Underground water seepage analysis and prediction method based on physical-data cooperative driving

Abstract

The invention discloses an underground water seepage analysis and prediction method based on physical-data cooperative driving, and belongs to the field of underground water seepage, and the method comprises the following steps: establishing a hydrogeological numerical model, carrying out seepage simulation calculation, and carrying out comparison verification with field monitoring data to realize accurate mapping; meanwhile, a machine learning data set is constructed by utilizing a numerical simulation result, a prediction model based on a Stacking integrated learning structure is established in combination with field data of a construction roadway, a water curtain layer and an oil storage cavern layer, and the water seepage amount or the underground water level after excavation of a rock mass in front of a tunnel face is predicted by taking geological, hydrological and construction parameters as input. According to the method, a physical mechanism and data driving are fused, the prediction precision and interpretability are remarkably improved, and a scientific basis is provided for cave depot project grouting and excavation optimization.

Description

Technical Field

[0001] This invention belongs to the field of groundwater seepage, and in particular relates to a groundwater seepage analysis and prediction method based on physical-data collaborative driving. Background Technology

[0002] In water-sealed underground storage projects, the oil storage caverns are generally located in rock strata below the groundwater level, utilizing groundwater in rock fissures to form a natural water curtain barrier to prevent oil leakage. The characteristics of groundwater seepage directly affect the safety of the storage facility and the rationality of the grouting design; therefore, accurate analysis and prediction of groundwater seepage characteristics are crucial for ensuring project safety. Currently, groundwater seepage analysis mainly relies on theoretical analytical methods, empirical formulas, and numerical simulation methods. Theoretical and empirical methods are suitable for scenarios with relatively simple geological conditions, while numerical simulation methods can accurately describe seepage characteristics and are therefore widely used. However, with the expansion of cavern project scale and the increase in burial depth, underground structures and hydrogeological conditions become increasingly complex. Traditional methods struggle to effectively characterize the spatial heterogeneity and multi-field coupling effects of groundwater seepage, leading to significant deviations between predicted results and actual conditions.

[0003] Current research faces several challenges. Firstly, physics-driven numerical simulations rely on precise values ​​for rock mass permeability parameters and boundary conditions. However, the difficulty in obtaining parameters in unexcavated areas and limitations in field testing often lead to calculations that rely on human experience and introduce significant errors. Secondly, while data-driven machine learning methods can achieve efficient predictions using historical engineering data, they lack interpretability of the physical mechanisms of groundwater seepage and cannot maintain reliability under changing geological conditions. Therefore, existing technologies struggle to achieve a unified representation of data characteristics and physical mechanisms under complex geological conditions, hindering accurate prediction of groundwater seepage in the unexcavated section ahead of the tunnel face, and consequently limiting scientific decision-making in grouting seepage reduction and construction optimization. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a groundwater seepage analysis and prediction method based on physical-data collaborative driving, comprising:

[0005] A three-dimensional hydrogeological numerical model was established based on engineering and hydrogeological conditions, cavern layout and construction design parameters, and the model was assigned differentiated values.

[0006] The numerical model is used to simulate the groundwater seepage in the engineering area, analyze the spatiotemporal evolution of the seepage field under different excavation processes, and compare it with the actual seepage field monitoring data to achieve accurate mapping between the numerical model and the field monitoring. The numerical simulation results are used as the output index of the seepage field prediction dataset of the machine learning model.

[0007] Based on the geological conditions and construction layers of the underground project, field data of the construction tunnel layer, water curtain layer and oil storage cavern layer were collected; based on the sparse hydrological monitoring data, the groundwater level above the excavated tunnel section was obtained by using the Kriging interpolation method in the numerical model, and the subsequent groundwater level and seepage volume were obtained through numerical simulation results, finally forming a groundwater seepage prediction dataset based on machine learning.

[0008] Based on the dataset, a groundwater seepage prediction model based on the Stacking ensemble learning structure was established. The model takes geological, construction design and test data, hydrological monitoring data and other indicators of each layer as inputs and the seepage volume or groundwater level of the cavern after the rock mass in front of the tunnel face is excavated as output. The output indicators of the dataset are derived from the numerical simulation results of the seepage field. The model is then trained and tested.

[0009] Optionally, the numerical model includes lithology, faults, groundwater level, and excavation area of ​​the cavern, and differentiates the values ​​of the model by combining the rock mass permeability coefficient and hydrological monitoring information. The groundwater level is obtained by using the Kriging space interpolation method based on the sparse groundwater level monitoring well data on site.

[0010] Optionally, during the simulation calculation using the numerical model, the evolution law of the seepage field is analyzed, the numerical model is dynamically corrected based on the newly revealed geological conditions and groundwater level changes during the excavation of the cave, and the seepage simulation calculation is re-performed to update the prediction results. The numerical simulation results are compared with the on-site seepage monitoring data to verify the reliability of the numerical model. Finally, the numerical simulation results are used as the output index in the seepage field prediction dataset of the machine learning model.

[0011] Optionally, the field data for the construction tunnel layer, water curtain layer, and oil storage cavern layer include: obtaining rock mass strength index and integrity index, and calculating the rock mass quality grade using the BQ formula; obtaining the pre-grouting permeability based on the water pressure test results; and obtaining the groundwater level and seepage volume based on hydrological monitoring equipment. Specifically, based on sparse groundwater level well monitoring data, the initial groundwater level above the excavated cavern is obtained using the Kriging interpolation method in numerical simulation software, and the groundwater level and seepage volume in the subsequent seepage field prediction dataset are obtained through numerical simulation results.

[0012] Optionally, in the step of constructing the groundwater seepage prediction dataset, the input indicators for the construction tunnel layer and the water curtain layer are rock mass quality classification, pre-grouting permeability, pre-excavation groundwater level, excavation advance, tunnel depth, seepage volume of advance boreholes and seepage pressure, and the output indicators are the seepage volume of the tunnel or the groundwater level after the rock mass in front of the tunnel face is excavated. The output indicators in the seepage field prediction dataset are derived from the numerical simulation results of the seepage field.

[0013] Optionally, in the step of constructing a groundwater seepage prediction dataset, when the stratum is an oil-bearing cavern layer, the seepage pressure, water curtain hole permeability coefficient, initial static pressure, water curtain hole recharge volume and recharge pressure in the surrounding rock of the oil-bearing cavern are further used as supplementary input indicators to improve the data input system.

[0014] Optionally, before establishing a groundwater seepage prediction model, the collected raw data is cleaned, subjected to correlation analysis and normalization, and low-correlation features are removed using the Pearson correlation method, and the dimensions are unified using the deviation standardization method.

[0015] Optionally, in the step of establishing a prediction model based on the Stacking ensemble learning structure, the Stacking model includes a random forest model, a support vector machine model, a K-nearest neighbor model, and a multilayer perceptron model as basic learners, and an extreme gradient boosting model (XGBoost) as a meta-learner.

[0016] Optionally, the mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²) can be used. 2 The predictive performance of the model was evaluated, and the accuracy of the prediction model was verified by comparing it with the seepage data after excavation.

[0017] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0018] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0019] Compared with the prior art, the present invention has the following advantages and technical effects:

[0020] This invention establishes a physical-data collaboratively driven method for groundwater seepage analysis and prediction, combining machine learning models with numerical simulation methods to achieve complementarity between data-driven and physical-driven approaches. Based on limited data from groundwater monitoring wells, this method uses numerical software to obtain the initial groundwater level in the engineering area, then the groundwater level above the excavation location of the tunnel, and uses numerical simulation software to perform numerical simulation of the excavation seepage field. It analyzes the spatiotemporal evolution of the seepage field in the engineering area, thereby achieving a precise mapping between the numerical results and actual seepage monitoring. The numerical simulation results of the seepage field are used as output indicators in the seepage field prediction dataset of the machine learning model. By collecting other field data from different strata, machine learning is used to quickly obtain the seepage volume and groundwater level of the cavern after the rock mass in front of the tunnel face is excavated. This enables scientific prediction of groundwater seepage after the rock mass in front of the tunnel face is excavated. Through this collaborative mechanism, the problems of strong parameter dependence and high degree of human experience in traditional numerical simulation can be effectively overcome. It also makes up for the lack of physical interpretability of machine learning methods, and achieves high-precision, interpretable and updatable prediction of groundwater seepage characteristics under complex geological conditions, providing a scientific basis for optimizing the grouting and excavation scheme of the cavern. Attached Figure Description

[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0022] Figure 1 This is a flowchart illustrating a research method for groundwater seepage analysis and prediction based on a physical-data dual-drive approach according to an embodiment of the present invention.

[0023] Figure 2 These are three elevation views of an embodiment of the present invention;

[0024] Figure 3 This is a data-driven groundwater seepage prediction index diagram according to an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the Stacking model according to an embodiment of the present invention;

[0026] Figure 5 This is an initial numerical model diagram of the engineering project area according to an embodiment of the present invention;

[0027] Figure 6 This is a front view of the numerical simulation of the groundwater seepage field according to an embodiment of the present invention;

[0028] Figure 7 This is a numerical simulation profile of the groundwater seepage field according to an embodiment of the present invention;

[0029] Among them, 1. oil storage cavern; 2. water curtain tunnel; 3. water curtain hole; 4. construction tunnel; 5. advance exploration hole. Detailed Implementation

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0031] 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.

[0032] Example 1

[0033] This embodiment provides a groundwater seepage analysis and prediction method based on physical-data collaborative driving, including:

[0034] A refined three-dimensional hydrogeological numerical model was established based on engineering geological and hydrogeological conditions, cavern layout and construction design parameters, and the model was assigned differentiated values.

[0035] The numerical model was used to simulate the groundwater seepage in the engineering area, analyze the spatiotemporal evolution of the seepage field under different excavation processes, and compare it with the seepage field monitoring data to verify the reliability of the numerical model and the seepage field results; and the numerical simulation results were used as the output index in the training set of the machine learning model.

[0036] Based on the geological conditions and construction layers of the underground project, field data of the construction tunnel layer, water curtain layer and oil storage cavern layer were collected, and groundwater seepage prediction dataset was constructed by combining the results of numerical simulation of seepage field.

[0037] Based on the dataset, a groundwater seepage prediction model based on the Stacking ensemble learning structure was established. The geological, hydrological and seepage characteristic parameters of each layer were used as inputs, and the seepage volume or groundwater level of the cavern after the rock mass in front of the tunnel face was excavated was used as outputs. The machine learning model was trained and tested, and the prediction performance of the model was evaluated from multiple aspects through different indicators. The prediction accuracy of the model was verified by comparing it with the seepage field data after excavation.

[0038] Establish a research system for groundwater seepage analysis and prediction methods driven by physics and data collaboration, combining machine learning methods with numerical simulation methods to achieve accurate analysis and prediction of groundwater seepage.

[0039] In terms of numerical simulation, an initial numerical model was constructed based on the on-site engineering and hydrogeological conditions, cavern layout, and construction design parameters. The established numerical model included: lithology, faults, groundwater level, and the cavern excavation area. The groundwater level was obtained from sparse hydrological monitoring data using the Kriging interpolation method in the numerical model. The model was then meshed, and differentiated values ​​were assigned based on rock permeability coefficients and hydrological monitoring information. The numerical model was continuously improved based on newly revealed geological structures during construction. Subsequently, groundwater seepage was numerically simulated, and the model was verified to meet requirements. By numerically simulating groundwater seepage under different excavation stages, the spatiotemporal evolution of the groundwater seepage field in the project area was obtained. This was compared and analyzed with actual seepage field monitoring data to verify the reliability of the numerical model. This provides a scientific and reliable basis for grouting seepage reduction and optimization of excavation construction schemes, thereby improving the scientific rigor and accuracy of cavern groundwater seepage analysis. The numerical simulation results were used as output indicators in the machine learning model dataset.

[0040] In the field of machine learning, the data collection for building machine learning algorithms is divided into three layers: the construction tunnel layer, the water curtain layer, and the oil storage cavern layer. In the construction tunnel layer and the water curtain layer, indicators such as geological conditions, construction design and experiments, and hydrological monitoring data are used as inputs for machine learning algorithms.

[0041] In the oil storage cavern layer, in addition to the data mentioned above, it is also necessary to collect data on the water curtain layer. The output will be the cavern seepage volume or groundwater level after excavation of the rock mass in front of the cavern face. The predictive performance of the model will be evaluated using different indicators, and compared with the seepage data after excavation, in order to achieve a scientific, objective, and rapid prediction of groundwater seepage in the cavern. The specific technical solution is as follows.

[0042] Data collection included geological conditions, field tests, and hydrological monitoring information from the project site. Based on this information, a refined three-dimensional hydrogeological model was constructed. Using sparse hydrological monitoring data, the initial groundwater level of the project area was obtained through Kriging interpolation in the numerical model, leading to a more accurate groundwater level above the excavated tunnel. The groundwater level and seepage volume in the subsequent seepage field prediction dataset were obtained through numerical simulation results. A groundwater seepage prediction dataset for the tunnel was constructed by combining the numerical simulation seepage field results. In the construction tunnel and water curtain layers: seven indicators were used as inputs: rock mass quality classification, pre-grouting permeability, pre-excavation groundwater level, excavation advance, tunnel depth, seepage volume from advance boreholes, and seepage pressure from advance boreholes. The outputs were either the tunnel seepage volume or the groundwater level after excavation of the rock mass in front of the tunnel face. In addition to the above-mentioned inputs, the water curtain permeability coefficient, initial static pressure, water inlet flow and pressure, and seepage pressure in the surrounding rock of the oil-bearing cavern are also required for the predicted cavern section. Data collected from the project site were cleaned and processed. Pearson correlation analysis was used to analyze the correlation of the indicators in the database. Deviation standardization was used to normalize the collected data, constructing a cavern groundwater seepage prediction dataset, which was then divided into training and testing sets in a 7:3 ratio.

[0043] A Stacking model is constructed, and grid search is used to optimize the hyperparameters of the model to obtain the optimal hyperparameters. K-fold cross-validation is used to further avoid overfitting and improve the generalization ability of the model. Random forest, support vector machine and other models are also built for comparative analysis.

[0044] The training set is input into the model for training. After the model is trained, the performance of the trained groundwater seepage prediction model is evaluated using the test set.

[0045] Choose the mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²). 2 As an evaluation index for prediction effectiveness, it is used to evaluate the error between the actual value of seepage or groundwater level after excavation of the rock mass in front of the tunnel face and the model prediction value.

[0046] This invention applies data-driven and physics-driven approaches to the analysis and prediction of groundwater seepage, providing a feasible method for quickly, objectively, and scientifically obtaining groundwater seepage information after the excavation of the rock mass in front of the tunnel face. It also offers a new approach and method for the analysis and prediction of groundwater seepage under complex geological conditions.

[0047] Example 2

[0048] like Figure 1 As shown, this embodiment provides a groundwater seepage analysis and prediction method based on physical-data collaborative driving, including:

[0049] A numerical model was established based on engineering and hydrogeological conditions, cavern layout, and construction design parameters. The established numerical model includes: lithology, faults, and the cavern excavation area. The model was further meshed. Differential values ​​were assigned to the model based on parameters such as rock mass permeability coefficient and hydrological monitoring information (groundwater level, seepage volume). The specific methods for obtaining this information are as follows:

[0050] Engineering geological conditions: By using various data such as preliminary geological survey reports, tunnel exposure during construction, and advanced borehole television imaging, we obtained information on the lithology and faults of the strata in the project area.

[0051] Hydrological monitoring information: Groundwater levels are obtained using pressure level gauges and monitoring boreholes. Initial groundwater levels in the project area are obtained using the Kriging interpolation method in numerical simulation software based on sparse monitoring borehole data. Seepage volume is obtained through pumping tests of advanced boreholes.

[0052] Rock permeability coefficient: The permeability coefficient of the surrounding rock is mainly obtained through water pressure tests of advance exploratory boreholes and grouting holes.

[0053] Cavern layout and construction design parameters: Obtain the cavern layout, dimensions, excavation progress, and construction parameters from the construction design plan.

[0054] First, based on preliminary exploration data, including stratigraphic lithology, fault information, and groundwater level monitoring well data, the initial groundwater level of the project area was obtained using the Kriging interpolation method based on the sparse groundwater level monitoring well data. This led to the establishment of an initial groundwater seepage analysis model for the project area. Subsequently, a mesh was generated, and the constructed mesh model was imported into finite element numerical simulation software. Considering the heterogeneity of rock mass seepage caused by discontinuous weak structural surfaces such as joints, fissures, weak interlayers, and faults, the boundary conditions of the model were determined, and differential values ​​were assigned to the model mesh elements based on the rock mass permeability coefficient and hydrological monitoring information. In the finite element numerical simulation software, the equivalent continuous medium method was used to simulate the underground rock mass seepage field. The equivalent continuous medium seepage model uniformly distributes the water flowing within the model's fissures throughout the entire rock mass structure in an equivalent manner. The fluid seepage in the medium is calculated using Darcy's law of saturation, with the specific formula as follows:

[0055] ;

[0056] In the formula: For effective porosity, For seepage velocity, Permeation tensor, For the total water head.

[0057] As the cavern is excavated, the model is continuously revised and updated based on new geological findings (fault extension range, alteration zone range), such as... Figure 5 As shown. Only when the numerical simulation results match the actual seepage conditions on site can numerical simulations of groundwater seepage during cavern excavation be conducted. For example... Figures 6-7 As shown, this study simulates groundwater seepage in a cavern at different excavation stages, analyzes the spatiotemporal evolution of the seepage field under different excavation processes, calculates the seepage volume and groundwater level of the surrounding rock of the cavern using the numerical simulation results, and compares them with actual seepage field monitoring data to verify the reliability of the numerical model. This achieves accurate mapping between the numerical model and field monitoring, providing a scientific and reliable basis for grouting seepage reduction and optimization of excavation construction schemes. The results of the numerical simulation are used as output indicators in the seepage field prediction dataset of the machine learning model. Based on the geological conditions and construction layers of the cavern project, field data of the construction tunnel layer, water curtain layer, and oil storage cavern layer, as well as the seepage calculation results of the numerical simulation, are collected to construct a groundwater seepage prediction dataset.

[0058] Based on the dataset, a groundwater seepage prediction model based on the Stacking ensemble learning structure was established. The geological conditions of each layer, cavern location and field test data, and hydrological monitoring data were used as inputs. If the cavern is located in an oil storage cavern layer, water curtain layer data also needs to be added. The seepage volume of the cavern or the groundwater level after the rock mass in front of the cavern face is used as the output to train and test the model.

[0059] Data-driven data collection: such as Figure 2 As shown, data collection at the three levels is mainly divided into: the construction tunnel level, the water curtain level, and the oil storage cavern level. Construction tunnel 4 is at the top level, water curtain tunnel 2 is in the middle level, and water curtain borehole 3 is located on both sides of the water curtain tunnel and is parallel to the axis of oil storage cavern 1, which is at the bottom level. The advance exploratory borehole 5, drilled in front of the oil storage cavern face, is located at the very front, forming a spatial arrangement sequence of "upper construction tunnel—middle water curtain tunnel—lower oil storage cavern—advance exploratory borehole". For the construction tunnel level and the water curtain level, data from the project site needs to be collected, including geological conditions, cavern location and test data, hydrological monitoring data, and numerical simulation results. For the oil storage cavern level, in addition to the above information, data on the water curtain level and seepage pressure also needs to be collected. Specific methods for obtaining characteristic factor information are as follows:

[0060] Geological condition data: Strength indicators (R) are obtained through rock point load or rebound tests. c ), and the rock mass integrity index (K) was obtained through acoustic testing of rock masses and rock blocks in the same area. v), using the BQ formula (BQ=100+3R) c +250K v ) Calculate the rock mass quality grade.

[0061] Cavern location and field test data: The permeability of the advance exploratory boreholes and grouting holes was obtained through water pressure tests, and the excavation progress and cavern depth were obtained based on the construction design scheme.

[0062] Hydrological monitoring data: Groundwater levels were obtained from monitoring wells using pressure gauges. Based on the sparse monitoring well data, the initial groundwater level of the project area was obtained using the Kriging interpolation method in numerical simulation software, which then yielded the groundwater level data above the excavated cavern. Seepage volume and pressure were obtained through pumping tests in advance boreholes and the installation of vibrating wire piezometers in boreholes. Subsequent groundwater levels and seepage volumes were obtained through numerical simulation results. Groundwater seepage pressure was obtained through piezometers installed in the surrounding rock of the oil storage cavern.

[0063] Water curtain layer data: The permeability coefficient and initial static pressure of the water curtain holes were obtained through water injection-drop tests; to ensure an effective water seal environment, the water supply volume and pressure of the water curtain holes were obtained through continuous water replenishment.

[0064] Construct a groundwater seepage prediction dataset: such as Figure 3 As shown, in the construction tunnel layer and water curtain layer, seven indicators were used as inputs: rock mass quality grading, pre-grouting permeability, pre-excavation groundwater level, excavation progress, tunnel depth, seepage volume of advance exploratory boreholes, and seepage pressure of advance exploratory boreholes. In the oil storage cavern layer, in addition to the above seven indicators, twelve additional indicators were added as inputs: seepage pressure in the surrounding rock of the oil storage cavern, permeability coefficient of the water curtain boreholes, initial static pressure, water replenishment volume and replenishment pressure of the water curtain boreholes. The outputs were the cavern seepage volume or groundwater level after the excavation of the rock mass in front of the tunnel face. Data cleaning was performed on the data collected at the project site, and statistical analysis was conducted to determine the distribution patterns and fluctuations of each characteristic factor.

[0065] Pearson correlation analysis is used to perform correlation analysis on the dataset, analyzing the correlation between input and output features. Features with low correlation are removed, while those with high correlation are retained, thus improving the model's prediction accuracy and speed. The formula for Pearson correlation analysis is as follows:

[0066] ;

[0067] In the formula: It is the Pearson correlation coefficient; and It means selecting any two of the parameters; and the covariance between the two variables; , and They are , and Expectations; and and yes and The standard deviation.

[0068] Because this embodiment collects different types of feature factors, there is a problem of inconsistent dimensions among the feature factors. Data normalization can effectively eliminate this inconsistency and improve the model's prediction accuracy and performance. Therefore, this embodiment uses deviation normalization (min-max) to process the collected data, as shown in the following formula:

[0069] ;

[0070] In the formula: The first of the current indicators One element, For the first Minimum value of each indicator; For the first The maximum value of each indicator; The value is a standardized value, ranging from [0,1].

[0071] Dataset partitioning: The processed dataset is divided into a training set and a test set in a 7:3 ratio. The training set is used for training the machine learning model, while the test set is used for prediction by the machine learning model. The results of numerical simulation of cavern seepage or groundwater level are used as the output indicators in the machine learning model dataset.

[0072] Model building: Stacking is an ensemble learning method that improves overall prediction accuracy by combining the predictions of multiple base models, such as... Figure 4 As shown, the XGBoost model learns from the original data using multiple base learners, and then uses the outputs of these base learners as input to the second-layer model for fitting. Compared with a single model, the XGBoost model has higher prediction accuracy and stability. Therefore, this embodiment uses four models—RF, SVM, KNN, and MLP—as base classifiers, and the XGBoost model as the meta-classifier. Grid search is used to optimize the hyperparameters of each model to obtain the optimal hyperparameters. K-fold cross-validation is used to further avoid overfitting and increase the effectiveness of model evaluation. Random forest and support vector machine models are also established for comparative analysis to verify the model's superiority.

[0073] Model training and prediction: The training set was input into the Stacking model for training. After training, the test set was input into the trained groundwater seepage prediction model for performance evaluation. The other comparison models were trained and predicted using the same steps. Finally, the accuracy of the machine learning model was verified by comparing it with the seepage field monitoring data after excavation.

[0074] For model evaluation, this embodiment selects mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²). 2 The evaluation index is used to assess the accuracy and reliability of the model's predictions. If all prediction indices are below the threshold, the model needs to be retrained; if the evaluation index is above the threshold, the model is used as the final model for predicting groundwater seepage after excavation of the rock mass in front of the tunnel face. The formula for the evaluation index is as follows:

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] in, For the true value, The average true value, is the predicted value, and N is the sample size.

[0080] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0081] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0082] The above are merely preferred embodiments of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A groundwater seepage analysis and prediction method based on physical-data collaborative driving, characterized in that, include: Numerical models were established based on engineering and hydrogeological conditions, cavern layout, and construction design parameters, and differentiated values ​​were assigned to the models. The numerical model is used to simulate the groundwater seepage in the engineering area, analyze the spatiotemporal evolution of the seepage field, and compare it with the actual seepage field monitoring data to achieve accurate mapping between the numerical model and the field monitoring; and the simulation results are used as the output index of the machine learning model dataset. Based on the geological conditions and construction layers of the underground project, field data of the construction tunnel layer, water curtain layer and oil storage cavern layer were collected, and groundwater seepage prediction dataset was constructed by combining numerical simulation information. Based on the dataset, a groundwater seepage prediction model based on the Stacking ensemble learning structure is established. The model takes parameters such as geological conditions, construction design and test data, and hydrological monitoring data of each layer as inputs, and takes the seepage volume of the surrounding rock of the cavern or the groundwater level after the rock mass in front of the tunnel face as outputs. The output indicators are derived from the numerical simulation results of the seepage field. The model is trained and tested to achieve accurate prediction of the seepage field.

2. The method according to claim 1, characterized in that, The numerical model includes lithology, faults, groundwater level, and excavation area of ​​the cavern, and differentiates the values ​​of the model by combining the rock mass permeability coefficient and hydrological monitoring information.

3. The method according to claim 2, characterized in that, In the process of using numerical models for simulation calculations, the evolution law of seepage field in the engineering area is analyzed; the numerical model is dynamically corrected based on the newly revealed geological conditions and groundwater level changes during the excavation of the cave, and seepage simulation calculations are performed again to update the prediction results. The results are compared with actual seepage field monitoring data to verify the reliability of the numerical model, and the numerical simulation results are used as the output index in the prediction dataset of the machine learning model.

4. The method according to claim 1, characterized in that, The collection of field data for the construction tunnel layer, water curtain layer, and oil storage cavern layer includes: obtaining rock mass strength and integrity indices, calculating the rock mass quality grade using the BQ formula; obtaining the pre-grouting permeability based on the water pressure test results; and obtaining the groundwater level and seepage volume based on hydrological monitoring equipment. Specifically, based on sparse groundwater level well monitoring data, the initial groundwater level above the excavated cavern is obtained using the Kriging interpolation method in numerical simulation software. The subsequent groundwater level and seepage volume are obtained through numerical simulation results and used as output indicators in the machine learning model dataset.

5. The method according to claim 1, characterized in that, In the step of constructing the groundwater seepage prediction dataset, the input indicators for the construction tunnel layer and the water curtain layer are rock mass quality classification, pre-grouting permeability, pre-excavation groundwater level, excavation advance, tunnel depth, seepage volume of advance boreholes and seepage pressure. The output indicators are the seepage volume of the tunnel or the groundwater level after the rock mass in front of the tunnel face is excavated. The output indicators in the dataset are derived from the numerical simulation results of the seepage field.

6. The method according to claim 1, characterized in that, In the step of constructing a groundwater seepage prediction dataset, when the stratum is an oil-bearing cavern, the seepage pressure, water curtain hole permeability coefficient, initial static pressure, water curtain hole recharge volume and recharge pressure in the surrounding rock of the oil-bearing cavern are further used as supplementary input indicators to improve the data input system.

7. The method according to claim 1, characterized in that, Before establishing a groundwater seepage prediction model, the collected raw data were cleaned, subjected to correlation analysis and normalization. The Pearson correlation method was used to remove low-correlation features, and the deviation standardization method was used to unify the dimensions.

8. The method according to claim 1, characterized in that, In the step of establishing a prediction model based on the Stacking ensemble learning structure, the Stacking model includes a random forest model, a support vector machine model, a K-nearest neighbor model, and a multilayer perceptron model as basic learners, and an extreme gradient boosting model as a meta-learner.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-8.

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

Images