CN110195592A - Shield driving pose intelligent Forecasting and system based on interacting depth study - Google Patents

Shield driving pose intelligent Forecasting and system based on interacting depth study Download PDF

Info

Publication number
CN110195592A
CN110195592A CN201910364559.5A CN201910364559A CN110195592A CN 110195592 A CN110195592 A CN 110195592A CN 201910364559 A CN201910364559 A CN 201910364559A CN 110195592 A CN110195592 A CN 110195592A
Authority
CN
China
Prior art keywords
shield
pose
shield machine
model
lstm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910364559.5A
Other languages
Chinese (zh)
Other versions
CN110195592B (en
Inventor
周诚
骆汉宾
吴惠明
魏林春
王志华
许恒诚
陈睿
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.)
Huazhong University of Science and Technology
Shanghai Tunnel Engineering Co Ltd
Original Assignee
Huazhong University of Science and Technology
Shanghai Tunnel Engineering 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 Huazhong University of Science and Technology, Shanghai Tunnel Engineering Co Ltd filed Critical Huazhong University of Science and Technology
Priority to CN201910364559.5A priority Critical patent/CN110195592B/en
Publication of CN110195592A publication Critical patent/CN110195592A/en
Application granted granted Critical
Publication of CN110195592B publication Critical patent/CN110195592B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • E—FIXED CONSTRUCTIONS
    • E21—EARTH DRILLING; MINING
    • E21D—SHAFTS; TUNNELS; GALLERIES; LARGE UNDERGROUND CHAMBERS
    • E21D9/00—Tunnels or galleries, with or without linings; Methods or apparatus for making thereof; Layout of tunnels or galleries
    • E21D9/003—Arrangement of measuring or indicating devices for use during driving of tunnels, e.g. for guiding machines
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH DRILLING; MINING
    • E21D—SHAFTS; TUNNELS; GALLERIES; LARGE UNDERGROUND CHAMBERS
    • E21D9/00—Tunnels or galleries, with or without linings; Methods or apparatus for making thereof; Layout of tunnels or galleries
    • E21D9/06—Making by using a driving shield, i.e. advanced by pushing means bearing against the already placed lining
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH DRILLING; MINING
    • E21F—SAFETY DEVICES, TRANSPORT, FILLING-UP, RESCUE, VENTILATION, OR DRAINING IN OR OF MINES OR TUNNELS
    • E21F17/00—Methods or devices for use in mines or tunnels, not covered elsewhere
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH DRILLING; MINING
    • E21F—SAFETY DEVICES, TRANSPORT, FILLING-UP, RESCUE, VENTILATION, OR DRAINING IN OR OF MINES OR TUNNELS
    • E21F17/00—Methods or devices for use in mines or tunnels, not covered elsewhere
    • E21F17/18—Special adaptations of signalling or alarm devices

Abstract

The invention discloses a kind of shield driving pose intelligent Forecastings and system based on interacting depth study, belong to subway shield tunnel construction field.This method uses PREDICTIVE CONTROL principle and artificial intelligence technology on the basis of shield misalignment mechanism, shield driving stage pose is predicted according to the interacting depth learning model WCNN-LSTM of foundation, and shield pose adjustable strategies are formulated, realize the problem of pre-adjusting and the Prior Control of operating parameter are to improve shield misalignment.Intelligent predicting of this method for pose variation subsequent in shield machine tunneling process, shield machine driver is supported to be adjusted in advance to shield pose, it solves the serpentine locomotion problem of shield machine, alleviate shield pose regulation hysteresis effect, realize the accurate control of shield machine driving axis, tunnel Forming Quality, engineering practical value with higher can effectively be promoted.

Description

Shield driving pose intelligent Forecasting and system based on interacting depth study
Technical field
The invention belongs to subway shield tunnel construction fields, dig more particularly, to a kind of shield based on interacting depth study Carry appearance intelligent Forecasting and system.
Background technique
Shield method is to build the main engineering method of subway engineering.In shield method work progress, mainly have unstability, failure, Three problems such as misalignment.Wherein, metro shield misalignment problem will cause the construction quality of engineering, progress, cost and safety complete The influence in orientation.Misalignment, is mainly shown as shield pose misalignment, i.e. shield driving direction deviates design axis, tunnel is caused to pass through Logical error and pipe sheet assembling are second-rate.On the one hand shield pose misalignment will form tunnel molding misalignment of axe, cause tunnel Piercing error causes security risk to following metro operation；On the other hand, shield pose is bad can also cause tunnel duct piece spelling The problem of dress, easily generation section of jurisdiction faulting of slab ends, breakage, the tunnels quality problems such as leakage.Therefore, shield Pose Control is to solve shield The key of pose misalignment problem.
Shield Pose Control system is a typical closed loop control system, control process are as follows: engineer is in advance by tunnel The data for designing axis input shield guidance system, are shield machine and tunnel then by terrestrial net measurement and connection survey Road design axis establishes unified earth coordinates, then obtains current shield machine and design axis by shield pose measurement system Pose deviation data, deviation data is read by shield machine driver, judged, analyze after assign control instruction, operate shield (mainly oil cylinder propulsion system, cutterhead, Tu Cang and screw conveyor (soil pressure formula) etc. play collaboration for machine Pose Control executing agency The effect of control) pose adjustment is carried out to shield machine.In tunneling process, external influence factors do shield machine position and posture It disturbs and exists always, need to realize closed-loop control by the way that shield machine position and posture is continuously measured and adjusted.But control process It must can just exert one's influence and tell on, belong to the thing in quality management after controlled volume deviates setting value and generates deviation Control method afterwards.In addition, subsequent control, which will lead to shield driving the shortcomings that control not in time, forms " snakelike " track, and control stagnant After be the inherent defect of this theoretical method, can not thoroughly eliminate.It is based on the shield pose adjustment technology controlled afterwards The main theory method of current shield Pose Control, under this control model, shield Pose Control effect is poor, shield serpentine locomotion It is difficult to avoid that, non-optimal control strategy.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of shields based on interacting depth study Structure tunnels pose intelligent Forecasting and system, it is intended that being instructed based on interacting deep learning model WCNN-LSTM Practice, establish the corresponding relationship of given input variable and pose output variable, the input so as to be obtained according to each moment becomes Amount predicts the pose variation tendency of shield machine and as a result, to be conducive to carry out early prediction and intervention to shield pose, is promoted Shield-tunneling construction quality.
To achieve the above object, according to one aspect of the present invention, a kind of shield based on interacting depth study is provided Tunnel pose intelligent Forecasting, including model training stage and pose forecast period, in which:
The model pre-training stage is carried out based on interacting deep learning model WCNN-LSTM, comprising:
Step 1: data are determining and acquire:
1.1, the input and output parameter of WCNN-LSTM is determined:
By driving speed, permeability, cutter head torque, gross thrust, cutterhead revolving speed, cutterhead electric current, cutterhead total torque, total extruding Power, practical excavated volume, cutter plate driver total current, cutterhead position, propelling cylinder stroke initial value, propelling cylinder thrust, cutterhead are stretched Contracting oil cylinder stroke, spoil disposal volume flow and bentonite inlet flow rate are as prediction model input variable；
By shield tail horizontal departure, shield tail vertical missing, shield head horizontal departure, shield head vertical missing, yaw angle and pitch angle As six output variables of prediction model, for characterizing shield machine pose；
1.2, assume that current time is t, the shield operation data according to specified time interval acquisition t to the t-n moment includes The input and output parameter that step 1.1 determines therefrom selects data continuously without missing and the complete ring of data as instructing Practice sample, training sample is divided into training set and test set, n is positive integer；
Step 2: the input parameter input WCNN-LSTM in training set is trained；
Step 3: trained WCNN-LSTM being tested using the input and output parameter of test set, and will The parameter of reality output and the output parameter of test set carry out deviation comparison, if deviation exceeds preset range, adjust WCNN- The internal reference and return step 3 of LSTM；If deviation is within a preset range, training is finished, and obtains the shield driving based on WCNN-LSTM Pose prediction model.
The pose forecast period includes:
Step 4: by t+j moment or t+1 to t+j period corresponding input variable, inputting the prediction of shield driving pose Model obtains the pre- of six output variables of the characterization shield machine pose in corresponding t+j moment or t+1 to t+j period Measured value, i.e. shield machine pose predicted value.
Further, further include as follows based on Prior Control shield pose correction the stage:
Step 5: calculating the inclined of t+j moment or shield machine pose predicted value in t+1 to t+j period and design value Difference adjusts the control being applied on shield machine in t+j moment or t+1 to t+j period if deviation exceeds allowed band in advance Amount processed, operation shield machine carry out pose adjustment, and realization is previously-completed correction operation before shield driving not yet misalignment.
Further, the pose adjustment process in step 5 is as follows:
Step 5.1: inscribed when according to t+j or t+1 to t+j period in shield pose predicted value and design value it is inclined Difference, calculating the corresponding moment needs to be applied to control amount on shield machine, and the control signal of shield machine is corrected according to the control amount；
Step 5.2: according to revised control signal, running in shield machine to when next sampling instant, adjust shield Seat in the plane appearance obtains the actual observed value of shield machine pose adjusted；Resampling and according to the obtained input variable of sampling into The prediction of line position appearance, the predicted value after obtaining the sampling instant resampling；
Step 5.3: according to the deviation of the predicted value after the actual observed value of shield machine pose adjusted and resampling, into One step corrects shield driving pose prediction model.
Further, in step 5.3, the method for correcting shield driving pose prediction model is as follows: resampling is obtained Training sample is added in the actual observed value of input variable and pose adjusted, is trained to shield driving pose prediction model And update.
To achieve the goals above, it is another aspect of this invention to provide that providing a kind of shield based on interacting depth study Structure tunnels pose intelligent predicting system, including processor, model training program module, WCNN-LSTM model and pose prediction Program module；
The model training program module executes foregoing model training stage when being called by the processor, WCNN-LSTM model is trained, the shield driving pose prediction model based on WCNN-LSTM is obtained；
The pose Prediction program module executes foregoing pose forecast period when being called by the processor, To obtain shield driving pose prediction result based on shield driving pose prediction model.
It further, further include pose correction program module；
The pose correction program module is executed when being called by the processor as previously described based on Prior Control Shield pose is rectified a deviation the stage, to carry out the correction of shield pose based on shield pose prediction result.
Further, further include Noise Elimination from Wavelet Transform program module, be used for when being called by the processor, to such as preceding institute The data acquired in the step 1 and step 4 stated carry out data de-noising by wavelet transformation, decompose to parameter time series, To remove ambient noise and systematic measurement error, signal reconstruction is carried out then to generate the data sequence after new denoising.
To achieve the goals above, it is another aspect of this invention to provide that providing a kind of shield based on interacting depth study Structure tunnels pose intelligent predicting system, including data acquisition and input module, wavelet transformation noise filter module, shield seat in the plane Appearance prediction model module and result output module；
The data acquisition and input module, for obtaining specified shield machine operating parameter, and by the wavelet transformation Shield machine operating parameter after the denoising of noise filter module inputs the shield machine pose prediction model module；
The wavelet transformation noise filter module, the shield machine for obtaining to the data acquisition and input module are transported Row parameter carries out data de-noising by wavelet transformation, decomposes to parameter time series, is surveyed with removing ambient noise and system Error is measured, carries out signal reconstruction then to generate data sequence after new denoising；
The shield machine pose prediction model module is based on WCNN-LSTM model, for being run according to the shield machine of input Parameter prediction shield machine pose variation tendency and result；
The result output module, for receiving and showing data that the data acquisition obtains with input module, described Data, the predictive analysis results of the shield machine pose prediction model module after the denoising of wavelet transformation noise filter module, And correction scheme.
In general, the above technical scheme conceived by the present invention compared with prior art, can obtain following beneficial to effect Fruit:
1, the present invention comprehensively considers shield ginseng for the defect of the shield pose adjustment technology controlled in the prior art afterwards It is deep to establish a kind of mixing with shield driving pose intelligent predicting function based on WCNN-LSTM for several dynamics and time variation Learning model is spent, the shield operating parameter at the moment can be obtained according to the shield operating parameter that each moment inputs, and predicts to connect The shield pose at a certain moment or period got off is conducive to construction worker and carries out early intervention and control to shield pose System, it is horizontal so as to improve shield Pose Control.
2, the present invention is predicted and is controlled by the integrated shield that prediction-correction-prediction-amendment combines, in shield mistake It is rolled in real time in journey and updates shield pose prediction model, the precision and timeliness of shield Pose Control can be greatly promoted, and then solve Certainly the serpentine locomotion problem of shield machine, alleviation shield pose regulate and control hysteresis effect, final to realize the accurate of shield machine driving axis It controls and effectively promotes tunnel Forming Quality.
Detailed description of the invention
Fig. 1 is shield Load Model schematic diagram provided by the invention；
Fig. 2 is the shield pose adjustment schematic diagram provided by the invention based on Prior Control；
Fig. 3 is a kind of shield pose dynamic prediction mixed model frame diagram based on deep learning provided by the invention；
Fig. 4 is a kind of shield pose dynamic prediction mixed model architecture based on deep learning provided by the invention Figure；
Fig. 5 is flow chart of the invention；
Fig. 6 is the RMSE mean value comparison of different prediction models.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below Not constituting a conflict with each other can be combined with each other.
As shown in figure 5, a kind of shield driving pose based on interacting depth study of the preferred embodiment of the present invention is intelligently pre- Survey method, including model training stage and pose forecast period, in which:
The model pre-training stage is carried out based on interacting deep learning model WCNN-LSTM, comprising:
Step 1: data are determining and acquire:
1.1, the input and output parameter of WCNN-LSTM is determined:
By driving speed, permeability, cutter head torque, gross thrust, cutterhead revolving speed, cutterhead electric current, cutterhead total torque, total extruding Power, practical excavated volume, cutter plate driver total current, cutterhead position, propelling cylinder stroke initial value, propelling cylinder thrust, cutterhead are stretched Contracting oil cylinder stroke, spoil disposal volume flow and bentonite inlet flow rate are as prediction model input variable；
By shield tail horizontal departure, shield tail vertical missing, shield head horizontal departure, shield head vertical missing, yaw angle and pitch angle As six output variables of prediction model, for characterizing shield machine pose；
1.2, assume that current time is t, the shield operation data according to specified time interval acquisition t to the t-n moment includes The input and output parameter that step 1.1 determines therefrom selects data continuously without missing and the complete ring of data as instructing Practice sample, training sample is divided into training set and test set, n is positive integer；
Step 2: the input parameter input WCNN-LSTM in training set is trained；
Step 3: trained WCNN-LSTM being tested using the input and output parameter of test set, and will The parameter of reality output and the output parameter of test set carry out deviation comparison, if deviation exceeds preset range, adjust WCNN- The internal reference and return step 3 of LSTM；If deviation is within a preset range, training is finished, and obtains the shield driving based on WCNN-LSTM Pose prediction model.
The pose forecast period includes:
Step 4: by t+j moment or t+1 to t+j period corresponding input variable, inputting the prediction of shield driving pose Model obtains the pre- of six output variables of the characterization shield machine pose in corresponding t+j moment or t+1 to t+j period Measured value, i.e. shield machine pose predicted value.
Preferably, after the future trends for predicting shield machine pose, then deviation can be carried out according to design value and sentenced It is disconnected, and accordingly make correction.
Correspondingly, the present embodiment additionally provides the correction stage of the shield pose based on Prior Control comprising following steps:
Step 5: calculating the inclined of t+j moment or shield machine pose predicted value in t+1 to t+j period and design value Difference adjusts the control being applied on shield machine in t+j moment or t+1 to t+j period if deviation exceeds allowed band in advance Amount processed, operation shield machine carry out pose adjustment, and realization is previously-completed correction operation before shield driving not yet misalignment.
Further, the pose adjustment process in step 5 is as follows:
Step 5.1: inscribed when according to t+j or t+1 to t+j period in shield pose predicted value and design value it is inclined Difference, calculating the corresponding moment needs to be applied to control amount on shield machine, and the control signal of shield machine is corrected according to the control amount；
Step 5.2: according to revised control signal, running in shield machine to when next sampling instant, adjust shield Seat in the plane appearance obtains the actual observed value of shield machine pose adjusted；Resampling and according to the obtained input variable of sampling into The prediction of line position appearance, the predicted value after obtaining the sampling instant resampling；
Step 5.3: according to the deviation of the predicted value after the actual observed value of shield machine pose adjusted and resampling, into One step corrects shield driving pose prediction model.The method for correcting shield driving pose prediction model is as follows: resampling is obtained Input variable and pose adjusted actual observed value be added training sample, shield driving pose prediction model is instructed Practice and updates.
Below in conjunction with FIG. 1 to FIG. 4, method and principle of the invention are described in detail with a concrete case.
The shield pose adjustment principle proposed by the present invention based on Prior Control is introduced first:
According to the analysis in relation to shield pose misalignment effects factor as a result, establishing shield Load Model, stress load is such as Shown in Fig. 1:
(1) from gravity F1.It mainly include that shield machine is self-possessed, and excavates the self weight of storehouse dregs.
(2) shield tail directed force F2.In shield machine tunneling process, the active force of section of jurisdiction ring and shield shell tail portion inner surface is main It will (its purposes be that segment assembly and shield machine is isolated, and section of jurisdiction end is made to form confined air by shield tail wire brush and seal grease Between be convenient for slip casting) generation effect.
(3) cylinder jack thrust F3.Hydraulic cylinder is circular layout along shield body, acts on and has been assembled into endless tube piece end face, Shield body is pushed ahead using reaction force.In practice of construction, the size direction of all cylinder jack thrusts is inconsistent, here Depending on its resultant force, and in view of the frictional force between lining cutting pipe ring and jack base, it is oriented parallel to shield machine axis.
(4) excavation face soil body resistance F4.Excavation face soil body resistance by soil body resistance and cutterhead periphery in front of cutterhead soil Body resistance composition.For earth pressure balanced shield and slurry-water balance type shield, soil body resistance between the two is slightly distinguished, but Equilibrium principle is similar.
(5) shield body outer surface soil body resistance F5.There is friction resistance when soil layer advances, between shield body shell and the soil body in shield machine Power.In addition, shield machine also will receive extruding of the soil body to shield shell to the extruding of the soil body when carrying out pose correction to shield Counter-force.Existing research shows that shield body outer surface soil body resistance is the most important influence factor for influencing shield machine Pose Control, and Soil pressure suffered by shield shell is also due to the variation of shield motor behavior causes surrounding soil to deform and change.
Shield machine tunnels in the soil body belongs to three-dimensional space motion, therefore the motor behavior of shield possesses 6 freedom degrees, respectively Change in displacement value Δ for shield machine in three X-axis, Y-axis and Z axis directionsx,Δy,ΔzAnd shield pitch angle α, yaw angle β and The variation of roll angle γ.The mechanical model of motion of shield machine is the stress Load Analysis based on shield machine, constructs the position of shield machine Attitude control parameter, Δx,Δy,Δz, α, β, the functional relation of γ and each load of shield.Shield mechanical model of motion can be by as follows Math equation expression, wherein t is time variable:
f(F1,F2,F3,F4,F5,Δx,Δy,Δz, α, β, γ, t) and=0 (1)
Then, on the basis of shield mechanical model of motion, comprehensively consider shield pose in shield tunneling process influence because Plain d (t), it is assumed that current time t obtains data of the shield pose correlated inputs variable within t to the t-n period, is denoted as { i (t), i (t-1) ..., i (t-n) }；By correlated inputs variable { i (t), i (t-1) ..., i (t-n) } input shield pose prediction Model, the shield pose predicted value (predicted value of i.e. six output variables of output t+j moment (or t+1 to t+j period)And pass through pose predicted valueThe pose measurement value c obtained with shield pose measurement systemm(t+j) Feedback compensation further corrects prediction model；
The deviation size e (t+j) for comparing pose predicted value and design value under the t+j moment, by shield machine operator judgement be It is no to need to rectify a deviation in advance, if deviation is excessive, shield pose is controlled by control signal u (t) and adjusts executive device, is realized to shield The Prior Control and adjustment of structure seat in the plane appearance, adjustment amount are q (t), and correction operation is previously-completed before shield driving not yet misalignment.
Inside forecasting system, when being in the t+j moment, shield pose measurement system collects the reality of pose at this time Value, then compares with predicted value, carries out feedback compensation to forecasting system.Meanwhile optimization process is with the rolling of certain time window Dynamic to implement, repeatedly on-line amending, guarantees that prediction model remains that higher precision of prediction, shield pose adjust schematic diagram such as Shown in Fig. 2.
In order to realize above-mentioned prediction and method for correcting error, it is also necessary to establish the shield machine pose based on interacting deep learning model Dynamic Forecasting System.
A kind of shield driving pose intelligent predicting system of interacting deep learning model provided in this embodiment, including data It obtains pre- with input module, wavelet transformation noise filter module, convolutional neural networks feature extractor module, long short-term memory Device module and result output module are surveyed, as shown in Figure 3.The WCNN-LSTM of embodiment is by convolutional neural networks (CNNs, sheet Embodiment choose Lenet-5) feature extractor module and long short-term memory predictor module (LSTM) composition.
The data acquisition and input module are mainly used for data acquisition, storage and corresponding pretreatment.Shield machine packet Containing hundreds of parameters, and mass data is generated during operation, this module will be recorded in shield machine plc data library with every 10 seconds In；
The wavelet transformation noise filter module, by wavelet transformation carry out data de-noising, to parameter time series into Row decompose, to remove ambient noise and systematic measurement error, then carry out signal reconstruction to generate data sequence after new denoising, And then improve the precision and accuracy of model prediction；
The CNNs feature extractor module, for automatically extracting the key characteristic of data.CNNs is that have multiple convolution The depth network structure of layer and pond layer, extracted data characteristic include the most data characteristics of source data, and more just The mode of time series data is identified in fallout predictor；
The LSTM predictor module, LSTM are the neural network on time dimension with depth structure, some notes Recalling unit can store the information of historical time sequence, and be trained on a large scale by using supervised learning method, real The multi-step prediction of existing target variable；
The result output module, the data transmitted for receiving computing system, and be shown in visualization interface.This mould The shield machine pose relevant parameter that block can also provide model prediction analysis result and prediction obtains, and the decision of adjustment is provided Scheme.
After putting up forecasting system, it can by way of model training, establish and learn mould based on interacting depth The shield machine pose prediction model of type, the prediction of line position of going forward side by side appearance and correction.
Step 1: data collection:
Shield operation data and is stored in shield machine local computer by the sensor collection of each subsystem, then The data center being transferred to by fiber optic network on ground.In the present embodiment, entire training set includes nearly 1000 parameters, with The frequency of 0.1HZ stores.Until tunnel holing through, database have stored nearly 1,000,000 data.
(1) model output parameters are determined.Interacting deep learning model in the present invention is intended to predict shield machine in underground sky Between pose in motion process and position.Laser-guided systems are responsible for pose of the real-time measurement shield machine in tunnel excavating process The position and.Wherein, shield tail horizontal departure (Horizontal deviation of shield tail, hereinafter referred to as HDST), shield Tail vertical missing (Vertical deviation of shield tail, hereinafter referred to as VDST), shield head horizontal departure (Horizontal deviation of shield head, hereinafter referred to as HDSH), shield head vertical missing (Vertical Deviation of shield head, hereinafter referred to as VDSH), yaw angle (Yaw angel) and pitch angle (Pitch angle) It is most important Con trolling index in shield machine tunneling process.Therefore, the present invention selects this six parameters as the defeated of prediction model Variable out；
(2) mode input parameter is determined.Shield machine is mainly by cutter disc system, propulsion system, dregs transportation system, guiding system Several subsystems such as system are constituted, and contain nearly thousand parameters in operation data.In view of shield machine motor behavior mainly by Propulsion system and cutter disc system are controlled, therefore the present invention is had chosen according to engineering experience based on the hybrid shield machine of extra large Rake 32 parameters such as driving speed, permeability, cutter head torque, gross thrust, cutterhead revolving speed are as input variable, as shown in table 1, and With the frequency of 0.1Hz along tunnel piercing continuous acquisition；
1 prediction model input variable of table summarizes
It should be noted that according to the shield machine of different brands, the group number and spoil disposal volume flow, muddy water of propelling cylinder The detection group number of circular flow and bentonite inlet flow rate has certain difference, and therefore, total detection parameters quantity is with shield machine Different and will be different, still, the type of detection parameters is type listed by table 1.
(3) samples selection.The continuity of data will affect the precision of prediction model, we select continuously 18 rings without missing Data, and the data of each ring are complete；
Step 2: model training:
A kind of interacting deep learning model WCNN-LSTM that the present embodiment proposes for predicting the pose of shield machine, and divides Analyse model prediction accuracy and efficiency.Wherein, model training frame uses the deep learning frame based on the rear end Tensorflow Frame Keras.
(1) data prediction.Data prediction consists of two parts, that is, rejects the non-driving time hop counts evidence of data sequence And Noise Elimination from Wavelet Transform is executed to time series data.Because wavelet transform (DiscreteWaveletTransform, Hereinafter referred to as DWT) redundancy coefficient can be reduced, it is widely used in engineering practice, the present invention uses wavelet transform Denoising is carried out to initial data, eliminates the data of non-driving state.
For continuous wavelet transform (ContinuousWaveletTransform, hereinafter referred to as CWT), formula can be passed through (2) small echo basic function is defined:
Wherein α and τ is scale factor and shift factor respectively, and t is the time.After sampling α and τ, Then wavelet function becomes as follows:
Wherein m=0,1,2 ..., n ∈ Z, integer m, n control the expansion and translation of small echo respectively, and are included in all integers Set in.a0It is specified fixed expansion step parameter, value is set greater than 1.b0It is location parameter and has to be larger than zero. Usual a0And b0Value 2 and 1 respectively, then formula (3) formula becomes formula (4):
Wavelet transformation TM, nSignal x (t) can be resolved into the difference sub- frequency spectrum horizontal with different resolution.It determines Justice is as follows:
A kind of fast algorithm for DWT for being known as Mallat is used in the present invention, and WT and is set up by spatial decomposition more The correlativity of resolution analysis.Time series data can be decomposed as follows by Mallat algorithm:
X (t)=An(t)+Dn(t)+Dn-1(t)+…D1(t) (6)
Wherein An(t) be original signal x (t) approximate part, DnIt (t) is relevant to the noise information in n-layer decomposition thin Save part.The data sequence of the Multiresolution Decomposition of x (t) is { An(t),Dn(t),Dn-1(t),…,D1(t)}.It generallys use Index (the lower its value the better) of the square error (MSE) as evaluation denoising effect, MSE can be calculated by formula (7):
Wherein yiCorrespond to input xiDesired output,Indicate the output valve of Denoising Algorithm, N indicates sample size.
The present embodiment uses a kind of quick DWT algorithm, training set include 38 variables (wherein 32 input variables and 6 output variables), in order to obtain optimal denoising effect, it is necessary to which each variables choice denoises the optimal wavelet basis function of effect. Then, representative variable is selected to execute Wavelet Algorithm from each classification.It is calculated finally, traversing all Db wavelet basis functions It denoises effect, and the index using mean square error (MSE) as evaluation denoising effect (the lower its value the better).
(2) prediction model structure.Interacting depth neural network of the invention consists of two parts, i.e. LeNet-5 and LSTM, General frame is as shown in Figure 4.
Prediction model is used as feature extractor using convolutional neural networks (CNN, LeNet-5).Each filter is one A weight matrix with part connection and shared weight, it can be by original image convolution to corresponding Feature Mapping, the spy The image that sign mapping can be considered as filter extraction indicates.
Convolutional calculation formula indicates are as follows:
Wherein xI, jIndicate the ith row and jth column of input picture, and wm,nIndicate the m row and n-th of k × k weight matrix Column, wbIndicate filter deviation, f indicates activation primitive (usually using ReLU function), ai,jIndicate the i-th row of Feature Mapping With the value of jth column.
Prediction model uses long short-term memory prediction (LSTM) as fallout predictor, all for dynamically capture sequence data The temporal information of element predicts the pose of shield machine.Only one is sensitive to short-term input for regular circulation neural network (RNN) Hidden state, but LSTM is added to the location mode of a storage long-term information, and controls it using three doors.As right The improvement of traditional RNN, LSTM successfully overcome disappearance gradient by introducing storage unit；
Wherein, three control door formula are as follows:
(1) forget door.Control the location mode c of previous momentt-1To the location mode c of current timetInfluence degree, Its formula can be with is defined as:
ft=σ (Wf·[ht-1,xt]+bf) (9)
(2) input gate.Control the input variable x at current timetTo the location mode c at current timetInfluence degree, Formula can be with is defined as:
it=σ (Wi·[ht-1,xt]+bi) (10)
(3) out gate.Control the location mode c at current timetTo current output htInfluence degree, formula can determine Justice are as follows:
ot=σ (Wo·[ht-1,xt]+bo) (13)
The mathematic sign of above formula is defined as follows: xtIt is the input vector in t moment.Wf、Wi、WoRespectively be forget door, The weight matrix of input gate and out gate.bf、bi、boIt is to forget door, the bias vector of input gate and out gate respectively.htWhen being Carve the value of the storage unit under t.ft、it、otIt is the value for forgeing door, input gate and out gate of moment t.Symbol " " indicates point Product, symbolIt is by element product.
(3) parameter setting.Training set and test set are respectively the 80% and 20% of total sample.We select MSE and Adam It must parameter, i.e. loss function and optimizer as two needed for compiling Keras model.In addition, model compilation has used English The big deep learning up to CUDA accelerates frame.
Step 3: being trained by training set, and verify training result using test set, if six output variables is pre- Within the allowable range, then the shield pose prediction model after training can come into operation the deviation of measured value and measured value.
Step 4: prediction of result:
According to given time or the input variable of period, six output variables of certain required precision are met by analysis The prediction result of WCNN-LSTM model obtains the future trends of shield machine pose and position.
Then, correction control and model modification are carried out according to step 5.
In the following, other three kinds of common prediction models, i.e. ARIMA, LSTM and WLSTM are introduced, with WCNN- of the invention LSTM model is compared, to verify the validity and precision of prediction for proposing method in the present invention.
(1) ARIMA model
ARIMA model is commonly used to the time series feature of forecasting research data, usually going through by univariate time series History value predicts the future value of numerical model to realize, therefore it inputs only one dimension of parameter, that is, predicts the historical series of target Data.It is usually designated as ARIMA (p, d, q), and p value indicates that numerical value is meant that the order (time lag number) of autoregression model, d Value indicates the difference number (order) that numerical value meaning is done by stationary sequence, and it is moving average model(MA model) that q value numerical value, which indicates that meaning is, Order.ARIMA model is indicated by following equation:
Wherein,Indicate auto-regressive parameter, θi(i=1,2 ..., q) indicates rolling average parameter, and And εtIt is to meet normal distribution N (0, σ2) bias term.{Xt-1,Xt-2,…,Xt-pIndicate time series input data, XtIt is Prediction result.Table 2 gives the parameter setting of six ARIMA models.
2 six ARIMA model parameter settings of table
Name variable Roll angle Pitch angle HDST VDST HDSH VDSH
(p,d,q) (2,1,2) (5,1,2) (5,1,5) (2,1,5) (5,1,3) (4,1,5)
(2) LSTM model
In deep learning model using LSTM model include at least Three Tiered Network Architecture, i.e. an input layer, one LSTM layers and an output layer.For specific forecasting problem, neuron in the input and output layer of LSTM network can be determined Quantity.Wherein input layer is numerically identical as input variable dimension numerical value (for 32), output layer numerically with Quantity contained by neuron is identical, is equal to 6, i.e. value in six time steps for predicting each target variable.Include 64 The LSTM layer of neuron and full articulamentum comprising 16 neurons, for constructing the middle layer of LSTM model.LSTM model is surplus Remaining hyper parameter setting is consistent with WCNN-LSTM.
(3) WLSTM model
Network structure and the hyper parameter setting of WLSTM model are identical as LSTM model.The difference is that being input to The data of WLSTM model have passed through Wavelet denoising.
As shown in table 3 and Fig. 6, the present invention compares prediction result of four models on test set, counts respectively Each model is calculated in prediction t+1, t+2 ..., the predicted value at t+6 moment and their RMSE (root-mean-square error), is finally demonstrated It is proposed that the validity of method and precision of prediction are much higher than other three kinds of methods in the present invention.Table 3 gives each model and calculates Obtained RMSE value, Fig. 6 are the RMSE mean values of each model.
Table 3 ARIMA, LSTM, WLSTM, the comparison of WCNN-LSTM model prediction accuracy
From table 3 and Fig. 6:
Firstly, for each prediction model, precision of prediction is on a declining curve with the increase of time step.This phenomenon with Practical expectation is consistent, and usually forecast interval is bigger, and the factor that may influence future outcomes is more unknowable, to reduce The accuracy of prediction.Time series models in the present invention are first 10 minutes of input data sequence come 1 minute after predicting value. In engineering practice, thus it is possible to vary the length of output sequence, to explore the maximum predicted time step within the scope of acceptable error It is long.
Secondly, table 3 shows that other than HDST, compared with other three models, WCNN-LSTM model of the invention is almost Optimal precision of prediction is all achieved to the prediction of all variables.It is in Fig. 6 the result shows that, ARIMA, LSTM, WLSTM and The performance of WCNN-LSTM model is gradually increased, the RMSE value being calculated on test set is respectively 0.9219,0.7839, 0.6754 and 0.5602.
The above result shows that the wavelet transformation and convolutional neural networks in mixed model proposed by the present invention play crucial work With.Wavelet transformation eliminates the ambient noise of initial data, and the changing pattern and trend for including data sequence are more easily examined It surveys.CNN has very strong independent learning ability, can detect highly complex and nonlinear data characteristics.CNN is merged automatically Multiple features extract the key feature for influencing prediction result again, and are transmitted to the part LSTM to improve prediction.
Finally, WCNN-LSTM mixed model proposed by the present invention all has relatively strong for precision of prediction and robustness Advantage.For different output variables, it is different that this method, which obtains precision,.For two variables of roll angle and pitch angle Speech, the precision of prediction of four kinds of models do not have especially significant difference.But it for its dependent variable, then has clear improvement, error is maximum It is reduced to 50%.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include Within protection scope of the present invention.

Claims (8)

1. a kind of shield driving pose intelligent Forecasting based on interacting depth study, which is characterized in that including model training Stage and pose forecast period, in which:
The model pre-training stage is carried out based on interacting deep learning model WCNN-LSTM, comprising:
Step 1: data are determining and acquire:
1.1, the input and output parameter of WCNN-LSTM is determined:
By driving speed, permeability, cutter head torque, gross thrust, cutterhead revolving speed, cutterhead electric current, cutterhead total torque, total extruding force, Practical excavated volume, cutter plate driver total current, cutterhead position, propelling cylinder stroke initial value, propelling cylinder thrust, the flexible oil of cutterhead Cylinder stroke, spoil disposal volume flow and bentonite inlet flow rate are as prediction model input variable；
Using shield tail horizontal departure, shield tail vertical missing, shield head horizontal departure, shield head vertical missing, yaw angle and pitch angle as Six output variables of prediction model, for characterizing shield machine pose；
1.2, assume that current time is t, the shield operation data according to specified time interval acquisition t to the t-n moment includes step 1.1 input and output parameters determined therefrom select the data continuously without missing and the complete ring of data as trained sample This, is divided into training set and test set for training sample, n is positive integer；
Step 2: the input parameter input WCNN-LSTM in training set is trained；
Step 3: trained WCNN-LSTM being tested using the input and output parameter of test set, and will be practical The parameter of output and the output parameter of test set carry out deviation comparison, if deviation exceeds preset range, adjust WCNN-LSTM's Internal reference and return step 3；If deviation is within a preset range, training is finished, and it is pre- to obtain the shield driving pose based on WCNN-LSTM Survey model.
The pose forecast period includes:
Step 4: by t+j moment or t+1 to t+j period corresponding input variable, shield driving pose prediction model is inputted, The predicted value of six output variables of the characterization shield machine pose in corresponding t+j moment or t+1 to t+j period is obtained, That is shield machine pose predicted value.
2. a kind of shield driving pose intelligent Forecasting based on interacting depth study as described in claim 1, feature Be, further include as follows based on Prior Control shield pose correction the stage:
Step 5: the deviation of t+j moment or shield machine pose predicted value and design value in t+1 to t+j period is calculated, if Deviation exceeds allowed band, then adjusts the control amount being applied on shield machine in t+j moment or t+1 to t+j period in advance, It operates shield machine and carries out pose adjustment, realization is previously-completed correction operation before shield driving not yet misalignment.
3. a kind of shield driving pose intelligent Forecasting based on interacting depth study as claimed in claim 2, feature It is, the pose adjustment process in step 5 is as follows:
Step 5.1: inscribed when according to t+j or t+1 to t+j period in shield pose predicted value and design value deviation, meter Calculating the corresponding moment needs to be applied to control amount on shield machine, and the control signal of shield machine is corrected according to the control amount；
Step 5.2: according to revised control signal, running in shield machine to when next sampling instant, adjust shield seat in the plane Appearance obtains the actual observed value of shield machine pose adjusted；Resampling simultaneously carries out position according to the input variable that sampling obtains Appearance prediction, the predicted value after obtaining the sampling instant resampling；
Step 5.3: according to the deviation of the predicted value after the actual observed value of shield machine pose adjusted and resampling, further Correct shield driving pose prediction model.
4. a kind of shield driving pose intelligent Forecasting based on interacting depth study as claimed in claim 3, feature It is, in step 5.3, the method for correcting shield driving pose prediction model is as follows: the input variable and tune that resampling is obtained Training sample is added in the actual observed value of pose after whole, and shield driving pose prediction model is trained and is updated.
5. a kind of shield driving pose intelligent predicting system based on interacting depth study, which is characterized in that including processor, mould Type training program module, WCNN-LSTM model and pose Prediction program module；
The model training program module is executed as described in claims 1 to 3 any one when being called by the processor Model training stage is trained WCNN-LSTM model, obtains the shield driving pose prediction model based on WCNN-LSTM；
The pose Prediction program module is executed as described in claims 1 to 3 any one when being called by the processor Pose forecast period, to obtain shield driving pose prediction result based on shield driving pose prediction model.
6. a kind of shield driving pose intelligent predicting system based on interacting depth study as claimed in claim 4, feature It is, further includes pose correction program module；
The pose correction program module executes as claimed in claim 2 or claim 3 when being called by the processor based in advance The shield pose of control is rectified a deviation the stage, to carry out the correction of shield pose based on shield pose prediction result.
7. a kind of shield driving pose intelligent predicting system based on interacting depth study as described in claim 4 or 5, special Sign is, further includes Noise Elimination from Wavelet Transform program module, is used for when being called by the processor, any to claims 1 to 3 The data acquired in step 1 and step 4 described in one, by wavelet transformation carry out data de-noising, to parameter time series into Row decomposes, and to remove ambient noise and systematic measurement error, carries out signal reconstruction then to generate the data sequence after new denoising Column.
8. a kind of shield driving pose intelligent predicting system based on interacting depth study, which is characterized in that including data acquisition With input module, wavelet transformation noise filter module, shield machine pose prediction model module and result output module；
The data acquisition and input module, for obtaining specified shield machine operating parameter, and by the wavelet transformation noise Shield machine operating parameter after filter module denoising inputs the shield machine pose prediction model module；
The wavelet transformation noise filter module, the shield machine for obtaining to the data acquisition and input module, which is run, joins Number carries out data de-noising by wavelet transformation, decomposes to parameter time series, is missed with removing ambient noise and systematic survey Then difference carries out signal reconstruction to generate data sequence after new denoising；
The shield machine pose prediction model module is based on WCNN-LSTM model, for the shield machine operating parameter according to input Predict shield machine pose variation tendency and result；
The result output module, for receiving and showing data, the small echo that the data acquisition and input module obtain Data, the predictive analysis results of the shield machine pose prediction model module after the denoising of converter noise filter module, and Correction scheme.
CN201910364559.5A 2019-04-30 2019-04-30 Intelligent shield tunneling pose prediction method and system based on hybrid deep learning Active CN110195592B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910364559.5A CN110195592B (en) 2019-04-30 2019-04-30 Intelligent shield tunneling pose prediction method and system based on hybrid deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910364559.5A CN110195592B (en) 2019-04-30 2019-04-30 Intelligent shield tunneling pose prediction method and system based on hybrid deep learning

Publications (2)

Publication Number Publication Date
CN110195592A true CN110195592A (en) 2019-09-03
CN110195592B CN110195592B (en) 2021-02-05

Family

ID=67752321

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910364559.5A Active CN110195592B (en) 2019-04-30 2019-04-30 Intelligent shield tunneling pose prediction method and system based on hybrid deep learning

Country Status (1)

Country Link
CN (1) CN110195592B (en)

Cited By (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110578529A (en) * 2019-09-20 2019-12-17 上海隧道工程有限公司 Shield tunneling machine excavation attitude vector self-adaptive adjustment method and system
CN110942210A (en) * 2019-12-11 2020-03-31 中国铁建重工集团股份有限公司 Shield TBM attitude deviation early warning method
CN111075489A (en) * 2020-02-19 2020-04-28 太原理工大学 Attitude description method for floating connection mechanism of hydraulic support and scraper conveyor
CN111209695A (en) * 2019-12-30 2020-05-29 浙江大学 LSTM-based structural dynamic response prediction method
CN111291500A (en) * 2020-03-06 2020-06-16 盾构及掘进技术国家重点实验室 Inversion modeling method based on earth pressure balance shield machine parameter data
CN111461927A (en) * 2020-04-01 2020-07-28 华中科技大学 Tower crane type selection system and method based on BIM and time-varying reliability analysis
CN111488694A (en) * 2020-04-16 2020-08-04 盾构及掘进技术国家重点实验室 Slurry shield tunneling machine tunneling key parameter prediction method based on multiple linear autoregression
CN111577708A (en) * 2020-04-27 2020-08-25 中联重科股份有限公司 Hydraulic oil cylinder piston stroke control method, equipment and system and hydraulic machinery
CN111636891A (en) * 2020-06-08 2020-09-08 中铁高新工业股份有限公司 Real-time shield attitude prediction system and construction method of prediction model
CN112016193A (en) * 2020-08-07 2020-12-01 震坤行工业超市(上海)有限公司 Online prediction method and system for lubrication failure of shield tunneling machine system
CN112100841A (en) * 2020-09-09 2020-12-18 中铁二十局集团有限公司 Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium
CN112796768A (en) * 2021-03-08 2021-05-14 盾构及掘进技术国家重点实验室 Construction tunneling parameter selection method for dual-mode tunnel boring machine
CN112879024A (en) * 2021-01-23 2021-06-01 西安建筑科技大学 Dynamic prediction method, system and equipment for shield attitude
CN113139340A (en) * 2021-04-23 2021-07-20 中铁十六局集团北京轨道交通工程建设有限公司 Real-time prediction method, terminal and medium for shield tunneling line
CN113282990A (en) * 2021-05-31 2021-08-20 中铁十六局集团北京轨道交通工程建设有限公司 Intelligent real-time determination method, terminal and medium for shield motion trail
CN113298220A (en) * 2021-05-31 2021-08-24 中铁十六局集团北京轨道交通工程建设有限公司 Neural network optimization-based shield tunneling machine tunneling speed prediction method
CN113311750A (en) * 2021-05-21 2021-08-27 上海大学 Automatic tunneling target cooperative control system and method
CN113344256A (en) * 2021-05-21 2021-09-03 上海隧道工程有限公司 System and method for predicting movement characteristics and evaluating control performance of multiple degrees of freedom of shield attitude
CN113374488A (en) * 2021-07-28 2021-09-10 中国铁建重工集团股份有限公司 Earth pressure balance shield machine guiding control method and device and readable storage medium
CN113408080A (en) * 2021-07-26 2021-09-17 中国铁建重工集团股份有限公司 Soil pressure dynamic characteristic modeling method, shield tunneling machine control system and shield tunneling machine
CN113435055A (en) * 2021-07-08 2021-09-24 上海交通大学 Self-adaptive migration prediction method and system in shield cutter head torque field
CN113486463A (en) * 2021-07-02 2021-10-08 中铁工程装备集团有限公司 Shield optimal autonomous tunneling control method based on deep reinforcement learning
CN113673059A (en) * 2021-08-26 2021-11-19 济南轨道交通集团有限公司 Shield tunneling parameter prediction method based on random forest and BP neural network
CN114117599A (en) * 2021-11-22 2022-03-01 中铁高新工业股份有限公司 Shield attitude position deviation prediction method
CN114233323A (en) * 2021-12-03 2022-03-25 中国水利水电第八工程局有限公司 Shield tunnel forward design method, system and medium based on BIM
CN114329810A (en) * 2021-11-16 2022-04-12 中国水利水电科学研究院 Real-time shield tunneling machine working attitude prediction method based on big data
CN114593735A (en) * 2022-01-26 2022-06-07 奥比中光科技集团股份有限公司 Pose prediction method and device
CN114636400A (en) * 2022-03-04 2022-06-17 华中科技大学 ANN deep learning-based relative pose measurement method and system between multiple shield bodies
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
CN117189239A (en) * 2023-09-07 2023-12-08 中国矿业大学 Tunnel surrounding rock damage monitoring method

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003240552A (en) * 2002-02-20 2003-08-27 Jfe Engineering Kk Method and device for propelling head position/direction measurement
JP2011168996A (en) * 2010-02-17 2011-09-01 Yasuda Engineering Kk Equipment for controlling excavation direction of shield machine for jacking shield tunneling method
JP2013113829A (en) * 2011-12-01 2013-06-10 Raito Kogyo Co Ltd Hole trajectory measuring instrument and measurement method
CN104101326A (en) * 2013-04-08 2014-10-15 同济大学 Automatic measurement method and apparatus for shield attitude based on spatial geometry analysis
CN105350971A (en) * 2015-09-30 2016-02-24 上海隧道工程有限公司 Pipe-roofing push bench provided with lock guiding devices and attitude control method thereof
CN104457719B (en) * 2014-12-28 2016-07-13 上海隧道工程有限公司 The attitude measuring of class rectangular shield construction and measuring method
CN106481344A (en) * 2016-10-25 2017-03-08 华中科技大学 A kind of balancing earth-pressure shielding machine tunnel piercing parameter intelligent control method
CN108614944A (en) * 2018-05-10 2018-10-02 西安电子科技大学 A kind of shield track axis correction parameter prediction technique
CN108710940A (en) * 2017-12-31 2018-10-26 中交第公路工程局有限公司 Method based on shield machine running orbit parameter in Neural Network Optimization dust stratum
CN108798692A (en) * 2018-06-12 2018-11-13 上海隧道工程有限公司 The automatic attitude rectification system of shield machine and modification method
CN108868805A (en) * 2018-06-08 2018-11-23 西安电子科技大学 Shield method for correcting error based on statistical analysis in conjunction with XGboost
CN108868807A (en) * 2018-09-07 2018-11-23 上海隧道工程有限公司 The intelligent control method of shield driving correction
CN109446716A (en) * 2018-11-15 2019-03-08 中铁第六勘察设计院集团有限公司 Based on BIM platform shield machine VMT and correction scheme three-dimensional display method
CN109519184A (en) * 2018-10-22 2019-03-26 中铁建大桥工程局集团第二工程有限公司 A kind of construction method of advance borehole control attitude of shield machine drift
CN109605372A (en) * 2018-12-20 2019-04-12 中国铁建重工集团有限公司 A kind of method and system of the pose for survey engineering mechanical arm

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003240552A (en) * 2002-02-20 2003-08-27 Jfe Engineering Kk Method and device for propelling head position/direction measurement
JP2011168996A (en) * 2010-02-17 2011-09-01 Yasuda Engineering Kk Equipment for controlling excavation direction of shield machine for jacking shield tunneling method
JP2013113829A (en) * 2011-12-01 2013-06-10 Raito Kogyo Co Ltd Hole trajectory measuring instrument and measurement method
CN104101326A (en) * 2013-04-08 2014-10-15 同济大学 Automatic measurement method and apparatus for shield attitude based on spatial geometry analysis
CN104457719B (en) * 2014-12-28 2016-07-13 上海隧道工程有限公司 The attitude measuring of class rectangular shield construction and measuring method
CN105350971A (en) * 2015-09-30 2016-02-24 上海隧道工程有限公司 Pipe-roofing push bench provided with lock guiding devices and attitude control method thereof
CN106481344A (en) * 2016-10-25 2017-03-08 华中科技大学 A kind of balancing earth-pressure shielding machine tunnel piercing parameter intelligent control method
CN108710940A (en) * 2017-12-31 2018-10-26 中交第公路工程局有限公司 Method based on shield machine running orbit parameter in Neural Network Optimization dust stratum
CN108614944A (en) * 2018-05-10 2018-10-02 西安电子科技大学 A kind of shield track axis correction parameter prediction technique
CN108868805A (en) * 2018-06-08 2018-11-23 西安电子科技大学 Shield method for correcting error based on statistical analysis in conjunction with XGboost
CN108798692A (en) * 2018-06-12 2018-11-13 上海隧道工程有限公司 The automatic attitude rectification system of shield machine and modification method
CN108868807A (en) * 2018-09-07 2018-11-23 上海隧道工程有限公司 The intelligent control method of shield driving correction
CN109519184A (en) * 2018-10-22 2019-03-26 中铁建大桥工程局集团第二工程有限公司 A kind of construction method of advance borehole control attitude of shield machine drift
CN109446716A (en) * 2018-11-15 2019-03-08 中铁第六勘察设计院集团有限公司 Based on BIM platform shield machine VMT and correction scheme three-dimensional display method
CN109605372A (en) * 2018-12-20 2019-04-12 中国铁建重工集团有限公司 A kind of method and system of the pose for survey engineering mechanical arm

Cited By (43)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110578529A (en) * 2019-09-20 2019-12-17 上海隧道工程有限公司 Shield tunneling machine excavation attitude vector self-adaptive adjustment method and system
CN110578529B (en) * 2019-09-20 2021-02-09 上海隧道工程有限公司 Shield tunneling machine excavation attitude vector self-adaptive adjustment method and system
CN110942210A (en) * 2019-12-11 2020-03-31 中国铁建重工集团股份有限公司 Shield TBM attitude deviation early warning method
CN110942210B (en) * 2019-12-11 2023-02-07 中国铁建重工集团股份有限公司 Shield TBM attitude deviation early warning method
CN111209695A (en) * 2019-12-30 2020-05-29 浙江大学 LSTM-based structural dynamic response prediction method
CN111075489A (en) * 2020-02-19 2020-04-28 太原理工大学 Attitude description method for floating connection mechanism of hydraulic support and scraper conveyor
CN111291500A (en) * 2020-03-06 2020-06-16 盾构及掘进技术国家重点实验室 Inversion modeling method based on earth pressure balance shield machine parameter data
CN111461927B (en) * 2020-04-01 2023-04-18 华中科技大学 Tower crane type selection system and method based on BIM and time-varying reliability analysis
CN111461927A (en) * 2020-04-01 2020-07-28 华中科技大学 Tower crane type selection system and method based on BIM and time-varying reliability analysis
CN111488694A (en) * 2020-04-16 2020-08-04 盾构及掘进技术国家重点实验室 Slurry shield tunneling machine tunneling key parameter prediction method based on multiple linear autoregression
CN111577708A (en) * 2020-04-27 2020-08-25 中联重科股份有限公司 Hydraulic oil cylinder piston stroke control method, equipment and system and hydraulic machinery
CN111577708B (en) * 2020-04-27 2022-05-20 中联重科股份有限公司 Hydraulic oil cylinder piston stroke control method, equipment and system and hydraulic machine
CN111636891A (en) * 2020-06-08 2020-09-08 中铁高新工业股份有限公司 Real-time shield attitude prediction system and construction method of prediction model
CN112016193B (en) * 2020-08-07 2021-05-04 震坤行工业超市(上海)有限公司 Online prediction method and system for lubrication failure of shield tunneling machine system
CN112016193A (en) * 2020-08-07 2020-12-01 震坤行工业超市(上海)有限公司 Online prediction method and system for lubrication failure of shield tunneling machine system
CN112100841A (en) * 2020-09-09 2020-12-18 中铁二十局集团有限公司 Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium
CN112100841B (en) * 2020-09-09 2024-04-19 中铁二十局集团有限公司 Method and device for predicting attitude of shield tunneling machine, terminal equipment and storage medium
CN112879024A (en) * 2021-01-23 2021-06-01 西安建筑科技大学 Dynamic prediction method, system and equipment for shield attitude
CN112879024B (en) * 2021-01-23 2024-01-16 西安建筑科技大学 Dynamic prediction method, system and equipment for shield attitude
CN112796768A (en) * 2021-03-08 2021-05-14 盾构及掘进技术国家重点实验室 Construction tunneling parameter selection method for dual-mode tunnel boring machine
CN113139340A (en) * 2021-04-23 2021-07-20 中铁十六局集团北京轨道交通工程建设有限公司 Real-time prediction method, terminal and medium for shield tunneling line
CN113139340B (en) * 2021-04-23 2023-08-22 中铁十六局集团北京轨道交通工程建设有限公司 Real-time prediction method, terminal and medium for shield tunneling line
CN113344256B (en) * 2021-05-21 2024-03-22 上海隧道工程有限公司 System and method for predicting multiple degrees of freedom motion characteristics of shield attitude and evaluating control performance
CN113344256A (en) * 2021-05-21 2021-09-03 上海隧道工程有限公司 System and method for predicting movement characteristics and evaluating control performance of multiple degrees of freedom of shield attitude
CN113311750A (en) * 2021-05-21 2021-08-27 上海大学 Automatic tunneling target cooperative control system and method
CN113282990A (en) * 2021-05-31 2021-08-20 中铁十六局集团北京轨道交通工程建设有限公司 Intelligent real-time determination method, terminal and medium for shield motion trail
CN113298220B (en) * 2021-05-31 2023-08-04 中铁十六局集团北京轨道交通工程建设有限公司 Neural network optimization-based shield tunneling speed prediction method
CN113298220A (en) * 2021-05-31 2021-08-24 中铁十六局集团北京轨道交通工程建设有限公司 Neural network optimization-based shield tunneling machine tunneling speed prediction method
CN113486463A (en) * 2021-07-02 2021-10-08 中铁工程装备集团有限公司 Shield optimal autonomous tunneling control method based on deep reinforcement learning
CN113435055A (en) * 2021-07-08 2021-09-24 上海交通大学 Self-adaptive migration prediction method and system in shield cutter head torque field
CN113408080A (en) * 2021-07-26 2021-09-17 中国铁建重工集团股份有限公司 Soil pressure dynamic characteristic modeling method, shield tunneling machine control system and shield tunneling machine
CN113374488A (en) * 2021-07-28 2021-09-10 中国铁建重工集团股份有限公司 Earth pressure balance shield machine guiding control method and device and readable storage medium
CN113673059A (en) * 2021-08-26 2021-11-19 济南轨道交通集团有限公司 Shield tunneling parameter prediction method based on random forest and BP neural network
CN114329810A (en) * 2021-11-16 2022-04-12 中国水利水电科学研究院 Real-time shield tunneling machine working attitude prediction method based on big data
CN114329810B (en) * 2021-11-16 2024-04-16 中国水利水电科学研究院 Real-time prediction method for working posture of shield tunneling machine based on big data
CN114117599A (en) * 2021-11-22 2022-03-01 中铁高新工业股份有限公司 Shield attitude position deviation prediction method
CN114233323A (en) * 2021-12-03 2022-03-25 中国水利水电第八工程局有限公司 Shield tunnel forward design method, system and medium based on BIM
CN114593735A (en) * 2022-01-26 2022-06-07 奥比中光科技集团股份有限公司 Pose prediction method and device
CN114636400A (en) * 2022-03-04 2022-06-17 华中科技大学 ANN deep learning-based relative pose measurement method and system between multiple shield bodies
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
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
CN117189239A (en) * 2023-09-07 2023-12-08 中国矿业大学 Tunnel surrounding rock damage monitoring method
CN117189239B (en) * 2023-09-07 2024-04-19 中国矿业大学 Tunnel surrounding rock damage monitoring method

Also Published As

Publication number Publication date
CN110195592B (en) 2021-02-05

Similar Documents

Publication Publication Date Title
CN110195592A (en) Shield driving pose intelligent Forecasting and system based on interacting depth study
CN107861508B (en) Local motion planning method and device for mobile robot
CN112052992B (en) Deep learning-based construction project progress prediction system and method
Shen et al. Real-time prediction of shield moving trajectory during tunnelling
CN110096827B (en) Shield tunneling machine parameter optimization method based on deep neural network
Jin et al. An accurate and adaptative cutterhead torque prediction method for shield tunneling machines via adaptative residual long-short term memory network
Fu et al. A hybrid deep learning approach for dynamic attitude and position prediction in tunnel construction considering spatio-temporal patterns
CN113128109B (en) Test and evaluation method for intelligent fully-mechanized mining robot production system
Wang et al. Prediction of axis attitude deviation and deviation correction method based on data driven during shield tunneling
CN115659749B (en) Foundation pit deformation prediction method and system, electronic equipment and storage medium
CN115422814B (en) Digital twin-driven closed-loop optimization design method for complex electromechanical product
CN114810100B (en) Shield tunneling attitude prediction method based on deep neural network
Fei et al. Research on tunnel engineering monitoring technology based on BPNN neural network and MARS machine learning regression algorithm
CN108595803A (en) Shale gas well liquid loading pressure prediction method based on recurrent neural network
CN112733440A (en) Intelligent fault diagnosis method, system, storage medium and equipment for offshore oil-gas-water well
CN115471679A (en) Method and intelligent system for synchronously assimilating water level and flow of natural river
CN115455727A (en) Digital twin-driven fully-mechanized mining equipment and coal bed coupling system optimization decision method
CN107368463B (en) Roadway nonlinear deformation prediction method based on fiber bragg grating sensor network data
CN112182955A (en) Construction method and application of shield general type segment point selection model
CN115481565A (en) Earth pressure balance shield tunneling parameter prediction method based on LSTM and ant colony algorithm
CN114019795A (en) Shield tunneling deviation rectifying intelligent decision-making method based on reinforcement learning
CN116933441A (en) Digital twin data driving model modeling method for autonomous shoveling process of loader
Kubota et al. Autopilot model for shield tunneling machines using support vector regression and its application to previously constructed tunnels
CN113703025B (en) GNSS (Global navigation satellite System) multiple failure state-oriented intelligent vehicle positioning error prediction method
CN115773127A (en) Intelligent decision-making method, system, equipment and medium for slurry balance shield

Legal Events

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