CN112287595A - Method for predicting shield tunnel wall back grouting thickness based on ground penetrating radar detection and machine learning - Google Patents

Method for predicting shield tunnel wall back grouting thickness based on ground penetrating radar detection and machine learning Download PDF

Info

Publication number
CN112287595A
CN112287595A CN202010975476.2A CN202010975476A CN112287595A CN 112287595 A CN112287595 A CN 112287595A CN 202010975476 A CN202010975476 A CN 202010975476A CN 112287595 A CN112287595 A CN 112287595A
Authority
CN
China
Prior art keywords
grouting
ground penetrating
penetrating radar
sample
shield tunnel
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010975476.2A
Other languages
Chinese (zh)
Other versions
CN112287595B (en
Inventor
谢雄耀
刘凤洲
曾里
周彪
石锦江
刘浩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongji University
Jinan Rail Transit Group Co Ltd
Original Assignee
Tongji University
Jinan Rail Transit Group Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tongji University, Jinan Rail Transit Group Co Ltd filed Critical Tongji University
Priority to CN202010975476.2A priority Critical patent/CN112287595B/en
Publication of CN112287595A publication Critical patent/CN112287595A/en
Application granted granted Critical
Publication of CN112287595B publication Critical patent/CN112287595B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21D—SHAFTS; TUNNELS; GALLERIES; LARGE UNDERGROUND CHAMBERS
    • E21D11/00—Lining tunnels, galleries or other underground cavities, e.g. large underground chambers; Linings therefor; Making such linings in situ, e.g. by assembling
    • E21D11/04—Lining with building materials
    • E21D11/10—Lining with building materials with concrete cast in situ; Shuttering also lost shutterings, e.g. made of blocks, of metal plates or other equipment adapted therefor
    • E21D11/105—Transport or application of concrete specially adapted for the lining of tunnels or galleries ; Backfilling the space between main building element and the surrounding rock, e.g. with concrete
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V3/00—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation
    • G01V3/12—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation operating with electromagnetic waves
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Geometry (AREA)
  • Structural Engineering (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Architecture (AREA)
  • Electromagnetism (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Civil Engineering (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Geology (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Mining & Mineral Resources (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Geophysics (AREA)
  • Medical Informatics (AREA)
  • Development Economics (AREA)
  • Software Systems (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Environmental & Geological Engineering (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)

Abstract

The invention discloses a method for predicting the thickness of shield tunnel postgrouting based on ground penetrating radar detection and machine learning. The method has the advantages that the method is more suitable for the actual situation of the shield tunnel wall postgrouting, and can effectively predict the ground penetrating radar image of the shield tunnel wall postgrouting, so that the possible dangerous case can be predicted in advance.

Description

Method for predicting shield tunnel wall back grouting thickness based on ground penetrating radar detection and machine learning
Technical Field
The invention relates to the field of shield tunnel backfill grouting detection, in particular to a method for predicting the thickness of shield tunnel backfill grouting based on ground penetrating radar detection and machine learning.
Background
The detection of the existing shield tunnel grouting after the wall mainly uses ground penetrating radar detection as a main part, and the thickness and the uniformity of the grouting are detected through a ground penetrating radar device. However, the image of the ground penetrating radar is not a direct image of the underground structure, and the return signal is an image formed by an electromagnetic wave signal, and it is necessary to analyze and explain the image. And (4) inspecting the tunnel site by a detector, and recording and identifying the ground penetrating radar image of the shield tunnel wall back grouting. The detection method has the advantages of low detection speed, strong dependence on experience of detection personnel and high subjectivity. Therefore, a method for rapidly, effectively and accurately detecting radar images based on machine learning is urgently needed. This is where the application needs to be focused on.
Disclosure of Invention
The invention aims to provide a method for predicting the thickness of the shield tunnel wall post-grouting based on ground penetrating radar detection and machine learning.
In order to solve the technical problems, the invention provides a method for predicting the thickness of the shield tunnel wall post-grouting based on ground penetrating radar detection and machine learning, which comprises the following steps:
s1, collecting signals and images in the shield tunnel by using a ground penetrating radar, and enabling the ground penetrating radar signals to correspond to scanning positions to form a data set;
s2, simulating a model test of ground penetrating radar shield tunnel backfill grouting, collecting ground penetrating radar images of the shield tunnel with known grouting thickness, and constructing a data set;
s3, preprocessing the data, and establishing an XGBoost grouting thickness prediction model based on the ground penetrating radar image according to the characteristics of the data set;
s4, optimizing parameters of the prediction model of the XGBoost grouting thickness;
and S5, predicting the ground penetrating radar image acquired by the ground penetrating radar in real time through the XGBoost grouting thickness prediction model after parameter optimization.
The step S1 includes the steps of:
s11, scanning and detecting lining and wall back grouting of a shield tail of the shield tunnel under construction by using a ground penetrating radar, wherein the position of the ground penetrating radar is close to a duct piece;
s12, recording the scanning position and the scanning path of the ground penetrating radar, collecting signals of the ground penetrating radar, and marking the corresponding position of each signal on the scanning path;
s13, numbering each collected radar signal from 0;
and S14, corresponding the ground penetrating radar signals to the scanning positions to form a data set.
The step S2 includes the steps of:
s21, selecting a shield tunnel segment and grouting slurry to perform a model test, and distributing the segment and the grouting slurry according to the ratio of 1: 1, manufacturing a model, placing pipe pieces in the same soil, injecting the same slurry, and placing a wave-absorbing material under the condition of boundary treatment;
s22, dividing the thickness of a grouting layer in the model into three types, namely under-grouting, normal grouting and over-grouting according to engineering ground conditions and engineering experience;
s23, performing model test by using the ground penetrating radar, and scanning the grouting layer with known thickness classification;
and S24, recording the scanning position and scanning path of the radar, collecting signals of the ground penetrating radar, marking the corresponding position of each signal on the scanning path, and simultaneously dividing the signals into under grouting, normal grouting and over grouting.
The step S3 includes the steps of:
s31, preprocessing data of radar signals collected in a model test and a shield tunnel, wherein the preprocessing steps of the data include deshifting, Butterworth band-pass filtering, moving average, F-K migration and normalization in sequence, the obtained radar signals are data between [ -1,1], and a sample set is formed;
s32, dividing a sample set into a training sample set I, a verification sample set and a test sample set in a random mode for radar data acquired by the preprocessed model test;
s33, calculating a loss function of the training sample set I in the current round according to the training sample set I:
Figure 831636DEST_PATH_IMAGE001
；
in the formula (I), the compound is shown in the specification,
Figure 387382DEST_PATH_IMAGE002
in order to be a function of the loss,
Figure 516881DEST_PATH_IMAGE003
in order to train the number of samples,
Figure 892499DEST_PATH_IMAGE004
to be aligned with
Figure 844274DEST_PATH_IMAGE005
A training sample
Figure 739680DEST_PATH_IMAGE006
The loss of (a) is reduced to (b),
Figure 56392DEST_PATH_IMAGE007
in order to train the true label value of the sample,
Figure 168573DEST_PATH_IMAGE008
is as followsiA sample is att-1The strong learner predicted value at the time of the sub-iteration,
Figure 533827DEST_PATH_IMAGE009
for the ith training sample
Figure 558108DEST_PATH_IMAGE010
In the first placetA weak learner function in sub-iterative training,
Figure 780141DEST_PATH_IMAGE011
and
Figure 458247DEST_PATH_IMAGE012
for the coefficients to be set manually, the coefficients are,
Figure 673197DEST_PATH_IMAGE013
the number of the leaf nodes is the number of the leaf nodes,
Figure 58042DEST_PATH_IMAGE014
is the first leaf node value;
s34, calculating the experience loss function of the current sample according to the training sample set I
Figure 201710DEST_PATH_IMAGE015
And based on the first order partial derivative and the second order partial derivative of the previous machine learning, summing the first order partial derivative and the second order partial derivative:
Figure 976899DEST_PATH_IMAGE016
；
in the formula (I), the compound is shown in the specification,
Figure 729960DEST_PATH_IMAGE017
in order to be a function of the loss,
Figure 234891DEST_PATH_IMAGE018
to fall into
Figure 546530DEST_PATH_IMAGE019
A set of training samples for each of the leaf nodes,
Figure 605753DEST_PATH_IMAGE020
is as follows
Figure 162505DEST_PATH_IMAGE021
Loss function of sample to
Figure 256363DEST_PATH_IMAGE021
The first partial derivative of the predicted value for each sample,
Figure 318997DEST_PATH_IMAGE022
is as follows
Figure 616248DEST_PATH_IMAGE021
Loss function of sample to
Figure 727424DEST_PATH_IMAGE021
The second partial derivative of the predicted value for each sample,
Figure 925056DEST_PATH_IMAGE023
and
Figure 299536DEST_PATH_IMAGE024
for manually set coefficients, empirical loss functionLIs part of the loss function, i.e., the remainder of the regularization is removed;
s35, the number of passing wheels istIteration of T, =1, 2.. XGboost prediction model is built.
The step S35 includes the steps of:
s351, calculating
Figure 661248DEST_PATH_IMAGE025
Empirical loss function of individual samples
Figure 323917DEST_PATH_IMAGE026
Based on
Figure 330050DEST_PATH_IMAGE027
First derivative of
Figure 921437DEST_PATH_IMAGE028
And second derivative
Figure 770444DEST_PATH_IMAGE029
Calculating the sum of the first derivatives of all samples
Figure 489002DEST_PATH_IMAGE030
Sum of second derivatives of all samples
Figure 365953DEST_PATH_IMAGE031
，
Figure 206870DEST_PATH_IMAGE032
=1，2，…，
Figure 418540DEST_PATH_IMAGE033
；
S352, based on the current node, trying to split the decision tree, and aiming at all the characteristicskCalculating a maximum score;
s353, maximum-based
Figure 3105DEST_PATH_IMAGE034
Splitting sub-trees by corresponding division characteristics and characteristic values;
s354, if it is maximum
Figure 295415DEST_PATH_IMAGE034
If 0, the current decision tree is established, and all leaf areas are calculated
Figure 182599DEST_PATH_IMAGE035
，
Figure 271778DEST_PATH_IMAGE035
Is as followstThe wheel training isjThe value of each leaf node is used to obtain the weak learner
Figure 752045DEST_PATH_IMAGE036
Updating strong learning device
Figure 977490DEST_PATH_IMAGE037
Entering the next weak learner iteration, if the maximum
Figure 97892DEST_PATH_IMAGE038
If not 0, go to step S352 to continue to attempt to split the decision tree.
The step S352 includes the steps of:
s3521, make the initial maximum score
Figure 533422DEST_PATH_IMAGE039
，
Figure 459790DEST_PATH_IMAGE040
、
Figure 415107DEST_PATH_IMAGE041
；
S3522, sample is characterized
Figure 768728DEST_PATH_IMAGE042
Arranged from small to large, are taken out in sequence
Figure 989756DEST_PATH_IMAGE043
And (3) each sample, after the current sample is placed into the left subtree, the sum of the first derivatives and the sum of the second derivatives of the left subtree and the right subtree are calculated in sequence:
Figure 595181DEST_PATH_IMAGE044
，
Figure 795218DEST_PATH_IMAGE045
；
Figure 178795DEST_PATH_IMAGE046
，
Figure 136387DEST_PATH_IMAGE047
；
in the formula (I), the compound is shown in the specification,
Figure 404557DEST_PATH_IMAGE048
it is indicated that the value is updated,
Figure 640909DEST_PATH_IMAGE049
is the sum of all sample first derivatives of the left sub-tree,
Figure 601912DEST_PATH_IMAGE050
is the sum of all sample first derivatives of the right sub-tree,
Figure 46800DEST_PATH_IMAGE051
for the sum of the first derivatives of all samples,
Figure 977715DEST_PATH_IMAGE052
is the sum of all sample first derivatives of the left sub-tree,
Figure 152345DEST_PATH_IMAGE053
is the sum of all sample second order derivatives of the right sub-tree,
Figure 628457DEST_PATH_IMAGE054
the sum of the second derivatives of all samples;
Figure 888537DEST_PATH_IMAGE055
and
Figure 921346DEST_PATH_IMAGE056
the first derivative and the second derivative of the ith sample entering the left subtree in the t round respectively;
s3523, updating the maximum score:
Figure 825848DEST_PATH_IMAGE057
。
the step S4 includes the steps of:
s41, introducing K-fold cross validation during XGboost algorithm training, estimating classifier generalization error, and evaluating classification accuracy of the XGboost algorithm under different parameters when only having training samples;
s42, adopting a grid search algorithm to search the multidimensional arrays in parallel from different growth directions, and determining the parameters of grid search.
The step S42 includes the steps of:
s421, determining the searching range of the parameter to be searched according to experience;
s422, setting the search step length of the searched parameters;
s423, calculating the classification accuracy of the XGboost according to a cross validation method for each group of values on the grid;
and S424, giving the classification accuracy of each group, and determining the optimal parameter value according to the classification accuracy.
The step S5 includes the steps of:
s51, preprocessing the ground penetrating radar data acquired in S1, wherein the preprocessing steps of the data include deshifting, Butterworth band-pass filtering, moving average, F-K migration and normalization in sequence, and the obtained radar signals are data between [ -1,1 ];
and S52, inputting the data into the established XGBoost prediction model to obtain the predicted value of the grouting layer thickness classification.
The invention has the following advantages: the machine learning prediction method applied to the field of shield tunnel wall post-grouting detection is used for rapidly, effectively and accurately identifying the ground penetrating radar image by utilizing the XGboost principle; compared with the prior art, the prediction method disclosed by the invention is more suitable for the actual situation of the shield tunnel wall postgrouting, and can effectively predict the ground penetrating radar image of the shield tunnel wall postgrouting, so that the possible dangerous situation can be predicted in advance, precious time is provided for timely supplementing grouting, the problem that the data of the ground penetrating radar detection wall postgrouting is difficult to explain is effectively solved, and real-time feedback and guidance are provided for grouting construction.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a flow chart of the present invention S1-S5;
FIG. 2 is a flow of the XGboost model in S3 according to the present invention;
FIG. 3 is a diagram illustrating the grid parameter search in S4 according to the present invention;
fig. 4 is a flowchart of the grid parameter search in S4 according to the present invention.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
FIG. 1 shows a flow chart of an embodiment of the present invention. As shown in fig. 1-4, the invention provides a method for predicting the thickness of the shield tunnel wall post-grouting based on ground penetrating radar detection and machine learning, which comprises the following steps:
s1, collecting signals and images by using a ground penetrating radar in the shield tunnel, and corresponding the ground penetrating radar signals with the scanning position to form a data set, wherein the method specifically comprises the following steps:
s11, scanning and detecting lining and wall back grouting of a shield tail of the shield tunnel under construction by using a ground penetrating radar, wherein the position of the ground penetrating radar is close to a duct piece;
s12, recording the scanning position and the scanning path of the ground penetrating radar, collecting signals of the ground penetrating radar, and marking the corresponding position of each signal on the scanning path;
s13, numbering each collected radar signal from 0;
and S14, corresponding the ground penetrating radar signals to the scanning positions to form a data set.
S2, simulating a model test of ground penetrating radar shield tunnel backfill grouting, collecting ground penetrating radar images of a shield tunnel with known grouting thickness, and constructing a data set, wherein the method specifically comprises the following steps:
s21, selecting a shield tunnel segment and grouting slurry to perform a model test, and distributing the segment and the grouting slurry according to the ratio of 1: 1, manufacturing a model, placing pipe pieces in the same soil, injecting the same slurry, and placing a wave-absorbing material under the condition of boundary treatment;
s22, dividing the thickness of a grouting layer in the model into three types, namely under-grouting, normal grouting and over-grouting according to engineering ground conditions and engineering experience;
s23, performing model test by using the ground penetrating radar, and scanning the grouting layer with known thickness classification;
and S24, recording the scanning position and scanning path of the radar, collecting signals of the ground penetrating radar, marking the corresponding position of each signal on the scanning path, and simultaneously dividing the signals into under grouting, normal grouting and over grouting.
S3, preprocessing the data, and establishing a prediction model of XGBoost grouting thickness based on the ground penetrating radar image according to the characteristics of a data set, as shown in FIG. 2, the method specifically comprises the following steps:
s31, preprocessing data of radar signals collected in a model test and a shield tunnel, wherein the preprocessing steps of the data include deshifting, Butterworth band-pass filtering, moving average, F-K migration and normalization in sequence (the obtained radar signals are data between [ -1,1], and a sample set is formed;
s32, dividing a sample set into a training sample set I, a verification sample set and a test sample set in a random mode for radar data acquired by the preprocessed model test;
s33, calculating a loss function of the training sample set I in the current round according to the training sample set I:
Figure 863074DEST_PATH_IMAGE058
；
in the formula (I), the compound is shown in the specification,
Figure 469505DEST_PATH_IMAGE059
in order to be a function of the loss,
Figure 555273DEST_PATH_IMAGE060
in order to train the number of samples,
Figure 438915DEST_PATH_IMAGE061
to be aligned with
Figure 270211DEST_PATH_IMAGE062
A training sample
Figure 239304DEST_PATH_IMAGE063
The loss of (a) is reduced to (b),
Figure 863184DEST_PATH_IMAGE064
in order to train the true label value of the sample,
Figure 991546DEST_PATH_IMAGE065
is as followsiA sample is att-1The strong learner predicted value at the time of the sub-iteration,
Figure 636154DEST_PATH_IMAGE066
for the ith training sample
Figure 702330DEST_PATH_IMAGE067
In the first placetA weak learner function in sub-iterative training,
Figure 457796DEST_PATH_IMAGE068
and
Figure 473288DEST_PATH_IMAGE069
for the coefficients to be set manually, the coefficients are,
Figure 164163DEST_PATH_IMAGE070
the number of the leaf nodes is the number of the leaf nodes,
Figure 842269DEST_PATH_IMAGE071
is as follows
Figure 526060DEST_PATH_IMAGE072
A number of leaf node values;
s34, calculating the experience loss function of the current sample according to the training sample set I
Figure 910905DEST_PATH_IMAGE073
Based on previous mechanicsThe first order partial derivative and the second order partial derivative are learned, and the sum of the first order partial derivative and the second order partial derivative is calculated as follows:
Figure 631737DEST_PATH_IMAGE074
；
in the formula (I), the compound is shown in the specification,
Figure 420308DEST_PATH_IMAGE075
in order to be a function of the loss,
Figure 189681DEST_PATH_IMAGE076
to fall into
Figure 22507DEST_PATH_IMAGE077
A set of training samples for each of the leaf nodes,
Figure 773295DEST_PATH_IMAGE078
is as follows
Figure 691572DEST_PATH_IMAGE079
Loss function of sample to
Figure 264636DEST_PATH_IMAGE079
The first partial derivative of the predicted value for each sample,
Figure 312488DEST_PATH_IMAGE080
is as follows
Figure 375122DEST_PATH_IMAGE079
Loss function of sample to
Figure 390483DEST_PATH_IMAGE079
The second partial derivative of the predicted value for each sample,
Figure 95134DEST_PATH_IMAGE081
and
Figure 558345DEST_PATH_IMAGE082
for manually set coefficients, empirical loss functionLIs part of the loss function, i.e., the remainder of the regularization is removed;
s35, the number of passing wheels istIteration of T, =1, 2.. XGboost prediction model is built, including the steps of:
s351, calculating
Figure 667246DEST_PATH_IMAGE083
Empirical loss function of individual samples
Figure 294537DEST_PATH_IMAGE084
Based on
Figure 957206DEST_PATH_IMAGE085
First derivative of
Figure 963339DEST_PATH_IMAGE086
And second derivative
Figure 633355DEST_PATH_IMAGE087
Calculating the sum of the first derivatives of all samples
Figure 669313DEST_PATH_IMAGE088
Sum of second derivatives of all samples
Figure 59974DEST_PATH_IMAGE089
，=1，2，…，
Figure 999242DEST_PATH_IMAGE090
；
S352, based on the current node, trying to split the decision tree, and aiming at all the characteristicskCalculating the maximum score, specifically comprising the following steps:
s3521, make the initial maximum score
Figure 574580DEST_PATH_IMAGE091
，
Figure 848567DEST_PATH_IMAGE092
、
Figure 88924DEST_PATH_IMAGE093
；
S3522, sample is characterized
Figure 69650DEST_PATH_IMAGE042
Arranged from small to large, are taken out in sequence
Figure 513093DEST_PATH_IMAGE094
And (3) each sample, after the current sample is placed into the left subtree, the sum of the first derivatives and the sum of the second derivatives of the left subtree and the right subtree are calculated in sequence:
Figure 71114DEST_PATH_IMAGE095
，
Figure 131474DEST_PATH_IMAGE096
；
Figure 215973DEST_PATH_IMAGE097
，
Figure 70797DEST_PATH_IMAGE098
；
in the formula (I), the compound is shown in the specification,
Figure 647271DEST_PATH_IMAGE099
it is indicated that the value is updated,
Figure 730896DEST_PATH_IMAGE100
is the sum of all sample first derivatives of the left sub-tree,
Figure 420635DEST_PATH_IMAGE101
is the sum of all sample first derivatives of the right sub-tree,
Figure 226786DEST_PATH_IMAGE102
for the sum of the first derivatives of all samples,
Figure 900343DEST_PATH_IMAGE103
is a derivative of all samples of the left sub-treeThe sum of the numbers is then calculated,
Figure 50309DEST_PATH_IMAGE104
is the sum of all sample second order derivatives of the right sub-tree,
Figure 719187DEST_PATH_IMAGE105
the sum of the second derivatives of all samples;
Figure 181393DEST_PATH_IMAGE106
and
Figure 529198DEST_PATH_IMAGE107
the first derivative and the second derivative of the ith sample entering the left subtree in the t round respectively;
s3523, updating the maximum score:
Figure 531789DEST_PATH_IMAGE108
；
s353, maximum-based
Figure 9169DEST_PATH_IMAGE109
Splitting sub-trees by corresponding division characteristics and characteristic values;
s354, if it is maximum
Figure 845538DEST_PATH_IMAGE109
If 0, the current decision tree is established, and all leaf areas are calculated
Figure 87163DEST_PATH_IMAGE110
，
Figure 80396DEST_PATH_IMAGE110
Is as followstThe wheel training isjThe value of each leaf node is used to obtain the weak learner
Figure 864812DEST_PATH_IMAGE111
Updating strong learning device
Figure 151044DEST_PATH_IMAGE112
Entering the next weak learner iteration, if the maximum
Figure 145544DEST_PATH_IMAGE113
If not 0, go to step S352 to continue to attempt to split the decision tree.
S4, optimizing the parameters of the XGboost grouting thickness prediction model, as shown in fig. 3 and 4, specifically including the following steps:
s41, introducing K-fold cross validation during XGboost algorithm training, estimating classifier generalization error, and evaluating classification accuracy of the XGboost algorithm under different parameters when only having training samples;
s42, adopting a grid search algorithm to search multidimensional arrays from different growth directions in parallel, and determining grid search parameters, comprising the following steps:
s421, determining the searching range of the parameter to be searched according to experience;
s422, setting the search step length of the searched parameters;
s423, calculating the classification accuracy of the XGboost according to a cross validation method for each group of values on the grid;
and S424, giving the classification accuracy of each group, and determining the optimal parameter value according to the classification accuracy.
S5, predicting a ground penetrating radar image acquired by a ground penetrating radar in real time through the XGBoost grouting thickness prediction model after parameter optimization, and specifically comprising the following steps:
s51, preprocessing the ground penetrating radar data acquired in S1, wherein the preprocessing steps of the data include deshifting, Butterworth band-pass filtering, moving average, F-K migration and normalization in sequence, and the obtained radar signals are data between [ -1,1 ];
and S52, inputting the data into the established XGBoost prediction model to obtain a predicted value of the grouting layer thickness classification, namely, the final output classification label in the figure 1.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (9)

1. A method for predicting the thickness of the shield tunnel backfill grouting based on ground penetrating radar detection and machine learning comprises the following steps:
s1, collecting signals and images in the shield tunnel by using a ground penetrating radar, and enabling the ground penetrating radar signals to correspond to scanning positions to form a data set;
s2, simulating a model test of ground penetrating radar shield tunnel backfill grouting, collecting ground penetrating radar images of the shield tunnel with known grouting thickness, and constructing a data set;
s3, preprocessing the data, and establishing an XGBoost grouting thickness prediction model based on the ground penetrating radar image according to the characteristics of the data set;
s4, optimizing parameters of the prediction model of the XGBoost grouting thickness;
and S5, predicting the ground penetrating radar image acquired by the ground penetrating radar in real time through the XGBoost grouting thickness prediction model after parameter optimization.
2. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 1, is characterized in that: the step S1 includes the steps of:
s11, scanning and detecting lining and wall back grouting of a shield tail of the shield tunnel under construction by using a ground penetrating radar, wherein the position of the ground penetrating radar is close to a duct piece;
s12, recording the scanning position and the scanning path of the ground penetrating radar, collecting signals of the ground penetrating radar, and marking the corresponding position of each signal on the scanning path;
s13, numbering each collected radar signal from 0;
and S14, corresponding the ground penetrating radar signals to the scanning positions to form a data set.
3. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 1, is characterized in that: the step S2 includes the steps of:
s21, selecting a shield tunnel segment and grouting slurry to perform a model test, and distributing the segment and the grouting slurry according to the ratio of 1: 1, manufacturing a model, placing pipe pieces in the same soil, injecting the same slurry, and placing a wave-absorbing material under the condition of boundary treatment;
s22, dividing the thickness of a grouting layer in the model into three types, namely under-grouting, normal grouting and over-grouting according to engineering ground conditions and engineering experience;
s23, performing model test by using the ground penetrating radar, and scanning the grouting layer with known thickness classification;
and S24, recording the scanning position and scanning path of the radar, collecting signals of the ground penetrating radar, marking the corresponding position of each signal on the scanning path, and simultaneously dividing the signals into under grouting, normal grouting and over grouting.
4. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 1, is characterized in that: the step S3 includes the steps of:
s31, preprocessing data of radar signals collected in a model test and a shield tunnel, wherein the preprocessing steps of the data include deshifting, Butterworth band-pass filtering, moving average, F-K migration and normalization in sequence, the obtained radar signals are data between [ -1,1], and a sample set is formed;
s32, dividing a sample set into a training sample set I, a verification sample set and a test sample set in a random mode for radar data acquired by the preprocessed model test;
s33, calculating a loss function of the training sample set I in the current round according to the training sample set I:
Figure 71740DEST_PATH_IMAGE001
；
in the formula (I), the compound is shown in the specification,
Figure 507401DEST_PATH_IMAGE002
in order to be a function of the loss,
Figure 967463DEST_PATH_IMAGE003
in order to train the number of samples,
Figure 121364DEST_PATH_IMAGE004
to be aligned with
Figure 190820DEST_PATH_IMAGE005
A training sample
Figure 808883DEST_PATH_IMAGE006
The loss of (a) is reduced to (b),
Figure 954694DEST_PATH_IMAGE007
in order to train the true label value of the sample,
Figure 78114DEST_PATH_IMAGE008
is as followsiA sample is att-1The strong learner predicted value at the time of the sub-iteration,
Figure 639677DEST_PATH_IMAGE009
for the ith training sample
Figure 564776DEST_PATH_IMAGE010
In the first placetA weak learner function in sub-iterative training,
Figure 678226DEST_PATH_IMAGE011
and
Figure 806719DEST_PATH_IMAGE012
for the coefficients to be set manually, the coefficients are,
Figure 719442DEST_PATH_IMAGE013
the number of the leaf nodes is the number of the leaf nodes,
Figure 187464DEST_PATH_IMAGE014
is as follows
Figure 658765DEST_PATH_IMAGE015
A number of leaf node values;
s34, calculating the experience loss function of the current sample according to the training sample set I
Figure 8975DEST_PATH_IMAGE016
And based on the first order partial derivative and the second order partial derivative of the previous machine learning, summing the first order partial derivative and the second order partial derivative:
Figure 734179DEST_PATH_IMAGE017
；
in the formula (I), the compound is shown in the specification,
Figure 853445DEST_PATH_IMAGE018
in order to be a function of the loss,
Figure 574276DEST_PATH_IMAGE019
to fall into
Figure 661049DEST_PATH_IMAGE020
A set of training samples for each of the leaf nodes,
Figure 164843DEST_PATH_IMAGE021
is as follows
Figure 154927DEST_PATH_IMAGE022
Loss function of sample to
Figure 718763DEST_PATH_IMAGE022
The first partial derivative of the predicted value for each sample,
Figure 496095DEST_PATH_IMAGE023
is as follows
Figure 803580DEST_PATH_IMAGE022
Loss function of sample to
Figure 645241DEST_PATH_IMAGE022
The second partial derivative of the predicted value for each sample,
Figure 379979DEST_PATH_IMAGE024
and
Figure 519973DEST_PATH_IMAGE025
for manually set coefficients, empirical loss functionLIs part of the loss function, i.e., the remainder of the regularization is removed;
s35, the number of passing wheels istIteration of T, =1, 2.. XGboost prediction model is built.
5. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 4, is characterized in that: the step S35 includes the steps of:
s351, calculating
Figure 880416DEST_PATH_IMAGE026
Empirical loss function of individual samples
Figure 828780DEST_PATH_IMAGE027
Based on
Figure 688414DEST_PATH_IMAGE028
First derivative of
Figure 987809DEST_PATH_IMAGE029
And second derivative
Figure 620784DEST_PATH_IMAGE030
Calculating the sum of the first derivatives of all samples
Figure 17130DEST_PATH_IMAGE031
Sum of second derivatives of all samples
Figure 93671DEST_PATH_IMAGE032
，=1，2，…，
Figure 628164DEST_PATH_IMAGE033
；
S352, based on the current node, trying to split the decision tree, and aiming at all the characteristicskCalculating a maximum score;
s353, maximum-based
Figure 549984DEST_PATH_IMAGE034
Splitting sub-trees by corresponding division characteristics and characteristic values;
s354, if it is maximum
Figure 128732DEST_PATH_IMAGE034
If 0, the current decision tree is established, and all leaf areas are calculated
Figure 969649DEST_PATH_IMAGE035
，
Figure 994368DEST_PATH_IMAGE035
Is as followstThe wheel training isjThe value of each leaf node is used to obtain the weak learner
Figure 719879DEST_PATH_IMAGE036
Updating strong learning device
Figure 12189DEST_PATH_IMAGE037
Entering the next weak learner iteration, if the maximum
Figure 633794DEST_PATH_IMAGE038
If not 0, go to step S352 to continue to attempt to split the decision tree.
6. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 5, is characterized in that: the step S352 includes the steps of:
s3521, make the initial maximum score
Figure 877300DEST_PATH_IMAGE039
，
Figure 672081DEST_PATH_IMAGE040
、
Figure 631947DEST_PATH_IMAGE041
；
S3522, sample is characterized
Figure 470459DEST_PATH_IMAGE042
Arranged from small to large, are taken out in sequence
Figure 719037DEST_PATH_IMAGE043
And (3) each sample, after the current sample is placed into the left subtree, the sum of the first derivatives and the sum of the second derivatives of the left subtree and the right subtree are calculated in sequence:
Figure 802662DEST_PATH_IMAGE044
，
Figure 289138DEST_PATH_IMAGE045
；
Figure 767393DEST_PATH_IMAGE046
，
Figure 237689DEST_PATH_IMAGE047
；
in the formula (I), the compound is shown in the specification,
Figure 436589DEST_PATH_IMAGE048
it is indicated that the value is updated,
Figure 68251DEST_PATH_IMAGE049
is the sum of all sample first derivatives of the left sub-tree,
Figure 264877DEST_PATH_IMAGE050
is the sum of all sample first derivatives of the right sub-tree,
Figure 940578DEST_PATH_IMAGE051
for the sum of the first derivatives of all samples,
Figure 84115DEST_PATH_IMAGE052
is the sum of all sample first derivatives of the left sub-tree,
Figure 561495DEST_PATH_IMAGE053
is the sum of all sample second order derivatives of the right sub-tree,
Figure 663443DEST_PATH_IMAGE054
the sum of the second derivatives of all samples;
Figure 170648DEST_PATH_IMAGE055
and
Figure 898301DEST_PATH_IMAGE056
the first derivative and the second derivative of the ith sample entering the left subtree in the t round respectively;
s3523, updating the maximum score:
Figure 213876DEST_PATH_IMAGE057
。
7. the method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 1, is characterized in that: the step S4 includes the steps of:
s41, introducing K-fold cross validation during XGboost algorithm training, estimating classifier generalization error, and evaluating classification accuracy of the XGboost algorithm under different parameters when only having training samples;
s42, adopting a grid search algorithm to search the multidimensional arrays in parallel from different growth directions, and determining the parameters of grid search.
8. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 7, is characterized in that: the step S42 includes the steps of:
s421, determining the searching range of the parameter to be searched according to experience;
s422, setting the search step length of the searched parameters;
s423, calculating the classification accuracy of the XGboost according to a cross validation method for each group of values on the grid;
and S424, giving the classification accuracy of each group, and determining the optimal parameter value according to the classification accuracy.
9. The method for predicting the thickness of the shield tunnel wall post-grouting based on the ground penetrating radar detection and the machine learning, according to claim 1 or 4, is characterized in that: the step S5 includes the steps of:
s51, preprocessing the ground penetrating radar data acquired in S1, wherein the preprocessing comprises the steps of drift removal, Butterworth band-pass filtering, moving average, F-K offset and normalization, and the obtained radar signals are data between [ -1,1 ];
and S52, inputting the data into the established XGBoost prediction model to obtain the predicted value of the grouting layer thickness classification.
CN202010975476.2A 2020-09-16 2020-09-16 Prediction method of post-grouting thickness of shield tunnel wall based on ground penetrating radar detection and machine learning Active CN112287595B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010975476.2A CN112287595B (en) 2020-09-16 2020-09-16 Prediction method of post-grouting thickness of shield tunnel wall based on ground penetrating radar detection and machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010975476.2A CN112287595B (en) 2020-09-16 2020-09-16 Prediction method of post-grouting thickness of shield tunnel wall based on ground penetrating radar detection and machine learning

Publications (2)

Publication Number Publication Date
CN112287595A true CN112287595A (en) 2021-01-29
CN112287595B CN112287595B (en) 2023-05-23

Family

ID=74419965

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010975476.2A Active CN112287595B (en) 2020-09-16 2020-09-16 Prediction method of post-grouting thickness of shield tunnel wall based on ground penetrating radar detection and machine learning

Country Status (1)

Country Link
CN (1) CN112287595B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113326661A (en) * 2021-06-17 2021-08-31 广西路桥工程集团有限公司 RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device
CN113326660A (en) * 2021-06-17 2021-08-31 广西路桥工程集团有限公司 Tunnel surrounding rock extrusion deformation prediction method based on GA-XGboost model

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1975112A (en) * 2006-12-14 2007-06-06 同济大学 Shield tunnel subsidence control method based on exploring radar
CN111025286A (en) * 2019-11-01 2020-04-17 长安大学 Ground penetrating radar map self-adaptive selection method for water damage detection
CN111144636A (en) * 2019-12-22 2020-05-12 同济大学 Slope deformation prediction method
CN111365015A (en) * 2020-03-05 2020-07-03 中建交通建设集团有限公司 Shield tunneling parameter feature extraction and attitude deviation prediction method based on XGboost

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1975112A (en) * 2006-12-14 2007-06-06 同济大学 Shield tunnel subsidence control method based on exploring radar
CN111025286A (en) * 2019-11-01 2020-04-17 长安大学 Ground penetrating radar map self-adaptive selection method for water damage detection
CN111144636A (en) * 2019-12-22 2020-05-12 同济大学 Slope deformation prediction method
CN111365015A (en) * 2020-03-05 2020-07-03 中建交通建设集团有限公司 Shield tunneling parameter feature extraction and attitude deviation prediction method based on XGboost

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
XIE XIONGYAO EL.: ""Real-time detection technology of synchronous grouting for shield tunnel and analysis of grouting effect"", 《2018 17TH INTERNATIONAL CONFERENCE ON GROUND PENETRATING RADAR (GPR)》 *
王鹏: ""基于机器学习的地铁隧道施工扰动控制研究"", 《现代隧道技术》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113326661A (en) * 2021-06-17 2021-08-31 广西路桥工程集团有限公司 RS-XGboost-based tunnel advanced drilling quantitative interpretation method and device
CN113326660A (en) * 2021-06-17 2021-08-31 广西路桥工程集团有限公司 Tunnel surrounding rock extrusion deformation prediction method based on GA-XGboost model
CN113326660B (en) * 2021-06-17 2022-11-29 广西路桥工程集团有限公司 Tunnel surrounding rock extrusion deformation prediction method based on GA-XGboost model

Also Published As

Publication number Publication date
CN112287595B (en) 2023-05-23

Similar Documents

Publication Publication Date Title
KR102008973B1 (en) Apparatus and Method for Detection defect of sewer pipe based on Deep Learning
CN110751099B (en) Unmanned aerial vehicle aerial video track high-precision extraction method based on deep learning
CN108020211A (en) A kind of method of unmanned plane aeroplane photography estimation instruction plant biomass
CN114548278A (en) In-service tunnel lining structure defect identification method and system based on deep learning
CN112287595A (en) Method for predicting shield tunnel wall back grouting thickness based on ground penetrating radar detection and machine learning
JP2011179874A (en) Method for extracting surface cracks of paved road and evaluating surface damage level of paved road
CN110472661A (en) Method for detecting automatic variation and system based on history background and current remote sensing image
CN112712169A (en) Model building method and application of full residual depth network based on graph convolution
CN116229380A (en) Method for identifying bird species related to bird-related faults of transformer substation
JP2020056254A (en) Geological evaluation system, geological evaluation method, and geological evaluation program
CN117741808A (en) Geological crack detection method based on geological exploration
Fakhri et al. Road crack detection using gaussian/prewitt filter
CN117574493B (en) Highway frozen soil range deformation identification method and system in permafrost region
CN114324580A (en) Intelligent knocking detection method and system for structural defects
US20240102833A1 (en) Weakly-supervised learning for manhole localization based on ambient noise
CN117496463A (en) Learning method and system for improving road perception precision
CN117351321A (en) Single-stage lightweight subway lining cavity recognition method and related equipment
CN114970346A (en) Ground penetrating radar and GAN-based shield tunnel wall back medium distribution prediction method
CN113077458B (en) Cloud and shadow detection method and system in remote sensing image
Ramachandraiah et al. Evaluation of Pavement Surface Distress Using Image Processing and Artificial Neural Network
Tundia et al. Aiding Irrigation Census in Developing Countries by Detecting Minor Irrigation Structures from Satellite Imagery.
CN116597296B (en) Wheat seedling density estimation method based on deep learning and thermodynamic diagram
CN118015198B (en) Pipe gallery fire risk assessment method based on convolutional neural network image processing
Wang et al. A Multistream Attention Network for Airport Runway Subsurface Target Segmentation
CN113963334A (en) Vehicle track defect data repairing method facing roadside sensing unit

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant