A method and system for predicting gas content in deep shale

Abstract

The present invention provides a method and system for predicting gas content in deep shale, which relates to the field of survey technology, including obtaining first information, second information and third information, wherein the first information is initial formation data, the second information is experimental data, and the third information includes real-time production data and monitoring data; performing principal component analysis based on the first information to extract key influencing factors to obtain fourth information; performing physical simulation processing based on the fourth information to obtain fifth information; performing scenario simulation processing based on the fifth information to generate sixth information, wherein the sixth information is a classified training data set; performing model construction processing on the sixth information based on a preset long short-term memory network to obtain an agent model; and inputting the third information into the agent model for prediction processing to obtain a prediction result. Through multi-scenario simulation and scenario optimization, the present invention can provide more reliable prediction results in the face of complex situations such as sudden changes in formation pressure, reopening of fractures, and redistribution of pore structures.

Description

Technical Field

[0001] The present invention relates to the field of surveying technology, and in particular to a method and system for predicting gas content in deep shale. Background Art

[0002] As an important part of unconventional natural gas resources, deep shale gas reservoirs have broad development prospects. However, the geological conditions of deep shale gas reservoirs are complex. Especially under high pressure and high temperature environments, the dynamic evolution of pore structure and fracture network is extremely complex, which brings great challenges to the accurate prediction of gas content. At present, gas content prediction usually relies on static geological models and numerical simulation methods. These methods are mainly based on formation data obtained by logging, seismic exploration, core analysis and other means, and estimate gas content by constructing basic geological models.

[0003] In deep shale gas reservoirs, phenomena such as formation pressure fluctuations, pore compression, and fracture reopening have a significant impact on gas content, and existing prediction methods are difficult to capture these dynamic processes. For example, when the formation pressure changes suddenly, the fracture network may expand or close significantly, and these microstructural changes directly affect the permeability and flow path of the gas. However, static geological models cannot reflect these changes in real time, resulting in gas content prediction results that often deviate from the actual situation. In addition, when dealing with deep and complex geological conditions, existing methods have large computational complexity and long simulation time, and have low prediction accuracy when facing nonlinear changes. These shortcomings make it difficult for existing technologies to provide high-precision, real-time gas content predictions in the exploitation of deep shale gas reservoirs, increasing the uncertainty and risk of development.

[0004] Based on the defects of the existing technology, there is an urgent need for a deep shale gas content prediction method and system. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for predicting the gas content of deep shale to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present application provides a method for predicting gas content in deep shale, comprising:

[0007] Acquire first information, second information and third information, wherein the first information is initial formation data including formation pressure, temperature, porosity, permeability and rock physical properties of the area to be predicted, the second information is experimental data of pore structure changes under high pressure conditions, and the third information includes real-time production data and monitoring data;

[0008] Performing principal component analysis on the first information to extract key influencing factors, and expressing the pore structure characteristics of the deep shale gas reservoir in the form of graph data by capturing the spatial topological relationship, thereby obtaining fourth information, wherein the fourth information includes a simplified input feature set and structured graph data;

[0009] Performing physical simulation processing according to the fourth information to simulate the pore compression, closing and reopening process under high pressure conditions, generating initial simulation data reflecting the dynamic change of the pore structure, and verifying and matching the initial simulation data with the second information to obtain fifth information, wherein the fifth information is the optimized dynamic change data of the pore structure;

[0010] Performing scenario simulation processing according to the fifth information, generating sixth information by combining the chain action mechanism of different pore structure state combinations, wherein the sixth information is a classified training data set;

[0011] The sixth information is processed by model building based on a preset long short-term memory network, and the characteristics of the dynamic changes of the pore structure with pressure and time are captured in combination with the structured graph data, and the first information and the second information are used for verification and correction to obtain a trained proxy model;

[0012] The third information is input into the proxy model for prediction processing to obtain a prediction result, which includes gas content change trend data and corresponding pore structure evolution results.

[0013] In a second aspect, the present application also provides a deep shale gas content prediction system, including:

[0014] A first acquisition unit is used to acquire first information, second information and third information, wherein the first information is initial formation data including formation pressure, temperature, porosity, permeability and rock physical properties of the area to be predicted, the second information is experimental data of pore structure changes under high pressure conditions, and the third information includes real-time production data and monitoring data;

[0015] a first extraction unit, configured to extract key influencing factors by performing principal component analysis based on the first information, and to represent the pore structure characteristics of the deep shale gas reservoir in the form of graph data by capturing the spatial topological relationship, so as to obtain fourth information, wherein the fourth information includes a simplified input feature set and structured graph data;

[0016] A first simulation unit is used to perform physical simulation processing according to the fourth information, simulate the pore compression, closing and reopening process under high pressure conditions, generate initial simulation data reflecting the dynamic change of the pore structure, and verify and match the initial simulation data with the second information to obtain fifth information, wherein the fifth information is the optimized dynamic change data of the pore structure;

[0017] A second simulation unit is used to perform scene simulation processing according to the fifth information, and generate sixth information by combining the chain action mechanism of different pore structure state combinations, wherein the sixth information is a classified training data set;

[0018] A first modeling unit performs model building processing on the sixth information based on a preset long short-term memory network, combines the structured graph data to capture the characteristics of the dynamic changes of the pore structure with pressure and time, and uses the first information and the second information for verification and correction to obtain a trained proxy model;

[0019] The first prediction unit is used to input the third information into the proxy model for prediction processing to obtain a prediction result, wherein the prediction result includes gas content change trend data and corresponding pore structure evolution results.

[0020] The beneficial effects of the present invention are:

[0021] The present invention uses principal component analysis to extract key geological influencing factors, and uses graph convolutional networks to capture the spatial topological characteristics of pore structures and fracture networks. At the same time, it combines long and short-term memory networks to model the time series of gas content changes, and constructs a high-precision hybrid proxy model. Through this model, the dynamic evolution of pore structures under deep high-pressure conditions can be analyzed in real time, and the changing trend of gas content can be accurately predicted. The present invention also provides more reliable prediction results in the face of complex situations such as sudden changes in formation pressure, reopening of fractures, and redistribution of pore structures through multi-scenario simulation and scenario optimization.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 It is a schematic diagram of the process of predicting the gas content of deep shale described in an embodiment of the present invention;

[0025] Figure 2 This is a flow chart of the k-means clustering algorithm described in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0028] Embodiment 1:

[0029] This embodiment provides a method for predicting the gas content of deep shale.

[0030] See also Figure 1 , the figure shows that the method includes step S100, step S200, step S300, step S400, step S500 and step S600.

[0031] Step S100, obtaining first information, second information and third information, wherein the first information is initial formation data including formation pressure, temperature, porosity, permeability and rock physical properties of the area to be predicted, the second information is experimental data of pore structure changes under high pressure conditions, and the third information includes real-time production data and monitoring data;

[0032] It is understandable that the first information includes the initial formation data of the area to be predicted, such as formation pressure, temperature, porosity, permeability and rock properties. These data are important bases for establishing geological models and understanding the physical properties of formations. They determine the initial state of pore structure and fracture network, and provide key input for subsequent analysis. The second information is the pore structure change data obtained under high pressure conditions in the laboratory or on-site. These data provide an accurate description of the dynamic evolution of pore structure in actual formations by simulating physical reactions in deep geological environments, such as pore compression, fracture opening and closing. The third information covers real-time production data and monitoring data, reflecting the dynamic response of formations and fluids during actual mining, such as pressure changes, gas production and flow characteristics. This step ensures the multi-dimensionality, accuracy and timeliness of the data used, laying a solid foundation for subsequent simulation and prediction.

[0033] Step S200: extract key influencing factors by performing principal component analysis based on the first information, and represent the pore structure characteristics of the deep shale gas reservoir in the form of graph data by capturing the spatial topological relationship, thereby obtaining fourth information, wherein the fourth information includes a simplified input feature set and structured graph data;

[0034] It is understandable that in this process, the initial formation data (such as pressure, porosity, permeability, etc.) must first be processed by principal component analysis, and the main factors that have the greatest impact on gas content prediction can be identified by calculating the covariance matrix and performing eigenvalue decomposition. Through this analysis step, the data dimension can be effectively reduced, redundant information can be removed, and the model can be made more efficient in processing complex geological data.

[0035] However, principal component analysis alone is not enough to fully capture the complex structural characteristics of deep shale gas reservoirs. To overcome this problem, this step further combines the graph data structure to represent the spatial topological relationship between the pore structure and the fracture network as structured graph data. In this step, spatially adjacent stratigraphic units (i.e., sample points) are connected by constructing a k-nearest neighbor graph to form an undirected weighted graph, in which the nodes represent stratigraphic samples and the edge weights reflect the geological similarities between the samples. Combined with a graph convolutional network, this graph data form can capture the correlation between different regions and the overall spatial characteristics. Through this comprehensive processing, a simplified input feature set and structured graph data are generated, providing an efficient and accurate feature input for subsequent dynamic simulation and prediction.

[0036] Step S300, performing physical simulation processing according to the fourth information, simulating the pore compression, closing and reopening process under high pressure conditions, generating initial simulation data reflecting the dynamic change of the pore structure, and verifying and matching the initial simulation data with the second information to obtain fifth information, which is the optimized pore structure dynamic change data;

[0037] Understandably, under deep high-pressure conditions, the pore structure is significantly compressed, a phenomenon that results in a decrease in permeability and affects gas mobility. Subsequently, as the production process proceeds, local pressure changes may trigger the reopening of fractures and partial recovery of pores. This series of mechanical behaviors is the main driving factor for changes in gas content in deep shale gas reservoirs.

[0038] In order to ensure the accuracy of the simulation, step S300 verifies and matches the generated initial simulation data with the second information (experimental data under high pressure conditions). By comparing and analyzing the experimental data and simulation results, possible deviations in the simulation process are discovered and adjusted so that the model can more accurately reflect the real formation behavior. In this process, a model parameter correction method based on difference analysis is used to gradually optimize the physical model by adjusting key parameters such as pore compression modulus and crack expansion coefficient. The fifth information finally obtained is the dynamic change data of the pore structure after verification and optimization, which not only reflects the real evolution of the pore and crack network under high pressure conditions, but also lays a solid foundation for the subsequent gas content prediction.

[0039] Step S400, performing scene simulation processing according to the fifth information, and generating sixth information by combining the chain action mechanism of different pore structure state combinations, where the sixth information is a classified training data set;

[0040] It can be understood that this step solves the problem that traditional methods are difficult to capture the complex nonlinear dynamic evolution in deep shale gas reservoirs through scenario simulation and in-depth analysis of chain reaction mechanisms, ensuring that the generated data set can cover all key stratigraphic change patterns, thereby greatly improving the generalization ability and accuracy of subsequent prediction models.

[0041] Step S500, constructing a model for the sixth information based on a preset long short-term memory network, combining the structured graph data to capture the characteristics of the dynamic changes of the pore structure with pressure and time, and using the first information and the second information for verification and correction to obtain a trained proxy model;

[0042] It should be noted that this step solves the problem that traditional methods cannot handle complex changes in time and space at the same time by integrating time series and spatial topological features. It also ensures the high accuracy and stability of the model through repeated verification of experimental data, enabling it to have strong predictive capabilities in practical applications.

[0043] Step S600: input the third information into the proxy model for prediction processing to obtain a prediction result, which includes gas content change trend data and corresponding pore structure evolution results.

[0044] This step combines the third information with the prediction capability of the proxy model to continuously track and predict the dynamic changes of the formation during the production process. This real-time prediction capability is of great significance for adjusting the mining strategy and optimizing the production plan. In addition, by comprehensively capturing the dynamic changes of the pore and fracture structure in the formation, the present invention can provide more accurate gas content prediction results, effectively cope with the complex geological conditions in deep shale gas reservoirs, reduce mining risks, and improve economic benefits.

[0045] Further, step S200 includes step S210 to step S240.

[0046] Step S210, performing principal component analysis processing according to the first information, calculating the covariance matrix of each feature in the initial formation data, performing eigenvalue decomposition, and selecting pressure gradient and porosity change as key influencing factors to obtain a simplified principal component feature set;

[0047] It should be noted that by calculating the covariance matrix, the correlation between the features can be quantified. Then, through eigenvalue decomposition, the eigenvalues ​​and eigenvectors that mainly affect the data changes are extracted. This step pays special attention to the changes in pressure gradient and porosity, which play a key role in the prediction of gas content in deep shale gas reservoirs. By selecting the principal component with the highest contribution to the explained variance, a simplified principal component feature set is obtained, which significantly reduces the data dimension and retains the main information that best explains the change in gas content. The covariance matrix calculation formula is:

[0048]

[0049] Where C represents the covariance matrix; n represents the number of samples; X represents the initial formation data matrix; and μ represents the characteristic mean vector.

[0050] Step S220: performing distance matrix calculation processing according to the principal component feature set, calculating the distance between each sample point in the principal component space by using the Euclidean distance formula, and combining the influence of the change of permeability and porosity in the principal component space on the distance calculation, to construct a symmetric distance matrix;

[0051] Specifically, this step calculates the weighted distance of each pair of sample points in the principal component space through the Euclidean distance formula. This distance not only takes into account the characteristics of the principal components, but also specifically combines the changes in permeability and porosity in the principal component space. By combining the changes in principal component characteristics, permeability, and porosity, a symmetric distance matrix that can reflect multiple geological characteristics is constructed. This matrix not only accurately captures the similarities between sample points, but also effectively expresses the complex spatial distribution of formation characteristics under high pressure conditions. This result provides richer and more accurate input data for subsequent graph convolution processing. The weighted Euclidean distance matrix calculation formula is:

[0052]

[0053] Where i and j represent the serial numbers of the sample points; p represents the serial number of the principal component; m represents the total number of principal components; d ij represents the weighted Euclidean distance between sample point i and sample point j in the principal component space; α and β are the weight factors of permeability and porosity respectively; z i x and z jp Respectively represent the coordinate values ​​of sample point i and sample point j on the pth principal component; Δk ij represents the permeability difference between sample points i and j; Δφ ij represents the porosity difference between sample points i and j.

[0054] Step S230, constructing a k-nearest neighbor graph according to the symmetric distance matrix, selecting k nearest neighbor sample points of each sample point by setting the neighbor threshold k, and constructing an undirected weighted graph based on the correlation between the pore structure and the pressure gradient to obtain a k-nearest neighbor graph, in which the nodes represent the sample points, and the edge weights represent the distance values ​​of the corresponding sample points;

[0055] Specifically, in this process, firstly, based on the previously calculated symmetric distance matrix, k nearest neighbor sample points are selected for each sample point by setting the neighbor threshold k. The setting of the neighbor threshold ensures that each sample point is connected to its most similar k neighbor points, thereby forming a k nearest neighbor graph. When constructing the k nearest neighbor graph, the correlation between pore structure and pressure gradient is particularly considered. Since these factors have an important influence on fluid flow and gas content changes in deep shale gas reservoirs, when selecting the nearest neighbor sample points, not only the geometric distance is considered, but also the correlation strength of these geological features is combined. This makes the constructed k nearest neighbor graph more reflective of the actual physical relationship and spatial structure of the formation. The nodes of the k nearest neighbor graph represent the sample points in the formation, and the edge weights are directly derived from the previously calculated distance values. Specifically, the edge weights between the sample points reflect their similarity in the principal component space. Combined with the pore structure and pressure gradient, these edge weights can further quantify the geological correlation between the sample points. Since the k nearest neighbor graph is an undirected weighted graph, the relationship between any two connected sample points is symmetrical, ensuring the integrity of the information in the graph.

[0056] Step S240, perform graph embedding processing according to the k-nearest neighbor graph, embed the graph data by applying a graph convolutional network, perform convolution operations to extract spatial topological features, extract the evolution of the fracture network and its impact on permeability, and generate a multi-layer embedding representation to obtain a simplified input feature set and structured graph data.

[0057] It should be noted that in the specific operation, the convolution operation of the graph convolutional network is performed between the node and its neighborhood, aggregating the information of the neighboring nodes layer by layer. This process enables the node to "sense" the characteristics of the neighboring nodes directly connected to it, and gradually expand to a wider area. This information aggregation method ensures that the model can capture complex spatial topological structures, especially the interaction between pore structure and fracture network. For deep shale gas reservoirs, the evolution of the fracture network directly affects the change of permeability, and the graph convolutional network can capture and quantify this evolution process and its impact on permeability through graph embedding processing. Through the graph convolution operation, the model extracts multiple layers of embedded representations, each layer aggregates neighbor information of different ranges, and gradually forms a global feature representation. These embedded representations integrate spatial topological features and fracture network evolution information, and finally generate a simplified input feature set and structured graph data. These data not only reduce redundant information, but also retain the spatial relationship and dynamic change characteristics that are crucial to predicting gas content.

[0058] Further, step S300 includes step S310 to step S340.

[0059] Step S310: performing finite element analysis according to the fourth information, discretizing the structured graph data, and establishing a finite element model of the porous medium to simulate the compression process of the pores under high pressure conditions, thereby obtaining initial pore compression state data;

[0060] It is understandable that the structured graph data already contains the spatial topological characteristics of the pore structure and fracture network. In order to perform finite element analysis, these graph data are further converted into discretized finite element meshes. This discretization process divides the continuous porous medium into multiple small units, each of which represents a tiny area in the formation, so that the stress and strain of each area can be calculated more accurately in the physical simulation. Next, based on the discretized data, a finite element model of the porous medium is established. This model simulates its mechanical response under high pressure conditions by treating the pore structure as a porous medium system. Specifically, the finite element analysis simulates the compression process of the pores under high pressure by solving the stress-strain relationship. This process takes into account key parameters such as formation pressure, rock properties, and porosity, and pays special attention to the deformation and volume shrinkage of the pore structure under high pressure. Each unit in the model will undergo different degrees of compression according to the applied pressure conditions, thereby generating an overall pore compression state. Through finite element analysis, the initial pore compression state data is finally obtained. These data not only reflect the overall compression trend of the pore structure under high-pressure conditions, but also include the stress distribution and deformation in each micro-region, providing accurate initial conditions for the subsequent simulation of crack opening and redistribution.

[0061] Step S320, performing contact mechanics analysis processing according to the initial pore compression state data, simulating the mechanical behavior of pore wall contact and separation, using the contact mechanics model to calculate the pore closing and reopening process under different pressures, and obtaining the pore wall mechanical state data;

[0062] Specifically, first, based on the initial pore compression state data, the pore wall areas where contact or separation may occur are identified. In these areas under high pressure, the pore walls will come into contact due to the increase in pressure, resulting in pore closure and reducing the effective permeability. Conversely, with the release of pressure or the reopening of local cracks, the pore walls may separate again, and the pores may recover part of their volume, thereby affecting the flow characteristics of the gas. In order to accurately simulate these phenomena, this step uses a contact mechanics model to calculate the contact and separation process of the pore walls. The contact mechanics model analyzes the contact force, friction force and normal force between the pore walls to determine whether the pore walls will contact and close or separate again under different pressure conditions.

[0063] During the simulation, the model dynamically adjusts the mechanical state of the pore wall and calculates the degree of closure and reopening possibility of each pore under a specific pressure. Especially for the high-pressure conditions in deep shale gas reservoirs, the slight deformation and contact behavior of the pore wall have a huge impact on the permeability of the entire formation. Therefore, contact mechanics analysis should not only consider the local pore wall behavior, but also integrate these behaviors into the changes in the overall formation structure in order to accurately predict the mechanical response of the formation.

[0064] Through this process, the mechanical state data of the pore wall are finally obtained, which describes in detail the closing and reopening of the pores under different pressure conditions, providing key input for the subsequent simulation of the evolution of the fracture network.

[0065] Step S330, performing fracture network evolution simulation processing according to the pore wall mechanical state data, by applying a tensor-based permeability model, combined with the dynamic changes of pore reopening and pressure distribution, to generate fracture network evolution data under high pressure conditions;

[0066] Specifically, first, based on the pore wall mechanical state data obtained in step S320, the area where crack expansion or reopening may occur is identified. As the formation pressure changes, the redistribution of the pore structure and the local stress concentration will directly affect the morphology of the fracture network. For example, when the pressure increases, the degree of pore closure increases, which may cause the expansion of cracks in adjacent areas; while when the pressure decreases, the reopening of the pores may cause the fracture network to recover or even expand. The evolution of this fracture network directly affects the permeability of the formation, thereby affecting the flow and distribution of gas content.

[0067] In order to simulate this complex process, this step uses a tensor-based permeability model. By expressing the permeability in tensor form, the model can more accurately describe the anisotropic characteristics of the fracture network and the impact of fracture expansion on permeability. The model integrates the dynamic effects of pore reopening, fracture expansion and closure, and pressure distribution changes to simulate the evolution of the fracture network under different pressure conditions. Especially in the high-pressure environment of deep shale gas reservoirs, the nonlinear expansion of fractures and pore redistribution have significant complexity and uncertainty. The tensor-based permeability model can capture this complexity and provide accurate fracture network evolution data.

[0068] Through this step, the final generated fracture network evolution data not only reflects the overall morphological changes of the fracture network under high pressure conditions, but also describes the permeability distribution of each region in detail.

[0069] Step S340: Verification processing is performed based on the fracture network evolution data, by fitting and matching the fracture network evolution data with the experimental data in the second information, and adjusting the physical model parameters to obtain optimized pore structure dynamic change data.

[0070] Specifically, the fracture network evolution data is first used to fit and match the experimental data in the second information. The second information contains the actual fracture network evolution results obtained under high pressure conditions in the laboratory or on-site, and these experimental data are regarded as the benchmark for model verification. Through fitting analysis, the differences between the simulation results and the experimental results can be evaluated, which may be due to insufficient model assumptions or inaccurate physical parameter settings. In the fitting process, the error between the simulation data and the experimental data is quantified using methods such as least squares or Bayesian updating. Based on the results of the error analysis, the model automatically adjusts physical parameters such as fracture expansion coefficient, pore compression modulus, permeability tensor, etc. to improve the accuracy of the model. For example, if the simulation results show that the fracture expansion is insufficient, the fracture expansion coefficient may need to be increased; if the pressure recovery effect is not significant, the pore compression modulus may need to be adjusted. This parameter adjustment process is repeated many times, so that the model gradually approaches the performance of the experimental data, and finally obtains the optimized pore structure dynamic change data. The verified and adjusted physical model can more realistically reflect the dynamic evolution process of pores and fractures in deep shale gas reservoirs, thereby improving the prediction ability and reliability of the model.

[0071] Further, step S400 includes step S410 to step S440.

[0072] Step S410, performing a combination state generation process according to the fifth information, identifying different pore structure state combinations by using a hierarchical clustering algorithm on the fifth information, and using the obtained cluster centers and cluster information as representatives of typical pore states to generate at least two pore structure combination states;

[0073] It is understandable that the application of hierarchical clustering algorithms can help analyze the diverse pore structure evolution patterns in deep shale gas reservoirs, especially when faced with complex factors such as different pressure conditions, fracture network evolution, and permeability changes.

[0074] Specifically, the fifth information is first processed using a hierarchical clustering algorithm, and the samples are gradually merged into clusters by calculating the similarities between samples (such as the degree of pore compression, the amplitude of fracture expansion, etc.). The algorithm starts with each sample as a separate cluster, and continuously merges clusters with high similarity until a cluster tree is finally formed. By analyzing the split points of the cluster tree, different combinations of pore structure states are identified, and the cluster center of each cluster is determined. These cluster centers and cluster information represent typical pore structure states in deep shale gas reservoirs, such as the extremely compressed state of pores under high pressure, the state of extensive reopening of fractures, etc. According to the clustering results, at least two different pore structure combination states are generated, which reflect the evolution characteristics of the formation under different pressure and stress conditions. These combination states provide a basis for the chain reaction analysis in subsequent scenario simulations.

[0075] Step S420, performing chain action mechanism simulation processing according to the combination state, constructing a state transition model based on a Markov chain, simulating the transition probability and path between different pore structure state combinations, and obtaining the state transition data of the pore structure;

[0076] It should be noted that Markov chain is a mathematical model used to describe the state transition of a system. It assumes that the transition of the current state depends only on the current state itself and has nothing to do with the previous state. This model is very suitable for simulating the nonlinear and random changes of pore structures under different geological conditions. Through the application of the Markov chain model, the complex transition relationship between different state combinations of pore structures is effectively captured. Compared with traditional static analysis methods, Markov chains can dynamically simulate the evolution of pore structures, especially capturing the nonlinear state transition caused by pressure changes and its chain effects.

[0077] Further, step S420 includes step S421 to step S424.

[0078] Step S421, performing state space construction processing according to the data generated by the combined state, by identifying the pore compressibility, fracture density, permeability change and gas desorption amount in each pore structure state combination, using the Gaussian mixture model to divide the state space, combining the gas desorption and pressure recovery effects caused by pore structure redistribution, to obtain a discrete state set representing different pore structure evolution stages;

[0079] Specifically, we first identify the key features of each pore structure state combination based on the data generated by the combined state, including pore compression, fracture density, permeability change, and gas desorption. These features are the main driving factors affecting the gas content of the formation and are also important parameters for describing the dynamic changes of the pore structure. For example, pore compression reflects the degree of deformation of the pores under high pressure, while fracture density directly affects the connectivity of the fracture network and the change of permeability.

[0080] In order to accurately divide the state space, the Gaussian mixture model is applied. The Gaussian mixture model is a clustering method based on probability distribution. It can identify multimodal features in the data and divide the data into several Gaussian distributions. By modeling multiple features of the pore structure state, the Gaussian mixture model can identify different evolution stages and divide these stages into discrete state sets. Each state set represents a typical pore structure evolution stage, such as the extremely compressed state of pores under high pressure, the state of fracture expansion, and the state of pressure recovery caused by gas desorption. When constructing the state space, the gas desorption and pressure recovery effects caused by the redistribution of the pore structure are specially considered. These effects are particularly significant under high pressure environments, because the redistribution of the pore structure will cause the gas to desorb from the adsorbed state and enter the fracture network, thereby causing the local pressure to rise. This complex physical process has an important impact on the overall evolution of the formation, so it is particularly concerned in the division of the state space. Finally, a discrete state set is generated, each of which represents a different pore structure evolution stage.

[0081] Step S422, performing state transition probability estimation processing according to the state set, by analyzing the transition frequency between different pore structure states, combining the Bayesian inference method to update the state transition probability, and combining the influence of pressure fluctuation and gas desorption on state transition under high pressure conditions to obtain a state transition probability matrix;

[0082] It should be noted that the transition frequency between each pair of states is first analyzed based on the discrete state set. The state transition frequency refers to the number of times the formation changes from one pore structure state to another in the simulation or actual data. These frequencies directly reflect the evolution law of the formation under specific conditions. For example, when the pressure gradually increases, the formation may more frequently transfer from the fracture expansion state to the pore closure state; conversely, when the pressure is released, it may transfer from the pore compression state to the fracture reopening state.

[0083] In order to enhance the estimation accuracy of the state transition probability, the Bayesian inference method is introduced. Bayesian inference updates the state transition probability by combining prior knowledge and observation data. Specifically, the initial state transition probability can be set as a prior distribution based on historical data or empirical estimates, and by analyzing the state transition frequency, the Bayesian inference method can gradually adjust these prior probabilities to generate a posterior distribution that is more in line with the actual situation. This method is suitable for dealing with geological systems with uncertainty and complexity, especially under high-pressure environments, where state transitions show nonlinear and non-uniform characteristics. In addition, this step specifically considers the effects of pressure fluctuations and gas desorption on state transitions under high-pressure conditions. Pressure fluctuations can cause sudden changes in pore structure, such as sudden expansion or closure of cracks, while gas desorption can cause local pressure to rise, thereby changing the state of the pore structure. These factors are reflected in the state transition probability estimation by weighting or correction to ensure that the transition probability can truly reflect these complex phenomena. The final generated state transition probability matrix describes in detail the transition possibility between each pair of states.

[0084] Step S423, performing path simulation processing according to the state transition probability matrix, simulating the transfer path of the pore structure state under different pressure conditions by constructing a Markov chain model, paying attention to the chain effect caused by pressure mutation, gas desorption and crack reopening, and obtaining a state transition path set;

[0085] Specifically, the core of step S423 is to use the state transition probability matrix to construct a Markov chain model and simulate the transfer path of the pore structure state under different pressure conditions. This step not only focuses on the transfer path of the conventional state, but also specifically analyzes the chain reaction and its evolution path of the pore structure state triggered by key events such as pressure mutation, gas desorption and crack reopening.

[0086] First, a Markov chain model is constructed based on the state transition probability matrix. The Markov chain model is suitable for describing the random evolution of the system state, especially when the future state of the system depends only on the current state. By taking each state in the state set as a node in the Markov chain and taking the values ​​in the transition probability matrix as the transition probability between nodes, a complete state transition network is established. This network can fully represent the dynamic evolution path of the pore structure under different pressure conditions. In the path simulation processing, the transfer path of the pore structure under different initial states when the pressure gradually changes, suddenly fluctuates, or the pressure rises due to gas desorption is analyzed. For example, sudden decompression of deep shale gas reservoirs under high pressure may cause the cracks to open rapidly, which in turn triggers changes in the pore structure of adjacent areas. This chain effect has a significant impact on the dynamic changes of gas content. The Markov chain model can reveal the key paths and important transfer nodes in the formation evolution by simulating the occurrence of these events and the transfer paths of the corresponding states. In addition, when processing the path simulation, the reopening effect of the fracture network under high pressure is also combined. The reopening of fractures often leads to significant changes in permeability, which not only affects the gas content distribution in the current area, but may also affect a wider area through a chain reaction. By simulating the state transition paths under different initial conditions and different pressure scenarios, the Markov chain model generates a set of state transition paths containing multiple possibilities.

[0087] Step S424: Perform chain reaction analysis based on the state transition path set, analyze the dependency relationship between the paths by applying graph theory, identify the chain reaction mechanism, and obtain state transition data.

[0088] It should be noted that this step first uses graph theory to model the state transition path set. Each path in the state transition path set can be regarded as a path in the graph, and each state in the state set corresponds to a node in the graph. In this way, the entire state transition process is abstracted into a graph structure, in which the weight of the edge is determined by the state transition probability. Graph theory provides a wealth of tools for analyzing the correlation between nodes and the dependencies between paths. For example, by calculating the overlap between paths and key nodes, it is possible to identify which state transitions have an important impact on the evolution of the overall system. Next, the chain reaction analysis focuses on identifying the complex chain effects caused by pressure fluctuations, gas desorption or crack reopening. In deep shale gas reservoirs, these events often lead to a series of interrelated changes within the formation. For example, the reopening of a fracture may cause a rapid drop in local pressure, thereby triggering the adjustment of the pore structure in the nearby area. This chain reaction may extend to the entire formation system. By analyzing the dependencies of the key paths in the graph, the trigger points and conduction paths of these chain reactions can be identified, thereby understanding how the entire system responds to changes in external conditions.

[0089] Through this graph theory analysis, not only important state transition paths are identified, but also the dependencies between the paths and their overall impact on the evolution of the formation are revealed. The final state transition data contains detailed information on these chain reaction mechanisms, providing an important reference for further prediction and optimization. These data not only describe the evolution of the formation under different scenarios, but also quantify the key mechanisms in these processes, enabling the model to better predict future state changes.

[0090] Step S430, performing scenario generation processing according to the state transition data, randomly sampling the transition processes of different states through Monte Carlo simulation, generating at least two pore structure evolution scenarios, and combining the state combination distribution under extreme pressure changes to obtain a scenario data set containing at least two evolution paths;

[0091] It can be understood that Monte Carlo simulation is a statistical method that estimates system behavior through a large number of random samples, which is suitable for dealing with geological evolution processes with uncertainty and complexity. In this step, each simulation randomly selects an initial state from the state transition data, and then simulates the transition path of the state under a series of pressure conditions based on the state transition probability matrix. Through multiple iterations, Monte Carlo simulation can generate a large number of different evolution paths, covering various possibilities from conventional pressure changes to sudden pressure fluctuations.

[0092] In particular, during the simulation, this step also combines the state combination distribution under extreme pressure changes. For example, when the pressure increases or decreases sharply, the pore structure and fracture network may undergo a sudden evolution. The state transition path under such extreme conditions usually triggers a violent chain reaction, resulting in significant changes in the formation permeability and gas content. In order to capture these key events, the sampling frequency of the paths under extreme conditions was particularly increased in the simulation, and the state combination distribution under these conditions was combined for scenario generation. Finally, through multiple Monte Carlo simulations and state sampling, multiple pore structure evolution scenarios were generated, and a scenario data set containing at least two different evolution paths was compiled, which not only showed the pore structure evolution process under normal and extreme pressure conditions, but also revealed the possible distribution and results of different evolution paths.

[0093] Step S440: Classify the scene data set, identify key pore structure evolution patterns, and mark them into different categories to generate a classified training data set.

[0094] Specifically, this step first conducts an in-depth analysis of the scenario data set to identify the key pore structure evolution patterns in each evolution path. Each evolution path represents a possible formation change process driven by complex factors such as pressure fluctuations, crack reopening, pore compression or expansion. In a large number of evolution scenarios, there may be some specific patterns, such as the fracture network expansion pattern under certain pressure conditions, or the pore redistribution pattern caused by gas desorption. These patterns are the main factors affecting formation behavior and gas content changes. In order to classify these patterns, pattern recognition and clustering algorithms are applied. Preferably, k-means clustering, such as Figure 2 As shown, Figure 2 The algorithm flow of k-means clustering is shown. First, feature vectors related to pore structure evolution are extracted from the scene data set, such as key factors such as pressure fluctuation, crack extension and pore compression. Subsequently, k cluster centers are initialized by the k-means clustering algorithm, and the initial position is randomly selected according to the distribution of feature vectors. The feature vector of each evolution path will be assigned to the nearest cluster center, and then the position of the cluster center is updated by recalculating the average value of the path in each cluster. This process is iterated until the position of the cluster center is stable, that is, there is no significant change. After stabilization, the evolutionary pattern in each cluster cluster is identified and marked, for example, paths showing similar crack extension and pore compression characteristics are classified into the same category. Finally, a classification training data set containing marked evolutionary patterns is generated to provide clear and efficient input for subsequent model training. This process not only effectively identifies key stratigraphic evolution paths, but also ensures the classification accuracy of the data set and the pertinence of model prediction. The algorithm identifies evolutionary patterns with high similarity by analyzing the feature vectors of different paths in the scene data set and classifies them into the same category. For example, if multiple paths show similar fracture extension and pore compression characteristics, these paths may be classified into the same evolutionary category. Each category represents a way that the formation evolves under specific conditions, and corresponds to different geological risks and gas content change trends. After the classification is completed, these key evolutionary patterns are marked as different categories, and finally a classified training data set is generated. This data set not only covers a variety of typical stratigraphic evolution paths, but also clearly identifies the geological characteristics and change patterns corresponding to each path. This classified data provides a clear input target for subsequent model training, enabling the model to learn and predict gas content changes under different evolutionary scenarios in a more targeted manner.

[0095] Further, step S500 includes step S510 to step S540.

[0096] Step S510, performing time series modeling processing according to the sixth information, by using a long short-term memory network to perform sequence training on the training data set, capturing the state evolution of the pore structure at different time points and its dependence on the pressure change, and obtaining a time series feature representation;

[0097] Specifically, the LSTM network is used to train and learn the time evolution patterns of pore structures in the classification training data set. In each time step, the LSTM network receives the input data at the current time point, and combines its internal state (i.e., the information at the previous time point) to gradually update its "memory unit" to capture long-term dependencies. Through multiple iterative training, the network gradually learns to identify the evolution laws of pore structures in the formation, such as the temporal change patterns of fracture expansion or pore redistribution under specific pressure conditions. At the same time, the LSTM network can also effectively identify the complex nonlinear relationships between different time points, especially the dynamic effects of pressure fluctuations on the evolution of pore structures. Through this process, time series feature representations are generated, which not only retain the temporal information of formation evolution, but also quantify the response characteristics of pore structures to pressure changes.

[0098] Step S520: Perform graph time series fusion processing according to the time series feature representation, by combining the time series features with the structured graph data, using the graph convolutional network to capture the spatial topological features and time dependency, and extracting the characteristics of the pore structure changing with pressure and time through multi-layer convolution operations, to obtain a spatiotemporal joint feature representation;

[0099] It should be noted that step S510 has generated time series feature representations, which capture the evolution patterns of pore structures at different time points and their dependence on pressure changes. However, time series analysis alone cannot fully reflect the complex spatial distribution of pore structures and their mutual influence. Therefore, this step further combines structured graph data to introduce spatial topological relationships into the model. Structured graph data originates from k-nearest neighbor graphs and graph embedding processing, representing the spatial correlation between different pore structures in the stratum. Specifically, the nodes of the graph represent the pore structure states of each stratum region, and the weights of the edges represent the spatial similarities or geological interactions between these regions. By combining time series features with structured graph data, spatiotemporal joint modeling of pore structures can be achieved.

[0100] In specific operations, graph convolutional networks are used to perform multi-layer convolution operations on graph data. Graph convolutional networks capture the spatial topological relationship between nodes layer by layer by aggregating node features within the neighborhood of the graph. Combined with time series features, graph convolutional networks can simultaneously consider dependencies in spatial and temporal dimensions. For example, at the same point in time, graph convolutional networks can identify how pressure changes in a certain area affect the pore structure of adjacent areas through the fracture network; and at different points in time, they can track how these effects evolve over time. Through multi-layer convolution operations, the characteristics of pore structure changes in time and space with pressure are gradually extracted, thus forming a highly integrated spatiotemporal joint feature representation. This spatiotemporal joint feature representation not only covers the dynamic evolution process of the stratum, but also retains the spatial distribution of the pore structure and fracture network and their interactions. This multi-dimensional feature representation provides richer and more accurate data support for subsequent model predictions.

[0101] Step S530: Perform initial model construction processing according to the spatiotemporal joint feature representation, perform back propagation training on the long short-term memory network and graph convolution network joint model, and optimize the network weights to obtain an initial training model;

[0102] Specifically, the LSTM network is responsible for processing time series features, capturing the evolution trend of pore structure at different time points and its dependence on pressure changes. On the other hand, the graph convolution network processes structured graph data and extracts the spatial correlation between different regions in the formation. The two networks complement each other through joint modeling to achieve a comprehensive understanding of formation evolution. In order to train and optimize the joint model, this step uses the back propagation algorithm. Back propagation is a widely used training method for neural networks. By calculating the error between the model output and the real data, the network weights are adjusted in reverse to minimize the error. The specific process is as follows: First, the model takes the spatiotemporal joint feature representation as input, and generates a prediction result after processing by the LSTM network and the graph convolution network. Then, the prediction result is compared with the actual gas content or pore structure state, and the error is calculated. Finally, through back propagation, the weights in the LSTM network and the graph convolution network are adjusted layer by layer to gradually optimize the model's prediction ability. This training process is iterative. By continuously updating the network weights, the model gradually learns how to more accurately capture the complex spatiotemporal dynamic changes in the formation. As the training progresses, the error gradually decreases and the prediction accuracy of the model gradually improves. Finally, after sufficient training and optimization, an initial training model is obtained, which has a strong ability to capture spatiotemporal features.

[0103] Step S540: using the initial formation data in the first information and the experimental data in the second information, the dynamic change results of the pore structure output by the initial training model are compared and verified, and the model parameters are adjusted by the gradient descent method to obtain the proxy model.

[0104] It is understandable that during the verification process, the dynamic changes in the pore structure output by the model are compared in detail with the experimental data. By analyzing the differences between the two, deviations in the model prediction can be identified. For example, if the crack expansion rate output by the model is lower than the experimental data, it indicates that the model is insufficiently responsive when dealing with pressure mutations; if the pore compression does not match the experimental data, the elastic parameters of the pore structure model are improperly set. To eliminate these deviations, the gradient descent method is used to adjust the model parameters. The gradient descent method is an optimization algorithm that gradually adjusts the parameter values ​​by calculating the gradient of the objective function (i.e., the prediction error) relative to the model parameters to minimize the error. Specifically, in each round of iteration, the error between the model output and the experimental data is calculated, and the weight parameters of the long short-term memory network and the graph convolution network in the model are reversely adjusted by the gradient descent method. As the iteration proceeds, the prediction accuracy of the model gradually improves, gradually approaching the real formation evolution behavior. Finally, through multiple rounds of iterations and error minimization processing, the optimized proxy model is obtained. This model can not only accurately reflect the complex spatiotemporal dynamic changes in deep shale gas reservoirs, but also has high robustness and generalization ability, and can adapt to the prediction of gas content under various geological conditions.

[0105] Further, step S600 includes step S610 to step S640.

[0106] Step S610: performing feature mapping processing according to the third information, converting the features in the third information into joint feature representations in a high-dimensional space by using the spatiotemporal joint feature mapping layer in the proxy model, and obtaining a mapped spatiotemporal feature vector;

[0107] It is understandable that the third information includes real-time production data and monitoring data, such as formation pressure, fluid production, temperature changes, and other key geological and engineering parameters. These data are dynamically changing and reflect the actual state of the current formation. However, the original data is multi-dimensional and complex, and directly inputting the model may lead to inefficient processing or failure to fully utilize the deep information in the data. Therefore, it is necessary to transform these data into high-dimensional feature representations suitable for model processing through feature mapping.

[0108] Step S620: Perform dynamic convolution processing according to the spatiotemporal feature vector, by applying an adaptive dynamic convolution layer in the proxy model, dynamically adjusting the parameters of the convolution kernel according to the time step, convolution processing the spatiotemporal features of different time points, and combining the evolution mode of the deep fracture network to obtain a time series feature map;

[0109] It should be noted that the geological conditions of deep shale gas reservoirs are complex and changeable, and the formation characteristics at different times may vary significantly. Therefore, it is difficult to effectively capture these changes using convolution operations with fixed parameters. To solve this problem, an adaptive dynamic convolution layer is introduced. Unlike traditional convolution, dynamic convolution can dynamically adjust the parameters of the convolution kernel according to the time step. This adaptability enables the convolution layer to flexibly adjust the focus of features according to changes in the formation state. For example, when the pressure changes drastically, the convolution kernel focuses more on capturing features related to the expansion of the fracture network; while in the stage where the pressure is relatively stable, the convolution kernel pays more attention to the subtle adjustments of the pore structure. In the dynamic convolution process, the spatiotemporal feature vector is used as input and convolution operations are performed through adaptive convolution kernels. The parameters of the convolution kernel are adjusted according to the data characteristics of the current time step to ensure that the convolution operation can accurately capture the geological changes at the current moment. This process aggregates and strengthens the feature information at different times through multi-level convolution operations to generate a time series feature map. The time series feature maps not only retain the spatiotemporal correlation information of the original data, but also contain key patterns related to the evolution of deep fracture networks.

[0110] Step S630, performing gas content prediction processing according to the time series feature graph, by inputting the time series feature graph into the length memory network layer of the proxy model, combining the gas content change trend in the historical data, predicting the gas content change in the future time period, and obtaining the gas content change trend data;

[0111] It should be noted that in the prediction process, the length memory network not only processes the time series feature graph at the current moment, but also combines the gas content change trend in the historical data. These historical data provide the behavior pattern of the formation under similar conditions, such as the fluctuation of gas content under a certain pressure condition in the past. By combining the current data with the historical trend, the change of gas content in the future time period can be predicted more accurately. For example, when historical data shows that a certain pressure pattern usually leads to a sharp drop in gas content, the length memory network layer can use this pattern to adjust the current prediction, thereby improving the prediction accuracy. The final output of the gas content change trend data describes in detail the possible change trajectory and trend of gas content in the future period, which can provide an important reference for production decision-making and resource management.

[0112] Step S640: perform pore structure evolution prediction processing based on the gas content variation trend data, predict the redistribution of the pore structure and the expansion of cracks under different pressure conditions by performing spatial topological deduction in the proxy model, and obtain the predicted pore structure evolution result.

[0113] It is understandable that this step uses the spatial topology deduction function in the proxy model to simulate and analyze the future state of the formation. During the deduction process, the model takes into account a variety of factors, including the current gas content change trend, pressure distribution, porosity and permeability. In particular, the model predicts the redistribution of pores based on pressure changes and identifies possible crack expansion or closure phenomena. Specifically, the model first evaluates how the pore structure responds under different pressure conditions. For each pressure condition at a moment, the model calculates the compression, expansion of pores and the reopening or closure of cracks. This calculation is based on previously learned formation response patterns, such as how cracks expand rapidly to release stress when pressure suddenly increases, or how pores reopen and affect gas flow paths when pressure drops. Through multiple iterations and deductions, the model generates a complete spatial topology evolution diagram, which shows the evolution path of pore structure and fracture network in future time periods. The final predicted pore structure evolution results include porosity changes, expansion and redistribution of fracture networks in various regions of the formation. These results provide a precise basis for understanding the sustainability of future gas reserves, helping to develop more effective production strategies and formation management options.

[0114] Embodiment 2:

[0115] This embodiment provides a deep shale gas content prediction system, the system comprising:

[0116] A first acquisition unit is used to acquire first information, second information and third information, wherein the first information is initial formation data including formation pressure, temperature, porosity, permeability and rock physical properties of the area to be predicted, the second information is experimental data of pore structure changes under high pressure conditions, and the third information includes real-time production data and monitoring data;

[0117] A first extraction unit is used to extract key influencing factors by performing principal component analysis based on the first information, and to express the pore structure characteristics of the deep shale gas reservoir in the form of graph data by capturing the spatial topological relationship, so as to obtain fourth information, wherein the fourth information includes a simplified input feature set and structured graph data;

[0118] A first simulation unit is used to perform physical simulation processing according to the fourth information, simulate the pore compression, closing and reopening process under high pressure conditions, generate initial simulation data reflecting the dynamic change of the pore structure, and verify and match the initial simulation data with the second information to obtain fifth information, which is the optimized pore structure dynamic change data;

[0119] A second simulation unit is used to perform scene simulation processing according to the fifth information, and generate sixth information by combining the chain action mechanism of different pore structure state combinations, where the sixth information is a classified training data set;

[0120] The first modeling unit constructs a model for the sixth information based on a preset long short-term memory network, combines the structured graph data to capture the characteristics of the dynamic changes of the pore structure with pressure and time, and uses the first information and the second information for verification and correction to obtain a trained proxy model;

[0121] The first prediction unit is used to input the third information into the proxy model for prediction processing to obtain a prediction result, wherein the prediction result includes gas content change trend data and corresponding pore structure evolution results.

[0122] In a specific embodiment disclosed in the present application, the first extraction unit includes:

[0123] The first processing subunit is used to perform principal component analysis processing according to the first information, calculate the covariance matrix of each feature in the initial formation data, perform eigenvalue decomposition, and select pressure gradient and porosity change as key influencing factors to obtain a simplified principal component feature set;

[0124] The first calculation subunit is used to perform distance matrix calculation processing according to the principal component feature set, calculate the distance between each sample point in the principal component space by using the Euclidean distance formula, and combine the influence of the change of permeability and porosity in the principal component space on the distance calculation to construct a symmetric distance matrix;

[0125] The first construction subunit is used to construct a k-nearest neighbor graph according to the symmetric distance matrix, select k nearest neighbor sample points of each sample point by setting the neighbor threshold k, and construct an undirected weighted graph based on the correlation between the pore structure and the pressure gradient to obtain a k-nearest neighbor graph, in which the nodes represent the sample points and the edge weights represent the distance values ​​of the corresponding sample points;

[0126] The first embedding subunit is used to perform graph embedding processing according to the k-nearest neighbor graph, embed the graph data by applying a graph convolutional network, perform convolution operations to extract spatial topological features, extract the evolution of the fracture network and its impact on the permeability, and generate a multi-layer embedding representation to obtain a simplified input feature set and structured graph data.

[0127] In a specific implementation disclosed in the present application, the first simulation unit includes:

[0128] The first analysis subunit is used to perform finite element analysis according to the fourth information, by discretizing the structured graph data and establishing a finite element model of the porous medium to simulate the compression process of the pores under high pressure conditions, thereby obtaining initial pore compression state data;

[0129] The second analysis subunit is used to perform contact mechanics analysis based on the initial pore compression state data. By simulating the mechanical behavior of pore wall contact and separation, the contact mechanics model is used to calculate the pore closing and reopening process under different pressures to obtain the pore wall mechanical state data.

[0130] The first simulation subunit is used to simulate the evolution of the fracture network according to the pore wall mechanical state data, and generate the fracture network evolution data under high pressure conditions by applying a tensor-based permeability model combined with the dynamic changes of pore reopening and pressure distribution;

[0131] The first verification subunit is used to perform verification processing based on the fracture network evolution data, by fitting and matching the fracture network evolution data with the experimental data in the second information, and adjusting the physical model parameters to obtain the optimized pore structure dynamic change data.

[0132] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting gas content in deep shale, characterized in that: include: Acquire first information, second information and third information, wherein the first information is initial formation data including formation pressure, temperature, porosity, permeability and rock physical properties of the area to be predicted, the second information is experimental data of pore structure changes under high pressure conditions, and the third information includes real-time production data and monitoring data; Performing principal component analysis on the first information to extract key influencing factors, and expressing the pore structure characteristics of the deep shale gas reservoir in the form of graph data by capturing the spatial topological relationship, thereby obtaining fourth information, wherein the fourth information includes a simplified input feature set and structured graph data; Performing physical simulation processing according to the fourth information to simulate the pore compression, closing and reopening process under high pressure conditions, generating initial simulation data reflecting the dynamic change of the pore structure, and verifying and matching the initial simulation data with the second information to obtain fifth information, wherein the fifth information is the optimized dynamic change data of the pore structure; Performing scenario simulation processing according to the fifth information, generating sixth information by combining the chain action mechanism of different pore structure state combinations, wherein the sixth information is a classified training data set; The sixth information is processed by model building based on a preset long short-term memory network, and the characteristics of the dynamic changes of the pore structure with pressure and time are captured in combination with the structured graph data, and the first information and the second information are used for verification and correction to obtain a trained proxy model; The third information is input into the proxy model for prediction processing to obtain a prediction result, which includes gas content change trend data and corresponding pore structure evolution results.

2. The method for predicting gas content in deep shale according to claim 1, characterized in that: According to the first information, principal component analysis is performed to extract key influencing factors, and the pore structure characteristics of the deep shale gas reservoir are represented as graph data by capturing the spatial topological relationship, so as to obtain fourth information, wherein the fourth information includes a simplified input feature set and structured graph data, including: Performing principal component analysis processing according to the first information, calculating the covariance matrix of each feature in the initial formation data, performing eigenvalue decomposition, and selecting pressure gradient and porosity change as key influencing factors to obtain a simplified principal component feature set; Performing distance matrix calculation processing according to the principal component feature set, calculating the distance between each sample point in the principal component space by using the Euclidean distance formula, and combining the influence of the change of permeability and porosity in the principal component space on the distance calculation, constructing a symmetric distance matrix; A k-nearest neighbor graph is constructed according to the symmetric distance matrix, k nearest neighbor sample points of each sample point are selected by setting an adjacency threshold k, and an undirected weighted graph is constructed based on the correlation between the pore structure and the pressure gradient to obtain a k-nearest neighbor graph, wherein the nodes in the k-nearest neighbor graph represent the sample points, and the weights of the edges represent the distance values ​​of the corresponding sample points; Graph embedding is performed according to the k-nearest neighbor graph. Graph data is embedded by applying a graph convolutional network. Convolution operations are performed to extract spatial topological features, and the evolution of fracture networks and their impact on permeability are extracted. A multi-layer embedding representation is generated to obtain a simplified input feature set and structured graph data.

3. The method for predicting gas content in deep shale according to claim 1, characterized in that: Performing physical simulation processing according to the fourth information, simulating the pore compression, closing and reopening process under high pressure conditions, generating initial simulation data reflecting the dynamic change of the pore structure, and verifying and matching the initial simulation data with the second information to obtain fifth information, the fifth information is the optimized pore structure dynamic change data, including: Performing finite element analysis according to the fourth information, discretizing the structured graph data, and establishing a finite element model of the porous medium to simulate the compression process of the pores under high pressure conditions, thereby obtaining initial pore compression state data; Performing contact mechanics analysis on the initial pore compression state data, simulating the mechanical behavior of pore wall contact and separation, and using a contact mechanics model to calculate the pore closing and reopening process under different pressures, thereby obtaining pore wall mechanical state data; The fracture network evolution is simulated based on the pore wall mechanical state data. By applying a tensor-based permeability model, combined with the dynamic changes of pore reopening and pressure distribution, the fracture network evolution data under high pressure conditions are generated. Verification processing is performed based on the fracture network evolution data, by fitting and matching the fracture network evolution data with the experimental data in the second information, and adjusting the physical model parameters to obtain optimized pore structure dynamic change data.

4. The method for predicting gas content in deep shale according to claim 1, characterized in that: The scenario simulation process is performed according to the fifth information, and the sixth information is generated by combining the chain action mechanism of different pore structure state combinations. The sixth information is a classified training data set, including: Performing a combination state generation process according to the fifth information, identifying different pore structure state combinations by using a hierarchical clustering algorithm on the fifth information, and using the obtained cluster centers and cluster information as representatives of typical pore states to generate at least two pore structure combination states; According to the combination state, a chain reaction mechanism simulation process is performed, and by constructing a state transition model based on a Markov chain, the transition probability and path between different pore structure state combinations are simulated to obtain the state transition data of the pore structure; Performing scenario generation processing according to the state transition data, randomly sampling the transition processes of different states through Monte Carlo simulation, generating at least two pore structure evolution scenarios, and combining the state combination distribution under extreme pressure changes to obtain a scenario data set containing at least two evolution paths; Classification processing is performed on the scene data set, and a classified training data set is generated by identifying key pore structure evolution patterns and marking them into different categories.

5. The method for predicting gas content in deep shale according to claim 4, characterized in that: According to the combination state, a chain reaction mechanism simulation process is performed, and by constructing a state transition model based on a Markov chain, the transition probability and path between different pore structure state combinations are simulated to obtain the state transition data of the pore structure, including: Performing state space construction processing based on the data generated by the combined state, by identifying the pore compressibility, fracture density, permeability change and gas desorption amount in each pore structure state combination, using a Gaussian mixture model to divide the state space, combining the gas desorption and pressure recovery effects caused by pore structure redistribution, to obtain a discrete state set representing different pore structure evolution stages; According to the state set, a state transition probability estimation process is performed, by analyzing the transition frequency between different pore structure states, combining the Bayesian inference method to update the state transition probability, and combining the influence of pressure fluctuation and gas desorption on state transition under high pressure conditions to obtain a state transition probability matrix; Performing path simulation processing according to the state transition probability matrix, simulating the transfer path of the pore structure state under different pressure conditions by constructing a Markov chain model, focusing on the chain effects caused by pressure mutation, gas desorption and crack reopening, and obtaining a state transition path set; A chain reaction analysis process is performed according to the state transition path set, and the dependency relationship between the paths is analyzed by applying graph theory to identify the chain reaction mechanism and obtain the state transition data.

6. The method for predicting gas content in deep shale according to claim 1, characterized in that: The sixth information is modeled based on a preset long short-term memory network, and the structured graph data is combined to capture the characteristics of the dynamic changes of the pore structure with pressure and time, and the first information and the second information are used for verification and correction to obtain a trained proxy model, including: Performing time series modeling processing according to the sixth information, by using a long short-term memory network to perform sequence training on the training data set, capturing the state evolution of the pore structure at different time points and its dependence on the pressure change, and obtaining a time series feature representation; Performing graph time series fusion processing according to the time series feature representation, combining the time series feature with the structured graph data, using a graph convolutional network to capture spatial topological features and time dependencies, and extracting features of pore structure changing with pressure and time through multi-layer convolution operations, to obtain a spatiotemporal joint feature representation; Performing an initial model construction process according to the spatiotemporal joint feature representation, performing back propagation training on a long short-term memory network and a graph convolutional network joint model, and optimizing network weights to obtain an initial training model; The initial formation data in the first information and the experimental data in the second information are used to compare and verify the dynamic change results of the pore structure output by the initial training model, and the model parameters are adjusted using the gradient descent method to obtain a proxy model.

7. The method for predicting gas content in deep shale according to claim 6, characterized in that: The third information is input into the proxy model for prediction processing to obtain a prediction result, wherein the prediction result includes gas content change trend data of gas content and corresponding pore structure evolution results, including: Performing feature mapping processing according to the third information, converting features in the third information into joint feature representations in a high-dimensional space by using a spatiotemporal joint feature mapping layer in the proxy model, to obtain a mapped spatiotemporal feature vector; Performing dynamic convolution processing according to the spatiotemporal feature vector, applying an adaptive dynamic convolution layer in the proxy model, dynamically adjusting the parameters of the convolution kernel according to the time step, convolution processing the spatiotemporal features at different time points, and combining the evolution pattern of the deep fracture network to obtain a time series feature map; Performing gas content prediction processing according to the time series feature graph, by inputting the time series feature graph into the length memory network layer of the proxy model, combining the gas content change trend in the historical data, predicting the gas content change in the future time period, and obtaining gas content change trend data; The pore structure evolution prediction processing is performed based on the gas content variation trend data, and the redistribution of the pore structure and the crack expansion under different pressure conditions are predicted by performing spatial topological deduction in the proxy model to obtain the predicted pore structure evolution result.

8. A deep shale gas content prediction system, characterized in that: include: A first acquisition unit is used to acquire first information, second information and third information, wherein the first information is initial formation data including formation pressure, temperature, porosity, permeability and rock physical properties of the area to be predicted, the second information is experimental data of pore structure changes under high pressure conditions, and the third information includes real-time production data and monitoring data; a first extraction unit, configured to extract key influencing factors by performing principal component analysis based on the first information, and to represent the pore structure characteristics of the deep shale gas reservoir in the form of graph data by capturing the spatial topological relationship, so as to obtain fourth information, wherein the fourth information includes a simplified input feature set and structured graph data; A first simulation unit is used to perform physical simulation processing according to the fourth information, simulate the pore compression, closing and reopening process under high pressure conditions, generate initial simulation data reflecting the dynamic change of the pore structure, and verify and match the initial simulation data with the second information to obtain fifth information, wherein the fifth information is the optimized dynamic change data of the pore structure; A second simulation unit is used to perform scene simulation processing according to the fifth information, and generate sixth information by combining the chain action mechanism of different pore structure state combinations, wherein the sixth information is a classified training data set; A first modeling unit performs model building processing on the sixth information based on a preset long short-term memory network, combines the structured graph data to capture the characteristics of the dynamic changes of the pore structure with pressure and time, and uses the first information and the second information for verification and correction to obtain a trained proxy model; The first prediction unit is used to input the third information into the proxy model for prediction processing to obtain a prediction result, wherein the prediction result includes gas content change trend data and corresponding pore structure evolution results.

9. The deep shale gas content prediction system according to claim 8, characterized in that: The first extraction unit comprises: A first processing subunit is used to perform principal component analysis processing according to the first information, calculate the covariance matrix of each feature in the initial formation data, perform eigenvalue decomposition, and select pressure gradient and porosity change as key influencing factors to obtain a simplified principal component feature set; A first calculation subunit is used to perform distance matrix calculation processing according to the principal component feature set, calculate the distance between each sample point in the principal component space by using the Euclidean distance formula, and combine the influence of the change of permeability and porosity in the principal component space on the distance calculation to construct a symmetric distance matrix; A first construction subunit is used to perform a k-nearest neighbor graph construction process according to the symmetric distance matrix, select k nearest neighbor sample points of each sample point by setting an adjacency threshold k, and perform an undirected weighted graph construction process based on the correlation between the pore structure and the pressure gradient to obtain a k-nearest neighbor graph, wherein the nodes in the k-nearest neighbor graph represent the sample points, and the weights of the edges represent the distance values ​​of the corresponding sample points; The first embedding subunit is used to perform graph embedding processing according to the k-nearest neighbor graph, embed the graph data by applying a graph convolutional network, perform convolution operations to extract spatial topological features, extract the evolution of the fracture network and its impact on the permeability, and generate a multi-layer embedding representation to obtain a simplified input feature set and structured graph data.

10. The deep shale gas content prediction system according to claim 8, characterized in that: The first simulation unit comprises: A first analysis subunit is used to perform finite element analysis according to the fourth information, by discretizing the structured graph data and establishing a finite element model of the porous medium to simulate the compression process of the pores under high pressure conditions, thereby obtaining initial pore compression state data; The second analysis subunit is used to perform contact mechanics analysis processing according to the initial pore compression state data, and obtain the pore wall mechanical state data by simulating the mechanical behavior of pore wall contact and separation and calculating the pore closing and reopening process under different pressures using a contact mechanics model; The first simulation subunit is used to simulate the evolution of the fracture network according to the pore wall mechanical state data, and generate the fracture network evolution data under high pressure conditions by applying a tensor-based permeability model combined with the dynamic changes of pore reopening and pressure distribution; The first verification subunit is used to perform verification processing based on the fracture network evolution data, by fitting and matching the fracture network evolution data with the experimental data in the second information, and adjusting the physical model parameters to obtain optimized pore structure dynamic change data.

Images