CN114742325A - Method and system for predicting land surface settlement during subway tunnel step method construction - Google Patents

Method and system for predicting land surface settlement during subway tunnel step method construction Download PDF

Info

Publication number
CN114742325A
CN114742325A CN202210581122.9A CN202210581122A CN114742325A CN 114742325 A CN114742325 A CN 114742325A CN 202210581122 A CN202210581122 A CN 202210581122A CN 114742325 A CN114742325 A CN 114742325A
Authority
CN
China
Prior art keywords
data
settlement
prediction
original
residual error
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.)
Pending
Application number
CN202210581122.9A
Other languages
Chinese (zh)
Inventor
李涛
杨腾宇
陈前
赵晶
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.)
China University of Mining and Technology Beijing CUMTB
Original Assignee
China University of Mining and Technology Beijing CUMTB
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 China University of Mining and Technology Beijing CUMTB filed Critical China University of Mining and Technology Beijing CUMTB
Priority to CN202210581122.9A priority Critical patent/CN114742325A/en
Publication of CN114742325A publication Critical patent/CN114742325A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/08—Construction
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A30/00—Adapting or protecting infrastructure or their operation
    • Y02A30/60—Planning or developing urban green infrastructure

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Mathematical Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Pure & Applied Mathematics (AREA)
  • Marketing (AREA)
  • Mathematical Optimization (AREA)
  • Computational Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Fuzzy Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Algebra (AREA)
  • Geophysics And Detection Of Objects (AREA)

Abstract

The application discloses a method and a system for predicting the land surface settlement in subway tunnel step method construction, wherein the method comprises the following steps: acquiring original ground surface settlement sequence data; processing the original settlement data; performing VMD decomposition on the processed original earth surface sedimentation sequence data to obtain subsequences of different central modes and a residual error, wherein the subsequences are sequentially arranged from low central frequency to high central frequency; setting and optimizing hyper-parameters of the GRU neural network according to the sub-sequences and the residual errors; training the optimized GRU neural network, and then inputting each subsequence and the residual error to obtain a prediction component of each subsequence and a prediction component of the residual error; and superposing the prediction component results to obtain a ground surface settlement prediction value. The ground surface settlement with high volatility and randomness has good prediction precision, and guidance is provided for subsequent subway tunnel step method construction.

Description

Method and system for predicting land surface settlement during subway tunnel step method construction
Technical Field
The application relates to the technical field of surface subsidence prediction, in particular to a method and a system for predicting surface subsidence in subway tunnel step method construction.
Background
At present, urban underground space engineering in China is rapidly developed, and the influence of surface subsidence caused by underground engineering construction on the surrounding environment is more and more obvious. Therefore, the prediction of the ground surface settlement rule caused by urban tunnel construction becomes an important problem of underground engineering.
At present, the main surface subsidence prediction methods include two types, physical model methods and statistical modeling methods. One type of physical model method is to analyze physical factors causing settlement, establish a physical mechanical model and calculate a surface settlement value to predict, and a statistical modeling method is to analyze and interpret the internal relation and development rule of a large amount of historical monitoring data by using a mathematical statistical analysis method to predict the settlement value. Aiming at the uncertainty of the surface subsidence prediction, the method is not ideal in prediction precision, along with the rising of a neural network, the neural network such as back propagation, generalized regression, long-term and short-term memory and the like also shows extremely high fitting degree to complex data, but the method is widely applied to the prior art, but the traditional neural network has the problem of gradient explosion or disappearance, and the calculation efficiency is low due to the complex structure.
The step method construction of the subway tunnel is a long-term process, the settlement period is long, the influence factors are many, the settlement curve is complex, and more fluctuation sections are provided, if the complex original data are directly trained and predicted, the problems of slow convergence, poor fitting of the data fluctuation sections, prediction distortion and the like are easy to occur, and if the original data are smoothed or denoised, the information carried by the data is easy to lose, so that the practicability of the prediction result is reduced.
At present, time series prediction is carried out by combining empirical mode decomposition with a traditional neural network, but the components lose the characteristics of a single characteristic scale due to the mode aliasing problem of the empirical mode decomposition, and the accuracy of a prediction result is low due to the difficulty of characteristic extraction, model training and mode recognition.
Disclosure of Invention
The application provides a method and a system for predicting surface subsidence in subway tunnel step method construction.
In order to achieve the purpose, the application provides a method for predicting the surface subsidence of subway tunnel construction by a step method, which comprises the following steps:
acquiring original ground surface settlement sequence data;
processing the original settlement data;
performing VMD decomposition on the processed original earth surface settlement sequence data to obtain subsequences of different central modes and a residual error, wherein each subsequence is sequentially arranged from low to high according to the central frequency;
setting and optimizing hyper-parameters of the GRU neural network according to the sub-sequences and the residual errors;
training the optimized GRU neural network, and then inputting each subsequence and the residual error to obtain a prediction component of each subsequence and a prediction component of the residual error; and superposing the prediction component results to obtain a ground surface settlement prediction value.
Optionally, the method for acquiring the data of the original surface subsidence sequence includes: on the basis of an elevation control network, the vertical displacement change of each measuring point on the ground around the subway is monitored to obtain the vertical displacement change.
Optionally, the method for processing the original settlement data includes: complementing interval period data in the original earth surface settlement sequence data by adopting the mean value of the data at two ends, and enabling the frequency of the whole settlement data time sequence to be 1 Hz.
Optionally, the method for performing VMD decomposition on the raw data of surface subsidence data includes: performing trial decomposition on the data according to K numbers respectively to obtain central frequency distribution graphs corresponding to different K numbers, and obtaining the optimal eigenmode component number K based on the central frequency distribution graphs; and decomposing the original settlement data by combining the optimal eigenmode component number K to obtain K subsequences with different central modes and a residual error.
Optionally, the GRU neural network includes an update gate, a reset gate and an output gate;
the GRU neural network combines a forgetting gate and an input gate in the LSTM network into an updating gate, retains the original resetting gate, and learns the advantages of the LSTM gating network to update the cell state and the hidden state; the updating gate is used for describing the influence degree of the characteristic information at the past moment on the current characteristic information, and the larger the threshold value of the updating gate is, the larger the influence of the characteristic information at the previous moment on the current moment is; the reset gate is used to describe the degree to which the status characteristic information at the past time is discarded, and the smaller the threshold value, the more the past information is discarded.
Optionally, the method for training each subsequence and the residual error through the GRU neural network includes:
forward propagation, sequentially carrying out weight operation on each input subsequence along a sequence propagation direction to output a target vector;
comparing with the test set and outputting an error through a target function;
the weight is changed by back propagation so as to extract the time series characteristics;
the forward propagation mathematical operation process is as follows:
Figure BDA0003663768200000041
in the formula: h ist-1And htFor hidden layer vectors at different times, XtAs input data, U, w is a weight parameter, is a gated activation function sigmoid, has an output value of 0 to 1, tan h as an activation function in generating a candidate memory, is a sign of Hadamard product, zt、rt、ht' update gate, reset gate, candidate state, respectively.
The application also provides a subway tunnel step method construction earth surface settlement prediction system which comprises a data collection module, a data processing module, a decomposition module, an optimization module and a prediction module;
the data collection module is used for acquiring original ground surface settlement sequence data;
the data processing module is used for processing the original settlement data;
the decomposition module is used for performing VMD decomposition on the processed original earth surface settlement sequence data to obtain subsequences of different central modes and a residual error, and the subsequences are sequentially arranged from low to high according to the central frequency;
the optimization module is used for setting and optimizing the hyper-parameters of the GRU neural network according to the sub-sequences and the residual errors;
the prediction module is used for training the optimized GRU neural network, and then inputting each subsequence and the residual error to obtain a prediction component of each subsequence and a prediction component of the residual error; the prediction module is further used for superposing the prediction component results to obtain an earth surface settlement prediction value.
Preferably, the workflow of the data processing module includes: and complementing interval period data in the acquired original earth surface settlement sequence data by adopting the mean value of the data at two ends, so that the frequency of the whole settlement data time sequence is 1 Hz.
Compared with the prior art, the beneficial effects of this application are as follows:
according to the method, the prediction results are obtained by superposing a plurality of prediction components obtained by the GUR network, and the prediction precision can be effectively improved compared with the traditional surface subsidence prediction method; the problem that the single characteristic scale of the component is lost due to mode aliasing existing in empirical mode decomposition is solved effectively.
Drawings
In order to more clearly illustrate the technical solution of the present application, the drawings needed to be used in the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for a person skilled in the art to obtain other drawings without any inventive exercise.
FIG. 1 is a schematic flow chart of the method in this embodiment;
FIG. 2 is a schematic diagram of the measured sedimentation curve in this embodiment;
FIG. 3 is a diagram illustrating the decomposition result of the VMD in the present embodiment;
FIG. 4 is a graph showing a comparison of the predicted settling curves of the GRU and VMD-GRU models in this example;
fig. 5 is a schematic structural diagram of a system for predicting surface subsidence in subway tunnel step method construction.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, the present application is described in further detail with reference to the accompanying drawings and the detailed description.
Example one
As shown in fig. 1, which is a schematic diagram of the technical process of the method of the embodiment, taking the data of surface subsidence constructed by the step method of the west subway tunnel with the line No. 11 in beijing as an example, the tunnel surrounding rock mainly comprises clay and clay silty soil, and has more interlaminar diving, more complex geological conditions and variable surface subsidence rules.
The data are collected from a No. 17 section DB1-17 point, and the monitoring time is from 6/month and 4 days in 2020 to 2/month and 11 days in 2021. For interval data, the mean value of the data at both ends is adopted to complement the interval data, so that the frequency of the whole sedimentation data time sequence is 1Hz, and 248 data sedimentation curves are shown in figure 2.
When the variation mode number is 6, the smaller the center frequency overlapping is, the more thorough the time series decomposition is, and the optimal mode number is obtained. And performing VMD decomposition on the surface subsidence time sequence to obtain 6 subsequences with different central modes and a residual error. The VMD decomposition method comprises the following steps:
s1, assuming a decomposed modal function ku(t), obtaining an analytic signal thereof by Hilbert transformation, and then converting the analytic signal into a single-side frequency spectrum:
Figure BDA0003663768200000061
wherein δ (t) is a unit impulse function, δ (t) is infinite when t is equal to 0, δ (t) is 0 when t is not equal to, and j is an imaginary unit;
s2, the frequency spectrum and the central frequency omega are further combinedkIndex term e-jωk tAliasing causes the signal to be translated to a baseband, and an expression of an analytic signal is obtained as follows:
Figure BDA0003663768200000071
s3, in order to ensure that when the original sequence is decomposed into K modal components, the sum of the estimated bandwidths of the modal components with the center frequency is minimum, and the sum of the modal components is equal to the original sequence, a constraint variation problem is obtained, and the expression is as follows:
Figure BDA0003663768200000072
in the formula: u. ofkAs decomposed K mode functions, ωkFor each of K mode functionsThe frequency of the heart is controlled by the heart rate,
Figure BDA0003663768200000073
to make a partial derivative about t, | | | | is norm, | is convolution, f (t) is the original time-series signal;
s4, introducing a penalty factor alpha and a Lagrange multiplier lambda to convert the constrained variation problem into an unconstrained variation problem to be solved, wherein the corresponding equation is as follows:
Figure BDA0003663768200000081
the method comprises performing dot product operation on the values of more than two vectors in the formula, and solving the unconstrained variational problem by using an alternative direction multiplier iterative algorithm to obtain lambda and uk、ωkIterative formula in corresponding frequency domain;
s5 is followed by inputting the initial uk (1)、ωk (1)、λ(1)Performing iteration on the Fourier transform form to obtain a formula (1) and a formula (2);
Figure BDA0003663768200000082
Figure BDA0003663768200000083
repeating the iteration of the formulas (1) and (2) until the following formula is satisfied:
Figure BDA0003663768200000084
in the formula, the upper right corner mark (n) and (n +1) represent iteration times, the upper mark ^ represents Fourier transformation, the lower right corner mark k represents the kth decomposition mode, epsilon is an allowable error, tau is noise tolerance, and epsilon and tau are constants and are larger than zero; and sequentially arranging the decomposed components according to the central frequency from low to high, and subtracting the modal components from the total data to obtain a residual error.
Finally, each settlement data component is shown in fig. 3, wherein the component 1 is a principal component, and basically conforms to the trend and the size of the total settlement data, so that the three-stage characteristic is obvious, and the three-stage characteristic has strong correlation with a process of excavating on site by adopting a double-step bench method. And the other components have small numerical values and are formed by other secondary factors and errors which influence the settlement, and the short-term construction settlement early warning judgment is mainly influenced on data.
The GRU neural network uses an Adam optimizer with a learning rate set to be 0.001, two layers of memories are respectively 80 and 100, the dropout rate is set to be 0.2, the batch size is set to be 64, the first 78 stages of settlement monitoring data are selected as training sets, and the later 170 stages of data are selected as testing sets; MSE is also used as a loss function, and MAE and root mean square error are used as auxiliary observation parameters.
The components obtained by the decomposition of the variation mode are arranged from large to small, and are sequentially input into a GRU neural network in the GRU neural network to train each component and residual error, single-step prediction and superposition are carried out to generate 238 single-step prediction data, then a data set is generated by adding 9 measured data and one prediction data, the same superposition is carried out to generate 238 double-step prediction data, stepping is sequentially repeated until seventh-step prediction data, namely prediction data of one week, is generated, and prediction in a longer period is provided for construction.
FIG. 4 shows the comparison between the single-step prediction effect of the GRU prediction model and the VMD-GRU prediction model at DB1-17 and the real sedimentation value, and it can be found that the fitting of the combined model to the nonlinear data is significantly better than that of a single neural network, and the data predicted by the combined model is more "coarse" and is more consistent with the measured data.
Table 1 shows the predicted value accuracy evaluation indexes of various prediction models at DB1-17 points, and it can be seen that the MSE, RMSE and MAE of the combined model are superior to those of a single neural network no matter single-step prediction or multi-step prediction is carried out, wherein the accuracy evaluation index of the VMD-GRU combined model is the best.
TABLE 1
Figure BDA0003663768200000101
The principle of the method is shown, the principle component trend obtained by variational modal decomposition of the validity of the method is basically similar to the trend and size of the total settlement data under the step method construction, the VMD-GRU combined model has good precision, and when the predicted value exceeds the standard, the method can give an early warning to guide the subsequent construction, thereby having good application prospect.
Example two
As shown in fig. 5, the schematic structural diagram of the system for predicting surface subsidence in subway tunnel step method construction in this embodiment includes a data collection module, a data processing module, a decomposition module, an optimization module, and a prediction module.
The data collection module is used for collecting the original ground subsidence sequence. Taking the surface subsidence data constructed by the bench method of the west subway tunnel of No. 11 line in Beijing city as an example, the tunnel surrounding rock mainly comprises clay and clay silty soil, more interlaminar diving, more complex geological conditions and variable surface subsidence rules.
The data are collected from a No. 17 section DB1-17 point, and the monitoring time is from 6/month and 4 days in 2020 to 2/month and 11 days in 2021. For interval data, the data processing module complements the interval data by adopting the average value of the data at two ends, so that the frequency of the whole sedimentation data time sequence is 1Hz, and 248 data sedimentation curves are shown in figure 2.
At this time, the decomposition module can know that when the eigenmode component number is 6, the smaller the center frequency overlap is, the more thorough the time series decomposition is, and the optimal mode number is obtained. And performing VMD decomposition on the surface subsidence time sequence through a decomposition module to obtain 6 subsequences with different central modes and a residual error. The VMD decomposition method comprises the following steps:
s1, assuming a decomposed modal function ku(t), which is Hilbert transformed to obtain its analytic signal, and then converted into a single-sided spectrum:
Figure BDA0003663768200000111
delta (t) is a unit impulse function, delta (t) is infinite when t is equal to 0, delta (t) is 0 when t is not equal to, and j is an imaginary number unit;
s2, the frequency spectrum and the central frequency omega are further combinedkIndex term e-jω k tAliasing translates it to baseband, yielding an analytic signal expressed as follows:
Figure BDA0003663768200000112
s3, in order to ensure that when the original sequence is decomposed into K modal components, the sum of the estimated bandwidths of the modal components with the center frequency is minimum, the sum of the modal components is equal to the original sequence, and a constraint variation problem is obtained, wherein the expression is as follows:
Figure BDA0003663768200000121
in the formula: u. ukIs a decomposed K mode functions, omegakThe center frequencies corresponding to the K mode functions respectively,
Figure BDA0003663768200000122
to derive a partial derivative with respect to t, | | | is a norm, | is a convolution operation, f (t) is an original time series signal;
s4, introducing a penalty factor alpha and a Lagrange multiplication operator lambda to convert the constraint variation problem into an unconstrained variation problem for solving, wherein the corresponding equation is as follows:
Figure BDA0003663768200000123
the method comprises calculating dot product of more than two vectors in the formula, and solving the unconstrained variational problem by using an alternative direction multiplier iterative algorithm to obtain lambda and uk、ωkIterative formula in corresponding frequency domain;
s5 is followed by inputting the initial uk (1)、ωk (1)、λ(1)Performing iteration on the Fourier transform form to obtain a formula (1) and a formula (2);
Figure BDA0003663768200000131
Figure BDA0003663768200000132
repeating the iteration of the expressions (1) and (2) until the following expression is satisfied:
Figure BDA0003663768200000133
in the formula, the upper right corner mark (n) and (n +1) represent iteration times, the upper mark ^ represents Fourier transformation, the lower right corner mark k represents the kth decomposition mode, epsilon is an allowable error, tau is noise tolerance, and epsilon and tau are constants and are larger than zero; and sequentially arranging the decomposed components according to the central frequency from low to high, and subtracting the modal components from the total data to obtain a residual error.
Finally, each settlement data component is shown in fig. 3, wherein the component 1 is a main component, and the trend and the size of the total settlement data are basically consistent, so that the three-stage characteristic is obvious, and the three-stage characteristic has strong correlation with a process of excavating by adopting a double-step bench method on site. And the other components have small numerical values and are formed by other secondary factors and errors which influence the settlement, and the short-term construction settlement early warning judgment is mainly influenced on data.
The optimization module controls the GRU neural network to use an Adam optimizer with the learning rate set to be 0.001, wherein two layers of memories are 80 and 100 respectively, the dropout rate is set to be 0.2, the batch size is set to be 64, the first 78 stages of settlement monitoring data are selected as training sets, and the later 170 stages of data are selected as test sets; the optimization module also uses MSE as a loss function and MAE and root mean square error as auxiliary observation parameters.
The components obtained by the decomposition of the variation mode are arranged from large to small, the components and the residual errors are sequentially input into a prediction module to be trained, single-step prediction and superposition are carried out to generate 238 single-step prediction data, then a data group is generated by adding 9 measured data and one prediction data, 238 double-step prediction data are generated by superposition, and stepping is sequentially repeated until seventh-step prediction data, namely one-week prediction data, is generated, so that the longer-term prediction is provided for construction.
FIG. 4 shows the comparison between the single-step prediction effect of the GRU prediction model and the VMD-GRU prediction model at DB1-17 and the real settlement value, and it can be found that the fitting of the combined model to the nonlinear data is significantly better than that of a single neural network, and the data predicted by the combined model is more "coarse" and is more consistent with the measured data.
Table 1 shows the predicted value accuracy evaluation indexes of various prediction models at DB1-17 points, and it can be seen that the MSE, RMSE and MAE of the combined model are superior to those of a single neural network no matter single-step prediction or multi-step prediction is carried out, wherein the accuracy evaluation index of the VMD-GRU combined model is the best.
TABLE 1
Figure BDA0003663768200000141
The principle of the method is shown, the principle component trend obtained by variational modal decomposition of the validity of the method is basically similar to the trend and size of the total settlement data under the step method construction, the VMD-GRU combined model has good precision, and when the predicted value exceeds the standard, the method can give an early warning to guide the subsequent construction, thereby having good application prospect.
The above-described embodiments are merely illustrative of the preferred embodiments of the present application, and do not limit the scope of the present application, and various modifications and improvements made to the technical solutions of the present application by those skilled in the art without departing from the spirit of the present application should fall within the protection scope defined by the claims of the present application.

Claims (8)

1. A method for predicting the land surface settlement in subway tunnel step construction is characterized by comprising the following steps:
acquiring original ground surface settlement sequence data;
processing the original settlement data;
performing VMD decomposition on the processed original earth surface settlement sequence data to obtain subsequences of different central modes and a residual error, wherein each subsequence is sequentially arranged from low to high according to the central frequency;
setting and optimizing hyper-parameters of the GRU neural network according to the sub-sequences and the residual errors;
training the optimized GRU neural network, and then inputting each subsequence and the residual error to obtain a prediction component of each subsequence and a prediction component of the residual error; and superposing the prediction component results to obtain a ground surface settlement prediction value.
2. The method for predicting the land surface subsidence in the subway tunnel step method construction according to claim 1, wherein the method for acquiring the data of the original land surface subsidence comprises the following steps: on the basis of an elevation control network, the vertical displacement change of each measuring point on the ground around the subway is monitored to obtain the vertical displacement change.
3. The method for predicting the land surface subsidence in the subway tunnel step method construction according to claim 1, wherein the method for processing the original subsidence data comprises the following steps: and complementing interval period data in the original earth surface settlement sequence data by adopting the average value of the data at two ends, so that the frequency of the whole settlement data time sequence is 1 Hz.
4. The method for predicting the land subsidence in the subway tunnel step method construction according to claim 1, wherein the method for performing VMD decomposition on the original land subsidence sequence data comprises: performing trial decomposition on the data according to K numbers respectively to obtain central frequency distribution graphs corresponding to different K numbers, and obtaining the optimal eigenmode component number K based on the central frequency distribution graphs; and decomposing the original settlement data by combining the optimal eigenmode component number K to obtain K subsequences with different central modes and a residual error.
5. The method for predicting the ground surface subsidence in the subway tunnel step method construction as claimed in claim 1, wherein said GRU neural network comprises an update gate, a reset gate and an output gate;
the GRU neural network combines a forgetting gate and an input gate in the LSTM network into an updating gate, retains the original reset gate, and learns the advantages of the LSTM gating network to update the cell state and the hidden state; the updating gate is used for describing the influence degree of the past time characteristic information on the current characteristic information, and the larger the threshold value of the updating gate is, the larger the influence of the previous time characteristic information on the current time is; the reset gate is used to describe the degree to which the status characteristic information at the past time is discarded, and the smaller the threshold value, the more the past information is discarded.
6. The method for predicting the land surface subsidence in the subway tunnel step method construction according to claim 1, wherein the method for training each subsequence and the residual error through a GRU neural network comprises the following steps:
forward propagation, sequentially carrying out weight operation on each input subsequence along a sequence propagation direction to output a target vector;
comparing with the test set and outputting an error through a target function;
the weight is changed by back propagation so as to extract the time series characteristics;
the forward propagation mathematical operation process is as follows:
Figure FDA0003663768190000031
in the formula: h ist-1And htFor hidden layer vectors at different times, XtAs input data, U, w is a weight parameter, is a gated activation function sigmoid, has an output value of 0 to 1, tan h as an activation function in generating a candidate memory, is a sign of Hadamard product, zt、rt、ht' update gate, reset gate, candidate state, respectively.
7. The system for predicting the land surface settlement during the subway tunnel step construction is characterized by comprising a data collection module, a data processing module, a decomposition module, an optimization module and a prediction module;
the data collection module is used for acquiring original earth surface settlement sequence data;
the data processing module is used for processing the original settlement data;
the decomposition module is used for performing VMD decomposition on the processed original earth surface settlement sequence data to obtain subsequences of different central modes and a residual error, and the subsequences are sequentially arranged from low to high according to the central frequency;
the optimization module is used for setting and optimizing the hyper-parameters of the GRU neural network according to the sub-sequences and the residual errors;
the prediction module is used for training the optimized GRU neural network, and then inputting each subsequence and the residual error to obtain a prediction component of each subsequence and a prediction component of the residual error; the prediction module is further used for superposing the prediction component results to obtain an earth surface settlement prediction value.
8. The system for predicting the land subsidence in the subway tunnel step-by-step construction as claimed in claim 7, wherein the workflow of said data processing module comprises: and complementing interval period data in the acquired original earth surface settlement sequence data by adopting the mean value of the data at two ends, so that the frequency of the whole settlement data time sequence is 1 Hz.
CN202210581122.9A 2022-05-26 2022-05-26 Method and system for predicting land surface settlement during subway tunnel step method construction Pending CN114742325A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210581122.9A CN114742325A (en) 2022-05-26 2022-05-26 Method and system for predicting land surface settlement during subway tunnel step method construction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210581122.9A CN114742325A (en) 2022-05-26 2022-05-26 Method and system for predicting land surface settlement during subway tunnel step method construction

Publications (1)

Publication Number Publication Date
CN114742325A true CN114742325A (en) 2022-07-12

Family

ID=82286708

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210581122.9A Pending CN114742325A (en) 2022-05-26 2022-05-26 Method and system for predicting land surface settlement during subway tunnel step method construction

Country Status (1)

Country Link
CN (1) CN114742325A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117114214A (en) * 2023-10-25 2023-11-24 辽宁东科电力有限公司 Substation equipment foundation settlement prediction method and system
CN117150445A (en) * 2023-10-30 2023-12-01 中铁建大桥工程局集团第三工程有限公司 Settlement monitoring and evaluating method for section tunnel short-distance downward river

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117114214A (en) * 2023-10-25 2023-11-24 辽宁东科电力有限公司 Substation equipment foundation settlement prediction method and system
CN117114214B (en) * 2023-10-25 2024-01-05 辽宁东科电力有限公司 Substation equipment foundation settlement prediction method and system
CN117150445A (en) * 2023-10-30 2023-12-01 中铁建大桥工程局集团第三工程有限公司 Settlement monitoring and evaluating method for section tunnel short-distance downward river
CN117150445B (en) * 2023-10-30 2024-02-23 中铁建大桥工程局集团第三工程有限公司 Settlement monitoring and evaluating method for section tunnel short-distance downward river

Similar Documents

Publication Publication Date Title
Giustolisi et al. Advances in data-driven analyses and modelling using EPR-MOGA
CN114742325A (en) Method and system for predicting land surface settlement during subway tunnel step method construction
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
Stojanovic et al. A self-tuning system for dam behavior modeling based on evolving artificial neural networks
Shiri et al. Estimation of daily suspended sediment load by using wavelet conjunction models
Mestav et al. State estimation for unobservable distribution systems via deep neural networks
CN113379107B (en) Regional ionosphere TEC forecasting method based on LSTM and GCN
CN110296833B (en) Soft measurement method and system for hydraulic cylinder comprehensive test board
CN103268519A (en) Electric power system short-term load forecast method and device based on improved Lyapunov exponent
CN111048163B (en) Shale oil hydrocarbon retention amount (S1) evaluation method based on high-order neural network
CN113687433B (en) Bi-LSTM-based magnetotelluric signal denoising method and system
CN110276441A (en) A kind of trapezoidal overlap kernel impulse response estimation method based on deep learning
CN115935834A (en) History fitting method based on deep autoregressive network and continuous learning strategy
CN113094860A (en) Industrial control network flow modeling method based on attention mechanism
CN114429077A (en) Time sequence multi-scale analysis method based on quantum migration
CN112381279B (en) Wind power prediction method based on VMD and BLS combined model
Cheng et al. Ensemble Kalman filter for GAN-ConvLSTM based long lead-time forecasting
CN116628442B (en) Groundwater reserve change space downscaling method based on artificial neural network
CN109061544B (en) Electric energy metering error estimation method
CN115577856A (en) Method and system for predicting construction cost and controlling balance of power transformation project
Wang et al. Inverse modeling for subsurface flow based on deep learning surrogates and active learning strategies
Silva et al. GAN for time series prediction, data assimilation and uncertainty quantification
Silva et al. Generative network-based reduced-order model for prediction, data assimilation and uncertainty quantification
Yang et al. Assessment of regional-scale geothermal production based on a hybrid deep learning model: A case study of the southern Songliao Basin, China
CN110829434B (en) Method for improving expansibility of deep neural network tidal current model

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