Tunnel rock mass attribute evaluation method and system based on physical law driven network

Abstract

The present invention belongs to the field of geophysical exploration and tunnel engineering. In order to solve the problem of poor prediction effect of the existing tunnel seismic wave advance detection technology, a tunnel rock mass attribute evaluation method and system based on a physical law driven network are provided. Among them, the tunnel rock mass attribute evaluation method based on the physical law driven network includes a deep neural network based on the multi-source characteristics of the rock mass and the trained multi-source information prediction. First, the elastic modulus and Poisson's ratio are predicted, and the elastic modulus and Poisson's ratio are corrected using the empirical formula; then the multi-source characteristics of the rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the ground stress and the water content of the rock mass; finally, the multi-source characteristics of the rock mass, the ground stress and the water content of the rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the surrounding rock classification, and then artificial adjustments are made to achieve multi-angle acquisition and evaluation of the tunnel rock mass attributes, and to achieve multi-angle acquisition and evaluation of the tunnel rock mass attributes, so as to ensure the safe and efficient construction of deep buried tunnels.

Description

Technical Field

[0001] The present invention belongs to the field of geophysical exploration and tunnel engineering, and in particular relates to a tunnel rock mass property evaluation method and system based on a physical law driven network. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As tunnel construction faces the complex geological conditions of "three highs and one disturbance" of high water pressure, high ground stress, high ground temperature, and excavation disturbance, deep and large faults, strong karst and other unfavorable geology are derived under the coupling of multiple disasters, which directly aggravates the complexity and disaster-causing nature of tunnel unfavorable geological conditions, leading to frequent disasters such as sudden water and mud, landslides, etc. in tunnel engineering, seriously affecting the safe and efficient construction of tunnels. Effectively obtaining information such as wave velocity, ground stress, water content and rock mass classification in the tunnel area is a prerequisite for making safe construction decisions and ensuring efficient tunnel construction. Among them, the tunnel seismic wave advance detection method can effectively obtain rock mass parameters such as wave velocity, Poisson's ratio, elastic model and so on in front of the tunnel, which is helpful to analyze the fragmentation of the geology in front of the tunnel. However, the existing method can only obtain information such as rock mass wave velocity and degree of fragmentation, and cannot obtain tunnel rock multi-attribute information, ground stress and surrounding rock classification. The existing technology analyzes the tunnel advance detection results based on artificial experience. Although it can preliminarily judge the surrounding rock classification and other information, it is seriously affected by subjective factors and is prone to misjudgment and omission. Summary of the invention

[0004] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a tunnel rock property evaluation method and system based on a physical law driven network, which obtains multi-component seismic data by conducting tunnel seismic advance detection, and then improves the anisotropic full waveform inversion and polarization analysis method, and at the same time uses deep learning algorithms and seismic attributes-rock physical properties to finally realize the evaluation and analysis of rock properties such as longitudinal and transverse wave velocities, density, anisotropic parameters, geostress, rock water content and surrounding rock classification, which can effectively improve the prediction effect of the existing tunnel seismic wave advance detection technology and ensure the safe and efficient construction of deep buried tunnels.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A first aspect of the present invention provides a tunnel rock mass attribute evaluation method based on a physical law driven network.

[0007] A tunnel rock mass attribute evaluation method based on a physical law driven network includes:

[0008] Acquire three-component seismic data, obtain P-wave velocity, S-wave velocity, density and anisotropy parameters through anisotropic full waveform inversion and use them as multi-source characteristics of rock mass;

[0009] Based on the multi-source characteristics of rock mass and the trained multi-source information prediction deep neural network, the elastic modulus and Poisson's ratio are first predicted, and the elastic modulus and Poisson's ratio are corrected using empirical formulas; then the multi-source characteristics of rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the ground stress and rock mass water content; finally, the multi-source characteristics of rock mass, ground stress and rock mass water content, and the corrected elastic modulus and Poisson's ratio are used to predict the surrounding rock classification;

[0010] Introducing artificial experience and combining the actual situation of the tunnel to artificially adjust the ground stress and rock water content, surrounding rock classification results and the corrected elastic modulus and Poisson's ratio, so as to achieve multi-angle acquisition and evaluation of the tunnel rock properties;

[0011] Among them, the total loss function for training the multi-source information prediction deep neural network is the weighted sum of the elastic modulus and Poisson's ratio prediction loss function, the ground stress and rock water content loss function, and the surrounding rock classification loss function; the elastic modulus and Poisson's ratio prediction loss function is the weighted sum of the prediction loss sub-function, the experience correction loss sub-function, and the rock sample field test loss sub-function; the ground stress and rock water content loss function and the surrounding rock classification loss function are both characterized by the cross entropy loss function.

[0012] As an implementation method, the expression of the elastic modulus and Poisson's ratio prediction loss function is:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] ;

[0018] in, is the elastic modulus and Poisson’s ratio prediction loss function, Predict loss subfunctions for elastic modulus and Poisson's ratio; Loss sub-function for empirical correction of elastic modulus and Poisson's ratio; Loss sub-function for field testing of rock samples for elastic modulus and Poisson's ratio; 、 They are actually elastic modulus and Poisson’s ratio respectively; 、 are the predicted elastic modulus and Poisson’s ratio, respectively; 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; To obtain the parameter correlation function based on the field rock sample measurement; are longitudinal wave velocity, shear wave velocity, and density respectively; is the anisotropy parameter; and is the weight coefficient; is the L2 norm; All are constant coefficients, obtained through on-site rock sample test analysis and regression.

[0019] As an implementation method, based on the rock property relationship and a pre-constructed tunnel rock quality database, the numerical relationship expressions of longitudinal wave velocity, transverse wave velocity, density, anisotropy parameters, elastic modulus and Poisson's ratio under different rock types and different geological conditions are analyzed to determine an empirical formula.

[0020] As an implementation method, the empirical formula is expressed as:

[0021]

[0022] in, 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; are longitudinal wave velocity, transverse wave velocity, and density respectively.

[0023] As an implementation mode, the multi-source information prediction deep neural network includes a multi-source information preprocessing module, an elastic modulus and Poisson's ratio prediction module, a ground stress and rock mass water content prediction module and a surrounding rock classification prediction module;

[0024] The multi-source information preprocessing module is composed of a multi-layer convolution block and a self-attention mechanism module, which is used to extract features and correlate the data of longitudinal wave velocity, shear wave velocity, density and anisotropy parameters to obtain multi-source features;

[0025] The elastic modulus and Poisson's ratio prediction module is composed of a multi-layer convolution block and a constraint layer; the convolution block is used to further analyze the multi-source features and predict the elastic modulus and Poisson's ratio; the constraint layer uses an empirical formula to correct the elastic modulus and Poisson's ratio;

[0026] The geostress and rock water content prediction module is composed of a self-attention mechanism module and a multi-layer convolution block, wherein the convolution block is used to extract the corrected elastic modulus and Poisson's ratio features, and then the self-attention mechanism module is used to fuse and weighted correlate the multi-source features and the extracted elastic modulus and Poisson's ratio features to predict geostress and rock water content;

[0027] The self-attention mechanism module is composed of a fully connected network layer, wherein the self-attention mechanism module is used to fuse and weightedly correlate multi-source features, ground stress and rock water content, and modified elastic modulus and Poisson's ratio features, and finally the fully connected network layer predicts the one-dimensional surrounding rock classification result along the front of the tunnel.

[0028] A second aspect of the present invention provides a tunnel rock mass attribute evaluation system based on a physical law driven network.

[0029] A tunnel rock mass attribute evaluation system based on a physical law driven network, comprising:

[0030] The rock mass multi-source feature acquisition module is used to acquire three-component seismic data, obtain the P-wave velocity, S-wave velocity, density and anisotropy parameters through anisotropic full waveform inversion and use them as the rock mass multi-source features;

[0031] The rock mass attribute prediction module is used to predict the deep neural network based on the multi-source characteristics of the rock mass and the trained multi-source information. First, the elastic modulus and Poisson's ratio are predicted, and the elastic modulus and Poisson's ratio are corrected using the empirical formula; then the multi-source characteristics of the rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the ground stress and the water content of the rock mass; finally, the multi-source characteristics of the rock mass, the ground stress and the water content of the rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the surrounding rock classification;

[0032] The rock mass property manual optimization module is used to introduce manual experience and adjust the ground stress, rock mass water content, surrounding rock classification results, and the corrected elastic modulus and Poisson's ratio in combination with the actual situation of the tunnel, so as to achieve multi-angle acquisition and evaluation of the tunnel rock mass properties;

[0033] Among them, the total loss function for training the multi-source information prediction deep neural network is the weighted sum of the elastic modulus and Poisson's ratio prediction loss function, the ground stress and rock water content loss function, and the surrounding rock classification loss function; the elastic modulus and Poisson's ratio prediction loss function is the weighted sum of the prediction loss sub-function, the experience correction loss sub-function, and the rock sample field test loss sub-function; the ground stress and rock water content loss function and the surrounding rock classification loss function are both characterized by the cross entropy loss function.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) The present invention utilizes a multi-task and multi-stage tunnel rock multi-source information prediction deep neural network, and predicts the elastic modulus and Poisson's ratio, geostress and water content, and surrounding rock classification in three stages, thereby achieving a preliminary prediction and evaluation of rock mass properties and accurate acquisition of information such as elastic modulus, Poisson's ratio, geostress, water content, and surrounding rock classification.

[0036] (2) The present invention introduces physical laws to drive the improvement of rock mass property prediction results. By analyzing the relationship between various geological properties and rock physical properties under different rock mass types and shapes, based on the rock property relationship and the pre-constructed tunnel rock mass quality database, the numerical relationship expressions of longitudinal wave velocity, transverse wave velocity, density, anisotropy parameters and elastic modulus and Poisson's ratio under different rock types and different geological conditions are analyzed, and the empirical formula is determined. Then, the elastic modulus and Poisson's ratio are corrected, thereby improving the prediction accuracy of tunnel rock mass properties. Finally, the results are optimized and improved by combining artificial experience and excavation exposure, realizing the tunnel rock mass property evaluation driven by both physical laws and data mining, which provides an important reference for tunnel safety construction.

[0037] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0039] Figure 1 is a flow chart of a tunnel rock mass attribute evaluation method based on a physical law driven network according to an embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of an anisotropic full waveform inversion method based on network reparameterization according to an embodiment of the present invention;

[0041] Figure 3 2. It is a schematic diagram of a multi-task and multi-stage tunnel rock mass multi-source information prediction deep neural network architecture according to an embodiment of the present invention;

[0042] Figure 4 Schematic diagram of the structure of a tunnel rock mass attribute evaluation system based on a physical law driven network according to an embodiment of the present invention;

[0043] FIG5( a ) is a longitudinal wave velocity diagram obtained by inversion according to an embodiment of the present invention;

[0044] FIG5( b ) is a shear wave velocity diagram obtained by inversion according to an embodiment of the present invention;

[0045] FIG5( c ) is a density map obtained by inversion according to an embodiment of the present invention;

[0046] FIG5( d ) is a diagram of predicted Poisson's ratio according to an embodiment of the present invention;

[0047] FIG5( e ) is a diagram of predicted elastic modulus according to an embodiment of the present invention;

[0048] Figure 6 It is the predicted surrounding rock classification result of the embodiment of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0051] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0052] Combination Figure 1 The embodiment of the present invention provides a tunnel rock mass attribute evaluation method based on a physical law driven network, which includes:

[0053] Step S101: Acquire three-component seismic data, obtain P-wave velocity, S-wave velocity, density and anisotropy parameters through anisotropic full waveform inversion and use them as multi-source characteristics of rock mass;

[0054] Step S102: Based on the multi-source characteristics of the rock mass and the trained multi-source information prediction deep neural network, first predict the elastic modulus and Poisson's ratio, and use the empirical formula to correct the elastic modulus and Poisson's ratio; then use the multi-source characteristics of the rock mass and the corrected elastic modulus and Poisson's ratio to predict the ground stress and the water content of the rock mass; finally, use the multi-source characteristics of the rock mass, the ground stress and the water content of the rock mass, and the corrected elastic modulus and Poisson's ratio to predict the surrounding rock classification;

[0055] Step S103: Introduce artificial experience and combine the actual situation of the tunnel to make artificial adjustments to the ground stress and rock water content, surrounding rock classification results and the corrected elastic modulus and Poisson's ratio to achieve multi-angle acquisition and evaluation of the tunnel rock properties.

[0056] The present invention improves the high-quality processing, precise inversion and intelligent parameter prediction of seismic data, combines the on-site geological conditions of the tunnel with the relationship between geological attributes and physical attributes, and realizes the evaluation and analysis of multiple attributes of the tunnel rock mass, further improving the level of existing tunnel advance detection technology and ensuring the safe construction of key and difficult tunnel projects.

[0057] In step S101, advance detection of tunnel seismic waves is carried out in the tunnel, including observation system setting, source excitation, three-component seismic data acquisition and other processes, and preprocessing of the three-component seismic data is performed.

[0058] The observation system setup includes setting the seismic source and geophone positions on the left and right walls of the tunnel. The key is to accurately set the position and direction of the three-component geophones to collect high-quality three-component seismic data.

[0059] The source excitation can be artificial hammer excitation, pneumatic source and broadband high-energy controllable source;

[0060] The preprocessing of tunnel three-component seismic data includes picking the initial value of the direct wave signal, analyzing the longitudinal and transverse wave velocities of the surrounding rock, and removing the direct wave signal.

[0061] Based on the pre-processed three-component seismic data, the full waveform inversion method calculation process of anisotropic media is carried out to obtain information such as longitudinal wave velocity, shear wave velocity, density, anisotropy parameters, etc. in front of the tunnel;

[0062] A method based on neural network reparameterization is used to realize full waveform inversion of anisotropic media. Specifically, the method based on neural network reparameterization is realized by using a tunnel geological multi-parameter inversion network. Among them, the tunnel geological multi-parameter inversion network is trained with the early rock mass properties of the tunnel as a pre-training set, effectively introducing pre-knowledge such as parameter range and parameter relationship. The three-component seismic data of the tunnel is used as the input of the inversion network, and the multi-parameter results in front of the tunnel are preliminarily predicted. The forward calculation of the anisotropic medium is carried out, the three-component seismic data corresponding to the current inversion result is calculated, and the difference between it and the collected three-component data is calculated. The error and gradient are calculated to update the inversion result and the anisotropic medium parameters.

[0063] like Figure 2 As shown, the tunnel geological multi-parameter inversion network includes two multi-component data feature extraction layers and a multi-parameter prediction decoding layer;

[0064] Two multi-component seismic data feature extraction layers use the three-component seismic data collected on the left and right walls of the tunnel as input for the three channel information, and analyze the relationship between the data itself and the data between different components;

[0065] The multi-parameter prediction decoding layer includes multiple convolutional neural network blocks, each of which includes a convolutional layer, a regularization layer, and an activation function layer. By increasing the number of channels and the convolution step size, the extracted three-channel seismic data characteristics are compressed and expanded in size, and finally the prediction results including five channels are predicted, namely, the longitudinal wave velocity, the shear wave velocity, the density, and two anisotropic parameters. Among them, Figure 5 (a) shows the longitudinal wave velocity map obtained by inversion; Figure 5 (b) shows the shear wave velocity map obtained by inversion; Figure 5 (c) shows the density map obtained by inversion.

[0066] The tunnel anisotropic medium full waveform inversion method of the present embodiment is a frequency domain elastic wave full waveform inversion method based on transverse isotropy.

[0067] In some optional embodiments, polarization analysis calculations are performed on the three-component seismic data after prediction processing, with a focus on obtaining fast and slow shear wave information in the seismic data, including arrival time and amplitude.

[0068] The three-component seismic data polarization analysis is to separate the fast and slow shear waves of the acquired tunnel data, obtain the arrival information and amplitude energy of the fast and slow shear waves of different components, and then analyze the distribution and arrangement direction of the cracks in front of the tunnel, providing information for the subsequent ground stress direction and strength constraints.

[0069] In step S102, a multi-source information prediction deep neural network is trained based on a pre-built tunnel rock quality database.

[0070] First, the preliminary survey data within the seismic detection range in front of the tunnel is collected and analyzed, such as geological information, surrounding rock quality and grade, etc., and then the rock type and quality, surrounding rock classification, stress, elastic modulus and other rock parameter information near the current construction tunnel mileage are analyzed. As the tunnel construction continues to advance, the entire process is tracked and dynamically updated to form a tunnel rock quality database for tunnel seismic advance detection.

[0071] It should be noted here that the constructed database includes but is not limited to:

[0072] Preliminary survey data: lithology, rock strength, surrounding rock grade, whether it is in a fault zone, whether there is groundwater, well logging information, etc.;

[0073] Tunnel face excavation reveals: face status, rough excavation face status, lithology, rock strength, surrounding rock grade, karst development degree, number of joint development groups and crack width, etc.;

[0074] The tunnel reveals the rock type and quality, surrounding rock classification, stress, elastic modulus and other rock parameter information of the surrounding rock.

[0075] In this embodiment, if Figure 3As shown, the multi-source information prediction deep neural network includes a multi-source information preprocessing module, an elastic modulus and Poisson's ratio prediction module, a ground stress and rock mass water content prediction module and a surrounding rock classification prediction module;

[0076] The multi-source information preprocessing module is composed of a multi-layer convolution block and a self-attention mechanism module, which is used to extract features and correlate the data of longitudinal wave velocity, shear wave velocity, density and anisotropy parameters to obtain multi-source features;

[0077] The elastic modulus and Poisson's ratio prediction module is composed of multiple layers of convolution blocks and constraint layers; the convolution blocks are used to further analyze multi-source features and predict the elastic modulus and Poisson's ratio; the constraint layers use empirical formulas to correct the elastic modulus and Poisson's ratio; as shown in Figures 5 (d) and 5 (e).

[0078] The geostress and rock water content prediction module is composed of a self-attention mechanism module and a multi-layer convolution block, wherein the convolution block is used to extract the corrected elastic modulus and Poisson's ratio features, and then the self-attention mechanism module is used to fuse and weighted correlate the multi-source features and the extracted elastic modulus and Poisson's ratio features to predict geostress and rock water content; the weighted coefficient is obtained through network training;

[0079] The self-attention mechanism module and the fully connected network layer are composed of the self-attention mechanism module, wherein the self-attention mechanism module is used to fuse and weightedly correlate the multi-source features, ground stress and rock mass water content, and the modified elastic modulus and Poisson's ratio features, and finally the fully connected network layer predicts the one-dimensional surrounding rock classification results along the front of the tunnel, such as Figure 6 As shown in Figure 2. The weighted coefficients are obtained through network training.

[0080] The multi-source information preprocessing module is used to extract features and correlate data such as longitudinal wave velocity, shear wave velocity, density and anisotropy parameters, which can be expressed as:

[0081] ,in It is a multi-source information preprocessing module, which consists of a multi-layer fully connected network. It is the output of the multi-source information preprocessing module;

[0082] The single-layer structure can be further expressed as:

[0083]

[0084]

[0085] in, Output results for single-layer structure functions; is a single-layer structure function; For the nLayer network 1 node data; For the n Layer 2 network node data; For the n Layer Network m Node data; For the n Network weight parameters of layer network 1 nodes; For the n Network weight parameters of layer 2 nodes; For the n Layer Network m The network weight parameters of the nodes are obtained through network training; For the n Layer 1 network nodes individually set weights; For the n Layer 2 network nodes individually set weights; For the n Layer Network m The node personal setting weight is 0 by default. It is set by manual experience through later analysis of the relationship between nodes. For the n The bias parameters of the layer network are obtained through network training; For the n The manual setting bias parameter of the layer is 0 by default. It is set based on manual experience through the subsequent analysis of the relationship between each node.

[0086] Among them, the total loss function for training the multi-source information prediction deep neural network is the weighted sum of the elastic modulus and Poisson's ratio prediction loss function, the ground stress and rock water content loss function, and the surrounding rock classification loss function; the elastic modulus and Poisson's ratio prediction loss function is the weighted sum of the prediction loss sub-function, the experience correction loss sub-function, and the rock sample field test loss sub-function; the ground stress and rock water content loss function and the surrounding rock classification loss function are both characterized by the cross entropy loss function.

[0087] In this embodiment, the expression of the elastic modulus and Poisson's ratio prediction loss function is:

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] ;

[0093] in, is the elastic modulus and Poisson’s ratio prediction loss function, Predict loss subfunctions for elastic modulus and Poisson's ratio; Loss sub-function for empirical correction of elastic modulus and Poisson's ratio; Loss sub-function for field testing of rock samples for elastic modulus and Poisson's ratio; 、 They are actually elastic modulus and Poisson’s ratio respectively; 、 are the predicted elastic modulus and Poisson’s ratio, respectively; 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; To obtain the parameter correlation function based on the field rock sample measurement; are longitudinal wave velocity, shear wave velocity, and density respectively; is the anisotropy parameter; and is the weight coefficient; is the L2 norm; All are constant coefficients, obtained through on-site rock sample test analysis and regression.

[0094] Among them, based on the rock property relationship and the pre-constructed tunnel rock quality database, the numerical relationship expressions of longitudinal wave velocity, transverse wave velocity, density, anisotropy parameters, elastic modulus and Poisson's ratio under different rock types and different geological conditions are analyzed to determine the empirical formula.

[0095] Specifically, the empirical formula is expressed as:

[0096]

[0097] in, 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; are longitudinal wave velocity, transverse wave velocity, and density respectively.

[0098] In step S103, artificial experience is introduced to analyze and judge the prediction results. By comparing the prediction results of multiple types of parameters, revealing verification and artificial experience, the weights of various parameters are adjusted, or the prediction results are directly optimized, ultimately achieving multi-angle acquisition and evaluation of tunnel rock properties.

[0099] The main network parameters and hardware conditions in this embodiment are as follows: The calculation is implemented on two graphics cards with 40G video memory. The network is built based on the PyTorch platform, the batch size of the Adam optimizer is 32, and the learning rate of the gradient optimization network parameters is set to 5e-4 .

[0100] This embodiment proposes a tunnel rock property evaluation method based on a network driven by physical laws based on three-component seismic data collected by tunnel seismic wave advance detection technology, combined with on-site geological conditions, seismic attribute-rock property relationship, etc., by successively carrying out anisotropic full waveform inversion, tunnel multi-element rock property prediction, physical law driven constraints, and artificial experience and excavation exposure optimization, the longitudinal wave velocity, transverse wave velocity, density, anisotropy parameters, elastic modulus, Poisson's ratio, ground stress, rock water content and surrounding rock classification in front of the tunnel can be predicted in succession. The prediction and analysis of the above results can help analyze the rock and geological conditions in front of the tunnel, and ultimately help to make safe and efficient tunnel construction decisions to ensure the penetration of key and difficult tunnels.

[0101] In one or more embodiments, Figure 4 As shown, a tunnel rock mass attribute evaluation system based on a physical law driven network is also provided, which specifically includes the following modules:

[0102] The rock mass multi-source feature acquisition module 401 is used to acquire three-component seismic data, obtain the P-wave velocity, S-wave velocity, density and anisotropy parameters through anisotropic full waveform inversion and use them as the rock mass multi-source features;

[0103] The rock mass attribute prediction module 402 is used to predict the deep neural network based on the multi-source characteristics of the rock mass and the trained multi-source information. First, the elastic modulus and Poisson's ratio are predicted, and the elastic modulus and Poisson's ratio are corrected using the empirical formula; then, the in-situ stress and water content of the rock mass are predicted using the multi-source characteristics of the rock mass and the corrected elastic modulus and Poisson's ratio; finally, the surrounding rock classification is predicted using the multi-source characteristics of the rock mass, the in-situ stress and water content of the rock mass and the corrected elastic modulus and Poisson's ratio;

[0104] The rock mass property manual optimization module 403 is used to introduce manual experience and adjust the ground stress and rock mass water content, surrounding rock classification results and the corrected elastic modulus and Poisson's ratio in combination with the actual situation of the tunnel, so as to achieve multi-angle acquisition and evaluation of the tunnel rock mass properties;

[0105] Among them, the total loss function for training the multi-source information prediction deep neural network is the weighted sum of the elastic modulus and Poisson's ratio prediction loss function, the ground stress and rock water content loss function, and the surrounding rock classification loss function; the elastic modulus and Poisson's ratio prediction loss function is the weighted sum of the prediction loss sub-function, the experience correction loss sub-function, and the rock sample field test loss sub-function; the ground stress and rock water content loss function and the surrounding rock classification loss function are both characterized by the cross entropy loss function.

[0106] In the specific implementation process, the expression of the elastic modulus and Poisson's ratio prediction loss function is:

[0107] ;

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] in, is the elastic modulus and Poisson’s ratio prediction loss function, Predict loss subfunctions for elastic modulus and Poisson's ratio; Loss sub-function for empirical correction of elastic modulus and Poisson's ratio; Loss sub-function for field testing of rock samples for elastic modulus and Poisson's ratio; 、 They are actually elastic modulus and Poisson’s ratio respectively; 、 are the predicted elastic modulus and Poisson’s ratio, respectively; 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; To obtain the parameter correlation function based on the field rock sample measurement; are longitudinal wave velocity, shear wave velocity, and density respectively; is the anisotropy parameter; and is the weight coefficient; is the L2 norm; All are constant coefficients, obtained through on-site rock sample test analysis and regression.

[0113] Based on the rock property relationship and the pre-constructed tunnel rock quality database, the numerical relationship expressions of the longitudinal wave velocity, shear wave velocity, density, anisotropy parameters, elastic modulus and Poisson's ratio under different lithologies and geological conditions are analyzed to determine the empirical formula.

[0114] Specifically, the empirical formula is expressed as:

[0115]

[0116] in, 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; are longitudinal wave velocity, transverse wave velocity, and density respectively.

[0117] The multi-source information prediction deep neural network includes a multi-source information preprocessing module, an elastic modulus and Poisson's ratio prediction module, a ground stress and rock mass water content prediction module, and a surrounding rock classification prediction module;

[0118] The multi-source information preprocessing module is composed of a multi-layer convolution block and a self-attention mechanism module, which is used to extract features and correlate the data of longitudinal wave velocity, shear wave velocity, density and anisotropy parameters to obtain multi-source features;

[0119] The elastic modulus and Poisson's ratio prediction module is composed of a multi-layer convolution block and a constraint layer; the convolution block is used to further analyze the multi-source features and predict the elastic modulus and Poisson's ratio; the constraint layer uses an empirical formula to correct the elastic modulus and Poisson's ratio;

[0120] The geostress and rock water content prediction module is composed of a self-attention mechanism module and a multi-layer convolution block, wherein the convolution block is used to extract the corrected elastic modulus and Poisson's ratio features, and then the self-attention mechanism module is used to fuse and weighted correlate the multi-source features and the extracted elastic modulus and Poisson's ratio features to predict geostress and rock water content;

[0121] The self-attention mechanism module is composed of a fully connected network layer, wherein the self-attention mechanism module is used to fuse and weightedly correlate multi-source features, ground stress and rock water content, and modified elastic modulus and Poisson's ratio features, and finally the fully connected network layer predicts the one-dimensional surrounding rock classification result along the front of the tunnel.

[0122] It should be noted here that the specific implementation process of each module in the tunnel rock mass property evaluation system based on physical law driven network is the same as the specific implementation process of each step in the tunnel rock mass property evaluation method based on physical law driven network, which will not be described in detail here.

[0123] In one or more implementations, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the steps in the tunnel rock mass property evaluation method based on a physical law driven network as described above are implemented.

[0124] In one or more embodiments, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the tunnel rock mass property evaluation method based on a physical law driven network as described above are implemented.

[0125] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0126] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0127] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A tunnel rock mass attribute evaluation method based on a physical law driven network, characterized in that: include: Acquire three-component seismic data, obtain P-wave velocity, S-wave velocity, density and anisotropy parameters through anisotropic full waveform inversion and use them as multi-source characteristics of rock mass; Based on the multi-source characteristics of rock mass and the trained multi-source information prediction deep neural network, the elastic modulus and Poisson's ratio are first predicted, and the elastic modulus and Poisson's ratio are corrected using empirical formulas; then the multi-source characteristics of rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the ground stress and rock mass water content; finally, the multi-source characteristics of rock mass, ground stress and rock mass water content, and the corrected elastic modulus and Poisson's ratio are used to predict the surrounding rock classification; Introducing artificial experience and combining the actual situation of the tunnel to artificially adjust the ground stress and rock water content, surrounding rock classification results and the corrected elastic modulus and Poisson's ratio, so as to achieve multi-angle acquisition and evaluation of the tunnel rock properties; Among them, the total loss function of training the multi-source information prediction deep neural network is the weighted sum of the elastic modulus and Poisson's ratio prediction loss function, the ground stress and rock water content loss function and the surrounding rock classification loss function; the elastic modulus and Poisson's ratio prediction loss function is the weighted sum of the prediction loss sub-function, the experience correction loss sub-function and the rock sample field test loss sub-function; the ground stress and rock water content loss function and the surrounding rock classification loss function are both characterized by the cross entropy loss function; The expression of the elastic modulus and Poisson's ratio prediction loss function is: ； ； ； ； ； in, is the elastic modulus and Poisson’s ratio prediction loss function, Predict loss subfunctions for elastic modulus and Poisson's ratio; Loss sub-function for empirical correction of elastic modulus and Poisson's ratio; Loss sub-function for field testing of rock samples for elastic modulus and Poisson's ratio; 、 They are actually elastic modulus and Poisson’s ratio respectively; 、 are the predicted elastic modulus and Poisson’s ratio, respectively; 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; To obtain the parameter correlation function based on the field rock sample measurement; are longitudinal wave velocity, shear wave velocity, and density respectively; is the anisotropy parameter; and is the weight coefficient; is the L2 norm; All are constant coefficients, obtained through on-site rock sample test analysis and regression; The multi-source information prediction deep neural network includes a multi-source information preprocessing module, an elastic modulus and Poisson's ratio prediction module, a ground stress and rock mass water content prediction module, and a surrounding rock classification prediction module; The multi-source information preprocessing module is composed of a multi-layer convolution block and a self-attention mechanism module, which is used to extract features and correlate the data of longitudinal wave velocity, shear wave velocity, density and anisotropy parameters to obtain multi-source features; The elastic modulus and Poisson's ratio prediction module is composed of a multi-layer convolution block and a constraint layer; the convolution block is used to further analyze the multi-source features and predict the elastic modulus and Poisson's ratio; the constraint layer uses an empirical formula to correct the elastic modulus and Poisson's ratio; The geostress and rock water content prediction module is composed of a self-attention mechanism module and a multi-layer convolution block, wherein the convolution block is used to extract the corrected elastic modulus and Poisson's ratio features, and then the self-attention mechanism module is used to fuse and weighted correlate the multi-source features and the extracted elastic modulus and Poisson's ratio features to predict geostress and rock water content; The self-attention mechanism module is composed of a fully connected network layer, wherein the self-attention mechanism module is used to fuse and weightedly correlate multi-source features, ground stress and rock water content, and modified elastic modulus and Poisson's ratio features, and finally the fully connected network layer predicts the one-dimensional surrounding rock classification result along the front of the tunnel.

2. The tunnel rock mass attribute evaluation method based on physical law driven network according to claim 1, characterized in that: Based on the rock property relationship and the pre-constructed tunnel rock quality database, the numerical relationship expressions of the longitudinal wave velocity, shear wave velocity, density, anisotropy parameters, elastic modulus and Poisson's ratio under different lithologies and geological conditions are analyzed to determine the empirical formula.

3. The tunnel rock mass attribute evaluation method based on physical law driven network according to claim 2, characterized in that: The expression of the empirical formula is: in, 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; are longitudinal wave velocity, transverse wave velocity, and density respectively.

4. A tunnel rock mass attribute evaluation system based on a physical law driven network, characterized in that: include: The rock mass multi-source feature acquisition module is used to acquire three-component seismic data, obtain the P-wave velocity, S-wave velocity, density and anisotropy parameters through anisotropic full waveform inversion and use them as the rock mass multi-source features; The rock mass attribute prediction module is used to predict the deep neural network based on the multi-source characteristics of the rock mass and the trained multi-source information. First, the elastic modulus and Poisson's ratio are predicted, and the elastic modulus and Poisson's ratio are corrected using the empirical formula; then the multi-source characteristics of the rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the ground stress and the water content of the rock mass; finally, the multi-source characteristics of the rock mass, the ground stress and the water content of the rock mass and the corrected elastic modulus and Poisson's ratio are used to predict the surrounding rock classification; The rock mass property manual optimization module is used to introduce manual experience and adjust the ground stress, rock mass water content, surrounding rock classification results, and the corrected elastic modulus and Poisson's ratio in combination with the actual situation of the tunnel, so as to achieve multi-angle acquisition and evaluation of the tunnel rock mass properties; Among them, the total loss function of training the multi-source information prediction deep neural network is the weighted sum of the elastic modulus and Poisson's ratio prediction loss function, the ground stress and rock water content loss function and the surrounding rock classification loss function; the elastic modulus and Poisson's ratio prediction loss function is the weighted sum of the prediction loss sub-function, the experience correction loss sub-function and the rock sample field test loss sub-function; the ground stress and rock water content loss function and the surrounding rock classification loss function are both characterized by the cross entropy loss function; The expression of the elastic modulus and Poisson's ratio prediction loss function is: ； ； ； ； ； in, is the elastic modulus and Poisson’s ratio prediction loss function, Predict loss subfunctions for elastic modulus and Poisson's ratio; Loss sub-function for empirical correction of elastic modulus and Poisson's ratio; Loss sub-function for field testing of rock samples for elastic modulus and Poisson's ratio; 、 They are actually elastic modulus and Poisson’s ratio respectively; 、 are the predicted elastic modulus and Poisson’s ratio, respectively; 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; To obtain the parameter correlation function based on the field rock sample measurement; are longitudinal wave velocity, shear wave velocity, and density respectively; is the anisotropy parameter; and is the weight coefficient; is the L2 norm; All are constant coefficients, obtained through on-site rock sample test analysis and regression; The multi-source information prediction deep neural network includes a multi-source information preprocessing module, an elastic modulus and Poisson's ratio prediction module, a ground stress and rock mass water content prediction module, and a surrounding rock classification prediction module; The multi-source information preprocessing module is composed of a multi-layer convolution block and a self-attention mechanism module, which is used to extract features and correlate the data of longitudinal wave velocity, shear wave velocity, density and anisotropy parameters to obtain multi-source features; The elastic modulus and Poisson's ratio prediction module is composed of a multi-layer convolution block and a constraint layer; the convolution block is used to further analyze the multi-source features and predict the elastic modulus and Poisson's ratio; the constraint layer uses an empirical formula to correct the elastic modulus and Poisson's ratio; The geostress and rock water content prediction module is composed of a self-attention mechanism module and a multi-layer convolution block, wherein the convolution block is used to extract the corrected elastic modulus and Poisson's ratio features, and then the self-attention mechanism module is used to fuse and weighted correlate the multi-source features and the extracted elastic modulus and Poisson's ratio features to predict geostress and rock water content; The self-attention mechanism module is composed of a fully connected network layer, wherein the self-attention mechanism module is used to fuse and weightedly correlate multi-source features, ground stress and rock water content, and modified elastic modulus and Poisson's ratio features, and finally the fully connected network layer predicts the one-dimensional surrounding rock classification result along the front of the tunnel.

5. The tunnel rock mass attribute evaluation system based on physical law driven network as claimed in claim 4, characterized in that: Based on the rock property relationship and the pre-constructed tunnel rock quality database, the numerical relationship expressions of the longitudinal wave velocity, shear wave velocity, density, anisotropy parameters, elastic modulus and Poisson's ratio under different lithologies and geological conditions are analyzed to determine the empirical formula.

6. The tunnel rock mass attribute evaluation system based on physical law driven network according to claim 5, characterized in that: The expression of the empirical formula is: in, 、 are the elastic modulus and Poisson's ratio calculated using empirical formulas, respectively; are longitudinal wave velocity, transverse wave velocity, and density respectively.

Images