Monitoring and optimizing method for carbon sequestration and geothermal exploitation cooperative system

Abstract

The invention discloses a monitoring and optimizing method for a carbon sequestration and geothermal exploitation cooperative system, and belongs to the technical field of carbon dioxide geological sequestration and deep geothermal energy exploitation. The method is characterized by comprising the following steps: firstly, generating a multi-fidelity heat-flow-force-chemical coupling data set; a greedy algorithm based on multivariate information minimum redundancy is adopted to optimize and arrange a monitoring network, and monitoring points with high information value and low redundancy are screened out; a multi-mode double-attention residual network substitution model is constructed, sparse monitoring data, operation parameters and geological parameters are fused, and rapid and high-precision prediction of the system state is achieved; and finally, an improved bald eagle search algorithm is utilized, the substitution model serves as an evaluator, and iterative optimization is conducted on the injection pressure and the flow so as to collaboratively optimize the thermal exploitation efficiency and the carbon dioxide storage amount. According to the method, the maximum monitoring information amount is extracted with the minimum monitoring cost, and finally the efficiency and precision of system operation parameter optimization are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of carbon dioxide geological storage and deep geothermal energy extraction technology, and particularly relates to a monitoring and optimization method for a synergistic system of carbon storage and geothermal extraction. Background Technology

[0002] With the advancement of dual-carbon goals, CS-EGS technology, which combines geological carbon dioxide sequestration with enhanced geothermal systems, has become a research hotspot. This system utilizes supercritical carbon dioxide as the working fluid for cyclic heat recovery, while simultaneously sequestering carbon dioxide through formation mineral reactions. However, this process involves complex thermo-hydrodynamic multi-physics coupling effects, and the states of underground reservoirs, such as pressure, temperature, carbon dioxide concentration, and stress fields, exhibit highly nonlinear and spatiotemporal evolution characteristics. To ensure system safety and assess efficiency, a monitoring well network must be deployed. In this practical engineering application scenario, the existing technologies currently relied upon face three core challenges, which directly restrict the large-scale and intelligent application of the technology.

[0003] Firstly, in terms of monitoring network deployment, existing technologies mainly rely on engineering experience or uniform grids for point placement. This experience-based or uniform placement method lacks quantitative evaluation of the information value of monitoring points, leading to monitoring redundancy in non-critical areas with gentle changes in the physical field, resulting in a waste of high drilling costs. Meanwhile, in critical change areas such as high seepage channels and stress concentration zones, insufficient point placement may result in incomplete information capture, failing to meet the engineering requirement of obtaining the maximum amount of information at the lowest cost.

[0004] Secondly, existing technical solutions have significant limitations in system simulation and prediction. One approach is to directly optimize parameters based on a coupled thermo-fluid-mechanical-chemical multiphysics model. Its core drawback is extremely low computational efficiency; a single high-fidelity simulation takes too long, making it unable to support the millions of iterations required by optimization algorithms and failing to meet the demands of real-time or near-real-time optimization. Another approach uses simple regression models or shallow machine learning models such as support vector machines as alternatives. However, its limitations include insufficient model capabilities, only able to handle low-dimensional data, losing crucial spatial correlations and nonlinear correlations between data in multiphysics data, and poor prediction accuracy with small-sample high-fidelity data, failing to reliably capture the nonlinear mapping relationship between injected parameters, geological conditions, and the complex system response.

[0005] Finally, at the data utilization level, the system involves vastly different data dimensions, including zero-dimensional wellhead control parameters, one-dimensional geological stratification attributes, and two-dimensional or three-dimensional spatial field data. Traditional single machine learning models struggle to effectively integrate these multimodal and heterogeneous data, leading to information fragmentation and an inability to fully utilize all available data to improve the accuracy of state perception and prediction. Solving these problems has been difficult, stemming from the inherent challenge of balancing high precision and high computational efficiency, as well as the lack of theories and tools for the unified quantification and fusion of multi-source heterogeneous information. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a monitoring and optimization method for a synergistic system of carbon sequestration and geothermal extraction, thereby resolving the issues present in the prior art.

[0007] In a first aspect, to achieve the above objectives, the present invention provides a monitoring and optimization method for a synergistic system of carbon sequestration and geothermal extraction, comprising the following steps: A dataset reflecting the system state is generated based on a thermo-fluid-mechanical-chemical multiphysics coupling model. Using the multimodal mutual information redundancy method, the optimal combination of monitoring points with high information value and low redundancy is evaluated and selected from the candidate monitoring points to form the optimal monitoring network. A multimodal deep learning alternative model is constructed and trained. Its inputs are sparse monitoring data, system operating parameters and geological parameters obtained by the monitoring network, and its output is a prediction of the overall state of the system and the optimization objective. The trained multimodal deep learning replacement model is used as an evaluator, and an optimization algorithm is used to iteratively optimize the system operating parameters to synergistically maximize thermal extraction efficiency and carbon dioxide sequestration.

[0008] Optionally, the process of selecting the optimal combination of monitoring points includes: Based on the dataset, the information contribution of each candidate monitoring point in different physical fields is calculated. A greedy algorithm is used, with cumulative joint entropy as the evaluation index, to iteratively select the monitoring points that contribute the most information and minimize the redundancy within the monitoring set. When the cumulative joint entropy of the selected set of monitoring points reaches the preset information coverage threshold, the iteration stops and the final combination of monitoring points is output.

[0009] Optionally, the process of generating the dataset includes: Run high-precision thermo-fluid-mechanical-chemical multiphysics coupled numerical simulations to generate the first number of high-fidelity samples; Run a low-precision model with lower computational cost to generate a second number of low-fidelity samples, the second number being greater than the first number; The high-fidelity samples and the low-fidelity samples together constitute a multi-fidelity dataset for model training.

[0010] Optionally, the multimodal deep learning alternative model is a dual-attention residual network that integrates channel attention and spatial attention mechanisms.

[0011] Optionally, the sparse monitoring data input to the multimodal deep learning alternative model is two-dimensional spatial field data, the system operating parameters are zero-dimensional scalar data, and the geological parameters are one-dimensional borehole data.

[0012] Optionally, the optimization objective output by the multimodal deep learning alternative model includes thermal mining power and carbon dioxide mineral sequestration.

[0013] Optionally, the optimization algorithm is an improved vulture search algorithm that incorporates piecewise chaotic mapping, Levy flight strategy, and dynamic weighting factors.

[0014] Optionally, after outputting the optimal system operating parameters, the method further includes: Response surface methodology was performed on the optimal system operating parameters to verify their robustness under parameter perturbations.

[0015] Secondly, the present invention also provides a computer terminal device, comprising: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the monitoring and optimization method for the carbon sequestration and geothermal extraction synergistic system in the first aspect described above.

[0016] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the monitoring and optimization method for the carbon sequestration and geothermal extraction synergistic system described in the first aspect above.

[0017] Compared with the prior art, the present invention has the following advantages and technical effects: This invention provides a monitoring and optimization method for a coordinated carbon sequestration and geothermal extraction system. By constructing a multi-fidelity dataset and optimizing the monitoring network layout using a multimodal mutual information redundancy method, it achieves the acquisition of maximum reservoir information with the fewest monitoring points, significantly reducing monitoring costs. By constructing a multimodal dual-attention residual network replacement model that integrates channel attention and spatial attention mechanisms, it effectively solves the bottleneck of long computation time in high-precision thermo-fluid-mechanical-chemical multiphysics coupled numerical simulations, and overcomes the difficulty of fusing zero-dimensional, one-dimensional, and two-dimensional multimodal heterogeneous data, achieving rapid and high-precision prediction of the system's overall state and optimization objectives. Based on this, an improved vulture search algorithm is combined to perform multi-objective collaborative optimization of injection pressure and flow rate, thereby achieving a dual improvement in thermal extraction efficiency and carbon dioxide sequestration, providing effective support for the safe, efficient operation and real-time intelligent control of the coordinated system. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a monitoring and optimization method for a synergistic system of carbon sequestration and geothermal extraction according to an embodiment of the present invention. Figure 2 This is a schematic diagram of high-precision forward modeling according to an embodiment of the present invention; Figure 3 The following are convergence diagrams of the monitoring network optimization process in an embodiment of the present invention, wherein (a) is the convergence diagram of the monitoring network optimization process of Con (CO2), and (b) is the convergence diagram of the monitoring network optimization process of p; Figure 4 This is a response surface plot showing the optimization results of an embodiment of the present invention. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0021] Example 1 like Figure 1 As shown in the figure, this embodiment provides a monitoring and optimization method for a synergistic system of carbon sequestration and geothermal extraction, including: A dataset reflecting the system state is generated based on a thermo-fluid-mechanical-chemical multiphysics coupling model. Using the multimodal mutual information redundancy method, the optimal combination of monitoring points with high information value and low redundancy is evaluated and selected from the candidate monitoring points to form the optimal monitoring network. A multimodal deep learning alternative model is constructed and trained. Its inputs are sparse monitoring data, system operating parameters and geological parameters obtained by the monitoring network, and its output is a prediction of the overall state of the system and the optimization objective. The trained multimodal deep learning replacement model is used as an evaluator, and an optimization algorithm is used to iteratively optimize the system operating parameters to synergistically maximize thermal extraction efficiency and carbon dioxide sequestration.

[0022] Furthermore, the process of selecting the optimal combination of monitoring points includes: Based on the dataset, the information contribution of each candidate monitoring point in different physical fields is calculated. A greedy algorithm is used, with cumulative joint entropy as the evaluation index, to iteratively select the monitoring points that contribute the most information and minimize the redundancy within the monitoring set. When the cumulative joint entropy of the selected set of monitoring points reaches the preset information coverage threshold, the iteration stops and the final combination of monitoring points is output.

[0023] Specifically, the implementation process of this embodiment includes: Optimization of monitoring network layout structure: To achieve the optimal deployment of monitoring points that obtain the maximum amount of information at the lowest cost.

[0024] ① Multiphysics Information Value Assessment: Quantitatively evaluate the value of each candidate monitoring point to avoid empiricism. Based on the MIMR (Maximum Information Minimum Redundancy) information theory method, evaluate the contribution of each spatial point in the four fields (T, H, M, C) to the overall information content, and eliminate points with high information redundancy.

[0025] ② Greedy algorithm based on MIMR: Find the optimal combination among all candidate points. Iteratively select from the point with the highest information value and calculate the cumulative joint entropy H(s) of the current monitoring point set S.

[0026] ③ Information Constraint Judgment: Ensure that the information coverage rate of the monitoring network meets the standard. Determine whether H(s) reaches the preset minimum information coverage rate Pct (i.e., H(s) ≥ Pct × H(C)). If satisfied, stop the selection and output the optimal monitoring network layout.

[0027] Furthermore, the process of generating the dataset includes: Run high-precision thermo-fluid-mechanical-chemical multiphysics coupled numerical simulations to generate the first number of high-fidelity samples; Run a low-precision model with lower computational cost to generate a second number of low-fidelity samples, the second number being greater than the first number; The high-fidelity samples and the low-fidelity samples together constitute a multi-fidelity dataset for model training.

[0028] Furthermore, the sparse monitoring data input into the multimodal deep learning alternative model is two-dimensional spatial field data, the system operating parameters are zero-dimensional scalar data, and the geological parameters are one-dimensional borehole data.

[0029] Specifically, the implementation process of this embodiment includes: THMC Coupled Model and Multi-Precision Dataset Generation and Input Structure: This approach constructs a physical model of the system and generates high / low precision datasets for training, addressing the efficiency bottleneck of THMC simulations.

[0030] ① High-precision THMC model: Accurately characterizes the water temperature geophysical processes of the system, providing high-precision data. Run the multi-field coupled ScCO2-EGS model to accurately simulate the dynamic evolution of temperature (T), fluid (H), stress (M), and chemical (C) fields, generating a small number of high-fidelity samples.

[0031] ② Low-precision data generation mechanism: Quickly generate large-scale data to meet the training needs of deep learning. By simplifying the THMC coupling model or utilizing pre-trained deep learning (DL) mapping relationships, a large number of low-precision datasets can be generated quickly with low computational cost.

[0032] ③ Multimodal data input interface: Unified reception and processing of data from different dimensions. Receiving: zero-dimensional operation parameters (P, Q), one-dimensional geological profile parameters (fracture spacing d), and two-dimensional spatial monitoring field (Con (CO2), P field distribution).

[0033] Furthermore, the multimodal deep learning alternative model is a dual-attention residual network that integrates channel attention and spatial attention mechanisms.

[0034] Specifically, the implementation process of this embodiment includes: Multimodal Deep Learning Alternative Model (MM-DARN) architecture: It enables deep fusion of multidimensional heterogeneous data and serves as an efficient alternative to thermo-fluid-mechanical-chemical multiphysics coupling models for prediction and optimization.

[0035] ① Residual Network (ResNet) backbone: Serves as the foundational architecture for deep learning, extracting features. Utilizing multiple layers of residual blocks avoids gradient vanishing during deep training, ensuring deep learning of complex nonlinear relationships within the THMC framework.

[0036] ② Channel Attention Module: Integrates information from different physical fields (feature channels). Through average pooling and max pooling, it assigns weights to feature channels of different physical fields (such as temperature, pressure, CO2 concentration), emphasizing the importance of key physical field features.

[0037] ③ Spatial Attention Module: Integrates information from different spatial regions. Through convolution operations, weights are assigned to different spatial locations of the feature map (i.e., different regions in the formation), focusing on the features of key areas such as the vicinity of injection and production wells and high-permeability channels.

[0038] ④ Predictive Output: Provides rapid predictions as an alternative to the THMC model. Outputs high-precision predictions for optimization objectives (thermal extraction efficiency and CO2 sequestration).

[0039] Furthermore, the optimization objectives output by the multimodal deep learning alternative model include thermal mining power and carbon dioxide mineral reserves.

[0040] Furthermore, the optimization algorithm is an improved vulture search algorithm that incorporates piecewise chaotic mapping, Levy flight strategy, and dynamic weighting factors.

[0041] Furthermore, after outputting the optimal system operating parameters, the method also includes: Response surface methodology was performed on the optimal system operating parameters to verify their robustness under parameter perturbations.

[0042] Specifically, the implementation process of this embodiment includes: Multi-objective joint optimization and verification structure: Determine the optimal operating parameters (P, Q) to maximize multiple objectives.

[0043] ① Optimization Objectives: The system's efficiency and safety objectives are defined. Objective 1: Maximize heat power (thermal extraction efficiency); Objective 2: Maximize CO2 sequestration.

[0044] ② Improved Vulture Search Algorithm (IBES): Achieves efficient global optimal parameter search. Improved strategies such as piecewise chaotic mapping, Levy flight strategy, and dynamic weighting factors are introduced to enhance the algorithm's exploration ability and convergence speed, overcoming the tendency of traditional optimization algorithms to get trapped in local optima.

[0045] ③ Iterative optimization: The optimization process is accelerated by using MM-DARN. Each time the optimization algorithm generates a set of parameters (P, Q), it quickly predicts the benefits (heat power, CO2 sequestration) through the third-structure MM-DARN to guide the next iteration, replacing the time-consuming high-precision THMC model.

[0046] ④ Response Surface Verification: Evaluate the stability and reliability of the optimization results. Perform response surface analysis on the optimized parameter combinations to verify their robustness under small-range perturbations of operating parameters.

[0047] Example 2 In this embodiment, a computer terminal device is provided, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the above-described monitoring and optimization method for a carbon sequestration and geothermal extraction synergistic system.

[0048] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described monitoring and optimization method for a carbon sequestration and geothermal extraction synergistic system.

[0049] An application example of this invention is as follows: Taking a deep saline aquifer CO2-EGS project as an example, this paper describes in detail how to apply the technical solution of this invention to achieve synergistic optimization of monitoring network deployment and operating parameters: (1) Implementation case setting and data preparation: ① Model establishment: A two-dimensional profile model was constructed using multi-field coupling software (such as OpenGeoSys) to simulate a geothermal reservoir at a depth of approximately 2500m. The model includes a caprock and bedrock, and incorporates an injector well and a producer well. The model involves four-field coupling of THMC (Thermal Thermal Capacity) as described above. Figure 2 As shown.

[0050] ② Parameter range: Control parameters and uncertainty parameters were set as follows: injection pressure P ranged from 10 to 24.4 MPa; flow rate Q ranged from 0.8 to 1.2 kg / s; and fracture spacing d was used as a geological uncertainty variable.

[0051] ③ Data generation: Run 10 high-precision THMC simulations (approximately 100 hours) as high-precision samples (Ground Truth); run 90 low-precision simulations (approximately 10 hours) to build a multi-fidelity sample library.

[0052] (2) Implementation of optimized monitoring network layout, such as Figure 3As shown, (a) is the convergence diagram of the monitoring network optimization process of Con (CO2), and (b) is the convergence diagram of the monitoring network optimization process of p; ① Candidate point setting: The simulation area was divided into 20×25 grid cells as candidate monitoring locations.

[0053] ② Information value assessment: MIMR information evaluation was performed on the data from the four fields: T, H, M, and C.

[0054] ③ Optimality judgment (ω determined): Run the MIMR greedy algorithm to calculate the cumulative joint entropy H(s) for different combinations of monitoring points. According to the convergence graph analysis, when ω=8 monitoring points are selected, the information gain curve tends to flatten out, satisfying the constraint H(s)≥Pct×H(C).

[0055] ④ Monitoring point layout output: The algorithm automatically recommends the optimal layout: Pressure sensor: Installed in the low-permeability cap layer around the injection well.

[0056] CO2 concentration sensors: deployed in production wells and central high-permeability channels (e.g., chemical field monitoring focuses on high-permeability areas).

[0057] (3) Training and application of MM-DARN alternative models: ① Data input: The sparse monitoring data (two-dimensional), corresponding operating parameters (P, Q, zero-dimensional), and geological parameters (d, one-dimensional) of the above 8 points are input into the MM-DARN network.

[0058] ② Model training: MM-DARN utilizes a dual attention mechanism (channel attention and spatial attention) to deeply fuse multimodal data and is trained using a residual network (ResNet) structure.

[0059] ③ Model accuracy: After training, the model achieved a root mean square error (RMSE) of only 0.005 for predicting overall pressure and CO2 concentration, with a prediction accuracy R0. 2 Up to 1.000. The model has extremely high prediction accuracy and can replace the THMC coupled model.

[0060] (4) Multi-objective joint optimization implementation, such as Figure 4 As shown: ① Optimize goal setting: The multi-objective function is set to maximize thermal power and CO2 mineral sequestration.

[0061] ② Optimization algorithm: We launched the improved vulture search algorithm (IBES) and used the MM-DARN model as a fast fitness evaluator.

[0062] ③ Iteration and convergence: The optimization algorithm converged after approximately 20 iterations.

[0063] ④ Optimal parameter output: The system outputs the optimal operating strategy: the injection pressure is adjusted to 22MPa and the flow rate is adjusted to 1.1kg / s.

[0064] ⑤ Benefit verification: Under optimal parameters, thermal extraction efficiency was improved by approximately 15% compared to the initial scheme, while CO2 mineral sequestration was increased by approximately 10%.

[0065] ⑥ Response surface validation: The global optimality and robustness of the optimization results were verified using response surface plots.

[0066] The core of this invention lies in constructing a data-driven, multi-objective collaborative optimization framework, aiming to solve the challenges of high-precision simulation, efficient monitoring, and intelligent control in the CO2 storage and geothermal extraction collaborative system (ScCO2-EGS). The patent's protection primarily focuses on the following three interrelated technological innovations: 1. Optimal Monitoring Network Deployment Method: This method employs the Multimodal Mutual Information Redundancy (MIMR) information theory approach to quantitatively evaluate the information value of multiple physical field parameters such as T, H, M, and C. Utilizing samples provided by multi-precision datasets, this method iteratively selects monitoring points using a greedy MIMR algorithm. This achieves the goal of acquiring the maximum amount of information with the minimum number of monitoring points while satisfying the information coverage constraint H(s) ≥ Pct × H(C), significantly reducing monitoring costs.

[0067] 2. Multimodal Deep Learning Alternative Model (MM-DARN): This paper innovatively proposes a multimodal dual-attention residual network (MM-DARN) as an intelligent alternative to the high-precision THMC model. The core of this model structure lies in the integration of channel attention and spatial attention modules, enabling deep fusion of heterogeneous and multimodal data such as zero-dimensional (operational parameters P, Q), one-dimensional (geological parameters d), and two-dimensional (spatial field monitoring data). This allows for rapid and high-precision prediction of target variables such as thermal extraction efficiency and CO2 sequestration.

[0068] 3. Multi-objective joint optimization algorithm: Based on the prediction model, this method achieves synergistic optimization of thermal extraction efficiency and CO2 storage capacity. It employs an improved vulture search algorithm with enhanced global search capabilities and integrates improved strategies such as Piecewise chaotic mapping, Levy flight strategy, and dynamic weighting factors. This significantly improves convergence and optimization performance in complex multi-objective problems, ultimately providing optimal operating parameters for the ScCO2-EGS system.

[0069] Alternative solution to the present invention: Alternatively, Deep Reinforcement Learning (DRL) or Graph Neural Networks (GNNs) can be used for monitoring network optimization. Monitoring point selection can be treated as a sequential decision problem, using a DRL agent to learn the optimal selection strategy in the candidate site space, with the reward function being the maximization of accumulated information (joint entropy). Alternatively, the monitoring network can be viewed as a graph structure, using GNNs to directly predict the optimal set of sites under given constraints. Advantages: Compared to the local optima of the MIMR greedy algorithm, DRL has the potential to search for a globally better monitoring network configuration.

[0070] Alternative solution to the present invention: Gaussian process regression (GPR) or multi-fidelity hybrid models (Co-Kriging) are employed. Co-Kriging models effectively fuse information from high-precision and low-precision data. It uses GPR to correct the output of the low-precision model, achieving predictive performance similar to the high-precision model. Advantages: GPR or Co-Kriging performs well in handling small sample data and quantifying prediction uncertainty, while having lower computational complexity than large deep learning networks.

[0071] Alternative solution three for this invention: Other high-performance evolutionary or multi-objective optimization algorithms can be employed, such as: NSGA-III with Adaptive Operations Selection (AOS) strategy integration: a classic algorithm for handling multi-objective optimization problems, integrating AOS improves the algorithm's efficiency and robustness in complex search spaces. TRIBES algorithm: a hybrid algorithm based on particle swarm optimization and evolutionary strategies, competitive in handling complex response surface optimization problems with multiple objectives. Advantages: It broadens the range of optimization algorithms, allowing the selection of the most suitable optimization strategy based on the response surface characteristics of the actual system (e.g., whether it is highly non-convex), to reach the optimal solution set more quickly.

[0072] This invention provides a monitoring and optimization method for a coordinated carbon sequestration and geothermal extraction system. By constructing a multi-fidelity dataset and optimizing the monitoring network layout using a multimodal mutual information redundancy method, it achieves the acquisition of maximum reservoir information with the fewest monitoring points, significantly reducing monitoring costs. By constructing a multimodal dual-attention residual network replacement model that integrates channel attention and spatial attention mechanisms, it effectively solves the bottleneck of long computation time in high-precision thermo-hydrodynamic coupling numerical simulations and overcomes the difficulty of fusing zero-dimensional, one-dimensional, and two-dimensional multimodal heterogeneous data, achieving rapid and high-precision prediction of the system's overall state and optimization objectives. Based on this, an improved vulture search algorithm is combined to perform multi-objective collaborative optimization of injection pressure and flow rate, thereby achieving a dual improvement in thermal extraction efficiency and carbon dioxide sequestration, providing effective support for the safe, efficient operation and real-time intelligent control of the coordinated system.

[0073] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A monitoring and optimization method for a synergistic system of carbon sequestration and geothermal extraction, characterized in that, Includes the following steps: A dataset reflecting the system state is generated based on a thermo-fluid-mechanical-chemical multiphysics coupling model. Using the multimodal mutual information redundancy method, the optimal combination of monitoring points with high information value and low redundancy is evaluated and selected from the candidate monitoring points to form the optimal monitoring network. A multimodal deep learning alternative model is constructed and trained. Its inputs are sparse monitoring data, system operating parameters and geological parameters obtained by the monitoring network, and its output is a prediction of the overall state of the system and the optimization objective. The trained multimodal deep learning replacement model is used as an evaluator, and an optimization algorithm is used to iteratively optimize the system operating parameters to synergistically maximize thermal extraction efficiency and carbon dioxide sequestration.

2. The method according to claim 1, characterized in that, The process of selecting the optimal combination of monitoring points includes: Based on the dataset, the information contribution of each candidate monitoring point in different physical fields is calculated. A greedy algorithm is used, with cumulative joint entropy as the evaluation index, to iteratively select the monitoring points that contribute the most information and minimize the redundancy within the monitoring set. When the cumulative joint entropy of the selected set of monitoring points reaches the preset information coverage threshold, the iteration stops and the final combination of monitoring points is output.

3. The method according to claim 1, characterized in that, The process of generating a dataset includes: Run high-precision thermo-fluid-mechanical-chemical multiphysics coupled numerical simulations to generate the first number of high-fidelity samples; Run a low-precision model with lower computational cost to generate a second number of low-fidelity samples, the second number being greater than the first number; The high-fidelity samples and the low-fidelity samples together constitute a multi-fidelity dataset for model training.

4. The method according to claim 1, characterized in that, The multimodal deep learning alternative model is a dual-attention residual network that integrates channel attention and spatial attention mechanisms.

5. The method according to claim 1 or 4, characterized in that, The sparse monitoring data input to the multimodal deep learning alternative model is two-dimensional spatial field data, the system operating parameters are zero-dimensional scalar data, and the geological parameters are one-dimensional borehole data.

6. The method according to claim 1, characterized in that, The optimization objectives output by the multimodal deep learning alternative model include thermal mining power and carbon dioxide mineral reserves.

7. The method according to claim 1, characterized in that, The optimization algorithm is an improved vulture search algorithm that incorporates piecewise chaotic mapping, Levy flight strategy, and dynamic weighting factors.

8. The method according to claim 1, characterized in that, After outputting the optimal system operating parameters, the method further includes: Response surface methodology was performed on the optimal system operating parameters to verify their robustness under parameter perturbations.

9. A computer terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the method as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-8.

Images