Horizontal stratum freezing construction thermal coupling analysis and intelligent agent model optimization method and system

Abstract

The invention discloses a horizontal stratum freezing construction thermal-mechanical coupling analysis and intelligent agent model optimization method. The method comprises the following steps that S1, horizontal stratum freezing construction parameters and response data are collected; s2, characteristic rules of freezing curtains and frost heaving deformation in the horizontal stratum freezing process of the construction area are obtained; s3, constructing a thermal coupling analysis data set based on a full-factor test design; s4, establishing a thermal-mechanical coupling agent model by adopting a machine learning algorithm; s5, constructing an optimization model of freezing construction parameters and a three-dimensional optimization decision drawing board; and S6, correcting the optimization model and the three-dimensional optimization decision drawing board based on engineering data. Aiming at the problems of low thermal-mechanical coupling analysis efficiency, high parameter optimization difficulty and the like in horizontal stratum freezing construction, a set of complete analysis and decision-making method is formed, the freezing curtain forming precision can be improved, frost heaving deformation can be controlled, the method can be effectively applied to actual horizontal freezing engineering, and the construction efficiency is improved. The method has a good guiding effect on construction parameter optimization and engineering safety control.

Description

Technical Field

[0001] This invention relates to the field of underground engineering construction and intelligent decision-making, specifically to a method and system for thermal coupling analysis and intelligent agent model optimization in horizontal stratum freezing construction. Background Technology

[0002] With the continuous development of urban underground space, the horizontal freezing method, as a key technology for reinforcing and sealing weak strata, has been widely used in projects such as subway tunnels and integrated utility tunnels. This technology forms a freezing curtain by horizontally deploying freezing pipes, providing a safe environment for excavation operations. However, the process involves complex coupling effects between temperature and displacement fields and is affected by multiple factors such as brine temperature, stratum parameters, and the arrangement of freezing holes.

[0003] Currently, the optimization of horizontal freezing construction parameters mainly relies on numerical simulation and engineering experience. Traditional numerical simulation methods (such as the finite element method) require the construction of complex thermo-mechanical coupling models, with a single calculation taking 3-5 hours, which is difficult to meet the needs of real-time decision-making on site. At the same time, on-site technicians often adjust parameters through trial and error, which can easily lead to risks such as insufficient freezing curtain thickness or excessive frost heave deformation.

[0004] In recent years, scholars have attempted to introduce surrogate models to simplify calculations. For example, existing technologies employ support vector machines to predict freezing temperature fields, but these do not consider multi-parameter coupling effects. There are also existing technologies that use neural networks to construct frost heave deformation prediction models, but their optimization objectives are singular, failing to achieve synergistic optimization of freezing effects and deformation control. Furthermore, existing methods lack a dedicated optimization framework for the characteristics of horizontal freezing, making them difficult to adapt to construction needs under complex geological conditions. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of existing methods for analyzing and optimizing horizontal strata freezing construction, and provides a highly efficient and accurate thermo-mechanical coupling analysis and intelligent optimization method to solve the following problems:

[0006] Traditional numerical simulation calculations are too time-consuming and cannot meet the needs of real-time on-site analysis;

[0007] Under the combined effect of multiple factors, the mapping relationship between frozen parameters and engineering response is difficult to quantify;

[0008] Existing optimization methods cannot balance the contradiction between the thickness of the frozen curtain and frost heave deformation, and lack systematic decision-making tools.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] A method for thermo-mechanical coupling analysis and intelligent agent model optimization during horizontal stratum freezing construction includes the following steps:

[0011] S1. Collect horizontal strata freezing construction parameters and response data;

[0012] S2. Based on the data collected in step S1, obtain the characteristic laws of freezing curtain and frost heave deformation during the freezing process of horizontal strata in the construction area.

[0013] S3. Based on full factorial experimental design, construct a thermo-coupling analysis dataset;

[0014] S4. Based on the feature patterns obtained in step S2 and the thermal coupling analysis dataset constructed in step S3, a thermal coupling proxy model is established using machine learning algorithms.

[0015] S5. Based on the thermo-coupling proxy model established in step S4, construct an optimization model for the frozen construction parameters and a three-dimensional optimization decision board.

[0016] S6. Correct the optimization model and the 3D optimization decision board based on engineering data.

[0017] Furthermore, in step S1, the collected horizontal formation freezing construction parameters include brine circulation temperature, initial formation temperature, thermal conductivity, and freezing hole spacing.

[0018] The response data includes curtain thickness and frost heave deformation, where curtain thickness is the average thickness of the frozen curtain and frost heave deformation is the maximum frost heave deformation in the construction area.

[0019] Further, in step S3, specifically, a full factorial experiment is designed using Minitab software, with brine circulation temperature, initial formation temperature, thermal conductivity and freezing hole spacing as input parameters. Each type of input parameter is normalized, and multiple horizontal gradients are set for each type of input parameter. Multiple sets of samples are generated by COMSOL numerical simulation as a thermo-coupling analysis dataset, and the multiple sets of samples are divided into training set, validation set and test set.

[0020] Further, in step S4, a horizontal freezing thermo-coupling proxy model is established using the GPR algorithm, referred to as the GPR proxy model. In the GPR proxy model, the brine circulation temperature and the spacing between freezing holes are used as design variables, while the initial formation temperature and thermal conductivity are fixed parameters. The output results are the predicted curtain thickness and frost heave deformation.

[0021] Furthermore, in step S5, when constructing the optimization model for freezing construction parameters, the GPR proxy model is used as the objective function, and a dual-objective collaborative optimization framework is adopted. The optimization objectives are "curtain thickness close to the target value" and "frost heave deformation is minimized". The NSGA-II algorithm is used to solve the Pareto optimal solution set, and the Pareto optimal solution set is sorted based on the TOPSIS method to determine the optimal parameter combination under different geological conditions.

[0022] Then, a three-dimensional optimization decision board is constructed based on the optimization model.

[0023] Furthermore, in step S6, when calibrating the optimization model based on engineering data, it is determined whether the deviation rate between the predicted result and the measured value is not greater than 5%. If so, no calibration is required; otherwise, return to step S4 and adjust the hyperparameters of the GPR surrogate model until the deviation rate meets the requirements.

[0024] Furthermore, after step S6, the following step is also included:

[0025] S7. Based on the corrected optimization model, a visual decision-making system for horizontal stratum freezing construction is developed. After the user sets the optimization model in the system and inputs the initial stratum temperature and thermal conductivity, the system can automatically output the optimal brine circulation temperature and freezing hole spacing.

[0026] Another aspect of the present invention provides a visual decision-making system for horizontal stratum freezing construction. The system is developed using the aforementioned method and includes a functional module, an interactive module, a view module, and an output module.

[0027] The functional modules integrate file management, model calculation, result output, model configuration and selection functions;

[0028] The interactive module includes a parameter input function, allowing users to perform numerical simulation operations;

[0029] The view module is built on the OpenGL graphics engine to construct a 3D rendering core, dynamically displaying multi-scale models and visualization output results;

[0030] The output module includes extreme value calculation, surrogate model prediction and target optimization execution, and log monitoring window, making the model calculation process and results transparent.

[0031] Compared with the prior art, the advantages of the present invention are as follows:

[0032] The present invention provides a method and system for thermal coupling analysis and intelligent surrogate model optimization in horizontal stratum freezing construction. It addresses the problems of low efficiency in thermal coupling analysis and high difficulty in parameter optimization in horizontal stratum freezing construction. It constructs a dataset based on full factorial experimental design, establishes a high-precision surrogate model using machine learning algorithms, and combines multi-objective optimization algorithms to achieve intelligent optimization of construction parameters, forming a complete set of analysis and decision-making methods, and providing technical support for horizontal freezing projects.

[0033] Specifically, this invention, based on full factorial experimental design and machine learning algorithms, constructs a high-precision thermo-mechanical coupling surrogate model, reducing the freezing response prediction time from 3-5 hours to milliseconds, significantly improving analysis efficiency. It also proposes a dual-objective collaborative optimization framework, using the NSGA-II algorithm to balance the contradiction between the thickness of the frozen curtain and frost heave deformation, providing multiple scenario-based options for on-site operations and reducing engineering risks. Furthermore, a 3D optimization decision board and visualization decision system have been developed, enabling visualized optimization of construction parameters. This can be operated without requiring specialized simulation knowledge, meeting the needs of on-site technicians. Engineering verification shows that after optimization using the method provided by this invention, the forming accuracy of the frozen curtain is improved by 15%, and frost heave deformation is controlled within 5mm, effectively improving the forming accuracy of the frozen curtain and controlling frost heave deformation. This invention can be effectively applied to actual horizontal freezing projects, suitable for complex horizontal freezing projects such as subway tunnels, and has a good guiding effect on construction parameter optimization and engineering safety control. Attached Figure Description

[0034] Figure 1 A flowchart illustrating the method for thermal coupling analysis and intelligent agent model optimization during horizontal stratum freezing construction provided by this invention;

[0035] Figure 2a and Figure 2b A comparison chart showing the numerical simulation results and the predicted values ​​from various surrogate models;

[0036] Figure 3a and Figure 3b Regression fitting plots for various surrogate models on the test set;

[0037] Figure 4a and Figure 4b The relative error distribution of various surrogate models on the test set;

[0038] Figure 5a and Figure 5b The regression fit plot of the GPR model;

[0039] Figure 6 This is an interface diagram of the visualization decision-making system for horizontal stratum freezing construction provided by the present invention. Detailed Implementation

[0040] To make the technical means, creative features, objectives and effects of this invention easier to understand, the following description, in conjunction with the accompanying drawings and specific embodiments, further explains how this invention is implemented.

[0041] In one specific embodiment, refer to Figure 1 As shown, this invention provides a method for thermo-mechanical coupling analysis and intelligent agent model optimization during horizontal stratum freezing construction.

[0042] In step S1, the horizontal strata freezing construction parameters and response data are collected.

[0043] Specifically, the collected horizontal strata freezing construction parameters include brine circulation temperature, initial strata temperature, thermal conductivity, and freezing hole spacing. Response data include curtain thickness and frost heave deformation, where curtain thickness is the average thickness of the freezing curtain, and frost heave deformation is the maximum frost heave deformation in the construction area.

[0044] Furthermore, engineering logging data and numerical simulation results were extracted to supplement the dataset and ensure that the parameters covered the actual range of the project. For example, the brine circulation temperature covered -20 to -35℃, the initial formation temperature covered 15 to 21℃, the thermal conductivity covered 1.20 to 1.95 W / (m·℃), and the freezing hole spacing covered 1100 to 1550 mm.

[0045] In step S2, based on the data collected in step S1, the characteristic patterns of freezing curtain and frost heave deformation during the freezing process of horizontal strata in the construction area are obtained.

[0046] In this embodiment, taking the freezing project of the section of the subway line 12 passing under the operating station of a certain city as an example, after on-site data collection, the freezing temperature field was recorded every 2 hours by a fiber optic thermometer, and the deformation of the station floor was monitored by a total station. The data sampling period was 1 day.

[0047] After data collection, the study found that when the brine temperature was -30℃, a curtain with a thickness of 2.8m was formed after 15 days of freezing, and reached 3.2m after 30 days. The frost heave deformation pattern was as follows: the deformation rate was fastest in the first 10 days (0.2mm per day), and tended to stabilize after 30 days, with a maximum deformation of 4.8mm, and the maximum deformation was concentrated in the middle of the station floor. When the spacing between freezing holes was 1200mm, the curtain thickness was 3.0m, and when the spacing was increased to 1300mm, the thickness decreased to 2.7m, and the maximum deformation increased by 15%.

[0048] The study clarified that low-temperature brine can accelerate curtain formation but increase the risk of frost heave; the correlation between freezing hole spacing and curtain thickness was analyzed, and it was found that when the spacing increases, the curtain thickness decreases, but the frost heave deformation increases; a coupling effect model of initial formation temperature and thermal conductivity was also established, revealing that freezing efficiency is increased by 20% in high thermal conductivity formations (1.95 W / (m·℃)); and the evolution law of frost heave deformation over time was summarized.

[0049] In step S3, a thermo-coupling analysis dataset is constructed based on a full factorial experimental design.

[0050] Specifically, a full factorial experiment was designed using Minitab software, with brine circulation temperature, initial formation temperature, thermal conductivity and freezing hole spacing as input parameters. Each type of input parameter was normalized, and multiple horizontal gradients were set for each type of input parameter. Multiple sets of samples were generated by COMSOL numerical simulation as a thermo-coupling analysis dataset, and the multiple sets of samples were divided into training set, validation set and test set.

[0051] In this embodiment, the thickness of the frozen curtain and the frost heave deformation corresponding to each parameter combination were obtained through COMSOL numerical simulation. Multiple horizontal gradients were set for each of the four input parameters: brine temperature: -20, -25, -30, -35℃; initial formation temperature: 15, 18, 21℃; thermal conductivity: 1.20, 1.45, 1.70, 1.95 W / (m·℃); borehole spacing: 1100~1550 mm, in 50 mm increments. 480 sets of sample data were generated through COMSOL numerical simulation. Stratified sampling was used to divide the dataset into a training set (70%), a validation set (15%), and a test set (15%). The data was normalized, and the parameters were scaled to the [0,1] interval to eliminate the influence of dimensions.

[0052] In step S4, based on the feature patterns obtained in step S2 and the thermal coupling analysis dataset constructed in step S3, a thermal coupling proxy model is established using a machine learning algorithm.

[0053] A thermally coupled proxy model can be established by using multiple different machine learning algorithms, and the optimal machine learning algorithm can be selected by comparing performance.

[0054] In this implementation, four algorithms—XGBoost, DNN, GPR, and SVR—were used to construct surrogate models. The brine circulation temperature and freezing hole spacing were used as design variables, while the initial formation temperature and thermal conductivity were fixed parameters. The outputs were the predicted curtain thickness and frost heave deformation. The specific settings for the four algorithms are as follows:

[0055] XGBoost: Set learning_rate=0.03, max_depth=8, and optimize hyperparameters using Bayesian methods;

[0056] DNN: Four-layer fully connected network, batch_size=32, Adam optimizer, learning_rate=0.001;

[0057] GPR: Composite kernel function (constant kernel × radial basis kernel), n_restarts_optimizer=5;

[0058] SVR: Radial basis kernel function, optimized by searching the logarithmic space grid with parameters C=100 and ε=0.1.

[0059] Bayesian optimization and grid search methods were used to adjust the model hyperparameters, and the model performance was evaluated using MAE, RMSE, and R² metrics to select the optimal model. In this embodiment, different models were processed separately, and then... Figure 2a and Figure 2b The image shows a comparison between numerical simulation (COMSOL) values ​​and predictions from various surrogate models. Figure 2a Corresponding to the average thickness of the frozen curtain, Figure 2b The maximum frost heave deformation corresponding to the station floor slab; refer to Figure 3a and Figure 3b The image shows the regression fitting plots for various surrogate models on the test set. Figure 3a Corresponding to the average thickness of the frozen curtain, Figure 3b The maximum frost heave deformation corresponding to the station floor slab; refer to Figure 4a and Figure 4b The figure shows the relative error distribution of various proxy models on the test set. Figure 4a Corresponding to the average thickness of the frozen curtain, Figure 4b This corresponds to the maximum frost heave deformation of the station's floor slab. As can be seen from the attached figures, the GPR model exhibits the lowest overall error and the best performance.

[0060] Reference Figure 5a and Figure 5b The image shown is a regression fit plot of the GPR model. Figure 5a Corresponding to the average thickness of the frozen curtain, Figure 5b The maximum frost heave deformation corresponds to the station floor slab. Under the GPR model, the curtain thickness corresponds to MAE=0.009m, RMSE=0.011m, R²=0.998, and the frost heave deformation corresponds to MAE=0.058mm, RMSE=0.075mm, R²=0.993.

[0061] The Gaussian Process Regression (GPR) model employs a composite kernel function consisting of a constant kernel multiplied by a radial basis kernel, and introduces a white noise term to enhance model robustness. MultiOutputRegressor encapsulates independent sub-models of multiple output objectives, achieving multi-objective modeling. During training, an automatic kernel function optimization and hyperparameter grid search strategy are combined, and multiple restarts are implemented to reduce the impact of initial parameters on model performance, thereby improving the stability of kernel function optimization.

[0062] Therefore, the GPR algorithm was subsequently used to establish a horizontal freezing thermo-coupling surrogate model, which is called the GPR surrogate model.

[0063] In step S5, based on the thermo-coupling proxy model established in step S4, an optimization model for frozen construction parameters and a three-dimensional optimization decision board are constructed.

[0064] In this embodiment, when constructing the optimization model for frozen construction parameters, the GPR proxy model is used as the objective function.

[0065] The design variables are: brine circulation temperature -20~-35℃; freezing hole spacing 1000~1600mm.

[0066] The fixed parameters are: initial formation temperature of 15, 18, and 21℃, and thermal conductivity of 1.20, 1.45, 1.70, and 1.95 W / (m·℃), forming 12 sets of working conditions.

[0067] A dual-objective collaborative optimization framework is adopted, with the optimization objectives being "curtain thickness approaching the target value (3m in this embodiment)" and "minimizing frost heave deformation." The NSGA-II algorithm is used to solve for the Pareto optimal solution set. The Nondominated Sorting Genetic Algorithm II (NSGA-II) with elitist strategy is an improved multi-objective optimization algorithm based on genetic algorithms, widely used to handle optimization problems with multiple conflicting objectives. While retaining the basic framework of standard genetic algorithms (GA), this algorithm integrates nondominated sorting, crowding assessment, and elitist retention mechanisms, significantly improving the convergence efficiency and solution set diversity of multi-objective optimization problems. Compared to weighted algorithms, NSGA-II does not rely on manually set objective weight coefficients but rather approximates the Pareto optimal front in the objective space based on Pareto dominance relationships.

[0068] The NSGA-II algorithm parameters are: population size 200, maximum number of generations 300, crossover probability 0.9, and mutation probability 0.1.

[0069] Generate the Pareto optimal solution set. For example, the optimal solution for condition 5 (initial temperature 18℃, thermal conductivity 1.20W / (m·℃)) is: brine circulation temperature -30.22℃, freezing hole spacing 1493mm, corresponding curtain thickness 3.05m, and frost heave deformation 5.55mm.

[0070] Furthermore, based on the TOPSIS method, the optimal solution set of Pareto is sorted to determine the optimal parameter combination under different geological conditions; then, a three-dimensional optimization decision board is constructed based on the optimization model.

[0071] In step S6, the optimization model and the three-dimensional optimization decision board are corrected based on engineering data.

[0072] Specifically, when correcting the optimization model based on engineering data, the field data of typical horizontal freezing projects are extracted, such as the freezing project of a certain urban rail transit section corresponding to this embodiment, including construction parameters and monitoring response.

[0073] Project the measured data onto the 3D optimization board, calculate the deviation rate, and determine whether the deviation rate between the predicted result and the measured value is no greater than 5%. If so, no correction is needed; otherwise, return to step S4, adjust the hyperparameters of the GPR surrogate model, and retrain until the deviation rate meets the requirements. Compare the prediction accuracy of the model before and after correction to ensure the consistency between the optimization result and the field data.

[0074] In a real-world test case, the brine circulation temperature was -30℃, the freezing hole spacing was 1400mm, the curtain thickness was 2.98m, and the deformation was 5.6mm. The model predicted the curtain thickness to be 3.01m and the deformation to be 5.5mm, with deviation rates of 1.0% and 1.8%, respectively. This indicates that the corrected 3D optimized drawing board matches the measured data well and meets the engineering accuracy requirements.

[0075] In step S7, based on the corrected optimization model, a visual decision-making system for horizontal stratum freezing construction is developed. After the user sets the optimization model in the system, the initial stratum temperature and thermal conductivity are input, and the system can automatically output the optimal brine circulation temperature and freezing hole spacing.

[0076] In addition, it can generate spatiotemporal evolution curves of the thickness of the frozen curtain and frost heave deformation, quantitatively characterize the safety risks in the construction process, and output parameter adjustment suggestions. For example, when the monitored deformation exceeds the limit, it automatically recommends reducing the brine temperature by 2°C or reducing the hole spacing by 50mm.

[0077] In a field test case, based on the method provided by this invention, the optimal construction scheme for the project is recommended: brine temperature -30.5℃, hole spacing 1420mm; field application results: after freezing for 30 days, the curtain thickness is 3.02m, and the maximum frost heave deformation is 5.2mm, both of which meet the design requirements; compared with the traditional scheme, the freezing period is shortened by 5 days and the deformation is reduced by 12%, which verifies the effectiveness of this invention.

[0078] Another aspect of this invention provides a visual decision-making system for horizontal stratum freezing construction. Developed using the aforementioned method, the system includes a functional module, an interactive module, a view module, and an output module. The functional module integrates file management, model calculation, result output, and model configuration and selection functions. The interactive module includes parameter input functionality for users to perform numerical simulation operations. The view module, based on an OpenGL graphics engine, constructs a 3D rendering core, dynamically displaying multi-scale models and visual output results. The output module encompasses extreme value calculation, surrogate model prediction and target optimization execution, and a log monitoring window, achieving transparency in the model calculation process and results.

[0079] Reference Figure 6The diagram illustrates the interface of a visualization decision-making system in a specific embodiment. The diagram shows the function area, interaction area, view area, and output area, corresponding to the aforementioned function module, interaction module, view module, and output module. This system can be deployed locally, packaged as a standalone application using the COMSOL Compiler tool, and run locally on multiple platforms including Windows, Linux, and macOS. Alternatively, it can be deployed in the cloud, utilizing COMSOL Server for server-side deployment. Users can access it via web login or a client, with all computational tasks performed by the server, improving the efficiency of computing resource utilization.

[0080] Figure 6 The functional area, located at the top of the interface, integrates file management, model calculation, result output, and model selection. It employs a cascading menu and shortcut icons to improve the efficiency of complex workflows. The interactive areas are symmetrically distributed on the left and right sides of the interface, using a tabbed categorization management mode. The left panel includes geometry and mesh, and soil parameter functions; the right panel includes a time regulator and saltwater circulation plan, facilitating numerical simulation operations. The view area, located in the center of the interface, uses an OpenGL graphics engine to build a 3D rendering core, dynamically displaying multi-scale models and visual output results. The output area, located at the bottom of the interface, covers extreme value calculation, surrogate model prediction and target optimization execution, and a log monitoring window, making the model calculation process and results transparent.

[0081] In summary, the method and system for thermo-mechanical coupling analysis and intelligent surrogate model optimization in horizontal stratum freezing construction provided by this invention addresses the problems of low efficiency in thermo-mechanical coupling analysis and high difficulty in parameter optimization in horizontal stratum freezing construction. It constructs a dataset based on full factorial experimental design, establishes a high-precision surrogate model using machine learning algorithms, and combines multi-objective optimization algorithms to achieve intelligent optimization of construction parameters, forming a complete set of analysis and decision-making methods, providing technical support for horizontal freezing projects.

[0082] Specifically, this invention, based on full factorial experimental design and machine learning algorithms, constructs a high-precision thermo-mechanical coupling surrogate model, reducing the freezing response prediction time from 3-5 hours to milliseconds, significantly improving analysis efficiency. It also proposes a dual-objective collaborative optimization framework, using the NSGA-II algorithm to balance the contradiction between the thickness of the frozen curtain and frost heave deformation, providing multiple scenario-based options for on-site operations and reducing engineering risks. Furthermore, a 3D optimization decision board and visualization decision system have been developed, enabling visualized optimization of construction parameters. This can be operated without requiring specialized simulation knowledge, meeting the needs of on-site technicians. Engineering verification shows that after optimization using the method provided by this invention, the forming accuracy of the frozen curtain is improved by 15%, and frost heave deformation is controlled within 5mm, effectively improving the forming accuracy of the frozen curtain and controlling frost heave deformation. This invention can be effectively applied to actual horizontal freezing projects, suitable for complex horizontal freezing projects such as subway tunnels, and has a good guiding effect on construction parameter optimization and engineering safety control.

[0083] Finally, it should be noted that the above description is only an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for thermo-mechanical coupling analysis and intelligent agent model optimization during horizontal stratum freezing construction, characterized in that, Includes the following steps: S1. Collect horizontal strata freezing construction parameters and response data; S2. Based on the data collected in step S1, obtain the characteristic laws of freezing curtain and frost heave deformation during the freezing process of horizontal strata in the construction area. S3. Based on full factorial experimental design, construct a thermo-coupling analysis dataset; S4. Based on the feature patterns obtained in step S2 and the thermal coupling analysis dataset constructed in step S3, a thermal coupling proxy model is established using machine learning algorithms. S5. Based on the thermo-coupling proxy model established in step S4, construct an optimization model for the frozen construction parameters and a three-dimensional optimization decision board. S6. Correct the optimization model and the 3D optimization decision board based on engineering data.

2. The method for thermo-mechanical coupling analysis and intelligent agent model optimization of horizontal stratum freezing construction according to claim 1, characterized in that, In step S1, the collected horizontal formation freezing construction parameters include brine circulation temperature, initial formation temperature, thermal conductivity, and freezing hole spacing. The response data includes curtain thickness and frost heave deformation, where curtain thickness is the average thickness of the frozen curtain and frost heave deformation is the maximum frost heave deformation in the construction area.

3. The method for thermo-mechanical coupling analysis and intelligent agent model optimization of horizontal stratum freezing construction according to claim 2, characterized in that, In step S3, specifically, a full factorial experiment is designed using Minitab software, with brine circulation temperature, initial formation temperature, thermal conductivity and freezing hole spacing as input parameters. Each type of input parameter is normalized, and multiple horizontal gradients are set for each type of input parameter. Multiple sets of samples are generated by COMSOL numerical simulation as a thermo-coupling analysis dataset, and the multiple sets of samples are divided into training set, validation set and test set.

4. The method for thermo-mechanical coupling analysis and intelligent agent model optimization of horizontal stratum freezing construction according to claim 3, characterized in that, In step S4, a horizontal freezing thermo-coupling proxy model is established using the GPR algorithm, referred to as the GPR proxy model. In the GPR proxy model, the brine circulation temperature and the spacing between freezing holes are used as design variables, while the initial formation temperature and thermal conductivity are fixed parameters. The output results are the predicted curtain thickness and frost heave deformation.

5. The method for thermo-mechanical coupling analysis and intelligent agent model optimization of horizontal stratum freezing construction according to claim 4, characterized in that, In step S5, when constructing the optimization model for freezing construction parameters, the GPR proxy model is used as the objective function, and a dual-objective collaborative optimization framework is adopted. The optimization objectives are "curtain thickness close to the target value" and "frost heave deformation is minimized". The NSGA-II algorithm is used to solve the Pareto optimal solution set, and the Pareto optimal solution set is sorted based on the TOPSIS method to determine the optimal parameter combination under different geological conditions. Then, a three-dimensional optimization decision board is constructed based on the optimization model.

6. The method for thermo-mechanical coupling analysis and intelligent agent model optimization of horizontal stratum freezing construction according to claim 5, characterized in that, In step S6, when calibrating the optimization model based on engineering data, it is determined whether the deviation rate between the predicted result and the measured value is no greater than 5%. If so, no calibration is required; otherwise, return to step S4 and adjust the hyperparameters of the GPR surrogate model until the deviation rate meets the requirements.

7. The method for thermo-mechanical coupling analysis and intelligent agent model optimization of horizontal stratum freezing construction according to claim 6, characterized in that, Following step S6, the following steps are also included: S7. Based on the corrected optimization model, a visual decision-making system for horizontal stratum freezing construction is developed. After the user sets the optimization model in the system and inputs the initial stratum temperature and thermal conductivity, the system can automatically output the optimal brine circulation temperature and freezing hole spacing.

8. A visual decision-making system for horizontal stratum freezing construction, characterized in that, The system was developed using the method described in claim 7, and includes a functional module, an interaction module, a view module, and an output module; The functional modules integrate file management, model calculation, result output, model configuration and selection functions; The interactive module includes a parameter input function, allowing users to perform numerical simulation operations; The view module is built on the OpenGL graphics engine to construct a 3D rendering core, dynamically displaying multi-scale models and visualization output results; The output module includes extreme value calculation, surrogate model prediction and target optimization execution, and log monitoring window, making the model calculation process and results transparent.

Images