Physical property parameter prediction method and device based on prior distribution characteristics
By utilizing probability density functions and Markov chain Monte Carlo stochastic simulations, combined with self-supervised and semi-supervised learning, the prediction model for physical property parameters was optimized. This solved the overfitting problem caused by insufficient samples in the prediction of physical property parameters, achieving accurate prediction of physical property parameters and promoting the development of oil and gas resource exploration.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2024-10-25
- Publication Date
- 2026-04-28
AI Technical Summary
Existing methods for predicting physical property parameters suffer from overfitting due to insufficient samples, and deep learning methods lack physical meaning, resulting in inaccurate predictions.
By utilizing probability density functions to characterize the prior distribution features of rock physical parameters, and combining Markov chain Monte Carlo stochastic simulation with self-supervised and semi-supervised learning algorithms to train the physical parameter prediction model, a bidirectional gated recurrent unit (GRU), one-dimensional convolutional layer, and deconvolutional layer are used to construct the model. Rock physical information and seismic reflection information are introduced as constraints to optimize the model weights.
This improves the accuracy of physical property parameter prediction, makes the prediction results conform to physical laws, and enhances the accuracy and efficiency of oil and gas resource exploration.
Smart Images
Figure CN121935504A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas geophysics, and in particular to a method and apparatus for predicting physical property parameters based on prior distribution characteristics. Background Technology
[0002] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section.
[0003] With the development of seismic rock physics technology, the relationship between physical property parameters and elastic parameters in existing technologies can be described using a deterministic rock physics model. Through seismic rock physics analysis and modeling, a deterministic functional expression between physical property parameters and elastic parameters can be obtained. With this deterministic functional expression, when high-precision seismic inversion results of elastic parameters are input, a suitable inversion algorithm can be used to further invert the elastic parameters into physical property parameters. This type of rock physics model-driven method has become the most popular means of predicting physical property parameters in this field. However, the correctness of establishing the forward and inversion relationship mainly depends on the accuracy of the theoretical rock physics model. In addition, since actual rock physics models are generally highly nonlinear, they are difficult to resolve analytically, resulting in extremely low inversion efficiency. In recent years, with the widespread application of deep learning technology in the field of geophysics, a class of deep learning-based physical property parameter prediction methods has received considerable attention. However, due to insufficient or incomplete training data, these methods often perform direct mapping based on deep networks without task physical meaning, leading to frequent overfitting of prediction results. Therefore, a physical property parameter prediction method is urgently needed to solve the above problems. Summary of the Invention
[0004] This invention provides a method for predicting physical property parameters based on prior distribution characteristics, to solve the overfitting problem caused by insufficient physical property parameter samples, improve the accuracy of physical property parameter prediction results, and ensure that the predicted results conform to physical laws. The method includes:
[0005] The prior distribution characteristics of the physical properties of rocks in the study area are characterized by using a pre-established probability density function.
[0006] Markov chain Monte Carlo stochastic simulations were performed on the prior distribution characteristics of the physical property parameters to obtain automatically expanded physical property parameters.
[0007] A self-supervised learning algorithm is used to train the physical property prediction model based on the pre-obtained actual physical property parameters. At the same time, a semi-supervised learning algorithm is used to train the physical property prediction model based on automatically expanded physical property parameters. During the training process, the weights of the physical property prediction model are obtained. The physical property prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers.
[0008] Based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information, the constraint loss function of the physical property parameter prediction model is determined; the physical mechanism information includes rock physics information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model.
[0009] The seismic data volume to be measured is input into the physical property parameter prediction model, and the predicted values of the physical property parameters in the seismic data volume are output.
[0010] This invention also provides a device for predicting physical property parameters based on prior distribution characteristics, to solve the overfitting problem caused by insufficient physical property parameter samples, improve the accuracy of physical property parameter prediction results, and ensure that the predicted physical property parameters conform to physical laws. The device includes:
[0011] The prior distribution characteristic characterization module of physical property parameters is used to characterize the prior distribution characteristics of physical property parameters of rocks in the study area using a pre-established probability density function.
[0012] The automatic expansion module for physical property parameters is used to perform Markov chain Monte Carlo stochastic simulation on the prior distribution characteristics of physical property parameters to obtain automatically expanded physical property parameters.
[0013] The first training module of the physical property parameter prediction model is used to train the physical property parameter prediction model using a self-supervised learning algorithm based on pre-obtained actual physical property parameters, and simultaneously uses a semi-supervised learning algorithm to train the physical property parameter prediction model based on automatically expanded physical property parameters. During the training process, the weights of the physical property parameter prediction model are obtained. The physical property parameter prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers.
[0014] The second training module of the physical property parameter prediction model is used to determine the constraint loss function of the physical property parameter prediction model based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model;
[0015] The physical property parameter prediction module is used to input the seismic data volume to be measured into the physical property parameter prediction model and output the predicted values of the physical property parameters in the seismic data volume.
[0016] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for predicting physical property parameters based on prior distribution characteristics.
[0017] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting physical property parameters based on prior distribution characteristics.
[0018] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method for predicting physical property parameters based on prior distribution characteristics.
[0019] In this embodiment of the invention, a prior distribution characteristic of the physical property parameters of rocks in the study area is characterized by using a pre-established probability density function; Markov chain Monte Carlo random simulation is performed on the prior distribution characteristic of the physical property parameters to obtain automatically expanded physical property parameters; a self-supervised learning algorithm is used to train the physical property parameter prediction model based on the pre-obtained actual physical property parameters, and a semi-supervised learning algorithm is used to train the physical property parameter prediction model based on the automatically expanded physical property parameters, obtaining the weights of the physical property parameter prediction model during the training process; the physical property parameter prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers; the constraint loss function of the physical property parameter prediction model is determined based on the weights in the physical property parameter prediction model, the prior distribution characteristic of the physical property parameters, and the pre-obtained physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model; the seismic data volume to be measured is input into the physical property parameter prediction model, and the predicted values of the physical property parameters in the seismic data volume to be measured are output. In the above process, to address the overfitting problem caused by direct mapping based on deep networks without physical meaning in existing physical property parameter prediction methods, this invention first characterizes the prior distribution characteristics of physical property parameters based on a pre-derived probability density function and automatically expands the physical property parameters to avoid overfitting due to insufficient physical property parameter samples. Second, it constructs a physical property parameter prediction model to make the extraction of physical property parameters more accurate. Finally, it uses the prior distribution characteristics of physical property parameters and pre-acquired physical mechanism information to constrain the iteration of the weights of the deep learning network for physical property parameters, so that the prediction results of physical property parameters conform to physical laws, which has a positive promoting effect and broad application prospects for future oil and gas resource exploration and discovery. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:
[0021] Figure 1 This is a flowchart of a method for predicting physical property parameters based on prior distribution characteristics in an embodiment of the present invention;
[0022] Figure 2 This is a graph representing the prior distribution characteristics of physical property parameters using a probability density function in an embodiment of the present invention.
[0023] Figure 3 This is a feature map of automatically expanded physical property parameters in an embodiment of the present invention;
[0024] Figure 4 This is a structural diagram of the physical property parameter prediction model in an embodiment of the present invention;
[0025] Figure 5 This is a schematic diagram of a physical property parameter prediction device based on prior distribution characteristics in an embodiment of the present invention. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.
[0027] Figure 1 This is a flowchart of a method for predicting physical property parameters based on prior distribution characteristics in an embodiment of the present invention. The method includes:
[0028] Step 101: Use a pre-established probability density function to characterize the prior distribution characteristics of the physical properties of rocks in the study area.
[0029] Step 102: Perform Markov chain Monte Carlo stochastic simulation on the prior distribution characteristics of the physical property parameters to obtain automatically expanded physical property parameters;
[0030] Step 103: A self-supervised learning algorithm is used to train the physical property parameter prediction model based on the pre-obtained actual physical property parameters. At the same time, a semi-supervised learning algorithm is used to train the physical property parameter prediction model based on automatically expanded physical property parameters. During the training process, the weights of the physical property parameter prediction model are obtained. The physical property parameter prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers.
[0031] Step 104: Determine the constraint loss function of the physical property parameter prediction model based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; train the physical property parameter prediction model according to the constraint loss function to obtain the trained physical property parameter prediction model.
[0032] Step 105: Input the seismic data volume to be measured into the physical property parameter prediction model, and output the predicted values of the physical property parameters in the seismic data volume to be measured.
[0033] Each step is explained in detail below.
[0034] In step 101, the prior distribution characteristics of the physical properties of rocks in the study area are characterized by a pre-established probability density function.
[0035] In one embodiment, before characterizing the prior distribution features of the physical properties of rocks in the study area using a pre-established probability density function, the process includes:
[0036] Rock physical property parameters were collected in the study area, and statistical methods were used to obtain the statistical characteristics of the physical property parameters samples.
[0037] Based on the statistical characteristics of the physical property parameter samples, a probability density function is pre-established.
[0038] In one embodiment, a pre-established probability density function is used to characterize the prior distribution features of the physical properties of rocks in the study area, including:
[0039] The prior distribution characteristics P(x) of the physical property parameters are characterized by the following probability density function relationship:
[0040]
[0041] Among them, the function γ is the integration variable, ranging from 0 to ∞; z is the parameter of the function f(z), which determines the shape and characteristics of the function. When f(z) is substituted into the prior distribution characteristics P(x), z is... or e -γ It is the exponential form of the natural exponential function, where e is the base of the natural logarithm, x represents the value of the physical property parameter, N represents the total number of lithofacies types in the study area, k represents the k-th lithofacies type in the study area, and α k μ represents the ratio of physical property parameters belonging to the k-th lithofacies to the total number of physical property parameters. k π represents the mean value of the physical properties of the k-th lithofacies, where π is the ratio of π to π, and p is the shape parameter.
[0042] Figure 2 This is a graph representing the prior distribution characteristics of physical property parameters using a probability density function in an embodiment of the present invention. Figure 2 The diagram illustrates, in one embodiment of the invention, the prior distribution characteristics of a physical property parameter, porosity, as represented by a probability density function (the histogram represents the prior distribution characteristics of the physical property parameter, and the black dashed line is the curve calculated by the probability density function). It can be seen that the pre-established probability density function accurately represents the prior distribution characteristics of the physical property parameter. When the shape parameter is small, the distribution corresponding to the pre-established probability density function is more "mild" and has a longer "tail" compared to a normal distribution. As the shape parameter becomes larger, the distribution corresponding to the pre-established probability density function becomes closer to a normal distribution.
[0043] In step 102, Markov chain Monte Carlo stochastic simulation is performed on the prior distribution characteristics of the physical property parameters to obtain automatically expanded physical property parameters.
[0044] In one embodiment, a Markov chain Monte Carlo stochastic simulation is performed on the prior distribution characteristics of the physical property parameters to obtain automatically expanded physical property parameters, including:
[0045] Based on the prior distribution characteristics of the physical property parameters, the Markov chain Monte Carlo random simulation method is used to randomly generate physical property parameters that conform to the prior distribution characteristics of the physical property parameters.
[0046] The randomly generated physical parameters were converted into angular domain pre-stack seismic data using a conventional rock physics model and the Zoeppritz equation.
[0047] The randomly generated physical property parameters and the corresponding angle-domain pre-stack seismic data are combined to form an automatically expanded physical property parameter dataset.
[0048] Figure 3 This is a feature map of automatically expanded physical property parameters in an embodiment of the present invention. In a specific embodiment, such as... Figure 3 As shown, the automatically expanded physical property parameters have different morphological characteristics, which can simulate the physical property characteristics of oil and gas reservoirs under different physical property conditions.
[0049] In step 103, a self-supervised learning algorithm is used to train the physical property parameter prediction model based on the pre-obtained actual physical property parameters. At the same time, a semi-supervised learning algorithm is used to train the physical property parameter prediction model based on automatically expanded physical property parameters. During the training process, the weights of the physical property parameter prediction model are obtained. The physical property parameter prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers.
[0050] Figure 4 This is a structural diagram of the physical property parameter prediction model in an embodiment of the present invention. In a specific embodiment, the physical property parameter prediction model consists of four sub-modules: these sub-modules are labeled as sequence modeling, local pattern analysis, upsampling, and multi-task network layer. Each of the four sub-modules plays a different role in the overall model.
[0051] The sequence modeling submodule consists of three bidirectional GRUs (gated cyclic units), which extract features from multiple sources that represent the overall geophysical property variation trends of geological sediments.
[0052] The local pattern analysis submodule consists of three parallel one-dimensional convolutional blocks with different dilation factors. It extracts local features from multiple information sources, which can represent the details of changes in geophysical properties.
[0053] The upsampling submodule consists of two deconvolution blocks with different kernel step sizes. It takes the sum of features generated by the previous submodule and performs vertical upsampling to compensate for the difference in resolution between seismic and well logging.
[0054] The multi-task network layer consists of multiple sub-task modules, which simultaneously and collaboratively predict multiple physical property parameters. Each sub-task module consists of two GRU units and two fully connected layers. They share the features extracted by the previous sub-module. The features include any one of porosity, permeability, and saturation. The amplified output of the corresponding feature domain is mapped to their respective target domain.
[0055] In step 104, the constraint loss function of the physical property parameter prediction model is determined based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model.
[0056] In one embodiment, the constraint loss function of the physical property parameter prediction model is determined based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and pre-acquired physical mechanism information, including:
[0057] Based on the least squares loss function, the prior distribution characteristics of physical property parameters and physical mechanisms are introduced as constraint terms to form a constrained loss function.
[0058] In one embodiment, the iteration is performed according to the following constraint loss function formula:
[0059]
[0060] Among them, L R To constrain the loss function, For self-monitoring item losses, For the loss of semi-supervised items, Loss due to physical mechanism constraints, The loss is the prior distribution characteristic constraint term for the physical property parameters; r i label For actual physical property parameter sample data, r i Flabel For automatically expanded sample data of physical property parameters; r i pre To use the actual physical property parameters to predict the obtained physical property parameter data, r i Fpre The physical property parameters are predicted using automatically expanded physical property parameters; Γ(r) i pre () represents the physical property data r predicted using actual physical property parameters based on conventional rock physics models and the Zoeppritz equation. i pre Converted into corresponding angle-domain pre-stack seismic data; A ilabel Pre-stack seismic data in the angular domain for actual physical property parameters; μ k pre The mean value and μ of the physical properties predicted using actual physical property parameters for the k-th lithofacies are statistically obtained. k label p represents the average value obtained statistically from actual physical property parameters of the k-th lithofacies group; k pre p k label These represent the shape parameters obtained from the statistical analysis of physical property parameters belonging to the k-th lithofacies type predicted using actual physical property parameters, and the shape parameters obtained from the statistical analysis of actual physical property parameters belonging to the k-th lithofacies type; S represents the actual number of physical property parameter samples, M represents the number of automatically expanded physical property parameter samples, and N represents the total number of lithofacies types in the study area; λ1, λ2, λ3, and λ4 represent the supervision weight of the actual physical property parameters, the pseudo-supervision weight of the automatically expanded physical property parameters, the physical mechanism constraint weight, and the prior distribution characteristic constraint weight of the physical property parameters, respectively.
[0061] In a specific embodiment, after completing network iteration based on the constraint loss function formula, the predictive performance of the physical property parameter prediction model is evaluated according to the convergence of the constraint loss function with the iteration cycle. If the constraint loss function fails to converge or the convergence error is too large, it indicates that the reliability of the physical property parameter prediction model is low, and the training parameters need to be fine-tuned and retrained. Based on the rate of decrease of the loss of each supervision and constraint term in the loss function with the iteration cycle, the weight parameters λ1, λ2, λ3, and λ4 are readjusted, following the principle that the faster the rate of decrease with the iteration cycle, the smaller the weight.
[0062] In step 105, the seismic data volume to be measured is input into the physical property parameter prediction model, and the predicted values of the physical property parameters in the seismic data volume to be measured are output.
[0063] In a specific embodiment, the angle-domain pre-stack seismic data volume is input into the physical property parameter prediction model to predict the physical property parameter data volume. The angle-domain pre-stack seismic data volume needs to have a sufficiently wide angle range, especially including more large-angle seismic information, and the data quality needs to meet the requirements of conventional pre-stack seismic inversion data. Based on the values of the final predicted physical property parameter data volume, areas whose physical property values meet the standards for commercial gas reservoirs are evaluated as favorable gas reservoir distribution areas.
[0064] This invention also provides a device for predicting physical property parameters based on prior distribution characteristics, as described in the following embodiments. Since the principle by which this device solves the problem is similar to that of the method for predicting physical property parameters based on prior distribution characteristics, the implementation of this device can refer to the implementation of the method for predicting physical property parameters based on prior distribution characteristics; repeated details will not be elaborated further.
[0065] Figure 5 This is a schematic diagram of a physical property parameter prediction device based on prior distribution characteristics in an embodiment of the present invention. The device includes:
[0066] The prior distribution characteristic characterization module 501 of physical property parameters is used to characterize the prior distribution characteristics of physical property parameters of rocks in the study area using a pre-established probability density function.
[0067] The automatic expansion module 502 for physical property parameters is used to perform Markov chain Monte Carlo stochastic simulation on the prior distribution characteristics of physical property parameters to obtain automatically expanded physical property parameters.
[0068] The first training module 503 of the deep learning network model for physical property parameters is used to train the deep learning network model for physical property parameters using a self-supervised learning algorithm based on pre-obtained actual physical property parameters, and simultaneously uses a semi-supervised learning algorithm to train the deep learning network model for physical property parameters based on automatically expanded physical property parameters. During the training process, the weights of the deep learning network model for physical property parameters are obtained. The deep learning network model for physical property parameters is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers.
[0069] The second training module 504 of the deep learning network model for physical property parameters is used to determine the constraint loss function of the deep learning network model for physical property parameters based on the weights in the deep learning network model for physical property parameters, the prior distribution characteristics of physical property parameters, and the pre-acquired physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; the deep learning network model for physical property parameters is trained according to the constraint loss function to obtain the trained deep learning network model for physical property parameters.
[0070] The physical property parameter prediction module 505 is used to input the seismic data volume to be measured into the physical property parameter deep learning network model and output the predicted values of the physical property parameters in the seismic data volume to be measured.
[0071] In one embodiment, the system further includes a probability density function establishment module, specifically used for:
[0072] Rock physical property parameters were collected in the study area, and statistical methods were used to obtain the statistical characteristics of the physical property parameters samples.
[0073] Based on the statistical characteristics of the physical property parameter samples, a probability density function is pre-established.
[0074] In one embodiment, the prior distribution feature characterization module 501 for physical property parameters is specifically used for:
[0075] The prior distribution characteristics P(x) of the physical property parameters are characterized by the following probability density function relationship:
[0076]
[0077] Among them, the function γ is the integration variable, ranging from 0 to ∞; z is the parameter of the function f(z), which determines the shape and characteristics of the function. When f(z) is substituted into the prior distribution characteristics P(x), z is... or e -γ It is the exponential form of the natural exponential function, where e is the base of the natural logarithm, x represents the value of the physical property parameter, N represents the total number of lithofacies types in the study area, k represents the k-th lithofacies type in the study area, and α k μ represents the ratio of physical property parameters belonging to the k-th lithofacies to the total number of physical property parameters. k π represents the mean value of the physical properties of the k-th lithofacies, where π is the ratio of π to π, and p is the shape parameter.
[0078] In one embodiment, the automatic expansion module 502 for physical property parameters is specifically used for:
[0079] Based on the prior distribution characteristics of the physical property parameters, the Markov chain Monte Carlo random simulation method is used to randomly generate physical property parameters that conform to the prior distribution characteristics of the physical property parameters.
[0080] The randomly generated physical parameters were converted into angular domain pre-stack seismic data using a conventional rock physics model and the Zoeppritz equation.
[0081] The randomly generated physical property parameters and the corresponding angle-domain pre-stack seismic data are combined to form an automatically expanded physical property parameter dataset.
[0082] In one embodiment, the second training module 504 of the deep learning network model for physical property parameters is specifically used for:
[0083] Based on the least squares loss function, the prior distribution characteristics of physical property parameters and physical mechanisms are introduced as constraint terms to form a constrained loss function.
[0084] In one embodiment, the second training module 504 of the deep learning network model for physical property parameters is specifically used for:
[0085] Iterate according to the following constraint loss function formula:
[0086]
[0087] Among them, L R To constrain the loss function, For self-monitoring item losses, For the loss of semi-supervised items, Loss due to physical mechanism constraints, The loss is the prior distribution characteristic constraint term for the physical property parameters; r i label For actual physical property parameter sample data, r i Flabel For automatically expanded sample data of physical property parameters; r i pre To use the actual physical property parameters to predict the obtained physical property parameter data, r i Fpre The physical property parameters are predicted using automatically expanded physical property parameters; Γ(r) i pre () represents the physical property data r predicted using actual physical property parameters based on conventional rock physics models and the Zoeppritz equation. i pre Converted into corresponding angle-domain pre-stack seismic data; A i la b el Pre-stack seismic data in the angular domain for actual physical property parameters; μ k pre The mean value and μ of the physical properties predicted using actual physical property parameters for the k-th lithofacies are statistically obtained. k label p represents the average value obtained statistically from actual physical property parameters of the k-th lithofacies group; k pre p k label These represent the shape parameters obtained from the statistical analysis of physical property parameters belonging to the k-th lithofacies type predicted using actual physical property parameters, and the shape parameters obtained from the statistical analysis of actual physical property parameters belonging to the k-th lithofacies type; S represents the actual number of physical property parameter samples, M represents the number of automatically expanded physical property parameter samples, and N represents the total number of lithofacies types in the study area; λ1, λ2, λ3, and λ4 represent the supervision weight of the actual physical property parameters, the pseudo-supervision weight of the automatically expanded physical property parameters, the physical mechanism constraint weight, and the prior distribution characteristic constraint weight of the physical property parameters, respectively.
[0088] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for predicting physical property parameters based on prior distribution characteristics.
[0089] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting physical property parameters based on prior distribution characteristics.
[0090] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method for predicting physical property parameters based on prior distribution characteristics.
[0091] In this embodiment of the invention, a prior distribution characteristic of the physical property parameters of rocks in the study area is characterized by using a pre-established probability density function; Markov chain Monte Carlo random simulation is performed on the prior distribution characteristic of the physical property parameters to obtain automatically expanded physical property parameters; a self-supervised learning algorithm is used to train the physical property parameter prediction model based on the pre-obtained actual physical property parameters, and a semi-supervised learning algorithm is used to train the physical property parameter prediction model based on the automatically expanded physical property parameters, obtaining the weights of the physical property parameter prediction model during the training process; the physical property parameter prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers; the constraint loss function of the physical property parameter prediction model is determined based on the weights in the physical property parameter prediction model, the prior distribution characteristic of the physical property parameters, and the pre-obtained physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model; the seismic data volume to be measured is input into the physical property parameter prediction model, and the predicted values of the physical property parameters in the seismic data volume to be measured are output. In the above process, to address the overfitting problem caused by direct mapping based on deep networks without physical meaning in existing physical property parameter prediction methods, this invention first characterizes the prior distribution characteristics of physical property parameters based on a pre-derived probability density function and automatically expands the physical property parameters to avoid overfitting due to insufficient physical property parameter samples. Second, it constructs a physical property parameter prediction model to make the extraction of physical property parameters more accurate. Finally, it uses the prior distribution characteristics of physical property parameters and pre-acquired physical mechanism information to constrain the iteration of the weights of the deep learning network for physical property parameters, so that the prediction results of physical property parameters conform to physical laws, which has a positive promoting effect and broad application prospects for future oil and gas resource exploration and discovery.
[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0096] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting physical property parameters based on prior distribution characteristics, characterized in that, include: The prior distribution characteristics of the physical properties of rocks in the study area are characterized by using a pre-established probability density function. Markov chain Monte Carlo stochastic simulations were performed on the prior distribution characteristics of the physical property parameters to obtain automatically expanded physical property parameters. A self-supervised learning algorithm is used to train the physical property prediction model based on the pre-obtained actual physical property parameters. At the same time, a semi-supervised learning algorithm is used to train the physical property prediction model based on automatically expanded physical property parameters. During the training process, the weights of the physical property prediction model are obtained. The physical property prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers. Based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information, the constraint loss function of the physical property parameter prediction model is determined; the physical mechanism information includes rock physics information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model. The seismic data volume to be measured is input into the physical property parameter prediction model, and the predicted values of the physical property parameters in the seismic data volume are output.
2. The method as described in claim 1, characterized in that, Before characterizing the prior distribution features of the physical properties of rocks in the study area using a pre-established probability density function, the following steps are included: Rock physical property parameters were collected in the study area, and statistical methods were used to obtain the statistical characteristics of the physical property parameters samples. Based on the statistical characteristics of the physical property parameter samples, a probability density function is pre-established.
3. The method as described in claim 2, characterized in that, Using a pre-established probability density function, the prior distribution characteristics of the physical properties of rocks in the study area are characterized, including: The prior distribution characteristics P(x) of the physical property parameters are characterized by the following probability density function relationship: Among them, the function γ is the integration variable, ranging from 0 to ∞; z is the parameter of the function f(z), which determines the shape and characteristics of the function. When f(z) is substituted into the prior distribution characteristics P(x), z is... or e -γ It is the exponential form of the natural exponential function, where e is the base of the natural logarithm, x represents the value of the physical property parameter, N represents the total number of lithofacies types in the study area, k represents the k-th lithofacies type in the study area, and α k μ represents the ratio of physical property parameters belonging to the k-th lithofacies to the total number of physical property parameters. k π represents the mean value of the physical properties of the k-th lithofacies, where π is the ratio of π to π, and p is the shape parameter.
4. The method as described in claim 1, characterized in that, Markov chain Monte Carlo stochastic simulations were performed on the prior distribution characteristics of the physical property parameters to obtain automatically expanded physical property parameters, including: Based on the prior distribution characteristics of the physical property parameters, the Markov chain Monte Carlo random simulation method is used to randomly generate physical property parameters that conform to the prior distribution characteristics of the physical property parameters. The randomly generated physical parameters were converted into angular domain pre-stack seismic data using a conventional rock physics model and the Zoeppritz equation. The randomly generated physical property parameters and the corresponding angle-domain pre-stack seismic data are combined to form an automatically expanded physical property parameter dataset.
5. The method as described in claim 1, characterized in that, Based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information, the constraint loss function of the physical property parameter prediction model is determined, including: Based on the least squares loss function, the prior distribution characteristics of physical property parameters and physical mechanisms are introduced as constraint terms to form a constrained loss function.
6. The method as described in claim 5, characterized in that, Iterate according to the following constraint loss function formula: Among them, L R To constrain the loss function, For self-monitoring item losses, For the loss of semi-supervised items, Loss due to physical mechanism constraints, The loss is the prior distribution characteristic constraint term for the physical property parameters; r i label For actual physical property parameter sample data, r i Flabel For automatically expanded sample data of physical property parameters; r i pre To use the actual physical property parameters to predict the obtained physical property parameter data, r i Fpre The physical property parameters are predicted using automatically expanded physical property parameters; Γ(r) i pre () represents the physical property data r predicted using actual physical property parameters based on conventional rock physics models and the Zoeppritz equation. i pre Converted into corresponding angle-domain pre-stack seismic data; A i la b el Pre-stack seismic data in the angular domain for actual physical property parameters; μ k pre The mean value and μ of the physical properties predicted using actual physical property parameters for the k-th lithofacies are statistically obtained. k label p represents the average value obtained statistically from actual physical property parameters of the k-th lithofacies group; k pre p k label These represent the shape parameters obtained from the statistical analysis of physical property parameters belonging to the k-th lithofacies type predicted using actual physical property parameters, and the shape parameters obtained from the statistical analysis of actual physical property parameters belonging to the k-th lithofacies type; S represents the actual number of physical property parameter samples, M represents the number of automatically expanded physical property parameter samples, and N represents the total number of lithofacies types in the study area; λ1, λ2, λ3, and λ4 represent the supervision weight of the actual physical property parameters, the pseudo-supervision weight of the automatically expanded physical property parameters, the physical mechanism constraint weight, and the prior distribution characteristic constraint weight of the physical property parameters, respectively.
7. A device for predicting physical property parameters based on prior distribution characteristics, characterized in that, include: The prior distribution characteristic characterization module of physical property parameters is used to characterize the prior distribution characteristics of physical property parameters of rocks in the study area using a pre-established probability density function. The automatic expansion module for physical property parameters is used to perform Markov chain Monte Carlo stochastic simulation on the prior distribution characteristics of physical property parameters to obtain automatically expanded physical property parameters. The first training module of the physical property parameter prediction model is used to train the physical property parameter prediction model using a self-supervised learning algorithm based on pre-obtained actual physical property parameters, and simultaneously uses a semi-supervised learning algorithm to train the physical property parameter prediction model based on automatically expanded physical property parameters. During the training process, the weights of the physical property parameter prediction model are obtained. The physical property parameter prediction model is constructed from multiple bidirectional gated recurrent units (GRUs), one-dimensional convolutional layers, deconvolutional layers, and multiple fully connected layers. The second training module of the physical property parameter prediction model is used to determine the constraint loss function of the physical property parameter prediction model based on the weights in the physical property parameter prediction model, the prior distribution characteristics of the physical property parameters, and the pre-acquired physical mechanism information; the physical mechanism information includes rock physical information and seismic reflection information; the physical property parameter prediction model is trained according to the constraint loss function to obtain the trained physical property parameter prediction model; The physical property parameter prediction module is used to input the seismic data volume to be measured into the physical property parameter prediction model and output the predicted values of the physical property parameters in the seismic data volume.
8. The apparatus as claimed in claim 7, characterized in that, It also includes a probability density function establishment module, specifically used for: Rock physical property parameters were collected in the study area, and statistical methods were used to obtain the statistical characteristics of the physical property parameters samples. Based on the statistical characteristics of the physical property parameter samples, a probability density function is pre-established.
9. The apparatus as claimed in claim 8, characterized in that, The prior distribution characteristic characterization module for physical property parameters is specifically used for: The prior distribution characteristics P(x) of the physical property parameters are characterized by the following probability density function relationship: Among them, the function γ is the integration variable, ranging from 0 to ∞; z is the parameter of the function f(z), which determines the shape and characteristics of the function. When f(z) is substituted into the prior distribution characteristics P(x), z is... or e -γ It is the exponential form of the natural exponential function, where e is the base of the natural logarithm, x represents the value of the physical property parameter, N represents the total number of lithofacies types in the study area, k represents the k-th lithofacies type in the study area, and α k μ represents the ratio of physical property parameters belonging to the k-th lithofacies to the total number of physical property parameters. k π represents the mean value of the physical properties of the k-th lithofacies, where π is the ratio of π to π, and p is the shape parameter.
10. The apparatus as claimed in claim 7, characterized in that, The automatic expansion module for physical property parameters is specifically used for: Based on the prior distribution characteristics of the physical property parameters, the Markov chain Monte Carlo random simulation method is used to randomly generate physical property parameters that conform to the prior distribution characteristics of the physical property parameters. The randomly generated physical parameters were converted into angular domain pre-stack seismic data using a conventional rock physics model and the Zoeppritz equation. The randomly generated physical property parameters and the corresponding angle-domain pre-stack seismic data are combined to form an automatically expanded physical property parameter dataset.
11. The apparatus as claimed in claim 7, characterized in that, The second training module of the physical property parameter prediction model is specifically used for: Based on the least squares loss function, the prior distribution characteristics of physical property parameters and physical mechanisms are introduced as constraint terms to form a constrained loss function.
12. The apparatus as claimed in claim 11, characterized in that, The second training module of the physical property parameter prediction model is specifically used for: Iterate according to the following constraint loss function formula: Among them, L R To constrain the loss function, For self-monitoring item losses, For the loss of semi-supervised items, Loss due to physical mechanism constraints, The loss is the prior distribution characteristic constraint term for the physical property parameters; r i label For actual physical property parameter sample data, r i Flabel For automatically expanded sample data of physical property parameters; r i pre To use the actual physical property parameters to predict the obtained physical property parameter data, r i Fpre The physical property parameters are predicted using automatically expanded physical property parameters; Γ(r) i pre () represents the physical property data r predicted using actual physical property parameters based on conventional rock physics models and the Zoeppritz equation. i pre Converted into corresponding angle-domain pre-stack seismic data; A i la b el Pre-stack seismic data in the angular domain for actual physical property parameters; μ k pre The mean value and μ of the physical properties predicted using actual physical property parameters for the k-th lithofacies are statistically obtained. k label p represents the average value obtained statistically from actual physical property parameters of the k-th lithofacies group; k pre p k label These represent the shape parameters obtained from the statistical analysis of physical property parameters belonging to the k-th lithofacies type predicted using actual physical property parameters, and the shape parameters obtained from the statistical analysis of actual physical property parameters belonging to the k-th lithofacies type; S represents the actual number of physical property parameter samples, M represents the number of automatically expanded physical property parameter samples, and N represents the total number of lithofacies types in the study area; λ1, λ2, λ3, and λ4 represent the supervision weight of the actual physical property parameters, the pseudo-supervision weight of the automatically expanded physical property parameters, the physical mechanism constraint weight, and the prior distribution characteristic constraint weight of the physical property parameters, respectively.
13. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.
14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.
15. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.