A Method and System for Dynamically Predicting Tunnel Lining Stress Based on GMS-MODFLOW and ABAQUS

Abstract

The present invention relates to the technical fields of tunnel engineering and hydrogeology, and provides a method and system for dynamically predicting the stress of tunnel lining based on GMS-MODFLOW and ABAQUS. The method dynamically predicts the distribution of tunnel lining stress through the linkage simulation of the GMS-MODFLOW model and the ABAQUS finite element model, in combination with the seepage field data under various precipitation conditions. At the same time, the long short-term memory network LSTM is used to learn and model the non-linear relationship between rainfall and lining stress, so as to provide accurate real-time prediction and risk warning. By establishing a multi-parameter database, rainfall, seepage field distribution and stress data are extracted to construct a time series model to realize the dynamic monitoring of tunnel lining stress. This method can effectively improve the safety, reliability and economic benefits of tunnel engineering, and meet the requirements of hydrogeological risk prediction and structural safety assessment in modern tunnel construction and operation.

Description

Technical Field

[0001] The present invention relates to the technical fields of tunnel engineering and hydrogeology. Specifically, it relates to a method and system for dynamically predicting the stress of tunnel lining based on GMS-MODFLOW and ABAQUS. Background Art

[0002] Tunnel engineering is an important part of transportation infrastructure construction. However, the construction and operation safety problems of tunnel engineering under complex geological conditions are becoming increasingly prominent. Especially in water-rich geological environments, the safety of tunnel lining structures is closely related to the dynamic changes of the seepage field of surrounding rocks. In a water-rich environment, tunnel engineering faces the coupling effect of the stress field and the seepage field. The dynamic change of the groundwater level will cause continuous disturbance of the system equilibrium state. This complex interaction is particularly obvious when the rainfall changes rapidly. Extreme rainfall events may cause a sharp increase in the tunnel water inflow, resulting in an increase in the lining water pressure, and further causing structural damage or even catastrophic consequences. In recent years, accidents such as tunnel cracking, water leakage, and roadbed heaving caused by heavy rainfall have been frequent, which not only threaten the safe operation of the tunnel but also cause huge economic losses and maintenance pressure.

[0003] Under this background, the stress distribution law of tunnel lining structures and their dynamic relationship with rainfall have become technical problems that need to be solved urgently. Existing methods mostly analyze the seepage field and stress distribution through physical models or numerical simulation means. However, these methods usually have difficulty providing reliable prediction results quickly and efficiently when facing complex geological conditions and diverse rainfall scenarios. In addition, due to the lack of real-time prediction tools coupled with rainfall conditions, the disaster warning ability during the tunnel operation period is limited. The limitations of traditional models have put forward higher requirements for the scientific nature of tunnel waterproof and drainage design and risk management during the construction and operation period.

[0004] Therefore, constructing a comprehensive method based on multi-parameter modeling and intelligent prediction, integrating seepage field simulation technology, structural mechanics analysis tools, and machine learning technology, and establishing a mapping model between rainfall and tunnel lining stress has become an effective way to solve the above problems. This method can realize the real-time prediction and risk warning of tunnel lining stress, provide scientific support for the safety management and design optimization of tunnel engineering, and has important application prospects and engineering value. Summary of the Invention

[0005] The content of the present invention is to provide a method and system for dynamically predicting the stress of tunnel lining based on GMS-MODFLOW and ABAQUS, which can solve the problems of insufficient prediction accuracy, poor real-time performance, and difficulty in effectively combining complex rainfall conditions for dynamic modeling in the prior art.

[0006] A dynamic prediction method for tunnel lining stress based on GMS-MODFLOW and ABAQUS provided by the present invention includes the following steps:

[0007] S1: Determine the research area and collect data;

[0008] S2: Establish a GMS-MODFLOW model, introduce the Richards equation to consider the unsaturated seepage process, and simulate the change of matrix suction during rainfall infiltration; use the Monte Carlo method for uncertainty analysis, quantify the influence of hydrogeological parameter uncertainty on the results and output the 95% confidence interval;

[0009] S3: Establish an ABAQUS model, refine the modeling of the lining, and construct a steel-concrete interface contact model considering the bond-slip effect;

[0010] S4: Establish a Python script to link the ABAQUS model with the GMS-MODFLOW seepage field results;

[0011] S5: Establish a parallel computing framework, use MPI distributed computing to accelerate the coupled simulation system of GMS-MODFLOW and ABAQUS, calculate the stress distribution characteristics of the tunnel lining under different rainfall conditions, and obtain the stress state of each position of the lining;

[0012] S6: Organize the calculation results under different rainfall conditions into a multi-parameter database, and the database content includes rainfall, the distribution characteristics of the seepage field in the tunnel site area, and the corresponding tunnel lining stress distribution data;

[0013] S7: Use the long short-term memory network LSTM to train the organized data, learn the dynamic relationship between rainfall and tunnel lining stress, and establish a lining stress prediction model;

[0014] S8: Verify and optimize the model;

[0015] S9: Save the optimized LSTM model and embed it into the tunnel lining stress prediction system to form an intelligent prediction module;

[0016] S10: By inputting real-time rainfall data, use the optimized LSTM model to predict the lining stress distribution at different tunnel positions, and realize the rapid prediction of the dynamic change of tunnel lining stress.

[0017] Preferably, in step S1, the data includes geological data, hydrogeological parameters, meteorological data, DEM data and groundwater level monitoring data.

[0018] Preferably, in step S2, it specifically includes the following steps:

[0019] S201: Establish a conceptual model;

[0020] S202: Define boundary conditions and clarify the distribution of aquifers and aquitards;

[0021] S203: Generate a three-dimensional grid model, divide grid cells and perform encryption processing on the tunnel area;

[0022] S204: Assign initial boundary conditions and establish a static model to simulate the static distribution of groundwater;

[0023] S205: Introduce the Richards equation to consider unsaturated seepage;

[0024] S206: Dynamically run the model, input data under different rainfall conditions, simulate the dynamic seepage field, and analyze parameter uncertainty using the Monte Carlo method, and output the results of the 95% confidence interval.

[0025] Preferably, in step S3, it specifically includes the following steps:

[0026] S301: According to the seepage field simulation results of the GMS-MODFLOW model, import the groundwater pressure distribution data into ABAQUS to establish a three-dimensional finite element model of the tunnel;

[0027] S302: Perform refined modeling on the lining, establish a steel-concrete interface contact model, and fully consider the bond-slip effect between the steel and the concrete;

[0028] S303: Set the tunnel lining material parameters and the initial stress field to simulate the stress changes of the tunnel lining under different precipitation conditions.

[0029] Preferably, in step S4, it specifically includes the following steps:

[0030] S401: Use a Python script to read the HED format file generated by GMS-MODFLOW, convert the file to CSV format, and extract the required head data;

[0031] S402: Extract the required time step data;

[0032] S403: According to the head corresponding to the original coordinates of the GMS-MODFLOW grid division, organize the head data according to the ABAQUS grid division, and use the interpolation method to map the head data of the GMS grid to the ABAQUS grid to ensure the reasonable distribution of the head data within the calculation domain of ABAQUS;

[0033] S404: Connect the water head data to the ABAQUS model through a Python script, and use the API interface of ABAQUS to input the sorted water head data into the ABAQUS finite element model to reflect the groundwater pressure under different precipitation conditions.

[0034] Preferably, in step S5, the stress state includes the water pressure on the inner and outer surfaces, the surrounding rock pressure, and the changes in the tensile stress and shear stress of the lining.

[0035] Preferably, in step S6, preprocess the extracted multi-dimensional data set, including normalization, time series reconstruction, and abnormal data cleaning, to construct a training data set suitable for time series analysis.

[0036] Preferably, in step S8, evaluate the model performance through cross-validation, and optimize the model parameters in combination with the loss function to improve the prediction accuracy and generalization ability of the model.

[0037] A tunnel lining stress dynamic prediction system based on GMS-MODFLOW and ABAQUS provided by the present invention adopts the above-mentioned tunnel lining stress dynamic prediction method based on GMS-MODFLOW and ABAQUS.

[0038] Through the linked simulation of the GMS-MODFLOW model and the ABAQUS finite element model, combined with the seepage field data under various precipitation conditions, the present invention dynamically predicts the distribution of tunnel lining stress. At the same time, the long short-term memory network LSTM is used to learn and model the non-linear relationship between rainfall and lining stress, so as to provide accurate real-time prediction and risk warning. By establishing a multi-parameter database, extracting rainfall, seepage field distribution, and stress data, and constructing a time series model, the dynamic monitoring of tunnel lining stress is realized. The present invention can effectively improve the safety, reliability, and economic benefits of tunnel engineering, and meet the requirements of hydrological risk prediction and structural safety assessment in modern tunnel construction and operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of a tunnel lining stress dynamic prediction method based on GMS-MODFLOW and ABAQUS in the embodiment.

[0040] Figure 2 It is a schematic diagram of the GMS-MODFLOW model in the embodiment.

[0041] Figure 3 It is a schematic diagram of the ABAQUS model in the embodiment.

[0042] Figure 4 It is a schematic diagram of the ABAQUS preprocessing Python script process in the embodiment.

[0043] Figure 5 It is a schematic diagram of the principle of the long short-term memory network (LSTM) in the embodiment. Specific implementation manners

[0044] To further understand the content of the present invention, the present invention will be described in detail in combination with the accompanying drawings and embodiments. It should be understood that the embodiments are only for explaining the present invention rather than limiting it.

[0045] Embodiment

[0046] As Figure 1 shown, this embodiment provides a method for dynamically predicting the stress of tunnel lining based on GMS-MODFLOW and ABAQUS, including the following steps:

[0047] S1: Determine the research area and collect data, and obtain the data related to the GMS-MODFLOW model. These data include geological data, hydrogeological parameters, meteorological data, DEM data, and groundwater level monitoring data. Specifically: geological data such as the regional stratigraphic distribution, lithological characteristics, the structure of aquifers and aquitards; hydrogeological parameters such as groundwater level, hydraulic conductivity, and storage coefficient; meteorological data such as historical rainfall, rainfall intensity, and evaporation (derived from public meteorological data or regional monitoring stations); digital elevation model (DEM data) used to generate topographic boundary conditions; and groundwater level data collected by arranging groundwater monitoring wells. These monitoring data are transmitted and located through an online acquisition terminal to provide initial and real-time data support for the model.

[0048] Specifically, obtain the geological data, hydrogeological parameters, and meteorological data of the area around the tunnel. Arrange groundwater monitoring wells in the tunnel area, collect groundwater level data in real time, and transmit the data through the GPRS module.

[0049] S101: Conduct on-site investigations using geological exploration technologies such as ground-penetrating radar and seismic wave methods, obtain underground rock and soil samples through drilling, and analyze their physical and chemical properties; use satellite or aerial remote sensing images to analyze the surface morphology to assist in judging the distribution of underground rock strata.

[0050] S102: Combine historical hydrogeological data with on-site investigations to understand the recharge, runoff, and discharge conditions of groundwater in the area; arrange a certain number of groundwater level monitoring wells around the tunnel, install GPRS modules in the groundwater level monitoring wells, which can realize the remote transmission of monitoring data, and monitor the changes in groundwater level regularly or in real time.

[0051] S103: Install a weather station near the tunnel to monitor meteorological elements such as temperature, humidity, and rainfall, and provide real-time meteorological information; combine meteorological satellite remote sensing data to obtain the meteorological change trend in a large area.

[0052] S2: Establish a GMS-MODFLOW model as shown in Figure 2 the following; according to the acquired data, first establish a conceptual model of the tunnel site area, specifically including determining boundary conditions, clarifying the distribution of aquifers and aquitards in the study area, and dividing hydrogeological units; then, generate a three-dimensional grid model, divide grid cells and perform densification processing on the tunnel area; subsequently, assign the initial boundary conditions in the conceptual model to the three-dimensional grid model and run the GMS-MODFLOW model to simulate the static distribution state of regional groundwater and establish a three-dimensional groundwater static model; finally, input dynamic data, including rainfall amounts and rainfall intensities under different rainfall conditions, to obtain the distribution of the dynamic seepage field, including key results such as water level changes, velocity distributions, and hydraulic gradients; specifically as follows:

[0053] S201: Based on the collected data, establish a seepage field model for the tunnel site area to simulate the dynamic flow characteristics of groundwater under different precipitation conditions.

[0054] Specifically, based on the acquired geological data (including the distribution of rock and soil layers, permeability coefficients, and porosities), hydrological data (such as historical precipitation, groundwater levels, and surface runoff), and topographic data (elevation data of the tunnel site area), start constructing the seepage field model for the tunnel site area. The initial conditions of the model are determined by historical observation data or regional average water levels and are used to set the initial water level and groundwater pressure distribution.

[0055] Next, use the three-dimensional grid division tool in GMS to divide the tunnel site area into regular grid cells. The selection of grid size needs to be comprehensively considered according to the complexity of the terrain, the characteristics of the geological structure, and the requirements of computing resources. Especially in the area close to the tunnel site, due to the more complex groundwater flow characteristics, finer grid division is required to accurately capture local flow details.

[0056] Furthermore, the setting of hydrogeological parameters is directly related to the accuracy of the model. First, the permeability coefficient (K) is a key parameter affecting groundwater flow, and corresponding permeability coefficients need to be set for different rock and soil layers according to geological data. Second, the storage coefficient (S) reflects the water storage capacity of the aquifer under pressure changes, and this parameter is usually determined by experimental data or empirical values. In addition, source-sink terms such as rainfall infiltration, surface runoff, and well pumping need to be considered in the model. These source-sink terms are used to simulate the recharge and discharge processes of groundwater to ensure that the model can comprehensively reflect the dynamic changes of the groundwater system.

[0057] S202: Input different precipitation scenarios in GMS to generate seepage field distribution data, including hydraulic gradients, groundwater pressures, and velocities, etc.

[0058] Specifically, during the design phase of precipitation scenarios, multiple precipitation scenarios were developed based on historical precipitation data and meteorological prediction information to cover possible situations with different rainfall intensities and durations. These scenarios mainly include:

[0059] Extreme precipitation: Simulate rare heavy rainfall events in history. Such events may lead to a large amount of rainfall infiltration in a short period, thus having a significant impact on the groundwater system.

[0060] Heavy precipitation: The rainfall is significantly higher than the average, close to large rainfall events in history, simulating a 50% increase in the average annual precipitation.

[0061] Moderate precipitation: Represent the average annual precipitation or seasonal precipitation pattern. This is a relatively common precipitation situation and is used to evaluate the groundwater dynamics under normal precipitation conditions.

[0062] Drought scenario: Simulate drought conditions with reduced precipitation or continuous lack of precipitation. This situation may affect the recharge of groundwater, thereby affecting the groundwater level and flow velocity.

[0063] In addition to the classified precipitation scenarios, time series data were generated for each scenario to simulate the changes in precipitation over different time periods, so as to more realistically reflect the impact of the actual precipitation process on the groundwater system. During the input stage of the GMS-MODFLOW model, we input the designed precipitation scenarios as the source-sink terms of the model. Specifically, the precipitation infiltration rate was used as the recharge term of the groundwater system. During this process, ensure that other hydrogeological parameters of the model, such as the permeability coefficient and storage coefficient, remain consistent, and only adjust the precipitation input to isolate the evaluation of the impact of precipitation changes on the seepage field.

[0064] S3: Combine the actual geometric dimensions, burial depth, lining structure parameters of the tunnel and the mechanical properties of the surrounding rock mass to establish an ABAQUS model of the tunnel and its surrounding area, as Figure 3 shown below:

[0065] S301: According to the seepage field simulation results of the GMS-MODFLOW model, import the groundwater pressure distribution data into ABAQUS to establish a three-dimensional finite element model of the tunnel.

[0066] S302: Conduct refined modeling of the lining, establish a reinforcement-concrete interface contact model, and fully consider the bond-slip effect between the reinforcement and the concrete.

[0067] S303: Set the material parameters and initial stress field of the tunnel lining to simulate the stress changes of the tunnel lining under different precipitation conditions.

[0068] S4: Based on the GMS seepage field simulation results under various precipitation conditions, the Python script inputs the water head distribution around the tunnel dynamically through the API interface in ABAQUS. According to the rainfall intensity and the change of the groundwater level, the dynamic loading process of the seepage pressure on the tunnel lining is simulated. The process is as Figure 4 shown, and specifically includes the following steps:

[0069] S401: Use the Python script to read the HED format file generated by GMS-MODFLOW, convert the file into CSV format, and extract the required water head data; the specific code is as follows:

[0070]

[0071] S402: Extract the required time step data; the specific code is as follows:

[0072]

[0073] S403: Corresponding to the original coordinates of the water head according to the GMS-MODFLOW grid division, and sorting out the water head data according to the ABAQUS grid division. Using the interpolation method, map the water head data of the GMS grid to the ABAQUS grid to ensure the reasonable distribution of the water head data within the calculation domain of ABAQUS; the code is as follows:

[0074]

[0075]

[0076] S404: Connect the water head data to the ABAQUS model through the Python script, and use the API interface of ABAQUS to input the sorted water head data into the ABAQUS finite element model to reflect the groundwater pressure under different precipitation conditions. The code is as follows:

[0077]

[0078] S5: Establish a parallel computing framework, adopt MPI distributed computing to accelerate the joint simulation of GMS-MODFLOW and ABAQUS, calculate the stress distribution characteristics of the tunnel lining under different rainfall conditions, and obtain the stress states of each position of the lining, including the water pressure on the inner and outer surfaces, the surrounding rock pressure, and the changes in the tensile stress and shear stress of the lining.

[0079] S6: Organize the calculation results under different rainfall conditions into a multi-parameter database. The database content includes rainfall, the distribution characteristics of the seepage field in the tunnel site area, and the corresponding tunnel lining stress distribution data, providing basic data for the training and dynamic prediction of subsequent machine learning models.

[0080] S601: Extract rainfall amounts, seepage field distributions, and lining stress data under different precipitation conditions from the results output by the GMS-MODFLOW and ABAQUS models.

[0081] S602: Preprocess this data, perform normalization, outlier handling, and construct a training dataset suitable for time series analysis.

[0082] Use the Min-Max normalization formula:

[0083] ;

[0084] where is the normalized data, is the original data, , are the maximum and minimum values of the original data.

[0085] S7: Use a long short-term memory network (LSTM) to train the sorted data, learn the dynamic relationship between rainfall and tunnel lining stress, and establish a lining stress prediction model.

[0086] As Figure 5 shown, the long short-term memory network (LSTM) is used to process time series data, such as the change in tunnel lining stress over rainfall time. LSTM is suitable for extracting complex time series features of tunnel lining stress data by memorizing past states and handling long-term dependence problems. The specific network structure is represented by the following formula:

[0087] The calculation formula for the LSTM cell is:

[0088] ;

[0089] where, is the forget gate, controlling how much historical information to discard; is the weight matrix of the forget gate; is the bias vector of the forget gate; is the input gate, determining the degree of memory of the current information; is the weight matrix of the input gate; input gate bias vector; is the output gate, finally determining the output; is the weight matrix of the output gate; output gate bias vector; is the cell state; is the cell state vector of the LSTM cell at time step t -1; is the candidate cell state; is at time step tThe hidden state vector of the LSTM cell at -1; is the current t hidden state vector of the LSTM cell at the current time step; is the input vector at time step t ; The bias vector used to generate the candidate cell state; is the sigmoid activation function that limits the value between 0 and 1.

[0090] The collected rainfall, seepage field distribution, and tunnel lining stress datasets are input into the LSTM network for training. The input of the network is time - series data, and the output is the predicted value of the tunnel lining stress. The LSTM network learns the complex non - linear relationship between rainfall and lining stress through multiple iterations, thus achieving accurate stress prediction.

[0091] S8: Verify and optimize the model; evaluate the model performance through cross - validation and optimize the model parameters in combination with the loss function to improve the prediction accuracy and generalization ability of the model. Specifically, it includes the following:

[0092] S801: Input the validation set into the trained model for prediction;

[0093] S802: Use the loss function to compare the strength prediction result output by the model with the actual strength;

[0094] The mean squared error MSE is used as the loss function to measure the error between the network - predicted strength value and the true strength value. The formula is as follows:

[0095] ;

[0096] where, is the actual strength, is the strength predicted by the model.

[0097] S803: Adjust the weight parameters and network topology structure of the LSTM network based on the comparison results to optimize the model performance. Ensure its learning ability for the lining stress. Evaluate the network performance through the validation set. If the accuracy and recall rate reach the expected effect, stop the optimization.

[0098] Specifically, input the test data not involved in training into the model, calculate the strength value predicted by the model and compare it with the actually calculated tunnel lining stress to verify the accuracy and generalization ability of the model. The model can be further optimized by adjusting network parameters (such as learning rate, number of layers).

[0099] S9: Save the optimized LSTM model and embed it into the tunnel lining stress prediction system to form an intelligent prediction module.

[0100] S10: By inputting rainfall data in real time, the optimized LSTM model is used to predict the lining stress distribution at different tunnel locations, and the dynamic changes of tunnel lining stress can be quickly predicted. Specifically:

[0101] S1001: During tunnel construction and operation, real-time data such as rainfall and groundwater level are collected through online monitoring equipment;

[0102] S1002: Input the real-time data into the LSTM model for prediction and calculate the real-time stress distribution of the tunnel lining.

[0103] S1003: Based on the output results of the LSTM model and the safety threshold of the tunnel design, the changes in the tunnel lining stress are evaluated in real time;

[0104] S1004: When the lining stress exceeds the safety range, a risk warning is automatically triggered to prompt construction personnel and operation management departments to take corresponding measures.

[0105] This embodiment provides a tunnel lining stress dynamic prediction system based on GMS-MODFLOW and ABAQUS, which adopts the above-mentioned tunnel lining stress dynamic prediction method based on GMS-MODFLOW and ABAQUS.

[0106] This embodiment realizes high-precision modeling of the influence of rainfall on seepage field and lining stress through the linkage simulation of GMS-MODFLOW and ABAQUS, and combines the LSTM model to learn complex nonlinear relationships, ultimately realizing dynamic prediction and risk warning of lining stress.

[0107] The present invention and its embodiments are described schematically above, and the description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it and designs a structural method and an embodiment similar to the technical solution without creativity without departing from the purpose of the invention, they shall all fall within the protection scope of the present invention.

Claims

1. A tunnel lining stress dynamic prediction method based on GMS-MODFLOW and ABAQUS, characterized in that: The following steps are involved: S1: Determine the study area and collect data; S2: Establish a GMS-MODFLOW model, introduce the Richards equation to consider the unsaturated seepage process, and simulate the change of matrix suction during rainfall infiltration; use the Monte Carlo method to perform uncertainty analysis, quantify the impact of hydrogeological parameter uncertainty on the results and output a 95% confidence interval; S3: Establish an ABAQUS model, refine the lining model, and construct a steel-concrete interface contact model considering the bond-slip effect; Step S3 specifically includes the following steps: S301: According to the simulation results of the seepage field of the GMS-MODFLOW model, the groundwater pressure distribution data is imported into ABAQUS to establish a three-dimensional finite element model of the tunnel; S302: Carry out detailed modeling of the lining, establish a steel-concrete interface contact model, and fully consider the bond-slip effect between the steel and concrete; S303: Setting tunnel lining material parameters and initial stress field to simulate the stress changes of tunnel lining under different precipitation conditions; S4: Establish a Python script to link the ABAQUS model with the GMS-MODFLOW seepage field results; Step S4 specifically includes the following steps: S401: Use Python script to read the HED format file generated by GMS-MODFLOW and convert the file into CSV format to extract the required head data; S402: extracting required time step data; S403: divide the hydraulic head corresponding to the original coordinates according to the GMS-MODFLOW grid, sort the hydraulic head data according to the ABAQUS grid, and use the interpolation method to map the hydraulic head data of the GMS grid to the ABAQUS grid to ensure that the hydraulic head data is reasonably distributed within the calculation domain of ABAQUS; S404: Connect the water head data to the ABAQUS model through the Python script, and use the ABAQUS API interface to input the sorted water head data into the ABAQUS finite element model to reflect the groundwater pressure under different precipitation conditions; S5: Establish a parallel computing framework and use MPI distributed computing to accelerate the joint simulation of GMS-MODFLOW and ABAQUS, calculate the stress distribution characteristics of the tunnel lining under different rainfall conditions, and obtain the stress state of each position of the lining; S6: The calculation results under different rainfall conditions are organized into a multi-parameter database, which includes rainfall, seepage field distribution characteristics in the tunnel site area, and corresponding tunnel lining stress distribution data; S7: Use the long short-term memory network (LSTM) to train the sorted data, learn the dynamic relationship between rainfall and tunnel lining stress, and establish a lining stress prediction model; S8: Validate and optimize the model; S9: Save the optimized LSTM model and embed it into the tunnel lining stress prediction system to form an intelligent prediction module; S10: By inputting rainfall data in real time, the optimized LSTM model is used to predict the lining stress distribution at different tunnel locations, thus realizing rapid prediction of the dynamic changes of tunnel lining stress.

2. The method for dynamic prediction of tunnel lining stress based on GMS-MODFLOW and ABAQUS according to claim 1 is characterized in that: In step S1, the data includes geological data, hydrogeological parameters, meteorological data, DEM data and groundwater level monitoring data.

3. The method for dynamic prediction of tunnel lining stress based on GMS-MODFLOW and ABAQUS according to claim 1 is characterized in that: Step S2 specifically includes the following steps: S201: Establish conceptual model; S202: Define boundary conditions and clarify the distribution of aquifers and aquitards; S203: Generate a three-dimensional grid model, divide the grid units and perform encryption processing on the tunnel area; S204: Assign initial boundary conditions and establish a static model to simulate the static distribution of groundwater; S205: Introduce Richards equation to consider unsaturated seepage; S206: Dynamically run the model, input data under different rainfall conditions, simulate the dynamic seepage field, and use the Monte Carlo method to analyze parameter uncertainties and output 95% confidence interval results.

4. The method for dynamic prediction of tunnel lining stress based on GMS-MODFLOW and ABAQUS according to claim 1 is characterized in that: In step S5, the stress state includes changes in internal and external surface water pressure, surrounding rock pressure, and lining tensile stress and shear stress.

5. The method for dynamic prediction of tunnel lining stress based on GMS-MODFLOW and ABAQUS according to claim 1 is characterized in that: In step S6, the extracted multidimensional data set is preprocessed, including normalization, time series reconstruction and abnormal data cleaning, to construct a training data set suitable for time series analysis.

6. The method for dynamic prediction of tunnel lining stress based on GMS-MODFLOW and ABAQUS according to claim 1 is characterized in that: In step S8, the model performance is evaluated through cross-validation, and the model parameters are optimized in combination with the loss function to improve the prediction accuracy and generalization ability of the model.

7. A tunnel lining stress dynamic prediction system based on GMS-MODFLOW and ABAQUS, characterized in that: It adopts a tunnel lining stress dynamic prediction method based on GMS-MODFLOW and ABAQUS as described in any one of claims 1 to 6.

Images