Carbonate rock pore type prediction method and device based on deep learning

Abstract

The invention discloses a carbonate rock pore type prediction method and device based on deep learning, and the method comprises the steps: obtaining training well data, building a sample data set, and constructing a CNN network model; inputting the training sample data set into the CNN network model to obtain a logging interpretation result, and dynamically adjusting parameters of the CNN network model, so that the calculation efficiency and prediction precision of the network model are higher; and obtaining the velocity and density of longitudinal and transverse waves of a test well, inputting the velocity and density into the trained CNN network model for inversion, and obtaining a prediction result of the pore type of the carbonate reservoir of the test well. And in combination with the trained CNN network model and the elastic parameters of the pre-stack seismic inversion, obtaining the spatial distribution characteristics of the multiple pore types. According to the method, data testing and learning network structure optimization are carried out on the basis of logging data in combination with a CNN (convolutional neural network) deep learning network, the learning network structure is applied to elastic parameters of pre-stack seismic inversion, and high-precision and stable quantitative characterization of spatial distribution of multiple pore types of the carbonate reservoir is realized.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration technology, specifically to a method and apparatus for predicting the pore type of carbonate rocks based on deep learning. Background Technology

[0002] Carbonate rocks contain approximately 60% of the world's remaining oil and gas resources, making their development a crucial area for increasing oil and gas reserves and production. Seismic prediction technology, as an important tool in reservoir exploration and development, has always been a research hotspot both domestically and internationally. However, the heterogeneity of carbonate reservoirs, often accompanied by diverse fracture and pore development, introduces uncertainty into their prediction. Pore structure is the primary cause of the heterogeneity and complex reservoir types in carbonate rocks; therefore, only by accurately characterizing the pore structure of carbonate rocks can reservoir parameters be accurately calculated.

[0003] Existing methods for evaluating pore structure mainly include experimental analysis, nuclear magnetic resonance (NMR) evaluation, imaging logging porosity spectrum analysis, and rock physics modeling. However, existing pore type methods primarily focus on the quantitative prediction of pore types under dual-pore combinations, and the limited input data makes it difficult to fully explore pore structure characteristics. Therefore, there is an urgent need to develop a deep learning-based method for quantitative prediction of pore types in carbonate reservoirs, achieving high-precision and stable quantitative characterization of carbonate rock physical parameters. Deep learning is an efficient feature extraction and information fusion method that can avoid the stringent assumptions involved in model-driven methods and directly mine information from the data. Currently, many deep learning algorithms have been introduced into the field of applied geophysics and have achieved breakthroughs in many aspects.

[0004] To address the shortcomings of existing pore type prediction methods, this invention quantitatively predicts the volume ratios of pores such as dissolution pores, matrix pores, and microfractures based on well logging data and establishes a sample dataset. It then combines a CNN (Convolutional Neural Network) deep learning network to test the model with actual measurement data, optimizes the learning network structure to improve the generalization performance and prediction accuracy of the deep learning algorithm, and achieves high-precision and stable quantitative characterization of multiple pore types in carbonate reservoirs. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for predicting the pore type of carbonate rocks based on deep learning, so as to achieve high-precision and stable quantitative characterization of the physical parameters of carbonate rocks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] According to a first aspect of the present invention, a method for predicting the pore type of carbonate rocks based on deep learning is provided, comprising the following parts:

[0008] Construct a CNN (Convolutional Neural Network) model;

[0009] Acquire density, acoustic wave, and resistivity data of training wells; calculate total porosity using the density of training wells, calculate matrix porosity using the acoustic wave of training wells, calculate microfracture porosity using the resistivity data of training wells, calculate cavern porosity using total porosity, matrix porosity, and microfracture porosity, and construct a training sample dataset of porosity and elastic parameters for multiple pore types using the total porosity, matrix porosity, microfracture porosity, and cavern porosity of training wells;

[0010] The training sample dataset is input into the CNN (Convolutional Neural Network) network model to obtain the well logging interpretation results. The parameters of the CNN (Convolutional Neural Network) network model are dynamically adjusted to make the input and output results match, thus obtaining the trained CNN (Convolutional Neural Network) network model.

[0011] The P-wave and S-wave velocities and densities of the test well are obtained and input into a trained CNN (convolutional neural network) model for inversion to obtain the test well porosity type prediction results.

[0012] In one possible implementation of the first aspect, the total porosity, matrix porosity, and microfracture porosity of the training well are obtained using the following steps:

[0013] S1, estimating total porosity φ using density logging data. D :

[0014]

[0015] In the formula: ρ ma ρ f and ρ b These represent the density values ​​of the rock skeleton, pore fluid, and actual well logging, respectively.

[0016] S2, estimating matrix porosity φ using acoustic logging data. S :

[0017]

[0018] In the formula: DT ma DT f DT and P-wave transit times for rock skeleton, pore fluid, and sonic logging, respectively;

[0019] S3, using resistivity obtained from shallow and deep lateral logging to calculate microfracture porosity φ f Alternatively, resistivity can be obtained using deep induction-microsphere focusing logging to calculate microfracture porosity φ. f :

[0020]

[0021] Where m is the porosity index, R D and R S Resistivity for light and dark water, R mf R is the resistivity of the mud filtrate. w The resistivity of formation water;

[0022] S4. Calculate the cavity porosity using the total porosity, matrix porosity, and microfracture porosity obtained above:

[0023] φ V =φ D -φ S -φ f

[0024] Where, φ D For total porosity, φ S For matrix porosity, φ f For microfracture porosity, φ V This refers to the porosity of the karst cave.

[0025] In one possible implementation of the first aspect, when training the CNN (Convolutional Neural Network) network model, a sample dataset is constructed using the P-wave and S-wave velocity and density logging data from multiple training wells and the porosity of different pore types interpreted by the logging. The CNN (Convolutional Neural Network) algorithm is then used to train the target layer data from multiple training wells in the research area, and the deep learning network structure parameters are dynamically adjusted to improve the generalization performance of the deep learning algorithm.

[0026] In one possible implementation of the first aspect, the research area has 6 training wells.

[0027] In one possible implementation of the first aspect, the trained CNN (convolutional neural network) algorithm is combined with the P-wave and S-wave velocities and densities of the test well to carry out quantitative prediction of the porosity of matrix pores, caverns and microfractures, and the results are compared and analyzed with the well logging interpretation results to achieve quantitative characterization of the pore type of the test well; the trained and tested CNN algorithm is combined with the elastic parameters obtained from pre-stack seismic inversion to achieve seismic quantitative characterization of the spatial distribution characteristics of multiple pore types in carbonate reservoirs.

[0028] According to a second aspect of the present invention, a device for predicting the pore type of carbonate rocks is provided, comprising:

[0029] Module 1: Used to build a CNN (Convolutional Neural Network) model; specifically, it uses 3 convolutional layers, 1 pooling layer, and 1 fully connected layer. The convolutional layers have 128 kernels, a stride of 1, and padding using the 'same' method. The kernel sizes of each convolutional layer are 3, 1, and 2 respectively.

[0030] The sample data module is used to acquire density, acoustic, and resistivity data from training wells. It then uses this data to calculate the total porosity, matrix porosity, microfracture porosity, and cavern porosity of the training wells, constructing a sample dataset. Specifically, this module receives density logging data to estimate total porosity, uses acoustic logging data to estimate matrix porosity, and uses resistivity to calculate microfracture porosity. Based on the matrix porosity, fracture porosity, total porosity, and cavern porosity interpreted from the acoustic, resistivity, and density logging data, it constructs the sample dataset required for training the deep learning network.

[0031] Training Module: This module trains and tests the CNN (Convolutional Neural Network) model using a sample dataset. By optimizing the network structure parameters, it improves the efficiency and accuracy of the CNN model in predicting multiple pore types. Specifically, this module uses the sample dataset and the CNN algorithm to train the target layer data of the research area; it dynamically adjusts the deep learning network structure parameters to improve the algorithm's generalization performance.

[0032] The prediction module is used to predict the pore type of test wells, obtain the P-wave and S-wave velocities and densities of the test wells, and input them into a trained CNN (Convolutional Neural Network) model for inversion, thereby achieving a quantitative characterization of the pore type of the test wells. Specifically, this module inputs the P-wave and S-wave velocity and density data of the test wells into a trained CNN model for inversion, performs quantitative prediction of the porosity of matrix pores, caverns, and microfractures, and compares and analyzes the results with the well logging interpretation results to achieve a quantitative characterization of the pore type of the test wells; it also combines the trained and tested CNN algorithm with the elastic parameters obtained from pre-stack seismic inversion to achieve a seismic quantitative characterization of the spatial distribution characteristics of multiple pore types in carbonate reservoirs.

[0033] According to a third aspect of the present invention, a processing apparatus is provided, the processing apparatus comprising at least a processor and a memory, the memory storing a computer program, wherein the processor executes the computer program to implement a deep learning-based method for predicting the pore type of carbonate rocks.

[0034] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, characterized in that it stores computer-readable instructions thereon, which can be executed by a processor to implement a deep learning-based method for predicting the pore type of carbonate rocks.

[0035] Compared with the prior art, the present invention has the following beneficial technical effects:

[0036] This invention provides a deep learning-based method for predicting pore types in carbonate reservoirs, offering a data-driven quantitative characterization method for multiple pore types in carbonate reservoirs. A sample dataset is constructed using elastic data from training wells and multiporosity interpreted from well logging. A deep learning algorithm is used to test the model against experimental data and dynamically adjust the structural parameters of the deep learning network. By combining this network with the elastic parameters of the test wells, accurate characterization of multiple pore types in carbonate reservoirs is achieved. Compared to traditional model-driven pore type inversion methods, the data-driven pore type prediction method can fully extract pore structure information from the data, overcoming challenges such as the dual-pore system assumption and limited input data. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of a deep learning-based method for predicting the pore type of carbonate rocks in an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the CNN network model structure in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the multi-porosity logging evaluation results of wells A, B, and C in this embodiment of the invention, where (a) is well A; (b) is well B; and (c) is well C; PORT: total porosity; PORV: cavern porosity; PORM: matrix porosity; and PORF: microfracture porosity.

[0040] Figure 4 This is a schematic diagram comparing the porosity of total porosity, matrix porosity, and microfractures based on CNN prediction with the well logging interpretation results in an embodiment of the present invention, wherein (a) is well A (trained); (b) is well B (trained); and (c) is well C (tested).

[0041] Figure 5 This is a schematic diagram of the P-wave velocity, S-wave velocity, and density obtained based on pre-stack AVO inversion in an embodiment of the present invention;

[0042] Figure 6 This is a schematic diagram illustrating the prediction of matrix porosity, dissolution pores, and microfractures from seismic inversion elastic parameters based on a CNN network in an embodiment of the present invention. Detailed Implementation

[0043] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0044] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0045] Example 1:

[0046] like Figure 1 As shown in the figure, this embodiment provides a method for predicting the pore type of carbonate rocks based on deep learning. The specific implementation method is as follows:

[0047] Step 1: Convolutional Neural Network (CNN) Learning Network Structure Construction. Specifically, a CNN mainly includes convolutional layers, pooling layers, and fully connected layers. The convolutional layer is the core of the CNN, responsible for extracting features from the input data. Convolutional layers are implemented by sliding convolutional kernels (or filters) across the input data and performing dot product operations. In this embodiment, the number of convolutional kernels is 128. Specifically, each layer will have 128 different convolutional kernels, each capturing different features in the input data. The stride is 1, meaning the convolutional kernel moves 1 pixel at a time as it slides across the input data. The padding method is 'same', meaning that zeros are added to the edges of the input data during convolution to maintain the same dimension as the input. The sizes of the convolutional kernels in each layer are 3, 1, and 2, representing the sizes of the convolutional kernels used in different convolutional layers. The size of the convolutional kernel affects the range of features it can capture.

[0048] The pooling layer, located after the convolutional layer, is used to reduce the dimensionality of the data while preserving important features. This helps reduce computation and avoid overfitting. In this embodiment, the pooling layer reduces the output of the convolutional layer, removing redundant features from the feature map while maintaining feature invariance.

[0049] The fully connected layer is responsible for combining and processing the features extracted from the previous layers to perform the final classification or regression task. It ensures that each neuron is connected to the neurons in the previous layer, performs weighted summation of the input features, and may apply activation functions to introduce nonlinearity.

[0050] The CNN (Convolutional Neural Network) model structure used in this embodiment includes three convolutional layers for feature extraction, one pooling layer for feature reduction and overfitting prevention, and one fully connected layer for final decision-making and output. By carefully designing the parameters and connections of these layers, an effective CNN model can be constructed to handle specific tasks.

[0051] Step 2: Evaluation of training well multi-porosity logging. Measure the density, sonic logging, and resistivity data of the training well and calculate the total porosity, matrix porosity, microfracture porosity, and cavern porosity of the training well.

[0052] Specifically, it includes the following process:

[0053] Estimating total porosity φ using density logging data D :

[0054]

[0055] In the formula: ρ ma ρ f and ρ b These represent the density values ​​of the rock skeleton, pore fluid, and actual well logging, respectively.

[0056] Estimating matrix porosity φ using acoustic logging data S :

[0057]

[0058] In the formula: DT ma DT f DT and P-wave transit times for rock skeleton, pore fluid, and sonic logging, respectively;

[0059] Calculating microfracture porosity φ using resistivity obtained from shallow and deep lateral logging. f Alternatively, resistivity can be obtained using deep induction-microsphere focusing logging to calculate microfracture porosity φ. f :

[0060]

[0061] Where m is the porosity index, R D and R S Resistivity for light and dark water, R mf R is the resistivity of the mud filtrate. w The resistivity of formation water;

[0062] Calculate the porosity of the karst cave using the total porosity, matrix porosity, and microfracture porosity obtained from the above calculations:

[0063] φ V =φ D -φ S -φ f

[0064] Where, φ D For total porosity, φ S For matrix porosity, φ f For microfracture porosity, φ V This refers to the porosity of the karst cave.

[0065] Specifically, such as Figure 3 The diagram shows the multi-porosity logging evaluation results of wells A, B and C in this embodiment. Figure (a) shows the logging evaluation results of well A, Figure (b) shows the logging evaluation results of well B, and Figure (c) shows the logging evaluation results of well C. PORT is the total porosity, PORV is the cavern porosity, PORM is the matrix porosity, and PORF is the microfracture porosity. Figure 3 This study demonstrates the calculation of total porosity, matrix porosity, and microfracture porosity in wells A, B, and C using acoustic, resistivity, and density logging methods, respectively. Training sample datasets were established using P-wave and S-wave velocity and porosity data from the entire well section in wells A and B, while well C served as the test sample dataset. The logging evaluation results show that the pore system of the target layer is characterized by well-developed matrix porosity, followed by cavernous porosity, with a small number of microfractures also present.

[0066] Step 3: Training and testing the CNN (Convolutional Neural Network) model based on well logging data. Specifically, the deep learning network is trained using matrix porosity, fracture porosity, and total porosity interpreted from acoustic, resistivity, and density well logging data, respectively. A sample dataset is constructed using P-wave and S-wave velocity and density well logging data, along with porosity interpretations for different pore types. The CNN algorithm is then used to train the model on target layer data from six training wells in the study area, dynamically adjusting the model as needed. Figure 1 The deep learning network structure parameters shown are used to improve the generalization performance of deep learning algorithms.

[0067] like Figure 4 As shown, Figure 4 (a) Demonstrates the use of a pre-established training sample dataset to train a CNN (Convolutional Neural Network) model, dynamically adjusting the network structure and model parameters to achieve a good match between input and output. The trained CNN model is then used for pore type evaluation in Well C. Comparison reveals that the total porosity and matrix porosity trained on the well logging data agree well with the well logging interpretation results. However, the microfracture porosity predicted based on the test data is slightly higher than the well logging interpretation results. Overall, however, the test results and well logging interpretation results maintain good consistency.

[0068] Step 4: Pre-stack seismic elastic parameter inversion. Specifically, firstly, a quantitative evaluation of multiple pore types is performed. An optimized CNN (convolutional neural network) is combined with the P-wave and S-wave velocities and densities from a test well to quantitatively predict the porosity of matrix pores, caverns, and microfractures. This prediction is then compared with the results of well logging interpretation to achieve a quantitative characterization of pore types. Next, a seismic survey line passing through three wells in the study area is selected, and pre-stack seismic AVO inversion is performed to obtain P-wave velocity, S-wave velocity, and density, such as... Figure 5 As shown in the figure, the structural high locations of wells A and B exhibit lower P-wave velocities and densities, which aligns with the production report indicating that wells A and B are gas-producing wells. This also demonstrates that AVO inversion results can effectively characterize the distribution of subsurface elastic parameters, providing reliable data support for the quantitative prediction of pore types in carbonate reservoirs.

[0069] Step 5: Predicting pore type earthquakes based on CNN (Convolutional Neural Network) network models. Figure 6 The porosity of total porosity, matrix porosity, and microfractures is predicted by combining the trained CNN (Convolutional Neural Network) model with the elastic parameters obtained from pre-stack inversion. As shown in the figure, the predicted total porosity and matrix porosity are relatively high in the structurally high regions of wells A and B, while microfractures are relatively well-developed in the structurally low regions. Considering that wells A and B are gas-producing wells, it can be concluded that matrix pores and cavern pores are the main pore spaces for hydrocarbon enrichment in the target layer, while microfractures provide good channels for hydrocarbon migration.

[0070] Example 2:

[0071] In one embodiment of the present invention, a device for predicting the pore type of carbonate rocks is provided, comprising:

[0072] Module 1: Used to build a CNN (Convolutional Neural Network) model; specifically, it uses 3 convolutional layers, 1 pooling layer, and 1 fully connected layer. The convolutional layers have 128 kernels, a stride of 1, and padding using the 'same' method. The kernel sizes of each convolutional layer are 3, 1, and 2 respectively.

[0073] The sample data module is used to acquire density, acoustic, and resistivity data from training wells. It then uses this data to calculate the total porosity, matrix porosity, microfracture porosity, and cavern porosity of the training wells, constructing a sample dataset. Specifically, this module receives density logging data to estimate total porosity, uses acoustic logging data to estimate matrix porosity, and uses resistivity to calculate microfracture porosity. Based on the matrix porosity, fracture porosity, total porosity, and cavern porosity interpreted from the acoustic, resistivity, and density logging data, it constructs the sample dataset required for training the deep learning network.

[0074] Training Module: This module trains and tests the CNN (Convolutional Neural Network) model using a sample dataset. By optimizing the network structure parameters, it improves the efficiency and accuracy of the CNN model in predicting multiple pore types. Specifically, this module uses the sample dataset and the CNN algorithm to train the target layer data of the research area; it dynamically adjusts the deep learning network structure parameters to improve the algorithm's generalization performance.

[0075] The prediction module is used to predict the pore type of the test well, obtain the P-wave and S-wave velocities and densities of the test well, and input them into a trained CNN (Convolutional Neural Network) model for inversion, thereby achieving a quantitative characterization of the pore type of the test well. Specifically, this module inputs the P-wave and S-wave velocity and density data of the test well into a trained CNN model for inversion, performs quantitative prediction of the porosity of matrix pores, caverns, and microfractures, and compares the results with the well logging interpretation results to achieve a quantitative characterization of the pore type of the test well.

[0076] The specific implementation methods of the above modules can be found in the relevant content disclosed in the foregoing embodiments, and will not be repeated here.

[0077] Example 3:

[0078] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to implement the operation of a method for diagnosing faults in intermediate bearings of aero-engines.

[0079] Example 4:

[0080] In one embodiment of the present invention, a deep learning-based method for predicting the pore type of carbonate rocks, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data.

[0081] The computer storage medium can be any available medium or data storage device that a computer can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs)), optical storage (e.g., CDs, DVDs, BDs, HVDs), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0082] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0083] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] 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.

[0085] 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.

Claims

1. A method for predicting carbonate pore types based on deep learning, characterized in that, The method comprises the following steps: constructing a CNN network model; obtaining density, acoustic wave and resistivity data of training wells; calculating total porosity by using the density of the training wells, calculating matrix porosity by using the acoustic wave of the training wells, calculating micro-fracture porosity by using the resistivity data of the training wells, calculating cave porosity by using the total porosity, the matrix porosity and the micro-fracture porosity, and constructing a training sample data set of multiple porosity types and elastic parameters by using the total porosity, the matrix porosity, the micro-fracture porosity and the cave porosity of the training wells; inputting the training sample data set into the CNN network model to obtain a well logging interpretation result, dynamically adjusting the parameters of the CNN network model so that the input and the output results are consistent, and obtaining a trained CNN network model; obtaining the P-wave and S-wave velocities and the density of a test well, inputting the P-wave and S-wave velocities and the density into the trained CNN network model for inversion, and obtaining a porosity type prediction result of the test well.

2. The method of claim 1, wherein the method is characterized by, The following steps are adopted to obtain the total porosity, the matrix porosity and the micro-fracture porosity of the training wells: S1, estimate total porosity φ using density logging data D : where: p ma , p f , and p b are the density values of the rock matrix, pore fluid, and actual log, respectively. S2, estimating the matrix porosity φ using sonic logging data S : where DT ma , DT f and DT are the slowness of the P-wave of the rock matrix, the pore fluid and the acoustic logging, respectively. S3, calculating microfracture porosity φ with resistivity f : where m is the porosity exponent, R D and R S are the deep and shallow resistivities, R mf is the mud filtrate resistivity, and R w is the formation water resistivity. S4, calculate the total porosity φ D , matrix porosity φ S and microfracture porosity φ f calculate the cave porosity φ V : φ V = φ D - φ s - φ f 3. The method of claim 2, wherein the method is characterized by, Using deep and shallow lateral logging to obtain resistivity to calculate microfracture porosity φ f .

4. The method of claim 2, wherein the method is characterized by, Resistivity calculated microfracture porosity φ using deep induction-microspherically focused logging f .

5. The method of claim 1, wherein the method is characterized by: When training the CNN network model, a sample data set is constructed by using the P-wave and S-wave velocities and the density logging data of multiple training wells and the porosities of different porosity types interpreted by well logging, the target layer data of the multiple training wells in the research area are trained by using the CNN algorithm, and the generalization performance of the deep learning algorithm is improved by dynamically adjusting the deep learning network structure parameters.

6. The method of claim 5, wherein the method is characterized by, The research area training wells are six.

7. The method of claim 1, wherein the method is based on deep learning. The trained CNN algorithm is combined with the P-wave and S-wave velocities and the density of a test well to quantitatively predict the porosities of matrix pores, caves and micro-fractures, and the prediction result is compared with the result of well logging interpretation to realize quantitative characterization of the porosity types of the test well; the trained and tested CNN algorithm is combined with the elastic parameters of pre-stack seismic inversion to realize seismic quantitative characterization of the spatial distribution characteristics of multiple porosity types in the carbonate reservoir.

8. A device for predicting carbonate rock pore types, characterized by, The method comprises the following steps: a construction module for constructing a CNN network model; a sample data module for obtaining density, acoustic wave and resistivity data of training wells, and calculating total porosity, matrix porosity, micro-fracture porosity and cave porosity of the training wells by using the acoustic wave, the density and the resistivity data of the training wells, and constructing a sample data set; a training module for training and testing the CNN network, optimizing the network structure parameters, and improving the efficiency and accuracy of the CNN network in predicting multiple porosity types; a prediction module for predicting the porosity types of a test well, inputting the P-wave and S-wave velocities and the density of the test well into the trained CNN network model for inversion, and realizing quantitative characterization of the porosity types of the test well.

9. A processing device comprising at least a processor and a memory having stored thereon a computer program, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable instructions stored thereon can be executed by the processor to implement the steps of the method according to any one of claims 1 to 7.

Images