An automatic inversion method for mechanical parameters of surrounding rock of tunnel with developed joints

Abstract

The invention relates to the field of geotechnical engineering, and in particular to an automatic inversion method for mechanical parameters of surrounding rocks of a tunnel with developed joints, comprising the steps of S1: obtaining initial support parameter values ​​and joint parameter values ​​of surrounding rocks at a certain section of the tunnel, and establishing a tunnel excavation numerical calculation model based on the initial support parameter values ​​and the joint parameter values; S2: taking multiple groups of values ​​of mechanical parameters and inputting them into the tunnel excavation numerical calculation model respectively to obtain multiple surrounding rock deformation curves, and obtaining the deformation amount of the surrounding rock at the end of the rapid deformation stage of the surrounding rock, taking the deformation amount as the feature of the training sample, and taking the multiple groups of values ​​of the mechanical parameters as the labels of the training samples to obtain multiple groups of training samples; S3: training a surrounding rock mechanical parameter inversion model using the multiple groups of training samples obtained in S2; and S4: inputting the deformation amount of the surrounding rock at the end of the rapid deformation stage measured at a construction site into the surrounding rock mechanical parameter inversion model trained in S3 to obtain the mechanical parameters of the surrounding rock at the construction site.

Description

Technical Field

[0001] The invention relates to a surrounding rock mechanical parameter inversion method, in particular to an automatic inversion method for surrounding rock mechanical parameter of a tunnel with developed joints. Background Art

[0002] As one of the important means to carry out tunnel safety risk assessment, ensuring the precision and reliability of numerical simulation results is the premise for accurate assessment of tunnel construction safety risks. The mechanical parameters of surrounding rock are the main quantitative input indicators in the process of assessing tunnel safety risks. Whether their values ​​are consistent with the actual values ​​of the mechanical parameters of the surrounding rock at the construction site directly determines whether the calculation results are reliable. However, in actual engineering, the mechanical parameters of surrounding rock are often difficult to obtain in a timely and accurate manner due to factors such as the compact construction process and time-consuming and labor-intensive indoor tests. More importantly, due to the fragmented state of the rock mass with developed joints, the on-site coring work alone is greatly restricted. There is no reliable method to obtain core samples that fully reflect the properties of the jointed rock mass and can be effectively used for indoor experiments. In view of the above reasons, the method of obtaining the mechanical parameters of surrounding rock through inversion has begun to be proposed. The research on the inversion of geotechnical engineering parameters began in the 1970s. It has gone through three stages of germination, development, and promotion and application. Its feasibility and effectiveness have been tested in practice.

[0003] With the advancement and development of computer technology, machine learning has been relatively comprehensively applied in the field of geotechnical engineering, and the inversion of rock mechanical parameters using artificial intelligence algorithms has become a research hotspot in recent years.

[0004] Generally speaking, the main research technical processes in this regard are:

[0005] 1. Establish a tunnel excavation numerical calculation model that matches the tunnel section to be inverted;

[0006] 2. Given a combination of mechanical parameters to be inverted (e.g. elastic modulus E of the surrounding rock, Poisson's ratio μ of the surrounding rock, internal friction angle of the surrounding rock The surrounding rock cohesion c, surrounding rock gravity γ, etc.) are input into the tunnel excavation numerical calculation model for calculation to obtain the deformation results of the surrounding rock. Then the input parameter combination is used as the label of the machine learning sample, and the obtained surrounding rock deformation results (such as: arch settlement, peripheral convergence) are used as the features of the machine learning sample to obtain multiple groups of sample data for machine learning model training;

[0007] 3. Train and modify the machine learning model based on the given samples, and finally obtain the corresponding regression prediction model;

[0008] 4. Based on the obtained regression prediction model, the measured surrounding rock deformation is used as parameter input to obtain the predicted mechanical parameter combination of the surrounding rock.

[0009] From the above, we can see that the key issue of this technical work lies in the establishment of machine learning samples and the subsequent selection of models. The current research on tunnel surrounding rock mechanical parameter inversion mainly focuses on machine learning algorithm optimization and model fusion to improve the reliability of parameter inversion results.

[0010] The reliability of inverting surrounding rock mechanical parameters using machine learning methods is not only subject to the adaptability of the inversion model, but also greatly depends on the accuracy and rationality of the learning sample data. The accuracy of the sample data mainly depends on whether the numerical calculation model of tunnel excavation can effectively reflect the situation of the surrounding rock on site, that is, the rationality of the constitutive model; in terms of the rationality of the sample data, since the deformation of the surrounding rock is a process that changes with time and construction progress, how to select the deformation data at the appropriate node as the sample feature is the key to improving the reliability of the inversion of surrounding rock mechanical parameters. However, the current research on inverting surrounding rock mechanical parameters using machine learning methods has problems in the following aspects:

[0011] 1. The numerical calculation model of tunnel excavation cannot reflect the actual surrounding rock conditions

[0012] In existing studies, most tunnel excavation numerical calculation models use the Mohr Coulomb general constitutive model. However, as a classic ideal elastic-plastic constitutive model, the Mohr Coulomb model is difficult to reflect the mechanical properties of engineering rock masses with developed joints and cracks. As a result, the tunnel excavation numerical calculation model cannot reflect the actual surrounding rock conditions at the construction site, and it is difficult to ensure the reliability of the sample data constructed based on this model. This greatly restricts the effectiveness and accuracy of the inversion results of the surrounding rock mechanical parameters and the deformation prediction results based on the inversion parameters.

[0013] 2. Select the time node of surrounding rock deformation data

[0014] "Analysis of the temporal and spatial effects of surrounding rock deformation in rock tunnels", Sun Yuanchun and Shang Yanjun, Journal of Engineering Geology, 16(02), pp. 211-215, 2008, gives a theoretical curve of the whole process of surrounding rock deformation during tunnel excavation, such as Figure 1As shown in the figure, after point D, the deformation of the surrounding rock is mainly affected by the time effect, in the BC section, it is mainly affected by the space effect, and in the CD section, it is affected by the space effect and the time effect. Compared with the time effect, the deformation of the surrounding rock generated under the space effect has a higher correlation with the mechanical parameters of the surrounding rock. Based on the statistical analysis of existing related studies, it is found that in current studies, the selection of the time to measure the deformation of the surrounding rock is relatively arbitrary. Most of them select the deformation of the surrounding rock from the beginning of the model excavation to the model excavation through and basically balanced as the characteristics of the training sample, and do not consider the different excavation methods and the process of the process The impact of the surrounding rock deformation caused by the process. There are two problems in doing this: First, the influence of the time effect on the surrounding rock deformation is not excluded, resulting in the selected surrounding rock deformation being difficult to effectively reflect the actual surrounding rock mechanical parameters; second, it is not combined with the actual deformation law of the surrounding rock monitored on site, ignoring the influence of the on-site construction process on the surrounding rock deformation. These two problems will eventually reduce the reliability of the inversion of the surrounding rock mechanical parameters.

[0015] 3. Machine Learning Model

[0016] Although a lot of research has been done on machine learning models, there are still some deficiencies in the application of this technology. In particular, the problem of obtaining multiple different types of surrounding rock mechanical parameters (i.e., multi-label regression prediction) based on multiple surrounding rock deformation data regression prediction involved in this technical work has not been well solved. In addition, the adjustment of model hyperparameters in most studies relies on manual settings, which makes it difficult for the model to achieve optimal performance and takes a long time. Summary of the invention

[0017] The purpose of the present invention is to provide an automatic inversion method for the mechanical parameters of surrounding rocks of tunnels with developed joints, in view of the problems that the numerical calculation model of tunnel excavation in the prior art is difficult to reflect the actual surrounding rock conditions at the construction site, resulting in unreliable sample data constructed and reduced accuracy of the mechanical inversion model of surrounding rock parameters; and the problem that when selecting the deformation of the surrounding rock as the sample feature of the training sample, the influence of the time effect on the deformation of the surrounding rock is not taken into account, resulting in inaccurate surrounding rock mechanical parameters output by the trained surrounding rock mechanical parameter inversion model and unreliable surrounding rock mechanical parameter inversion model.

[0018] A method for automatically inverting mechanical parameters of surrounding rock of a tunnel with developed joints, comprising the steps of:

[0019] S1: obtaining the initial support parameter values ​​and joint parameter values ​​of the surrounding rock at a certain section in the tunnel, and establishing a tunnel excavation numerical calculation model based on the initial support parameter values ​​and the joint parameter values, wherein the rock mass within the loose zone of tunnel excavation adopts a joint-diffuse constitutive model, and the rock mass outside the loose zone adopts a Mohr-Coulomb constitutive model;

[0020] S2: taking multiple sets of values ​​of mechanical parameters and inputting them into the tunnel excavation numerical calculation model respectively, to obtain multiple surrounding rock deformation curves, all of which are recorded as second deformation curves; obtaining the deformation amount of the surrounding rock at the end of the rapid deformation stage of the surrounding rock on each of the second deformation curves, taking the deformation amount as the feature of the training sample, taking the multiple sets of values ​​of the mechanical parameters as the labels of the training samples, and combining the features of the training samples with the labels of the training samples to obtain multiple sets of training samples;

[0021] S3: using the multiple groups of training samples obtained in S2 to train the surrounding rock mechanical parameter inversion model;

[0022] S4: Inputting the deformation of the surrounding rock at the end of the rapid deformation stage measured at the construction site into the surrounding rock mechanical parameter inversion model trained in S3, to obtain the mechanical parameters of the surrounding rock at the construction site.

[0023] In the above technical solution, the numerical calculation model of tunnel excavation is used to provide training samples for the inversion model of the mechanical parameters of the surrounding rock. The numerical calculation model of tunnel excavation includes the pervasive joint constitutive model and the Mohr-Coulomb constitutive model. Joints control the mechanical characteristics of the rock mass, such as strength and stress, to a large extent, and also determine the failure mechanism of the rock mass. Therefore, the numerical calculation model of tunnel excavation considering the influence of joints can better reflect the mechanical properties of engineering rock mass with developed joints and cracks, simulate the deformation and stress of the tunnel surrounding rock, and construct training sample data that is closer to the actual engineering.

[0024] The tunnel excavation numerical calculation model in the above technical solution needs to provide initial support parameters, joint parameters and mechanical parameters of the surrounding rock. During the tunnel excavation process, within the scope of the tunnel of tens of meters in length, it can be considered that the values ​​of the initial support parameters and the values ​​of the joint parameters are unchanged. Therefore, it is only necessary to input the initial support parameter values ​​and joint parameter values ​​of a certain section of the surrounding rock measured in the tunnel excavation numerical calculation model in step S1, and then input multiple sets of values ​​of the surrounding rock mechanical parameters into the tunnel excavation numerical calculation model to obtain multiple sets of deformations of the surrounding rock. The multiple sets of deformations of the surrounding rock and the multiple sets of values ​​of the surrounding rock mechanical parameters are combined into training samples, which can be used to train the surrounding rock mechanical parameter inversion model. By inputting the surrounding rock deformation measured at the construction site into the surrounding rock mechanical parameter inversion model after the training is completed, the mechanical parameters of the surrounding rock at that location can be obtained, and the obtained surrounding rock parameters can be used to assess tunnel risks. The above scheme is used to automatically obtain the mechanical parameters of the surrounding rock through the surrounding rock mechanical parameter inversion model. Compared with obtaining the mechanical parameters of the surrounding rock through on-site sampling, it is more convenient and efficient and can ensure the accuracy of the data.

[0025] Among them, the deformation amount of the surrounding rock at the end of the rapid deformation stage of the surrounding rock obtained on each second deformation curve is used as the feature of the training sample. The feature of the training sample is highly correlated with the mechanical parameters of the surrounding rock (the label of the training sample) and is not affected by the time effect. Therefore, when training the surrounding rock mechanical parameter inversion model and using the surrounding rock mechanical parameter inversion model to obtain the numerical value of the surrounding rock mechanical parameter, the deformation amount that has occurred at the end of the rapid deformation stage of the surrounding rock is used, which can reduce the error between the numerical value of the surrounding rock mechanical parameter obtained by inversion and the actual numerical value of the surrounding rock mechanical parameter, and improve the reliability of the surrounding rock mechanical parameter inversion model.

[0026] In addition, since the training samples are provided by the numerical calculation model of tunnel excavation, there is no need to sample the surrounding rock at multiple cross-section positions during the tunnel excavation process and then measure the surrounding rock deformation at each cross-section position. This overcomes the problem that it is difficult to obtain training samples for the inversion model of surrounding rock mechanical parameters in a timely and accurate manner due to many factors such as the compact construction process and time-consuming and labor-intensive indoor experiments.

[0027] As a preferred embodiment of the present invention, in S1, the deformation of the surrounding rock of the section during the tunnel excavation process is monitored to obtain a deformation curve of the surrounding rock at the section over time, which is recorded as a first deformation curve; the point on the first deformation curve where the rapid deformation stage of the surrounding rock ends is recorded as point C′, and the time t elapsed from the start of excavation to point C′ of the section is recorded; in S2, the method for obtaining the deformation of the surrounding rock at the end of the rapid deformation stage of the surrounding rock from each of the second deformation curves is as follows: from each of the second deformation curves, the deformation of the surrounding rock at the tunnel section corresponding to t time after the excavation of the tunnel section is selected, and the deformation is the deformation of the surrounding rock at the end of the rapid deformation stage of the surrounding rock.

[0028] Since the construction procedures of the excavation process of each section of the tunnel are the same in the same project, the deformation law of the surrounding rock at each section of the tunnel after excavation should also be consistent. Therefore, since the surrounding rock at a certain position monitored at the construction site ends the sharp deformation stage after excavation time t, the deformation of the surrounding rock at other positions in the tunnel at time t after excavation is the deformation at the end of the sharp deformation stage of the surrounding rock. Through the above method, the deformation of the surrounding rock at the end of the sharp deformation stage can be quickly and accurately obtained from the second deformation curve generated by the tunnel excavation numerical calculation model. It is only necessary to monitor the deformation of the surrounding rock of the tunnel section from excavation to time t, and there is no need to compare each generated surrounding rock change curve with the theoretical tunnel excavation surrounding rock deformation process curve, which improves efficiency, can quickly obtain the training sample data of the surrounding rock mechanical parameter inversion model, and ensure the accuracy of the trained surrounding rock parameter mechanical inversion model.

[0029] As a preferred embodiment of the present invention, in S4, the deformation amount of the surrounding rock measured at the construction site at the end of the rapid deformation stage is the deformation amount of the surrounding rock produced at t time after the excavation of the section.

[0030] As a preferred embodiment of the present invention, the surrounding rock mechanical parameter inversion model is an RC-XGBoost model, and the RC-XGBoost model is a regressor chain that uses a gradient boosting algorithm for optimization and solution.

[0031] The gradient boosting algorithm (XGBoost) builds multiple weak evaluators on the data one by one, and gradually accumulates and summarizes the modeling results of multiple weak evaluators through multiple iterations to obtain better regression or classification performance than a single model. Regressor Chain is a chain method based on a single-target model. The training process of the Regressor Chain model can be simply understood as: selecting a random sequence chain of a set of target variables, and then sequentially establishing a multi-target regression prediction model for each target. Since the prediction results of the previous target are continuously added to the training set during the modeling process, it becomes a feature of the next target prediction, thereby realizing the consideration of the correlation between multiple targets. For the prediction of multiple targets with specific potential correlations, better prediction performance can be achieved. Using the XGBoost algorithm to optimize and solve the Regressor Chain model can consider the correlation between targets and make the prediction results more accurate.

[0032] As a preferred embodiment of the present invention, the mechanical parameters include the elastic modulus E of the surrounding rock, the Poisson's ratio μ of the surrounding rock, the internal friction angle The cohesion c of the surrounding rock and the density γ of the surrounding rock; the initial support parameters include the thickness of the initial support, the elastic modulus of the initial support, the Poisson's ratio of the initial support and the density of the initial support; the joint parameters include the inclination angle of the joint, the tendency of the joint, the tensile strength of the joint, the cohesion of the joint and the internal friction angle of the joint.

[0033] As a preferred solution of the present invention, in S2, multiple groups of values ​​of mechanical parameters serving as training sample labels are obtained by orthogonal test design of the mechanical parameters.

[0034] The data used as training sample labels is obtained through orthogonal experimental design, which can reduce the number of training samples and save training time, while ensuring that the training samples are representative.

[0035] As a preferred solution of the present invention, in the tunnel excavation numerical calculation model, a rapid stress boundary method is used to simulate the initial geostress field.

[0036] The rapid stress boundary method is used to simulate the initial geostress field, which can quickly and effectively simulate the initial stress state of the deep buried tunnel.

[0037] As a preferred solution of the present invention, the distance between the left and right boundaries of the tunnel excavation numerical calculation model is five times the diameter of the tunnel to be excavated, the upper boundary is located within 40m to 50m upward from the arch line of the tunnel to be excavated, and the lower boundary is located within 40m to 50m downward from the arch line of the tunnel to be excavated, and the length of the tunnel excavation numerical calculation model along the tunnel excavation direction is 50m to 60m.

[0038] Through the above method, the relationship between the accuracy of numerical calculation results and the calculation efficiency can be balanced.

[0039] As a preferred embodiment of the present invention, the end of the rapid deformation stage of the surrounding rock is the moment when the slope of the deformation curve of the surrounding rock is 1.

[0040] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0041] 1. Aiming at the problem that it is difficult to obtain the surrounding rock mechanical parameters of jointed rock tunnels, the present invention uses numerical simulation and machine learning technology to establish a mapping relationship between tunnel surrounding rock mechanical parameters and surrounding rock deformation, thereby constructing an inversion model of surrounding rock mechanical parameters. The inverted surrounding rock mechanical parameters can be used for tunnel surrounding rock deformation prediction and support structure stability analysis, effectively solving the problem that it is difficult to quickly obtain surrounding rock mechanical parameters for analysis and calculation at the tunnel site;

[0042] 2. By introducing the Mohr-Coulomb and ubiquitous-joint joint constitutive model into the tunnel excavation numerical calculation model and reasonably assigning the surrounding rock mechanical parameters through orthogonal tests based on the results of on-site joint investigation, the numerical calculation model of tunnel excavation is ensured to have good consistency with the surrounding rock conditions on the tunnel site, so that the numerical simulation deformation results are in good agreement with the on-site monitoring deformation results;

[0043] 3. By analyzing the deformation curve of the whole process of tunnel surrounding rock deformation, the start and end time nodes of the surrounding rock deformation monitoring value in the process of using surrounding rock deformation to invert surrounding rock mechanical parameters are reasonably determined, which can effectively improve the reliability of sample data, thereby improving the accuracy of the output results of the surrounding rock mechanical parameter inversion model;

[0044] 4. After actual engineering verification, the surrounding rock mechanical parameters inverted by this method have high accuracy for forward analysis and fully meet the accuracy requirements of tunnel engineering analysis and calculation. At the same time, the surrounding rock mechanical parameter inversion model constructed by this method can quickly and automatically invert the surrounding rock mechanical parameters after inputting the actual surrounding rock deformation. Using it for analysis and calculation to guide construction can fully meet the timeliness of engineering construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is the theoretical curve of the whole process of surrounding rock deformation during tunnel excavation;

[0046] Figure 2 This is a schematic diagram of training the RC-XGBoost model;

[0047] Figure 3 It is a structural diagram of the tunnel lining support form;

[0048] Figure 4 It is a schematic diagram of the numerical calculation model of tunnel excavation;

[0049] Figure 5 It is a schematic diagram of setting deformation monitoring points on the tunnel section;

[0050] Figure 6 is through Figure 4 The tunnel deformation curve diagram obtained by plotting the monitoring points in;

[0051] Figure 7 It is a statistical chart of the accuracy verification results of the parameter inversion model. DETAILED DESCRIPTION

[0052] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] Example 1

[0055] An automatic inversion method for mechanical parameters of surrounding rock in tunnels with developed joints, such as Figure 2 As shown, by constructing a surrounding rock mechanics parameter inversion data set (training sample data set), the constructed training sample data set is sent to the surrounding rock mechanics parameter inversion model, the surrounding rock mechanics parameter inversion model is trained, and the model that meets the accuracy requirements after training is saved and used in actual engineering.

[0056] Taking the section K32+400~k32+360 at the right exit of Changning Tunnel as an example, the tunnel burial depth within the section is about 300m, the surrounding rock is mainly composed of strongly weathered metamorphic sandstone intercalated with schist, with developed joints and fissures, and the rock mass is mostly layered. There is earthy yellow mud filling between the structural surfaces, and the interlayer bonding is poor. The designed surrounding rock grade is V, and composite lining support is adopted. The support form is as follows: Figure 3 During the on-site construction, engineering diseases such as face collapse, support structure damage and large deformation occurred from time to time. The tunnel surrounding rock is a typical joint-developed rock mass.

[0057] Before inverting the mechanical parameters of the surrounding rock in this section, a numerical calculation model of tunnel excavation was established to obtain training samples. The FLAC3D finite difference program was used to calculate the tunnel excavation model according to the right tunnel exit K32+400~K32+350 of Changning Tunnel (the size parameters of the tunnel are as follows: Figure 3 The numerical calculation model of tunnel excavation is constructed as shown in Figure 4 As shown in the figure, the core rock mass (Core Rock) within the loose zone of tunnel excavation adopts the ubiquitous-joint constitutive model, and the outer rock mass (OutRock) of the loose zone adopts the Mohr-Coulomb constitutive model. When establishing the numerical calculation model of tunnel excavation, the distance between the left and right boundaries is 5 times the width of the tunnel to be excavated, the upper boundary is located within 40m to 50m above the arch line of the tunnel to be excavated, and the lower boundary is located within 40m to 50m below the arch line of the tunnel to be excavated.

[0058] The buried depth of the Changning Tunnel research section is 300m. During the formation of the initial geostress field, the tunnel excavation numerical calculation model does not set velocity boundary conditions. Only stress boundary conditions are applied on the model surface according to the distribution of the geostress field and kept constant. This is equivalent to adding tectonic stress boundary conditions to the model. The node forces at the contact surface between the surface unit and the internal unit of the model propagate into the model, so that the surface stress field distribution pattern diffuses into the model until equilibrium is reached. At this time, the stress field obtained can be considered as the result of the superposition of the tectonic stress field and the self-weight stress field. Since the model has no displacement boundary, a large displacement will be generated during the model equilibrium process, but this can be eliminated by clearing the displacement of all nodes in the model after the initial stress field is generated. At the same time, before the subsequent calculation, a boundary condition with zero velocity can be added. This method can quickly simulate the initial geostress field of the deep buried tunnel. That is, the rapid stress boundary method (SB method) is used to simulate the initial geostress field of this study section.

[0059] Before tunnel excavation, the mechanical parameters of the surrounding rock measured are recorded in the geological survey data. The mechanical parameters of the surrounding rock can be selected according to the recommended values ​​of the previous geological survey data. For the joint parameters that need to be input in the pervasive joint model, the inclination and inclination of the joints are determined according to the on-site joint measurement, the tensile strength and cohesion of the joints are 1% of the intact rock mass, and the friction angle of the joints is 30°. The thickness of the initial support, the elastic modulus of the initial support, the Poisson's ratio of the initial support and other parameters are directly or converted according to the design data. Considering that the deformation after the secondary lining is mainly affected by creep and seepage, and has a low correlation with the mechanical parameters of the surrounding rock, the tunnel excavation numerical calculation model does not establish the tunnel secondary lining structure. The values ​​of the input parameters of the tunnel excavation numerical calculation model are shown in Table 1.

[0060] Table 1 Input parameter values ​​of tunnel excavation numerical calculation model

[0061]

[0062] At the right exit of Changning Tunnel K32+375 section, the following Figure 5 There are five displacement deformation monitoring points shown in the figure, among which A0, A1, and A2 are the arch sinking monitoring points, and the line segment L formed by the monitoring points A3 and A4 is the peripheral convergence monitoring line; sampling is carried out at the section to measure the mechanical parameter values ​​of the surrounding rock at that location (the mechanical parameter values ​​of the surrounding rock at that location can also be selected based on the previous geological survey data), and the measured values ​​are input into the tunnel excavation numerical calculation model to obtain the deformation curve of the surrounding rock. The deformation curve of the surrounding rock drawn from the excavation of the tunnel section by the five monitoring points is as follows Figure 6 As shown, Figure 6 The surrounding rock deformation curve in Figure 1 The theoretical tunnel excavation surrounding rock deformation curves in the whole process are compared. Figure 6 Find the point corresponding to point C on the surrounding rock deformation curve in (i.e., the point where the rapid deformation stage of the surrounding rock ends in the surrounding rock deformation curve monitored at the construction site), and record it as C′, and the time corresponding to C′ is recorded as t.

[0063] Elastic modulus E of surrounding rock, Poisson's ratio μ of surrounding rock, internal friction angle of surrounding rock The cohesion c of the surrounding rock and the specific gravity γ of the surrounding rock have a significant impact on the deformation of the surrounding rock. In order to reduce the amount of calculation and ensure the effectiveness and rationality of the training sample data provided by the tunnel excavation numerical calculation model for the surrounding rock mechanical parameter model, the above five parameters E, μ, c, γ are the parameters output by the inversion model of surrounding rock mechanical parameters. According to the geological survey data and relevant specifications, combined with the tunnel site conditions, the range of each parameter is comprehensively determined and each parameter is divided into 9 levels (Table 2). 5 ) to construct 81 sets of numerical tests, and some of the parameter values ​​are shown in Table 3. The 81 sets of values ​​constructed by the orthogonal test table are respectively sent to the tunnel excavation numerical calculation model to obtain the deformation curve of the surrounding rock under different values ​​of the mechanical parameters. The deformation amount of the surrounding rock at the end of the rapid deformation stage on the surrounding rock deformation curve (that is, the deformation amount generated after t time) is used as the feature of the training sample, and the corresponding mechanical parameter value is used as the label of the training sample to obtain multiple sets of training sample data. The multiple sets of training sample data obtained are used to train the surrounding rock mechanical parameter inversion model. For example, the training sample is [a0, a1, a2, ΔL, E, μ, c, γ], where a0 is the surrounding rock deformation detected at A0, a1 is the surrounding rock deformation detected at A1, a2 is the surrounding rock deformation detected at A2, ΔL is the deformation of line segment L, and the deformations in the above training samples are all the deformations of the surrounding rock at the end of the rapid deformation stage; E, μ, c and γ are the values ​​of the mechanical parameters corresponding to the above deformations.

[0064] Table 2 Parameter 9 horizontal division table

[0065]

[0066]

[0067] Table 3L81(9 5 ) Orthogonal test parameter combination table (part)

[0068]

[0069] In this embodiment, since there is a certain correlation between the mechanical parameters of the surrounding rock, the model used in this embodiment for inverting the mechanical parameters of the surrounding rock is the RC-XGBoost model. The RC-XGBoost model includes a Regressor Chain and XGBoost (i.e., a gradient boosting algorithm) for optimization and solution, so that the surrounding rock mechanical parameters inverted by the RC-XGBoost model are more accurate.

[0070] When training the model, the training sample data set is divided into a training set and a prediction set. The training set accounts for 80%, a total of 65 groups of training samples, which are used to train the surrounding rock mechanical parameter inversion model; the prediction set accounts for 20%, a total of 16 groups of prediction samples. After training the surrounding rock mechanical parameter inversion model with satisfactory performance, the displacement deformation [a0, a1, a2, ΔL] in the 16 groups of prediction samples in the prediction set are used as the input vectors of the RC-XGBoost model, and the output is E, μ, c. The predicted value of γ is compared with the actual value (label) in the predicted sample by the predicted value output by the inversion model of surrounding rock mechanical parameters, such as Figure 7 As shown in Figure 2, it can be seen that the predicted values ​​of each parameter are close to the actual values, E, μ, The average prediction accuracy of c and γ are 92.8%, 91.0%, 72.4%, 90.9% and 87.2% respectively, and the inversion results have high reliability. Among them, satisfactory performance refers to the average determination coefficient R 2 is the objective function. When the objective function reaches 0.8, it is considered that the model performance is satisfactory and the training is terminated. 2 Determined by the following test:

[0071]

[0072] Where: is the standard deviation of the indicator (i.e., the surrounding rock deformation output by the surrounding rock mechanical parameter inversion model after inputting the characteristics of the training samples in the training set), is the average value of the indicator (the labels of the training samples in the training set).

[0073] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An automatic inversion method for mechanical parameters of surrounding rock of a tunnel with developed joints, characterized in that: The following steps are involved: S1: obtaining the initial support parameter values ​​and joint parameter values ​​of the surrounding rock at a certain section in the tunnel, and establishing a tunnel excavation numerical calculation model based on the initial support parameter values ​​and the joint parameter values, wherein the rock mass within the loose zone of tunnel excavation adopts a joint-diffuse constitutive model, and the rock mass outside the loose zone adopts a Mohr-Coulomb constitutive model; S2: taking multiple sets of values ​​of mechanical parameters and inputting them into the tunnel excavation numerical calculation model respectively, to obtain multiple surrounding rock deformation curves, all of which are recorded as second deformation curves; obtaining the deformation amount of the surrounding rock at the end of the rapid deformation stage of the surrounding rock on each of the second deformation curves, taking the deformation amount as the feature of the training sample, taking the multiple sets of values ​​of the mechanical parameters as the labels of the training samples, and combining the features of the training samples with the labels of the training samples to obtain multiple sets of training samples; S3: using the multiple groups of training samples obtained in S2 to train the surrounding rock mechanical parameter inversion model; S4: inputting the deformation of the surrounding rock measured at the construction site at the end of the rapid deformation stage into the surrounding rock mechanical parameter inversion model trained in S3 to obtain the mechanical parameters of the surrounding rock at the construction site; In S1, the deformation of the surrounding rock of the section during the tunnel excavation process is monitored to obtain a deformation curve of the surrounding rock at the section over time, which is recorded as a first deformation curve; the point on the first deformation curve where the rapid deformation stage of the surrounding rock ends is recorded as point C′, and the time t elapsed from the start of excavation to point C′ of the section is recorded; in S2, the method for obtaining the deformation of the surrounding rock at the end of the rapid deformation stage of the surrounding rock from each of the second deformation curves is as follows: from each of the second deformation curves, the deformation of the surrounding rock at the tunnel section corresponding to t time after the tunnel section is excavated is selected, and the deformation is the deformation of the surrounding rock at the end of the rapid deformation stage of the surrounding rock.

2. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to claim 1 is characterized in that: In S4, the deformation amount of the surrounding rock measured at the construction site at the end of the rapid deformation stage is the deformation amount of the surrounding rock generated at t time after the excavation of the section.

3. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to claim 1 is characterized in that: The surrounding rock mechanical parameter inversion model is an RC-XGBoost model, and the RC-XGBoost model is a regressor chain that uses a gradient boosting algorithm for optimization and solution.

4. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to claim 1 is characterized in that: The mechanical parameters include the elastic modulus E of the surrounding rock, the Poisson's ratio μ of the surrounding rock, the internal friction angle of the surrounding rock, the cohesion c of the surrounding rock, and the density γ of the surrounding rock; the initial support parameters include the thickness of the initial support, the elastic modulus of the initial support, the Poisson's ratio of the initial support and the density of the initial support; the joint parameters include the inclination angle of the joint, the tendency of the joint, the tensile strength of the joint, the cohesion of the joint and the internal friction angle of the joint.

5. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to claim 1 is characterized in that: In S2, multiple groups of values ​​of mechanical parameters serving as training sample labels are obtained by orthogonal experimental design of the mechanical parameters.

6. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to claim 1 is characterized in that: In the tunnel excavation numerical calculation model, the rapid stress boundary method is used to simulate the initial geostress field.

7. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to claim 1 is characterized in that: The distance between the left and right boundaries of the tunnel excavation numerical calculation model is five times the diameter of the tunnel to be excavated, the upper boundary is located within 40m to 50m upward from the arch line of the tunnel to be excavated, and the lower boundary is located within 40m to 50m downward from the arch line of the tunnel to be excavated. The length of the tunnel excavation numerical calculation model along the tunnel excavation direction is 50m to 60m.

8. The method for automatic inversion of mechanical parameters of surrounding rock of a tunnel with developed joints according to any one of claims 1 to 7, characterized in that: The end of the rapid deformation stage of the surrounding rock is the moment when the slope of the deformation curve of the surrounding rock reaches 1.

Images