A real-time inversion method for fluid-solid coupling parameters of an arch cover method station in a water-rich area

Abstract

The present invention discloses a real-time inversion method for fluid-solid coupling parameters of arch-covered stations in water-rich areas, comprising: setting up an automated monitoring system for arch-covered stations in water-rich areas; establishing a three-dimensional numerical calculation model of the stratum and station excavation; establishing and training a convolutional neural network model to obtain parameters w and b, and a mapping relationship between the input and output of the convolutional neural network model; determining the optimal hyperparameter combination of the convolutional neural network model; obtaining the optimal convolutional neural network model and the mapping relationship between the input and output of the optimal convolutional neural network model; obtaining fluid-solid coupling parameters of the surrounding rock mass; and adjusting the support form of the arch-covered stations in water-rich areas. The present invention achieves the inversion of fluid-solid coupling parameters of arch-covered stations in water-rich areas by establishing and optimizing a convolutional neural network model. The method has the advantages of low workload, low computational time, low cost, high precision, and high reliability, greatly improving the efficiency of fluid-solid coupling parameter inversion of the surrounding rock mass, and having strong universality.

Description

Technical Field

[0001] The present invention relates to the technical field of fluid-solid coupling parameter identification for arch-covered stations in water-rich areas, and in particular to a real-time inversion method for fluid-solid coupling parameters for arch-covered stations in water-rich areas. Background Art

[0002] In the construction of large-section arch-covered stations in water-rich areas, the project span is large and the specific engineering geological conditions and surrounding environment are complex, so it is difficult to directly apply the successful experience of existing projects. In this case, the structures of new subway stations are mostly constructed using a step-by-step excavation and support method. The actual seepage and mechanical properties of the rock mass during construction are relatively complex, and the selection of support form and the determination of fluid-solid coupling parameters are particularly critical. Improper selection and application in construction numerical simulation will result in results that are inconsistent with reality, which will lead to large amounts of water gushing and instability in the existing structure.

[0003] Because conventional forward analysis or single-step prediction methods cannot reflect the seepage and mechanical property transformation process under the current construction sequence, a dynamic inversion prediction model was proposed. This method performs an inverse analysis of the original formation fluid-solid coupling parameters, and then uses the parameters obtained from the inverse analysis to perform forward calculations on the model to predict the seepage and deformation values of the existing structure. This method can more effectively reflect the disturbance of the existing structure by the current construction, dynamically establish the interaction relationship between the new tunnel, soil, and the existing structure, and thus more accurately predict the seepage and deformation values of the existing structure in the next construction step, providing a basis for the selection of the next support form.

[0004] However, in the process of numerical simulation construction, numerical simulation must be based on reasonable mechanical parameters. Testing rock-related mechanical parameters by indoor tests or in-situ tests is expensive and difficult to reflect the actual situation within the entire project. The determined mechanical parameters have large deviations from the actual situation. The determination of surrounding rock parameters is the guarantee for the accuracy of the sample library.

[0005] Today, some relatively mature deep learning algorithms have been widely applied in various fields and have also made significant progress in underground engineering construction. Deep learning algorithms analyze existing databases to identify corresponding relationships between data. Convolutional neural networks, as a highly efficient recognition method, can easily extract features from various data types. However, when working on specific tasks or projects, hyperparameter adjustment and optimization are required based on individual needs. This manual operation is not only time-consuming and labor-intensive, but also difficult to achieve optimal training results. How to select hyperparameter combinations for specific projects to ensure the accuracy and computational efficiency of convolutional neural network algorithms, and how to improve existing deep learning algorithms to enhance model performance, are both issues for further research. Summary of the Invention

[0006] The present invention provides a real-time inversion method for fluid-solid coupling parameters of arch-covered stations in water-rich areas, so as to overcome the bottleneck problems of the current traditional inversion method, such as the huge time consumption in calculation, the inability to consider the time effect of the fluid-solid coupling mechanical response of the surrounding rock mass, and the inability to predict and adjust the fluid-solid coupling parameters of the surrounding rock mass in real time.

[0007] In order to achieve the above object, the technical solution of the present invention is:

[0008] A real-time inversion method for fluid-solid coupling parameters of a station with an arch cover method in a water-rich area comprises the following steps:

[0009] S1: Set up an automated monitoring system for arch-covered stations in water-rich areas to obtain the water inflow during the excavation of the main rock mass below the station, the convergence displacement of the high side walls on both sides, the settlement displacement of the arch crown, the settlement displacement of the first arch waist, and the settlement displacement of the second arch waist;

[0010] S2: Based on the supporting structure and geological conditions of the arch-covered station in the water-rich area, a three-dimensional numerical calculation model of the stratum and station excavation is established to obtain the simulated water inflow, the convergence displacement of the high side walls on both sides, the simulated arch crown settlement displacement, the simulated first arch waist settlement displacement, the simulated second arch waist settlement displacement, and the simulated surrounding rock fluid-solid coupling parameters;

[0011] The surrounding rock includes surrounding rock layer I and surrounding rock layer II;

[0012] The simulated surrounding rock fluid-solid coupling parameters include: elastic modulus of simulated surrounding rock layer I, permeability coefficient of simulated surrounding rock layer I, elastic modulus of simulated surrounding rock layer II, permeability coefficient of simulated surrounding rock layer II;

[0013] S3: establishing a sample set with the simulated water inflow, the convergence displacement values of the simulated high side walls on both sides, the simulated arch crown settlement displacement values, the simulated first arch waist settlement displacement values, and the simulated second arch waist settlement displacement values as inputs, and the simulated surrounding rock fluid-solid coupling parameters as outputs, to obtain a training set and a test set;

[0014] S4: Based on the training set, establish a convolutional neural network model and perform training to obtain convolutional neural network parameters w and b, and a mapping relationship between the input and output of the convolutional neural network model;

[0015] S5: Determine the optimal hyperparameter combination of the convolutional neural network model based on the test set;

[0016] S6. Obtain an optimal convolutional neural network model and a mapping relationship between the input and output of the optimal convolutional neural network model according to the convolutional neural network parameters w, b and the optimal hyperparameter combination of the convolutional neural network;

[0017] S7: According to the mapping relationship between the input and output of the optimal convolutional neural network model, the water inflow, the convergence displacement value of the high side walls on both sides, the arch crown settlement displacement value, the first arch waist settlement displacement value and the second arch waist settlement displacement value are input into the optimal convolutional neural network model to obtain the surrounding rock fluid-solid coupling parameters;

[0018] The surrounding rock fluid-solid coupling parameters include: elastic modulus of surrounding rock layer I, permeability coefficient of surrounding rock layer I, elastic modulus of surrounding rock layer II, permeability coefficient of surrounding rock layer II;

[0019] S8: According to the surrounding rock fluid-solid coupling parameters, the support form of the arch-covering station in the water-rich area is adjusted.

[0020] Beneficial effects: The present invention provides a real-time inversion method for fluid-solid coupling parameters of arch-covered stations in water-rich areas. By establishing a convolutional neural network model and optimizing the convolutional neural network model, the inversion of fluid-solid coupling parameters of arch-covered stations in water-rich areas is completed through the optimized convolutional neural network model. By considering the time effect of the fluid-solid coupling mechanical response of the surrounding rock mass and real-time prediction and adjustment of the fluid-solid coupling parameters of the surrounding rock mass, this method consumes less computation time and has the advantages of small workload, low cost, high precision, and high reliability. It can greatly improve the efficiency of the inversion of the fluid-solid coupling parameters of the surrounding rock mass and has strong universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 This is a schematic diagram of the research section of the water-rich area arch-covered station of the present invention;

[0023] Figure 2 Schematic diagram of formation parameters in an embodiment of the present invention;

[0024] Figure 3 Schematic diagram of a numerical calculation model in an embodiment of the present invention;

[0025] Figure 4 A schematic diagram of arranging monitoring points in an embodiment of the present invention;

[0026] Figure 5 Output a convergence curve graph for the convolutional neural network model of the present invention;

[0027] Figure 6 A cross-sectional layout diagram of a station in an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of the overall arrangement of terminal surface sensors at a station in an embodiment of the present invention;

[0029] Figure 8 is a flow chart of the inversion method of the present invention;

[0030] Figure 9 This is a simplified diagram of the convolutional neural network model structure of the present invention;

[0031] Figure 10 Obtaining a process diagram for the convolutional neural network model of the present invention;

[0032] Figure 11 Schematic diagram of the parameter identification process of the present invention;

[0033] Figure 12 This is a graph showing changes in monitoring data of the laser rangefinder of the present invention.

[0034] Among them, 1. Automated monitoring and collection; 2. Laser rangefinder; 3. Static level; 4. Water level gauge; 5. Launch box; 6. Soil layer; 7. Surrounding rock layer I; 8. Surrounding rock layer II. DETAILED DESCRIPTION

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0036] This embodiment provides a real-time inversion method for fluid-solid coupling parameters of a station with an arch cover method in a water-rich area; the method comprises the following steps, as shown in the attached figure: Figure 8 As shown:

[0037] S1: Set up an automated monitoring system for arch-covered stations in water-rich areas to obtain the water inflow during the excavation of the main rock mass below the station, the convergence displacement of the high side walls on both sides, the settlement displacement of the arch crown, the settlement displacement of the first arch waist, and the settlement displacement of the second arch waist;

[0038] The automatic monitoring system includes: several automatic monitoring and collection boxes 1, a water level meter 4, a laser rangefinder 2, several static levels 3, a transmitting box 5, a remote server and a terminal;

[0039] As attached Figure 6 and attached Figure 7 As shown, the water level gauge is placed inside the sump; the sump is set on the inner side of the high side walls on both sides of the arch-covered station in the water-rich area;

[0040] The laser rangefinder is fixedly installed on the high side wall of the arch-covered station in the rich water area;

[0041] The static level is fixedly arranged on both sides of the arch waist and the arch top respectively;

[0042] The water level meter, laser rangefinder and static level are all electrically connected to the automated monitoring and collection box;

[0043] The automated monitoring and collection box is connected to the transmitting box;

[0044] The transmitting box is communicatively connected to the remote server;

[0045] The remote server is in communication connection with the terminal.

[0046] In this embodiment, the rock mass of the lower body of the station is excavated according to the sequence of the excavation steps of the lower body of the station (i.e., the sequential method). The rock mass excavation adopts the method of first trenching the middle part and then excavating the rock mass on both sides. After the initial lining construction of the lower high side wall is completed, the automatic monitoring system is immediately deployed. The automatic monitoring system is used to obtain the water inflow from the start of the excavation of the lower body of the station to the completion of a certain excavation step, the convergence displacement values of the high side walls on both sides, the arch crown settlement displacement values, the first arch waist settlement displacement values, and the second arch waist settlement displacement values.

[0047] A water collection pit was built on the inside of the high side wall of the arch-covered station in the water-rich area. Due to the large amount of water at the arch foot of the upper steps and the bottom of the pilot tunnel during station construction, an open drainage method was adopted. A 300mm wide drainage ditch was reserved on one side of the station pilot tunnel to collect water into the water collection pit. In addition, soft permeable pipe blind ditches were set up circumferentially and longitudinally behind the lining, and double-side ditches were set up in the tunnel for drainage. A ring of circumferential permeable pipe blind ditches and a centralized water outlet were set up every 8 to 10 meters in the longitudinal direction. Longitudinal blind ditches were set up at the corners of the walls on both sides. Groundwater was introduced into the inner side ditches of the tunnel every 8 to 12 meters. A water level gauge was built into the water collection pit to monitor the amount of water gushing during the excavation of the lower main body.

[0048] During the lower main excavation, a laser rangefinder was suspended in the middle of the high sidewall after the initial lining was completed to monitor the high sidewall's convergence displacement. Specifically, 10 laser rangefinders were installed within 100 meters of the lower main excavation, one every 10 meters. The laser rangefinders were used to measure the distance between the high sidewalls on both sides, i.e., the horizontal convergence value (BC) of the sidewalls.

[0049] To monitor the displacement of the arch crown, the first arch waist, and the second arch waist, first use a total station to determine the monitoring position, and then install static levels in the designated positions using expansion bolts. The same set of static levels are connected in series using liquid connecting pipes, and a liquid storage tank is arranged on one side. The height of the liquid storage tank is 10 to 30 cm higher than the level group. After installation, pure water is injected into the liquid storage tank (antifreeze should be injected when used in low temperature areas) to fill the entire set of equipment to monitor the displacement of the arch crown. Z and the first arch waist settlement displacement value E Z and the second arch waist settlement displacement value F Z .

[0050] The above sensors are deployed immediately after the initial lining construction of the lower high side wall is completed;

[0051] The monitoring data of the water level meter, laser rangefinder and static level are collected uniformly through a preset automatic monitoring collection box, and the collected data are transmitted to a remote server for reception. The data can be accessed and obtained through terminals such as mobile phones or PCs, and the changes in the monitoring data during the construction process can be monitored in real time. The problem of the long distance between multiple groups of water level meters, laser rangefinders and static levels can be solved by setting up multiple data collection boxes, as shown in the attached figure. Figure 7 As shown in the figure, data from multiple collection boxes is uploaded to the cloud server through a common transmission box, thereby achieving the goal of multi-terminal data release. The automated collection box aggregates measurement data information from different monitoring points to achieve regular recording of data from water level gauges, static levels, and laser rangefinders.

[0052] S2: Based on the supporting structure and geological conditions of the arch-covered station in the water-rich area, a three-dimensional numerical calculation model of the stratum and station excavation is established to obtain the simulated water inflow, the convergence displacement of the high side walls on both sides, the simulated arch crown settlement displacement, the simulated first arch waist settlement displacement, the simulated second arch waist settlement displacement, and the simulated surrounding rock fluid-solid coupling parameters;

[0053] The surrounding rock comprises surrounding rock layer I 7 and surrounding rock layer II 8; in this embodiment, surrounding rock layer I and surrounding rock layer II are respectively highly weathered quartzite and moderately weathered quartzite;

[0054] The simulated surrounding rock fluid-solid coupling parameters include: elastic modulus of simulated surrounding rock layer I, permeability coefficient of simulated surrounding rock layer I, elastic modulus of simulated surrounding rock layer II, permeability coefficient of simulated surrounding rock layer II;

[0055] In an embodiment of the present invention, the method for establishing the three-dimensional numerical calculation model of the stratum and the station excavation is as follows: according to the determined support structure design scheme of the arch-covered station in the water-rich area and the geological conditions of the arch-covered station in the water-rich area, the FLAC 3D The software creates a three-dimensional numerical model of the stratum and station excavation, and numerically simulates the construction process of the main excavation of the lower part of the station, such as Figure 3 ; Among them, the design plan of the support structure refers to the on-site working drawings; this modeling process is a common technology in the field, so it will not be described here.

[0056] S3: establishing a sample set with the simulated water inflow, the convergence displacement values of the simulated high side walls on both sides, the simulated arch crown settlement displacement values, the simulated first arch waist settlement displacement values, and the simulated second arch waist settlement displacement values as inputs, and the simulated surrounding rock fluid-solid coupling parameters as outputs, to obtain a training set and a test set;

[0057] Specifically, in an embodiment of the present invention, the sample set is established as follows: after determining the value range of the surrounding rock fluid-solid coupling parameter according to the geological conditions of the station, the surrounding rock fluid-solid coupling parameter is uniformly selected to form a corresponding orthogonal scheme, and simulation is performed according to the three-dimensional numerical model to obtain the simulated water inflow, the convergence displacement value of the simulated high side walls on both sides, the simulated arch crown settlement displacement value, the simulated first arch waist settlement displacement value, and the simulated second arch waist settlement displacement value generated by the three-dimensional numerical model simulation;

[0058] Specifically, based on the sample set, the method for obtaining the training set and the test set is as follows: m data are randomly extracted with replacement from the data in the sample set as the training set; and s data are randomly extracted with replacement as the test set, where m:s = 4:1. In this embodiment, 40 data are randomly extracted with replacement as the training set for training; correspondingly, 10 data are randomly extracted with replacement as the test set.

[0059] S4: Based on the training set, a convolutional neural network model is established and trained to obtain convolutional neural network parameters w, b and a mapping relationship between the input and output of the convolutional neural network model;

[0060] Specifically, in an embodiment of the present invention, the method for training a convolutional neural network model is: initializing the convolutional neural network model parameters, obtaining the mapping relationship between the input and output of the convolutional neural network model based on a training set, and obtaining a fitness function based on a test set.

[0061] Specifically, the convolutional neural network model includes an input layer, a convolution layer, a downsampling layer and a fully connected layer; Figure 5 Shown is a graph of the output convergence curve of the convolutional neural network model of the present invention;

[0062] The convolutional neural network model training process is as follows:

[0063] S41: Initialize the convolutional neural network training parameters w and b; where w is the weight matrix and b is the bias;

[0064] S42: Inputting the training set into the input layer of the convolutional neural network model, and normalizing the data of the training set;

[0065] The method for normalizing the training set data is as follows:

[0066]

[0067] Where: X normi is the data value of the normalized input column; X i is the real data value of the input i-th column; X mini is the minimum value of the input column; X maxi is the maximum value of the input column; Y normj is the data value of the normalized output column; Y j Output the real data value of column j; Y minj is the minimum value of the output column; Y maxj is the maximum value of the output column; X is the input data vector matrix; Y is the output data vector matrix;

[0068] S43: extracting sample features of the training set through a convolutional layer of a convolutional neural network;

[0069] Specifically, in this embodiment, the sample features of the training set are extracted by performing a convolution operation on the activation function Relu and the convolution kernel; the activation function Relu and the convolution operation on the activation function and the convolution kernel are both existing mature technologies, and this is only the use of the existing technology, so it will not be described in detail.

[0070] S44: reducing the dimension of the sample features through a downsampling layer;

[0071] Specifically, in this embodiment, the method of reducing the dimension of the sample features through the downsampling layer is: calculating the eigenvalues from the local area of the sample features of the convolution layer, selecting without overlap to reduce the dimension of the sample features, and keeping the feature scale unchanged. This is a common technology in the field, so it will not be described in detail.

[0072] S45: Connect all features of the previous layer of the fully connected layer through a fully connected layer, and use the fully connected layer as an output layer to output the convolution surrounding rock fluid-solid coupling parameters (i.e., the output result of the output layer);

[0073] S46: Establish a loss function, compare the error information between the convolution surrounding rock fluid-solid coupling parameter and the surrounding rock fluid-solid coupling parameter, and adjust the values of w and b according to the error information; wherein, the method of adjusting the values of w and b according to the error information is a commonly used method in the field, and therefore will not be described in detail here.

[0074] Specifically, in this embodiment, the data in the training set is trained using a convolutional neural network model: the goal of the training is to make the predicted value as close as possible to the target true value. Among them, S41 to S45 are forward propagation, the purpose of which is to obtain the predicted value, and the back propagation is to adjust the model training parameters w, b by comparing the error information between the output value of the convolutional neural network model (i.e., the convolution surrounding rock fluid-solid coupling parameter) and the target true value (i.e., the surrounding rock fluid-solid coupling parameter in the training set). The error information is the loss function;

[0075] The loss function is:

[0076]

[0077] Among them, m is the number of training sets, q is the number of training set data in the training set; y q is the true value of Y (i.e., the actual value of the surrounding rock fluid-solid coupling parameter); is the output value of the convolutional neural network (i.e., the predicted value of the surrounding rock fluid-solid coupling parameters), w is the weight matrix, and b is the bias;

[0078] S47: Obtain the mapping relationship between the input and output of the convolutional neural network model:

[0079] Y norm =CNN(X norm ) (3)

[0080]

[0081]

[0082] Where: Y norm is the output data after standardization; CNN represents the convolutional neural network model; v is the dimension (column) of the output data, X norm is the normalized input data, and u is the dimension (column) of the input data.

[0083] It should be noted here that the convolutional neural network model itself has model parameters such as weights and biases, which are optimized through the back propagation of this embodiment; in addition, there are many hyperparameters that need to be set manually. The performance of the machine learning algorithm depends to a large extent on the configuration of its initial hyperparameters. Different hyperparameter configurations are often required for different tasks. For convolutional neural networks, the hyperparameters that need to be optimized include the number of network training times, whether to use the dropout strategy, the number of convolution modules, pooling modules, fully connected modules, pooling methods, and batch size. Therefore, the appropriate selection of hyperparameter combinations is crucial to model performance. In this embodiment, the hyperparameter combination is the number of convolution layers, the number of fully connected layers, the convolution kernel size, the batch size, and the number of training epochs. See the schematic diagram of the convolutional neural network model described in the present invention. Figure 9 .

[0084] S5: Determine the optimal hyperparameter combination of the convolutional neural network model based on the test set;

[0085] Specifically, the method for determining the optimal hyperparameter combination of the convolutional neural network model in this embodiment is: using the firefly algorithm to determine the optimal hyperparameter combination of the convolutional neural network model;

[0086] The steps to determine the optimal hyperparameter combination for the convolutional neural network model are as follows. Figure 11 As shown:

[0087] S51: Initialize the parameters of the firefly algorithm. According to the number n of hyperparameters that need to be optimized in the convolutional neural network model, randomly generate a firefly population with N firefly individuals in the n-dimensional search space D to obtain the initial position of the firefly population. The dimension of the search space D is determined by the number n of hyperparameters that need to be optimized, and the range of the space D is determined by the upper and lower limits of each parameter in the selected hyperparameter combination. (The specific hyperparameter combination in this embodiment is the number of convolution layers (2 to 10), the number of fully connected layers (2 to 5), the convolution kernel size ([2×2] to [5×5]), the batch size (1 to 200), and the number of training epochs (1000 to 50000)).

[0088] Specifically, in this embodiment, the parameters of the firefly algorithm include the upper and lower limits of each hyperparameter of the optimal hyperparameter combination, the fluorescein volatility factor ρ = 0.4, the fluorescein update rate γ = 0.6, the dynamic decision domain update rate β = 0.08, and the n t The threshold n of the number of firefly individuals in the neighbor set of firefly individual I at time t is represented by t =5, fixed moving step step = 0.03, initial value of fluorescein l0 = 5, number of iterations T = 5000).

[0089] S52. Based on the fitness function, the initial positions of the firefly population are introduced into the convolutional neural network model for training and testing to obtain an updated fitness function value. In this algorithm, the mean square error (MSE) is selected as the objective function. The mapping relationship in the convolutional neural network model is predicted using the normalized test sample set. To minimize the objective function, the fitness function is constructed using an exponential transformation.

[0090] The fitness function is:

[0091] F(X I )=e -αMSE (4)

[0092] Where: F(X I ) represents the fitness function of firefly I; I is the individual number of the firefly population, X I is the hyperparameter combination represented by individual I, that is, the location information of firefly individual I, α is the gap adjustment coefficient. The smaller the value, the smaller the gap between the new fitness value of the individual with larger fitness before the transformation and the new fitness value of the individual with smaller fitness before the transformation. Conversely, the larger the value, the gap between the new fitness values of individuals with similar fitness before the transformation can be widened; MSE represents the objective function;

[0093]

[0094] Where: t is the evolutionary generation; N is the number of individuals in the group; f ave is the average fitness of the group; f min is the minimum fitness of the individual; f max is the individual maximum fitness; f is the individual fitness;

[0095]

[0096] Among them, y p represents the actual value of the pth surrounding rock parameter, is the predicted value of the p-th surrounding rock parameter; p is the number of the test set data in the test set;

[0097] S53: Update the fluorescein value of firefly I at time t;

[0098] In the embodiment of the present invention, the fluorescein value of the firefly I at time t is determined, and the fluorescein values of all individual fireflies in the firefly population are updated. The current fluorescein value of the individual firefly is l I (t) is affected by the fitness function F(X I (t)) and the fluorescence l of the previous iteration I the joint influence of (t-1);

[0099] then,

[0100] l I (t)=(1-ρ)l I (t-1)+γF(X I (t)) (7)

[0101] Where: F(X I (t)) represents the fitness function of firefly I at time t; ρ is the fluorescein volatilization factor, which represents the consumption rate of the light emitted by firefly individual I during the transmission process; γ is the fluorescein update rate, which represents the rate at which the brightness of firefly individual I increases in each iteration; l I (t) represents the fluorescein value of firefly I at time t; l I (t-1) represents the fluorescein value of firefly I at time t-1;

[0102] S54: Determine the set of neighbors of firefly individual I at time t to find the neighbors of firefly individual I;

[0103] Specifically, the method for finding the neighbors of firefly individual I in this embodiment is as follows:

[0104]

[0105] Where: N I (t) represents the set of neighbors of firefly individual I at time t; represents the radius of the dynamic decision domain of firefly individual I at time t; ||X J (t)-X I (t)|| represents the Euclidean distance between firefly individual I and firefly individual J; l J (t) represents the fluorescein value of firefly J at time t; r s is the perception radius of individual fireflies;

[0106] S55: Update the firefly population position, if N I (t) If the set is not empty, the firefly position is updated using the proportional selection method. If the set is empty, the firefly position is supplemented;

[0107] Use the proportional selection method to update the firefly position and calculate the new position state of firefly individual I as follows:

[0108]

[0109] Where: X I (t+1) is the position of the first firefly at the t+1th iteration; X I (t) is the position of firefly individual I at time t; X J(t) is the position of firefly individual J at time t; step is the fixed moving step length;

[0110] in:

[0111] J=max(p I ),

[0112]

[0113] Where: p IJ (t) is the transition probability of firefly individual I moving to firefly individual J;

[0114] If the domain is an empty set, the firefly supplementary position update is performed, and the formula is:

[0115]

[0116] θ is a random vector of corresponding dimension;

[0117] S56: If the updated position of the firefly population is not within the space D, then the updated position of the firefly population is corrected; otherwise, directly execute S57;

[0118] The formula for correcting the position of the updated firefly population is:

[0119]

[0120] Where: x Ik (t+1) represents the position state X of firefly individual I in the t+1th iteration I The kth component of (t+1), Represents the position component x k The upper limit of the value (that is, the upper limit of the value of the kth hyperparameter of the convolutional neural network), Represents the position component x k The lower limit of the value (that is, the lower limit of the value of the kth hyperparameter of the convolutional neural network):

[0121] S57: Update the dynamic decision domain radius of individual fireflies;

[0122] When the firefly individual completes its movement, adjust its decision domain radius r d I (t+1) is as follows:

[0123]

[0124] Where β represents the dynamic decision domain update rate, n t represents the threshold number of firefly individuals in the set of neighbors of firefly individual I at time t; represents the radius of the dynamic decision domain of firefly individual I at time t+1;

[0125] S58: If the number of iterations does not reach the set number of iterations, return to S52 and continue iterating;

[0126] If the number of iterations has reached the set number of iterations, the optimal firefly position is obtained, that is, the optimal hyperparameter combination of the convolutional neural network model;

[0127] S6. According to the convolutional neural network parameters w, b and the optimal hyperparameter combination of the convolutional neural network, the optimal convolutional neural network model, namely the FC model, and the mapping relationship between the input and output of the FC model are obtained. Figure 10 Shown is a diagram of the FC (convolutional neural network prediction model) acquisition process.

[0128] The optimal hyperparameter combination is introduced into CNN (convolutional neural network) to obtain the FC model, which is trained using the training set to obtain the mapping relationship between the input and output of the FC model.

[0129] S7: Obtaining fluid-solid coupling parameters of the surrounding rock based on the water inflow, the convergence displacement values of the high side walls on both sides, the arch crown settlement displacement value, the first arch waist settlement displacement value, and the second arch waist settlement displacement value;

[0130] The surrounding rock fluid-solid coupling parameters include: elastic modulus of surrounding rock layer I, permeability coefficient of surrounding rock layer I, elastic modulus of surrounding rock layer II, permeability coefficient of surrounding rock layer II;

[0131] Specifically, the water inflow, the convergence displacement values of the high side walls on both sides, the arch crown settlement displacement value, the first arch waist settlement displacement value and the second arch waist settlement displacement value are input into the FC model, so that the actual surrounding rock fluid-solid coupling parameters obtained by inversion can be obtained.

[0132] S8: Adjust the support form of the arch-covered station in the water-rich area according to the fluid-solid coupling parameters of the surrounding rock. The support form of the arch-covered station in the water-rich area in this embodiment is designed before construction, and the specific plan of the support form can be obtained according to the construction drawings.

[0133] Dynamic adjustment of support form: The inverted surrounding rock fluid-solid coupling parameters are input into the three-dimensional numerical model to obtain the water inflow generated by the next construction step in the simulation process, the convergence displacement value of the high side walls on both sides, the arch crown settlement displacement value, the first arch waist settlement displacement value and the second arch waist settlement displacement value, predict the stable state of the overall environment inside the station, and realize dynamic adjustment of the support form during the construction process.

[0134] Specifically, this embodiment deploys automated monitoring devices to obtain on-site water inflow, arch crown settlement, arch haunch settlement, and high sidewall horizontal convergence. A three-dimensional numerical model of the stratum and station excavation is created, and a sample set is generated. A training set and a test set are obtained to train a convolutional neural network model. The firefly algorithm is then used to determine the optimal hyperparameter combination for the convolutional neural network model based on a fitness function to obtain the optimal convolutional neural network model, namely the FC model. The FC model is then used to invert the fluid-solid coupling parameters of the surrounding rock mass, enabling dynamic adjustment of the support structure. This method addresses the bottleneck issues of traditional inversion methods, such as the time-consuming nature of computations, the inability to consider the time effects of the fluid-solid coupling mechanical response of the surrounding rock mass, and the inability to predict and adjust the fluid-solid coupling parameters of the surrounding rock mass in real time. It offers the advantages of low workload, low cost, high accuracy, and high reliability. It can greatly improve the efficiency of inversion of fluid-solid coupling parameters of the surrounding rock mass and is highly universal.

[0135] A specific embodiment of the present invention is as follows:

[0136] Choose FLAC 3D The software builds a three-dimensional calculation model to simulate the station construction process. Figure 1 As shown in Figure 6, the station soil layer 6 is mainly composed of four layers: fill layer, strongly weathered quartzite layer, medium weathered quartzite layer, and slightly weathered quartzite layer. Table 1 shows the soil and mechanical parameters of each layer. The stability of the cover-and-cut station foundation pit project is mainly affected by the medium weathered quartzite layer and the strongly weathered quartzite layer above it. Figure 2 As shown, the goal of this study is to invert the numerical model of these two formation parameters. Figure 3 The monitoring points are arranged and the distribution of the monitoring points among the five points of the side wall BC, arch A and arch waist E and F is shown as follows: Figure 4 According to the geological report and expert opinions, the rock mass and mechanical parameters of each layer are shown in Table 1, and the range of values of the fluid-solid coupling parameters of the surrounding rock mass to be inverted at the station is determined as shown in Table 2.

[0137] Table 1 Soil layers and mechanical parameters

[0138]

[0139] Table 2 Orthogonal calculation parameter value range

[0140]

[0141] According to the value range of the parameters in Table 2, 25 orthogonal design parameter combination schemes are formed, and the parameters corresponding to each experimental scheme are brought into FLAC 3DNumerical calculations were performed in the process of excavating the main body of the station, recording the water flow, side wall displacement convergence, and settlement of the arch and arch waist. The water flow q, the convergence value BC of the side wall in the horizontal direction, and the arch settlement D were obtained. Z and the settlement E on both sides of the arch waist Z 、F Z The orthogonal parameter schemes and the calculation results of the schemes are shown in Table 3.

[0142] The data in Table 3 were randomly divided into training samples and test samples with replacement at a ratio of 4:1. The convolutional neural network model for fluid-structure interaction parameter inversion was trained according to the method proposed in this paper, and the convolutional neural network model optimized by the firefly algorithm was obtained.

[0143] Table 3 Orthogonal scheme and calculation results

[0144]

[0145] After construction, based on on-site monitoring results, a back-analysis of the fluid-solid coupling parameters of the surrounding rock mass was performed. The obtained on-site stress response data is shown in Table 5. The back-analysis results are shown in Table 6.

[0146] Table 5 Field monitoring results

[0147]

[0148] Table 6 Back analysis results of surrounding rock parameters

[0149]

[0150] Using the fluid-solid coupling parameters of the surrounding rock mass to be inverted obtained by the inversion method of the present invention, a numerical simulation analysis of the station excavation was performed, resulting in stress response data changes at the monitoring points, as shown in Table 7. The numerical simulation results were compared with the field monitoring data. As can be seen from the results, the maximum relative error between the two methods is 3.18%, verifying the accuracy of the results obtained by the inversion method of the present invention.

[0151] Table 7 Back-analysis calculation results and field monitoring results

[0152]

[0153] Considering the high risk of the double-layer overlapping primary support arch cover method for large-span underground excavation stations and the lack of experience to draw on, an independently developed automated data collection system monitors multiple information during the construction process and dynamically adjusts the construction plan based on the monitored information. High side wall information construction process control technology:

[0154] During the excavation of the lower rock mass of Shikui Road Station, it was determined that the station would be constructed using the sequential construction method, with the lower main rock mass excavated by grooving in the middle followed by construction on both sides, omitting steel supports. To ensure the smooth implementation of the above plan, the dynamic changes of the high side walls during construction were monitored in real time to ensure the smooth implementation of the optimization plan and achieve dynamic adjustments to the construction process.

[0155] During construction, the horizontal convergence value of the high sidewall reflects its stability and the safety of the construction environment. During the excavation of the second internal rock layer, numerical simulations using the inverse analysis parameters from the previous step revealed a sudden displacement change at monitoring point CJ-04, but the change remained within control. Based on the actual construction conditions on site, it was proposed to add anchor cable support within a 30-meter radius along the station axis within the area where monitoring point CJ-04 is located to ensure the safety of subsequent construction.

[0156] After the anchor cable support was installed, monitoring data stabilized at approximately 2.3 mm, with no further sudden displacement changes. This data indicates that the high sidewall remained stable during the later stages of construction. The high sidewall information construction process control technology enabled dynamic adjustments during the construction process, ensuring safe construction.

[0157] From the internal environmental conditions of the station during the on-site high side wall construction phase, it can be seen that the overall internal environment of the station is in a stable state, and the dynamic adjustment of the construction plan based on monitoring data has achieved good results.

[0158] Laser rangefinders are used to monitor the convergence displacement of the side walls during the excavation of the lower part of the station. Ten laser rangefinders are installed within 100m of the lower main excavation, with one installed every 10m. The data measured by the laser rangefinders is used to monitor and issue early warnings for the lower main excavation of the station.

[0159] In one embodiment of the present invention, when using the inversion method of the present invention, a laser rangefinder, a static level, a water pressure gauge, and a water level gauge are used to obtain the actual water inflow, the convergence displacement values of the high side walls on both sides, the arch crown settlement displacement value, the first arch waist settlement displacement value, and the second arch waist settlement displacement value; and a collection box is used to realize the collection of timed data. When inverting the fluid-solid coupling parameters of the arch cover method station in the water-rich area, the device used in this embodiment is arranged as follows:

[0160] (a) Monitoring of high side wall displacement during the lower excavation of the station: During the excavation of the lower main body, a laser rangefinder is hung at the middle position of the high side wall after the initial lining is completed to monitor the convergence displacement value of the high side wall, as shown in the attached figure. Figure 12 As shown. The laser rangefinder is used to measure the distance between the two measuring points B and C of the target, that is, the convergence value BC in the horizontal direction of the side wall;

[0161] (b) Monitoring the settlement and displacement of the high side wall during the excavation of the lower main body: During the excavation of the lower main body, a static level is hung in the middle of the high side wall to monitor the vertical settlement and displacement of the high side wall, that is, the side wall settlement D Z .

[0162] (c) Select appropriate locations and arrange automated collection boxes to collect measurement data from different monitoring points and record the data of the static level and laser rangefinder at regular intervals.

[0163] (d) Water flow monitoring during lower main body excavation: During the excavation of the lower main body, a water level gauge is placed in the drainage trough at the bottom of the high side wall to monitor the water flow value during the excavation process and to monitor the water level changes during the excavation process and obtain the water flow value.

[0164] The problem of long distances between multiple sensor groups is solved by setting up multiple data collection boxes. The data from multiple collection boxes can be uploaded to the cloud server through a common transmission box, thus achieving the goal of multi-terminal data distribution.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time inversion method for fluid-solid coupling parameters of a station with an arch cover method in a water-rich area, characterized in that: The steps include: S1: Set up an automated monitoring system for arch-covered stations in water-rich areas to obtain the water inflow during the excavation of the main rock mass below the station, the convergence displacement of the high side walls on both sides, the settlement displacement of the arch crown, the settlement displacement of the first arch waist, and the settlement displacement of the second arch waist; S2: Based on the supporting structure and geological conditions of the arch-covered station in the water-rich area, a three-dimensional numerical calculation model of the stratum and station excavation is established to obtain the simulated water inflow, the convergence displacement of the high side walls on both sides, the simulated arch crown settlement displacement, the simulated first arch waist settlement displacement, the simulated second arch waist settlement displacement, and the simulated surrounding rock fluid-solid coupling parameters; The surrounding rock includes surrounding rock layer I and surrounding rock layer II; The simulated surrounding rock fluid-solid coupling parameters include: elastic modulus of simulated surrounding rock layer I, permeability coefficient of simulated surrounding rock layer I, elastic modulus of simulated surrounding rock layer II, permeability coefficient of simulated surrounding rock layer II; S3: establishing a sample set with the simulated water inflow, the convergence displacement values of the simulated high side walls on both sides, the simulated arch crown settlement displacement values, the simulated first arch waist settlement displacement values, and the simulated second arch waist settlement displacement values as inputs, and the simulated surrounding rock fluid-solid coupling parameters as outputs, to obtain a training set and a test set; S4: Based on the training set, establish a convolutional neural network model and perform training to obtain convolutional neural network parameters w and b, and a mapping relationship between the input and output of the convolutional neural network model; In S4, the convolutional neural network model is established as follows: S41: Initialize the convolutional neural network training parameters w and b; where w is the weight matrix and b is the bias; S42: Inputting the training set into the input layer of the convolutional neural network model, and normalizing the data of the training set; The method for normalizing the training set data is as follows: Where: X normi is the data value of the normalized input column; X i is the real data value of the input i-th column; X mini is the minimum value of the input column; X maxi is the maximum value of the input column; Y normj is the data value of the normalized output column; Y j Output the real data value of column j; Y minj is the minimum value of the output column; Y maxj is the maximum value of the output column; X is the input data vector matrix; Y is the output data vector matrix; S43: extracting sample features of the training set through a convolutional layer of a convolutional neural network; S44: reducing the dimension of the sample features through a downsampling layer; S45: Connect all features of the previous layer through a fully connected layer, and use the fully connected layer as the output layer to output the convolution surrounding rock fluid-solid coupling parameters; S46: establishing a loss function, comparing error information between the convolution surrounding rock fluid-solid coupling parameter and the surrounding rock fluid-solid coupling parameter, and adjusting the values of w and b according to the error information; S47: Obtain the mapping relationship between the input and output of the convolutional neural network model: Y norm =CNN(X norm ) (3) Where: Y norm is the output data after standardization; CNN represents the convolutional neural network model; v is the dimension of the output data, X norm is the standardized input data, u is the dimension of the input data; S5: Determine the optimal hyperparameter combination of the convolutional neural network model based on the test set; S6. Obtain an optimal convolutional neural network model and a mapping relationship between the input and output of the optimal convolutional neural network model according to the convolutional neural network parameters w, b and the optimal hyperparameter combination of the convolutional neural network; S7: According to the mapping relationship between the input and output of the optimal convolutional neural network model, the water inflow, the convergence displacement value of the high side walls on both sides, the arch crown settlement displacement value, the first arch waist settlement displacement value and the second arch waist settlement displacement value are input into the optimal convolutional neural network model to obtain the surrounding rock fluid-solid coupling parameters; The surrounding rock fluid-solid coupling parameters include: elastic modulus of surrounding rock layer I, permeability coefficient of surrounding rock layer I, elastic modulus of surrounding rock layer II, permeability coefficient of surrounding rock layer II S8: According to the surrounding rock fluid-solid coupling parameters, the support form of the arch-covering station in the water-rich area is adjusted.

2. The real-time inversion method for fluid-solid coupling parameters of a water-rich area arch-covering station according to claim 1 is characterized in that: The automated monitoring system includes: several automated monitoring and collection boxes, a water level meter, a laser rangefinder, several static levels, a transmitter box, a remote server and a terminal; The water level gauge is placed inside the sump; the sump is set inside the high side walls on both sides of the arch-covered station in the water-rich area; The laser rangefinder is fixedly installed on the high side wall of the arch-covered station in the rich water area; The static level is fixedly arranged on both sides of the arch waist and the arch top respectively; The water level meter, laser rangefinder and static level are all electrically connected to the automated monitoring and collection box; The automated monitoring and collection box is connected to the transmitting box; The transmitting box is communicatively connected to the remote server; The remote server is in communication connection with the terminal.

3. The real-time inversion method for fluid-solid coupling parameters of a water-rich area arch-covering station according to claim 2 is characterized in that: In S46, the loss function is: Among them, m is the number of training sets, q is the number of training set data in the training set; y q is the true value of Y; is the output value of the convolutional neural network, w is the weight matrix, and b is the bias.

4. The real-time inversion method for fluid-solid coupling parameters of a water-rich area arch-covering station according to claim 3 is characterized in that: The steps for determining the optimal hyperparameter combination for the convolutional neural network model are as follows: S51: Initialize the parameters of the firefly algorithm. According to the number of hyperparameters n that need to be optimized in the convolutional neural network model, randomly generate a firefly population with N firefly individuals in the n-dimensional search space D to obtain the initial position of the firefly population. S52: According to the fitness function, the initial position of the firefly population is brought into the convolutional neural network model for training and testing to obtain an updated fitness function value; The fitness function is: F(X I )=e -αMSE (4) Where: F(X I ) represents the fitness function of firefly I; I is the individual number of the firefly population, X I is the hyperparameter combination represented by individual I, i.e., the position information of firefly individual I, α is the gap adjustment coefficient; MSE represents the objective function; Where: t is the evolutionary generation; N is the number of individuals in the population; f ave is the average fitness of the group; f min is the minimum fitness of the individual; f max is the individual maximum fitness; f is the individual fitness; Among them, y p represents the actual value of the pth surrounding rock parameter, is the predicted value of the p-th surrounding rock parameter; p is the number of the test set data in the test set; S53: Update the fluorescein value of firefly I at time t; then, l I (t)=(1-ρ)l I (t-1)+γF(X I (t)) (7) Where: F(X I (t)) represents the fitness function of firefly I at time t; ρ is the fluorescein volatilization factor, which represents the consumption rate of the light emitted by firefly individual I during the transmission process; γ is the fluorescein update rate, which represents the rate at which the brightness of firefly individual I increases in each iteration; l I (t) represents the fluorescein value of firefly I at time t; l I (t-1) represents the fluorescein value of firefly I at time t-1; S54: Determine the set of neighbors of firefly individual I at time t to find the neighbors of firefly individual I; Specifically, the method for finding the neighbors of firefly individual I in this embodiment is as follows: Where: N I (t) represents the set of neighbors of firefly individual I at time t; represents the radius of the dynamic decision domain of firefly individual I at time t; ||X J (t)-X I (t)|| represents the Euclidean distance between firefly individual I and firefly individual J; l J (t) represents the fluorescein value of firefly J at time t; r s is the perception radius of individual fireflies; S55: Update the firefly population position, if N I (t) If the set is not empty, the firefly position is updated using the proportional selection method. If the set is empty, the firefly position is supplemented; Use the proportional selection method to update the firefly position and calculate the new position state of firefly individual I as follows: Where: X I (t+1) is the position of the first firefly at the t+1th iteration; X I (t) is the position of firefly individual I at time t; X J (t) is the position of firefly individual J at time t; step is the fixed moving step length; in: Where: is the probability that firefly individual I moves to the NI(t)th individual in the set of its neighbors at time t; Where: p IJ (t) is the transition probability of firefly individual I moving to firefly individual J; If the domain is an empty set, the firefly supplementary position update is performed, and the formula is: θ is a random vector of corresponding dimension; S56: If the updated position of the firefly population is not within the space D, then the updated position of the firefly population is corrected; otherwise, directly execute S57; The formula for correcting the position of the updated firefly population is: Where: x Ik (t+1) represents the position state X of firefly individual I in the t+1th iteration I The kth component of (t+1), Represents the position component x k The upper limit of the value, Represents the position component x k The value of limit S57: Update the dynamic decision domain radius of individual fireflies; Where β represents the dynamic decision domain update rate, n t represents the threshold number of firefly individuals in the set of neighbors of firefly individual I at time t; represents the radius of the dynamic decision domain of firefly individual I at time t+1; S58: If the number of iterations does not reach the set number of iterations, return to S52 and continue iterating; If the number of iterations has reached the set number of iterations, the optimal hyperparameter combination is obtained.

Images