Carbonate pore structure inversion method and related equipment

Abstract

The invention provides a carbonate pore structure inversion method and related equipment, and relates to the technical field of geological prospecting, and the method comprises the steps: obtaining test data, employing a trained deep learning prediction model to carry out pore structure inversion through employing a Bayesian estimation algorithm for the test data, so as to obtain a pore structure prediction result, the deep learning prediction model comprises a Bayesian convolutional neural network layer. According to the technical scheme, the test data are input into the trained deep learning prediction model, and the Bayesian estimation algorithm is adopted to perform pore structure inversion so as to obtain the pore structure prediction result including the confidence interval of the pore structure parameters. Therefore, the uncertainty quantification of the prediction result can be realized, the uncertainty is further expressed, the confidence interval is obtained, and the prediction precision can be effectively improved, namely the accuracy of the pore structure prediction result is improved. The method has great application potential in the field of data-driven pore structure prediction, and better decision support is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of geological exploration, and more specifically, to a carbonate rock pore structure inversion method, a carbonate rock pore structure inversion device, an electronic device and a storage medium. Background Art

[0002] Seismic inversion is a powerful tool for studying the internal structure of the earth and carrying out reservoir description. Pore structure is an important parameter for reservoir evaluation of carbonate reservoirs, and it is also a very important task in carbonate reservoir prediction. Generally, seismic inversion methods are divided into two categories. One is the deterministic method, including generalized linear inversion and sparse pulse inversion. The other is called the statistical inversion method, which performs statistical sampling of the posterior distribution based on the Bayesian theoretical framework. Due to the band-limited nature of seismic data, seismic inversion has non-uniqueness and instability problems. Both deterministic methods and statistical methods require prior information to regularize the inversion process, so prior information plays an important role in obtaining the optimal solution for inversion. For mathematical convenience, some simple prior assumptions are usually made on the properties of the inversion solution, such as the sparse assumption or the minimum length assumption. However, these prior assumptions may cause the inversion results to be too smooth, and the application effect in thin reservoir resolution is limited.

[0003] Deep learning has been a hot topic of research in recent years. In the fields of computer vision and pattern recognition, it is a technology that imitates the perception process of the human brain mechanism. It adopts a multi-layer structure and imitates the hierarchical transmission characteristics of the human brain's nervous system. It has received widespread attention and application in the fields of computer vision and pattern recognition, and has achieved good results.

[0004] However, the pore structure prediction performed by traditional deep neural networks can only obtain point estimates of the prediction results, so prediction bias may occur in practical applications, resulting in losses.

[0005] Therefore, a new technical solution is urgently needed to solve the above technical problems. Summary of the invention

[0006] A series of simplified concepts are introduced in the Summary of the Invention, which will be further described in detail in the Detailed Description of the Invention. The Summary of the Invention does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the scope of protection of the claimed technical solution.

[0007] In a first aspect, the present invention provides a carbonate rock pore structure inversion method, comprising:

[0008] Get test data;

[0009] For the test data, the trained deep learning prediction model is used to perform pore structure inversion using the Bayesian estimation algorithm to obtain the pore structure prediction results, wherein the deep learning prediction model includes a Bayesian convolutional neural network layer.

[0010] Optionally, the method further comprises:

[0011] Using rock physics analysis and forward modeling to build labeling technology, a label training data set is generated based on the original data, where the label training data set includes seismic data and pore structure parameters;

[0012] The deep learning prediction model is trained based on the labeled training dataset.

[0013] Optionally, generate a labeled training dataset based on the original data, including:

[0014] Simulate the depositional environment of carbonate rocks;

[0015] Obtain the lithofacies and petrophysical parameters of the target reservoir;

[0016] Based on lithofacies and rock physical parameters, seismic records are generated, wherein the seismic data, pore structure parameters and seismic records in the label training data set have a one-to-one correspondence.

[0017] Optionally, the method further comprises:

[0018] Compute the posterior distribution over the weights of a Bayesian convolutional neural network layer.

[0019] Optionally, compute the posterior distribution of the weights of a Bayesian convolutional neural network layer, including:

[0020] Using variational inference, the first distribution controlled by the first parameter is made to approach the posterior distribution to obtain the objective function;

[0021] Based on the objective function and the labeled training dataset, the loss function of the Bayesian convolutional neural network layer is obtained;

[0022] Obtain a weight sampling value to optimize the first parameter using the loss function and the weight sampling value.

[0023] Optionally, obtain the weight sampling value, including:

[0024] Sampling from each Gaussian distribution of the first parameter to obtain weight sample values;

[0025] The method also includes:

[0026] After multiple samplings, multiple pore structure prediction results are obtained;

[0027] Multiple pore structure prediction results are averaged to obtain the final prediction result.

[0028] Optionally, the deep learning prediction model also includes:

[0029] A fully connected layer, a pooling layer, a long short-term memory network, and a convolutional neural network, wherein the Bayesian convolutional neural network layer is connected to the pooling layer, and is connected to the long short-term memory network and the convolutional neural network via a fully connected layer;

[0030] The method also includes:

[0031] According to the data scale of the test data, select the corresponding sliding window size.

[0032] Secondly, a carbonate rock pore structure inversion device is also proposed, comprising:

[0033] Acquisition module, used to obtain test data;

[0034] The inversion module is used to perform pore structure inversion on the test data using a trained deep learning prediction model and a Bayesian estimation algorithm to obtain a pore structure prediction result, wherein the deep learning prediction model includes a Bayesian convolutional neural network layer.

[0035] In a third aspect, an electronic device is also proposed, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are run by the processor, they are used to execute the carbonate rock pore structure inversion method as described above.

[0036] In a fourth aspect, a storage medium is also proposed, on which program instructions are stored, and the program instructions are used to execute the carbonate rock pore structure inversion method as described above when running.

[0037] According to the above technical solution, the test data is input into the trained deep learning prediction model, and the Bayesian estimation algorithm is used to perform pore structure inversion to obtain the pore structure prediction results, including the confidence intervals of the pore structure parameters. In this way, not only can the uncertainty of the prediction results be quantified, and then the uncertainty can be expressed and the confidence interval can be obtained, but also the prediction accuracy can be effectively improved, that is, the accuracy of the pore structure prediction results can be improved. It has great application potential in the field of data-driven pore structure prediction, which facilitates better decision support.

[0038] The carbonate rock pore structure inversion method of the present invention, other advantages, objectives and features of the present invention will be partially reflected through the following description, and will also be partially understood by those skilled in the art through research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present specification. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0040] Figure 1 A schematic flow chart of a carbonate rock pore structure inversion method according to an embodiment of the present invention is shown;

[0041] Figure 2 A schematic diagram showing a probability distribution of a pore structure predicted according to an embodiment of the present invention is shown;

[0042] Figure 3 A schematic diagram showing a pore structure prediction result according to an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of the structure of a deep learning prediction model according to an embodiment of the present invention is shown;

[0044] Figure 5 A schematic diagram showing a pore structure prediction result obtained by using a deep learning prediction model including a Bayesian convolutional neural network layer or a deep learning prediction model not including a Bayesian convolutional neural network layer according to one embodiment of the present invention;

[0045] Figure 6 A schematic diagram showing a method of generating a label training data set by using rock physical analysis and forward simulation to construct a labeling technology according to an embodiment of the present invention is shown;

[0046] Figure 7 A schematic block diagram showing a carbonate rock pore structure inversion device according to an embodiment of the present invention; and

[0047] Figure 8 A schematic block diagram of an electronic device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0048] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0049] According to a first aspect of the present invention, the present invention provides a carbonate rock pore structure inversion method. Figure 1 A schematic flow chart of a carbonate rock pore structure inversion method 100 according to an embodiment of the present invention is shown. The method 100 may include the following steps.

[0050] Step S110, obtaining test data.

[0051] For example, by conducting seismic acquisition and development in carbonate rock areas, a large amount of test data can be obtained, which may include seismic data and drilling and logging information. Among them, pore structure is an important parameter for reservoir evaluation of carbonate reservoirs. Generally, pore structure parameters can be predicted on a single well.

[0052] Step S130, for the test data, using the trained deep learning prediction model and a Bayesian estimation algorithm to perform pore structure inversion to obtain a pore structure prediction result, wherein the deep learning prediction model includes a Bayesian convolutional neural network layer.

[0053] Exemplarily, the test data can be used as input and input into the trained deep learning prediction model to use the Bayesian estimation algorithm to predict the pore structure. Among them, the deep learning prediction model includes a Bayesian convolutional neural network layer. For example, the deep learning prediction model can be a traditional deep learning neural network that introduces a Bayesian convolutional neural network layer. It can be understood that after the introduction of the Bayesian convolutional neural network layer, the model parameters can be replaced by probability distributions from specific values. Therefore, by training the deep learning prediction model, the probability distribution results of the pore structure prediction can be output. Optionally, the probability distribution results obtained in the training process can be used as a prior distribution. After the test data is input into the trained deep learning prediction model, the Bayesian estimation algorithm can be used, combined with the aforementioned prior distribution, to calculate the confidence interval of the pore structure parameters, and then the posterior distribution can be obtained. Figure 2 FIG. 4 shows a schematic diagram of the probability distribution of the predicted pore structure according to an embodiment of the present invention. Figure 2 As shown in the figure, the horizontal axis represents the confidence interval, and the vertical axis represents the frequency of occurrence in the confidence interval, where "True" represents the true pore structure distribution, and "BCNN" represents the probability distribution of pore structure prediction. The probability distribution results obtained by the trained deep learning prediction model are normally distributed as a whole. Compared with the traditional neural network model, its mean is close to the point estimate result predicted by the traditional model, which can indicate that the prediction accuracy is not affected. At the same time, the function of quantifying the uncertainty of the model through probability distribution is added, that is, the confidence interval is obtained. Figure 3 FIG. 1 is a schematic diagram showing the pore structure prediction results according to an embodiment of the present invention. The horizontal axis represents time, and the vertical axis represents the pore structure parameters. The solid line corresponding to "True" represents the true pore structure value, and the dotted line corresponding to "BCNN" represents the predicted pore structure value. It is not difficult to see that the matching degree between the two is still relatively high. Figure 3 The gray shaded part is the 95% confidence interval. Combined with the above, compared with the traditional prediction model, the deep learning prediction model with the Bayesian convolutional neural network layer added to the pore structure prediction can additionally obtain the 95% confidence interval of the prediction result.

[0054] According to the above technical solution, the test data is input into the trained deep learning prediction model, and the Bayesian estimation algorithm is used to perform pore structure inversion to obtain the pore structure prediction results, including the confidence intervals of the pore structure parameters. In this way, not only can the uncertainty of the prediction results be quantified, and then the uncertainty can be expressed and the confidence interval can be obtained, but also the prediction accuracy can be effectively improved, that is, the accuracy of the pore structure prediction results can be improved. It has great application potential in the field of data-driven pore structure prediction, which facilitates better decision support.

[0055] Optionally, the deep learning prediction model may also include: a fully connected layer (Dense Layer), a pooling layer, a long short-term memory network (Long Short-Term Memory, LSTM) and a convolutional neural network (Convolutional Neural Networks, CNN), wherein the Bayesian convolutional neural network layer is connected to the pooling layer, and is connected to the long short-term memory network and the convolutional neural network via the fully connected layer. Figure 4 The schematic diagram of the structure of the deep learning prediction model according to an embodiment of the present invention is shown. The deep learning prediction model mainly includes: a composite CNN-LSTM composed of a Bayesian convolutional neural network layer, a pooling layer, and an LSTM layer. Specifically, combined with Figure 4 , where "Feature Dimension" represents the feature dimension and "TimeSequence" represents the time sequence. First, the deep space of the input sample is extracted on the time axis through three layers of 1D Bayesian convolutional neural network layers. Optionally, the input sample can be test data or training data. Each layer of Bayesian convolutional neural network is connected to the pooling layer, and connected to CNN and LSTM through a fully connected layer. The structure of LSTM is set to a two-layer network, and finally the model is output through a fully connected layer.

[0056] Combined with what has been mentioned above, in the specific application process of Bayesian convolutional neural networks, it is necessary to make major improvements to traditional deep neural networks. Since the model parameters are replaced by probability distributions instead of specific numerical values, many additional hyperparameters will be introduced into the neural network. In addition, the reasoning process and the structure of the deep neural network itself are relatively complex, which will greatly increase the amount of modeling calculations and cause huge time consumption in the process of model training and prediction. However, with the advancement of computer technology and the improvement of computing power, the use of a deep learning prediction model with the above structure can effectively overcome the problem of computing power and obtain better pore structure prediction accuracy while taking into account the measurement of prediction uncertainty.

[0057] Preferably, the method may further include: selecting a corresponding sliding window size according to the data scale of the test data. It is understandable that for different test data, the data scale may be different. Therefore, a corresponding and appropriate sliding window size can be selected for different data scales, that is, dynamic control of the sliding window can be achieved, which can ensure the validity of data during data transmission and avoid blocking. Optionally, during training or testing, the channels can be set to 24, the dropout rate can be set to 0.2, and the activation functions of all network layers can be set to ReLU.

[0058] Figure 5A schematic diagram showing a pore structure prediction result obtained by using a deep learning prediction model including a Bayesian convolutional neural network layer or a deep learning prediction model not including a Bayesian convolutional neural network layer according to an embodiment of the present invention. Figure 3 Similarly, the horizontal axis represents time and the vertical axis represents pore structure parameters. Figure 5 As shown in the figure, the dotted line is the pore structure value predicted by CNN-LSTM, and the solid line is the pore structure value predicted by CNN-LSTM with the Bayesian convolutional neural network layer. From the prediction results, the overall trend of the reservoir pore structure is consistent, with some differences in details, which is also allowed by geophysical reservoir prediction. The CNN-LSTM prediction structure based on the introduction of the Bayesian convolutional neural network layer is closer to the true value.

[0059] Optionally, the method may further include:

[0060] Step S121, using rock physics analysis and forward simulation to construct a labeling technology, generating a label training data set based on the original data, wherein the label training data set includes seismic data and pore structure parameters.

[0061] Figure 6 A schematic diagram of generating a label training data set using a labeling technique based on rock physical analysis and forward simulation according to an embodiment of the present invention is shown.

[0062] exist Figure 6 In the illustrated embodiment, step S121 may include the following steps of generating a labeled training data set based on the original data: step S121a, simulating the depositional environment of carbonate rocks; step S121b, acquiring the lithofacies and rock physical property parameters of the target reservoir; and step S121c, generating seismic records based on the lithofacies and rock physical property parameters, wherein the seismic data, pore structure parameters, and seismic records in the labeled training data set have a one-to-one correspondence.

[0063] Specifically, the pore structure obtained by well point logging and the seismic data of the well point can constitute a sample label. However, the number of drilled wells is usually limited, which greatly reduces the amount of available labeled training data. Therefore, the observed well data can be analyzed using rock physics and geological methods to artificially generate labeled training data sets for training models. In fact, from a geological point of view, the formation of underground strata is a physical, non-random process. Therefore, artificially increasing oil well data and generating synthetic reservoir models must at least consider the spatial correlation caused by the position change process. Therefore, in the label construction process, a carbonate sedimentary environment is first assumed. There are two main facies in the reservoir profile, limestone (called phase 1, Figure 6 The mud (called phase 2, Figure 6In this case, the lithofacies and rock physical properties can be obtained from the well logging curve: Porosity (Porosity, After that, 2000 200-meter-thick one-dimensional geological models are simulated using geostatistical techniques. Optionally, several lithofacies with predefined lithofacies proportions and variograms can be generated based on sequential indicator simulation. Similarly, the physical property parameters of each lithofacies are filled using a predefined variogram based on sequential Gaussian simulation, see Figure 6 b in which “Dolomite” represents marble and “Limestone” represents limestone. Both simulations assume that the variogram type is spherical. The generated physical properties are then related to the product of P-wave velocity and density using a multi-facies rock physics model to obtain the pore structure, see Figure 6 On this basis, based on the reflectivity method, given the seismic wavelet, the corresponding seismogram can be synthesized, see Figure 6 d in , where “Amplitude” represents amplitude and “Time” represents time. These synthetic seismic records have a one-to-one correspondence with pore structure parameters and seismic data.

[0064] Step S122: training the deep learning prediction model based on the labeled training data set.

[0065] After artificially synthesizing the seismic records, the corresponding pore structure parameters and seismic data are combined as a labeled training data set, which is used as the input training data of the deep learning prediction model to train the deep learning prediction model.

[0066] This can effectively increase the amount of data available for training, thereby improving training accuracy, that is, improving model prediction results. It can also enhance the robustness of the model.

[0067] Optionally, the method further includes: step S140, calculating the posterior distribution of the weights of the Bayesian convolutional neural network layer.

[0068] For example, for the training set D, assuming that the input variable is X, the output is the distribution of the predicted value Y, and W is the weight in the Bayesian convolutional neural network, the Bayesian convolutional neural network can be regarded as a model that satisfies the conditional distribution P(Y|X, W). As mentioned above, the Bayesian estimation algorithm can be used to introduce a priori distribution, and then the posterior distribution P(W|D) of W can be obtained, introducing uncertainty into the prediction of the deep learning prediction model and optimizing the pore structure prediction results.

[0069] Optionally, step S140 of calculating the posterior distribution of the weights of the Bayesian convolutional neural network layer may include the following steps.

[0070] Step S141, using variational inference, so that the first distribution controlled by the first parameter approaches the posterior distribution to obtain the objective function.

[0071] It can be understood that since the prediction result is a probability distribution, based on the weight W, the probability model for predicting Y from the input variable X becomes formula (1): P(Y|X) = E P(w|D) [P(Y|X, W)]. Using variational inference, we can use a distribution q(W|θ) controlled by a set of parameters θ to approximate the true posterior P(W|D). Thus, the problem of obtaining the posterior distribution can be transformed into an optimization problem of θ. For example, this can be achieved by minimizing the KL divergence (relative entropy) of the two distributions, which can be specifically achieved using formula (2):

[0072]

[0073]

[0074] Then we can obtain formula (3), which is the objective function:

[0075] F(D,θ)=D KL [q(W|θ)||P(W)]-E q(W|θ) [log P(D|W)]. The first term on the right is the complexity cost, which is the KL divergence between the variational inference posterior and prior, and describes the fit between the weights and the prior. The second term is the likelihood cost, whose value depends on the training data and describes the fit to the sample.

[0076] Step S142, based on the objective function and the labeled training data set, obtain the loss function of the Bayesian convolutional neural network layer.

[0077] Generally, the training set D may include input variables X and output prediction values ​​Y. Therefore, formula (4) can be obtained: q(W|θ) [log P(D|W)]=E q(W|θ) [log q(Y|W, X)]+P(X). It can be understood that for a random variable ε and probability density q(ε), as long as q(ε)dε=q(W|θ)dW is satisfied, an unbiased estimate of the expected derivative can be obtained. Since P(X) is a known distribution and can be ignored, combined with formula (4), the Monte Carlo approximation of formula (3) can be obtained as formula (5):

[0078] Among them, W i Represents the weight sampling when processing the i-th data point, and the Gaussian distribution of the i-th weight is θ i={μ i , σ i}, when m=1, W i Independent of each other, W i With θ i are independent of each other, θ i are also independent of each other. Therefore, given the training data {x j ,y j}, we can get formula (6), which is the loss function:

[0079]

[0080] Step S143, obtaining a weight sampling value to optimize the first parameter using the loss function and the weight sampling value.

[0081] For example, W can be obtained by sampling i The specific value of can be substituted into formula (6) to calculate the loss function value, and then the first parameter θ can be calculated by back propagation. i Optimize the training parameters of the entire network.

[0082] Therefore, by modeling the posterior probability of parameters, we can effectively reduce over-learning in the training process, reduce the impact of noise in the training process, and improve the generalization ability of unknown data, that is, improve the generalization ability of the model.

[0083] Optionally, step S143 of obtaining the weight sampling value may include: sampling from each Gaussian distribution of the first parameter to obtain the weight sampling value.

[0084] For example, when predicting, we can get i ={μ i ,σ i} to obtain the weighted sampling value W i .

[0085] In this embodiment, the method may further include:

[0086] Step S150, after multiple samplings, multiple pore structure prediction results are obtained.

[0087] It can be understood that the Bayesian neural network is different from the general neural network in that its weight parameters are random variables that satisfy a certain distribution, such as a Gaussian distribution, rather than a definite value. Therefore, the weight parameters can be sampled multiple times, so that according to the above, multiple prediction results can be obtained.

[0088] Step S160, averaging multiple pore structure prediction results to obtain a final prediction result.

[0089] It can be understood that the prediction result is a probability distribution, and multiple probability distribution results can be averaged to obtain the final prediction result.

[0090] Thus, accidental events are avoided and the accuracy of prediction results is ensured.

[0091] According to the second aspect of the present invention, a carbonate rock pore structure inversion device is also proposed. Figure 7 FIG. 8 is a schematic block diagram of a carbonate rock pore structure inversion device 700 according to an embodiment of the present invention. Figure 7 As shown, the apparatus 700 may include:

[0092] An acquisition module 710 is used to acquire test data;

[0093] The inversion module 720 is used to perform pore structure inversion on the test data using a trained deep learning prediction model and a Bayesian estimation algorithm to obtain a pore structure prediction result, wherein the deep learning prediction model includes a Bayesian convolutional neural network layer.

[0094] According to a third aspect of the present invention, an electronic device is also provided. Figure 8 FIG. 8 is a schematic block diagram of an electronic device 800 according to an embodiment of the present invention. Figure 8 As shown, the electronic device 800 includes a processor 810 and a memory 820, wherein the memory 820 stores computer program instructions, and the computer program instructions are used to execute the carbonate rock pore structure inversion method as described above when the processor 810 is running.

[0095] According to a fourth aspect of the present invention, a storage medium is also provided, on which program instructions are stored, and the program instructions are used to execute the carbonate rock pore structure inversion method as described above when running. The storage medium may include, for example, a storage component of a tablet computer, a hard disk of a computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disk read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0096] A person skilled in the art can understand the specific details and beneficial effects of the carbonate rock pore structure inversion device, electronic device and storage medium by reading the above description of the carbonate rock pore structure inversion method, which will not be described here for the sake of brevity.

[0097] In the several embodiments provided in the present application, it should be understood that the disclosed devices and / or equipment can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0098] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0099] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0100] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0101] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A carbonate rock pore structure inversion method, characterized in that: include: Get test data; For the test data, a Bayesian estimation algorithm is used to perform pore structure inversion using a trained deep learning prediction model to obtain a pore structure prediction result, wherein the deep learning prediction model includes a Bayesian convolutional neural network layer.

2. The carbonate rock pore structure inversion method according to claim 1, characterized in that: The method further comprises: Using rock physics analysis and forward modeling to construct labeling technology, a label training data set is generated based on the original data, wherein the label training data set includes seismic data and pore structure parameters; The deep learning prediction model is trained based on the labeled training data set.

3. The carbonate rock pore structure inversion method according to claim 2, characterized in that: The step of generating a label training data set based on the original data includes: Simulating the depositional environment of the carbonate rock; Obtain the lithofacies and petrophysical parameters of the target reservoir; Based on the lithofacies and rock physical property parameters, a seismic record is generated, wherein the seismic data, the pore structure parameters and the seismic record in the label training data set have a one-to-one correspondence.

4. The carbonate rock pore structure inversion method according to claim 2 or 3, characterized in that: The method further comprises: Compute the posterior distribution of the weights of the Bayesian convolutional neural network layer.

5. The carbonate rock pore structure inversion method according to claim 4, characterized in that: The calculating the posterior distribution of the weights of the Bayesian convolutional neural network layer comprises: Using variational inference, a first distribution controlled by a first parameter is made to approximate the posterior distribution to obtain an objective function; Based on the objective function and the labeled training data set, obtaining a loss function of the Bayesian convolutional neural network layer; Obtain a weight sampling value to optimize the first parameter using the loss function and the weight sampling value.

6. The carbonate rock pore structure inversion method according to claim 5, characterized in that: The obtaining of the weight sampling value comprises: Sampling from each Gaussian distribution of the first parameter to obtain the weight sampling value; The method further comprises: After multiple samplings, multiple pore structure prediction results are obtained; The multiple pore structure prediction results are averaged to obtain a final prediction result.

7. The carbonate rock pore structure inversion method according to claim 1, characterized in that: The deep learning prediction model also includes: A fully connected layer, a pooling layer, a long short-term memory network and a convolutional neural network, wherein the Bayesian convolutional neural network layer is connected to the pooling layer, and is connected to the long short-term memory network and the convolutional neural network via the fully connected layer; The method further comprises: According to the data scale of the test data, a corresponding sliding window size is selected.

8. A carbonate rock pore structure inversion device, characterized in that: include: Acquisition module, used to obtain test data; An inversion module is used to perform pore structure inversion on the test data using a trained deep learning prediction model and a Bayesian estimation algorithm to obtain a pore structure prediction result, wherein the deep learning prediction model includes a Bayesian convolutional neural network layer.

9. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are run by the processor, they are used to execute the carbonate rock pore structure inversion method according to any one of claims 1 to 7.

10. A storage medium having program instructions stored thereon, wherein the program instructions are used to execute the carbonate rock pore structure inversion method according to any one of claims 1 to 7 when running.

Images