ScCO2 geological sequestration multi-mode safety evaluation system and method

Abstract

The invention discloses an scCO2 geological sequestration multi-mode safety assessment system and method, and relates to the technical field of carbon dioxide geological sequestration safety monitoring and risk assessment. The scCO2 geological sequestration multi-modal safety evaluation system mainly comprises a numerical simulation construction module, a multi-source monitoring data acquisition and analysis module, a simulation spatial feature coding module, a monitoring data time feature coding module, a cross-modal attention feature fusion module, a physical constraint and mechanical degradation modeling module and a safety state classification and early warning module. By implementing the scCO2 geological sequestration multi-mode safety evaluation system and method provided by the invention, accurate identification and early warning of the scCO2 geological sequestration safety state can be realized.

Description

Technical Field

[0001] This invention relates to the field of carbon dioxide geological storage safety monitoring and risk assessment technology, and more specifically, to a multimodal safety assessment system and method for scCO2 geological storage. Background Technology

[0002] scCO2 geological storage is one of the important technological pathways for achieving large-scale carbon emission reduction. In the process of scCO2 geological storage, the carbon source often comes from production industries such as factories (e.g., power plants). The carbon dioxide is captured and converted into liquid form, then injected deep into sealed geological structures, such as saline aquifers, depleted oil and gas fields, or unminable coal seams, to limit its release into the atmosphere. Although this method is considered the most feasible scCO2 geological storage method, it carries the following risks: Injecting large amounts of high-pressure liquid carbon dioxide underground will subject the reservoir-caprock system to multiple coupled physical and chemical processes, including pore pressure changes, effective stress redistribution, fracture initiation and propagation, permeability evolution, and scCO2-coal-rock-water interactions. This can disrupt the reservoir's mechanical balance and easily induce caprock failure or CO2 leakage risks. Therefore, assessing the safety of scCO2 geological storage is essential. Existing technologies for assessing the safety of scCO2 geological storage mainly include threshold criteria methods based on numerical simulation, empirical or statistical analysis methods based on monitoring data, and data-driven methods based on machine learning. These methods generally have the following shortcomings: (1) Reliance on a single information source makes it difficult to simultaneously reflect underground physical evolution and monitoring response; (2) Safety thresholds are mostly fixed and do not take into account the degradation of mechanical parameters caused by scCO2 dissolution; (3) Pure data-driven methods lack physical constraints and there is a risk that the prediction results are inconsistent with the basic physical laws.

[0003] How to accurately identify and provide early warning of the safety status of scCO2 geological storage is an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a multimodal safety assessment system and method for scCO2 geological storage, which can accurately identify and provide early warning of the safety status of scCO2 geological storage.

[0005] This invention provides a multimodal safety assessment system for scCO2 geological storage, comprising a numerical simulation construction module, a multi-source monitoring data acquisition and analysis module, a simulation spatial feature encoding module, a monitoring data temporal feature encoding module, a cross-modal attention feature fusion module, a physical constraint and mechanical degradation modeling module, and a safety status classification and early warning module. The numerical simulation construction module is used to simulate the coupled process of fluid flow-geomechanics-fracture evolution using numerical simulation methods, constructing multi-channel simulation images and simulation detection data to reflect the strong coupling relationship between underground pore pressure, stress, permeability, and CO2 distribution. The multi-source monitoring data acquisition and analysis module is used to collect field monitoring data and organize it into a multivariate time series, capturing dynamic response signals during the CO2 storage process to provide real-time monitoring data support for safety assessment. The simulation spatial feature encoding module is used to extract spatial features from the multi-channel simulation images to obtain potential spatial feature representations to characterize pressure concentration areas, high-stress areas, and fractures. The key spatial characteristics of the evolution of the guide zone and CO2 plume are described. The monitoring data time feature encoding module is used to process the multivariate time series, using a self-attention mechanism to capture long-term time dependence, abrupt changes, and potential leakage precursor information to obtain a time feature representation. The cross-modal attention feature fusion module is used to learn the correlation between the monitoring signal and a specific spatial state based on the potential spatial feature representation and the time feature representation, using a cross-modal attention mechanism to obtain a comprehensive feature representation that integrates multimodal information. The physical constraint and mechanical degradation modeling module is used to quantify the impact of CO2 dissolution on reservoir mechanical parameters, correct the mechanical degradation effect and safety threshold in safety assessment, and guide the model prediction to conform to basic physical laws using pressure-stress consistency constraints, permeability evolution constraints, CO2 mass conservation constraints, and total loss function. The safety status classification and early warning module is used to identify the safety status and provide risk warnings for CO2 storage sites based on the joint feature representation of the integrated multimodal information.

[0006] This invention also provides a multimodal safety assessment method for scCO2 geological storage, which utilizes the aforementioned multimodal safety assessment system for scCO2 geological storage to conduct safety monitoring and risk assessment of carbon dioxide geological storage.

[0007] Implementing the multimodal safety assessment system and method for scCO2 geological storage provided by this invention has the following beneficial effects: This invention constructs a multimodal safety assessment system for scCO2 geological storage, comprising a numerical simulation construction module, a multi-source monitoring data acquisition and analysis module, a simulation spatial feature encoding module, a monitoring data temporal feature encoding module, a cross-modal attention feature fusion module, a physical constraint and mechanical degradation modeling module, and a safety status classification and early warning module. The numerical simulation construction module uses numerical simulation methods to simulate the coupled process of fluid flow, geomechanics, and fracture evolution, constructing multi-channel simulation images and simulated detection data to reflect the strong coupling relationship between underground pore pressure, stress, permeability, and CO2 distribution. The multi-source monitoring data acquisition and analysis module collects field monitoring data and organizes it into multivariate time series, capturing dynamic response signals during CO2 storage to provide real-time monitoring data support for safety assessment. The simulation spatial feature encoding module extracts spatial features from the multi-channel simulation images to obtain potential spatial feature representations, characterizing pressure concentration areas, high-stress areas, and fracture zones. The system identifies key spatial characteristics of the flow channel and CO2 plume evolution; a monitoring data temporal feature encoding module processes multivariate time series data, using a self-attention mechanism to capture long-term time dependence, abrupt anomalies, and potential leak precursor information to obtain temporal feature representations; a cross-modal attention feature fusion module learns the correlation between monitoring signals and specific spatial states based on potential spatial feature representations and temporal feature representations, using a cross-modal attention mechanism to obtain a comprehensive feature representation that integrates multimodal information; a physical constraint and mechanical degradation modeling module quantifies the impact of CO2 dissolution on reservoir mechanical parameters, corrects mechanical degradation effects and safety thresholds in safety assessments, and uses pressure-stress consistency constraints, permeability evolution constraints, and CO2 mass conservation constraints, using a total loss function to guide model predictions to conform to basic physical laws; and a safety status classification and early warning module classifies CO2 storage sites according to the joint feature representation of integrated multimodal information to achieve safety status identification and risk early warning. In summary, this invention proposes a multimodal intelligent safety assessment framework for scCO2 geological storage that integrates numerical simulation fields, multi-source monitoring data, and physical constraints. By integrating numerical simulations, multi-source monitoring data, and introducing physical constraints, a multimodal joint representation of numerical simulation fields and multi-source monitoring data is constructed. Physical constraints and a scCO2 dissolution-induced mechanical degradation model are introduced during the training process of the intelligent model to achieve accurate identification and early warning of the safety status of scCO2 geological storage. This framework is applicable to the field of safety monitoring and risk assessment of CO2 geological storage, and can be used for safety status identification, risk warning, and engineering decision support in supercritical carbon dioxide (scCO2) geological storage projects in deep saline aquifers, depleted oil and gas reservoirs, and coal seams. Attached Figure Description

[0008] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a block diagram of the multimodal safety assessment system for scCO2 geological storage provided by the present invention; Figure 2 This is the overall architecture diagram of the scCO2 geological storage multimodal safety assessment system provided by the present invention; Figure 3 This is a flowchart of the formation reconstruction process provided by the present invention; Figure 4 This is a schematic diagram of well logging data provided by the present invention; Figure 5 This is a three-dimensional bottom-layer schematic diagram provided by the present invention; Figure 6 This is a flowchart of the simulation solution provided by the present invention; Figure 7 This is a flowchart of the simulation detection data correction process provided by the present invention; Figure 8 This is a schematic diagram of the multi-channel simulated image provided by the present invention; Figure 9 This is a schematic diagram of monitoring data from the surface station 1 provided by the present invention; Figure 10 This is a schematic diagram of monitoring data from surface station 2 provided by the present invention; Figure 11 This is a schematic diagram of monitoring data after well closure at the injection wellhead provided by the present invention; Figure 12 This is a schematic diagram of multi-channel tensor stacking provided by the present invention; Figure 13 This is a block diagram of the cross-modal attention feature fusion module provided by the present invention; Figure 14 This is a schematic diagram of the degradation effect provided by the present invention. Detailed Implementation

[0009] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0010] Figure 1A schematic diagram of the scCO2 geological storage multimodal safety assessment system of this embodiment is shown. In this embodiment, the scCO2 geological storage multimodal safety assessment system includes: a numerical simulation construction module, a multi-source monitoring data acquisition and analysis module, a simulation spatial feature encoding module, a monitoring data temporal feature encoding module, a cross-modal attention feature fusion module, a physical constraint and mechanical degradation modeling module, and a safety status classification and early warning module. The numerical simulation construction module is used to simulate the coupled process of fluid flow-geomechanics-fracture evolution using numerical simulation methods, constructing multi-channel simulation images and simulation detection data to reflect the strong coupling relationship between underground pore pressure, stress, permeability, and CO2 distribution. The multi-source monitoring data acquisition and analysis module is used to collect field monitoring data and organize it into a multivariate time series, capturing dynamic response signals during the CO2 storage process to provide real-time monitoring data support for safety assessment. The simulation spatial feature encoding module is used to extract spatial features from the multi-channel simulation images to obtain potential spatial feature representations to characterize pressure concentration areas, high-stress areas, and fractures. The key spatial characteristics of the evolution of the guide zone and CO2 plume are described. The monitoring data time feature encoding module is used to process the multivariate time series, using a self-attention mechanism to capture long-term time dependence, abrupt changes, and potential leakage precursor information to obtain a time feature representation. The cross-modal attention feature fusion module is used to learn the correlation between the monitoring signal and a specific spatial state based on the potential spatial feature representation and the time feature representation, using a cross-modal attention mechanism to obtain a comprehensive feature representation that integrates multimodal information. The physical constraint and mechanical degradation modeling module is used to quantify the impact of CO2 dissolution on reservoir mechanical parameters, correct the mechanical degradation effect and safety threshold in safety assessment, and guide the model prediction to conform to basic physical laws using pressure-stress consistency constraints, permeability evolution constraints, CO2 mass conservation constraints, and total loss function. The safety status classification and early warning module is used to identify the safety status and provide risk warnings for CO2 storage sites based on the joint feature representation of the integrated multimodal information.

[0011] In one exemplary embodiment, the simulated detection data is used to correct multi-channel simulated images and complete multivariate time series.

[0012] In one exemplary embodiment, the field monitoring data includes CO2 concentration from monitoring wells, CO2 concentration from ground stations, and microseismic waveform-derived attributes.

[0013] In one exemplary embodiment, the simulated spatial feature encoding module includes a multi-channel convolutional neural network.

[0014] In one exemplary embodiment, the monitoring data temporal feature encoding module includes multiple Transformer encoder layers and a self-attention mechanism.

[0015] In one exemplary embodiment, the cross-modal attention feature fusion module includes a multimodal feature input unit, a modal projection and alignment unit, a cross-modal attention calculation unit, and a fused feature output unit. The multimodal feature input unit is used to receive temporal feature representations of multiple modal data. The modal projection and alignment unit is used to project the temporal feature representations onto a unified feature space using a linear mapping. The cross-modal attention calculation unit is used to calculate cross-modal attention weights by using the target modality as a query and the other auxiliary modalities as keys and values ​​to obtain a comprehensive feature representation that fuses multimodal information. The fused feature output unit is used to output the comprehensive feature representation that fuses multimodal information.

[0016] In one exemplary embodiment, the total loss function is specifically: , , , , , , in, This is the total loss function; The cross-entropy loss function; This is due to pressure-stress consistency loss; This represents the loss due to the evolution of penetration rate. For the constraint of mass conservation; For the loss clause of the CO2 dissolution-induced mechanical degradation model; , , , The weights are respectively for the loss clauses of pressure-stress consistency loss, permeability evolution loss, mass conservation constraint, and CO2 dissolution-induced mechanical degradation model; For fluid pressure; This is the maximum principal stress; The critical pore pressure; Tensile strength; It is an L2 norm; This ensures that the permeability gradient is a non-negative function; This represents the partial derivative of CO2 concentration with respect to time. This represents the net outflow rate of CO2 mass or volume fraction per unit volume due to fluid transport; N is the sample size; C is the number of safety levels. This represents a true state of security. Predict samples for the model i Categoryj The probability of; To take chemical dissolution into account t At any given moment, the tensile strength is maintained. To take chemical dissolution into account t Tensile strength at -1 time; The chemical dissolution coefficient; Let t be the CO2 concentration at time t.

[0017] This embodiment provides a multimodal safety assessment method for scCO2 geological storage, which utilizes the aforementioned multimodal safety assessment system for scCO2 geological storage to conduct safety monitoring and risk assessment of carbon dioxide geological storage.

[0018] In some embodiments, the above-described scCO2 geological storage multimodal safety assessment system can also be implemented in the following ways.

[0019] like Figure 2 The diagram shows the overall architecture of the scCO2 geological storage multimodal safety assessment system. In this embodiment, the specific contents of the scCO2 geological storage multimodal safety assessment system are as follows.

[0020] 1. Overall Technical Solution The multimodal safety assessment framework for scCO2 geological storage of the present invention includes a numerical simulation construction module, a multi-source monitoring data acquisition and analysis module, a simulation spatial feature encoding module, a monitoring data temporal feature encoding module, a cross-modal attention feature fusion module, a physical constraint and mechanical degradation modeling module, and a safety status classification and early warning module. The modules work together to achieve a comprehensive assessment of the safety status of scCO2 geological storage.

[0021] 2. Numerical Simulation Construction Module Formation reconstruction refers to the process of using geological, geophysical, and well logging data, along with computer modeling techniques, to quantitatively reconstruct and visualize the structure, properties, and spatial distribution of underground rock strata in three dimensions. In carbon sequestration simulation, formation reconstruction can accurately characterize the geometry, porosity, permeability, and caprock features of the target sequestration layer (such as saline aquifers and abandoned oil and gas reservoirs). This provides a reliable geological model basis for simulating the migration path, pressure evolution, storage safety, and long-term sequestration effect of carbon dioxide after injection, and is a key support for assessing leakage risks.

[0022] Based on site geological data, including drilling data, core data, and well logging curves, the drilling data includes the well location coordinates, total well depth, encountered formations, and well inclination data for the coalbed methane wells. The core data primarily consists of coal samples from parameter wells. gas content Adsorption time, Langevin volume, Randomization pressure, porosity, and density, among other parameters, are used in well logging. The main data collected include longitudinal wave sonic transit time (AC), density (DEN), natural gamma ray (GR), spontaneous potential (SP), borehole diameter (CAL), deep dual lateral resistivity (RD), shallow dual lateral resistivity (RS), and temperature (TEMP). This data provides the foundation for establishing three-dimensional heterogeneous geological models of coal reservoirs. Figure 3 The diagram shown is a flowchart of the formation reconstruction process.

[0023] Stratigraphic reconstruction steps: (1) Determine the coordinates of the inflection points in the reconstructed region; (2) Collect geological borehole data of the study area, clarify the stratigraphic properties, and determine the target stratigraphy and sub-target stratigraphy (determine whether a sub-target stratigraphy exists based on research needs). (3) If it is necessary to establish surface data, use a total station to acquire surface data; (4) Collect gravity, magnetic or ground-penetrating radar data of the study area and perform inversion to obtain stratigraphic data; (5) Import all data into the software, use interpolation to complete the data of the unmonitored areas, and draw the stratigraphic model.

[0024] like Figure 4 The image shown is a schematic diagram of well logging data. Figure 5 This is a schematic diagram of the three-dimensional bottom layer.

[0025] The safety of deep CO2 geological reservoirs is governed by a coupled process of fluid flow, geomechanics, and fracture evolution. Injection of scCO2 alters pore pressure, effective stress, and permeability, leading to fracture initiation, fracture propagation, and ultimately scCO2 leakage.

[0026] The basic control relationship can be summarized as follows: (1) Conservation of fluid mass: , (2) Darcy's Law: , (3) Effective stress principle: , (4) Fracture-controlled permeability evolution: , First, geostress equilibrium calculations are performed to obtain the initial geostress field. This initial stress field is used as a boundary condition or initial field input and is simultaneously coupled with the mass transfer process of the surrounding rock and coal seam fluids to obtain the initial stress distribution and flow field under fluid-solid coupling.

[0027] Subsequently, the displacement field u is solved in the mechanics module. This displacement field allows for the further calculation of the unknown principal strains and their directions. Next, the pressure P is calculated using the seepage module, and the CO2 volume fraction is determined using the mass transfer module.

[0028] Finally, the permeability change is calculated based on the pressure P, and the adsorption expansion force and fluid pressure are transmitted to the mechanical module.

[0029] To achieve multi-field coupled calculations, it is necessary to organically couple the reservoir-capsule fluid mass transfer process with the geostress equilibrium process in order to solve for their respective unknowns. For example... Figure 6 As shown, the specific implementation steps are as follows.

[0030] First, all physical field variables, including displacement and pore pressure fields, are spatially discretized. To simplify the solution of multi-field coupling, a staggered scheme is used for iterative solving, where each physical field variable is solved as an independent system of equations in each iteration. For time discretization, all differential equations employ implicit generalized... The method employs integration to ensure unconditional stability of the time integral, maintaining numerical stability even at large time steps. Each time step contains several iterations, and the relative errors of each physics field are calculated after each iteration to evaluate convergence. To accelerate iterative convergence, Anderson acceleration is used, effectively reducing the number of iterations and improving computational efficiency.

[0031] Numerical simulation data may become distorted over time due to assumptions made in the model and approximate input of mechanical / seepage parameters. Therefore, it is necessary to compare the simulation data with monitoring data during the simulation process. When the deviation between the simulated data and the actual monitoring data exceeds a threshold, the simulation parameters should be modified based on the new monitoring data.

[0032] Simulated detection data can be used to correct multi-channel simulated images and complete multivariate time series. For example... Figure 7 The diagram shown is a flowchart of the simulation test data correction process.

[0033] By using equations to strongly couple fluid pressure, stress, permeability, and CO2 distribution, a joint learning framework can be formed, and uniform interpolation can be performed to a regular grid to construct a multi-channel simulation image. For example... Figure 8 The image shown is a schematic diagram of a multi-channel simulation, representing fluid pressure, stress, permeability, and CO2 distribution from left to right.

[0034] 3. Multi-source monitoring data (acquisition and analysis) representation module The field monitoring data includes: CO2 concentration in monitoring wells, CO2 concentration at ground stations, and microseismic waveform derived attributes (such as event energy, dominant frequency, and b-value).

[0035] Data point layout requirements: (1) Layout of injection and production wells; (2) Layout in locations where CO2 fluid tends to accumulate, such as regional uplifts; (3) Layout on the main migration path, such as the highest point in the study area or the highest elevation in the region; (4) Layout in the direction of reservoir permeability where the topographic relief is not obvious; (5) If there is a fault in the study area, layout near the fault; if there is a fault outside the study area, layout at the location closest to the fault within the area; (6) Layout in areas with weak caprock. Figure 9 This represents monitoring data from surface station 1. Figure 10 This indicates monitoring data from surface station 2. Figure 11 This indicates monitoring data after the injection wellhead is closed. Figure 12 This is a schematic diagram of multichannel tensor stacking; these measurements are organized into a multivariate time series, as shown in the following equation: .

[0036] 4. Simulated Spatial Feature Encoding Module To extract spatial features from numerical simulations, a multi-channel convolutional neural network (CNN) was employed. Given an input simulation image... The CNN encoder produces a latent feature representation: , The extracted features are used to characterize the pressure concentration zone, high stress zone, fracture flow zone, and CO2 plume evolution characteristics.

[0037] (1) CNN convolution feature extraction formula 1) Multi-channel input definition Let the first The physical field inputs are: , The multimodal input tensor is represented as: .

[0038] 2) Convolution operation No. l Layer j The convolution of each feature map is calculated as follows: , in, For convolution kernel; For bias terms; This is a convolution operation; It is a non-linear activation function.

[0039] (2) Feature vectorization Flatten the feature map output by the CNN as follows: .

[0040] 5. Monitoring data time feature coding module The monitoring data exhibits a strong time dependence and delayed response to underground processes. To effectively simulate long-range temporal correlation and abrupt anomalies, a Transformer-based encoder was employed. Each monitoring vector x... t .

[0041] First, it is projected into a high-dimensional embedding space: Utilizing self-attention mechanisms to capture information about long-term dependencies, mutational anomalies, and potential leak precursors: , Then, multi-head self-attention is used to process the embedded sequence: , By stacking multiple Transformer encoder layers, a compact temporal feature representation is obtained: .

[0042] 6. Cross-modal attention feature fusion module In the process of monitoring surface CO2 concentration and assessing the safety of underground CO2 storage, the system typically acquires multiple heterogeneous monitoring data simultaneously, such as: surface CO2 concentration time series; engineering parameters such as injection pressure and pore pressure; geophysical information such as the energy and frequency of microseismic events; and environmental background data such as temperature and humidity. These multi-source data exhibit significant differences in dimensions, sampling frequency, physical meaning, and noise characteristics. Traditional feature splicing or weighted averaging methods are insufficient to effectively characterize the dynamic relationships between different modes and can easily obscure key precursory information during anomaly evolution. This paper proposes a cross-modal attention feature fusion module to achieve adaptive correlation modeling and weight allocation of multimodal information at the feature level.

[0043] The cross-modal attention feature fusion module is located after the Transformer encoder. Its overall structure includes a multimodal feature input unit, a modal projection and alignment unit, a cross-modal attention calculation unit, and a fused feature output unit. Figure 13 As shown, its function is to dynamically measure the influence of different modal features on the target modality (such as CO2 concentration evolution) based on the attention mechanism, and generate a unified fusion feature representation.

[0044] Input feature format: Assume the system simultaneously acquires M modal data (M=3), and after processing by their respective Transformer encoders, the corresponding time feature representations are obtained: , in, (1) This represents the temporal characteristics of surface CO2 concentration. (2) Features related to pressure or stress; (3) This is a microseismic or energy characteristic.

[0045] Modal projection and feature alignment: Since the feature dimensions of different modalities may be inconsistent, they are first projected onto a unified feature space through a linear mapping: , in: ( ) Let m be the projection matrix of mode m; the feature dimensions are unified after projection. .

[0046] Cross-modal attention calculation mechanism: Regarding attention role allocation, in this embodiment, the target modality (e.g., surface CO2 concentration features) is used as the Query, and the remaining auxiliary modalities are used as the Key and Value. , The formula for calculating cross-modal attention weights is: , in: This represents the contribution weight of mode m to the target mode at time t.

[0047] The comprehensive feature representation that integrates multimodal information is defined as: , in: This represents the integrated time characteristics after cross-modal fusion.

[0048] Therefore, this embodiment proposes a cross-modal attention feature fusion module to achieve adaptive association modeling and weight allocation of multimodal information at the feature level.

[0049] From the perspective of complementarity between numerical simulation and monitoring data, a deep interaction between modalities is established, and a cross-modal attention mechanism is introduced.

[0050] Specifically, the simulation function acts as the query, while the monitoring function acts as the key and value: , This mechanism enables the model to automatically learn the correlation between monitoring signals and specific spatial states, such as whether a sudden increase in CO2 concentration is related to enhanced permeability pathways in high-stress areas. (Fusing features) Linking underground processes with surface and wellbore observations.

[0051] 7. CO2 Dissolution-Induced Mechanical Degradation Model During CO2 sequestration, supercritical or dissolved CO2 reacts with formation water to form a weakly acidic fluid (primarily carbonic acid), which undergoes a dissolution reaction with reservoir and caprock minerals, leading to the dissolution or weakening of pore walls and cement. This process increases porosity, connectivity, and effective permeability, altering the seepage channel structure; on the other hand, it weakens the rock skeleton and intergranular cementation strength, causing a decrease in elastic modulus, strength, and shear resistance, and may even induce fracture propagation or reopening in some areas. The coupling of mechanical degradation and seepage enhancement results in a synergistic change characteristic of "structural weakening—seepage enhancement" in the sequestration system during long-term evolution.

[0052] The CO2 dissolution-induced mechanical degradation model can dynamically adjust the threshold based on the input simulation parameters.

[0053] In the process of CO2 geological sequestration and dissolution, the chemical reactions involved can be mainly categorized into two types: CO2-water system reactions and mineral dissolution / precipitation reactions. Typical reaction formulas are as follows: (1) The dissolution and ionization reaction of CO2 in water , , , (2) Dissolution reaction of carbonate minerals calcite ( ): , or , dolomite( ): .

[0054] (3) Dissolution reaction of silicate minerals Feldspars (taking potassium feldspar as an example): , Calcium plagioclase (simplified representation): .

[0055] (4) Secondary mineral precipitation When the solution is supersaturated, redeposition may occur: , , After CO2 is injected, it forms a carbonic acid system through hydration and ionization, preferentially dissolving carbonate minerals and slowly acting on silicate minerals, leading to the evolution of pore structure and mechanical weakening. At the same time, on a long-term scale, it may be accompanied by carbonate redeposition, realizing a dynamic coupling process of dissolution and precipitation.

[0056] The following methods were used to quantitatively determine the dissolution effect: Initial characterization of the physical properties and structure (porosity, permeability, wave velocity, CT structure) of core samples taken from the field was performed. Subsequently, CO2-formation water immersion tests were conducted under high temperature and high pressure conditions, controlling temperature, pressure, CO2 partial pressure, and treatment time to obtain different degrees of dissolution. Uniaxial / triaxial compression, Brazilian fracturing, and steady-state or unsteady-state seepage tests were conducted before and after immersion to compare changes in elastic modulus, peak strength, Poisson's ratio, and permeability. Simultaneously, the mass loss rate and ion concentration were considered. , Evolution of pH and changes in CT pore structure are used as quantitative indicators of the degree of dissolution. A quantitative response model of the degree of dissolution to core mechanical deterioration and enhanced seepage is established by establishing the correspondence between "dissolution index - mechanical parameters - seepage parameters". Figure 14 The diagram shown illustrates the degradation effect.

[0057] Through the degradation model: , Definition of loss clauses: .

[0058] 8. Physical Constraints and Physical Guidance Loss Module Construct the following physical constraint loss term: While data-driven multimodal models can effectively capture complex patterns, purely statistical learning can lead to predictions that are inconsistent with physical laws, especially when monitoring data is limited or extrapolation analysis is performed. In CO2 geological storage, safety assessments must adhere to fundamental hydraulic principles, including mass conservation, stress-dependent permeability evolution, and fracture initiation criteria.

[0059] Incorporating physical guidance constraints into the training process guides the model to solve problems in a way that conforms to physical laws, while maintaining the flexibility of data-driven approaches.

[0060] (1) Pressure-stress consistency constraint The increased pore pressure caused by injection alters the effective stress state of the reservoir and caprock. According to the effective stress principle: , To ensure consistency between the predicted safe state and the simulated stress-pressure coupling, a stress-stress consistency loss is defined: , This function is designed to predict and prevent high-pressure risks without corresponding stress responses, and vice versa, to ensure mechanical consistency.

[0061] (2) Constraints on the evolution of penetration rate The activation and propagation of fractures lead to an irreversible increase in permeability, affecting CO2 transport pathways. To enhance the monotonic permeability behavior after fracture initiation, a permeability evolution loss is introduced: , Once an increase in permeability occurs due to a crack, this constraint prevents the model permeability from decreasing, thus aligning the prediction with observations of crack mechanics.

[0062] (3) CO2 mass conservation constraint CO2 leakage risk is directly related to mass transport behavior. To ensure consistency between the predicted risk and CO2 distribution, a mass conservation constraint is applied: , This function ensures that the theoretical CO2 migration trend matches the simulation inference.

[0063] Based on fused multimodal features, a fully connected classification method is used to determine the safety status of CO2 storage sites: , Three safety states were considered: safe, warning, and failure.

[0064] The model is trained using the cross-entropy loss function: , The final training objective, which combines the physical information objective function, integrates data-driven classification loss and physical information constraints: , The selection coefficients were validated through experiments to balance prediction accuracy and physical consistency.

[0065] 9. Safety Status Classification (and Early Warning) Module In practical CO2 geological storage projects, there are typically no direct ground-based true value labels to guarantee storage safety. Therefore, safety labels are constructed by integrating numerical simulation results, monitoring responses, and engineering threshold standards, forming a physically-informed labeling strategy. A comprehensive risk index is defined. is, quantifies the storage security status at each time step: , , in, , , , The weights are respectively for fluid pressure, stress, permeability, and CO2 concentration variables; Let t be the permeability. This represents the initial penetration rate.

[0066] The weighting coefficients reflect the relative importance of hydraulic, mechanical, and transport risks and can be determined based on expert analysis or sensitivity analysis.

[0067] According to the risk index The storage system is divided into three security states: , A safe state corresponds to stable storage conditions with no significant pressure build-up or leakage risk. A warning state indicates potential instability or early leakage, requiring enhanced monitoring. A failure state represents a high-risk condition associated with crack propagation, increased permeability, or abnormal CO2 migration.

[0068] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A multimodal safety assessment system for scCO2 geological storage, characterized in that, The system includes a numerical simulation construction module, a multi-source monitoring data acquisition and analysis module, a simulation spatial feature encoding module, a monitoring data temporal feature encoding module, a cross-modal attention feature fusion module, a physical constraint and mechanical degradation modeling module, and a safety status classification and early warning module. The numerical simulation construction module is used to simulate the coupled process of fluid flow-geomechanics-fracture evolution using numerical simulation methods, constructing multi-channel simulation images and simulated detection data. The multi-source monitoring data acquisition and analysis module is used to collect field monitoring data and organize it into a multivariate time series, capturing dynamic response signals during CO2 sequestration. The simulation spatial feature encoding module is used to extract spatial features from the multi-channel simulation images to obtain latent spatial feature representations. The monitoring data temporal feature encoding module is used to process the multivariate time series, utilizing self-attention... The force mechanism captures long-term time dependence, abrupt changes, and potential leak precursor information to obtain a temporal feature representation. The cross-modal attention feature fusion module is used to learn the correlation between monitoring signals and specific spatial states based on the potential spatial feature representation and the temporal feature representation, using the cross-modal attention mechanism to obtain a comprehensive feature representation that integrates multimodal information. The physical constraint and mechanical degradation modeling module is used to quantify the impact of CO2 dissolution on reservoir mechanical parameters, correct the mechanical degradation effect and safety threshold in safety assessment, and guide the model prediction to conform to basic physical laws using pressure-stress consistency constraints, permeability evolution constraints, CO2 mass conservation constraints, and total loss function. The safety status classification and early warning module is used to identify the safety status and provide risk warnings for CO2 storage sites based on the joint feature representation of the integrated multimodal information.

2. The multimodal safety assessment system for scCO2 geological storage according to claim 1, characterized in that, The simulated detection data is used to correct multi-channel simulated images and complete multivariate time series.

3. The multimodal safety assessment system for scCO2 geological storage according to claim 1, characterized in that, The on-site monitoring data includes CO2 concentration from monitoring wells, CO2 concentration from ground stations, and microseismic waveform derived attributes.

4. The multimodal safety assessment system for scCO2 geological storage according to claim 1, characterized in that, The simulated spatial feature encoding module includes a multi-channel convolutional neural network.

5. The multimodal safety assessment system for scCO2 geological storage according to claim 1, characterized in that, The monitoring data time feature encoding module includes multiple Transformer encoder layers and a self-attention mechanism.

6. The multimodal safety assessment system for scCO2 geological storage according to claim 1, characterized in that, The cross-modal attention feature fusion module includes a multimodal feature input unit, a modal projection and alignment unit, a cross-modal attention calculation unit, and a fused feature output unit. The multimodal feature input unit is used to receive temporal feature representations of multiple modal data. The modal projection and alignment unit is used to project the temporal feature representations onto a unified feature space using a linear mapping. The cross-modal attention calculation unit is used to calculate cross-modal attention weights by using the target modality as a query and the other auxiliary modalities as keys and values ​​to obtain a comprehensive feature representation that fuses multimodal information. The fused feature output unit is used to output the comprehensive feature representation that fuses multimodal information.

7. The multimodal safety assessment system for scCO2 geological storage according to claim 1, characterized in that, The total loss function is specifically as follows: in, This is the total loss function; The cross-entropy loss function; This is due to pressure-stress consistency loss; This represents the loss due to the evolution of penetration rate. For the constraint of mass conservation; For the loss clause of the CO2 dissolution-induced mechanical degradation model; , , , These are the weights of the loss clauses for pressure-stress consistency loss, permeability evolution loss, mass conservation constraint, and CO2 dissolution-induced mechanical degradation model, respectively. For fluid pressure; This is the maximum principal stress; The critical pore pressure; Tensile strength; It is an L2 norm; This ensures that the permeability gradient is a non-negative function; This represents the partial derivative of CO2 concentration with respect to time. This represents the net outflow rate of CO2 mass or volume fraction per unit volume due to fluid transport; N is the sample size; C is the number of safety levels. This represents a true state of security. Predicting samples for the model i Category j The probability of; To take chemical dissolution into account t At any given moment, the tensile strength is maintained. To take chemical dissolution into account t Tensile strength at -1 time; The chemical dissolution coefficient; Let t be the CO2 concentration at time t.

8. A multimodal safety assessment method for scCO2 geological storage, characterized in that, The scCO2 geological storage multimodal safety assessment system as described in any one of claims 1-7 is used to conduct safety monitoring and risk assessment of carbon dioxide geological storage.

Images