CN113282990A - Intelligent real-time determination method, terminal and medium for shield motion trail - Google Patents

Intelligent real-time determination method, terminal and medium for shield motion trail Download PDF

Info

Publication number
CN113282990A
CN113282990A CN202110597488.0A CN202110597488A CN113282990A CN 113282990 A CN113282990 A CN 113282990A CN 202110597488 A CN202110597488 A CN 202110597488A CN 113282990 A CN113282990 A CN 113282990A
Authority
CN
China
Prior art keywords
shield
parameters
term memory
long
short term
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
CN202110597488.0A
Other languages
Chinese (zh)
Inventor
卫海梁
王磊
沈水龙
陈露明
林松顺
王念
陈贺
李志坡
王猛
李明俊
韩非
邓川宁
季少雷
杨帅
罗园
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.)
Shantou University
China Railway 16th Bureau Group Co Ltd
Beijing Rail Transit Engineering Construction Co Ltd of China Railway 16th Bureau Group Co Ltd
Original Assignee
Shantou University
China Railway 16th Bureau Group Co Ltd
Beijing Rail Transit Engineering Construction Co Ltd of China Railway 16th Bureau 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 Shantou University, China Railway 16th Bureau Group Co Ltd, Beijing Rail Transit Engineering Construction Co Ltd of China Railway 16th Bureau Group Co Ltd filed Critical Shantou University
Priority to CN202110597488.0A priority Critical patent/CN113282990A/en
Publication of CN113282990A publication Critical patent/CN113282990A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • 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
    • G06N3/048—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

Abstract

The invention discloses an intelligent real-time determination method, a terminal and a medium for a shield motion trail, which comprise the following steps: collecting hydrogeological parameters; collecting shield operation parameters in tunnel shield construction, and performing discrete wavelet transform processing on the shield operation parameters; taking hydrogeological parameters and shield operation parameters as data sets, and dividing the data sets into training sets and testing sets; and establishing a long-short term memory neural network, inputting a training set in the long-short term memory neural network for training, finishing the training when the test set reaches the precision standard, and storing the long-short term memory neural network for determining the movement track of the shield. The method realizes high-precision and high-efficiency prediction of the shield attitude and the shield machine position deviation, and further realizes accurate control of the shield machine motion track by regulating and controlling the deviation in real time.

Description

Intelligent real-time determination method, terminal and medium for shield motion trail
Technical Field
The invention belongs to the field of subway shield construction, and particularly relates to an intelligent real-time determination method, a terminal and a medium for a shield motion trail.
Background
With the advance of urbanization, shield construction has become one of the most common construction methods in tunnel and infrastructure construction. The shield construction refers to a mechanical construction process that a shield machine excavates soil, transports muck and assembles segments under the ground according to the design axis of a tunnel. The movement track of the shield machine is generally determined by the posture and the position of the shield machine, if the movement track of the shield machine deviates from a design axis, an excavation route is changed, further the subsequent segment assembling operation and assembling quality are influenced, and the engineering problems of water seepage in a tunnel, overlarge ground settlement or uplift and the like can also be caused. Therefore, the accurate control of the movement track of the shield tunneling machine has important influence on the tunnel construction quality. However, in general, due to complex geological structure, frictional resistance between shield parts and deviation of a shield operation process, a motion track of a shield machine inevitably deviates from a design axis, and how to monitor and adjust the deviation in real time has important significance on accurate control of the motion track. At present, in actual engineering, a feedback technology is mostly adopted to control the movement track of the shield machine, but the control based on the feedback is started to adjust after the deviation of the movement track of the shield machine is monitored, so that certain time delay is realized, and unexpected construction risks can be caused. Therefore, the motion trajectory needs to be predicted in advance, the attitude and the position of the shield tunneling machine are adjusted in real time, and serious trajectory deviation and snake-shaped motion are avoided. In the existing research, the traditional machine learning methods such as neural network and fuzzy logic are mostly adopted to predict and adjust the tunnel motion trail. However, the accuracy of the network is still unsatisfactory when dynamic and non-linear time series data are encountered. With the development of deep neural networks, such as long-short term memory (LSTM) neural networks, superior performance is exhibited due to their good processing power for time series data. On the other hand, in most of the existing methods for determining the movement track of the shield, the complex geological environment around the shield is not considered, and a kriging interpolation method, a local space interpolation method based on probability aiming at regionalized variables, is adopted in the data processing of the method for determining the distribution condition and the physical and mechanical properties of the stratum between the drill holes, so that the problem of discontinuous geological parameter space obtained by drilling is solved.
Through the search of the prior technical documents, the application patent numbers are as follows: 201910364559.5, publication number: CN110195592A, patent name: the patent states that a model WCNN-LSTM based on mixed deep learning is trained, and a given corresponding relation between an input variable and a pose output variable is established, so that the pose change trend and the result of a shield machine can be predicted according to the input variable acquired at each moment, early prediction and intervention on the shield pose are facilitated, and the shield construction quality is improved. Although the accuracy and the time effect of shield pose control can be improved to a certain extent, the prediction accuracy of shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation and shield tail vertical deviation is only 50%, the result is not optimized, and the result is not an optimal solution. Meanwhile, the forecasting model does not consider tunnel hydrogeological parameters and cannot reflect the influence of the geological parameters on the movement track.
Disclosure of Invention
In order to overcome the defects of low prediction precision, no consideration of geological parameters and the like in the prior art, the invention provides an intelligent real-time determination method, a terminal and a medium for a shield motion track.
The invention provides an intelligent real-time determination method of a shield motion track, which comprises the following steps:
s100, collecting hydrogeological parameters, including: determining the stratum distribution of each drilling hole and hydrogeological parameters of engineering geology through geological exploration along the tunnel, and further determining the stratum distribution of an undetected area between the drill holes and the hydrogeological parameters of the engineering geology by adopting a kriging interpolation method;
s200, collecting shield operation parameters in tunnel shield construction, and performing discrete wavelet transform processing on the shield operation parameters;
s300, taking the hydrogeological parameters after the kriging interpolation and the shield operation parameters after the discrete wavelet transform processing as data sets, and dividing the data sets into training sets and test sets;
s400, establishing a long-short term memory neural network, inputting a training set in the long-short term memory neural network for training, finishing training when the test set reaches the precision standard, and storing the long-short term memory neural network for determining the shield motion track.
The invention provides a terminal for intelligently determining the shield motion trail in real time, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor is used for executing the method for intelligently determining the shield motion trail in real time when executing the program.
In a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when being executed by a processor, is configured to perform the method for intelligently determining a shield motion trajectory in real time.
The invention establishes a prediction model (LSTM network) of the motion trail, the model predicts the motion trail of the shield by inputting stratum hydrogeology parameters and shield operation parameters, and the motion trail of the shield can be expressed by the shield posture and the shield position deviation condition. The shield attitude comprises a rolling angle (R) and a pitch angle (P), the shield machine position deviation comprises shield head Horizontal Deviation (HDSH), shield tail Horizontal Deviation (HDST), shield head Vertical Deviation (VDSH) and shield tail Vertical Deviation (VDST), the high-precision and high-efficiency prediction of the shield attitude and the shield machine position deviation condition is realized by the model, and the accurate control of the motion track of the shield machine is realized by regulating and controlling the deviation in real time.
The method adopts a real-time determination model of the shield motion track based on a data driving technology, and the model predicts the dynamic behavior of the shield motion based on the attitude and position parameters, can process unbalanced data and time-related data, and improves the prediction precision; geological conditions and operational parameters are considered; the shield tunneling machine can help a driver to adjust shield tunneling operation parameters in advance; and a foundation is laid for the research of an automatic driving system in the future.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings and accompanying tables:
FIG. 1 is a flow chart of a method for intelligently determining a shield motion trajectory in real time according to an embodiment of the present invention;
FIG. 2 is a graph illustrating the convergence behavior of discrete wavelet transforms in combination with optimized long and short term memory models, in accordance with an embodiment of the present invention;
FIG. 3 is a diagram illustrating real-time prediction data of shield tail level deviation of an output variable according to an embodiment of the present invention.
Detailed Description
The present invention will be described in detail below. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications can be made by persons skilled in the art without departing from the spirit of the invention. All falling within the scope of the present invention. The operations not specifically described in this embodiment are performed by referring to the methods already described in the summary of the invention, and are not described herein again.
The embodiment of the invention provides a method for determining a shield motion track in real time, which can be specifically carried out by referring to the following steps:
step one, collecting geological data: and determining the stratum distribution of each drilling hole and hydrogeological parameters such as engineering geology and the like through geological exploration along the tunnel, and determining the stratum distribution of an undetected area between the drill holes and the hydrogeological parameters such as the engineering geology and the like by adopting a kriging interpolation method.
The hydrogeological parameters refer to the buried depth of the underground water level and the physical and mechanical parameters of each soil layer. The geological exploration refers to that drilling holes are arranged along two sides of a proposed tunnel interval in a crossed mode, and stratum distribution information, stratum physical and mechanical properties and the like are obtained through soil sampling samples. The drill holes are arranged at the positions 3-5m outside the tunnel structure, and the hole distance is 50-70 m. The stratum distribution refers to the types of all the stratums and the thicknesses Th of the upper and lower interfaces of all the stratums measured on the basis of drilling holes along the tunnel.
The underground water level burial depth refers to the distance between the underground water level along the tunnel and the ground surface. The physical and mechanical parameters of each soil layer refer to the physical and mechanical properties of the soil layer obtained through field or laboratory geotechnical tests, such as uniaxial compressive strength, plasticity index, consistency index, cohesion, internal friction angle and permeability.
The kriging interpolation method is a local space interpolation method based on probability for regional variables, a half variation function is obtained through measured data of drilled points of the regional variables, and the variation function model is used as an interpolation function to estimate the stratum distribution and the hydrogeological parameters of the non-drilled positions. Specifically, the semivariance function is a fitting function of semivariances of spatial data pairs with different spatial distances h and distances h, and the expression of the semivariance is formula (1).
Figure BDA0003091711400000041
Where γ (h) is the data pair (x) that is a distance of hiAnd xiA half variance of + h), xiFor the ith probed borehole, xi+ h is equal to xiProbed holes at a distance h, z (x)i) Is xiThe value of the area variable at the position, n being equal to xiThe number of probed boreholes at distance h.
And secondly, collecting shield operation parameters in tunnel shield construction, and performing discrete wavelet transform processing on the parameters.
The shield operation parameters refer to construction parameters of the shield tunneling machine during excavation, and comprise Thrust (TF), cutter head torque (CT), Specific Energy (SE), cutter rotating speed (RPM), tunneling speed (PR), spiral unearthing Speed (SR), Soil Pressure (SP), Grouting Pressure (GP), grouting amount (GV), shield head Horizontal Deviation (HDSH), shield head Vertical Deviation (VDSH), shield tail Horizontal Deviation (HDST), shield tail Vertical Deviation (VDST), rolling angle (R) and pitch angle (P).
Discrete wavelet transform processing is a discretization processing model based on wavelet transform, discretization is carried out on signals x (t) according to a scale factor a and a translation factor b through a Malay algorithm, signals with various resolutions and translation characteristics are obtained through stepwise decomposition, and accordingly noise in data is reduced, and the discrete wavelet transform processing is defined as an equation (2).
Figure BDA0003091711400000042
Wherein the content of the first and second substances,
Figure BDA0003091711400000043
a0＞1，b0j belongs to Z, j is the j level frequency resolution, and k is the k level translation transformation.
Further, the maratt algorithm is an effective tool for obtaining a multi-level resolution signal by discrete wavelet transform, mainly decomposes the signal into a low-frequency signal and a high-frequency signal by using a high-frequency filter and a low-frequency filter, can further decompose the decomposed low-frequency signal continuously, and finally decomposes the time series function into equation (3).
x(t)＝An(t)+Dn(t)+Dn-1(t)+···+D1(t) (3)
Wherein A isn(t) denotes an approximation of the original signal x (t), Dn(t) is the detail component of the noise data in the nth order decomposition.
And thirdly, dividing the data set into a training set and a testing set.
The data set includes hydrogeological parameters in the first step and shield operation parameters in the second step.
The hydrogeological parameters comprise the types of all the layers, the thicknesses of upper and lower interfaces of all the layers, the buried depth of underground water level and the physical and mechanical parameters of all the layers, such as uniaxial compressive strength, plasticity index, consistency index, cohesive force, internal friction angle and permeability; the shield operation parameters refer to construction parameters of the shield machine during excavation, and comprise Thrust (TF), Cutter Torque (CT), Specific Energy (SE), cutter rotating speed (RPM), tunneling speed (PR), spiral unearthing Speed (SR), Soil Pressure (SP), Grouting Pressure (GP) and grouting quantity (GV).
The training set is to select partial data from the total data set as input data for training the long-term and short-term memory model. The test set refers to model input data and labels for verifying the prediction accuracy of the long-term and short-term memory model, which are left except the training set, and the data type of the test set is the same as that of the training set.
The input data comprises hydrogeological parameters after kriging interpolation and shield operation parameters after discrete wavelet transformation processing.
And fourthly, establishing a long-short term memory neural network, inputting a training set in the long-short term memory neural network for training, finishing training when the test set reaches the precision standard, and storing the long-short term memory model.
In this step, the long-short term memory neural network is a long-short term memory recurrent neural network formed by a plurality of long-short term memory units. Compared with the traditional circulating neural network, the cell state is introduced into the long-term and short-term memory unit to store, update and transmit information, and the gate control unit is introduced to control the update of the cell state, so that the problem of gradient disappearance existing in long dependence is solved.
Specifically, the gate control unit comprises a forgetting gate, an input gate and an output gate, and is used for controlling the previous unit to output information ht-1Current input information xtRetention and forgetting. Wherein:
forgetting to control the previous cell state ct-1By inputting ht-1And xtOutput ft，ftTake on a value of [0,1]Interval, the closer the value is to 0, the more ct-1The greater the degree to which the state is forgotten, the more expression (4) can be given.
ft＝σ(wf[ht-1,xt]+bf) (4)
Wherein f istThe value of the forgetting gate at the time t; sigma is sigmoid function; matrix wfIs a weight matrix; vector bfIs an offset vector.
The input gate controls the degree of renewal of the cell state, via input ht-1And xtOutput it，itTake on a value of [0,1]Interval, the closer the value is to 1, the more ct-1Will update more information to form a new cell state ctAnd can be expressed as formula (5).
it＝σ(wi[ht-1,xt]+bi) (5)
Wherein itThe gate value is entered at time t; matrix wiIs a weight matrix; vector biIs an offset vector.
The output gate determines the output of the cell state via input ht-1And xtOutput ot，otTake on a value of [0,1]The interval indicates that the more the updated cell state is outputted as the current hidden state h as the value is closer to 1tAnd can be expressed as formula (6).
ot＝σ(w0[ht-1,xt]+b0) (6)
Wherein o istThe value of the output gate at time t; matrix woIs a weight matrix; vector boIs an offset vector.
Further, the updating of the cell state comprises the following steps:
(a) obtaining candidate cell information of cell state update
Figure BDA0003091711400000064
Such as (7)
Figure BDA0003091711400000065
Wherein the tanh activation function adjusts the output value to [ -1,1 [ ]]And (3) a range. Matrix wcIs a weight matrix; vector bcIs an offset vector.
(b) Determining the degree of updating the candidate cell information into the new cell state according to the input gate, and determining the degree of remaining the previous cell to the new cell according to the forgetting gate, as shown in equation (8)
Figure BDA0003091711400000061
Wherein the content of the first and second substances,
Figure BDA0003091711400000062
representing the element intelligence product.
(c) Determining the hidden state h of the next moment according to the output gatetAs shown in equation (9)
Figure BDA0003091711400000063
In this step, the training of the LSTM network means that the LSTM network weight is optimized by using an adam algorithm, so that the cost function of the predicted value and the true value meets the convergence requirement, and the model predicts the effect.
Further, the adam algorithm is a gradient descent optimization algorithm based on first-order gradients, and independent adaptive learning rates of various parameters are estimated by considering first-order moment and second-order moment estimated values of the gradients. The method comprises the following general steps:
a) selecting a random target at a time step t
b) Updating an offset first moment estimate
c) Updating biased second-order raw moment estimates
d) Calculating a corrected bias first moment estimate
e) Computing a corrected biased second-order raw moment estimate
f) The convergence was checked.
The cost function is the Root Mean Square Error (RMSE) between the predicted value and the measured value, and the calculation formula is shown in formula (10).
Figure BDA0003091711400000071
Wherein, yiWhich represents the actual value of the measured value,
Figure BDA0003091711400000072
representing the predicted value and n representing the total number of data sets.
The embodiment of the invention adopts the long-short term memory neural network obtained by training as a shield track determination model, and corresponding parameters are input, so that the corresponding shield track can be obtained in real time.
The embodiment of the invention realizes the high-precision and high-efficiency prediction of the shield attitude and the position deviation of the shield machine, and further realizes the precise control of the movement track of the shield machine by regulating and controlling the deviation in real time.
In another embodiment of the present invention, a terminal for intelligently determining a shield motion trajectory in real time is provided, which includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor is configured to execute the method for intelligently determining a shield motion trajectory in real time when executing the program.
In another embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when being executed by a processor, is configured to perform the method for intelligently determining a shield motion trajectory in real time.
In order to better illustrate the real-time determination method of the shield motion trajectory, the following description is made in conjunction with detailed engineering operations, but the following embodiments are not intended to limit the present invention:
an inter-city railway project is one of the largest infrastructure projects in China. The project connects a certain train station with a certain international airport. The length of the tunnel between the station areas is about 3.3km, and the buried depth is 8.0-22.0 m. The tunnel is excavated by an Earth Pressure Balance (EPB) shield machine. The mounting of a 1.6m wide, 400mm thick segment ring was achieved by airfoil vacuum mounting.
The method for determining the shield motion trajectory in real time based on the combination of the discrete wavelet transform and the optimized long-term and short-term memory model in the embodiment specifically comprises the following steps (see a flow chart in fig. 1):
step one, collecting geological data: and determining the stratum distribution of each drilling hole and hydrogeological parameters such as engineering geology and the like through geological exploration along the tunnel, and determining the stratum distribution of an undetected area between the drill holes and the hydrogeological parameters such as the engineering geology and the like by adopting a kriging interpolation method.
In this embodiment, the geological survey refers to that drill holes are arranged along two sides of the proposed tunnel interval in a crossing manner, and stratum distribution information, stratum physical and mechanical properties and the like are obtained by taking a soil sample.
In the embodiment, the drilling holes are arranged at the positions 3-5m outside the tunnel structure, the hole distance is 50-70m, and 32 drilling holes are formed.
In this embodiment, the stratum distribution refers to the thickness Th of the upper and lower interfaces of each stratum type measured based on the drilled holes along the tunnel. The main distribution stratum is backfill soil, clay, silty clay, weathered rock, strongly weathered granite and medium weathered granite in sequence.
In this embodiment, the hydrogeological parameters refer to the buried depth of the groundwater level and the physical and mechanical parameters of each soil layer.
In this embodiment, the groundwater level burial depth refers to the distance from the groundwater level along the tunnel to the ground surface. The water level is between 1.63m and 3.63m below the ground surface.
In this embodiment, the physical mechanical parameters of each soil layer refer to the physical mechanical properties of the soil layer obtained through a field or laboratory geotechnical test, such as uniaxial compressive strength, plasticity index, consistency index, cohesion, internal friction angle, and permeability. Through a laboratory geotechnical test, the plasticity index of the field collected soil sample is measured to be between 11.90 and 25.10, and the consistency index is measured to be below 1.
In this embodiment, the kriging interpolation method is a local spatial interpolation method based on probability for the regionalized variable, and a semi-variation function is obtained from measured data of a drilled point of the regionalized variable, and the variation function model is used as an interpolation function to estimate the formation distribution and the hydrogeological parameters of the non-drilled position.
In this embodiment, the semivariance function refers to a fitting function of semivariances of spatial data pairs with different spatial distances h and distances h, and the expression of the semivariance is formula (1).
Step two, collecting shield operation parameters: collecting shield operation parameters in tunnel shield construction and carrying out discrete wavelet transform processing on the parameters.
In this embodiment, the shield operation parameters refer to construction parameters of the shield machine during excavation, including Thrust (TF) from rings 378 to 1578, Cutter Torque (CT), Specific Energy (SE), cutter rotation speed (RPM), tunneling speed (PR), spiral unearthing Speed (SR), Soil Pressure (SP), Grouting Pressure (GP), grouting amount (GV), shield head Horizontal Deviation (HDSH), shield head Vertical Deviation (VDSH), shield tail Horizontal Deviation (HDST), shield tail Vertical Deviation (VDST), roll angle (R), and pitch angle (P).
In this embodiment, a key dimension based on data established by a previous researcher is selected. Selecting 19 parameters, and taking [ I1,I2，…I19]The method is used for dynamically predicting the motion trail of the shield. And recording the posture and the position of the shield tunneling machine in real time by using a laser navigation system.
In this embodiment, the discrete wavelet transform processing is a discretization processing model based on wavelet transform, and discretizes the signal x (t) by the scale factor a and the translation factor b through the maratt algorithm, and decomposes step by step to obtain signals with various resolutions and translation characteristics, so as to reduce noise in data, and is defined as equation (2).
In this embodiment, the Daubechies wavelet family with the best noise reduction effect is selected as the parent wavelet, and five decomposition levels are adopted in the model to denoise the data.
In this embodiment, the marant algorithm is an effective tool for obtaining a multi-level resolution signal by discrete wavelet transform, and mainly decomposes a signal into a low-frequency signal and a high-frequency signal by using a high-frequency filter and a low-frequency filter, and can further decompose the decomposed low-frequency signal, and finally decompose a time series function into equation (3).
And step three, forming a data set by the shield operation parameters in the step two and the hydrogeological parameters in the step one, and dividing the data set into a training set and a testing set.
In this embodiment, the training set refers to that 80% of samples are selected from 1200 sample data as input data for training the LSTM model.
In this embodiment, the input data includes hydrogeological parameters after kriging interpolation and shield operation parameters after discrete wavelet transform processing. The underground water level burial depth refers to the distance between the underground water level along the tunnel and the ground surface; the types and thicknesses of all soil layers, and the physical and mechanical parameters of all soil layers refer to the physical and mechanical properties of the soil layers obtained through field or laboratory geotechnical tests, such as uniaxial compressive strength, plasticity index, consistency index, cohesive force, internal friction angle and permeability; the shield operation parameters refer to construction parameters of the shield machine during excavation, and comprise Thrust (TF), Cutter Torque (CT), Specific Energy (SE), cutter rotating speed (RPM), tunneling speed (PR), spiral unearthing Speed (SR), Soil Pressure (SP), Grouting Pressure (GP) and grouting quantity (GV).
In this embodiment, the test set refers to the model input data and the labels for verifying the prediction accuracy of the LSTM model, which are obtained from 20% of samples except the training set, and the data types of the test set are the same as those of the training set.
And step four, establishing a long-short term memory neural network, inputting a training set into the model for training, finishing the training when the test set reaches the precision standard, and storing the long-short term memory model for determining the shield motion track.
In this embodiment, the long-short term memory neural network is a long-short term memory recurrent neural network formed by a plurality of long-short term memory units. Compared with the traditional circulating neural network, the cell state is introduced into the long-term and short-term memory unit to store, update and transmit information, and the gate control unit is introduced to control the update of the cell state, so that the problem of gradient disappearance existing in long dependence is solved.
In this embodiment, the gate control unit includes a forgetting gate, an input gate, and an output gate, and is used to control the previous unit to output information ht-1Current input information xtRetention and forgetting.
In this embodiment, the forgetting gate controls the previous cell state ct-1By inputting ht-1And xtOutput ft，ftTake on a value of [0,1]Interval, the closer the value is to 0, the more ct-1The greater the degree to which the state is forgotten, the more expression (4) can be given.
In this embodiment, the input gate controls the degree of updating of the cell state by inputting ht-1And xtOutput it，itTake on a value of [0,1]Interval, the closer the value is to 1, the more ct-1Will update more information to form a new cell state ctAnd can be expressed as formula (5).
In this embodiment, outputThe output of the gate determining the state of the cell, via input ht-1And xtOutput ot，otTake on a value of [0,1]The interval indicates that the more the updated cell state is outputted as the current hidden state h as the value is closer to 1tAnd can be expressed as formula (6).
In this embodiment, the updating of the cell state includes the following steps:
(a) obtaining candidate cell information of cell state update
Figure BDA0003091711400000101
Such as (7)
(b) Determining the degree of updating the candidate cell information into the new cell state according to the input gate, and determining the degree of remaining the previous cell state into the new cell state according to the forgetting gate, as shown in equation (8)
(c) Determining the hidden state h of the next moment according to the output gatetAs shown in equation (9)
In this embodiment, the model training is to optimize the long-term and short-term memory neural network weight by using an adam algorithm, so that the cost function of the predicted value and the true value meets the convergence requirement, and the model predicts the effect.
In this embodiment, the adam algorithm is a gradient descent optimization algorithm based on a first-order gradient, and estimates the independent adaptive learning rate of each parameter by considering the first-order moment and the second-order moment estimation values of the gradient.
In this embodiment, the cost function is a Root Mean Square Error (RMSE) between the predicted value and the measured value, and the calculation formula is as shown in formula (10). The calculation results were that HDSH was 2.82, HDST was 3.05, VDSH was 7.05, VDST was 7.80, P was 2.08, and R was 10.43.
In this example, 300 epochs were set in order to fit the discrete wavelet variations in combination with the optimization of the long-short term memory model. As shown in fig. 2, the model converges to the optimal fitness function after about 100 time periods and then to a minimum with subtle local oscillations. The result shows that the model reaches the optimal solution in 300 periods, and the searching operation can be stopped.
In this embodiment, in order to describe the performance of the prediction model (LSTM network), a relationship between the measured data and the predicted data of 6 output parameters (HDST, HDSH, VDST, VDSH, R, P) is given. As shown in fig. 3 (taking HDST as an example), the output parameters of the shield machine can be successfully predicted by wavelet change in combination with the optimized long-short term memory model, and the attitude and position parameters can be accurately predicted.
While the present invention has been described in detail with reference to the preferred embodiments, it should be understood that the above description should not be taken as limiting the invention. Various modifications and alterations to this invention will become apparent to those skilled in the art upon reading the foregoing description. Accordingly, the scope of the invention should be determined from the following claims.

Claims (10)

1. An intelligent real-time determination method for a shield motion track is characterized by comprising the following steps:
s100, collecting hydrogeological parameters, including: determining the stratum distribution of each drilling hole and hydrogeological parameters of engineering geology through geological exploration along the tunnel, and further determining the stratum distribution of an undetected area between the drill holes and the hydrogeological parameters of the engineering geology by adopting a kriging interpolation method;
s200, collecting shield operation parameters in tunnel shield construction, and performing discrete wavelet transform processing on the shield operation parameters;
s300, taking the hydrogeological parameters after the kriging interpolation and the shield operation parameters after the discrete wavelet transform processing as data sets, and dividing the data sets into training sets and test sets;
s400, establishing a long-short term memory neural network, inputting a training set in the long-short term memory neural network for training, finishing training when the test set reaches the precision standard, and storing the long-short term memory neural network for determining the shield motion track.
2. The intelligent real-time determination method of the shield motion trail according to claim 1, wherein the hydrogeological parameters refer to the underground water level burial depth and the physical and mechanical parameters of each soil layer.
3. The intelligent real-time determination method of the shield motion trail according to claim 1, wherein the kriging interpolation method is a local space interpolation method based on probability for regional variables, a semi-variation function is obtained through measured data of drilled points of the regional variables, and the variation function is used as an interpolation function to estimate the stratum distribution and hydrogeological parameters of the non-drilled positions;
the semivariance function is a fitting function of semivariances of spatial data pairs with different spatial distances h and distances h, and the semivariance expression is as shown in formula (1):
Figure FDA0003091711390000011
where γ (h) is the data pair (x) that is a distance of hiAnd xiA half variance of + h), xiFor the ith probed borehole, xi+ h is equal to xiProbed holes at a distance h, z (x)i) Is xiThe value of the area variable at the position, n being equal to xiThe number of probed boreholes at distance h.
4. The intelligent real-time shield motion trajectory determination method according to claim 1, wherein the shield operation parameters refer to construction parameters of a shield machine during excavation, and include Thrust (TF), Cutter Torque (CT), Specific Energy (SE), cutter rotation speed (RPM), tunneling speed (PR), spiral unearthing Speed (SR), Soil Pressure (SP), Grouting Pressure (GP), grouting amount (GV), shield head Horizontal Deviation (HDSH), shield head Vertical Deviation (VDSH), shield tail Horizontal Deviation (HDST), shield tail Vertical Deviation (VDST), roll angle (R) and pitch angle (P).
5. The intelligent real-time determination method for the shield motion trail according to claim 1, wherein the discrete wavelet transform process is a discretization process model based on wavelet transform, discretizing the signal x (t) by a scale factor a and a translation factor b through a maratt algorithm, and decomposing the signal step by step to obtain signals with various resolutions and translation characteristics, so as to reduce the noise in the data, and the method is defined as equation (2):
Figure FDA0003091711390000021
wherein the content of the first and second substances,
Figure FDA0003091711390000022
a0＞1，b0j belongs to Z, j is the j level frequency resolution, k is the k level translation transformation, and t is the time step.
6. The intelligent real-time shield motion trail determination method according to claim 1, wherein the long-short term memory neural network is a long-short term memory recurrent neural network composed of a plurality of long-short term memory units, the long-short term memory units are introduced with cell states for storing, updating and transmitting information, and the long-short term memory units are introduced with gate control units for controlling the updating of the cell states.
7. The intelligent real-time shield motion trail determination method according to claim 6, wherein the gate control unit comprises a forgetting gate, an input gate and an output gate, and is used for controlling a previous unit to output information ht-1Current input information xtRetention and forgetting of, wherein:
the forgetting gate controls the state c of the previous cellt-1By inputting ht-1And xtOutput ft，ftTake on a value of [0,1]Interval, the closer the value is to 0, the more ct-1The greater the degree to which the state is forgotten;
the input gate controls the degree of renewal of the cell state by input ht-1And xtOutput it，itTake on a value of [0,1]Interval, the closer the value is to 1, the more ct-1Will update more information to form a new cell state ct；
The output gateOutput for determining cell state by input ht-1And xtOutput ot，otTake on a value of [0,1]The interval indicates that the more the updated cell state is outputted as the current hidden state h as the value is closer to 1t。
8. The intelligent real-time determination method for the shield motion trail according to claim 7, wherein the updating of the cell state comprises:
(a) obtaining candidate cell information of cell state update
Figure FDA0003091711390000023
As shown in formula (7):
Figure FDA0003091711390000024
wherein the tanh activation function adjusts the output value to [ -1,1 [ ]]Range, matrix wcIs a weight matrix; vector bcIs an offset vector;
(b) determining the degree of updating the candidate cell information into the new cell state according to the input gate, and determining the degree of remaining the previous cell to the new cell according to the forgetting gate, as shown in equation (8):
Figure FDA0003091711390000025
wherein the content of the first and second substances,
Figure FDA0003091711390000031
representing an element intelligence product;
(c) determining the hidden state h of the next moment according to the output gatetIs given by equation (9):
Figure FDA0003091711390000032
9. an intelligent real-time determination terminal for a shield motion trail, comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor is used for executing the intelligent real-time determination method for the shield motion trail according to any one of claims 1-8 when executing the program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, is adapted to carry out the method for intelligent real-time determination of a shield motion trajectory according to any one of claims 1 to 8.
CN202110597488.0A 2021-05-31 2021-05-31 Intelligent real-time determination method, terminal and medium for shield motion trail Pending CN113282990A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110597488.0A CN113282990A (en) 2021-05-31 2021-05-31 Intelligent real-time determination method, terminal and medium for shield motion trail

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110597488.0A CN113282990A (en) 2021-05-31 2021-05-31 Intelligent real-time determination method, terminal and medium for shield motion trail

Publications (1)

Publication Number Publication Date
CN113282990A true CN113282990A (en) 2021-08-20

Family

ID=77282583

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110597488.0A Pending CN113282990A (en) 2021-05-31 2021-05-31 Intelligent real-time determination method, terminal and medium for shield motion trail

Country Status (1)

Country Link
CN (1) CN113282990A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114810100A (en) * 2022-06-28 2022-07-29 中铁工程服务有限公司 Shield tunneling attitude prediction method based on deep neural network

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110195592A (en) * 2019-04-30 2019-09-03 华中科技大学 Shield driving pose intelligent Forecasting and system based on interacting depth study
CN111242377A (en) * 2020-01-15 2020-06-05 重庆大学 Short-term wind speed prediction method integrating deep learning and data denoising
CN112505784A (en) * 2020-12-02 2021-03-16 中国矿业大学(北京) Method for optimizing excavation route of underground coal mine shield machine
CN112784493A (en) * 2021-01-27 2021-05-11 武汉轻工大学 Geographic space prediction method and system based on self-adaptive deep Q network
US20210148213A1 (en) * 2017-08-21 2021-05-20 Landmark Graphics Corporation Neural Network Models For Real-Time Optimization of Drilling Parameters During Drilling Operations

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210148213A1 (en) * 2017-08-21 2021-05-20 Landmark Graphics Corporation Neural Network Models For Real-Time Optimization of Drilling Parameters During Drilling Operations
CN110195592A (en) * 2019-04-30 2019-09-03 华中科技大学 Shield driving pose intelligent Forecasting and system based on interacting depth study
CN111242377A (en) * 2020-01-15 2020-06-05 重庆大学 Short-term wind speed prediction method integrating deep learning and data denoising
CN112505784A (en) * 2020-12-02 2021-03-16 中国矿业大学(北京) Method for optimizing excavation route of underground coal mine shield machine
CN112784493A (en) * 2021-01-27 2021-05-11 武汉轻工大学 Geographic space prediction method and system based on self-adaptive deep Q network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
刘玉良: "深度学习", 西安电子科技大学出版社, pages: 185 - 187 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114810100A (en) * 2022-06-28 2022-07-29 中铁工程服务有限公司 Shield tunneling attitude prediction method based on deep neural network
CN114810100B (en) * 2022-06-28 2022-12-02 中铁工程服务有限公司 Shield tunneling attitude prediction method based on deep neural network

Similar Documents

Publication Publication Date Title
Shen et al. Real-time prediction of shield moving trajectory during tunnelling
Zhu et al. Prediction of rockhead using a hybrid N-XGBoost machine learning framework
Gan et al. Two-level intelligent modeling method for the rate of penetration in complex geological drilling process
CN105849363B (en) Control the computer implemented method and wellbore system of shaft bottom sub-assembly
CN110500034B (en) Method for establishing neural network model, determining torsional pendulum drill string parameters and directionally drilling
US20210027144A1 (en) Petroleum reservoir behavior prediction using a proxy flow model
Hajihassani et al. 3D prediction of tunneling-induced ground movements based on a hybrid ANN and empirical methods
RU2687668C1 (en) Method and system for combined tracking of a well drilling process
Wang et al. Application of artificial neural network in tunnel engineering: A systematic review
CN105917074A (en) Managing wellbore operations using uncertainty calculations
CN106030031A (en) Controlling wellbore operations
US20210310307A1 (en) Process for real time geological localization with reinforcement learning
Kaunda et al. Neural network modeling applications in active slope stability problems
KR102294384B1 (en) Method for constructing drilling driving guide model to predict drilling rate using machine learning and system for predicting drilling rate using thereof
US20220090481A1 (en) Method and system for processing well log data from multiple wells using machine learning
CN111967079A (en) Foundation pit deformation prediction method based on improved artificial bee colony algorithm and BP neural network
US20210293133A1 (en) Process for real time geological localization with stochastic clustering and pattern matching
CN107861917A (en) Method of the multivariate data combined calculation well to stratigraphic boundary distance in horizontal well
US20210293129A1 (en) Process for real time geological localization with kalman filtering
CN113298220B (en) Neural network optimization-based shield tunneling speed prediction method
CN113255990A (en) Real-time prediction system and method for soil quality of excavation surface in tunnel construction by shield method
Gowida et al. Synthetic well-log generation: New approach to predict formation bulk density while drilling using neural networks and fuzzy logic
CN113282990A (en) Intelligent real-time determination method, terminal and medium for shield motion trail
Zhdaneev et al. Predictive systems for the well drilling operations
US20210293132A1 (en) Process for real time geological localization with greedy monte carlo

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