Carbon dioxide storage prediction method and device based on mechanism and data fusion drive

Abstract

The present invention discloses a carbon dioxide storage prediction method and device driven by mechanism and data fusion, wherein the method comprises: obtaining reservoir parameters, well parameters and fluid parameters of a shale reservoir; establishing a multi-scale seepage control model based on the reservoir parameters, the well parameters and the fluid parameters; embedding the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model; and performing storage prediction on the reservoir parameters, the well parameters and the fluid parameters using the shale reservoir carbon dioxide storage prediction model to obtain a prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon capture and storage (CCUS), and in particular to a carbon dioxide storage prediction method and device driven by mechanism and data fusion. Background Art

[0002] With the development of industrialization, large amounts of carbon dioxide (CO2) emissions have triggered global warming. Carbon capture and storage (CCUS) technology is considered an important means of addressing climate change. Geological storage, by injecting CO2 into deep underground reservoirs (such as shale formations), can both mitigate the rise in atmospheric CO2 concentration and increase oil and gas recovery. The widespread micro- and nano-scale pores within shale reservoirs provide excellent geological conditions for CO2 storage. However, after volume fracturing, the internal structure of shale reservoirs becomes more complex, containing numerous nanoscale organic pores and micro- and nano-scale clay mineral pores, as well as natural fractures ranging from micrometers to millimeters and larger artificial fractures, forming a highly heterogeneous, multi-scale storage and permeability system. This complex pore and fracture structure results in significant multi-scale permeability characteristics during CO2 storage, severely limiting the reliability of CO2 storage dynamic predictions.

[0003] Existing technologies primarily rely on numerical simulation methods for dynamic predictions of CO2 storage. Numerical simulations can accurately describe the seepage process by constructing a mathematical model of the reservoir and performing a discrete solution. However, current numerical simulators for CO2 storage in shale reservoirs lack systemic deficiencies in integrating microscopic and macroscopic multiscale transport mechanisms (such as surface diffusion, slip flow, adsorption-desorption, and Darcy flow). These simulators lack effective cross-scale physical mechanism coupling, resulting in a lack of precise connection between microscopic processes and macroscopic flows, which in turn leads to significant deviations in key prediction parameters. Furthermore, traditional numerical simulation methods are limited by meshing and discretization efficiency. When faced with complex reservoir structures ranging from nanopores to kilometer-scale fractures, the computational complexity and single simulation cycles are enormous, making it difficult to meet the demands of rapid response and real-time decision-making for dynamic CO2 storage management.

[0004] In recent years, the use of numerical reservoir simulators to generate large numbers of training samples and machine learning methods to establish a nonlinear mapping relationship between input conditions and storage dynamic results has achieved highly efficient predictions. Although this proxy modeling approach balances prediction efficiency and accuracy to a certain extent, the lack of physical constraints during training results in a typical black-box model, making it difficult to clearly infer the internal learning mechanism, reducing its reliability and credibility in engineering applications. Therefore, there is an urgent need to develop a new intelligent prediction method that integrates multi-scale seepage physics mechanisms while maintaining high computational efficiency. This method can significantly reduce the generalization error of CO2 storage dynamic predictions and provide a technical solution that combines efficiency, accuracy, and physical interpretability for actual storage projects.

[0005] This section is intended to provide a background or context to the embodiments of the invention that are recited in the claims. No statement herein is admitted to be prior art by virtue of its inclusion in this section. Summary of the Invention

[0006] The embodiment of the present invention provides a carbon dioxide storage prediction method based on mechanism and data fusion drive, which is used to realize the fusion modeling of multi-scale seepage mechanism and carbon dioxide storage deep learning model, thereby improving the prediction accuracy and computational efficiency of the carbon dioxide storage process.

[0007] Carbon dioxide storage prediction methods based on mechanism and data fusion include:

[0008] Obtaining reservoir parameters, well parameters and fluid parameters of shale reservoirs;

[0009] establishing a multi-scale seepage control model according to the reservoir parameters, the well parameters and the fluid parameters;

[0010] Embedding the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model;

[0011] The shale reservoir carbon dioxide storage prediction model is used to perform storage prediction on the reservoir parameters, the well parameters and the fluid parameters to obtain a prediction result.

[0012] In one embodiment, establishing a multi-scale seepage control model based on the reservoir parameters, the well parameters, and the fluid parameters includes:

[0013] Establishing a horizontal well injection model according to the reservoir parameters, the well parameters and the fluid parameters;

[0014] A multi-scale seepage control model corresponding to the horizontal well injection model is established according to the reservoir parameters and the fluid parameters.

[0015] In one embodiment, establishing a horizontal well injection model according to the reservoir parameters, the well parameters, and the fluid parameters includes:

[0016] determining injection engineering parameters according to the reservoir parameters, the well parameters, and the fluid parameters;

[0017] Determine the system composition of the horizontal well injection model and its corresponding fluid seepage mechanism based on the reservoir parameters and the fluid parameters

[0018] Determining initial conditions and boundary conditions for carbon dioxide storage based on the injection engineering parameters and the system composition of the horizontal well injection model;

[0019] The horizontal well injection model is established according to the system composition of the horizontal well injection model and its corresponding fluid seepage mechanism, the injection engineering parameters, the initial conditions and the boundary conditions.

[0020] In one embodiment, the steps of training the carbon dioxide sequestration deep learning model include:

[0021] Establishing a multimodal training data set based on the reservoir parameters, the fluid parameters, and the injection engineering parameters;

[0022] Inputting the high-dimensional data in the multimodal training dataset into the carbon dioxide sequestration deep learning model for encoding processing to obtain a low-dimensional feature space;

[0023] Performing Transformer coupling in the low-dimensional feature space to obtain a coupling relationship feature vector;

[0024] The coupling relationship feature vector is decoded to obtain high-dimensional prediction data.

[0025] In one embodiment, performing Transformer coupling in the low-dimensional feature space to obtain a coupling relationship feature vector includes:

[0026] Performing a linear transformation operation on the low-dimensional feature space to generate a query vector, a keyword vector, and a value vector respectively;

[0027] The coupled relationship feature vector after fusion is determined according to the query vector, the keyword vector, the value vector and the softmax function.

[0028] In one embodiment, embedding the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model includes:

[0029] Discretizing the multi-scale seepage control model to obtain a multi-scale seepage control model in a differential form;

[0030] Inputting the high-dimensional prediction data into the multi-scale seepage control model in differential form to obtain physical residual loss;

[0031] Determining data fitting loss, boundary condition residuals, and initial condition residuals respectively according to the high-dimensional prediction data;

[0032] Determining a total loss function according to the physical residual loss, the data fitting loss, the boundary condition residual, the initial condition residual and their corresponding weight factors;

[0033] The model parameters are updated using the Adam optimizer and the total loss function, and the above steps are repeated for iterative training until the total loss function converges, thereby obtaining the shale reservoir carbon dioxide sequestration prediction model.

[0034] In one embodiment, establishing a multimodal training data set based on the reservoir parameters, the fluid parameters, and the injection engineering parameters includes:

[0035] Construct a multi-field coupling numerical model through a multi-physics simulation platform;

[0036] A multimodal training data set is constructed based on the reservoir parameters, the fluid parameters, the injection engineering parameters and the multi-field coupling numerical model.

[0037] In one embodiment, establishing a multimodal training data set based on the reservoir parameters, the fluid parameters, the injection engineering parameters, and the multi-field coupling numerical model includes:

[0038] Performing Latin hypercube sampling on the reservoir parameters, the fluid parameters, and the injection engineering parameters to generate multiple groups of input samples;

[0039] Inputting each group of input samples into the multi-field coupling numerical model for simulation to obtain corresponding output samples;

[0040] Performing standardization on each group of input samples;

[0041] An input-output training sample pair is constructed based on the normalized input sample and the corresponding output sample to obtain the multimodal training data set.

[0042] In one embodiment, performing Latin hypercube sampling on the reservoir parameters, the fluid parameters, and the injection engineering parameters to generate multiple sets of input samples includes:

[0043] Determining the value ranges of the reservoir parameters, the fluid parameters, and the injection engineering parameters;

[0044] The value range corresponding to each input parameter is divided into equal probability according to the preset number of samples to generate multiple sample intervals;

[0045] Performing random sampling within each of the sample intervals to obtain a sample value corresponding to each of the sample intervals;

[0046] The sample values ​​sampled from each input parameter are randomly combined to obtain multiple groups of input samples.

[0047] In one embodiment, the reservoir parameters include: a first pore structure parameter; the fluid parameters include: a first fluid parameter; the horizontal well injection model includes an artificial fracture system; and establishing a multi-scale seepage control model corresponding to the horizontal well injection model based on the reservoir parameters and the fluid parameters includes:

[0048] Based on the first pore structure parameters and the first fluid parameters, a multi-scale seepage control model corresponding to the artificial fracture system is determined; wherein the first pore structure parameters include: the permeability, artificial fracture width and reservoir thickness of the artificial fracture system; the first fluid parameters include: the gas viscosity and gas volume coefficient of the artificial fracture system.

[0049] In one embodiment, the reservoir parameters further include: a second pore structure parameter and a third pore structure parameter; the fluid parameters further include: a second fluid parameter and a third fluid parameter; the horizontal well injection model further includes a natural fracture system; and establishing a multi-scale seepage control model corresponding to the horizontal well injection model based on the reservoir parameters and the fluid parameters includes:

[0050] determining a multi-scale seepage control model corresponding to the natural fracture system according to the second pore structure parameter, the third pore structure parameter, the second fluid parameter, and the third fluid parameter;

[0051] Among them, the second pore structure parameters include: the radius of the natural fracture system, the permeability of the natural fracture system, the stress sensitivity modulus of the natural fracture, the reservoir pressure of the natural fracture system, the interface radius between the fracture system and the inorganic matrix system, the porosity of the natural fracture system and the reservoir temperature of the natural fracture system; the third pore structure parameters include: the shape factor of the inorganic matrix system, the permeability of the inorganic matrix system, the reservoir pressure of the inorganic matrix system and the radius of the inorganic matrix system; the second fluid parameters include: the gas density in the natural fracture system, the gas viscosity of the natural fracture system, the gas compressibility coefficient, the compressibility factor and the ideal gas constant; the third fluid parameters include: the gas density of the inorganic matrix system.

[0052] In one embodiment, the reservoir parameters further include: a fourth pore structure parameter; the fluid parameters further include: a fourth fluid parameter; the horizontal well injection model further includes an inorganic matrix system; and establishing a multi-scale seepage control model corresponding to the horizontal well injection model based on the reservoir parameters and the fluid parameters includes:

[0053] Determining a multi-scale seepage control model corresponding to the inorganic matrix system according to the third pore structure parameter, the fourth pore structure parameter, the third fluid parameter, the fourth fluid parameter, and the preset first adsorption model parameter;

[0054] Among them, the third pore structure parameter also includes: clay mineral content, interface radius between kerogen system and inorganic matrix system, and porosity of inorganic matrix system; the fourth pore structure parameter includes: kerogen content, shape factor of kerogen system, permeability of kerogen system, reservoir pressure of kerogen system and radius of kerogen system; the third fluid parameter also includes: gas slip coefficient in inorganic matrix system and gas compressibility coefficient in inorganic matrix system; the fourth fluid parameter includes: gas density in kerogen system; the first adsorption model parameter includes: isothermal adsorption volume of inorganic matrix system and isothermal adsorption pressure of inorganic matrix system.

[0055] In one embodiment, the reservoir parameters further include: a fifth pore structure parameter; the fluid parameters further include: a fifth fluid parameter; the horizontal well injection model further includes a porous kerogen system; and establishing a multi-scale seepage control model corresponding to the horizontal well injection model based on the reservoir parameters and the fluid parameters includes:

[0056] Determining a multi-scale seepage control model corresponding to the porous kerogen system according to the fourth pore structure parameter, the fifth pore structure parameter, the fourth fluid parameter, the fifth fluid parameter, and the preset second adsorption model parameter;

[0057] Among them, the fourth pore structure parameter also includes: the porosity of the kerogen system; the fifth pore structure parameter includes: the shape factor of the organic matter system and the particle radius of the organic matter system; the fourth fluid parameter also includes: the gas slip coefficient of the kerogen system and the gas compressibility coefficient of the kerogen system; the fifth fluid parameter includes: the gas density in the organic matter system, the effective diffusion coefficient of the gas and the gas concentration in the organic matter system; the second adsorption model parameter includes: the isothermal adsorption volume of the kerogen system and the isothermal adsorption pressure of the kerogen system.

[0058] In one embodiment, the horizontal well injection model further includes: the organic matter system, and the multi-scale seepage control model corresponding to the horizontal well injection model is established according to the reservoir parameters and the fluid parameters, including:

[0059] A multi-scale seepage control model corresponding to the organic matter system is determined according to the fifth pore structure parameter and the fifth fluid parameter.

[0060] An embodiment of the present invention also provides a carbon dioxide storage prediction device driven by mechanism and data fusion, which is used to realize the fusion modeling of multi-scale seepage mechanism and carbon dioxide storage deep learning model, thereby improving the prediction accuracy and computational efficiency of the carbon dioxide storage process.

[0061] The CO2 storage prediction device based on mechanism and data fusion drive includes:

[0062] A parameter acquisition module, used to obtain reservoir parameters, well parameters and fluid parameters of the shale reservoir;

[0063] A model building module, configured to establish a multi-scale seepage control model based on the reservoir parameters, the well parameters, and the fluid parameters;

[0064] A model fusion module is used to embed the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model;

[0065] The prediction output module is used to perform storage prediction on the reservoir parameters, the well parameters and the fluid parameters using the shale reservoir carbon dioxide storage prediction model to obtain prediction results.

[0066] An embodiment of the present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned carbon dioxide sequestration prediction method based on mechanism and data fusion is implemented.

[0067] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned carbon dioxide sequestration prediction method based on mechanism and data fusion.

[0068] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the above-mentioned carbon dioxide sequestration prediction method based on mechanism and data fusion.

[0069] The present invention provides a mechanism- and data-driven CO2 storage prediction method and device. In developing a multiscale seepage control model, the reservoir system is divided into multiple scales, including artificial fractures, natural fractures, inorganic matrix, kerogen, and organic matter, taking into account the complex heterogeneous structure of shale reservoirs. These systems then construct corresponding differential form seepage control equations for each scale. This multiscale seepage control model fully considers the coupled migration behaviors of multiple mechanisms, including Darcy flow, slip flow, dissolution-diffusion, and multicomponent competitive adsorption, comprehensively reflecting the migration and storage characteristics of CO2 within different scale subdomains. This effectively enhances the multiscale seepage control model's ability to physically characterize and adapt to the dynamic process of CO2 storage. Furthermore, the multiscale seepage control model is embedded in a U-Net neural network architecture to construct a deep learning model for CO2 storage that incorporates physical constraints. By introducing the physical residual term, boundary condition consistency term, and initial condition constraint term into the differential governing equations, a hybrid loss function is constructed. This improves model fitting accuracy while ensuring that prediction results strictly conform to physical field constraints, enhancing the model's interpretability and predictive stability. The shale reservoir CO2 storage prediction model efficiently predicts the CO2 storage process by inputting any combination of reservoir, well, and fluid parameters. It rapidly outputs injection dynamics, pressure distribution, and phase transition results at multiple time scales. Compared to traditional numerical simulation methods, this method significantly improves prediction accuracy and computational efficiency, demonstrating excellent engineering practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0071] Figure 1 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in one embodiment of the present invention;

[0072] Figure 2 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0073] Figure 3 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0074] Figure 4This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0075] Figure 5 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0076] Figure 6 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0077] Figure 7 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0078] Figure 8 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0079] Figure 9 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0080] Figure 10 Schematic diagram of the structure of the horizontal well injection model in an embodiment of the present invention;

[0081] Figure 11 Schematic diagram of the structure of the shale reservoir carbon dioxide storage prediction model in an embodiment of the present invention;

[0082] Figure 12 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0083] Figure 13 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0084] Figure 14 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive in another embodiment of the present invention;

[0085] Figure 15 This is a diagram showing the pressure difference prediction effect of the shale reservoir carbon dioxide storage prediction model under a numerical simulation data set in an embodiment of the present invention;

[0086] Figure 16 This is a diagram showing the effect of pressure difference prediction based on actual shale reservoir production data using the shale reservoir carbon dioxide storage prediction model in an embodiment of the present invention;

[0087] Figure 17 This is a diagram showing the pressure difference prediction effect during the carbon dioxide injection process according to an embodiment of the present invention;

[0088] Figures 18a to 18c Graph showing the parameter inversion and dynamic prediction effects of the carbon dioxide storage process under different data noise conditions in an embodiment of the present invention;

[0089] Figure 19 This is a diagram showing the dynamic prediction effects of different storage mechanisms during the entire carbon dioxide storage process in a shale reservoir according to an embodiment of the present invention;

[0090] Figure 20 This is a flow chart of a carbon dioxide storage prediction method based on mechanism and data fusion drive of a carbon dioxide storage prediction device based on mechanism and data fusion drive in an embodiment of the present invention;

[0091] Figure 21 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0092] To make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0093] The information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0094] Provide users with corresponding operation entrances for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, they will enter the expert decision-making process.

[0095] To address the technical challenges of existing CO2 storage prediction methods driven by mechanism and data fusion, such as high numerical simulation computational costs, large parameter uncertainty, and a lack of physical consistency, the present invention provides a CO2 storage prediction method based on mechanism and data fusion. This method, based on a multi-physics simulation platform (such as COMSOL), constructs a multi-field coupled numerical model that couples the seepage, chemical, temperature, and mechanical fields. It combines Latin hypercube sampling with SOBOL global sensitivity analysis to generate a high-quality training dataset. Based on a U-Net neural network, it introduces the discrete residuals of the multiscale seepage equation and constructs a physically consistent loss function, enabling rapid and accurate prediction of CO2 storage in shale reservoirs. This method effectively improves the computational efficiency and physical rationality of storage dynamic predictions, enhancing the model's adaptability and generalization capabilities to complex reservoir conditions.

[0096] like Figure 1 As shown, the carbon dioxide storage prediction method based on mechanism and data fusion drive includes steps 101 to 104.

[0097] Step 101: Obtain reservoir parameters, well parameters, and fluid parameters of a shale reservoir.

[0098] Step 102: Establish a multi-scale seepage control model based on reservoir parameters, well parameters, and fluid parameters.

[0099] Step 103: Embed the multi-scale seepage control model into the pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model.

[0100] Step 104: Perform storage prediction on the reservoir parameters, the well parameters, and the fluid parameters using a shale reservoir carbon dioxide storage prediction model to obtain a prediction result.

[0101] In the present embodiment, during the development of a multiscale seepage control model, the reservoir system is divided into multiple scale levels, including artificial fracture systems, natural fracture systems, inorganic matrix systems, kerogen systems, and organic matter systems, taking into account the complex heterogeneous structure of shale reservoirs. Corresponding differential form seepage control equations are then constructed for each level. This multiscale seepage control model fully considers the coupled migration behaviors of multiple mechanisms, including Darcy flow, slip flow, dissolution-diffusion, and multicomponent competitive adsorption, comprehensively reflecting the migration and storage characteristics of carbon dioxide within different scale subdomains. This effectively enhances the multiscale seepage control model's ability to physically characterize and adapt to the dynamic process of carbon dioxide storage. Furthermore, this multiscale seepage control model is embedded in a U-Net neural network architecture to construct a deep learning model for carbon dioxide storage that incorporates physical constraints. By introducing the physical residual term, boundary condition consistency term, and initial condition constraint term into the differential governing equations, a hybrid loss function is constructed. This improves model fitting accuracy while ensuring that prediction results strictly conform to physical field constraints, enhancing the model's interpretability and predictive stability. The shale reservoir CO2 storage prediction model efficiently predicts the CO2 storage process by inputting any combination of reservoir, well, and fluid parameters. It rapidly outputs injection dynamics, pressure distribution, and phase transition results at multiple time scales. Compared to traditional numerical simulation methods, this method significantly improves prediction accuracy and computational efficiency, demonstrating excellent engineering practicality.

[0102] like Figure 1 As shown, the carbon dioxide storage prediction method based on mechanism and data fusion drive includes steps 101 to 104.

[0103] Step 101: Obtain reservoir parameters, well parameters, and fluid parameters of a shale reservoir.

[0104] Specifically, reservoir parameters such as reservoir temperature, reservoir pressure, porosity, permeability, kerogen content, clay mineral content, natural fracture permeability and reservoir compressibility of shale reservoirs can be obtained through core analysis experiments, logging data interpretation, well testing and geological modeling, but the present invention is not limited to this.

[0105] Furthermore, well parameters related to fracture structure, such as artificial fracture width, fracture stimulation radius, and number of artificial fractures, can be obtained based on fracturing design documents, fracturing operation records, and microseismic monitoring results. Furthermore, wellbore design data and wellbore trajectory measurement data can be combined to obtain wellbore parameters related to wellbore structure, such as wellbore trajectory, wellbore radius, and wellbore storage coefficient. However, the present invention is not limited to this.

[0106] Furthermore, fluid parameters such as CO2 diffusion coefficient, CO2 solubility coefficient, gas density, gas viscosity, gas compressibility, adsorption volume, adsorption pressure and slip coefficient can be obtained through PVT experiments, EOS (equation of state) modeling and other methods, but the present invention is not limited to this.

[0107] The above reservoir parameters, well parameters and fluid parameters are shown in Table 1 below.

[0108] Table 1

[0109]

[0110] Step 102: Establish a multi-scale seepage control model based on reservoir parameters, well parameters, and fluid parameters.

[0111] In one embodiment, if Figure 2 As shown, step 102 includes steps 201 and 202.

[0112] Step 201: Establish a horizontal well injection model (i.e., a multi-scale system physical model for carbon dioxide storage in shale reservoirs) based on reservoir parameters, well parameters, and fluid parameters.

[0113] Specifically, volumetric fracturing can be used to create a dense artificial fracture network in target shale reservoirs, enhancing their permeability. Using horizontal wells as injection wells for CO2 injection operations can create a complete injection-storage scenario, enabling simulation modeling and dynamic prediction of CO2 storage behavior within shale reservoirs.

[0114] like Figure 10As shown in the figure, the established multi-scale system physical model for carbon dioxide storage in shale reservoirs takes into account the heterogeneity of the reservoir medium and the multi-scale characteristics of the pore structure. The multi-scale system includes: artificial fracture system, natural fracture system, inorganic matrix system, kerogen system and organic matter system, which correspond to different fluid seepage mechanisms and storage behaviors.

[0115] In one embodiment, if Figure 3 As shown, step 201 includes steps 301 to 304.

[0116] Step 301: Determine injection engineering parameters based on reservoir parameters, well parameters, and fluid parameters.

[0117] Specifically, combining reservoir parameters such as reservoir porosity, permeability and natural fracture distribution characteristics, well parameters such as wellbore structure, artificial fracture geometry and its conductivity, and fluid parameters such as CO2 density, CO2 viscosity, compressibility factor, CO2 solubility and diffusivity, a comprehensive analysis of the fluid seepage mechanism and storage behavior during CO2 injection is conducted to determine the injection engineering parameters (Injection Engineering Parameters) and engineering design conditions (Design Configuration Parameters) suitable for the CO2 storage process in shale reservoirs.

[0118] Engineering design conditions include: injection stage settings, well type selection, and reservoir stimulation methods. Injection engineering parameters include: injection medium type, injection rate, injection time, injection pressure, and total injection volume.

[0119] Furthermore, the well type adopts a horizontal well structure arranged along the direction of shale bedding to increase the contact area between the wellbore and the reservoir. The reservoir transformation method adopts volume fracturing technology, which uses hydraulic force to form a high-density artificial fracture system in the reservoir, which can improve the overall permeability of the shale reservoir. The injection medium is carbon dioxide in a supercritical state, which has high solubility and fluidity. The injection rate should be controlled within a safe range that does not cause abnormal crack expansion or reservoir stress sensitivity effects. The injection pressure should be higher than the reservoir pore pressure to ensure that the injected fluid enters the reservoir smoothly. At the same time, the injection pressure should be controlled within the allowable range of the wellbore pressure bearing capacity to ensure the safety of the injection process. The injection time and total injection volume can be optimized according to the storage target, reservoir capacity and injection safety.

[0120] Step 302: Determine the system components of the horizontal well injection model and its corresponding fluid seepage mechanism based on reservoir parameters and fluid parameters. The horizontal well injection model is a multi-scale system physical model for carbon dioxide storage in shale reservoirs.

[0121] Specifically, in artificial fracture systems, carbon dioxide migrates mainly along the fracture channels in the form of limited conductivity, and its fluid seepage mechanism adopts the modified Darcy flow equation to characterize the fluid transport characteristics under the condition of limited fracture conductivity.

[0122] In natural fracture systems, the seepage mechanism of carbon dioxide follows the traditional Darcy's law, and the permeability can be modified according to the stress sensitivity characteristics, that is, a stress-sensitive permeability function is introduced to reflect the influence of fracture conductivity changes with formation stress.

[0123] In an inorganic matrix system, this system primarily consists of an inorganic pore structure formed by clay minerals and other non-organic minerals, as well as micro- and nanoscale inorganic channels distributed within organic matter. CO2 undergoes slip flow within these micro- and nanopores, requiring the use of a slip-corrected Darcy flow equation to characterize the discontinuous flow behavior under scale effects. Furthermore, the influence of pore size on the Knudsen number must be considered to perform scale corrections on the traditional percolation equation. Furthermore, CO2 molecules undergo reversible physical adsorption on the surfaces of clay minerals and organic matter, with adsorption behavior obeying the Langmuir isotherm model, which characterizes the nonlinear relationship between adsorption capacity and partial pressure.

[0124] In kerogen systems, CO2 primarily flows by slip within nanoscale pores, accompanied by competitive adsorption between native gases (such as methane) and CO2. Simultaneously, a certain proportion of CO2 molecules undergo dissolution and diffusion. To accurately characterize this non-convective migration process, a dissolution-diffusion term is introduced into the model to describe the diffusion and adsorption of CO2 molecules into the organic matrix, thereby simulating molecular-scale sequestration effects.

[0125] In the organic matter system, the migration of carbon dioxide still exhibits slip flow characteristics, and there is also a competitive adsorption process with native gas. The adsorption behavior conforms to the multi-component Langmuir isotherm adsorption model.

[0126] The seepage mechanisms of the above-mentioned systems together constitute the physical flow model of the multi-scale system of CO2 injection and storage in shale reservoirs.

[0127] Step 303: Determine the initial conditions and boundary conditions for carbon dioxide storage based on the injection engineering parameters and the system composition of the horizontal well injection model.

[0128] Specifically, initial conditions primarily include information about the physical state of the shale reservoir prior to CO2 injection, including reservoir pressure and temperature fields, porosity and permeability distributions at various scales, and concentrations of native gases (such as methane). These initial conditions can be determined based on historical production data, interpreted well logging data, well test analysis, and geological modeling data to accurately characterize the reservoir's original physical state prior to CO2 storage.

[0129] Boundary conditions are set based on the constructed storage scenario and typically include injection wellbore boundaries, far-field boundaries, and vertical boundaries (caprock and floor). The injection well boundary can be set to a constant injection rate boundary or a constant injection pressure boundary based on engineering design parameters to control the injection drive conditions. The far-field boundary can be set to a constant pressure boundary or a no-flow boundary (Neumann boundary) to simulate the constraints or responses of the storage system's peripheral areas to seepage behavior. Vertical boundaries are typically set to adiabatic or impermeable boundaries to ensure the system's vertical containment and meet the physical constraints of shale storage.

[0130] In addition, in order to accurately simulate the seepage and migration behavior of carbon dioxide in a multi-scale system, appropriate continuity conditions need to be imposed at the interfaces of different subdomains in the horizontal well injection model, including pressure continuity conditions and mass flux continuity conditions, to ensure the stability and physical consistency of the horizontal well injection model during the numerical solution process.

[0131] Step 304: Establish a horizontal well injection model based on the system components of the horizontal well injection model and its corresponding fluid seepage mechanism, injection engineering parameters, initial conditions, and boundary conditions.

[0132] Specifically, based on the spatial structure and distribution characteristics of artificial fracture systems, natural fracture systems, inorganic matrix systems, kerogen systems and organic matter systems and their corresponding fluid seepage mechanisms, combined with injection engineering parameters, initial conditions and boundary conditions, they are uniformly integrated into a coupled physical modeling framework to construct a multi-scale system physical model of the entire process of carbon dioxide injection, migration and storage in shale reservoirs.

[0133] Based on the spatial structure and distribution characteristics of artificial fracture systems, natural fracture systems, inorganic matrix systems, kerogen systems, and organic matter systems, as well as the corresponding seepage mechanisms and migration behaviors of each system, combined with injection engineering parameters such as injection medium type, injection rate, and injection pressure, as well as initial conditions and boundary constraints such as the pre-injection pressure field, temperature field, and component distribution, these are integrated into a multiscale coupled physical modeling framework to construct a multiscale system physical model that describes the entire process of CO2 injection, migration, and storage in shale reservoirs. This multiscale system physical model accurately reflects the complex flow characteristics and multi-mechanism storage behavior within porous media, providing a theoretical basis and boundary input for subsequent multi-physics field numerical simulations and U-Net neural network training.

[0134] Step 202: Establish a multi-scale seepage control model corresponding to the horizontal well injection model based on reservoir parameters and fluid parameters.

[0135] Specifically, based on the constructed multi-scale physical model of CO2 storage in shale reservoirs, and combining the spatial structural characteristics and distribution patterns of artificial fracture systems, natural fracture systems, inorganic matrix systems, kerogen systems, and organic matter systems, the seepage control equations for CO2 migration behavior in each scale system were established (i.e., the multi-scale seepage control model for CO2 injection and storage in shale reservoirs). This multi-scale seepage control model fully accounts for the differences between different media types in pore size, seepage mechanisms, adsorption behavior, and non-convective migration mechanisms, and can accurately characterize the multi-scale flow patterns of CO2 in shale reservoirs during the injection, migration, and storage process.

[0136] In one embodiment, the reservoir parameters include first pore structure parameters, the fluid parameters include first fluid parameters, and the horizontal well injection model includes an artificial fracture system.

[0137] Step 202 specifically includes determining a multi-scale seepage control model (i.e., a multi-scale seepage control equation) corresponding to the artificial fracture system based on the first pore structure parameter and the first fluid parameter. The first pore structure parameter includes the permeability K of the artificial fracture system. hf 、Artificial crack width ω hf The first fluid parameters include the injected gas viscosity μ and the gas volume coefficient B.

[0138] Specifically, after carbon dioxide is compressed on the ground and enters a supercritical state, it is injected into the shale reservoir through a horizontal wellbore and first enters a fracture system composed of an artificial fracture system and a natural fracture system.

[0139] Within an artificial fracture system, carbon dioxide migrates along the fracture channel in a limited conductivity form. Its seepage behavior is affected by the fracture geometry constraints and stress sensitivity effects. Therefore, a modified Darcy seepage model is used to describe it, and limited conductivity and stress sensitivity correction terms are introduced to characterize the actual fluid transport characteristics under the condition of limited fracture conductivity.

[0140] The multi-scale seepage control equation of the artificial fracture system is as follows:

[0141]

[0142] Among them, K hf is the permeability of the artificial fracture system, ω hf is the width of the artificial fracture, h is the vertical thickness of the fracture segment penetrating the reservoir, μ is the viscosity of the injected gas, B is the gas volume coefficient, and p hf is the pressure distribution function along the crack direction in the artificial crack section, q schf(x,t) is the gas flow rate function of the unit length fracture segment at position x and time t, where x is the spatial coordinate of the artificial fracture segment and t is the time variable.

[0143] In one embodiment, the reservoir parameters further include: a second pore structure parameter and a third pore structure parameter. The fluid parameters further include: a second fluid parameter and a third fluid parameter. The horizontal well injection model further includes a natural fracture system.

[0144] Step 202 specifically includes determining a multi-scale seepage control model corresponding to the natural fracture system based on the second pore structure parameter, the third pore structure parameter, the second fluid parameter, and the third fluid parameter. The second pore structure parameter includes: the radius r of the natural fracture system, the permeability K of the natural fracture system, and the second fluid parameter. lf , natural fracture stress sensitivity modulus α, reservoir pressure of the natural fracture system, interface radius R between the fracture system and the inorganic matrix system m , porosity of natural fracture system φ fir and reservoir temperature T. The third pore structure parameter includes: the shape factor β of the inorganic matrix system m , the permeability K of the inorganic matrix system m , reservoir pressure p of inorganic matrix system m and the radius r of the inorganic matrix system m The second fluid parameter includes: gas density ρ in the natural fracture system ir , gas viscosity μ, gas molar mass M g , gas compressibility coefficient C gir , compressibility factor Z and ideal gas constant R. The third fluid parameter includes: gas density ρ of the inorganic matrix system m .

[0145] Specifically, in natural fracture systems, the migration process of carbon dioxide follows the traditional Darcy seepage law, and its permeability can be corrected by a stress-sensitive permeability function to characterize the changing characteristics of fracture conductivity with changes in formation stress.

[0146] The multi-scale seepage control equation of the natural fracture system is as follows:

[0147]

[0148] Where r is the radius of the natural fracture system, ρ ir is the gas density in the natural fracture system, K lf is the permeability of the natural fracture system, α is the stress sensitivity modulus of the natural fracture, and p i is the initial reservoir pressure of the natural fracture system, p ir is the current reservoir pressure of the natural fracture system, μ is the gas viscosity, R m is the interface radius between the crack system and the inorganic matrix system, ρ m is the gas density in the inorganic matrix system, β m is the shape factor of the inorganic matrix system, K m is the permeability of the inorganic matrix system, p m is the reservoir pressure of the inorganic matrix system, r m is the radius of the inorganic matrix system, M g is the molar mass of the gas, C gir is the gas compressibility coefficient of the natural fracture system, φ fir is the porosity of the natural fracture system, Z is the compressibility factor, R is the ideal gas constant, T is the reservoir temperature, and t is the time variable.

[0149] In one embodiment, the reservoir parameters further include a fourth pore structure parameter, the fluid parameters further include a fourth fluid parameter, and the horizontal well injection model further includes an inorganic matrix system.

[0150] Step 202 specifically includes: determining a multi-scale seepage control model corresponding to the inorganic matrix system according to the third pore structure parameter, the fourth pore structure parameter, the third fluid parameter, the fourth fluid parameter and the preset first adsorption model parameter.

[0151] Among them, the third pore structure parameter also includes: clay mineral content f c , the interface radius R between the kerogen system and the inorganic matrix system k and the porosity φ of the inorganic matrix system m The fourth pore structure parameter includes: kerogen content f k , the shape factor β of the kerogen system k , the permeability of the kerogen system K k, reservoir pressure p of kerogen system k and the radius r of the kerogen system k The third fluid parameter also includes: the gas slip coefficient α in the inorganic matrix system m and the gas compressibility coefficient C of the inorganic matrix system gm The fourth fluid parameter includes: gas density ρ in the kerogen system k The first adsorption model parameters include: the isothermal adsorption volume V of the inorganic matrix system Lc and the isothermal adsorption pressure p of the inorganic matrix system Lc .

[0152] Specifically, CO2 migrates from the fracture system into the inorganic matrix system within the shale matrix. This inorganic matrix system is composed of inorganic pore structures at the micro- and nanoscale. The small pore size and the free path of gas molecules are comparable to the pore size, significantly affecting the percolation behavior of CO2. In this context, the flow of CO2 in the inorganic matrix system exhibits distinct slip flow characteristics. The overall percolation mechanism is described using a slip-corrected Darcy flow model to reflect the physical characteristics of discontinuous flow in micro- and nanoscale pores.

[0153] Furthermore, carbon dioxide molecules can compete with native gas molecules (such as methane) for adsorption on the surfaces of clay particles within the inorganic pore structure, forming a reversible storage mechanism primarily based on physical adsorption. This physical adsorption process follows the Langmuir isotherm adsorption model, which can be used to characterize the nonlinear relationship between adsorption capacity and local partial pressure, thereby reflecting the static storage capacity of carbon dioxide in the shale matrix system.

[0154] The multi-scale seepage control equation of the inorganic matrix system is as follows:

[0155]

[0156] Among them, r m is the radius of the inorganic matrix system, ρ m is the gas density in the inorganic matrix system, β m is the shape factor of the inorganic matrix system, α m is the gas slip coefficient in the inorganic matrix system, K m is the permeability of the inorganic matrix system, μ is the gas viscosity, and p m is the reservoir pressure of the inorganic matrix system, f c is the clay mineral content, V Lc is the Langmuir isothermal adsorption volume of clay minerals, p Lc is the Langmuir isothermal adsorption pressure of clay minerals, R k The interface radius between the kerogen system and the inorganic matrix system, ρ kis the gas density in the kerogen system, f k is the kerogen content, β k is the shape factor of the kerogen system, K k Permeability of the kerogen system, p k is the reservoir pressure of the kerogen system, r k is the radius of the kerogen system, M g is the molar mass of the gas, C gm is the gas compressibility coefficient of the inorganic matrix system, φ m is the porosity of the inorganic matrix system, Z is the compressibility factor, R is the ideal gas constant, T is the reservoir temperature, and t is the time variable.

[0157] In one embodiment, the reservoir parameters further include a fifth pore structure parameter, the fluid parameters further include a fifth fluid parameter, and the horizontal well injection model further includes a porous kerogen system.

[0158] Step 202 specifically includes: determining a multi-scale seepage control model corresponding to the porous kerogen system based on the fourth pore structure parameter, the fifth pore structure parameter, the fourth fluid parameter, the fifth fluid parameter and the preset second adsorption model parameter.

[0159] Among them, the fourth pore structure parameter also includes: φ k is the porosity of the kerogen system. The fifth pore structure parameter includes: the shape factor β of the organic matter system o and the particle radius r of the organic matter system o The fourth fluid parameter also includes: the gas slip coefficient α of the kerogen system k and the gas compressibility coefficient C of the kerogen system gk The fifth fluid parameter includes: gas density ρ in the organic system o , gas effective diffusion coefficient D E and gas concentration C in organic matter system o The second adsorption model parameters include: the isothermal adsorption volume V of the kerogen system Lk and the isothermal adsorption pressure p of the kerogen system Lk .

[0160] Specifically, CO2 migrates from the inorganic matrix system to the organic matter system dominated by kerogen. Because kerogen systems typically have small pore sizes at the nanoscale, CO2 flows primarily through slip flow, significantly influenced by the coupling effect between pore size and gas free path. Furthermore, CO2 molecules can compete with native gases (such as methane) for adsorption on pore walls. This adsorption process is reversible, and the adsorption mechanism can be described using a multicomponent Langmuir isotherm adsorption model, accurately characterizing the adsorption distribution of different gas components on the kerogen pore surface.

[0161] To accurately simulate the complex migration mechanisms of carbon dioxide within kerogen systems, a coupled mechanism combining slip flow, competitive adsorption, and dissolution-diffusion is introduced into a multiscale seepage control model of kerogen systems. This coupled mechanism not only reflects the convective migration of carbon dioxide within kerogen pores, but also simulates the non-convective migration of carbon dioxide through pore walls into the organic matrix, thereby enabling a quantitative characterization of the entire process of carbon dioxide sequestration in organic matter systems.

[0162] The multi-scale seepage control equation of the porous kerogen system is as follows:

[0163]

[0164] Among them, r k is the radius of the kerogen system, ρ k is the gas density in the kerogen system, β k is the shape factor of the kerogen system, α k is the gas slip coefficient of the kerogen system, K k is the permeability of the kerogen system, μ is the gas viscosity, and p k is the reservoir pressure of the kerogen system, V Lk is the Langmuir isotherm adsorption volume of the kerogen system, p Lk is the Langmuir isothermal adsorption pressure of the kerogen system, φ k is the porosity of the kerogen system, f k is the kerogen content, R o is the initial particle radius of the organic matter system, ρ o Gas density in organic matter system, β o Form factor of the organic matter system, D E is the effective diffusion coefficient of gas, C o is the gas concentration in the organic matter system, r o is the particle radius of the organic matter system, M g is the molar mass of the gas, C gk is the gas compressibility coefficient of the kerogen system, Z is the compression factor, R is the ideal gas constant, T is the reservoir temperature, and t is the time variable.

[0165] In one embodiment, the horizontal well injection model also includes an organic system. Step 202 specifically includes: determining a multi-scale seepage control model corresponding to the organic system based on the fifth pore structure parameter and the fifth fluid parameter.

[0166] Specifically, CO2 diffuses from the inner wall of nanoscale kerogen pores along a concentration gradient into the interior of organic matter particles, dissolving in the organic system as molecules, resulting in solid-phase storage dominated by dissolved CO2. To accurately characterize this non-convection-dominated migration behavior, a multiscale seepage control model of the organic system was introduced, incorporating slip flow, competitive adsorption, multicomponent Langmuir isothermal adsorption, and dissolution-diffusion coupling mechanisms to simulate the complex migration and storage of CO2 within microscopic organic structures.

[0167] The multi-scale seepage control equation of the organic matter system is as follows:

[0168]

[0169] Among them, r o is the particle radius of the organic matter system, ρ o is the gas density in the organic matter system, β o is the shape factor of the organic matter system, D E is the effective diffusion coefficient of gas, C o is the gas concentration in the organic matter system, and t is the time variable.

[0170] In one embodiment, if Figure 4 As shown, the training steps of the carbon dioxide sequestration deep learning model include steps 401 to 404.

[0171] Step 401: Establish a multimodal training data set based on reservoir parameters, fluid parameters, and injection engineering parameters.

[0172] In one embodiment, if Figure 5 As shown, step 401 includes steps 501 to 502.

[0173] Step 501: Construct a multi-field coupling numerical model through a multi-physics field simulation platform.

[0174] Specifically, based on the COMSOL Multiphysics simulation platform, a numerical modeling method combining the finite element method and the volume averaging method was adopted to comprehensively consider the multi-scale seepage characteristics of shale reservoirs and the multi-physical process coupling mechanism of carbon dioxide-water-shale. Based on the above-constructed multi-scale system physical model of carbon dioxide storage in shale reservoirs and the corresponding multi-scale seepage control model, a multi-field coupling numerical model of coupled seepage field, chemical field, temperature field and mechanical field was established.

[0175] like Figure 12 As shown, step 501 includes steps 1201 to 1206.

[0176] Step 1201: Construct a geometric model of a multi-scale system in the COMSOL multi-physics simulation platform.

[0177] For example, in the COMSOL multi-physics simulation platform, create a new three-dimensional component (such as Component1 (3D)) and select a basic physical interface suitable for complex fluid structures (such as Porous Media Flow or Transport of Diluted Species, etc.).

[0178] like Figure 13 As shown, step 120 includes steps 1301 to 1304 .

[0179] Step 1301: Construct the representative elementary volume (REV) shell geometry.

[0180] For example, a cube or rectangular parallelepiped structure (for example, with a size of 1m × 1m × 1m) is drawn under the Geometry node as the computational domain of the multi-scale shale reservoir.

[0181] Step 1302: embed a discrete fracture network (DFN) inside REV to establish an artificial fracture system and a natural fracture system.

[0182] For example, multiple planar cracks are created inside the REV geometry through the Work Plane node. The cracks can be modeled as ultra-thin cuboids or thin film volumes. The crack thickness is set to the millimeter level, and the spatial position, distribution direction and geometric dimensions of the cracks are set to truly reflect the structural characteristics of artificial cracks and natural cracks.

[0183] Step 1303: Construct an inorganic matrix system and a kerogen system inside REV.

[0184] For example, third-party software (such as Avizo, PoreSpy, or OpenPNM) is used to generate point cloud or volume mesh data of the microscopic pore network and save this data in STL or OBJ format. This microscopic pore structure is then imported into the Import Geometry node of the COMSOL multiphysics simulation platform. The imported pore-throat network is designated as a "hollow domain," and the corresponding porosity and slip-corrected flow parameters are set to establish a 3D geometric model of the inorganic matrix and kerogen systems.

[0185] Step 1304: After completing the modeling of the artificial fracture system, the natural fracture system, the inorganic matrix system, and the kerogen system, physically connect the different systems to obtain a multi-scale geometric model.

[0186] For example, the Identity Pair node or ContinuityBoundary Conditions node in the COMSOL multi-physics simulation platform is used to set the physical coupling between the boundaries of each system to ensure that the flow variables (such as pressure and concentration) maintain continuity at the boundaries of each system, which can meet the physical closure of the coupling model.

[0187] Furthermore, during the modeling process, different porous media parameters can be set based on the characteristics of the systems at each scale. For example, the fracture system region can be set to high permeability, high porosity, and low adsorption capacity; the inorganic matrix region can be set to low permeability, medium porosity, and a slip flow correction term; and the kerogen system region can be set to very low permeability and high adsorption capacity to reflect the dominant role of kerogen in CO2 storage.

[0188] Step 1202: Construct multiple physical field control modules in the COMSOL multi-physics field simulation platform, wherein the physical fields include seepage field, chemical field, temperature field, and mechanical field.

[0189] For example, in the Component node of the COMSOL project file, add the following modules in sequence: Darcy's Law module (for seepage field), Transport of Diluted Species module (for chemical field), Heat Transfer in Porous Media module (for temperature field), and Solid Mechanics module (for mechanical field).

[0190] Step 1203: Set key physical parameters of each physical field control module.

[0191] For example, in the Darcy's Law module, the porosity, permeability, and physical properties related to fluid flow, such as gas density and gas viscosity, of the porous media region are set. In the Transport of Diluted Species module, the concentration and diffusion coefficient of the injected medium (such as CO2 and related mineral ions) are set, and reaction equations describing the CO2 dissolution reaction and mineral precipitation reaction are added. In the Heat Transfer in Porous Media module, heat transfer parameters such as thermal conductivity, specific heat capacity, and initial temperature field are set. In the Solid Mechanics module, mechanical parameters such as the elastic modulus, Poisson's ratio, and Biot coefficient of the rock material are set to describe the poroelastic coupling behavior, but the present invention is not limited thereto.

[0192] Step 1204: Establish coupling relationships between physical fields in the Multiphysics node of the COMSOL project file.

[0193] For example, the Poroelasticity Coupling interface can be used to establish a coupling relationship between the Darcy's Law module and the Solid Mechanics module, thereby realizing a bidirectional coupling between the seepage field and the mechanical field. Furthermore, the Non-Isothermal Flow Coupling interface is used to realize the coupling between the Darcy's Law module and the Heat Transfer in Porous Media module. In the non-isothermal flow coupling process, the fluid velocity field is set to drive the thermal convection term, and the temperature dependence of the fluid parameters (such as gas density and gas viscosity, etc.) is configured to characterize the feedback regulation effect of temperature changes on the seepage behavior. In addition, in the Transport of Diluted Species module, the mass flux term caused by the seepage field is added through the Source Term node to construct a coupling mechanism between the seepage field and the chemical field, thereby realizing the linkage modeling of carbon dioxide migration and mineral reaction processes.

[0194] Step 1205: In the COMSOL multi-physics simulation platform, set a custom partial differential equation. The partial differential equation includes: a mineral dissolution rate equation, a porosity evolution equation, and a permeability dynamic change equation, but the present invention is not limited thereto.

[0195] For example, in the Add Physics node of the COMSOL project file, add a Weak Form PDE module and enter a custom PDE expression.

[0196] For example:

[0197] The mineral dissolution rate equation is as follows:

[0198]

[0199] Where C is the mineral concentration or the mass of undissolved minerals, t is the time variable, k diss is the mineral dissolution rate constant, C co2 is the CO2 concentration in the reaction environment.

[0200] The porosity evolution equation is as follows:

[0201]

[0202] Among them, φ is the porosity at the current moment, φ0 is the initial porosity, △φ reaction Indicates the amount of porosity change caused by the reaction between minerals and fluids.

[0203] The dynamic change equation of permeability is as follows:

[0204]

[0205] Among them, K is the penetration rate at the current moment, K o is the initial permeability, φ is the porosity at the current moment, φ0 is the initial porosity, and n is an empirical index, which is usually 2-5.

[0206] Step 1206: Based on the multi-physics field control module, injection parameters, boundary conditions and initial fields are set, and simulation is performed to obtain a multi-field coupling numerical model.

[0207] For example, in the Darcy's Law module, a boundary condition is applied to the boundary corresponding to the injection wellbore, and a constant injection rate boundary or a constant injection pressure boundary can be selected.

[0208] In the Transport of Diluted Species module, apply an Inflow condition to the same injection boundary and set the concentration of the injected medium (e.g., CO2). In the Heat Transfer in Porous Media module, set the temperature condition for the injection boundary and the temperature of the injected medium.

[0209] Through the above parameter settings and boundary configurations, after completing the physical field assignment of the model's initial state, the simulation calculation can be started to obtain a multi-field coupled numerical model including seepage field, chemical field, temperature field and mechanical field, providing basic data support for the subsequent carbon dioxide dynamic storage process simulation and U-Net neural network training.

[0210] Step 502: Construct a multimodal training data set based on reservoir parameters, fluid parameters, injection engineering parameters, and a multi-field coupling numerical model.

[0211] In one embodiment, if Figure 6 As shown, step 502 includes: steps 601 to 604.

[0212] Step 601: Perform Latin hypercube sampling on reservoir parameters, fluid parameters, and injection engineering parameters to generate multiple groups of input samples.

[0213] Specifically, Latin hypercube sampling is used to generate samples of input parameters to obtain multiple groups of input samples, where the input parameters include reservoir parameters, fluid parameters, and injection engineering parameters.

[0214] In one embodiment, if Figure 7 As shown, step 601 includes: steps 701 to 704.

[0215] Step 701: Determine the value ranges of reservoir parameters, fluid parameters, and injection engineering parameters.

[0216] Illustratively, the value ranges of the input parameters are shown in Table 2 below, but the present invention is not limited thereto.

[0217] Table 2

[0218]

[0219] Step 702: Divide the value range corresponding to each input parameter into equal probability according to a preset number of samples to generate multiple sample intervals.

[0220] Exemplarily, the number of samples is usually set to 100 groups, but the present invention is not limited thereto.

[0221] The porosity [0.03, 0.05], matrix permeability [10 -18 , 10 -15 ], crack permeability [10 -16 , 10 -13 ], fracture density [0.01, 0.1], injection pressure [20, 40], and injection temperature [310, 340] are divided into 100 equal probability sample intervals respectively.

[0222] Step 703: Perform random sampling in each sample interval to obtain the sample value corresponding to each sample interval.

[0223] Specifically, a sample value is randomly selected from each sample interval corresponding to each input parameter.

[0224] For example, in the 23rd sample interval of porosity [0.042, 0.0432], a sample value, for example, 0.0427, is randomly selected as the sample value corresponding to the sample interval.

[0225] Step 704: Randomly combine all sample values ​​sampled from each input parameter to obtain multiple groups of input samples.

[0226] Specifically, each set of input samples consists of a sample value vector, and the input sample set is X={x1,x2,…,x k ,…,x n}, where x k Input sample for group k, where n is the total number of samples.

[0227] For example:

[0228] x k={porosity, permeability, fracture density, injection pressure, ..., injection temperature}={0.035,1.5×10 -16 ,0.05,28.7,…,321}.

[0229] Step 602: Input each set of input samples into a multi-field coupling numerical model for simulation to obtain corresponding output samples.

[0230] Specifically, the above input samples are sequentially input into the constructed multi-field coupling numerical model for simulation, and the corresponding shale reservoir carbon dioxide storage dynamic data under different input parameter combinations are obtained, namely, output samples.

[0231] The output sample set is Y={y1,y2,…,y k ,…,y n}.

[0232] Among them, y k The kth group of output samples, n is the total number of output sample groups, y k Includes variables such as injection dynamics, pressure distribution and concentration field during the carbon dioxide storage process.

[0233] Step 603: Perform standardization processing on each group of input samples.

[0234] Specifically, the following normalization formula is used to normalize the input sample vector x k Each component x ki Perform normalization processing.

[0235]

[0236] Among them, x* ki is the i-th input component after normalization, x ki is the original value of the i-th input component before normalization, μ(x k ) is the sample mean, σ(x k ) is the sample standard deviation.

[0237] Step 604: construct input and output training sample pairs based on the normalized input samples and their corresponding output samples to obtain a multimodal training dataset.

[0238] Specifically, the standardized input sample set X*={x1*,x2*,…, x k *,…, x n *} and the corresponding output sample set Y={y1,y2,…,y k ,…,y n} are paired to form a set of input and output training sample pairs (X*, Y), thereby forming a multimodal training dataset.

[0239] Preferably, in one embodiment, Figure 14 As shown, the key control parameters in the carbon dioxide storage process of shale reservoirs are screened out by the SOBOL global sensitivity analysis method, including steps 1401 to 1408.

[0240] Step 1401: Determine the value range of input parameters and establish correlation constraints between input parameters. Input parameters include reservoir parameters, fluid parameters, well parameters, injection engineering parameters, etc.

[0241] For example, the porosity can be set to a value range between 0.03 and 0.15, and the matrix permeability can be set to a value range between 10 -18 and 10 -15 m 2 and stipulates that the correlation coefficient between porosity and matrix permeability is 0.8, indicating a high positive correlation.

[0242] Step 1402: Perform Saltelli extended sampling on each input parameter to generate an independent Sobol sequence sample matrix for each input parameter.

[0243] For example, Saltelli extended sampling is performed on porosity and permeability respectively to obtain corresponding sample matrices A and B1.

[0244] Step 1403: Perform Copula transformation on the generated independent Sobol sequence sample matrices to construct a joint sample matrix that satisfies the correlation constraint.

[0245] Exemplarily, Copula transformation is performed on sample matrices A and B1 to obtain a joint sample matrix C with a correlation coefficient of 0.8.

[0246] Step 1404: Perform an inverse transformation on the joint sample matrix to obtain a plurality of Sobol input samples.

[0247] Specifically, the joint sample matrix after Copula transformation is mapped from the standard normal space back to the original physical quantity space to obtain the Sobol input samples that meet the set correlation coefficient.

[0248] Step 1405: Input the Sobol input samples into the multi-field coupling numerical model for simulation, obtain corresponding output samples, and construct a set of input and output sample pairs.

[0249] Step 1406: Bootstrap resampling is performed on the input and output sample pairs, and the Sobol sensitivity index is calculated after each round of sampling. The Sobol sensitivity index includes: first-order effect index S i , total effect index S Tiand the second-order effect index S ij The present invention is not limited thereto.

[0250] Specifically, the input and output sample pairs are resampled with replacement by the Bootstrap method. After each round of resampling, the first-order effect index S is calculated. i , total effect index S Ti and the second-order effect index S ij The above process is repeated until the preset sampling times (e.g., 1000 times) are reached to construct a sample set for each sensitivity index.

[0251] For example, the first-order effect index S of porosity and permeability is calculated i , the first-order effect index of porosity is 0.32, and the first-order effect index of permeability is 0.06.

[0252] Step 1407: Perform statistical analysis on the sample set of each sensitivity index to obtain the confidence interval of the sensitivity index.

[0253] Specifically, each set of sensitivity index samples was sorted from small to large according to the numerical value, and the 2.5th percentile and the 97.5th percentile were extracted respectively to obtain the corresponding 95% confidence interval.

[0254] For example, the first-order effect index of porosity is 0.32, and its confidence interval CI is [0.28, 0.35]. The first-order effect index of permeability is 0.06, and its confidence interval CI is [0.02, 0.09].

[0255] Step 1408: Filter the sensitivity index according to a preset threshold to obtain a set of key control parameters.

[0256] For example, the screening criteria can be set as follows: if the first-order effect index S of a certain input parameter is i >0.05 or the second-order effect index S with other parameters ij >0.01, the input parameter is a key control parameter.

[0257] For example, if the first-order effect index of porosity is 0.32>0.05, and the first-order effect index of permeability is 0.06>0.05, then both can be determined as key control parameters.

[0258] Step 402: Input the high-dimensional data from the multimodal training dataset into the CO2 storage deep learning model for encoding, generating a low-dimensional feature space. The CO2 storage deep learning model is a neural network built on a U-Net architecture, comprising an encoder, a decoder, and a Transformer layer based on an attention mechanism.

[0259] Specifically, High-dimensional data φ at time t , input into the U-Net neural network model, and perform multi-layer convolution and pooling operations through the encoder of the U-Net neural network model to transform the input high-dimensional data φ t Mapped to a low-dimensional high-order feature space, low-dimensional compressed features are obtained. t It is a multimodal training dataset.

[0260] Step 403: Perform Transformer coupling in the low-dimensional feature space to obtain a coupling relationship feature vector.

[0261] Specifically, in the low-dimensional feature space, the self-attention mechanism fusion operation between the input data is performed through the Transformer layer in the U-Net neural network to obtain the coupling relationship feature vector Attention(q,k,v).

[0262] In one embodiment, if Figure 8 As shown, step 403 includes steps 801 to 802.

[0263] Step 801: Perform a linear transformation operation on the low-dimensional feature space to generate a query vector q, a keyword vector k, and a value vector v respectively.

[0264] Specifically, the low-dimensional compressed features output by the encoder are linearly transformed through the fully connected layer in the U-Net neural network to obtain the query vector q, keyword vector k and value vector v.

[0265] Step 802: Determine the fused coupling relationship feature vector Attention(q, k, v) based on the query vector q, the keyword vector k, the value vector v and the Softmax function.

[0266] Specifically, according to the formula of the attention mechanism, the coupling relationship feature vector Attention(q,k,v) is calculated.

[0267]

[0268] Among them, q is the query vector, k is the keyword vector, v is the value vector, d k is a vector dimension constant.

[0269] First, by calculating the similarity between the query vector q and the keyword vector k, the attention weight coefficient qk is obtained T Secondly, through the Softmax function and the vector dimension constant d k After normalization, the attention weights between the feature points are obtained Finally, the value vector v is weighted and summed based on the attention weight to obtain the fused coupling relationship feature vector Attention(q,k,v).

[0270] Step 404: Decode the coupling relationship feature vector Attention(q, k, v) to obtain high-dimensional prediction data.

[0271] Specifically, the decoder in the U-Net neural network performs multi-layer upsampling and convolution operations on the coupling relationship feature vector Attention(q,k,v), restores the spatial resolution layer by layer, and finally reconstructs the multimodal high-dimensional data φ at time t+δt t+δt , which can realize the time series prediction of CO2 storage dynamics in shale reservoirs.

[0272] Step 103: Embed the multi-scale seepage control model into the pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model.

[0273] In one embodiment, if Figure 9 and Figure 11 As shown, step 103 includes steps 901 to 905.

[0274] Step 901: Discretize the multi-scale seepage control model to obtain a multi-scale seepage control model in differential form.

[0275] Specifically, the multiscale seepage control model includes the seepage control equations for artificial fracture systems, natural fracture systems, inorganic matrix systems, kerogen systems, and organic matter systems. These seepage control equations are discretized using the finite difference method or the finite volume method to obtain differential expressions.

[0276] Step 902: Input the high-dimensional prediction data into the multi-scale percolation control model in differential form to obtain the physical residual loss.

[0277] Specifically, the high-dimensional prediction data φ output by the U-Net neural network model at time t+δt t+δt Substitute it into the differential expression of the multi-scale seepage control equation, calculate the discrete equation residual of each physical system item by item, and obtain the physical residual value.

[0278] Furthermore, based on the discrete equation residuals of each physical system, the physical residual loss L of the U-Net neural network model is calculated according to the following formula (11): f .

[0279]

[0280] Among them, Residual iIt represents the residual value corresponding to the i-th prediction sample in the differential expression, and N represents the total number of prediction samples.

[0281] Step 903: Determine the data fitting loss L based on the high-dimensional prediction data s , boundary condition residual L b and the initial condition residual L i .

[0282] Specifically, the high-dimensional prediction result φ output by the U-Net neural network model at time t+δt is t+δt , construct data fitting loss L s , boundary condition residual L b and the initial condition residual L i and other loss functions.

[0283] Among them, the data fitting loss L s Used to measure the prediction output of the U-Net neural network model (i.e. the high-dimensional prediction result φ t+δt The error between the numerical simulation result (i.e., the real observation data) and the high-quality numerical simulation result can be calculated using the mean square error (MSE) or the mean absolute error (MAE).

[0284] Boundary condition residual L b Used to evaluate the prediction results (i.e., high-dimensional prediction results φ t+δt ) satisfies the consistency constraints of the preset physical boundary conditions (such as fixed pressure boundary, zero flow boundary or injection boundary) on the spatial boundary. t+δt Substitute it into the control equation at the boundary position and calculate its residual.

[0285] Initial condition residual loss L i Used to evaluate the prediction results of the U-Net neural network model (i.e., the high-dimensional prediction results φ t+δt ) is consistent with the initial state conditions, ensuring that the starting state of the predicted CO2 storage process is consistent with the initial configuration of the physical model. This is calculated by comparing the error between the predicted result at time t=0 and the initial field configuration.

[0286] Step 904: According to the physical residual loss L f , data fitting loss L s , boundary condition residual L b and the initial condition residual L i and its corresponding weight factors to determine the total loss function L.

[0287] Specifically, the total loss function L is calculated according to formula (12).

[0288]

[0289] Among them, L s is the data fitting loss, L f is the physical residual loss, L b is the boundary condition residual, L i is the initial condition residual, λ s ,λ f ,λ b and λ i is the weight factor.

[0290] Step 905: Update the model parameters using the Adam optimizer and the total loss function, and repeat the above steps for iterative training until the total loss function converges to obtain a shale reservoir CO2 storage prediction model.

[0291] Specifically, the Adam optimizer is used to update the parameters of the U-Net neural network model to minimize the above-mentioned total loss function L, and steps 902 to 905 are iteratively executed until the loss function converges or the set number of training rounds is reached, thereby completing the training of the shale reservoir carbon dioxide storage prediction model.

[0292] In this embodiment of the present invention, a shale reservoir CO2 storage prediction model incorporating physical constraints is constructed by differentially discretizing the multiscale seepage governing equations and embedding them into a U-Net neural network structure. This significantly reduces the computational overhead of simulating the storage process and improves overall computational efficiency. Furthermore, by introducing physical residual terms, boundary condition consistency terms, and initial condition constraints, a hybrid loss function is constructed, which not only ensures prediction accuracy but also enhances the physical interpretability and convergence stability of the model results. Furthermore, combined with a multi-field coupling model constructed using the COMSOL multiphysics simulation platform, the model covers the coupling relationships between seepage, temperature, mechanical, and chemical fields. It accurately characterizes the migration mechanism of CO2 in multiscale shale reservoir systems (including artificial fracture systems, natural fracture systems, inorganic matrix systems, kerogen systems, and nano-organic matter systems), and comprehensively reflects the physical and chemical behaviors of supercritical CO2 and native gas in actual engineering, such as competitive adsorption, slip flow, dissolution and diffusion, and mineralization reactions. Furthermore, Latin hypercube sampling was used to generate a representative sample set, and key influencing parameters were screened based on the SOBOL global sensitivity analysis method. This improved the diversity and representativeness of the training dataset, enhancing the generalization and stability of the prediction model. The proposed shale reservoir CO2 storage prediction model is adaptable to both numerical simulation data and field monitoring data, demonstrating excellent engineering applicability and suitability for dynamic prediction and safety assessment of CO2 storage under complex geological conditions.

[0293] Step 104: Perform storage prediction on the reservoir parameters, the well parameters, and the fluid parameters using a shale reservoir carbon dioxide storage prediction model to obtain a prediction result.

[0294] Specifically, based on the trained shale reservoir carbon dioxide storage prediction model, the reservoir parameters, well parameters and fluid parameters of the target shale reservoir are input, and the dynamic changes of the carbon dioxide injection process are quickly predicted and quantitatively evaluated.

[0295] like Figure 15 The figure shows the pressure difference prediction effect of the shale reservoir carbon dioxide storage prediction model under the numerical simulation data set. Among them, the horizontal axis is the prediction result Δp of the shale reservoir carbon dioxide storage prediction model. 预测 (Unit: MPa), the vertical axis is the simulation result Δp output by the COMSOL multi-physics simulation platform 模拟 (Unit: MPa).

[0296] The Pearson correlation coefficient ρ and goodness of fit R 2 Evaluate the prediction accuracy of the CO2 storage prediction model for shale reservoirs. Figure 15 As shown, the calculation results show that ρ=0.993, R 2 =0.986, indicating that the shale reservoir CO2 storage prediction model has extremely high prediction accuracy and consistency on the multimodal training dataset.

[0297] like Figure 16 As shown in the figure, the shale reservoir carbon dioxide storage prediction model predicts the pressure difference during the carbon dioxide injection process after inputting the actual shale reservoir production data. Among them, the horizontal axis is the prediction result Δp of the shale reservoir carbon dioxide storage prediction model. 预测 (Unit: MPa), the vertical axis is the actual result Δp measured on site 实测 (Unit: MPa).

[0298] The Pearson correlation coefficient ρ and goodness of fit R 2 Evaluate the prediction accuracy of the CO2 storage prediction model for shale reservoirs. Figure 16 As shown, the calculation results show that ρ=0.990, R 2 =0.980, further verifying that the shale reservoir CO2 storage prediction model also has a high generalization ability in processing actual engineering data.

[0299] like Figure 17 Figure 2 shows the CO2 injection dynamics prediction curves for two typical samples in the test dataset using the shale reservoir CO2 storage prediction model. The blue curve represents the simulated data, and the red curve represents the prediction results. It can be seen that the model's predicted curves are highly consistent with the simulated curves at different time scales.

[0300] like Figures 18a to 18c As shown in Figure 3, the dynamic prediction accuracy of the shale reservoir CO2 storage prediction model is evaluated under different observation noise levels (such as 0%, 10%, and 20%). Figure 18a This is the prediction result of the shale reservoir carbon dioxide storage prediction model without noise interference. The pressure change curve constructed by the shale reservoir carbon dioxide storage prediction model almost completely coincides with the real data. Figure 18b and Figure 18c The following table shows the prediction results of the shale reservoir CO2 storage prediction model under 10% and 20% noise perturbation conditions. Although the predicted curves show slight fluctuations, the overall trend remains well consistent with the true curve, demonstrating that the shale reservoir CO2 storage prediction model has good robustness to observation errors and parameter inversion capabilities.

[0301] Figure 19 This is a dynamic prediction diagram of the effects of different storage mechanisms during the entire process of carbon dioxide storage in shale reservoirs. As the storage period progresses, carbon dioxide gradually migrates from the supercritical state to the dissolved state, adsorbed state, ionized state, and mineralized state. It is predicted that after 300 years, the supercritical state storage capacity will reach 5.12×10 5 tons, adsorption storage capacity 3.08×10 5 tons and dissolved CO2 is 9.58×10 4 Tons. The total storage capacity based on physical storage mechanism is 9.158×10 5 tons, and the mineralized and ionized conversion parts are sealed in total 1.082×10 5 ton. Figure 19 Demonstrates the quantitative capability of the shale reservoir CO2 storage prediction model in predicting long-term conversion pathways.

[0302] In the embodiments of the present invention, the shale reservoir carbon dioxide storage prediction model can not only accurately predict injection dynamics and pressure changes, but also has good adaptability and robustness. It can effectively reflect the multi-field coupling mechanism of the shale reservoir and the time evolution process of the storage path. It can be widely used in engineering scenarios such as shale reservoir carbon dioxide storage assessment and optimized injection design.

[0303] The present invention also provides a device for predicting carbon dioxide storage based on a mechanism-driven data fusion approach, as described in the following examples. Because the principles underlying this device are similar to those of the method for predicting carbon dioxide storage based on a mechanism-driven data fusion approach, the implementation of this device can be found in the implementation of the method for predicting carbon dioxide storage based on a mechanism-driven data fusion approach, and any repetitions will not be repeated.

[0304] like Figure 20As shown, the carbon dioxide storage prediction device 2000 driven by mechanism and data fusion includes: a parameter acquisition module 2001, a model construction module 2002, a model fusion module 2003 and a prediction output module 2004.

[0305] The parameter acquisition module is used to obtain reservoir parameters, well parameters and fluid parameters of the shale reservoir;

[0306] The model building module is used to establish a multi-scale seepage control model according to the reservoir parameters, the well parameters and the fluid parameters;

[0307] The model fusion module is used to embed the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model;

[0308] The prediction output module is used to perform storage prediction on the reservoir parameters, the well parameters and the fluid parameters through the shale reservoir carbon dioxide storage prediction model to obtain prediction results.

[0309] In one embodiment, the model building module includes: a horizontal well injection model building unit and a seepage mechanism modeling unit.

[0310] The horizontal well injection model building unit is used to build a horizontal well injection model according to the reservoir parameters, the well parameters and the fluid parameters;

[0311] The seepage mechanism modeling unit is used to establish a multi-scale seepage control model corresponding to the horizontal well injection model according to the reservoir parameters and the fluid parameters.

[0312] In one embodiment, the horizontal well injection model building unit includes: a parameter determination unit, a system identification unit, a boundary initial value setting unit and a model building execution unit.

[0313] The parameter determination unit is used to determine the injection engineering parameters according to the reservoir parameters, the well parameters and the fluid parameters;

[0314] The system identification unit is used to determine the system composition of the horizontal well injection model and its corresponding fluid seepage mechanism according to the reservoir parameters and the fluid parameters;

[0315] The boundary initial value setting unit is used to determine the initial conditions and boundary conditions of carbon dioxide storage according to the injection engineering parameters and the system composition of the horizontal well injection model;

[0316] The model building execution unit is used to establish the horizontal well injection model according to the system composition, seepage mechanism, injection parameters, initial conditions and boundary conditions.

[0317] In one embodiment, the training module of the carbon dioxide sequestration deep learning model includes a dataset construction module, an encoding module, a feature coupling module, and a decoding module.

[0318] The data set construction module is used to establish a multimodal training data set based on the reservoir parameters, the fluid parameters and the injection engineering parameters;

[0319] The encoding module is used to input the high-dimensional data in the multimodal training data set into the carbon dioxide sequestration deep learning model for encoding processing to obtain a low-dimensional feature space;

[0320] The feature coupling module is used to perform Transformer coupling in the low-dimensional feature space to obtain a coupling relationship feature vector;

[0321] The decoding module is used to decode the coupling relationship feature vector to obtain high-dimensional prediction data.

[0322] In one embodiment, the feature coupling module includes: a vector generation unit and a coupling fusion unit.

[0323] The vector generating unit is used to perform a linear transformation operation on the low-dimensional feature space to generate a query vector, a keyword vector and a value vector respectively;

[0324] The coupling fusion unit is used to determine the fused coupling relationship feature vector according to the query vector, keyword vector, value vector and softmax function.

[0325] In one embodiment, the model fusion module includes: a physical model discretization unit, a physical loss acquisition unit, a boundary loss calculation unit, a total loss function construction unit and an optimization training unit.

[0326] The physical model discretization unit is used to discretize the multi-scale seepage control model to obtain a multi-scale seepage control model in a differential form;

[0327] The physical loss acquisition unit is used to input the high-dimensional prediction data into the multi-scale seepage control model in the differential form to obtain the physical residual loss;

[0328] The boundary loss calculation unit is used to determine the data fitting loss, the boundary condition residual and the initial condition residual according to the high-dimensional prediction data;

[0329] The total loss function construction unit is used to determine the total loss function according to the physical residual loss, data fitting loss, boundary condition residual and initial condition residual and their corresponding weight factors;

[0330] The optimization training unit is used to update the model parameters through the Adam optimizer and the total loss function, and repeat the above steps to perform iterative training until the total loss function converges, thereby obtaining the shale reservoir carbon dioxide storage prediction model.

[0331] In one embodiment, the data set construction module includes: a simulation model construction unit and a sample data generation unit.

[0332] The simulation model building unit is used to build a multi-field coupling numerical model through a multi-physics field simulation platform;

[0333] The sample data generating unit is used to construct a multimodal training data set according to the reservoir parameters, the fluid parameters, the injection engineering parameters and the multi-field coupling numerical model.

[0334] In one embodiment, the sample data generating unit includes: a sampling unit, a simulation unit, a sample standardization unit and a training sample construction unit.

[0335] The sampling unit is used to perform Latin hypercube sampling on the reservoir parameters, the fluid parameters and the injection engineering parameters to generate multiple groups of input samples;

[0336] The simulation unit is used to input each group of input samples into the multi-field coupling numerical model for simulation to obtain corresponding output samples;

[0337] The sample standardization unit is used to perform standardization processing on each group of input samples;

[0338] The training sample construction unit is used to construct input-output training sample pairs according to the input samples after standardization and the corresponding output samples to obtain the multimodal training data set.

[0339] In one embodiment, the sampling module includes: a parameter determination unit, an interval generation unit, a random sampling unit and a random combination unit.

[0340] The parameter determination unit is used to determine the value ranges of the reservoir parameters, the fluid parameters and the injection engineering parameters;

[0341] The interval generation unit is used to divide the value range corresponding to each input parameter into equal probability according to the preset number of samples to generate multiple sample intervals;

[0342] The random sampling unit is used to perform random sampling in each of the sample intervals to obtain a sample value corresponding to each of the sample intervals;

[0343] The random combination unit is used to randomly combine the sample values ​​sampled from each input parameter to obtain multiple groups of input samples.

[0344] In one embodiment, the model building module further includes an artificial fracture modeling unit.

[0345] The artificial fracture modeling unit is used to determine a multi-scale seepage control model corresponding to the artificial fracture system according to the first pore structure parameter and the first fluid parameter;

[0346] The first pore structure parameters include: the permeability of the artificial fracture system, the width of the artificial fracture and the reservoir thickness; the first fluid parameters include: the gas viscosity and the gas volume coefficient of the artificial fracture system.

[0347] In one embodiment, the model building module further includes a natural fracture modeling unit.

[0348] The natural fracture modeling unit is used to determine a multi-scale seepage control model corresponding to the natural fracture system according to the second pore structure parameter, the third pore structure parameter, the second fluid parameter and the third fluid parameter;

[0349] Among them, the second pore structure parameters include: the radius of the natural fracture system, the permeability of the natural fracture system, the stress sensitivity modulus of the natural fracture, the reservoir pressure of the natural fracture system, the interface radius between the fracture system and the inorganic matrix system, the porosity of the natural fracture system and the reservoir temperature of the natural fracture system; the third pore structure parameters include: the shape factor of the inorganic matrix system, the permeability of the inorganic matrix system, the reservoir pressure of the inorganic matrix system and the radius of the inorganic matrix system; the second fluid parameters include: the gas density in the natural fracture system, the gas viscosity of the natural fracture system, the gas compressibility coefficient, the compressibility factor and the ideal gas constant; the third fluid parameters include: the gas density of the inorganic matrix system.

[0350] In one embodiment, the model building module further includes an inorganic matrix modeling unit.

[0351] The inorganic matrix modeling unit is used to determine a multi-scale seepage control model corresponding to the inorganic matrix system according to the third pore structure parameter, the fourth pore structure parameter, the third fluid parameter, the fourth fluid parameter and the preset first adsorption model parameter;

[0352] Among them, the third pore structure parameter also includes: clay mineral content, interface radius between kerogen system and inorganic matrix system, and porosity of inorganic matrix system; the fourth pore structure parameter includes: kerogen content, shape factor of kerogen system, permeability of kerogen system, reservoir pressure of kerogen system and radius of kerogen system; the third fluid parameter also includes: gas slip coefficient in inorganic matrix system and gas compressibility coefficient in inorganic matrix system; the fourth fluid parameter includes: gas density in kerogen system; the first adsorption model parameter includes: isothermal adsorption volume of inorganic matrix system and isothermal adsorption pressure of inorganic matrix system.

[0353] In one embodiment, the model building module further includes a porous kerogen modeling unit.

[0354] The porous kerogen modeling unit is used to determine a multi-scale seepage control model corresponding to the porous kerogen system according to the fourth pore structure parameter, the fifth pore structure parameter, the fourth fluid parameter, the fifth fluid parameter and the preset second adsorption model parameter;

[0355] Among them, the fourth pore structure parameter also includes: the porosity of the kerogen system; the fifth pore structure parameter includes: the shape factor of the organic matter system and the particle radius of the organic matter system; the fourth fluid parameter also includes: the gas slip coefficient of the kerogen system and the gas compressibility coefficient of the kerogen system; the fifth fluid parameter includes: the gas density in the organic matter system, the effective diffusion coefficient of the gas and the gas concentration in the organic matter system; the second adsorption model parameter includes: the isothermal adsorption volume of the kerogen system and the isothermal adsorption pressure of the kerogen system.

[0356] In one embodiment, the model building module further includes an organic matter system modeling unit.

[0357] The organic system modeling unit is used to determine a multi-scale seepage control model corresponding to the organic system according to the fifth pore structure parameter and the fifth fluid parameter.

[0358] Figure 21 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as Figure 5 As shown, the electronic device includes: a processor (processor) 2101, a memory (memory) 2102 and a bus 2103.

[0359] The processor 2101 and the memory 2102 communicate with each other via the bus 2103 .

[0360] The processor 2101 is used to call the program instructions in the memory 2102 to execute the methods provided by the above-mentioned method embodiments.

[0361] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the above-mentioned carbon dioxide sequestration prediction method based on mechanism and data fusion.

[0362] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the above-mentioned carbon dioxide sequestration prediction method based on mechanism and data fusion.

[0363] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0364] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0365] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0366] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0367] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A carbon dioxide storage prediction method based on mechanism and data fusion, characterized in that: include: Obtaining reservoir parameters, well parameters and fluid parameters of shale reservoirs; establishing a multi-scale seepage control model according to the reservoir parameters, the well parameters and the fluid parameters; Embedding the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model; Performing storage prediction on the reservoir parameters, the well parameters, and the fluid parameters using the shale reservoir carbon dioxide storage prediction model to obtain a prediction result; The establishing of a multi-scale seepage control model according to the reservoir parameters, the well parameters and the fluid parameters includes: establishing a horizontal well injection model according to the reservoir parameters, the well parameters and the fluid parameters; establishing a multi-scale seepage control model corresponding to the horizontal well injection model according to the reservoir parameters and the fluid parameters; The multi-scale seepage control model is embedded in a pre-trained carbon dioxide sequestration deep learning model to obtain a shale reservoir carbon dioxide sequestration prediction model, including: discretizing the multi-scale seepage control model to obtain a differential multi-scale seepage control model; inputting high-dimensional prediction data into the differential multi-scale seepage control model to obtain a physical residual loss; determining data fitting loss, boundary condition residuals, and initial condition residuals based on the high-dimensional prediction data; determining a total loss function based on the physical residual loss, the data fitting loss, the boundary condition residuals, the initial condition residuals, and their corresponding weight factors; updating model parameters using an Adam optimizer and the total loss function, and repeating the above steps for iterative training until the total loss function converges, thereby obtaining the shale reservoir carbon dioxide sequestration prediction model.

2. The method according to claim 1, characterized in that The establishing of a horizontal well injection model according to the reservoir parameters, the well parameters and the fluid parameters comprises: determining injection engineering parameters according to the reservoir parameters, the well parameters, and the fluid parameters; Determining the system composition of the horizontal well injection model and its corresponding fluid seepage mechanism according to the reservoir parameters and the fluid parameters; Determining initial conditions and boundary conditions for carbon dioxide storage based on the injection engineering parameters and the system composition of the horizontal well injection model; The horizontal well injection model is established according to the system composition of the horizontal well injection model and its corresponding fluid seepage mechanism, the injection engineering parameters, the initial conditions and the boundary conditions.

3. The method according to claim 2, characterized in that The training steps of the carbon dioxide sequestration deep learning model include: Establishing a multimodal training data set based on the reservoir parameters, the fluid parameters, and the injection engineering parameters; Inputting the high-dimensional data in the multimodal training dataset into the carbon dioxide sequestration deep learning model for encoding processing to obtain a low-dimensional feature space; Performing Transformer coupling in the low-dimensional feature space to obtain a coupling relationship feature vector; The coupling relationship feature vector is decoded to obtain high-dimensional prediction data.

4. The method according to claim 3, characterized in that The Transformer coupling is performed in the low-dimensional feature space to obtain a coupling relationship feature vector, including: Performing a linear transformation operation on the low-dimensional feature space to generate a query vector, a keyword vector, and a value vector respectively; The coupled relationship feature vector after fusion is determined according to the query vector, the keyword vector, the value vector and the softmax function.

5. The method according to claim 3, characterized in that The establishing of a multimodal training data set based on the reservoir parameters, the fluid parameters, and the injection engineering parameters includes: Construct a multi-field coupling numerical model through a multi-physics simulation platform; A multimodal training data set is constructed based on the reservoir parameters, the fluid parameters, the injection engineering parameters and the multi-field coupling numerical model.

6. The method according to claim 5, characterized in that The step of establishing a multimodal training data set based on the reservoir parameters, the fluid parameters, the injection engineering parameters, and the multi-field coupling numerical model includes: Performing Latin hypercube sampling on the reservoir parameters, the fluid parameters, and the injection engineering parameters to generate multiple groups of input samples; Inputting each group of input samples into the multi-field coupling numerical model for simulation to obtain corresponding output samples; Performing standardization on each group of input samples; An input-output training sample pair is constructed based on the normalized input sample and the corresponding output sample to obtain the multimodal training data set.

7. The method according to claim 6, characterized in that The Latin hypercube sampling is performed on the reservoir parameters, the fluid parameters, and the injection engineering parameters to generate multiple groups of input samples, including: Determining the value ranges of the reservoir parameters, the fluid parameters, and the injection engineering parameters; The value range corresponding to each input parameter is divided into equal probability according to the preset number of samples to generate multiple sample intervals; Performing random sampling within each of the sample intervals to obtain a sample value corresponding to each of the sample intervals; The sample values ​​sampled from each input parameter are randomly combined to obtain multiple groups of input samples.

8. The method according to claim 1, characterized in that The reservoir parameters include: a first pore structure parameter; the fluid parameters include: a first fluid parameter; the horizontal well injection model includes an artificial fracture system; and the multi-scale seepage control model corresponding to the horizontal well injection model is established based on the reservoir parameters and the fluid parameters, including: Based on the first pore structure parameters and the first fluid parameters, a multi-scale seepage control model corresponding to the artificial fracture system is determined; wherein the first pore structure parameters include: the permeability, artificial fracture width and reservoir thickness of the artificial fracture system; the first fluid parameters include: the gas viscosity and gas volume coefficient of the artificial fracture system.

9. The method according to claim 1, characterized in that The reservoir parameters further include: a second pore structure parameter and a third pore structure parameter; the fluid parameters further include: a second fluid parameter and a third fluid parameter; the horizontal well injection model further includes a natural fracture system; and establishing a multi-scale seepage control model corresponding to the horizontal well injection model based on the reservoir parameters and the fluid parameters includes: determining a multi-scale seepage control model corresponding to the natural fracture system according to the second pore structure parameter, the third pore structure parameter, the second fluid parameter, and the third fluid parameter; Among them, the second pore structure parameters include: the radius of the natural fracture system, the permeability of the natural fracture system, the stress sensitivity modulus of the natural fracture, the reservoir pressure of the natural fracture system, the interface radius between the fracture system and the inorganic matrix system, the porosity of the natural fracture system and the reservoir temperature of the natural fracture system; the third pore structure parameters include: the shape factor of the inorganic matrix system, the permeability of the inorganic matrix system, the reservoir pressure of the inorganic matrix system and the radius of the inorganic matrix system; the second fluid parameters include: the gas density in the natural fracture system, the gas viscosity of the natural fracture system, the gas compressibility coefficient, the compressibility factor and the ideal gas constant; the third fluid parameters include: the gas density of the inorganic matrix system.

10. The method according to claim 9, characterized in that The reservoir parameters further include: a fourth pore structure parameter; the fluid parameters further include: a fourth fluid parameter; the horizontal well injection model further includes an inorganic matrix system; the multi-scale seepage control model corresponding to the horizontal well injection model is established based on the reservoir parameters and the fluid parameters, including: Determining a multi-scale seepage control model corresponding to the inorganic matrix system according to the third pore structure parameter, the fourth pore structure parameter, the third fluid parameter, the fourth fluid parameter, and the preset first adsorption model parameter; Among them, the third pore structure parameter also includes: clay mineral content, interface radius between kerogen system and inorganic matrix system, and porosity of inorganic matrix system; the fourth pore structure parameter includes: kerogen content, shape factor of kerogen system, permeability of kerogen system, reservoir pressure of kerogen system and radius of kerogen system; the third fluid parameter also includes: gas slip coefficient in inorganic matrix system and gas compressibility coefficient in inorganic matrix system; the fourth fluid parameter includes: gas density in kerogen system; the first adsorption model parameter includes: isothermal adsorption volume of inorganic matrix system and isothermal adsorption pressure of inorganic matrix system.

11. The method according to claim 10, characterized in that The reservoir parameters further include: a fifth pore structure parameter; the fluid parameters further include: a fifth fluid parameter; the horizontal well injection model further includes a porous kerogen system; the multi-scale seepage control model corresponding to the horizontal well injection model is established based on the reservoir parameters and the fluid parameters, including: Determining a multi-scale seepage control model corresponding to the porous kerogen system according to the fourth pore structure parameter, the fifth pore structure parameter, the fourth fluid parameter, the fifth fluid parameter, and the preset second adsorption model parameter; Among them, the fourth pore structure parameter also includes: the porosity of the kerogen system; the fifth pore structure parameter includes: the shape factor of the organic matter system and the particle radius of the organic matter system; the fourth fluid parameter also includes: the gas slip coefficient of the kerogen system and the gas compressibility coefficient of the kerogen system; the fifth fluid parameter includes: the gas density in the organic matter system, the effective diffusion coefficient of the gas and the gas concentration in the organic matter system; the second adsorption model parameter includes: the isothermal adsorption volume of the kerogen system and the isothermal adsorption pressure of the kerogen system.

12. The method according to claim 11, characterized in that The horizontal well injection model further includes: the organic matter system; the multi-scale seepage control model corresponding to the horizontal well injection model is established according to the reservoir parameters and the fluid parameters, including: A multi-scale seepage control model corresponding to the organic matter system is determined according to the fifth pore structure parameter and the fifth fluid parameter.

13. A carbon dioxide storage prediction device based on mechanism and data fusion drive, characterized in that: include: A parameter acquisition module, used to obtain reservoir parameters, well parameters and fluid parameters of the shale reservoir; A model building module, configured to establish a multi-scale seepage control model based on the reservoir parameters, the well parameters, and the fluid parameters; A model fusion module is used to embed the multi-scale seepage control model into a pre-trained carbon dioxide storage deep learning model to obtain a shale reservoir carbon dioxide storage prediction model; A prediction output module, configured to perform storage prediction on the reservoir parameters, the well parameters, and the fluid parameters using the shale reservoir carbon dioxide storage prediction model to obtain a prediction result; The model building module includes: a horizontal well injection model building unit, configured to build a horizontal well injection model according to the reservoir parameters, the well parameters, and the fluid parameters; A seepage mechanism modeling unit, configured to establish a multi-scale seepage control model corresponding to the horizontal well injection model according to the reservoir parameters and the fluid parameters; The model fusion module includes: A physical model discretization unit is used to discretize the multi-scale seepage control model to obtain a multi-scale seepage control model in a differential form; A physical loss acquisition unit, configured to input high-dimensional prediction data into the differential multi-scale seepage control model to obtain a physical residual loss; a boundary loss calculation unit, configured to determine data fitting loss, boundary condition residuals, and initial condition residuals, respectively, based on the high-dimensional prediction data; A total loss function construction unit is used to determine a total loss function based on the physical residual loss, data fitting loss, boundary condition residual and initial condition residual and their corresponding weight factors; The optimization training unit is used to update the model parameters using the Adam optimizer and the total loss function, and repeat the above steps to perform iterative training until the total loss function converges, thereby obtaining the shale reservoir carbon dioxide sequestration prediction model.

14. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 12 is implemented.

15. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

16. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 12 is implemented.

Images