A reliability evaluation method for soft rock tunnels

Abstract

This invention relates to a method for reliability evaluation of soft rock tunnels, comprising: generating a specified number of parameter datasets and determining a numerical test scheme; performing numerical simulation calculations using FLAC3D finite difference software to obtain the displacement of specified measuring points, and then combining the parameter datasets to obtain an overall parameter-displacement dataset; constructing a fitting model of parameters and displacements using a multilayer neural network algorithm, and optimizing the hyperparameters of the multilayer neural network model using the firefly algorithm to obtain the optimal fitting model; collecting measured displacement data around the soft rock tunnel, and calculating the probability distribution of geotechnical parameters based on the pre-reserved deformation values ​​around the tunnel, combined with the optimal fitting model, using a Bayesian inference algorithm, and inverting the surrounding rock parameters to obtain optimized surrounding rock mechanical parameters, so as to further calculate the corresponding reliability index of the initial support structure of the soft rock tunnel. Compared with the prior art, this invention can quickly and accurately obtain the reliability evaluation results of soft rock tunnel structures.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering technology, and in particular to a method for evaluating the reliability of soft rock tunnels. Background Technology

[0002] With the rapid development of highway and high-speed railway construction in my country, soft rock tunnels, especially ultra-deep buried soft rock tunnels, are significantly affected by the uncertainties in terrain, route selection, and surrounding rock mechanical parameters. This impacts the stress state of the tunnel structure during excavation and support construction, posing substantial safety hazards. Therefore, it is necessary to conduct numerical simulation analysis and on-site monitoring measurements for soft rock tunnels to scientifically evaluate the reliability of the initial support structure.

[0003] Currently, when designing soft rock tunnel structures, commonly used methods for constructing soft rock tunnel function functions include polynomial fitting, neural network models, rational fractional models, and Kriging models. However, these methods often have limitations such as difficulty in solving, frequent oscillations in the results that are difficult to converge, and complex parameter estimation processes. In addition, most existing reliability assessments of soft rock tunnels adopt traditional machine learning methods, which on the one hand makes it difficult to guarantee the accuracy of the assessment results, and on the other hand, there is still considerable room for improvement in terms of computational efficiency. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for evaluating the reliability of soft rock tunnels, so as to obtain the structural reliability evaluation results of soft rock tunnels quickly and accurately.

[0005] The objective of this invention can be achieved through the following technical solution: a method for evaluating the reliability of soft rock tunnels, comprising the following steps:

[0006] S1. Generate a specified number of parameter datasets using the Latin hypercube sampling method and determine the numerical experiment plan;

[0007] S2. Based on the numerical test scheme determined in step S1, numerical simulation calculations are performed using FLAC3D finite difference software to obtain the displacement of the specified measuring point.

[0008] Combined with the parameter dataset generated in step S1, the overall parameter-displacement dataset is obtained;

[0009] S3. Construct a fitting model of parameters and displacement through a multi-layer neural network algorithm, and use the firefly algorithm to optimize the hyperparameters of the multi-layer neural network model, select the optimal combination of hyperparameters, and obtain the optimal fitting model.

[0010] S4. Collect measured displacement data around the soft rock tunnel, and based on the pre-reserved deformation value around the tunnel, combine the optimal fitting model and use the Bayesian inference algorithm to calculate the probability distribution of the rock and soil mechanical parameters. Based on this, the surrounding rock parameters to be determined are inverted to obtain the optimized surrounding rock mechanical parameters.

[0011] S5. Based on the calculation results of step S4, the corresponding reliability indicators of the initial support structure of the soft rock tunnel are further calculated.

[0012] Furthermore, step S1 specifically includes the following steps:

[0013] S11. Determine the surrounding rock parameters to be evaluated and their range of variation through on-site measured data;

[0014] S12. Using the Latin hypercube sampling method, generate a specified number of surrounding rock parameter datasets and determine the numerical test scheme.

[0015] Furthermore, the surrounding rock parameters include the elastic modulus of the surrounding rock, Poisson's ratio, the cohesion of the surrounding rock, and the internal friction angle of the surrounding rock.

[0016] Furthermore, step S2 specifically includes the following steps:

[0017] S21. Obtain the FLAC3D calculation file;

[0018] S22. Call the FLAC3D calculation file, record the displacement changes of each measuring point around the tunnel, and merge it with the parameter dataset generated in step S1.

[0019] S23. Determine whether the number of times the FLAC3D calculation file has been called has reached the preset threshold. If yes, proceed to step S24; otherwise, return to step S22.

[0020] S24. Output the merged overall parameter-displacement dataset.

[0021] Furthermore, step S3 specifically includes the following steps:

[0022] S31. Divide the overall parameter-displacement dataset into a training set and a test set according to a set ratio;

[0023] S32. By using a multi-layer neural network algorithm, establish the relationship between input and output sample data to construct a fitting model of parameters and displacement;

[0024] S33. Using the Firefly algorithm, combined with the training set and the test set, the hyperparameters of the fitting model constructed in step S32 are tuned to obtain the optimal fitting model.

[0025] Furthermore, the hyperparameters in step S33 include the learning rate, regularization parameter, number of neurons in each hidden layer, neuron activation function, number of layers in the neural network, scale of data participating in training the model, number of epochs for learning, and size of minibatch data.

[0026] Furthermore, step S33 specifically includes the following steps:

[0027] S331. Determine the adaptive movement step size α and the adaptive light absorption coefficient γ in the firefly algorithm;

[0028] S332. The Firefly Algorithm is used to iteratively call the relationship between input and output sample data. By solving the optimal solution of the objective function, the optimal hyperparameters of the fitting model are determined, thus obtaining the optimal fitting model.

[0029] Furthermore, the adaptive movement step size α and the adaptive light absorption coefficient γ in step S331 are specifically as follows:

[0030]

[0031]

[0032] Where α0 and γ0 are the initial movement step size and initial light intensity absorption coefficient, respectively, c is an integer determining the random decay rate, and Itr max Itr is the maximum number of iterations. i This represents the current iteration number.

[0033] Furthermore, step S4 specifically includes the following steps:

[0034] S41. Collect measured displacement data around the soft rock tunnel, and use the Bayesian inference algorithm to calculate the probability distribution of the geotechnical parameters corresponding to the measured displacement data around the soft rock tunnel, combined with the optimal fitting model.

[0035] S42. Based on the probability distribution of the geotechnical parameters corresponding to the measured displacement data around the soft rock tunnel in step S41, the confidence interval of the geotechnical parameters is obtained. The parameter value corresponding to the mean of the probability distribution is the optimized inverted surrounding rock mechanical parameters. The optimized inverted surrounding rock mechanical parameters include the calculated values ​​of the elastic modulus, cohesion, and internal friction angle of the surrounding rock corresponding to the reserved deformation value around the initial support structure of the soft rock tunnel.

[0036] Furthermore, the specific reliability indicators of the initial support structure for soft rock tunnels in step S5 are as follows:

[0037]

[0038] Where β is the reliability index, μ E and σ E These are the mean and standard deviation of the elastic modulus of the surrounding rock, μ c and σ c These are the mean and standard deviation of the cohesion of the surrounding rock, respectively. and E represents the mean and standard deviation of the internal friction angle of the surrounding rock, respectively. * c * , These are the calculated values ​​of the surrounding rock elastic modulus, cohesion, and internal friction angle corresponding to the reserved deformation value around the tunnel perimeter of the initial support structure of the soft rock tunnel obtained from the inversion calculation in step S42.

[0039] Compared with existing technologies, this invention is based on a constructed fitting model. By optimizing the fitting model and then using a Bayesian inference algorithm to calculate the probability distribution of geotechnical mechanical parameters of the optimal fitting model, the mean, variance, and standard deviation of the posterior distribution of the surrounding rock mechanical parameters can be determined, thereby obtaining statistically significant confidence intervals. This can effectively improve the calculation accuracy. Since the uncertainty of the regression calculation results of the surrounding rock mechanical parameters is fully considered, the reliability index is obtained more accurately, thus ensuring the accuracy of the reliability evaluation results.

[0040] This invention uses FLAC3D calculation files to record displacement changes at various measuring points around the tunnel, combines the generated parameter dataset, and merges them to obtain a parameter-displacement dataset. Then, a multi-layer neural network algorithm is used to construct a parameter-displacement fitting model, and the hyperparameters of the fitting model are tuned using the Firefly algorithm to obtain the optimal fitting model. This not only allows for the rapid and reliable acquisition of the optimal fitting model, but also ensures the accuracy of subsequent calculations to optimize the surrounding rock mechanical parameters. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0042] Figure 2 This is a schematic diagram of the reliability evaluation results in the embodiment. Detailed Implementation

[0043] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0044] Example

[0045] This technical solution was supported by the China Railway 20th Bureau Group Science and Technology Project (qzsycsd-202010-00002) and the Zhejiang Provincial Department of Transportation Science and Technology Project (2020035). The following is a combination of... Figure 1 and Figure 2 The contents of this technical solution are described in detail.

[0046] like Figure 1 As shown, a reliability evaluation method for soft rock tunnels includes the following steps:

[0047] Step 1: Generate a specified number of parameter datasets using the Latin hypercube sampling method and determine the numerical experiment plan;

[0048] Step 2: Based on the numerical test scheme determined in Step 1, numerical simulation calculations are performed using FLAC3D finite difference software to obtain the displacement of the specified measuring points. Combined with the parameter dataset, the overall parameter-displacement dataset is obtained.

[0049] Step 3: Construct a fitting model of parameters and displacement using a multilayer neural network algorithm, and use the firefly algorithm to optimize the hyperparameters of the multilayer neural network model, select the optimal combination of hyperparameters, and obtain the optimal fitting model;

[0050] Step 4: Collect measured displacement data around the soft rock tunnel, and based on the pre-designed deformation value around the tunnel, combine the optimal fitting model and use the Bayesian inference algorithm to calculate the probability distribution of the geotechnical parameters. Based on this, the surrounding rock parameters to be determined are inverted to obtain the optimized surrounding rock mechanical parameters.

[0051] Step 5: Based on the calculation results of Step 4, calculate the corresponding reliability index of the initial support structure for soft rock tunnels.

[0052] Step 1 specifically includes the following steps:

[0053] Step 1-1: Determine the parameters and range of the surrounding rock to be evaluated based on on-site measured data;

[0054] Steps 1-2: Use the Latin hypercube sampling method to generate a specified number of parameter datasets and determine the numerical experiment scheme.

[0055] Step 2 specifically includes the following steps:

[0056] Step 2-1: Obtain the FLAC3D calculation file;

[0057] Step 2-2: Call the FLAC3D calculation file from Step 2-1, record the displacement changes of the measuring points, and merge it with the parameter dataset from Step 1;

[0058] Step 2-3: Determine whether the current number of calls has reached the preset threshold. If yes, proceed to step 2-4; otherwise, return to step 2-2.

[0059] Steps 2-4: Obtain the global parameter-displacement dataset.

[0060] Step 3 specifically includes the following steps:

[0061] Step 3-1: In this embodiment, 70-80% of the total number of parameter and displacement data is used as the training set, and 20-30% of the total number of parameter and displacement data is used as the test set to complete the division of the parameter-displacement dataset;

[0062] Step 3-2: Establish a fitting model of parameters and displacement through a multilayer neural network algorithm. The multilayer neural network algorithm contains a network with multiple hidden layers and has the ability to handle linearly inseparable problems. It has shown its powerful performance in many fields. The multilayer neural network algorithm used in this technical solution can effectively replace the time-consuming calculation of the simulation model and achieve high accuracy.

[0063] Step 3-3: In the firefly algorithm, the adaptive movement step size α and the adaptive light absorption coefficient γ are calculated as follows:

[0064]

[0065]

[0066] In the formula, α0 and γ0 are the initial moving step size and the initial light intensity absorption coefficient, respectively; c is an integer that determines the random decay rate, generally taken as c = 5; Itr max Itr represents the maximum number of iterations. i This represents the current iteration number.

[0067] The Firefly Algorithm is used to fine-tune the hyperparameters of the fitted model constructed in step 3-2, including the learning rate, regularization parameter, number of neurons in each hidden layer, neuron activation function, encoding method of output neurons, choice of cost function, weight initialization method, number of neural network layers, scale of training data, number of epochs, and size of minibatch data, in order to obtain the optimal fitted model.

[0068] Step 4 specifically includes the following steps:

[0069] Step 4-1: Collect measured displacement data around the soft rock tunnel. Based on the optimal fitting model in Step 3-3, use the Bayesian inference algorithm to calculate the probability distribution of the geotechnical parameters corresponding to the measured displacement data around the soft rock tunnel.

[0070] Step 4-2: Based on the probability distribution of the geotechnical parameters corresponding to the measured displacement data around the soft rock tunnel in Step 4-1, obtain the confidence interval of the geotechnical parameters (which can be considered as needed in engineering). The parameter value corresponding to the mean of the probability distribution is the optimized inverted surrounding rock mechanical parameters.

[0071] Step 5 involves calculating the reliability index β of the initial support structure of the soft rock tunnel based on the surrounding rock mechanical parameters corresponding to the pre-reserved deformation values ​​around the tunnel obtained in Step 4-2, using the following formula:

[0072]

[0073] In the formula, μ E and σ E These are the mean and standard deviation of the elastic modulus of the surrounding rock, μ c and σ c These are the mean and standard deviation of the cohesion of the surrounding rock, respectively. and E represents the mean and standard deviation of the internal friction angle of the surrounding rock, respectively. * c * , These are the calculated values ​​of the surrounding rock elastic modulus, cohesion, and internal friction angle corresponding to the pre-reserved deformation value around the tunnel perimeter of the initial support structure for soft rock tunnels obtained in step 4-2.

[0074] In this embodiment, a soft rock tunnel on a highway is selected as the evaluation object. The surrounding rock of the tunnel is fully weathered granite gneiss, with a height of 10.23m and a span of 12.46m. The initial support design parameters are as follows: anchor length φ25-5 = 3.5m; grid steel frame spacing φ25 = 0.6m; sprayed concrete thickness 25cm; steel mesh φ8 = 20×20cm. The tunnel depth is 148m. The horizontal ground stress in the tunnel area is 4.125~4.411MPa. The reserved deformation of the initial support structure is 38cm. The range of surrounding rock parameters is shown in Table 1. Since the surrounding rock parameters of the tunnel have a large range of variation, to ensure construction safety and reliability, the reliability of the reserved deformation of the initial support structure is evaluated based on monitoring data.

[0075] Table 1 Range of variation of surrounding rock parameters

[0076]

[0077] Based on the value ranges shown in Table 1, 2000 sets of surrounding rock parameter data were generated using the Latin hypercube sampling method. A FLAC3D model of the tunnel was then established, and the displacement values ​​at each measuring point were calculated. The established dataset was input into a multilayer neural network algorithm for learning, and the hyperparameters of the multilayer neural network algorithm model were optimized using the firefly algorithm to obtain the optimal fitting model between the surrounding rock parameters and displacement.

[0078] In this embodiment, the probability distribution of the geotechnical mechanical parameters corresponding to the measured displacement data around the soft rock tunnel is calculated using a Bayesian inference algorithm, yielding the confidence intervals of the geotechnical mechanical parameters (which can be considered as needed in engineering). The parameter values ​​corresponding to the mean of the probability distribution are the optimized inverted surrounding rock mechanical parameters. Finally, the reliability index β of the initial support structure of the soft rock tunnel is calculated using the following formula:

[0079]

[0080] Finally, the reliability indices of each measuring point in the soft rock tunnel were obtained as follows: Figure 2 As shown in the figure (the numbers in the figure are reliability indicators).

[0081] In summary, this technical solution can obtain the mean, variance, and standard deviation of the posterior distribution of the surrounding rock mechanical parameters, and thus obtain statistically significant confidence intervals, which has the advantages of high precision and high accuracy.

Claims

1. A method for evaluating reliability of a soft rock tunnel, characterized by, The method comprises the following steps: S1, generating a specified number of parameter data sets by using a Latin hypercube sampling method, and determining a numerical test scheme; S2, performing numerical simulation calculation by using FLAC3D finite difference software according to the numerical test scheme determined in step S1 to obtain displacement values of specified measuring points; and combining the parameter data sets generated in step S1 to obtain an overall parameter-displacement data set; S3, constructing a fitting model of parameters and displacement by using a multilayer neural network algorithm, and optimizing hyperparameters of the multilayer neural network model by using a glowworm algorithm to select an optimal hyperparameter combination to obtain an optimal fitting model; S4, collecting measured displacement data of a soft rock tunnel hole, and combining the optimal fitting model to calculate the probability distribution of geotechnical mechanics parameters by using a Bayesian inference algorithm, thereby inversely calculating the to-be-solved surrounding rock parameters to obtain optimized surrounding rock mechanics parameters; The surrounding rock parameters include the elastic modulus of the surrounding rock, the Poisson's ratio, the cohesion of the surrounding rock and the internal friction angle of the surrounding rock; Step S3 specifically comprises the following steps: S31, dividing the overall parameter-displacement data set into a training set and a test set according to a set proportion; S32, establishing a relationship between input and output sample data by using a multilayer neural network algorithm to construct a fitting model of parameters and displacement; S33, optimizing hyperparameters of the fitting model constructed in step S32 by using a glowworm algorithm in combination with the training set and the test set to obtain an optimal fitting model, wherein the hyperparameters include a learning rate, a regularization parameter, the number of neurons in each hidden layer, a neuron activation function, the number of layers of the neural network, the size of the data participating in the training model, the number of learning rounds Epoch, and the size of the minibatch data; Step S33 specifically comprises the following steps: S331, determining an adaptive moving step α and an adaptive light absorption coefficient γ in the glowworm algorithm: Wherein, a0, g0 are initial moving step and initial light intensity absorption coefficient respectively, c is an integer determining the randomness decay speed, Itr max is the maximum iteration number, Itr i is the current iteration number; S332, continuously calling the relationship between input and output sample data by using the glowworm algorithm to determine the optimal hyperparameters of the fitting model by solving the optimal solution of the objective function, that is, to obtain the optimal fitting model; Step S4 specifically comprises the following steps: S41, collecting measured displacement data of a soft rock tunnel hole, combining the optimal fitting model, and calculating the probability distribution of geotechnical mechanics parameters corresponding to the measured displacement data of the soft rock tunnel hole by using a Bayesian inference algorithm; S42, obtaining the confidence interval of the geotechnical mechanics parameters according to the probability distribution of the geotechnical mechanics parameters corresponding to the measured displacement data of the soft rock tunnel hole in step S41, wherein the parameter value corresponding to the mean of the probability distribution is the optimized inverse surrounding rock mechanics parameter, and the optimized inverse surrounding rock mechanics parameter includes the calculated values of the elastic modulus, the cohesion and the internal friction angle of the surrounding rock corresponding to the reserved deformation value of the initial support structure of the soft rock tunnel hole; S5, further calculating the corresponding reliability index of the initial support structure of the soft rock tunnel according to the calculation result of step S4: wherein β is a reliability index, μ E and σ E are the mean value and standard deviation of the elastic modulus of the surrounding rock, μ c and σ c are the mean value and standard deviation of the cohesion of the surrounding rock, and are the mean value and standard deviation of the internal friction angle of the surrounding rock, E * , c * , are the calculated values of the elastic modulus, cohesion and internal friction angle of the surrounding rock corresponding to the reserved deformation value of the initial support structure around the soft rock tunnel obtained by the inversion calculation in step S42.

2. The soft rock tunnel reliability evaluation method according to claim 1, characterized in that, The step S1 specifically comprises the following steps: S11, determining the to-be-evaluated surrounding rock parameters and their variation ranges through field measurement data; S12, a Latin hypercube sampling method is used to generate a specified number of surrounding rock parameter data sets, and a numerical test scheme is determined.

3. The soft rock tunnel reliability evaluation method of claim 1, wherein, The step S2 specifically comprises the following steps: S21, obtaining a FLAC3D calculation file; S22, calling the FLAC3D calculation file, recording the displacement change of each measuring point around the hole, and merging with the parameter data set generated in step S1; S23, judging whether the number of times of calling the FLAC3D calculation file reaches a preset threshold value, if the judgment is yes, step S24 is executed, otherwise step S22 is returned; S24, outputting the obtained merged overall parameter-displacement data set.

Images