Intelligent prediction method and system for geometric deformation of twin in tunnel construction

Abstract

The invention discloses a tunnel construction twin geometric deformation intelligent prediction method and system, and solves the problem that tunnel excavation contour deformation cannot be accurately predicted because internal change factors of tunnel geometric deformation cannot be clarified due to complex and changeable tunnel construction environments. The method comprises the following steps of: extracting semantic features of geometric deformation of a tunnel excavation contour based on a tunnel excavation construction process, and constructing an incidence relation model of geometric deformation features of the tunnel excavation contour; based on the incidence relation model, deformation factors are extracted, a grey relational degree model is adopted to carry out grey relational degree analysis on the deformation factors, and key factors of geometric deformation of the tunnel excavation contour are obtained; and obtaining a deformation characteristic data organization model based on the key factors and the excavation contour deformation quantity, and predicting tunnel excavation contour deformation information by adopting the tunnel excavation contour geometric deformation prediction neural network model. The method is used for intelligently predicting the geometric deformation of the twin in the tunnel construction.

Description

Technical Field

[0001] A method and system for intelligently predicting geometric deformation of twin bodies during tunnel construction, which are used for intelligently predicting geometric deformation of twin bodies during tunnel construction, belong to the technical field of tunnel digital twins. Background Art

[0002] As an important part of transportation construction, tunnels can cross mountains and greatly reduce the construction mileage. At the same time, tunnels in complex and dangerous mountainous areas account for a large proportion, which requires higher requirements in terms of construction quality and safety control. Tunnel construction is often affected by complex geological conditions, and it is very easy to cause structural deformation, affecting its stability. Therefore, by monitoring and evaluating the geometric deformation state of the tunnel, it provides a basis and guidance for tunnel construction risk control, construction plan adjustment and management decision-making, which plays a vital role in ensuring the quality of tunnel engineering.

[0003] At present, the commonly used research on tunnel geometric deformation prediction includes manual observation, mathematical statistics and finite element analysis. These methods can establish tunnel models, analyze contour deformation characteristics, realize the monitoring of structural displacement, deformation and external environment, and provide reasonable prediction results within a certain range. However, there are also problems such as low prediction efficiency, long processing time series, and poor collaborative analysis capabilities. When dealing with structural interactions and nonlinear behaviors in large and complex tunnel projects, the effect is limited.

[0004] Digital twin is an advanced technology that realizes the interaction between virtual and real space. It aims to simulate, collect information and create feedback loops across the life cycle of products or systems to effectively monitor, simulate, predict, diagnose and make decisions on physical entities, and provide precise guidance for intelligent management and control of the construction process. By monitoring the deformation of railway construction through digital twins, timely feedback can be obtained, effectively improving management standards and full life cycle application management, and ensuring the safe and efficient development of railway construction.

[0005] Therefore, the existing technology has the following technical difficulties when performing deformation diagnosis of railway engineering twins:

[0006] 1. The tunnel construction environment is complex and changeable, which makes it impossible to clarify the geometric deformation factors of the tunnel, and thus it is impossible to accurately predict the deformation of the tunnel excavation contour.

[0007] 2. There are many data on geometric deformation factors of tunnel excavation profiles, the amount of prediction input data is extremely large, and there is a lack of accurate positioning of key deformation factors, resulting in low prediction efficiency;

[0008] 3. The prediction input data and prediction model used cannot achieve the preparatory prediction of tunnel excavation contour deformation. Summary of the invention

[0009] The purpose of the present invention is to provide a method and system for intelligently predicting the geometric deformation of twin bodies during tunnel construction, so as to solve the problem that the internal changing factors of the geometric deformation of the tunnel cannot be clarified due to the complex and changeable tunnel construction environment, and thus the deformation of the tunnel excavation contour cannot be accurately predicted.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is:

[0011] A method for intelligently predicting geometric deformation of twin bodies during tunnel construction comprises the following steps:

[0012] Step 1: extracting semantic features of geometric deformation of tunnel excavation contour based on tunnel excavation construction process, and constructing an association relationship model of geometric deformation features of tunnel excavation contour;

[0013] Step 2: Extract deformation factors based on the correlation model, and use the grey correlation model to perform grey correlation analysis on the deformation factors to obtain the key factors of the geometric deformation of the tunnel excavation profile;

[0014] Step 3: Obtain a deformation feature data organization model based on key factors and excavation profile deformation variables, and use a tunnel excavation profile geometric deformation prediction neural network model to predict tunnel excavation profile deformation information.

[0015] Further, the specific steps of step 1 are:

[0016] Step 1.1, the tunnel excavation construction process is artificially decomposed into four stages of construction processes: advance support, blasting excavation, initial support and secondary lining construction, among which the tunnel excavation construction process is the blasting method construction process;

[0017] Step 1.2, based on the decomposed construction procedures, the semantic features of the geometric deformation of the tunnel excavation profile are extracted, wherein the semantic features of the advance support stage include the material, length, diameter and density of the pipe shed, advance anchor and small guide tube; the semantic features of the blasting excavation stage include the type, position, number and size of the blasthole, the type of surrounding rock and the size of the heterodyne angle; the semantic features of the initial support stage include the size, material, position and density of the anchor, shotcrete and steel mesh; the semantic features of the secondary lining construction stage include the size of the arch frame, material density and the material of the concrete;

[0018] Step 1.3: Based on the basic data, the interaction relationship in the geometric deformation process of the tunnel excavation profile is artificially clarified, and the semantic features within different construction processes and between different construction processes are established, as well as the hierarchical relationship, attribute relationship and semantic relationship between each semantic feature and the deformation factor. That is, the association relationship model of the geometric deformation characteristics of the tunnel excavation profile is constructed, in which the basic data includes standards and specifications, geological data and literature basis.

[0019] Further, the specific steps of step 2 are:

[0020] Step 2.1, deformation mechanism analysis based on the correlation model is performed to extract deformation factors involved in the excavation profile deformation process during tunnel excavation construction to form a deformation factor set, wherein the deformation factors include excavation environment factors, deformation process factors and deformation risk factors, wherein the deformation mechanism analysis refers to the correlation analysis, regression analysis or graph data reasoning between the deformation factors;

[0021] Step 2.2: Use the grey correlation model to calculate the weight of each deformation factor in the deformation factor set, specifically:

[0022] The reference sequence for deforming the tunnel excavation profile is:

[0023] X0=[x0(k)|k=1,2,…,n] (1)

[0024] The comparison sequence of each deformation factor in the deformation factor set is:

[0025] X i =[x i (k)|k=1,2,…,n] (2)

[0026] In the formula, i = 1, 2, 3, ..., m, m represents the total number of deformation factors, n represents the number of time series contained in each reference sequence or comparison sequence, i represents the i-th deformation factor, X i represents the comparison sequence of the i-th deformation factor, k represents the k-th time point in the i-th deformation factor, x i (k) represents the comparison sequence of the i-th deformation factor, and x0(k) represents the reference sequence at the k-th time point;

[0027] The mean method is used to perform dimensionless processing on the reference sequence and the comparison sequence of each deformation factor. The formula is:

[0028]

[0029]

[0030] In the formula, x′0(k) represents the averaging result of the reference sequence, represents the mean value of the reference sequence, x′ i (k) is the average result of the comparison sequence of the i-th deformation factor, is the average value of the comparison sequence of the i-th deformation factor;

[0031] Calculate the absolute difference between the averaged comparison sequence and the reference sequence to obtain the difference sequence Δx i (k):

[0032] Δxi (k)=|x′0(k)-x′ i (k)| (5)

[0033] Based on the difference sequence, we get X0 and X i The grey correlation coefficient is:

[0034]

[0035] In the formula, min i min k Δx i (k) represents the minimum absolute difference of all difference sequences, max i max k Δx i (k) represents the maximum absolute difference of all difference sequences, and ρ represents the resolution coefficient;

[0036] The grey correlation coefficients at each moment are concentrated into one value, that is, the average value of m grey correlation coefficients is calculated as a quantitative representation of the correlation degree between the comparison sequence of each deformation factor and the reference sequence. The formula is:

[0037]

[0038] In the formula, γ 0i Represents the grey relational degree of the comparison sequence of the i-th deformation factor to the reference sequence;

[0039] Based on each grey correlation value, the minimum and maximum intervals of the grey correlation value distribution are determined as [min(γ), max(γ)]. After determination, they are evenly divided into 9 sub-intervals, and the indicator importance index is assigned in descending order of grey correlation degree, where the indicator importance index of the sub-interval with the highest correlation degree to the indicator importance index of the sub-interval with the lowest correlation degree are 1-9 respectively;

[0040] According to the index importance index, construct the judgment matrix B, and the element b of the judgment matrix B ij It represents the ratio of the index importance index of the i-th deformation factor to the j-th deformation factor. If I i j , If I i >I j , Among them, I j Represents the index importance index of the jth deformation factor, I i Represents the index importance index of the i-th deformation factor;

[0041] Calculate the maximum eigenvalue λ of the judgment matrix B max and its corresponding eigenvector W = [w1,w2,w​i …,w n ] T , the obtained eigenvector element w i The indicator importance index I i The weight of the corresponding deformation factor i;

[0042] Step 2.3: Based on the weights of all deformation factors of the tunnel excavation profile calculated in step 2.2, sort them from large to small, obtain the deformation factors corresponding to the front weights after sorting, and obtain the key factors of the geometric deformation of the tunnel excavation profile.

[0043] Further, the specific steps of step 3 are:

[0044] Step 3.1, by comparing the multi-phase point cloud data of tunnel excavation construction, the difference between the point cloud data is calculated to obtain the excavation profile deformation variable, and the prediction sample library of the geometric deformation of the tunnel excavation profile is formed together with the key factors;

[0045] Step 3.2, projecting the tunnel excavation construction point cloud data in the prediction sample library onto a plane, so that the points on the plane form a one-to-one correspondence with the point cloud data, and thus obtaining a deformation feature data organization model;

[0046] Step 3.3, using CNN and RNN connected in sequence to construct a neural network model for predicting geometric deformation of tunnel excavation profile;

[0047] Step 3.4: The tunnel excavation contour geometric deformation prediction neural network model is used to predict the deformation feature data organization model, that is, after the deformation space features are extracted by CNN, they are connected to the RNN to extract the time features, and finally the deformation spatiotemporal features are obtained, which are the tunnel excavation contour deformation information.

[0048] Further, the specific steps of step 3.2 are:

[0049] Step 3.21, dividing the projection plane into a plurality of overlapping p×q grids according to the deformation factor, wherein p represents the number of arcs involved in predicting the tunnel section contour along the tunnel length direction, and q represents the number of points on the arc along the tunnel width direction;

[0050] Step 3.22: Based on the prediction sample library, the points on the tunnel surface are projected downward onto the grid along the direction perpendicular to the ground, so that the grid stores the key factors of the excavation contour deformation variables and the geometric deformation of the tunnel excavation contour of the corresponding points on the tunnel surface. The final grid is the deformation feature data organization model.

[0051] An intelligent prediction system for geometric deformation of twin bodies in tunnel construction, comprising:

[0052] Association model building module: extracts semantic features of geometric deformation of tunnel excavation contour based on tunnel excavation construction process, and builds an association model of geometric deformation features of tunnel excavation contour;

[0053] Key factor acquisition module: Extract deformation factors based on the correlation model, and use the grey correlation model to perform grey correlation analysis on the deformation factors to obtain the key factors of the geometric deformation of the tunnel excavation profile;

[0054] Tunnel excavation profile deformation information prediction module: The deformation feature data organization model is obtained based on key factors and excavation profile deformation variables, and the tunnel excavation profile geometric deformation prediction neural network model is used to predict the tunnel excavation profile deformation information.

[0055] Furthermore, the specific implementation steps of the association relationship model building module are:

[0056] Step 1.1, the tunnel excavation construction process is artificially decomposed into four stages of construction processes: advance support, blasting excavation, initial support and secondary lining construction, among which the tunnel excavation construction process is the blasting method construction process;

[0057] Step 1.2, based on the decomposed construction procedures, the semantic features of the geometric deformation of the tunnel excavation profile are extracted, wherein the semantic features of the advance support stage include the material, length, diameter and density of the pipe shed, advance anchor and small guide tube; the semantic features of the blasting excavation stage include the type, position, number and size of the blasthole, the type of surrounding rock and the size of the heterodyne angle; the semantic features of the initial support stage include the size, material, position and density of the anchor, shotcrete and steel mesh; the semantic features of the secondary lining construction stage include the size of the arch frame, material density and the material of the concrete;

[0058] Step 1.3: Based on the basic data, the interaction relationship in the geometric deformation process of the tunnel excavation profile is artificially clarified, and the semantic features within different construction processes and between different construction processes are established, as well as the hierarchical relationship, attribute relationship and semantic relationship between each semantic feature and the deformation factor. That is, the association relationship model of the geometric deformation characteristics of the tunnel excavation profile is constructed, in which the basic data includes standards and specifications, geological data and literature basis.

[0059] Furthermore, the specific implementation steps of the key factor acquisition module are:

[0060] Step 2.1, deformation mechanism analysis based on the correlation model is performed to extract deformation factors involved in the excavation profile deformation process during tunnel excavation construction to form a deformation factor set, wherein the deformation factors include excavation environment factors, deformation process factors and deformation risk factors, wherein the deformation mechanism analysis refers to the correlation analysis, regression analysis or graph data reasoning between the deformation factors;

[0061] Step 2.2: Use the grey correlation model to calculate the weight of each deformation factor in the deformation factor set, specifically:

[0062] The reference sequence for deforming the tunnel excavation profile is:

[0063] X0=[x0(k)|k=1,2,…,n] (1)

[0064] The comparison sequence of each deformation factor in the deformation factor set is:

[0065] X i =[x i (k)|k=1,2,…,n] (2)

[0066] In the formula, i = 1, 2, 3, ..., m, m represents the total number of deformation factors, n represents the number of time series contained in each reference sequence or comparison sequence, i represents the i-th deformation factor, X i represents the comparison sequence of the i-th deformation factor, k represents the k-th time point in the i-th deformation factor, x i (k) represents the comparison sequence of the i-th deformation factor, and x0(k) represents the reference sequence at the k-th time point;

[0067] The mean method is used to perform dimensionless processing on the reference sequence and the comparison sequence of each deformation factor. The formula is:

[0068]

[0069]

[0070] In the formula, x′0(k) represents the averaging result of the reference sequence, represents the mean value of the reference sequence, x′ i (k) is the average result of the comparison sequence of the i-th deformation factor, is the average value of the comparison sequence of the i-th deformation factor;

[0071] Calculate the absolute difference between the averaged comparison sequence and the reference sequence to obtain the difference sequence Δx i (k):

[0072] Δx i (k)=|x′0(k)-x′ i (k)| (5)

[0073] Based on the difference sequence, we get X0 and X i The grey correlation coefficient is:

[0074]

[0075] In the formula, mini min k Δx i (k) represents the minimum absolute difference of all difference sequences, max i max k Δx i (k) represents the maximum absolute difference of all difference sequences, and ρ represents the resolution coefficient;

[0076] The grey correlation coefficients at each moment are concentrated into one value, that is, the average value of m grey correlation coefficients is calculated as a quantitative representation of the correlation degree between the comparison sequence of each deformation factor and the reference sequence. The formula is:

[0077]

[0078] In the formula, γ 0i Represents the grey relational degree of the comparison sequence of the i-th deformation factor to the reference sequence;

[0079] Based on each grey correlation value, the minimum and maximum intervals of the grey correlation value distribution are determined as [min(γ), max(γ)]. After determination, they are evenly divided into 9 sub-intervals, and the indicator importance index is assigned in descending order of grey correlation degree, where the indicator importance index of the sub-interval with the highest correlation degree to the indicator importance index of the sub-interval with the lowest correlation degree are 1-9 respectively;

[0080] According to the index importance index, construct the judgment matrix B, and the element b of the judgment matrix B ij It represents the ratio of the index importance index of the i-th deformation factor to the j-th deformation factor. If I i j , If I i >I j , Among them, I j Represents the index importance index of the jth deformation factor, I i Represents the index importance index of the i-th deformation factor;

[0081] Calculate the maximum eigenvalue λ of the judgment matrix B max and its corresponding eigenvector W = [w1,w2,w i …,w n ] T , the obtained eigenvector element w i The indicator importance index I i The weight of the corresponding deformation factor i;

[0082] ​Step 2.3: Based on the weights of all deformation factors of the tunnel excavation profile calculated in step 2.2, sort them from large to small, obtain the deformation factors corresponding to the front weights after sorting, and obtain the key factors of the geometric deformation of the tunnel excavation profile.

[0083] Furthermore, the specific implementation steps of the tunnel excavation profile deformation information prediction module are as follows:

[0084] Step 3.1, by comparing the multi-phase point cloud data of tunnel excavation construction, the difference between the point cloud data is calculated to obtain the excavation profile deformation variable, and the prediction sample library of the geometric deformation of the tunnel excavation profile is formed together with the key factors;

[0085] Step 3.2, projecting the tunnel excavation construction point cloud data in the prediction sample library onto a plane, so that the points on the plane form a one-to-one correspondence with the point cloud data, and thus obtaining a deformation feature data organization model;

[0086] Step 3.3, using CNN and RNN connected in sequence to construct a neural network model for predicting geometric deformation of tunnel excavation profile;

[0087] Step 3.4: The tunnel excavation contour geometric deformation prediction neural network model is used to predict the deformation feature data organization model, that is, after the deformation space features are extracted by CNN, they are connected to the RNN to extract the time features, and finally the deformation spatiotemporal features are obtained, which are the tunnel excavation contour deformation information.

[0088] Further, the specific steps of step 3.2 are:

[0089] Step 3.21, dividing the projection plane into a plurality of overlapping p×q grids according to the deformation factor, wherein p represents the number of arcs involved in predicting the tunnel section contour along the tunnel length direction, and q represents the number of points on the arc along the tunnel width direction;

[0090] Step 3.22: Based on the prediction sample library, the points on the tunnel surface are projected downward onto the grid along the direction perpendicular to the ground, so that the grid stores the key factors of the excavation contour deformation variables and the geometric deformation of the tunnel excavation contour of the corresponding points on the tunnel surface. The final grid is the deformation feature data organization model.

[0091] Compared with the prior art, the advantages of the present invention are:

[0092] 1. The present invention sorts out the tunnel drilling and blasting construction process, analyzes the geometric scene object relationship and deformation-related factors of the tunnel excavation contour, constructs a tunnel excavation contour association model, and realizes an accurate description of the geometric deformation process of the tunnel excavation contour under complex environments;

[0093] Second, the present invention uses grayscale correlation analysis to analyze external deformation factors, including temperature, humidity, surrounding rock and other conditions, screens key factors and forms a prediction sample library of tunnel excavation profile geometric deformation with excavation profile deformation variables, realizes the optimal input of prediction data, can accurately locate the key factors of tunnel excavation profile geometric deformation, and improves prediction efficiency;

[0094] 3. The present invention generates a two-dimensional array through data reconstruction (i.e., projection) of the collected point cloud data to obtain a deformation feature data organization model, and fully considers the demand for accurate deformation prediction during tunnel excavation, explores the characteristics and advantages of different deformation prediction neural network models, and establishes a CNN-RNN deformation accuracy prediction model (tunnel excavation contour geometric deformation prediction neural network model) that takes into account both time continuity and spatial proximity. The spatial features and time series features are directly extracted on the two-dimensional array to obtain the tunnel excavation contour deformation information, breaking through the difficult problem of low accuracy in predicting the geometric deformation of the tunnel excavation contour. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0096] Figure 1 This is the overall research idea of ​​the present invention, wherein: Figure 1 The deep learning neural network shown in Figure 4 The neural network model for predicting geometric deformation of tunnel excavation profile shown in ;

[0097] Figure 2 It is an analysis diagram of the geometric deformation process of the tunnel excavation profile with semantic constraints in the present invention;

[0098] Figure 3 It is a geometric deformation factor evaluation diagram of the tunnel excavation profile in the present invention;

[0099] Figure 4 Schematic diagram of geometric deformation prediction of tunnel excavation profile in the present invention, wherein U represents the connection weight from the input layer to the hidden layer in the recurrent time network, v represents the connection weight from the hidden layer to the hidden layer, and w represents the connection weight from the hidden layer to the output layer;

[0100] Figure 5 This is an example diagram for inputting the deformation prediction model in the present invention. DETAILED DESCRIPTION

[0101] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0102] The overall research idea is as follows Figure 1 As shown in the figure, it mainly includes three aspects: analysis of geometric deformation process during tunnel construction, construction of evaluation model of geometric deformation factors of tunnel twins and prediction of tunnel geometric deformation information.

[0103] The analysis of the geometric deformation process of tunnel excavation contour under semantic constraints is to extract the semantic features of the geometric deformation of tunnel excavation contour based on the tunnel excavation construction process, and to build a correlation model of the geometric deformation features of tunnel excavation contour, specifically:

[0104] Tunnel engineering procedures are complicated, and the construction process involves many factors. There are many factors affecting the structural deformation during the tunnel construction process, and the correlation is complex. How to combine these factors to establish a unified tunnel deformation description model and fully display the structural deformation mechanism is a problem that needs to be considered. Based on the tunnel excavation construction process, the present invention constructs a system of factors affecting the deformation of the tunnel excavation profile, analyzes the hierarchical relationship, attribute relationship, and semantic relationship between structural deformation and influencing factors, and realizes an accurate description of the multi-factor characteristics and coupling relationship of the geometric deformation of the tunnel engineering.

[0105] The tunnel excavation construction process is decomposed into four stages: advance support, blasting excavation, initial support and secondary lining construction. Among them, the tunnel excavation construction process is a blasting method construction process.

[0106] The semantic features of the geometric deformation of the tunnel excavation profile are extracted based on the decomposed construction procedures. Among them, the semantic features of the advanced support stage include the material, length, diameter, density, etc. of the pipe shed, advanced anchor, and small guide tube; the semantic features of the blasting excavation stage include the type, position, number, size of the blasthole, the type of surrounding rock, the size of the heterodyne angle, etc.; the semantic features of the initial support stage include the size, material, position, density, etc. of the anchor, shotcrete, and steel mesh; the semantic features of the secondary lining construction stage include the size, material density, and material of the arch frame.

[0107] Further, based on a variety of basic data references, such as standards and specifications, geological data, literature basis and other data, the interaction relationship in the geometric deformation process of the tunnel excavation profile, the semantic features of each other within and between different construction processes, and the hierarchical relationship, attribute relationship and semantic relationship between each semantic feature and the deformation factors are clarified, and a correlation relationship model of the geometric deformation factors of the tunnel excavation profile is constructed to achieve a complete analysis of the tunnel excavation profile deformation process.

[0108] Evaluation of geometric deformation factors of tunnel excavation profile based on grey correlation model, that is, extracting deformation factors based on correlation model, and using grey correlation model to perform grey correlation analysis on deformation factors, to obtain key factors of geometric deformation of tunnel excavation profile, specifically:

[0109] The evaluation method of geometric deformation factors of tunnel excavation profile is as follows: Figure 4 As shown:

[0110] Through the construction of the correlation model of the geometric deformation factors of the tunnel excavation profile mentioned above, the complex conditions of the tunnel construction excavation events can be comprehensively sorted out. That is, the deformation factors involved in the excavation profile deformation process during tunnel excavation construction are extracted through deformation mechanism analysis to form a set of deformation factors, among which the deformation factors include excavation environment factors, deformation process factors, deformation risk factors, etc. Among them, deformation mechanism analysis refers to the correlation analysis, regression analysis or graph data reasoning between deformation factors.

[0111] The grey correlation model is used to calculate the weight of each deformation factor in the deformation factor set, specifically:

[0112] The reference sequence for deforming the tunnel excavation profile is:

[0113] X0=[x0(k)|k=1,2,…,n] (1)

[0114] The comparison sequence of each deformation factor in the deformation factor set is:

[0115] X i =[x i (k)|k=1,2,…,n] (2)

[0116] In the formula, i = 1, 2, 3, ..., m, m represents the total number of deformation factors, n represents the number of time series contained in each reference sequence or comparison sequence, i represents the i-th deformation factor, X i represents the comparison sequence of the i-th deformation factor, k represents the k-th time point in the i-th deformation factor, x i(k) represents the comparison sequence of the i-th deformation factor, and x0(k) represents the reference sequence at the k-th time point; due to the large differences in magnitude and measurement units between the deformation factors, it is impossible to directly compare and calculate the deformation factors, so it is necessary to convert the analysis data into dimensionless values. The deformation factor data is processed using data processing operators to convert them into dimensionless data with roughly similar magnitudes.

[0117] The mean method is used to perform dimensionless processing on the reference sequence and the comparison sequence of each deformation factor. The formula is:

[0118]

[0119]

[0120] In the formula, x′0(k) represents the averaging result of the reference sequence, represents the mean value of the reference sequence, x′ i (k) is the average result of the comparison sequence of the i-th deformation factor, is the average value of the comparison sequence of the i-th deformation factor;

[0121] Calculate the absolute difference between the averaged comparison sequence and the reference sequence to obtain the difference sequence Δx i (k):

[0122] Δx i (k)=|x′0(k)-x′ i (k)| (5)

[0123] Based on the difference sequence, we get X0 and X i The grey correlation coefficient is:

[0124]

[0125] In the formula, min i min k Δx i (k) is the minimum value of all absolute differences, max i max k Δx i (k) is the maximum value of all absolute differences, ρ is the resolution system number, ranging from 0 to 1, usually 0.5.

[0126] The correlation coefficients at each moment are concentrated into one value, that is, the average value of m grey correlation coefficients is calculated as a quantitative representation of the correlation degree between the comparison sequence and the reference sequence:

[0127]

[0128] In the formula, γ 0iRepresents the grey relational degree of the comparison sequence of the i-th deformation factor to the reference sequence;

[0129] Based on each grey correlation value, the minimum and maximum intervals of the grey correlation value distribution are determined as [min(γ), max(γ)]. After determination, they are evenly divided into 9 subintervals, and the indicator importance index is assigned in descending order of grey correlation degree. Among them, the indicator importance index of the subinterval with the highest correlation degree to the indicator importance index of the subinterval with the lowest correlation degree are 1-9, that is, 1, 2, 3, 4, 5, 6, 7, 8, 9 respectively;

[0130] According to the index importance index, construct the judgment matrix B, and the element b of the judgment matrix B ij It represents the ratio of the index importance index of the i-th deformation factor to the j-th deformation factor. If I i j , If I i >I j , Among them, I j Represents the index importance index of the jth deformation factor, I i Represents the index importance index of the i-th deformation factor;

[0131] Calculate the maximum eigenvalue λ of the judgment matrix B max and its corresponding eigenvector W = [w1,w2,w i …,w n ] T , the obtained eigenvector element w i The indicator importance index I i The weight of the corresponding deformation factor i; based on the calculated weights of all deformation factors of the tunnel excavation profile, sort them from large to small, obtain the deformation factor corresponding to the front weight after sorting, and obtain the key factors of the geometric deformation of the tunnel excavation profile.

[0132] The greater the correlation, the greater its impact on deformation. Based on the above grey correlation key index screening model, the correlation between the deformation factors of tunnel excavation profile and deformation is measured to achieve the extraction of key indicators (i.e. key factors) of tunnel excavation profile deformation.

[0133] Tunnel geometric deformation information prediction, that is, obtaining the deformation feature data organization model based on key factors and excavation profile deformation variables, and using the tunnel excavation profile geometric deformation prediction neural network model to predict the tunnel excavation profile deformation information, specifically:

[0134] ​Firstly, according to the extraction of key factors in tunnel excavation contour deformation, key knowledge is effectively screened out from a large number of potential factors, thereby simplifying data input and achieving optimal sample library input. The excavation contour deformation variable is obtained by comparing the multi-phase point cloud data of tunnel excavation construction and calculating the difference between the point cloud data, which together with the key factors of tunnel excavation contour deformation constitutes a prediction sample library of tunnel excavation contour geometric deformation.

[0135] The tunnel excavation construction point cloud data in the prediction sample library is projected onto a plane, and the points on the plane image form a one-to-one correspondence with the point cloud data, and the deformation feature data organization model is obtained. That is, in order to use the neural network for feature extraction, the tunnel surface is projected onto a two-dimensional plane, and the projection plane is divided into sufficiently dense grids. Each grid stores the deformation size and deformation key factors of the corresponding points on the tunnel surface. Specifically, the projection plane is divided into multiple superimposed p×q grids according to the deformation factors, where p represents the number of arc segments involved in predicting the tunnel segment contour along the tunnel length direction, and q represents the number of points on the arc segment along the tunnel width direction; based on the prediction sample library, the points on the tunnel surface are projected downward onto the grid along the vertical ground direction, so that the grid stores the excavation contour deformation of the corresponding points on the tunnel surface and the key factors of the geometric deformation of the tunnel excavation contour. The grid finally obtained is the deformation feature data organization model. In this way, the deformation characteristics of the tunnel are effectively organized and stored, which is convenient for subsequent analysis and visualization.

[0136] A neural network model for predicting geometric deformation of tunnel excavation contour is constructed by using sequentially connected CNNs (including sequentially connected convolutional layers, pooling layers, convolutional layers and fully connected layers) and RNNs. Convolutional neural networks (CNNs) can better extract main features while ignoring the influence of interference items. Therefore, the present invention is based on convolutional neural networks and can predict and output tunnel excavation contour deformation on the basis of fully extracting the spatial features of tunnel excavation contour deformation.

[0137] For complex nonlinear relationships, the prediction accuracy of the neural network model is higher. The characteristic of RNN is that all nodes (i.e., recurrent units) are connected in a chain structure. It is a recursive neural network with sequence data as input and recursion in the direction of sequence evolution. As a widely used deep learning model, RNN is unique in concept, mainly in terms of its cross-time steps and multiple hidden layers between input and output. Each layer is represented by a time step. The data features learned by the neural network at each step can be shared in all time steps, and the input of each time step is carried out from front to back, with strong learning ability.

[0138] like Figure 4As shown in the figure, after the data (deformation feature data organization model) is input, the internal work of the neural network is mainly divided into the following parts:

[0139] 1) Convolution calculation

[0140] Convolution calculation is the most basic part in the convolutional layer neural network model. The information of the model will be convolutionally calculated by the convolution layer to obtain the characteristic parameters of the input information. The calculation formula of convolution is as follows:

[0141] C=∑l ef *m ef (8)

[0142] In the formula, C represents the input two-dimensional matrix, l ef represents the element in the e-th row and f-th column of the input matrix, m ef Represents the element in the e-th row and f-th column of the convolution kernel;

[0143] After entering the convolutional layer, the data will be processed into the following format:

[0144] input=[batch, steps, channel] (9)

[0145] In the formula, batch represents the number of samples, steps represents the step size, and channel represents the channel.

[0146] Convolution operation is mainly used to explore the mapping relationship between the deformation values ​​of key deformation indicators, which is mainly achieved through window sliding operation. During the operation, key parameters such as sliding step size, convolution kernel size and number need to be set according to the input data sample information to extract data features that are more suitable for the prediction model. After the convolution operation, the output feature vector is in the form of:

[0147] output=[batch,new_steps,filters] (10)

[0148] Where new_steps is the output step size, which is because the input data step size will change with padding and stride changes; filters is the number of convolution kernels, which will be updated with the convolution operation.

[0149] 2) Pooling calculation

[0150] In the convolutional layer of the network structure, adding a pooling layer can effectively reduce the parameters of the network image information, reduce the sampling pressure, and thus effectively improve the quality of the model's image data, and can improve the field of modeling, which is a downsampling operation. The pooling structure is also used as more priors on image features, which can make the model structure emphasize global information more, and increase the fault tolerance of the entire network structure while ensuring the fidelity of the original image information. At the same time, since the pooling layer has no weight parameters to learn, it can also reduce the number of network parameters to a certain extent, which has the effect of avoiding overfitting.

[0151] 3) Cyclic Time Network

[0152] The forward propagation process of RNN is essentially the propagation of a simple neural network in a continuous cycle, and the time steps are interrelated, with the hidden layer information of the current time step being passed to the next time step. The RNN input-to-hidden connection is parameterized by the weight matrix U, the hidden-to-hidden recurrent connection is parameterized by the weight matrix W, and the hidden-to-output connection is parameterized by the weight matrix V. The value h of the RNN hidden layer is determined by the current input value x and the previous hidden layer value h. As shown in the RNN calculation expansion diagram, at time t, the input value x t , hidden layer h t , output value o t .h t The value of x t and h t The calculation method of RNN is as follows:

[0153] o t =g(Vh t ) (11)

[0154] h t =f(Ux t +Wh t-1 ) (12)

[0155] 4) Result output

[0156] After training, the output of CNN is input into RNN, and the output of RNN is connected to the fully connected layer to achieve the final output, that is, the predicted value of tunnel excavation contour deformation will be output.

[0157] Finally, the trained tunnel excavation contour geometric deformation prediction neural network model is used to predict the deformation feature data organization model. That is, the deformation space features are extracted by CNN and then connected to RNN to extract the time features. Finally, the deformation spatiotemporal features are obtained, which are the tunnel excavation contour deformation information.

[0158] In response to the technical difficulties, the present invention is based on theoretical methods such as tunnel construction monitoring, deep learning, and virtual geographic environment. Aiming at the need for accurate prediction of the geometric deformation of engineering bodies during the construction process, the present invention designs a twin deformation information model of tunnel engineering, integrates the correlation characteristics of structural deformation and influencing factors, and establishes an intelligent prediction method and system for the geometric deformation of tunnel construction twins based on the grey correlation method and the improved long and short-time neural network method.

Claims

1. An intelligent prediction method for geometric deformation of twin bodies in tunnel construction, characterized in that: The steps include: Step 1: extracting semantic features of geometric deformation of tunnel excavation contour based on tunnel excavation construction process, and constructing an association relationship model of geometric deformation features of tunnel excavation contour; Step 2: Extract deformation factors based on the correlation model, and use the grey correlation model to perform grey correlation analysis on the deformation factors to obtain the key factors of the geometric deformation of the tunnel excavation profile; Step 3: Obtain a deformation feature data organization model based on key factors and excavation profile deformation variables, and use a tunnel excavation profile geometric deformation prediction neural network model to predict tunnel excavation profile deformation information.

2. The method for intelligently predicting geometric deformation of twin bodies during tunnel construction according to claim 1, characterized in that: The specific steps of step 1 are: Step 1.1, the tunnel excavation construction process is artificially decomposed into four stages of construction processes: advance support, blasting excavation, initial support and secondary lining construction, among which the tunnel excavation construction process is the blasting method construction process; Step 1.2, based on the decomposed construction procedures, the semantic features of the geometric deformation of the tunnel excavation profile are extracted, wherein the semantic features of the advance support stage include the material, length, diameter and density of the pipe shed, advance anchor and small guide tube; the semantic features of the blasting excavation stage include the type, position, number and size of the blasthole, the type of surrounding rock and the size of the heterodyne angle; the semantic features of the initial support stage include the size, material, position and density of the anchor, shotcrete and steel mesh; the semantic features of the secondary lining construction stage include the size of the arch frame, material density and the material of the concrete; Step 1.3: Based on the basic data, the interaction relationship in the geometric deformation process of the tunnel excavation profile is artificially clarified, and the semantic features within different construction processes and between different construction processes are established, as well as the hierarchical relationship, attribute relationship and semantic relationship between each semantic feature and the deformation factor. That is, the association relationship model of the geometric deformation characteristics of the tunnel excavation profile is constructed, in which the basic data includes standards and specifications, geological data and literature basis.

3. The method for intelligent prediction of geometric deformation of twin bodies in tunnel construction according to claim 2 is characterized by: The specific steps of step 2 are: Step 2.1, deformation mechanism analysis based on the correlation model is performed to extract deformation factors involved in the excavation profile deformation process during tunnel excavation construction to form a deformation factor set, wherein the deformation factors include excavation environment factors, deformation process factors and deformation risk factors, wherein the deformation mechanism analysis refers to the correlation analysis, regression analysis or graph data reasoning between the deformation factors; Step 2.2: Use the grey correlation model to calculate the weight of each deformation factor in the deformation factor set, specifically: The reference sequence for deforming the tunnel excavation profile is: x0=[x0(k)|k=1,2,…,n] (1) The comparison sequence of each deformation factor in the deformation factor set is: X i =[x i (k)|k=1,2,…,n] (2) In the formula, i = 1, 2, 3, ..., m, m represents the total number of deformation factors, n represents the number of time series contained in each reference sequence or comparison sequence, i represents the i-th deformation factor, X i represents the comparison sequence of the i-th deformation factor, k represents the k-th time point in the i-th deformation factor, x i (k) represents the comparison sequence of the i-th deformation factor, and x0(k) represents the reference sequence at the k-th time point; The mean method is used to perform dimensionless processing on the reference sequence and the comparison sequence of each deformation factor. The formula is: In the formula, x′0(k) represents the average result of the reference sequence, represents the mean value of the reference sequence, x′ i (k) is the average result of the comparison sequence of the i-th deformation factor, is the average value of the comparison sequence of the i-th deformation factor; Calculate the absolute difference between the averaged comparison sequence and the reference sequence to obtain the difference sequence Δx i (k): Δx i (k)=|x′0(k)-x′ i (k)| (5) Based on the difference sequence, we get X0 and X i The grey correlation coefficient is: In the formula, min i min k Δx i (k) represents the minimum absolute difference of all difference sequences, max i max k Δx i (k) represents the maximum absolute difference of all difference sequences, and ρ represents the resolution coefficient; The grey correlation coefficients at each moment are concentrated into one value, that is, the average value of m grey correlation coefficients is calculated as a quantitative representation of the correlation degree between the comparison sequence of each deformation factor and the reference sequence. The formula is: In the formula, γ 0i Represents the grey relational degree of the comparison sequence of the i-th deformation factor to the reference sequence; Based on each grey correlation value, the minimum and maximum intervals of the grey correlation value distribution are determined as [min(γ), max(γ)]. After determination, they are evenly divided into 9 sub-intervals, and the indicator importance index is assigned in descending order of grey correlation degree, where the indicator importance index of the sub-interval with the highest correlation degree to the indicator importance index of the sub-interval with the lowest correlation degree are 1-9 respectively; According to the index importance index, construct the judgment matrix B, and the element b of the judgment matrix B ij It represents the ratio of the index importance index of the i-th deformation factor to the j-th deformation factor. If I i j , If I i >I j , Among them, I j Represents the index importance index of the jth deformation factor, I i Represents the index importance index of the i-th deformation factor;​ Calculate the maximum eigenvalue λ of the judgment matrix B max and its corresponding eigenvector W = [w1,w2,w i …,w n ] T , the obtained eigenvector element w i The indicator importance index I i The weight of the corresponding deformation factor i; Step 2.3: Based on the weights of all deformation factors of the tunnel excavation profile calculated in step 2.2, sort them from large to small, obtain the deformation factors corresponding to the front weights after sorting, and obtain the key factors of the geometric deformation of the tunnel excavation profile.

4. The method for intelligent prediction of geometric deformation of twin bodies in tunnel construction according to claim 3 is characterized by: The specific steps of step 3 are: Step 3.1, by comparing the multi-phase point cloud data of tunnel excavation construction, the difference between the point cloud data is calculated to obtain the excavation profile deformation variable, and the prediction sample library of the geometric deformation of the tunnel excavation profile is formed together with the key factors; Step 3.2, projecting the tunnel excavation construction point cloud data in the prediction sample library onto a plane, so that the points on the plane form a one-to-one correspondence with the point cloud data, and thus obtaining a deformation feature data organization model; Step 3.3, using CNN and RNN connected in sequence to construct a neural network model for predicting geometric deformation of tunnel excavation profile; Step 3.4: The tunnel excavation contour geometric deformation prediction neural network model is used to predict the deformation feature data organization model, that is, after the deformation space features are extracted by CNN, they are connected to the RNN to extract the time features, and finally the deformation spatiotemporal features are obtained, which are the tunnel excavation contour deformation information.

5. The method for intelligently predicting geometric deformation of twin bodies during tunnel construction according to claim 4 is characterized in that: The specific steps of step 3.2 are: Step 3.21, dividing the projection plane into a plurality of overlapping p×q grids according to the deformation factor, wherein p represents the number of arcs involved in predicting the tunnel section contour along the tunnel length direction, and q represents the number of points on the arc along the tunnel width direction; Step 3.22: Based on the prediction sample library, the points on the tunnel surface are projected downward onto the grid along the direction perpendicular to the ground, so that the grid stores the key factors of the excavation contour deformation variables and the geometric deformation of the tunnel excavation contour of the corresponding points on the tunnel surface. The final grid is the deformation feature data organization model.

6. An intelligent prediction system for geometric deformation of twin bodies in tunnel construction, characterized in that: include: Association model building module: extracts semantic features of geometric deformation of tunnel excavation contour based on tunnel excavation construction process, and builds an association model of geometric deformation features of tunnel excavation contour; Key factor acquisition module: Extract deformation factors based on the correlation model, and use the grey correlation model to perform grey correlation analysis on the deformation factors to obtain the key factors of the geometric deformation of the tunnel excavation profile; Tunnel excavation profile deformation information prediction module: The deformation feature data organization model is obtained based on key factors and excavation profile deformation variables, and the tunnel excavation profile geometric deformation prediction neural network model is used to predict the tunnel excavation profile deformation information.

7. The intelligent prediction system for geometric deformation of twin bodies in tunnel construction according to claim 6 is characterized by: The specific implementation steps of the association relationship model building module are: Step 1.1, the tunnel excavation construction process is artificially decomposed into four stages of construction processes: advance support, blasting excavation, initial support and secondary lining construction, among which the tunnel excavation construction process is the blasting method construction process; Step 1.2, based on the decomposed construction procedures, the semantic features of the geometric deformation of the tunnel excavation profile are extracted, wherein the semantic features of the advance support stage include the material, length, diameter and density of the pipe shed, advance anchor and small guide tube; the semantic features of the blasting excavation stage include the type, position, number and size of the blasthole, the type of surrounding rock and the size of the heterodyne angle; the semantic features of the initial support stage include the size, material, position and density of the anchor, shotcrete and steel mesh; the semantic features of the secondary lining construction stage include the size of the arch frame, material density and the material of the concrete; Step 1.3: Based on the basic data, the interaction relationship in the geometric deformation process of the tunnel excavation profile is artificially clarified, and the semantic features within different construction processes and between different construction processes are established, as well as the hierarchical relationship, attribute relationship and semantic relationship between each semantic feature and the deformation factor. That is, the association relationship model of the geometric deformation characteristics of the tunnel excavation profile is constructed, in which the basic data includes standards and specifications, geological data and literature basis.

8. The intelligent prediction system for geometric deformation of twin bodies in tunnel construction according to claim 7 is characterized by: The specific implementation steps of the key factor acquisition module are: Step 2.1, deformation mechanism analysis based on the correlation model is performed to extract deformation factors involved in the excavation profile deformation process during tunnel excavation construction to form a deformation factor set, wherein the deformation factors include excavation environment factors, deformation process factors and deformation risk factors, wherein the deformation mechanism analysis refers to the correlation analysis, regression analysis or graph data reasoning between the deformation factors; Step 2.2: Use the grey correlation model to calculate the weight of each deformation factor in the deformation factor set, specifically: The reference sequence for deforming the tunnel excavation profile is: X0=[x0(k)|k=1,2,…,n] (1) The comparison sequence of each deformation factor in the deformation factor set is: X i =[x i (k)|k=1,2,…,n] (2) In the formula, i = 1, 2, 3, ..., m, m represents the total number of deformation factors, n represents the number of time series contained in each reference sequence or comparison sequence, i represents the i-th deformation factor, X i represents the comparison sequence of the i-th deformation factor, k represents the k-th time point in the i-th deformation factor, x i (k) represents the comparison sequence of the i-th deformation factor, and x0(k) represents the reference sequence at the k-th time point; The mean method is used to perform dimensionless processing on the reference sequence and the comparison sequence of each deformation factor. The formula is: In the formula, x′0(k) represents the averaging result of the reference sequence, represents the mean value of the reference sequence, x′ i (k) is the average result of the comparison sequence of the i-th deformation factor, is the average value of the comparison sequence of the i-th deformation factor; Calculate the absolute difference between the averaged comparison sequence and the reference sequence to obtain the difference sequence Δx i (k): Δx i (k)=|x′0(k)-x′ i (k)| (5) Based on the difference sequence, we get X0 and X i The grey correlation coefficient is: In the formula, min i min k Δx i (k) represents the minimum absolute difference of all difference sequences, max i max k Δx i (k) represents the maximum absolute difference of all difference sequences, and ρ represents the resolution coefficient; The grey correlation coefficients at each moment are concentrated into one value, that is, the average value of m grey correlation coefficients is calculated as a quantitative representation of the correlation degree between the comparison sequence of each deformation factor and the reference sequence. The formula is: In the formula, γ 0i Represents the grey relational degree of the comparison sequence of the i-th deformation factor to the reference sequence; Based on each grey correlation value, the minimum and maximum intervals of the grey correlation value distribution are determined as [min(γ), max(γ)]. After determination, they are evenly divided into 9 sub-intervals, and the indicator importance index is assigned in descending order of grey correlation degree, where the indicator importance index of the sub-interval with the highest correlation degree to the indicator importance index of the sub-interval with the lowest correlation degree are 1-9 respectively; According to the index importance index, construct the judgment matrix B, and the element b of the judgment matrix B ij It represents the ratio of the index importance index of the i-th deformation factor to the j-th deformation factor. If I i j , If I i >I j , Among them, I j Represents the index importance index of the jth deformation factor, I i Represents the index importance index of the i-th deformation factor;​ Calculate the maximum eigenvalue λ of the judgment matrix B max and its corresponding eigenvector W = [w1,w2,w i …,w n ] T , the obtained eigenvector element w i The indicator importance index I i The weight of the corresponding deformation factor i; Step 2.3: Based on the weights of all deformation factors of the tunnel excavation profile calculated in step 2.2, sort them from large to small, obtain the deformation factors corresponding to the front weights after sorting, and obtain the key factors of the geometric deformation of the tunnel excavation profile.

9. The intelligent prediction system for geometric deformation of twin bodies in tunnel construction according to claim 8, characterized in that: The specific implementation steps of the tunnel excavation profile deformation information prediction module are as follows: Step 3.1, by comparing the multi-phase point cloud data of tunnel excavation construction, the difference between the point cloud data is calculated to obtain the excavation profile deformation variable, and the prediction sample library of the geometric deformation of the tunnel excavation profile is formed together with the key factors; Step 3.2, projecting the tunnel excavation construction point cloud data in the prediction sample library onto a plane, so that the points on the plane form a one-to-one correspondence with the point cloud data, and thus obtaining a deformation feature data organization model; Step 3.3, using CNN and RNN connected in sequence to construct a neural network model for predicting geometric deformation of tunnel excavation profile; Step 3.4: The tunnel excavation contour geometric deformation prediction neural network model is used to predict the deformation feature data organization model, that is, after the deformation space features are extracted by CNN, they are connected to the RNN to extract the time features, and finally the deformation spatiotemporal features are obtained, which are the tunnel excavation contour deformation information.

10. The intelligent prediction system for geometric deformation of twin bodies in tunnel construction according to claim 9, characterized in that: The specific steps of step 3.2 are: Step 3.21, dividing the projection plane into a plurality of overlapping p×q grids according to the deformation factor, wherein p represents the number of arcs involved in predicting the tunnel section contour along the tunnel length direction, and q represents the number of points on the arc along the tunnel width direction; Step 3.22: Based on the prediction sample library, the points on the tunnel surface are projected downward onto the grid along the direction perpendicular to the ground, so that the grid stores the key factors of the excavation contour deformation variables and the geometric deformation of the tunnel excavation contour of the corresponding points on the tunnel surface. The final grid is the deformation feature data organization model.

Images