.

[0027] 3. According to the probability distribution of physical parameters and elastic parameters, the unlabeled data of physical parameters and elastic parameters are generated by Monte Carlo random simulation. The unlabeled data Y of physical parameters generated by Monte Carlo random simulation is * and elastic parameters unlabeled data X * The probability distribution that satisfies the Gaussian mixture model is formed by forming a labeled data set (X, Y) based on the well logging physical property parameter observation data Y and the elastic parameter observation data X.

[0028] In the embodiment of the present invention, the number of Gaussian components N c =3, and the total number of Monte Carlo sample sets is 5000.

[0029] 4. Build a residual convolutional neural network ResNet model consisting of convolutional layers, pooling layers, fully connected layers and residual modules, and determine its hyperparameters.

[0030] (1) The construction process of the ResNet model is: ; Among them, Conv is the convolution layer used to extract features; ReLU is the activation function used to insert nonlinearity in the calculation; Pool is the pooling layer used to reduce the spatial dimension of the features and reduce the amount of calculation; ResidualBlock is a combination of residual blocks used to solve the gradient disappearance problem in deep networks; FC is the fully connected layer used to perform classification or regression tasks.

[0031] (2) The structure of the residual block satisfies: ; Where: It is the output of the convolution layer and activation function in the residual block, representing the residual of network learning, x is the input, and y is the output.

[0032] 5. Use labeled data to train the ResNet model and build the initial ResNet model, which is expressed as: ; Where: are the parameters of the network structure; is the training process of the residual convolutional neural network model, is i labeled data.

[0033] In an embodiment of the present invention, the parameters for building a residual convolutional neural network model are: the convolution kernel size is 3, the number of input and output channels is the same, the step size is 1, the padding is 1, and it is followed by a ReLU activation function; the residual block output and the convolution layer output are directly added to form a residual learning mechanism; the network end is connected to a fully connected layer to compress the number of channels to the target dimension.

[0034] In the embodiment of the present invention, the Figure 2 The well logging observation data shown is used as labeled data to train the residual convolutional neural network model and construct the initial ResNet model.

[0035] 6. Input the elastic parameter unlabeled data into the initial ResNet model to generate physical parameter data to form the first part of pseudo-label data; the first part of pseudo-label data The expression is: ; Where: is the initial ResNet model; is the unlabeled data of i elastic parameters, is the prediction result of the initial ResNet model.

[0036] 7. For the unlabeled data of physical parameters, the elastic parameter data is predicted through the rock physics model to form the second part of pseudo-labeled data, which is combined with the first part of pseudo-labeled data to form a full pseudo-labeled data set.

[0037] (1) The second part of pseudo-label data The expression is: ; Where: RPM is the rock physics model; is the unlabeled data of k physical property parameters; are rock physics model parameters; Predict results for rock physics models.

[0038] (2) Full pseudo-label dataset The expression is: ; Where: is the first part of pseudo-label data; is the second part of pseudo-label data; the addition operation represents concatenation.

[0039] 8. Fuse the labeled dataset and the fully pseudo-labeled dataset, iteratively train the initial ResNet model, and obtain the final ResNet model, which is expressed as: ; Where: θ *is the parameter of the final network structure; ResNet train is the training process of the residual convolutional neural network model, and (X, Y) is a labeled dataset.

[0040] 9. Input pre-stack seismic inversion or logging elastic parameter observation data into the final ResNet model to predict reservoir physical properties. The expression is: ; Where: is the final residual convolutional neural network model; x is the elastic parameter observation data including P-wave velocity, S-wave velocity, and density; y is the predicted reservoir physical property parameters including porosity, water saturation, and mud content.

[0041] In the embodiment of the present invention, Figure 2 The data curve diagram of well logging observation ((a) is porosity; (b) is water saturation; (c) is mud content; (d) is P-wave velocity; (e) is S-wave velocity; (f) is density). Figure 3 Comparison between the physical property parameter prediction results of the initial residual convolutional neural network model and the logging observation data ((a) is porosity; (b) is water saturation; (c) is mud content). Although the trends of the physical property parameter prediction results are basically consistent with the observed data, there are still certain deviations in the values (such as water saturation). Figure 4 The physical property parameter prediction results of the final residual convolutional neural network model are compared with the logging observation data ((a) is porosity; (b) is water saturation; (c) is mud content). The comparison shows that the physical property parameter prediction results of the final residual convolutional neural network model (red dotted line) are consistent with the observation data (blue solid line). It can be seen that the present invention combines the rock physics model with the seismic reflectivity equation to establish a forward model that links reservoir physical parameters with seismic response, directly estimates the physical property parameters of tight sandstone reservoirs from seismic data, and more accurately describes the nonlinear relationship between physical property parameters and elastic parameters; when processing geological data, a Gaussian mixture model is used as a priori model to handle the complex prior distribution of reservoir physical property parameters; at the same time, a pseudo-label data set is added for model training to make up for the limited label data of physical property-elastic parameters, thereby improving the physical interpretability and generalization ability of the deep learning model; it can effectively improve the accuracy of reservoir physical property parameter prediction, and is expected to play an important role in the prediction of complex reservoir physical property parameters, providing technical support for oil and gas exploration and development.

[0042] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for intelligent prediction of reservoir physical parameters driven by both rock physics and deep learning, characterized in that: The following steps are involved: Step 1: Based on the physical properties and elastic parameter observation data of the calibration logging, a rock physics model is constructed to predict the elastic parameters and match them with the elastic parameter observation data of the logging; Step 2: Based on the physical and elastic parameter observation data of the calibration logging, the probability distribution of the physical and elastic parameters are estimated using the Gaussian mixture model. Step 3: Based on the probability distribution of physical property parameters and elastic parameters, unlabeled data of physical property parameters and elastic parameters are generated through Monte Carlo random simulation, and then combined with the well logging observation data to form a labeled data set; Step 4: Build a Residual Convolutional Neural Network (ResNet) model consisting of convolutional layers, pooling layers, fully connected layers, and residual modules, and determine its hyperparameters. Step 5: Use labeled data to train the ResNet model and build the initial ResNet model; Step 6: Input the elastic parameter unlabeled data into the initial ResNet model to generate physical parameter data, forming the first part of pseudo-labeled data; Step 7: For the unlabeled data of physical parameters, elastic parameter data is predicted by rock physics model to form the second part of pseudo-labeled data, which is combined with the first part of pseudo-labeled data to form a full pseudo-labeled data set; Step 8: Fuse the labeled dataset and the fully pseudo-labeled dataset, iteratively train the initial ResNet model, and obtain the final ResNet model; Step 9: Input pre-stack seismic inversion or logging elastic parameter observation data into the final ResNet model to predict reservoir physical properties.

2. The rock physics and deep learning dual-driven reservoir property parameter intelligent prediction method according to claim 1 is characterized in that: Step 1 includes the following sub-steps: Step 11: Calculate the bulk modulus K of the rock matrix using the VRH average model m and shear modulus μ m , and its calculation formula is: ； ； Where: K mi 、μ mi 、v mi are the bulk modulus, shear modulus and percentage content of the i-th mineral in the rock matrix, respectively, and n is the total number of mineral types; Step 12: Calculate the bulk modulus K of the pore fluid using the Voigt model fl , and its calculation formula is: ； Where: S w is water saturation; K w , K g are the bulk moduli of the liquid and gas in the pores, respectively; Step 13: Calculate the bulk modulus K of the dry rock skeleton using the Kuster-Toksoz model dry and shear modulus μ dry , and its calculation formula is ； ； Where: K * and μ * are the bulk modulus and shear modulus of the containing material, respectively; v is the pore volume fraction, α is the pore aspect ratio, T ijij and T iijj are the pore structure factors; Step 14: Calculate the bulk modulus K of fluid-saturated rock using the Biot-Gassmann equation sat and shear modulus μ sat , and its calculation formula is: ； ； Where: is the porosity; Step 15: Based on K sat 、μ sat and density ρ to calculate the longitudinal wave velocity V p , shear wave velocity V s , and its calculation formula is: ； ； ； Where: ρ fl and ρ m is the density of the fluid and rock matrix.

3. The rock physics and deep learning dual-driven reservoir property parameter intelligent prediction method according to claim 2 is characterized in that: The probability distribution function of the Gaussian mixture model in step 2 is: ； Where: m is the physical parameter or elastic parameter; N c is the number of Gaussian components; ω k is the weight; is the probability density function of the kth Gaussian component, μ k and are the mean and covariance matrices.

4. The rock physics and deep learning dual-driven reservoir property parameter intelligent prediction method according to claim 3 is characterized in that: The unlabeled data Y of physical property parameters generated by Monte Carlo random simulation in step 3 * and elastic parameters unlabeled data X * A probability distribution that satisfies a Gaussian mixture model.

5. The method for intelligent prediction of reservoir physical parameters driven by rock physics and deep learning according to claim 4 is characterized in that: The process of building the ResNet model in step 4 is: ； Among them, Conv is the convolution layer, which is used to extract features; ReLU is the activation function, which is used to insert nonlinearity in the calculation; Pool is the pooling layer, which is used to reduce the spatial dimension of the features and reduce the amount of calculation; ResidualBlock is a combination of residual blocks, which is used to solve the gradient disappearance problem in deep networks; FC is the fully connected layer, which is used to perform classification or regression tasks; Wherein: the structure of the residual block satisfies: ； Where: It is the output of the convolution layer and activation function in the residual block, representing the residual of network learning, x is the input, and y is the output.

6. The rock physics and deep learning dual-driven intelligent prediction method for reservoir physical parameters according to claim 5 is characterized in that: In step 5, the initial ResNet model is constructed, and its expression is: ； Where: are the parameters of the network structure; is the training process of the residual convolutional neural network model, is i labeled data.

7. The method for intelligent prediction of reservoir physical parameters driven by rock physics and deep learning according to claim 6 is characterized in that: The first part of pseudo-label data in step 6 The expression is: ； Where: is the initial ResNet model; is the j elastic parameter unlabeled data, is the prediction result of the initial ResNet model.

8. The rock physics and deep learning dual-driven intelligent prediction method for reservoir physical property parameters according to claim 7 is characterized in that: The second part of pseudo-label data in step 7 The expression is: ； Where: RPM is the rock physics model; is the unlabeled data of k physical property parameters; are rock physics model parameters; predicting results for rock physics models; The expression of the full pseudo-label dataset is: ； Where: is the first part of pseudo-label data; is the second part of pseudo-label data; the addition operation represents concatenation.

9. The method for intelligent prediction of reservoir physical parameters driven by rock physics and deep learning according to claim 8, characterized in that: The expression for the fused dataset in step 8 is: ； Where: θ * is the parameter of the final network structure; ResNet train is the training process of the residual convolutional neural network model, and (X, Y) is a labeled dataset.

10. The rock physics and deep learning dual-driven intelligent prediction method for reservoir physical property parameters according to claim 9 is characterized in that: The expression of the reservoir physical property parameters predicted in step 9 is: ； Where: is the final residual convolutional neural network model; x is the elastic parameter observation data including P-wave velocity, S-wave velocity, and density; y is the predicted reservoir physical property parameters including porosity, water saturation, and mud content.

Images