Tunnel surrounding rock deformation measurement device and measurement method

Abstract

The present invention discloses a tunnel surrounding rock deformation measurement device and measurement method, which belongs to the technical field of tunnel surrounding rock deformation calculation model, including: obtaining geological parameters, groundwater parameters and ground stress state of the current tunnel; inputting into a deformation discrimination model, outputting label results, and continuing when the label results indicate that the deformation attribute of the tunnel surrounding rock is suitable for prediction; obtaining tunnel geometric parameters and initial support construction parameters, and inputting into a deformation prediction model together with geological parameters, groundwater parameters and ground stress state, and outputting deformation data of the tunnel surrounding rock over time; when one or more deformation amounts of the deformation data are greater than a preset safety threshold, adjusting the construction parameters of the initial support until the safety threshold requirement is met; based on the preset surrounding rock deformation stability conditions during the initial support period, selecting the construction time of the secondary lining by comparing the deformation data. The present invention can accurately predict the deformation data of the tunnel surrounding rock over time and improve the reliability of the deformation prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel surrounding rock deformation calculation model, and more specifically, to a tunnel surrounding rock deformation measurement device and measurement method. Background Art

[0002] The deformation of the surrounding rock of a tunnel may cause cracking and spalling of the initial support and shotcrete of the tunnel, and even the twisting and deformation of the steel arch frame, thus affecting the integrity and stability of the tunnel structure. If serious surrounding rock deformation is not handled in time, it may cause engineering accidents, resulting in casualties and property and financial losses. In tunnel construction in high-stress soft rock areas, the large deformation of the surrounding rock will lead to great difficulties and challenges in the safe construction and construction management of the tunnel. The deformation of the surrounding rock may cause the instability of the tunnel face, and then cause tunnel collapse, which will seriously endanger the construction of the tunnel project. For shallow buried tunnels, the deformation of the surrounding rock may develop to the surface, causing surface deformation and cracking, and even collapse and roof fall. Dealing with the deformation of the surrounding rock requires additional time and resources, which may lead to delays in the construction period and increase in construction costs. The presence of karst water may cause water and mud inrush disasters in the tunnel. These disasters are sudden and destructive, directly threatening the personal safety of workers and may cause tunnel construction to stop. Under the long-term immersion of groundwater, the lining structure may soften, the creep amount increases, and the lining structure is damaged. Therefore, the measurement and control of tunnel surrounding rock deformation is an important task in tunnel engineering design and construction, and effective prediction, monitoring and support measures are needed to ensure construction safety and engineering quality. This paper is dedicated to studying the accurate prediction of tunnel surrounding rock deformation, so as to reduce the occurrence of engineering accidents and provide a data basis for improving support performance.

[0003] Some existing technologies have also developed methods based on deep neural networks to predict tunnel surrounding rock deformation, which can achieve very high classification accuracy on training data. However, minimizing the loss function on the training set is not equivalent to minimizing the loss function on the test set, which is prone to overfitting problems. This means that the model performs well on training data, but may perform poorly on unseen data, and the prediction reliability is seriously insufficient. Summary of the invention

[0004] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0005] In order to achieve these purposes and other advantages according to the present invention, a method for measuring deformation of surrounding rock of a tunnel is provided, comprising the following steps:

[0006] S1. Obtain the geological parameters, groundwater parameters and ground stress state of the current tunnel;

[0007] S2, inputting the data obtained in step S1 into a deformation discrimination model, and outputting a label result, wherein the deformation discrimination model is used to judge the deformation attribute of the tunnel surrounding rock, and when the output label result indicates that the deformation attribute of the tunnel surrounding rock is suitable for prediction, the execution continues; wherein the deformation discrimination model is constructed and trained based on a feedforward neural network FNN;

[0008] S3, obtaining tunnel geometry parameters and initial support construction parameters, and inputting them together with the geological parameters, groundwater parameters and ground stress state obtained in step S1 into a deformation prediction model, and outputting deformation data of the tunnel surrounding rock over time; wherein the deformation prediction model is constructed and trained based on an RBF neural network;

[0009] S4, when one or more deformation amounts in the deformation data of the tunnel surrounding rock over time obtained in step S3 is greater than a preset safety threshold, adjusting the construction parameters of the initial support, and repeating step S3 until the deformation amounts in the deformation data are all no greater than the preset safety threshold;

[0010] S5. Based on the preset surrounding rock deformation stability conditions during the initial support period, the timing for the construction of the secondary lining is selected by comparing the deformation data of the tunnel surrounding rock over time.

[0011] Preferably, when the timing of secondary lining construction selected in step S5 is greater than a preset construction period threshold, the construction parameters of the initial support are adjusted, and steps S3 and S4 are repeated until the timing of secondary lining construction is not greater than the preset construction period threshold.

[0012] Preferably, constructing and training the deformation prediction model comprises the following steps:

[0013] Establish a data set sample, which includes the collected geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, initial support construction parameters, and deformation data of tunnel surrounding rock at multiple time points from the initial support stage to before the secondary lining is constructed, and perform normalization processing;

[0014] The input layer is the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, and construction parameters of the initial support of the dataset samples;

[0015] The hidden layer contains multiple neurons, each of which uses the radial basis function as the activation function. The number and position of neurons are determined by the k-means method based on the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, and construction parameters of the initial support of the data set samples;

[0016] The output layer is the deformation data of the tunnel surrounding rock at multiple predicted time points;

[0017] Weight training: The LMS algorithm is used to update the weight of each neuron in the training hidden layer based on the output deformation data of the tunnel surrounding rock at multiple time points and the deformation data of the tunnel surrounding rock at multiple time points corresponding to the data sample set. When the error is reduced to below the preset target error, the training is completed and the deformation prediction model that has been trained is obtained.

[0018] Preferably,

[0019] The geological parameters include: uniaxial compressive strength, elastic modulus, rock quality index, joint spacing, and rock mass integrity;

[0020] The groundwater parameters include: groundwater level, water quality, and water pressure;

[0021] The in-situ stress state includes: the magnitude and direction of the initial in-situ stress, and the stress concentration area;

[0022] The tunnel geometric parameters include: the cross-sectional dimensions of the tunnel, which include the width, height and profile of the tunnel;

[0023] The construction parameters of the initial support include: support type, support timing, and support density, wherein the support type includes shotcrete, steel frame support, and anchor rod;

[0024] The deformation data of the tunnel surrounding rock at multiple time points from the initial support stage to before the secondary lining is constructed include one or more combinations of: preceding displacement, initial displacement, rapid increase zone, slow deformation zone, and stable deformation zone.

[0025] Preferably, constructing and training the deformation discrimination model comprises the following steps:

[0026] Randomly sample samples from the data set, input the deformation prediction model, output the prediction results, and calculate the corresponding loss function MSE;

[0027] Obtain a data set other than the data set sample, in which the data difference between the geological parameters, groundwater parameters and ground stress state and the corresponding parameters in the data set sample is greater than a certain value, and obtain the corresponding tunnel geometry parameters and initial support construction parameters, and then input them into the deformation prediction model, output the prediction results, and calculate the corresponding loss function MSE;

[0028] Set the preferred value and critical value of MSE, where the critical value is greater than the preferred value. When MSE is not greater than the preferred value, it is marked as Class I. When MSE is between the preferred value and the critical value, it is marked as Class II. When MSE is greater than the critical value, it is marked as Class III.

[0029] The normalized data of geological parameters, groundwater parameters, and ground stress state are used as input data, and the MSE category is used as the label to form training samples;

[0030] Construct a feedforward neural network FNN:

[0031] The input layer is the normalized geological parameters, groundwater parameters, and ground stress state;

[0032] The hidden layer is set to 1 layer, the number of neurons in the hidden layer is the same as the number of parameters in the input layer, and the ReLU function is used as the activation function;

[0033] The number of neurons in the output layer is set to 3, and a fully connected layer is used to connect the hidden layer, and the softmax function is used as the activation function to output the label result;

[0034] A reflection propagation algorithm is used to train a feedforward neural network, wherein a cross entropy loss function is used to calculate the correctness of the output label, and then the gradient is calculated and the weight is updated; when the correctness reaches the target value, the training is completed, and the trained deformation discrimination model is obtained.

[0035] A measuring device for executing the tunnel surrounding rock deformation measuring method is provided, comprising:

[0036] Data acquisition module, which is used to obtain the geological parameters, groundwater parameters and ground stress state of the current tunnel;

[0037] A deformation attribute discrimination module, which is used to input the geological parameters, groundwater parameters and ground stress state of the current tunnel into a deformation discrimination model, and output a label result, wherein the deformation discrimination model is used to judge the deformation attribute of the tunnel surrounding rock, and when the output label result indicates that the deformation attribute of the tunnel surrounding rock is suitable for prediction, the execution continues; wherein the deformation discrimination model is constructed and trained based on a feedforward neural network FNN;

[0038] The deformation measurement module is used to obtain the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters and initial support construction parameters of the current tunnel, and input them into the deformation prediction model, and output the deformation data of the tunnel surrounding rock over time; wherein the deformation prediction model is constructed and trained based on the RBF neural network;

[0039] The deformation correction module is used to adjust the construction parameters of the initial support and re-predict when one or more deformation amounts in the output deformation data of the tunnel surrounding rock over time are greater than the preset safety threshold value until the deformation amounts in the deformation data are no greater than the preset safety threshold value.

[0040] The present invention has at least the following beneficial effects:

[0041] First, by introducing data sets other than data set samples with large data differences, prediction is performed based on the deformation prediction model to obtain a data set whose prediction results are greatly different from the actual measured data (the loss function MSE is large). Then, a deformation discrimination model is constructed and trained to screen out which tunnel projects can be adapted to the deformation prediction model of this solution, thereby improving the prediction reliability of the deformation prediction model.

[0042] Second, by collecting geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters and other parameters, and combining with RBF neural network, a surrounding rock time-history deformation prediction model with strong nonlinear relationship can be automatically constructed, which is very beneficial for dealing with complex geological problems. Combined with the radial basis function to respond locally to input data, this local approximation capability enables the deformation prediction model to simulate the local characteristics of surrounding rock deformation more finely, improve the accuracy of prediction, and provide scientific decision-making support for tunnel construction. And the output layer of this solution is a linear weighting of the hidden layer, avoiding lengthy calculations like back propagation, and thus has a higher computing speed.

[0043] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The figure is a flow chart of the measuring method according to one of the technical solutions of the present invention. DETAILED DESCRIPTION

[0045] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0046] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the orientation or positional relationship indicated by the terms is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0047] like Figure 1 As shown, the present invention provides a method for measuring deformation of surrounding rock of a tunnel, comprising the following steps:

[0048] S1. Obtain the geological parameters, groundwater parameters and geostress state of the current tunnel; geological parameters include, for example, the uniaxial compressive strength and elastic modulus of rock, groundwater parameters include groundwater level and water pressure, and geostress state includes the magnitude and direction of initial geostress; geological parameters can be obtained through geological surveys, laboratory tests and field tests. Groundwater parameters, such as water level, can be measured by a water level meter to record the depth or height of the groundwater level. Geostress state can be obtained by geostress measuring equipment.

[0049] S2. Input the data obtained in step S1 into the deformation discrimination model, and output a label result, wherein the deformation discrimination model is used to judge the deformation properties of the tunnel surrounding rock. When the output label result indicates that the deformation properties of the tunnel surrounding rock are suitable for prediction, the execution continues; wherein the deformation discrimination model is constructed and trained based on the feedforward neural network FNN; there are different levels of surrounding rock, such as level V surrounding rock, level IV surrounding rock, level III surrounding rock, etc., but it is not accurate to directly judge whether the deformation properties of the surrounding rock deformation of the current tunnel are suitable for the deformation prediction model trained by the present invention based on the existing surrounding rock level, because the situation of the tunnel surrounding rock itself is particularly complex, and the situations presented during the tunnel construction process are also various, and the training samples used for the deformation prediction model are also limited. Therefore, by designing a deformation discrimination model for the deformation prediction model of the present invention, the reliability of the deformation prediction model prediction can be significantly improved.

[0050] S3, obtain the tunnel geometry parameters and the construction parameters of the initial support, and input them into the deformation prediction model together with the geological parameters, groundwater parameters and ground stress state obtained in step S1, and output the deformation data of the tunnel surrounding rock over time; wherein, the deformation prediction model is constructed and trained based on the RBF neural network; for example, the tunnel geometry parameters include the cross-sectional dimensions of the tunnel, i.e., the width and height of the tunnel, etc., and the construction parameters of the initial support include the support type and support density, etc., which together with the geological parameters, groundwater parameters and ground stress state form the basis of the prediction data, and accurately predict the deformation data of the tunnel surrounding rock over time, so that before the initial support is applied, the surrounding rock deformation that will occur after the tunnel excavation can be known, so as to evaluate in advance whether the construction parameters of the designed initial support are perfect and whether they need to be adjusted. That is, the construction parameters of the initial support can be accurately judged through the prediction of the deformation of the tunnel surrounding rock, so as to adjust them in time, and complete the initial support and secondary lining within the specified construction period, and ensure safety. The tunnel geometry parameters can be obtained through the tunnel design drawings and on-site measurements. The construction parameters of the initial support can be obtained through construction records and on-site monitoring. The surrounding rock deformation data during RBF neural network training can be obtained in real time through monitoring equipment installed in the tunnel.

[0051] S4. When one or more deformations in the deformation data of the tunnel surrounding rock over time obtained in step S3 are greater than the preset safety threshold, adjust the construction parameters of the initial support, and repeat step S3 until the deformations in the deformation data are no greater than the preset safety threshold; when the deformation of the tunnel surrounding rock reaches a certain level, it indicates that the potential safety hazard is relatively large, so from a safety perspective, it is necessary to adjust the construction parameters of the initial support to ensure safety. The construction parameters of the initial support can be adjusted, for example, the thickness of the shotcrete, the spacing of the steel mesh, the spacing of the steel arch frame, the length and spacing of the anchor rods, etc.

[0052] S5. Based on the preset surrounding rock deformation stability conditions during the initial support period, the timing of secondary lining construction is selected by comparing the deformation data of the tunnel surrounding rock over time. Under normal circumstances, secondary lining construction can be carried out when the surrounding rock deformation is in a stable state. For example, when the peripheral displacement rate is less than 0.1~0.2 mm / d, or the arch sinking speed is less than 0.07~0.15 mm / d, it can be considered that the surrounding rock deformation has stabilized, that is, secondary lining construction can be carried out.

[0053] In the above technical solution, firstly, the geological parameters, groundwater parameters, ground stress state and deformation discrimination model of the current tunnel are used to determine whether the deformation of the surrounding rock of the current tunnel can be predicted, so as to ensure the usefulness of the prediction data of the deformation prediction model. Then, the deformation data of the surrounding rock of the tunnel over time is obtained by predicting the geological parameters, groundwater parameters, ground stress state, tunnel geometric parameters, construction parameters of the initial support and deformation prediction model of the current tunnel, and then the construction parameters of the initial support can be adjusted based on this to determine the safety.

[0054] In another technical solution, when the timing of secondary lining construction selected in step S5 is greater than a preset construction period threshold, the construction parameters of the initial support are adjusted, and steps S3 and S4 are repeated until the timing of secondary lining construction is not greater than the preset construction period threshold.

[0055] In the above technical solution, according to the aforementioned selected construction timing of the secondary lining, when compared with the prescribed construction period, the construction parameters of the initial support can be adjusted in the opposite direction, so as to increase the probability of completing the initial support and the secondary lining within the prescribed construction period while ensuring safety. Here, the construction parameters of the initial support can, for example, adjust the construction timing of the initial support.

[0056] In another technical solution, building and training the deformation prediction model includes the following steps:

[0057] A data set sample is established, which includes the collected geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, construction parameters of initial support, and deformation data of tunnel surrounding rock at multiple time points from the initial support stage to the secondary lining construction, and normalized. Since there are many parameters in the data set sample, normalization is used to ensure the consistency of scales of different features, which is helpful for the training and convergence of the deformation prediction model. The data is scaled to a small, specified range by normalization, which is [0, 1] in this solution. For some graded data, such as the integrity of the rock mass, it is simplified to 0 or 1, 0 for incomplete and 1 for complete.

[0058] The input layer is the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, and construction parameters of the initial support of the dataset samples;

[0059] The hidden layer contains multiple neurons, each of which uses the radial basis function as the activation function. The number and position of neurons are determined by the k-means method based on the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, and construction parameters of the initial support of the data set samples. The k-means method can be used to determine the center (i.e., the center point of the radial basis function) and width (i.e., the standard deviation of the Gaussian function) of the hidden layer neurons.

[0060] The output layer is the deformation data of the tunnel surrounding rock at multiple predicted time points;

[0061] Weight training: The LMS algorithm is used to update the weight of each neuron in the training hidden layer based on the deformation data of the tunnel surrounding rock at multiple time points output and the deformation data of the tunnel surrounding rock at multiple time points corresponding to the data sample set. When the error is reduced to below the preset target error, the training is completed, and the deformation prediction model that has been trained is obtained. In the RBF neural network, the weight is the connection strength between the hidden layer neurons and the output layer neurons. Each neuron corresponds to a radial basis function, which exists in the form of a Gaussian function. The output of the hidden layer neurons, that is, the value of the radial basis function, will be passed to the output layer. The weight connecting the hidden layer neurons and the output layer neurons determines how these outputs are combined to produce the final output. This solution uses the LMS algorithm to adjust the weights and biases to minimize the error between the RBF neural network output and the target output.

[0062] In the above technical solution, by collecting geological parameters, groundwater parameters, ground stress state, tunnel geometric parameters and other parameters, and combining with RBF neural network, a surrounding rock time-history deformation prediction model with strong nonlinear relationship can be automatically constructed, which is very beneficial for dealing with complex geological problems. Combined with the radial basis function to respond locally to input data, this local approximation capability enables the deformation prediction model to simulate the local characteristics of surrounding rock deformation more finely, improve the accuracy of prediction, and provide scientific decision-making support for tunnel construction. And the output layer of this solution is a linear weighting of the hidden layer, avoiding lengthy calculations like back propagation, and thus has a higher computing speed.

[0063] In another technical solution, the geological parameters include: uniaxial compressive strength, elastic modulus, rock quality index (RQD), joint spacing, and rock mass integrity; these parameters affect the stability of the surrounding rock and support design. For example, soft rocks with low uniaxial compressive strength and elastic modulus, such as argillaceous sandstone and sandy mudstone, are prone to deformation and damage and require special support measures.

[0064] The groundwater parameters include: groundwater level, water quality, and water pressure; the presence of groundwater will reduce the stability of the rock mass and increase the difficulty of tunnel construction. In fractured rock mass with groundwater, it is necessary to prevent large seepage pressure to reduce the impact of water on the surrounding rock.

[0065] The in-situ stress state includes: the magnitude and direction of the initial in-situ stress, the stress concentration area; under high in-situ stress conditions, the surrounding rock is prone to rock bursts, posing a threat to tunnel construction. The correspondence between the quantitative value and the qualitative description of the in-situ stress state, for example, under extremely high in-situ stress conditions, rock bursts occur during the excavation of hard rock, and the displacement of soft rock is extremely significant.

[0066] The tunnel geometric parameters include: the cross-sectional dimensions of the tunnel, which include the width, height and contour shape of the tunnel; the cross-sectional dimensions of the tunnel reflect the excavation size and the degree of disturbance to the surrounding rock, and are therefore an important influencing factor.

[0067] The construction parameters of the initial support include: support type, support timing, and support density, among which the support types include shotcrete, steel frame support, and anchor rods; the construction parameters of the initial support are crucial to controlling the deformation and loosening of the surrounding rock. The initial support should closely follow the excavation surface, and the shotcrete operation must be completed at least 4 hours before the next blasting to ensure the support effect. The support density includes, for example, the thickness of shotcrete. Tunnels with different surrounding rock levels have different requirements for the thickness of shotcrete. For example, the V-level surrounding rock section takes 130~150 mm, the IV-level surrounding rock section takes 120~130 mm, and the III-level surrounding rock section takes 80~120 mm.

[0068] The deformation data of the tunnel surrounding rock at multiple time points from the initial support stage to the construction of the secondary lining include one or more combinations of the preceding displacement, initial displacement, rapid increase zone, slow deformation zone, and stable deformation zone. Specifically, after the initial support construction and construction is completed, the displacement deformation law of the tunnel surrounding rock over time is usually manifested in the following stages:

[0069] Pre-displacement: occurs within a certain range in front of the tunnel face, mainly due to the stress release of the surrounding rock in front and the consolidation of the surrounding rock due to water loss caused by the excavation of the working face;

[0070] Initial displacement: occurs at the tunnel face, is closely related to geological conditions, and occurs immediately after excavation;

[0071] Rapid increase area: In the area 1 to 3D behind the tunnel face, the deformation rate of the surrounding rock increases rapidly and the deformation increases sharply, where D represents the equivalent diameter of the tunnel;

[0072] Slow deformation zone: In the area 3 to 5D behind the tunnel face, the deformation increases slowly and the deformation curve begins to converge;

[0073] Stable deformation zone: In the area beyond 5D behind the excavation surface, the deformation of the surrounding rock increases slowly and the surrounding rock tends to be stable.

[0074] In the above technical solution, when predicting the deformation of tunnel surrounding rock, it is precisely because the types and amounts of data that need to be collected are large, therefore, not all machine learning methods and models can handle it, which can easily lead to problems such as lengthy calculations, insufficient computing power, and overfitting.

[0075] In another technical solution, constructing and training the deformation discrimination model includes the following steps:

[0076] Randomly sample from the data set sample, input the deformation prediction model, output the prediction result, and calculate the corresponding loss function MSE; this type of sample that can be accurately predicted represents the scope that the deformation prediction model of this application can adapt to, because the tunnel surrounding rock itself is particularly complex, the situation presented during the tunnel construction process is also diverse, and the training samples for the deformation prediction model are also limited. Moreover, it is not based on the level of the surrounding rock in experience that can be simply judged, but also needs to rely on other data such as groundwater parameters and ground stress state.

[0077] A data set outside the data set sample is obtained, wherein the data difference between the geological parameters, groundwater parameters and ground stress state and the corresponding parameters in the data set sample is greater than a certain value, and the corresponding tunnel geometry parameters and initial support construction parameters are obtained, and then input into the deformation prediction model, the prediction results are output, and the corresponding loss function MSE is calculated; the data set outside the data set sample can be considered when the number of parameters reaches 70% or more of the total number of parameters and the corresponding numerical difference reaches 50% or more.

[0078] Set the preferred value and critical value of MSE, where the critical value is greater than the preferred value. When the MSE is not greater than the preferred value, it is marked as Class I. When the MSE is between the preferred value and the critical value, it is marked as Class II. When the MSE is greater than the critical value, it is marked as Class III. The selection of the preferred value affects the scope of application of the deformation prediction model. The larger the preferred value, the wider the allowable accuracy. The preferred value can be selected as 0.05%. When it is between the preferred value and the critical value, it indicates that the deformation prediction model of this application can be used for prediction, but the prediction result is only for reference by construction personnel. The critical value can be selected as 2%~5%. When it is higher than the critical value, it means that even if the deformation prediction model is used for prediction, its prediction result is not for reference. Based on the above method, the prediction reliability of the deformation prediction model of this application can be significantly improved.

[0079] The normalized data of geological parameters, groundwater parameters, and ground stress state are used as input data, and the MSE category is used as the label to form training samples;

[0080] Construct a feedforward neural network FNN:

[0081] The input layer is the normalized geological parameters, groundwater parameters, and ground stress state;

[0082] The hidden layer is set to 1 layer, the number of neurons in the hidden layer is the same as the number of parameters in the input layer, and the ReLU function is used as the activation function; although there are many types of input data, since only the correlation with one parameter MSE needs to be judged, the hidden layer can be set to 1 layer, but the number of neurons is set to the same as the number of parameters in the input layer, which can reduce computational redundancy and will not affect accuracy. The ReLU activation function is used, which can speed up the training process, reduce the gradient disappearance problem, and is simple to calculate.

[0083] The number of neurons in the output layer is set to 3, and a fully connected layer is used to connect the hidden layer, and the softmax function is used as the activation function to output the label result; because this scheme is divided into three categories, namely Class I (accurate prediction), Class II (prediction for reference), and Class III (prediction without reference significance), therefore, the number of neurons in the output layer only needs to be set to 3, and the fully connected layer is used to solve the problem of dimensional change from the hidden layer to the output layer. The Softmax function can be used to convert a K-dimensional vector (3-dimensional in this scheme) into a probability distribution, where the value of each element is between 0 and 1, and the sum of all elements is 1, which can easily explain the prediction probability of each category.

[0084] The reflection propagation algorithm is used to train the feedforward neural network, in which the cross entropy loss function is used to calculate the correctness of the output label, and then the gradient is calculated and the weight is updated; when the correctness reaches the target value, the training is completed, and the deformation discrimination model that has been trained is obtained. On the basis of the Softmax function, combined with the cross entropy loss function, the difference between the predicted probability distribution and the probability distribution of the true annotation label can be measured very intuitively, and its gradient calculation is relatively simple, so that the parameters of the deformation discrimination model can be efficiently updated during the reflection propagation process.

[0085] In the above technical scheme, by introducing data sets other than data set samples with large data differences, prediction is performed based on the deformation prediction model to obtain a data set with a large difference between the prediction result and the actual measured data (large loss function MSE), and then constructing and training a deformation discrimination model to screen out which tunnel projects can be adapted to the deformation prediction model of this scheme, that is, when the result output by the deformation discrimination model is Class I, it means that the deformation data of the tunnel surrounding rock of the current tunnel project over time can be accurately predicted; when it is Class II, it means that the predicted value of the deformation data of the tunnel surrounding rock of the current tunnel project over time can be used for reference; when it is Class III, it means that the deformation data of the tunnel surrounding rock of the previous tunnel project over time is not suitable for reference, and there is no need to use the deformation prediction model for prediction.

[0086] Based on the same inventive concept, the present invention also provides a tunnel surrounding rock deformation measurement device, which is used to perform the tunnel surrounding rock deformation measurement method, comprising:

[0087] Data acquisition module, which is used to obtain the geological parameters, groundwater parameters and ground stress state of the current tunnel; geological parameters include, for example, the uniaxial compressive strength and elastic modulus of rock, groundwater parameters include groundwater level and water pressure, and ground stress state includes the magnitude and direction of initial ground stress;

[0088] A deformation attribute discrimination module is used to input the geological parameters, groundwater parameters and ground stress state of the current tunnel into a deformation discrimination model, and output a label result, wherein the deformation discrimination model is used to judge the deformation attributes of the tunnel surrounding rock, and when the output label result indicates that the deformation attributes of the tunnel surrounding rock are suitable for prediction, the execution continues; wherein the deformation discrimination model is constructed and trained based on a feedforward neural network FNN; and is used to judge whether the deformation attributes of the surrounding rock deformation of the current tunnel are suitable for the deformation prediction model trained by the present invention.

[0089] The deformation measurement module is used to obtain the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters and initial support construction parameters of the current tunnel, and input them into the deformation prediction model, and output the deformation data of the tunnel surrounding rock over time; wherein, the deformation prediction model is constructed and trained based on the RBF neural network; the tunnel geometry parameters and initial support construction parameters together with the geological parameters, groundwater parameters and ground stress state form the basis of the prediction data, and the deformation data of the tunnel surrounding rock over time is accurately predicted, so that the surrounding rock deformation that will occur after tunnel excavation can be known before the initial support is applied

[0090] The deformation correction module is used to adjust the construction parameters of the initial support and re-predict when one or more deformations in the output deformation data of the tunnel surrounding rock over time are greater than the preset safety threshold, until the deformations in the deformation data are no greater than the preset safety threshold. When the deformation of the tunnel surrounding rock reaches a certain level, it indicates that there is a great safety hazard, so from a safety perspective, it is necessary to adjust the construction parameters of the initial support to ensure safety.

[0091] In the above technical solution, firstly, the geological parameters, groundwater parameters, ground stress state and deformation discrimination model of the current tunnel are used to determine whether the deformation of the surrounding rock of the current tunnel can be predicted, so as to ensure the usefulness of the prediction data of the deformation prediction model. Then, the deformation data of the surrounding rock of the tunnel over time is obtained by predicting the geological parameters, groundwater parameters, ground stress state, tunnel geometric parameters, construction parameters of the initial support and deformation prediction model of the current tunnel, and then the construction parameters of the initial support can be adjusted based on this to determine the safety.

[0092] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present invention may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0093] The present invention also provides an electronic device, comprising:

[0094] one or more processors;

[0095] and a memory connected to the processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the tunnel surrounding rock deformation measurement method. The electronic device can be any terminal device including a mobile phone, a laptop computer, a desktop computer, a tablet computer, a PDA (Personal Digital Assistant), a POS (Point of Sales), a vehicle-mounted computer, etc.

[0096] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the above-mentioned tunnel surrounding rock deformation measurement method is implemented.

[0097] Through the description of the above implementation mode, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present invention, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, RandomAccess Memory), a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0098] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for measuring deformation of surrounding rock of a tunnel, characterized in that: The following steps are involved: S1. Obtain the geological parameters, groundwater parameters and ground stress state of the current tunnel; S2, inputting the data obtained in step S1 into a deformation discrimination model, and outputting a label result, wherein the deformation discrimination model is used to judge the deformation attribute of the tunnel surrounding rock, and when the output label result indicates that the deformation attribute of the tunnel surrounding rock is suitable for prediction, the execution continues; wherein the deformation discrimination model is constructed and trained based on a feedforward neural network FNN; S3, obtaining tunnel geometry parameters and initial support construction parameters, and inputting them together with the geological parameters, groundwater parameters and ground stress state obtained in step S1 into a deformation prediction model, and outputting deformation data of the tunnel surrounding rock over time; wherein the deformation prediction model is constructed and trained based on an RBF neural network; S4, when one or more deformation amounts in the deformation data of the tunnel surrounding rock over time obtained in step S3 is greater than a preset safety threshold, adjusting the construction parameters of the initial support, and repeating step S3 until the deformation amounts in the deformation data are all no greater than the preset safety threshold; S5. Based on the preset surrounding rock deformation stability conditions during the initial support period, the timing of secondary lining construction is selected by comparing the deformation data of the tunnel surrounding rock over time; When the construction timing of the secondary lining selected in step S5 is greater than the preset construction period threshold, the construction parameters of the initial support are adjusted, and steps S3 and S4 are repeated until the construction timing of the secondary lining is not greater than the preset construction period threshold; Building and training the deformation prediction model includes the following steps: Establish a data set sample, which includes the collected geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, initial support construction parameters, and deformation data of tunnel surrounding rock at multiple time points from the initial support stage to before the secondary lining is constructed, and perform normalization processing; The input layer is the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, and construction parameters of the initial support of the dataset sample; The hidden layer contains multiple neurons, each of which uses the radial basis function as the activation function. The number and position of neurons are determined by the k-means method based on the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters, and construction parameters of the initial support of the data set samples; The output layer is the deformation data of the tunnel surrounding rock at multiple predicted time points; Weight training: The LMS algorithm is used to update the weight of each neuron in the training hidden layer based on the output deformation data of the tunnel surrounding rock at multiple time points and the deformation data of the tunnel surrounding rock at multiple time points corresponding to the data sample set. When the error is reduced to below the preset target error, the training is completed and the deformation prediction model that has been trained is obtained.

2. The tunnel surrounding rock deformation measurement method according to claim 1, characterized in that: The geological parameters include: uniaxial compressive strength, elastic modulus, rock quality index, joint spacing, and rock mass integrity; The groundwater parameters include: groundwater level, water quality, and water pressure; The in-situ stress state includes: the magnitude and direction of the initial in-situ stress, and the stress concentration area; The tunnel geometric parameters include: the cross-sectional dimensions of the tunnel, which include the width, height and profile of the tunnel; The construction parameters of the initial support include: support type, support timing, and support density, wherein the support type includes shotcrete, steel frame support, and anchor rod; The deformation data of the tunnel surrounding rock at multiple time points from the initial support stage to before the secondary lining is constructed include one or more combinations of: preceding displacement, initial displacement, rapid increase zone, slow deformation zone, and stable deformation zone.

3. The tunnel surrounding rock deformation measurement method according to claim 1, characterized in that: Building and training the deformation discrimination model includes the following steps: Randomly sample samples from the data set, input the deformation prediction model, output the prediction results, and calculate the corresponding loss function MSE; Obtain a data set other than the data set sample, in which the data difference between the geological parameters, groundwater parameters and ground stress state and the corresponding parameters in the data set sample is greater than a certain value, and obtain the corresponding tunnel geometry parameters and initial support construction parameters, and then input them into the deformation prediction model, output the prediction results, and calculate the corresponding loss function MSE; Set the preferred value and critical value of MSE, where the critical value is greater than the preferred value. When MSE is not greater than the preferred value, it is marked as Class I. When MSE is between the preferred value and the critical value, it is marked as Class II. When MSE is greater than the critical value, it is marked as Class III. The normalized data of geological parameters, groundwater parameters, and ground stress state are used as input data, and the MSE category is used as the label to form training samples; Construct a feedforward neural network FNN: The input layer is the normalized geological parameters, groundwater parameters, and ground stress state; The hidden layer is set to 1 layer, the number of neurons in the hidden layer is the same as the number of parameters in the input layer, and the ReLU function is used as the activation function; The number of neurons in the output layer is set to 3, and a fully connected layer is used to connect the hidden layer, and the softmax function is used as the activation function to output the label result; A reflection propagation algorithm is used to train a feedforward neural network, wherein a cross entropy loss function is used to calculate the correctness of the output label, and then the gradient is calculated and the weight is updated; when the correctness reaches the target value, the training is completed, and the trained deformation discrimination model is obtained.

4. A device for executing the tunnel surrounding rock deformation measurement method according to any one of claims 1 to 3, characterized in that: include: Data acquisition module, which is used to obtain the geological parameters, groundwater parameters and ground stress state of the current tunnel; A deformation attribute discrimination module, which is used to input the geological parameters, groundwater parameters and ground stress state of the current tunnel into a deformation discrimination model, and output a label result, wherein the deformation discrimination model is used to judge the deformation attribute of the tunnel surrounding rock, and when the output label result indicates that the deformation attribute of the tunnel surrounding rock is suitable for prediction, the execution continues; wherein the deformation discrimination model is constructed and trained based on a feedforward neural network FNN; The deformation measurement module is used to obtain the geological parameters, groundwater parameters, ground stress state, tunnel geometry parameters and initial support construction parameters of the current tunnel, and input them into the deformation prediction model, and output the deformation data of the tunnel surrounding rock over time; wherein the deformation prediction model is constructed and trained based on the RBF neural network; The deformation correction module is used to adjust the construction parameters of the initial support and re-predict when one or more deformation amounts in the output deformation data of the tunnel surrounding rock over time are greater than the preset safety threshold value until the deformation amounts in the deformation data are no greater than the preset safety threshold value.

Images