CN110084322A - A kind of prediction technique of shield machine boring parameter neural network based - Google Patents
A kind of prediction technique of shield machine boring parameter neural network based Download PDFInfo
- Publication number
- CN110084322A CN110084322A CN201910382958.4A CN201910382958A CN110084322A CN 110084322 A CN110084322 A CN 110084322A CN 201910382958 A CN201910382958 A CN 201910382958A CN 110084322 A CN110084322 A CN 110084322A
- Authority
- CN
- China
- Prior art keywords
- data
- layer
- neuron
- neural network
- output
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Neurology (AREA)
- Excavating Of Shafts Or Tunnels (AREA)
Abstract
The invention discloses a kind of prediction techniques of shield machine boring parameter neural network based.Method of the invention proposes the method that shield engineering big data is carried out feature learning by machine learning method after over cleaning and statistical disposition, by means of the data high-level characteristic Extracting Ability of machine learning algorithm, to operating parameter and operational efficiency High Efficiency Modeling, to make the suggestion of directiveness to the parameter regulation of shield in engineering, the characteristic with artificial intelligence.
Description
Technical field
The present invention relates to a kind of prediction techniques of shield machine boring parameter neural network based.
Background technique
Shield machine is named as shielding tunnel excavator entirely, is that a kind of special engineering of tunnel piercing is mechanical, has and excavate cutting
The functions such as the soil body, conveying soil quarrel, assembled tunnel-liner, synchronous grouting, measurement guiding correction.It needs to set in shield machine operational process
Various parameters are set, and the quality of parameter setting can directly affect the driving speed of shield machine.Therefore suitable method is selected
It precisely and effectively predicts boring parameter, has important practical significance.
With the development of data mining technology, the boring parameter sets method based on machine learning is gradually taken seriously.Mesh
It is preceding existing, using SVM, the mathematical prediction model of compound stratum shield machine fltting speed is established, improves the driving speed of shield machine
Degree；Shield driving parameter analogy setting method (SAPAS) is proposed in conjunction with K-means algorithm according to by empirical equation setting, is led to
It crosses and the data of history is clustered, realize the extraction to boring parameter and Auto-matching；Also have using simple network, to shield
In structure machine tunneling process the sedimentation of earth's surface carry out prediction and to constructing tunnel during carried out sedimentation modeling and settlement prediction.
But in practical application, still there is model and be difficult to train in these methods；It is accurate in parameter prediction value to train the model come
It is poor to spend；The problems such as data user rate that shield machine generates is low.
Shield machine boring parameter is predicted in view of the above-mentioned problems, can use neural network.Neural network is a kind of
Operational model is constituted by being coupled to each other between a large amount of node.A kind of each specific output function of node on behalf, is referred to as motivated
Function.Connection between every two node all represents a weighted value for passing through the connection signal, referred to as weight, this is quite
In the memory of artificial neural network.Exactly because there are these features, neural network is made to possess extremely strong learning ability and fitting energy
Power.Neural network preferably can sufficiently learn the data of shield machine generation, and can train more stable model,
And the more traditional machine learning of model performance is also more also excellent.
Summary of the invention
It is an object of the present invention to provide a kind of prediction techniques of shield machine boring parameter neural network based, are surveying
It tries under data, the model predication value is consistent with initial data changing rule, and mean error is within 12%, and model structure light weight
Grade, generalization ability meet site operation requirement, provide a set of effective scheme for the prediction of shield machine boring parameter, have certain
Practical value.
Technical solution to facilitate the understanding of the present invention carries out the nerual network technique used in the present invention program following
Illustrate:
Neural network is a kind of operational model, is constituted by being coupled to each other between a large amount of node (or neuron).Each
A kind of specific output function of node on behalf, referred to as excitation function (activation function).Company between every two node
It connects and all represents a weighted value for passing through the connection signal, referred to as weight, this is equivalent to the memory of artificial neural network.
The output of network then according to the connection type of network, the difference of weighted value and excitation function and it is different.And network itself is usually all
Certain algorithm of nature or function are approached, it is also possible to the expression to a kind of logic strategy.
Recently during the last ten years, the research work of artificial neural network deepens continuously, and has been achieved for very big progress,
The fields such as pattern-recognition, intelligent robot, automatic control, predictive estimation, biology, medicine, economy have successfully solved many
The insoluble practical problem of modern computer, shows good intelligent characteristic.
The research of neural network can be divided into two broad aspect of theoretical research and application study.
Theoretical research can be divided into following two categories:
1, nervous physiology and cognitive science research human thinking and intelligent mechanism are utilized.
2, using the research achievement of neural basal theory, with mathematics Research on Methods function is more perfect, performance is more superior
Neural network model, further investigate network algorithm and performance, such as: stability, convergence, fault-tolerance, robustness；Exploitation is new
Network mathematical theory, such as: neural network dynamics, non-linear neural field.
Application study can be divided into following two categories:
1, the software simulation and hard-wired research of neural network.
2, the research that neural network is applied in every field.These fields specifically include that pattern-recognition, signal processing,
Knowledge engineering, expert system, optimum organization, robot control etc..With neural network theory itself and correlation theory, correlation
The application of the continuous development of technology, neural network will more be goed deep into surely.
The technical solution of the present invention is as follows:
A kind of preprocess method based on shield machine big data, comprising the following steps:
Step 1: collecting sample data acquire certain city's specific model in conjunction with actual construction experience and historgraphic data recording
One section of excavation historical data of shield machine model.
Step 2: sample data is subjected to data point and data processing.
Data analysis phase has carried out Data Dimensionality Reduction, correlation analysis and feature extraction to initial data, and special to extracting
The data of sign have done comprehensive analysis.Data processing stage mainly carries out data smoothing operations to treated data.Through
After crossing the above processing, it will imperfect, inconsistent edge data in removal initial data extracts the higher spy of correlation
Sign, to improve the quality of data and model to the utilization rate of data.
Step 3: determining the input and output vector of model.
Input data is constituted by characteristic, input data is normalized respectively using minimax method, is turned
Turn to the value between [0,1].
Step 4: the structure of initialization multilayer DNN model, the hidden layer of model are 10 layers, maximum number of iterations 2000,
Activation primitive is sigmoid function, and each hidden layer neuron number is identical.
Step 5: model being carried out using step 3 pretreated input data using back-propagation algorithm is carried out at any time
Training, specific as follows:
1) output valve of each neuron of forward calculation；
2) the error entry value of each neuron of retrospectively calculate, it is partial derivative of the loss function to neuron weighting input；
3) gradient of each weight is calculated, then updates weight with batch gradient descent algorithm, judges whether loss function is received
Hold back or whether reach maximum number of iterations, loss function is not converged and not up to maximum number of iterations then returns to the 1) step；
If 4) reach maximum number of iterations, and the loss function of model no longer changes substantially, then training terminates.?
The model thought is optimal models, can be applied to the parameter prediction of shield machine.
Compared with the prior art, the advantages of the present invention are as follows:
One, the big data processing method based on machine learning is applied in the processing of shield project data for the first time, it is different
In the data digging method based on statistics.
Two, it proposes and shield engineering big data is subjected to feature by machine learning method after over cleaning and statistical disposition
The method of study, it is efficient to operating parameter and operational efficiency by means of the data high-level characteristic Extracting Ability of machine learning algorithm
Modeling, so that the suggestion of directiveness is made to the parameter regulation of shield in engineering, the characteristic with artificial intelligence.
Three, more intelligent deep learning algorithm is extracted applied to data model, utilizes the powerful data of deep learning
Fitting learning ability may be implemented to effectively improve the operational efficiency of shield, therefore deep learning is applied to propose engineering efficiency
It is the important innovative point of the present invention in the problem of liter.
Detailed description of the invention
Fig. 1 is that data of the invention analyze content；
Fig. 2 is data processing content of the invention；
Fig. 3 is BP neural network model of the invention；
Fig. 4 is that the present invention is based on the parameter prediction result figures of neural network.
Specific embodiment
Bright technical solution of the present invention is described in detail with reference to the accompanying drawing.
In order to intuitively embody the practicability of the present invention program, in following scheme, data is divided into training set simultaneously and are surveyed
Examination collection, by the test to test set, to prove the practicability of the present invention program；
The prediction technique of shield machine boring parameter neural network based is as follows:
(1) collecting sample data acquire certain city's specific model shield in conjunction with actual construction experience and historgraphic data recording
One section of excavation historical data of type number.Sample data is subjected to data point and data processing.Data analysis phase, such as Fig. 1, it is right
Initial data has carried out Data Dimensionality Reduction, correlation analysis and feature extraction, and has done comprehensive analysis to the data for extracting feature.Number
According to processing stage, such as Fig. 2, data smoothing operations mainly are carried out to treated data.After handling above, it will
Edge data imperfect, inconsistent in initial data is removed, the higher feature of correlation is extracted, to improve the quality of data
With model to the utilization rate of data.
(2) the input and output vector of model is determined.The feature of input includes ring number, general power, cutter head torque, cutterhead pressure
Power, cutterhead revolving speed, propelling pressure, overall driving force, the current cumulative amount of foam mixing liquid, left support pressure, upper left support pressure,
Lower-left support pressure, bottom right support pressure, the right side in support pressure；Output valve of the shield machine boring parameter as model.Then will
Data set is divided into training set and test set according to " reserving method "；Training set and test set are carried out respectively using minimax method
Normalized, the value being converted between [0,1].
(3) structure of multilayer DNN model is initialized, as shown in figure 3, the hidden layer of model is 10 layers, maximum number of iterations
It is 2000, activation primitive is sigmoid function, and each hidden layer neuron number is identical.Then reversed using carrying out at any time
Propagation algorithm is trained using pretreated trained the set pair analysis model.
(4) X indicates training sample, X=[X1,X2,…Xi…Xn], wherein XiIndicate a wherein record, Xi=[x1,
x2,…,xk],xiIt indicates a kind of input feature vector, shares k input feature vector；Export result Y=[y1,y2,…,yi,…,yn].It is first
First, the output valve of each neuron of forward calculation；The calculation formula of input layer are as follows: net=w1x1+w2x2+…+wkxkIn formula: net
For summation unit；wiFor neuron weight, i=1,2 ..., k.The calculation formula of output layer are as follows:f
It (net) is sigmoid activation primitive.
(5) then, the error entry value of each neuron of retrospectively calculate, it is loss function to the inclined of neuron weighting input
Derivative.Wherein output error function are as follows:In formula: dtIt is defeated for t-th of neuron expectation of input layer
Out, ytFor the output valve of neural network, m is the number of output layer neuron, t=1,2 ..., m.
(6) secondly, calculating the gradient of each weight, then with batch gradient descent algorithm update weight, judge loss function
Maximum number of iterations whether is restrained or whether reaches, loss function is not converged and not up to maximum number of iterations then returns to
(4) otherwise step terminates to train.Weight is adjusted according to gradient function when training, thus obtain hidden layer and output layer it
Between adjustment weight:
In formula: η is adjustment factor, 0 < η < 1.vjtIndicate j-th of unit of unit hidden layer to t-th of output layer of connection weight
Value, l and t indicate the neuron number of respective layer.Weight between hidden layer and input layer adjusts calculation formula are as follows:
Wherein, wijConnection weight for i-th of unit of input layer to j-th of unit of output layer is bjFor j-th unit
Biasing, n and l indicate the neuron number of respective layer.
(7) finally, being predicted using the neural network prediction model that training is completed, and sample predictions result and mind are calculated
Through the relative error between network simulation result.Specific data are had chosen between certain 115 ring of ring -1807 of city, section, somewhere
The data of total 1693 rings are trained as sample data, and the data of later 50 ring are as prediction data.As a result such as Fig. 4 institute
Show.
Claims (1)
1. a kind of prediction technique of shield machine boring parameter neural network based, which comprises the following steps:
S1, to shield machine to be predicted, according to its history construction note collecting sample data, the sample data includes the shield
The ring number of machine, general power, cutter head torque, cutterhead pressure, cutterhead revolving speed, propelling pressure, overall driving force, foam mixing liquid currently tire out
Metering, left support pressure, upper left support pressure, lower-left support pressure, bottom right support pressure, the right side in support pressure；
S2, data analysis and data processing are carried out to sample data, the data analysis be sample data is carried out Data Dimensionality Reduction,
Correlation analysis and feature extraction, the data processing are to carry out data smoothing operations to the data obtained after data analysis, from
And edge data imperfect, inconsistent in initial data is removed, and extract the higher characteristic of correlation, composing training
Data, while training data is normalized；
S3, observation index and prediction index are splitted data into, using observation index as input, prediction index is as output, to mind
It is trained through network, the neural network includes input layer, hidden layer and output layer, and the activation primitive of neural network is
Sigmoid function is trained model using training data using back-propagation algorithm is carried out at any time, specifically:
S31, X is enabled to indicate training sample, X=[X1, X2... Xi...Xn], wherein XiIndicate a wherein record, Xi=[x1,
x2..., xk],xiIt indicates a kind of input feature vector, shares k input feature vector；Export result Y=[y1, y2..., yi..., yn]；
Firstly, the output valve of each neuron of forward calculation；The calculation formula of input layer are as follows:
Net=w1x1+w2x2+…+wkxk
Wherein, net is summation unit, wiFor neuron weight, i=1,2 ..., k；
The calculation formula of output layer are as follows:
Wherein, f (net) is sigmoid activation primitive；
The error entry value of each neuron of S32, retrospectively calculate, it is partial derivative of the loss function to neuron weighting input, defeated
Entering t-th of neuron desired output of layer is dt, then output error function are as follows:
Wherein, ytFor the output valve of neural network, m is the number of output layer neuron, t=1,2 ..., m；
S33, the gradient for calculating each weight, then weight is updated with batch gradient descent algorithm:
Wherein, η is adjustment factor, 0 < η < 1, vjtIndicate j-th of unit of unit hidden layer to t-th of output layer of connection weight, l
The neuron number that respective layer is indicated with t, the weight between hidden layer and input layer, which adjusts, to be calculated are as follows:
Wherein, wijConnection weight for i-th of unit of input layer to j-th of unit of output layer is bjFor the inclined of j-th unit
It sets, the neuron number of n and l expression respective layer；
Judge maximum number of iterations is restrained or whether reached to loss function whether, loss function is not converged and not up to maximum changes
Generation number then return step S31；
If convergence reaches maximum number of iterations, training training terminates, and obtained model is defined as optimal models, uses
In the parameter prediction of shield machine.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910382958.4A CN110084322A (en) | 2019-05-09 | 2019-05-09 | A kind of prediction technique of shield machine boring parameter neural network based |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910382958.4A CN110084322A (en) | 2019-05-09 | 2019-05-09 | A kind of prediction technique of shield machine boring parameter neural network based |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110084322A true CN110084322A (en) | 2019-08-02 |
Family
ID=67419297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910382958.4A Pending CN110084322A (en) | 2019-05-09 | 2019-05-09 | A kind of prediction technique of shield machine boring parameter neural network based |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084322A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111832223A (en) * | 2020-06-29 | 2020-10-27 | 上海隧道工程有限公司 | Neural network-based shield construction surface subsidence 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 |
| CN113516287A (en) * | 2021-05-17 | 2021-10-19 | 广东粤海珠三角供水有限公司 | Correlation method of geological parameters and shield construction key parameters based on Elman neural network model |
| CN113535748A (en) * | 2021-07-02 | 2021-10-22 | 中铁十五局集团有限公司 | Shield tunneling machine model selection system and method based on historical cases |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095962A (en) * | 2015-07-27 | 2015-11-25 | 中国汽车工程研究院股份有限公司 | Method for predicting dynamic mechanical property of material based on BP artificial neural network |
| CN107101813A (en) * | 2017-04-26 | 2017-08-29 | 河北工业大学 | A kind of frame-type circuit breaker mechanical breakdown degree assessment method based on vibration signal |
-
2019
- 2019-05-09 CN CN201910382958.4A patent/CN110084322A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105095962A (en) * | 2015-07-27 | 2015-11-25 | 中国汽车工程研究院股份有限公司 | Method for predicting dynamic mechanical property of material based on BP artificial neural network |
| CN107101813A (en) * | 2017-04-26 | 2017-08-29 | 河北工业大学 | A kind of frame-type circuit breaker mechanical breakdown degree assessment method based on vibration signal |
Non-Patent Citations (4)
| Title |
|---|
| ZYVV: "反向传播算法（过程及公式推导）", 《HTTPS://WWW.CNBLOGS.COM/WLZY/P/7751297.HTML》 * |
| 周纯择 等: "南昌上软下硬地层中盾构施工地表沉降的BP神经网络预测方法", 《防灾减灾工程学报》 * |
| 李超 等: "基于BP神经网络的复合地层盾构掘进参数预测与分析", 《土木工程学报》 * |
| 牟松 等: "基于权重优化神经网络的盾构机掘进参数预测方法", 《中国工程机械学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111832223A (en) * | 2020-06-29 | 2020-10-27 | 上海隧道工程有限公司 | Neural network-based shield construction surface subsidence prediction method |
| CN113516287A (en) * | 2021-05-17 | 2021-10-19 | 广东粤海珠三角供水有限公司 | Correlation method of geological parameters and shield construction key parameters based on Elman neural network model |
| 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 |
| CN113535748A (en) * | 2021-07-02 | 2021-10-22 | 中铁十五局集团有限公司 | Shield tunneling machine model selection system and method based on historical cases |
| CN113535748B (en) * | 2021-07-02 | 2024-05-07 | 中铁十五局集团有限公司 | Shield tunneling machine type selection system and type selection method based on historical cases |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084322A (en) | A kind of prediction technique of shield machine boring parameter neural network based | |
| CN110096827B (en) | Shield tunneling machine parameter optimization method based on deep neural network | |
| Chen et al. | Global and local surrogate-model-assisted differential evolution for waterflooding production optimization | |
| Liu et al. | A deep-learning-based prediction method of the estimated ultimate recovery (EUR) of shale gas wells | |
| Pan et al. | Data-driven multi-output prediction for TBM performance during tunnel excavation: An attention-based graph convolutional network approach | |
| CN102622418B (en) | Prediction device and equipment based on BP (Back Propagation) nerve network | |
| CN103870677B (en) | Setting method for tunneling parameters of tunneling machine | |
| CN109635461A (en) | A kind of application carrys out the method and system of automatic identification Grades of Surrounding Rock with brill parameter | |
| CN102622515B (en) | A kind of weather prediction method | |
| Zhang et al. | Integrated optimization design for horizontal well placement and fracturing in tight oil reservoirs | |
| Liu et al. | Multi-objective optimization control for tunnel boring machine performance improvement under uncertainty | |
| Forouzanfar et al. | Simultaneous and sequential estimation of optimal placement and controls of wells with a covariance matrix adaptation algorithm | |
| KR101967978B1 (en) | Apparatus for predicting net penetration rate of shield tunnel boring machine and method thereof | |
| CN103984788B (en) | A kind of coal entry anchor rod support automated intelligent design and optimization system | |
| CN106481344A (en) | A kind of balancing earth-pressure shielding machine tunnel piercing parameter intelligent control method | |
| CN110069893A (en) | A kind of prediction technique of the shield machine boring parameter based on polynomial regression | |
| CN113435128A (en) | Oil and gas reservoir yield prediction method and device based on condition generation type countermeasure network | |
| CN116084892B (en) | Automatic perforation system based on fuzzy neural network | |
| CN118191967B (en) | Intelligent early warning system and method for full-space three-dimensional monitoring of water damage risk of coal seam roof | |
| CN115773127A (en) | Intelligent decision-making method, system, equipment and medium for slurry balance shield | |
| CN107939372A (en) | For the optimal well location dispositions method and device of craven fault block reservoirs | |
| Bai et al. | A new measuring method of dredging concentration based on hybrid ensemble deep learning technique | |
| Fan et al. | Performance optimization of hard rock tunnel boring machine using multi-objective evolutionary algorithm | |
| Liu et al. | Autonomous intelligent control of earth pressure balance shield machine based on deep reinforcement learning | |
| Qiu et al. | An optimized system of random forest model by global harmony search with generalized opposition-based learning for forecasting TBM advance rate |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190802 |