A method for evaluating tunnel surrounding rock stability based on numerical simulation and deep learning

Abstract

A tunnel surrounding rock stability assessment method based on numerical simulation and deep learning, obtains the actual parameters of the formation rock under the actual geological environment; establishes a two-dimensional geological model according to the drilling column chart near the underground coal mine tunnel and the formation rock parameters determined by the experiment; changes the influencing factors, and records the deformation of the tunnel side, acceleration value and whether obvious damage occurs after changing the influencing factors, then uses the extracted influencing factors and the dynamic response characteristics of the tunnel surrounding rock as labels, and then combines the two to obtain a data set, and finally forms a database with the obtained multiple data sets; divides the data set into a training set and a validation set according to a set ratio, and inputs the training set into a PSO-BP neural network and a GA-SVM deep learning model for deep learning, obtains a preliminary tunnel surrounding rock stability assessment model based on numerical simulation and deep learning, and then uses the validation set to adjust and verify the preliminary tunnel surrounding rock stability assessment model, and obtains an optimized tunnel surrounding rock stability assessment model; and performs stability assessment. This method can quickly and accurately evaluate the stability of the tunnel surrounding rock.

Description

Technical Field

[0001] The present invention belongs to the technical field of safety prediction and assessment of surrounding rock in mining tunnels, and specifically relates to a tunnel surrounding rock stability assessment method based on numerical simulation and deep learning. Background Art

[0002] With the increase in the depth of coal mining and the deterioration of resource conditions in my country, safety accidents caused by the instability of tunnel surrounding rocks are becoming increasingly serious. Tunnel surrounding rock stability assessment is a key link in mine safety production. Its purpose is to timely discover potential tunnel surrounding rock instability risks and take corresponding preventive measures in time, so as to avoid or reduce the occurrence of safety accidents. However, there are still some problems in the assessment of tunnel surrounding rock stability. The current assessment methods are mostly based on empirical formulas and theoretical calculations, but these methods often fail to fully consider the complexity and variability of tunnel surrounding rocks. For example, factors such as the lithology, structure, stress state, and groundwater conditions of tunnel surrounding rocks may affect their stability, and these factors are often difficult to accurately quantify and consider in actual assessments. At the same time, since theoretical models are often oversimplified, the existing theoretical research results are often difficult to directly apply to actual projects, and thus cannot fully reflect the complex situations in actual projects. At the same time, the relevant theoretical research on the stability assessment of tunnel surrounding rocks, anti-impact support and pressure relief measures in rock burst mines still lags behind on-site engineering practice.

[0003] Therefore, there is an urgent need to provide a method that can accurately and quickly evaluate the stability of the tunnel to effectively ensure the safe production of the mine. Summary of the invention

[0004] In response to the problems existing in the above-mentioned prior art, the present invention provides a tunnel surrounding rock stability assessment method based on numerical simulation and deep learning. The method has good flexibility and a wide range of adaptability. It can realize a rapid and accurate evaluation of the tunnel surrounding rock stability under the same geological conditions, different dynamic loads, support unloading, etc., and can be adjusted and optimized according to actual conditions.

[0005] In order to achieve the above object, the present invention provides a tunnel surrounding rock stability assessment method based on numerical simulation and deep learning, which specifically includes the following steps:

[0006] Step 1: Determine the underground tunnel of the coal mine to be evaluated, collect rocks on the coal seam roof as samples, and use test equipment in the laboratory to test the physical and mechanical properties of the samples to obtain the actual parameters of the formation rock under the actual geological environment;

[0007] Step 2: Based on the borehole histogram near the underground tunnel of the coal mine and the experimentally determined formation rock parameters, a two-dimensional geological model is established through numerical simulation;

[0008] Step 3: In the two-dimensional geological model, the influencing factors are changed, including dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress degree, support parameters and pressure relief measures, and the deformation amount, acceleration value and whether obvious damage occurs in the tunnel side after the influencing factors are changed are recorded. Then, the extracted influencing factors and the dynamic response characteristics of the tunnel surrounding rock are used as labels, and the two are combined to obtain a data set. Finally, the obtained multiple data sets are formed into a database;

[0009] Step 4: Divide the data set into a training set and a validation set according to a set ratio, and input the training set into the PSO-BP neural network and GA-SVM deep learning model for deep learning to obtain a preliminary roadway surrounding rock stability evaluation model based on numerical simulation and deep learning. Then, use the validation set to adjust and verify the preliminary roadway surrounding rock stability evaluation model to obtain an optimized roadway surrounding rock stability evaluation model.

[0010] Step 5: Use the optimized tunnel surrounding rock stability evaluation model to conduct stability assessment on other mining stages of the tunnel to determine whether tunnel surrounding rock instability and failure will occur under the influence of dynamic loads under certain geological conditions, support methods and pressure relief measures.

[0011] Furthermore, in order to fully reflect the mechanical conditions of the coal rock mass, in step one, the actual parameters include density, bulk modulus, shear modulus, friction angle and cohesion.

[0012] As a preferred embodiment, in step 2, a three-dimensional geological model is established using UDEC numerical simulation software.

[0013] Furthermore, in order to establish a two-dimensional geological model that can accurately reflect the deformation and response characteristics of the tunnel surrounding rock, in step 2, the process of establishing the two-dimensional geological model is as follows:

[0014] S21: Set the boundary conditions of the model: set the horizontal displacement in the x and y directions to zero, the lower boundary of the model is fixed, that is, the displacement of the lower boundary is zero, the upper boundary is the free boundary with equivalent load, and the yield criterion is selected Yield criterion, and ignoring plastic flow and dilatancy, the model is established with reference to the comprehensive column chart, and the model is parameterized based on the formation rock parameters in step 1;

[0015] S22: applying loads on the top and both sides of the two-dimensional geological model according to the actual burial depth and the ground stress detection results, wherein the deadweight load on the top of the two-dimensional geological model is calculated by formula (1), and the loads applied on both sides are obtained by referring to the ground stress test results, and are calculated under given mechanical conditions and displacement boundary conditions, so that the model reaches an initial stress equilibrium state from an initial state;

[0016] P = γH(1);

[0017] Where, γ is the average bulk density of the rock layer, which is 25 kN / m3; H is the height of the upper boundary of the model from the ground surface, in meters;

[0018] S23: The numerical model is divided into five layers using the crack command in UDEC, of ​​which four are roof layers, namely siltstone, argillaceous siltstone, siltstone and mudstone, and one is a coal seam in which the roadway is arranged.

[0019] Furthermore, in order to obtain multiple sets of data sets quickly and conveniently, in step three, the process of obtaining multiple sets of data sets is as follows:

[0020] A tunnel is excavated in the built two-dimensional geological model, and survey lines are arranged on the sides and bottom of the tunnel. The survey lines are used to monitor the dynamic response characteristics of the tunnel surrounding rock after changing the influencing factors, that is, the deformation amount and acceleration value of the tunnel side are monitored, and whether the tunnel is damaged under this condition is recorded. The recorded influencing factors and the dynamic response characteristics of the tunnel surrounding rock are then used as labels and the two are combined into a data set; multiple groups of numerical simulation models are established according to different conditions, and multiple groups of data sets are obtained according to the above method.

[0021] Furthermore, in order to obtain a model that can quickly and accurately evaluate the stability of the tunnel surrounding rock, in step 4, the establishment process of the optimized tunnel surrounding rock stability evaluation model is as follows:

[0022] S41: First, use the Randperm function in MATLAB to randomly shuffle the order of multiple data sets in the database, and then screen them into training set and validation set in a ratio of 7:3;

[0023] S42: Establish PSO-BP neural network;

[0024] S421, determining the topological structure and number of nodes of the BP neural network;

[0025] S422, setting the initialization parameters of the particle swarm algorithm: learning factor, number of trainings, target error, learning rate, number of population updates, population size, position extreme value and velocity extreme value of particles, and optimizing the initialization weight and threshold of the BP neural network;

[0026] S423, determining that the evaluation function of the BP neural network performance is RMSE;

[0027] S424, determining the individual optimal solution and the overall optimal solution in the PSO algorithm;

[0028] S425, updating each particle according to the particle update formula (2);

[0029]

[0030] Where i = 1, 2, ..., N, N is the total number of particles in this group; v i is the velocity vector of particle i in d dimension; rand() is a random number between (0,1); x i is the position vector of particle i in d dimension; c1 and c2 are learning factors, c1=c2=2;

[0031] S426, when the number of iterations in the particle swarm optimization algorithm is equal to the preset number of training times or meets the target error, the iterative optimization process ends, otherwise returns to S423 for re-optimization;

[0032] S427, the optimal output value at the end of the iterative optimization, that is, the optimized weight and threshold, is substituted into the BP neural network for data training to achieve effective, accurate and fast data fitting;

[0033] S43: Establish GA-SVM deep learning model;

[0034] S431, setting initial parameters, kernel function K(x,z) and appropriate penalty function C in the support vector machine;

[0035] S432, setting initialization parameters of the genetic algorithm to optimize the initialization parameters in the support vector machine;

[0036] S433, determining that the evaluation function of the support vector machine is the mean square error;

[0037] S434, when the number of iterations in the genetic algorithm is equal to the preset number of training times or meets the target error, the iterative optimization ends, otherwise returns to S432 for re-optimization;

[0038] S435, substituting the optimal output value at the end of the iterative optimization, that is, the optimized c and g parameters, into the support vector machine for data training to achieve effective, accurate and fast data fitting;

[0039] S44: Construct a preliminary model for evaluating the stability of surrounding rock in tunnels;

[0040] S441: In the BP neural network model optimized based on particle swarm algorithm, vertical stress, horizontal stress, dynamic load intensity, dynamic load distance, side anchor length, bottom anchor length and bottom anchor angle in multiple data sets are used as evaluation indicators, and the maximum displacement and maximum speed of the side are used as target values. After importing the training set and the validation set, the PSO-BP algorithm is used to improve the evaluation model;

[0041] S442: In the support vector machine model optimized based on genetic algorithm, the maximum displacement and maximum speed of the gangway in multiple data sets are used as evaluation indicators, and whether the tunnel is damaged is used as the target value. After importing the training set and the verification set, the GA-SVM algorithm is used to improve the evaluation model.

[0042] As a preferred embodiment, in step 5, the process of using the optimized tunnel surrounding rock stability evaluation model to evaluate the stability of other mining stages of the tunnel is as follows:

[0043] The specific parameters of dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress degree, support parameters and pressure relief measures are input into the optimized tunnel surrounding rock stability evaluation model. The optimized tunnel surrounding rock stability evaluation model determines whether the tunnel surrounding rock will become unstable and damaged under this condition through data prediction.

[0044] The present invention is based on numerical simulation, deep learning and engineering measured data. First, the influence of factors such as dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress level, support parameters and pressure relief measures on the stability of tunnel surrounding rock is analyzed by numerical simulation, and the dynamic load response characteristics of tunnel surrounding rock are recorded. It can be conveniently and comprehensively determined by data prediction whether the tunnel surrounding rock will be unstable and damaged under the working conditions where each influencing factor changes. Then, the deep learning part determines the nonlinear correspondence between the influencing factors and the dynamic load response characteristics of the tunnel surrounding rock based on the numerical simulation results, and then establishes an evaluation model, which can facilitate the use of the evaluation model to quickly evaluate the stability of the tunnel surrounding rock. Finally, the evaluation model can be corrected and improved by field data, which can significantly improve the accuracy of the evaluation model.

[0045] The present invention constructs an accurate, objective, reasonable and reliable model for evaluating the stability of surrounding rock of a tunnel, which overcomes the problems of large engineering workload, time-consuming and labor-intensive, high labor costs, and inability to conduct large-scale or all-round field measurements in traditional methods for evaluating the stability of surrounding rock of a tunnel. The method has the advantages of accurate, objective, reasonable and reliable prediction of the stability grade of surrounding rock of a mining tunnel and the model maturity evaluation results, and a high degree of consistency with the actual situation. It is helpful for scientifically designing tunnel anti-impact support and support methods, and can effectively improve the impact resistance of surrounding rock of a tunnel, thereby significantly reducing the degree and scope of impact damage. The present invention is of great significance for protecting the safety of construction personnel in the mining space, improving the safety of operations, and ensuring the safe production of mines. The method has good flexibility and a wide range of adaptability. It can realize rapid and accurate multi-group evaluation of the stability of surrounding rock of a tunnel under the same geological conditions, different dynamic loads, support unloading, etc., and can be adjusted and optimized according to the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of the present invention;

[0047] Figure 2 It is a flow chart of the BP neural network model optimized based on the particle swarm algorithm in the present invention;

[0048] Figure 3 A flowchart of optimizing a nonlinear support vector machine based on a genetic algorithm in the present invention;

[0049] Figure 4 It is a schematic diagram of the block division of the UDEC model in the present invention;

[0050] Figure 5 is a schematic diagram of a calculation model in the present invention;

[0051] Figure 6 This is a comparison curve of the predicted value and the simulated value of the maximum displacement of the central helper verified by PSO-BP in the present invention;

[0052] Figure 7 This is a comparison curve of the predicted value and the simulated value of the maximum speed of the central helper verified by PSO-BP in the present invention;

[0053] Figure 8 It is the numerical fitting regression curve of the maximum displacement of the central part verified by PSO-BP in the present invention;

[0054] Fig. 9 It is the numerical fitting regression curve of the maximum speed of the central helper verified by PSO-BP in the present invention;

[0055] Fig.10 This is a line graph comparing the predicted values ​​and simulated values ​​in the GA-SVM validation set used in the present invention. DETAILED DESCRIPTION

[0056] like Figures 1 to 10 As shown, the present invention provides a method for evaluating the stability of tunnel surrounding rock based on numerical simulation and deep learning. In order to explain the present invention in detail, the tunnel data of a coal mine working face is used as an example, which specifically includes the following steps:

[0057] Step 1: Determine the underground tunnel of the coal mine to be evaluated, collect the rock on the coal seam roof as a sample, and use the test equipment in the laboratory to test the physical and mechanical properties of the sample to obtain the actual parameters of the formation rock under the actual geological environment. In order to fully reflect the mechanical conditions of the coal rock mass, the actual parameters include density, bulk modulus, shear modulus, friction angle and cohesion. The specific mechanical parameters of the coal rock mass are shown in Table 1;

[0058] Table 1: Mechanical parameters of coal rock mass

[0059]

[0060] Step 2: Based on the borehole histogram near the underground tunnel of the coal mine and the experimentally determined formation rock parameters, a two-dimensional geological model is established through numerical simulation;

[0061] In order to establish a two-dimensional geological model that can accurately reflect the deformation and response characteristics of the tunnel surrounding rock, the UDEC numerical simulation software is used to establish a three-dimensional geological model. The process of establishing a two-dimensional geological model is as follows:

[0062] S21: Set the boundary conditions of the model: set the horizontal displacement in the x and y directions to zero, the lower boundary of the model is fixed, that is, the displacement of the lower boundary is zero, the upper boundary is the free boundary with equivalent load, and the yield criterion is selected Yield criterion, and ignoring plastic flow and dilatancy, the model is established with reference to the comprehensive column chart, and the model is parameterized based on the formation rock parameters in step 1;

[0063] S22: applying loads on the top and both sides of the two-dimensional geological model according to the actual burial depth and the ground stress detection results, wherein the deadweight load on the top of the two-dimensional geological model is calculated by formula (1), and the loads applied on both sides are obtained by referring to the ground stress test results, and are calculated under given mechanical conditions and displacement boundary conditions, so that the model reaches an initial stress equilibrium state from an initial state;

[0064] P = γH(1);

[0065] Where, γ is the average bulk density of the rock layer, which is 25 kN / m3; H is the height of the upper boundary of the model from the ground surface, in meters;

[0066] The buried depth of the coal mine selected in this embodiment is 536m, so H is taken as 536m, and the self-weight load on the top of the two-dimensional geological module is calculated to be P=γH=25×536=13.4Mpa, which is in the z-axis direction, and the loads applied on both sides are σ1=16.1MPa, which is in the x-axis direction, and σ2=15.5MPa, which is in the y-axis direction;

[0067] S23: See Figure 4 The overall height of the numerical model is 60m and the width is 40m. The simulated excavated tunnel is 3.6m high and 5.4m wide. The numerical model is divided into 5 layers using the crack command in UDEC, of ​​which 4 layers are roofs, namely siltstone, muddy siltstone, siltstone and mudstone, and 1 layer is a coal seam, in which the tunnel is arranged. In order to explore the deformation and response characteristics of the tunnel surrounding rock under dynamic and static combined loading, the 16m high and 20m wide area near the tunnel in the numerical simulation model is divided into 0.6m triangular blocks, and 1m irregular blocks are used in other areas of the numerical simulation model;

[0068] Step 3: In the two-dimensional geological model, change the influencing factors, including dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress degree, support parameters and pressure relief measures, and record the deformation of the tunnel side, acceleration value and whether obvious damage occurs after changing the influencing factors. Then, use the extracted influencing factors and the dynamic response characteristics of the tunnel surrounding rock as labels, and then combine the two to obtain a data set. Finally, the obtained multiple data sets form a database;

[0069] In order to obtain multiple sets of data sets quickly and conveniently, the process of obtaining multiple sets of data sets is as follows: excavate a tunnel in the built two-dimensional geological model, and arrange survey lines on the side and bottom of the tunnel. Use the survey lines to monitor the dynamic response characteristics of the tunnel surrounding rock after changing the influencing factors, that is, monitor the deformation and acceleration values ​​of the tunnel side, and record whether the tunnel is damaged under this condition. Then, the recorded influencing factors and the dynamic response characteristics of the tunnel surrounding rock are used as labels, and the two are combined into a data set; establish multiple sets of numerical simulation models according to different conditions, and obtain multiple sets of data sets according to the above method.

[0070] In this embodiment, the obtained data set is 66 groups, as shown in Table 2;

[0071] Table 2: Evaluation index and target value statistics

[0072]

[0073]

[0074]

[0075]

[0076] Step 4: Divide the data set into a training set and a validation set according to a set ratio, and input the training set into the PSO-BP neural network and GA-SVM deep learning model for deep learning to obtain a preliminary roadway surrounding rock stability evaluation model based on numerical simulation and deep learning. Then, use the validation set to adjust and verify the preliminary roadway surrounding rock stability evaluation model to obtain an optimized roadway surrounding rock stability evaluation model.

[0077] In order to obtain a model that can quickly and accurately evaluate the stability of tunnel surrounding rock, the process of establishing the optimized tunnel surrounding rock stability evaluation model is as follows:

[0078] S41: First, use the Randperm function in MATLAB to randomly shuffle the order of multiple data sets in the database (66 data sets in this embodiment), and then screen them into a training set and a validation set in a ratio of 7:3;

[0079] S42: Establish PSO-BP neural network;

[0080] S421, determining the topological structure and number of nodes of the BP neural network;

[0081] S422, setting the initialization parameters of the particle swarm algorithm: learning factor, number of trainings, target error, learning rate, number of population updates, population size, position extreme value and velocity extreme value of particles, and optimizing the initialization weight and threshold of the BP neural network;

[0082] S423, determining an evaluation function of the BP neural network performance, and using a root mean square error (RMSE) to reflect the accuracy of the output data;

[0083] S424, determining the individual optimal solution and the overall optimal solution in the PSO algorithm;

[0084] S425, updating each particle according to the particle update formula (2);

[0085]

[0086] Where i = 1, 2, ..., N, N is the total number of particles in this group; v i is the velocity vector of particle i in d dimension; rand() is a random number between (0,1); x i is the position vector of particle i in d dimension; c1 and c2 are learning factors, c1=c2=2;

[0087] S426, when the number of iterations in the particle swarm optimization algorithm is equal to the preset number of training times or meets the target error, the iterative optimization process ends, otherwise returns to S423 for re-optimization;

[0088] S427, the optimal output value at the end of the iterative optimization, that is, the optimized weight and threshold, is substituted into the BP neural network for data training to achieve effective, accurate and fast data fitting;

[0089] S43: Establish GA-SVM deep learning model;

[0090] S431, setting initial parameters, kernel function K(x,z) and appropriate penalty function C in the support vector machine;

[0091] S432, setting initialization parameters of the genetic algorithm to optimize the initialization parameters in the support vector machine;

[0092] S433, determining an evaluation function of the support vector machine, and using mean square error (MSE) to reflect the accuracy of the output data;

[0093] S434, when the number of iterations in the genetic algorithm is equal to the preset number of training times or meets the target error, the iterative optimization ends, otherwise returns to S432 for re-optimization;

[0094] S435, substituting the optimal output value at the end of the iterative optimization, that is, the optimized c and g parameters, into the support vector machine for data training to achieve effective, accurate and fast data fitting;

[0095] S44: Construct a preliminary model for evaluating the stability of surrounding rock in tunnels;

[0096] S441: In the BP neural network model optimized based on the particle swarm algorithm, the vertical stress, horizontal stress, dynamic load intensity, dynamic load distance, side anchor length, bottom anchor length and bottom anchor angle in multiple data sets (66 data sets in this embodiment) are used as evaluation indicators, and the maximum displacement and maximum speed of the side are used as target values. After importing the training set and the validation set, the PSO-BP algorithm is used to improve the evaluation model; the numerical fitting regression prediction results in the validation set are shown in Figure 5 , the generalization ability of the tunnel safety evaluation model was evaluated using the validation set data. In the maximum displacement test set of the bottom plate, R = 0.92, and in the maximum velocity test set of the side, R = 0.93. Among them, R represents the correlation between the predicted value and the test value, and the value range is between 0 and 1. The larger the R value, the better the correlation. The predicted value is close to the measured value of the numerical simulation. The R value is close to 1, indicating that the model has a good degree of fit and high accuracy.

[0097] S442: In the support vector machine model optimized based on the genetic algorithm, the maximum displacement and maximum speed of the side parts in multiple data sets (66 data sets in this embodiment) are used as evaluation indicators, and whether the roadway is damaged is used as the target value. By quantifying whether it is damaged as 0 or 1, the evaluation model can make predictions. After importing the training set and the validation set, the GA-SVM algorithm is used to improve the evaluation model. The comparison between the predicted value and the simulated value in the validation set is shown in FIG. Figure 6 ,Compared with the actual damage of the tunnel in the numerical simulation, the prediction results are highly accurate, and the model accuracy is 100%;

[0098] Step 5: Use the optimized tunnel surrounding rock stability evaluation model to conduct stability evaluation on other mining stages of the tunnel to determine whether tunnel surrounding rock instability and damage will occur under the influence of dynamic loads under certain geological conditions, support methods and pressure relief measures. The evaluation result is that the tunnel is stable and no instability and damage occurs. Combined with the fact that no impact ground pressure occurs in the tunnel after a period of later construction, it can be seen that the tunnel surrounding rock stability evaluation of the present invention is more accurate.

[0099] Among them, the process of using the optimized tunnel surrounding rock stability evaluation model to evaluate the stability of the tunnel in other mining stages is as follows:

[0100] The specific parameters of dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress degree, support parameters and pressure relief measures are input into the optimized tunnel surrounding rock stability evaluation model. The optimized tunnel surrounding rock stability evaluation model determines whether the tunnel surrounding rock will become unstable and damaged under this condition through data prediction.

[0101] The present invention is based on numerical simulation, deep learning and engineering measured data. First, the influence of factors such as dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress level, support parameters and pressure relief measures on the stability of tunnel surrounding rock is analyzed by numerical simulation, and the dynamic load response characteristics of tunnel surrounding rock are recorded. It can be conveniently and comprehensively determined by data prediction whether the tunnel surrounding rock will be unstable and damaged under the working conditions where each influencing factor changes. Then, the deep learning part determines the nonlinear correspondence between the influencing factors and the dynamic load response characteristics of the tunnel surrounding rock based on the numerical simulation results, and then establishes an evaluation model, which can facilitate the use of the evaluation model to quickly evaluate the stability of the tunnel surrounding rock. Finally, the evaluation model can be corrected and improved by field data, which can significantly improve the accuracy of the evaluation model.

[0102] The present invention constructs an accurate, objective, reasonable and reliable model for evaluating the stability of surrounding rock of a tunnel, which overcomes the problems of large engineering workload, time-consuming and labor-intensive, high labor costs, and inability to conduct large-scale or all-round field measurements in traditional methods for evaluating the stability of surrounding rock of a tunnel. The method has the advantages of accurate, objective, reasonable and reliable prediction of the stability grade of surrounding rock of a mining tunnel and the model maturity evaluation results, and a high degree of consistency with the actual situation. It is helpful for scientifically designing tunnel anti-impact support and support methods, and can effectively improve the impact resistance of surrounding rock of a tunnel, thereby significantly reducing the degree and scope of impact damage. The present invention is of great significance for protecting the safety of construction personnel in the mining space, improving the safety of operations, and ensuring the safe production of mines. The method has good flexibility and a wide range of adaptability. It can realize rapid and accurate multi-group evaluation of the stability of surrounding rock of a tunnel under the same geological conditions, different dynamic loads, support unloading, etc., and can be adjusted and optimized according to the actual situation.

[0103] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A tunnel surrounding rock stability assessment method based on numerical simulation and deep learning, characterized in that: The specific steps include: Step 1: Determine the underground tunnel of the coal mine to be evaluated, collect rocks on the coal seam roof as samples, and use test equipment in the laboratory to test the physical and mechanical properties of the samples to obtain the actual parameters of the formation rock under the actual geological environment; Step 2: Based on the borehole histogram near the underground tunnel of the coal mine and the experimentally determined formation rock parameters, a two-dimensional geological model is established through numerical simulation; Step 3: In the two-dimensional geological model, the influencing factors are changed, including dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress degree, support parameters and pressure relief measures, and the deformation amount, acceleration value and whether obvious damage occurs in the tunnel side after the influencing factors are changed are recorded. Then, the extracted influencing factors and the dynamic response characteristics of the tunnel surrounding rock are used as labels, and the two are combined to obtain a data set. Finally, the obtained multiple data sets are formed into a database; Step 4: Divide the data set into a training set and a validation set according to a set ratio, and input the training set into the PSO-BP neural network and GA-SVM deep learning model for deep learning to obtain a preliminary roadway surrounding rock stability evaluation model based on numerical simulation and deep learning. Then, use the validation set to adjust and verify the preliminary roadway surrounding rock stability evaluation model to obtain an optimized roadway surrounding rock stability evaluation model. Step 5: Use the optimized tunnel surrounding rock stability evaluation model to conduct stability assessment on other mining stages of the tunnel to determine whether tunnel surrounding rock instability and failure will occur under the influence of dynamic loads under certain geological conditions, support methods and pressure relief measures.

2. A tunnel surrounding rock stability assessment method based on numerical simulation and deep learning according to claim 1, characterized in that: In step one, the actual parameters include density, bulk modulus, shear modulus, friction angle and cohesion.

3. A tunnel surrounding rock stability assessment method based on numerical simulation and deep learning according to claim 1 or 2, characterized in that: In step 2, the UDEC numerical simulation software is used to establish a three-dimensional geological model.

4. The method for evaluating tunnel surrounding rock stability based on numerical simulation and deep learning according to claim 3 is characterized in that: In step 2, the process of establishing a two-dimensional geological model is as follows: S21: Set the boundary conditions of the model: set the horizontal displacement in the x and y directions to zero, the lower boundary of the model is fixed, that is, the displacement of the lower boundary is zero, the upper boundary is the free boundary with equivalent load, and the yield criterion is selected Yield criterion, and ignoring plastic flow and dilatancy, the model is established with reference to the comprehensive column chart, and the model is parameterized based on the formation rock parameters in step 1; S22: applying loads on the top and both sides of the two-dimensional geological model according to the actual burial depth and the ground stress detection results, wherein the deadweight load on the top of the two-dimensional geological model is calculated by formula (1), and the loads applied on both sides are obtained by referring to the ground stress test results, and are calculated under given mechanical conditions and displacement boundary conditions, so that the model reaches an initial stress equilibrium state from an initial state; P = γH(1); Where, γ is the average bulk density of the rock layer, which is 25 kN / m3; H is the height of the upper boundary of the model from the ground surface, in meters; S23: The numerical model is divided into five layers using the crack command in UDEC, of ​​which four are roof layers, namely siltstone, argillaceous siltstone, siltstone and mudstone, and one is a coal seam in which the roadway is arranged.

5. The method for evaluating tunnel surrounding rock stability based on numerical simulation and deep learning according to claim 3 is characterized in that: In step three, the process of obtaining multiple sets of data sets is as follows: A tunnel is excavated in the built two-dimensional geological model, and survey lines are arranged on the sides and bottom of the tunnel. The survey lines are used to monitor the dynamic response characteristics of the tunnel surrounding rock after changing the influencing factors, that is, the deformation amount and acceleration value of the tunnel side are monitored, and whether the tunnel is damaged under this condition is recorded. The recorded influencing factors and the dynamic response characteristics of the tunnel surrounding rock are then used as labels and the two are combined into a data set; multiple groups of numerical simulation models are established according to different conditions, and multiple groups of data sets are obtained according to the above method.

6. The method for evaluating tunnel surrounding rock stability based on numerical simulation and deep learning according to claim 5, characterized in that: In step 4, the process of establishing the optimized tunnel surrounding rock stability evaluation model is as follows: S41: First, use the Randperm function in MATLAB to randomly shuffle the order of multiple data sets in the database, and then screen them into training set and validation set in a ratio of 7:3; S42: Establishing PSO-BP neural network; S421, determining the topological structure and number of nodes of the BP neural network; S422, setting the initialization parameters of the particle swarm algorithm: learning factor, number of trainings, target error, learning rate, number of population updates, population size, position extreme value and velocity extreme value of particles, and optimizing the initialization weight and threshold of the BP neural network; S423, determining that the evaluation function of the BP neural network performance is RMSE; S424, determining the individual optimal solution and the overall optimal solution in the PSO algorithm; S425, updating each particle according to the particle update formula (2); Where i = 1, 2, ..., N, N is the total number of particles in this group; v i is the velocity vector of particle i in d dimension; rand() is a random number between (0,1); x i is the position vector of particle i in d dimension; c1 and c2 are learning factors, c1=c2=2; S426, when the number of iterations in the particle swarm optimization algorithm is equal to the preset number of training times or meets the target error, the iterative optimization process ends, otherwise returns to S423 for re-optimization; S427, the optimal output value at the end of the iterative optimization, that is, the optimized weight and threshold, is substituted into the BP neural network for data training to achieve effective, accurate and fast data fitting; S43: Establish GA-SVM deep learning model; S431, setting initial parameters, kernel function K(x,z) and appropriate penalty function C in the support vector machine; S432, setting initialization parameters of the genetic algorithm to optimize the initialization parameters in the support vector machine; S433, determining that the evaluation function of the support vector machine is the mean square error; S434, when the number of iterations in the genetic algorithm is equal to the preset number of training times or meets the target error, the iterative optimization ends, otherwise returns to S432 for re-optimization; S435, substituting the optimal output value at the end of the iterative optimization, that is, the optimized c and g parameters, into the support vector machine for data training to achieve effective, accurate and fast data fitting; S44: Construct a preliminary model for evaluating the stability of surrounding rock in tunnels; S441: In the BP neural network model optimized based on particle swarm algorithm, vertical stress, horizontal stress, dynamic load intensity, dynamic load distance, side anchor length, bottom anchor length and bottom anchor angle in multiple data sets are used as evaluation indicators, and the maximum displacement and maximum speed of the side are used as target values. After importing the training set and the validation set, the PSO-BP algorithm is used to improve the evaluation model; S442: In the support vector machine model optimized based on genetic algorithm, the maximum displacement and maximum speed of the gangway in multiple data sets are used as evaluation indicators, and whether the tunnel is damaged is used as the target value. After importing the training set and the verification set, the GA-SVM algorithm is used to improve the evaluation model.

7. The method for evaluating tunnel surrounding rock stability based on numerical simulation and deep learning according to claim 5, characterized in that: In step 5, the process of using the optimized tunnel surrounding rock stability evaluation model to evaluate the stability of other mining stages of the tunnel is as follows: The specific parameters of dynamic load intensity, dynamic load action distance, dynamic load action time, static load stress degree, support parameters and pressure relief measures are input into the optimized tunnel surrounding rock stability evaluation model. The optimized tunnel surrounding rock stability evaluation model determines whether the tunnel surrounding rock will become unstable and damaged under this condition through data prediction.

Images