CN110096827B - Shield tunneling machine parameter optimization method based on deep neural network - Google Patents
Shield tunneling machine parameter optimization method based on deep neural network Download PDFInfo
- Publication number
- CN110096827B CN110096827B CN201910382961.6A CN201910382961A CN110096827B CN 110096827 B CN110096827 B CN 110096827B CN 201910382961 A CN201910382961 A CN 201910382961A CN 110096827 B CN110096827 B CN 110096827B
- Authority
- CN
- China
- Prior art keywords
- data
- layer
- model
- weight
- neural network
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Geometry (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Excavating Of Shafts Or Tunnels (AREA)
Abstract
The invention discloses a shield tunneling machine parameter optimization method based on a deep neural network. The invention utilizes a neural network algorithm to establish a forward model of the shield tunneling machine excavation parameters for predicting complex geological conditions, if an expected output value cannot be obtained in an output layer, the sum of squares of an output error and an expected error is taken as a target function, backward propagation is carried out, partial derivatives of the target function to each neuron weight are calculated layer by layer to form a gradient of the target function to the weight vector, the gradient is used as a basis for modifying the weight, and the learning of the network is completed in the weight modifying process. When real data are used for prediction, the model prediction value is consistent with the change rule of original data, the average error is within 12%, the lightweight and generalization capability of the model structure meet the field construction requirements, an effective scheme is provided for predicting the excavation parameters of the shield tunneling machine, and the method has certain practical value.
Description
Technical Field
The invention relates to a shield tunneling machine parameter optimization method based on a deep neural network.
Background
The shield machine is a special large-scale device for tunnel driving, and has wide application in subway construction and track construction. Various parameters need to be set according to different tunneling conditions in the operation process of the shield tunneling machine, and the parameters also need to be continuously adjusted in the whole tunneling process, because the tunneling speed of the shield tunneling machine can be directly influenced by the quality of the parameter setting. Therefore, the method selects a proper method to accurately and effectively predict the tunneling parameters, and has important practical significance.
The operation condition analysis of the existing shield machine mainly depends on the experience judgment of decision-makers, and in the process of engineering, the decision-makers are difficult to quickly and accurately analyze a large amount of complex information and data generated by the sensors, and the method for adjusting parameters by the aid of engineering experience cannot guarantee the excavation efficiency under different geological conditions. Currently, suggestions are provided for decision-making personnel based on a computer expert system, but the system can only provide suggestions of single parameters, and a large amount of construction historical information under different topographic and geological features, which is collected by the current shield cloud platform, is not sufficiently analyzed and utilized.
Disclosure of Invention
On the basis of the model control method, the invention combines the advantages of the neural network to establish a method for intelligently predicting and optimizing the operation parameters of the shield machine, and can arrange and analyze the real-time operation information of the shield machine and quickly transmit the information to decision-making personnel, so as to be beneficial to quick decision-making in the construction process and achieve the purpose of optimally controlling the shield construction parameters.
In order to facilitate understanding of the technical solution of the present invention, the following description is made on the neural network technology adopted in the solution of the present invention:
deep neural networks are currently the basis for many human intelligence applications, and can use statistical learning methods to extract high-level features from raw sensory data, obtaining efficient characterization of the input space in large amounts of data, which enables neural networks to exceed human accuracy in many fields. The neural network establishes M hidden layers, establishes the connection between the input layer and the hidden layers in sequence, and finally establishes the connection between the hidden layers and the output layer. Each layer is composed of a large number of nodes (or neurons), and the connection between layers is realized by the connection of each neuron of the layer. The connections between nodes represent a weighted value, called weight, for the signal passing through the connection, and a specific output function, called excitation function, is selected for each node of each hidden layer. The network itself is an approximation to some algorithm or function that solves for the weight of each connection and the bias value of each node itself.
In order to achieve the purpose, the invention adopts the technical scheme that:
a shield tunneling machine parameter optimization method based on a deep neural network comprises the following steps:
step 1: collecting sample data, combining actual construction experience and historical data records, and collecting a section of excavation historical data of a shield machine model of a specific model in a certain city, wherein the data sample generally comprises the ring number, total power, cutter torque, cutter pressure, cutter rotating speed, propulsion pressure, total propulsion force, current accumulated amount of foam mixed liquid, left middle soil bin pressure, left upper soil bin pressure, left lower soil bin pressure, right middle soil bin pressure and current propulsion speed of the shield machine;
step 2: and carrying out targeted data prediction mathematical theory on the acquired data samples.
Most of the collected data are collected by the shield machine sensor, outlier/abnormal data appear in the data due to different sensitivities of the sensor in the collecting process, dimensions of the data are not uniform, and the uniform dimensions are used as input in the deep neural network. The data processing stage includes the dimensionality reduction, denoising and feature selection of data. Through the processing, high-quality data with abnormal points removed, nonlinear correlation removed and unified dimension is obtained.
And step 3: and establishing and initializing a model.
The model is divided into two parts, namely a forward prediction model formed by a deep neural network and a parameter generation model formed by a weight optimization matrix, and the weight of each layer of node is initialized randomly at the beginning, wherein the weight matrix of the weight optimization layer is initialized to be an identity matrix.
And 4, step 4: in the model training stage, data are divided into observation data and prediction data, the prediction data in the model is the tunneling speed, and the observation data are the rest of the data after data preprocessing. And then locking the weight optimization matrix and not participating in the training at the stage, obtaining a predicted value y ^ of the model, namely the predicted speed, by using a forward propagation algorithm of the model, calculating an error function with the actual value y, namely the real speed recorded in the original data, and obtaining the optimal parameter by using a gradient descent algorithm. On the first training, the weight optimization layer will be set to untrained and the back propagation will update the parameters in the neural network layer such that an optimum is achieved between input and output.
And 5: after the forward prediction model is trained to the right most, the weight optimization layer parameters are unlocked, the gradient can be normally transferred to the layer, and other hidden layer weights are locked.
And 6, applying the forward propagation algorithm again to obtain a predicted value and y (1+ delta), wherein delta is an error function calculated by the manual intervention factor, and updating parameters through a gradient descent algorithm.
On the second training, the weight optimization layer will be set to trainable and the neural network layer set to untrained. At this point, the back propagation will update the parameters in the weight optimization layer, and the parameters of the neural network layer are frozen and cannot be trained to modify.
And 7, taking the predicted value of the weight optimization layer as a final output result.
After two times of training, the output obtained by the input through the weight optimization layer is the final required output value. In the invention, the output content of the weight optimization layer is consistent with the received content and is a predicted value of the observation index after optimization and modification.
Compared with the prior art, the invention has the advantages that: the invention utilizes a neural network algorithm to establish a forward model of shield tunneling machine excavation parameters for predicting complex geological conditions, if an expected output value cannot be obtained in an output layer, the square sum of the output and the expected error is taken as a target function, backward propagation is carried out, the partial derivative of the target function to each neuron weight is calculated layer by layer to form the gradient of the target function to the weight vector, the gradient is used as the basis for modifying the weight, and the learning of the network is completed in the weight modifying process. When real data are used for prediction, the model prediction value is consistent with the change rule of original data, the average error is within 12%, the lightweight and generalization capability of the model structure meet the field construction requirements, an effective scheme is provided for predicting the excavation parameters of the shield tunneling machine, and the method has certain practical value.
Drawings
FIG. 1 hidden layer parameter training;
FIG. 2 is a process flow;
FIG. 3 test cases of different models;
FIG. 4 training scenarios for different models;
fig. 5 inverse model training scenario.
Detailed Description
The technical scheme of the invention is described in detail in the following with reference to the accompanying drawings.
The method mainly comprises the following steps:
(1) the method comprises the steps of collecting sample data, and collecting a section of mining historical data of a shield machine model of a specific model in a certain city by combining actual construction experience and historical data records, wherein the common data samples comprise the ring number, the total power, the cutter torque, the cutter pressure, the cutter rotating speed, the propelling pressure, the total propelling force, the current accumulated amount of foam mixed liquid, the left middle soil bin pressure, the left upper soil bin pressure, the left lower soil bin pressure, the right middle soil bin pressure and the current propelling speed of the shield machine.
(2) And carrying out targeted data prediction mathematical theory on the acquired data samples.
Most of collected data are collected by a shield machine sensor, outlier/abnormal data appear in the data due to different sensitivities of the sensor in the collection process, dimensions of the data are not uniform, and the uniform dimensions are used as input in a deep neural network. The data processing stage includes the dimensionality reduction, denoising and feature selection of data. Through the processing, high-quality data with abnormal points removed, nonlinear correlation removed and unified dimension is obtained.
(3) Determining an input and output vector of a model, outputting the input and output vector as a predicted tunneling speed in a forward prediction stage, and inputting data preprocessed except for the speed, namely, in the figure 1, X represents a training sample, and X is ═ X 1 ,X 2 ,…X i …X n ]Then, n training samples are represented, and each sample X i ＝[x 1 ,x 2 ,…,x k ]Where k represents the input data in how many dimensions there are, the output data is y', and in the backward prediction phase, unlike the forward prediction phase, the input is X i ＝[x 1 ,x 2 ,…,x k ]To transportOut is optimized post x' i ＝[x 1′ ,x 2′ ,…,x k′ ]The initial weight matrix is an identity matrix and is represented by I; the connection weight value from the ith unit of the input layer to the jth unit of the output layer is w ij The connection weight value from the jth unit of the hidden layer to the tth unit of the output layer is v jt The threshold values of the cells of the hidden layer and the output layer are respectively represented as theta j ，γ j . The input and output layer relationships are represented using activation functions.
(4) Before the first training, the weight optimization layer needs to be initialized to the identity matrix, and the input will not change after passing through the weight optimization layer. On the first training, the weight optimization layer will be set to untrained and the back propagation will update the parameters in the neural network layer so that the optimization between input and output is achieved.
(5) On the second training, the weight optimization layer will be set to trainable and the neural network layer set to untrained. At this point, the back propagation will update the parameters in the weight optimization layer, and the parameters of the neural network layer are frozen and cannot be trained to modify.
(6) After two times of training, the output obtained by the input through the weight optimization layer is the final required output value.
(7) In the inverse model, samples with a speed of 40-50 are used as training data, and samples with a speed of 48-60 are used as comparison data. And finally obtaining the distribution of the predicted values and the real values of the total propelling force, the cutter head torque, the propelling pressure, the total power, the cutter head rotating speed, the soil bin pressure and the foam mixed liquid ring accumulation amount by taking the ring average data as a training sample, wherein the distribution of the predicted values and the real values is shown in a fifth graph, the predicted values of the model are consistent with the change rule of the original data, and the average error is within 12 percent.
Claims (1)
1. A shield tunneling machine parameter optimization method based on a deep neural network is characterized by comprising the following steps:
s1, collecting sample data according to the historical construction record of the shield machine to be predicted, wherein the sample data comprises the ring number, the total power, the cutter torque, the cutter pressure, the cutter rotation speed, the propelling pressure, the total propelling force, the current accumulated amount of foam mixed liquid, the left middle soil bin pressure, the left upper soil bin pressure, the left lower soil bin pressure, the right middle soil bin pressure and the current propelling speed of the shield machine;
s2, carrying out data processing on the sample data, including dimension reduction, denoising and feature selection of the data;
s3, establishing a model, wherein the model comprises a forward prediction model formed by a deep neural network and a parameter generation model formed by a weight optimization matrix, and the model specifically comprises an input layer, a weight optimization layer, a hidden layer and an output layer in sequence; during initialization, randomly initializing the weight of each layer of nodes, wherein the weight matrix of the weight optimization layer is initialized to be an identity matrix; dividing the data into observation data and prediction data, defining the prediction data as the tunneling speed of the shield tunneling machine, and defining the observation data as the data obtained in the step S2;
s4, first training:
locking the weight optimization matrix without participating in the training, obtaining a predicted value y' of the model by utilizing a forward propagation algorithm of the model, namely a predicted speed, calculating an error function with an actual value y, namely a real speed recorded in original data, and obtaining an optimal parameter through a gradient descent algorithm; because the weight optimization layer is set to be untrainable, the parameters in the deep neural network are updated by back propagation, and the forward prediction model is trained to be optimal;
s5, second training:
unlocking the weight optimization matrix, and locking the deep neural network without participating in the training; applying the forward propagation algorithm again to obtain a predicted value and y (1+ delta), wherein delta is an error function calculated by the manual intervention factor, updating the parameters in the weight optimization layer through the gradient descent algorithm, and training the parameter generation model to be optimal;
and S6, optimizing the parameters of the shield machine by adopting the trained model, and taking the value output by the parameter generation model as a final output result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910382961.6A CN110096827B (en) | 2019-05-09 | 2019-05-09 | Shield tunneling machine parameter optimization method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910382961.6A CN110096827B (en) | 2019-05-09 | 2019-05-09 | Shield tunneling machine parameter optimization method based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110096827A CN110096827A (en) | 2019-08-06 |
| CN110096827B true CN110096827B (en) | 2022-08-05 |
Family
ID=67447524
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910382961.6A Active CN110096827B (en) | 2019-05-09 | 2019-05-09 | Shield tunneling machine parameter optimization method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110096827B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
| CN112065421B (en) * | 2020-10-10 | 2022-04-22 | 中国铁建重工集团股份有限公司 | Automatic positioning method for heading machine cutter head |
| CN113111467B (en) * | 2021-05-11 | 2023-05-12 | 上海同岩土木工程科技股份有限公司 | Tunneling parameter prediction method, medium and equipment based on hybrid neural network |
| CN113486463B (en) * | 2021-07-02 | 2023-05-02 | 中铁工程装备集团有限公司 | Shield optimal autonomous tunneling control method based on deep reinforcement learning |
| 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 |
| CN115982515B (en) * | 2023-01-05 | 2023-09-29 | 西南交通大学 | Method for obtaining optimal value of attitude control parameter of shield tunneling machine |
| CN116079730B (en) * | 2023-02-06 | 2024-06-25 | 北京大学 | Control method and system for operation precision of arm of elevator robot |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000345793A (en) * | 1999-06-07 | 2000-12-12 | Okumura Corp | Rock-bed judging system |
| WO2014060001A1 (en) * | 2012-09-13 | 2014-04-24 | FRENKEL, Christina | Multitransmitter model of the neural network with an internal feedback |
| CN108468549A (en) * | 2018-03-11 | 2018-08-31 | 辽宁石油化工大学 | The optimization method and device of earth pressure balanced shield, EPBS |
| CN108979648A (en) * | 2018-08-28 | 2018-12-11 | 中铁工程服务有限公司 | A kind of shield machine segment ring number calculation method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8103606B2 (en) * | 2006-12-08 | 2012-01-24 | Medhat Moussa | Architecture, system and method for artificial neural network implementation |
| US20150324690A1 (en) * | 2014-05-08 | 2015-11-12 | Microsoft Corporation | Deep Learning Training System |
-
2019
- 2019-05-09 CN CN201910382961.6A patent/CN110096827B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000345793A (en) * | 1999-06-07 | 2000-12-12 | Okumura Corp | Rock-bed judging system |
| WO2014060001A1 (en) * | 2012-09-13 | 2014-04-24 | FRENKEL, Christina | Multitransmitter model of the neural network with an internal feedback |
| CN108468549A (en) * | 2018-03-11 | 2018-08-31 | 辽宁石油化工大学 | The optimization method and device of earth pressure balanced shield, EPBS |
| CN108979648A (en) * | 2018-08-28 | 2018-12-11 | 中铁工程服务有限公司 | A kind of shield machine segment ring number calculation method |
Non-Patent Citations (4)
| Title |
|---|
| "Application. of neural networks to automatic soil pressure balance control for shield tunneling";I-Cheng Yeh;《Automtion in Construction》;19990218;第5卷(第5期);第421-426页 * |
| "反向传播算法（过程及公式推导）";ZYVV;《https://www.cnblogs.com/wlzy/p/7751297.html》;20171029;第1-15页 * |
| "基于LSTM循环神经网络的盾构机故障预测系统设计";周奇才等;《现代机械》;20181031(第5期);第35-40页 * |
| "支持向量机在围岩变形监测中的应用";张彬等;《世界科技研究与发展》;20110630;第33卷(第3期);第425-427页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110096827A (en) | 2019-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110096827B (en) | Shield tunneling machine parameter optimization method based on deep neural network | |
| Boussabaine | The use of artificial neural networks in construction management: a review | |
| CN110110419B (en) | TBM tunneling parameter prediction method based on multi-target learning | |
| CN110195592A (en) | Shield driving pose intelligent Forecasting and system based on interacting depth study | |
| CN109165664A (en) | A kind of attribute missing data collection completion and prediction technique based on generation confrontation network | |
| CN102622515B (en) | A kind of weather prediction method | |
| CN109635461A (en) | A kind of application carrys out the method and system of automatic identification Grades of Surrounding Rock with brill parameter | |
| CN110807557A (en) | Drilling rate prediction method based on BP neural network and drilling rate optimization method based on BP neural network and particle swarm optimization | |
| CN109829236A (en) | A kind of Compressor Fault Diagnosis method based on XGBoost feature extraction | |
| CN110852423A (en) | Tunnel boring machine excavation performance and control parameter prediction method based on transfer learning | |
| KR101967978B1 (en) | Apparatus for predicting net penetration rate of shield tunnel boring machine and method thereof | |
| Fu et al. | A hybrid deep learning approach for dynamic attitude and position prediction in tunnel construction considering spatio-temporal patterns | |
| CN110210169B (en) | LSTM-based shield tunneling machine fault prediction method | |
| CN108595803A (en) | Shale gas well liquid loading pressure prediction method based on recurrent neural network | |
| CN114810100B (en) | Shield tunneling attitude prediction method based on deep neural network | |
| CN112733440A (en) | Intelligent fault diagnosis method, system, storage medium and equipment for offshore oil-gas-water well | |
| CN112879024A (en) | Dynamic prediction method, system and equipment for shield attitude | |
| CN110084322A (en) | A kind of prediction technique of shield machine boring parameter neural network based | |
| Xu et al. | Predicting shield position deviation based on double-path hybrid deep neural networks | |
| CN115773127A (en) | Intelligent decision-making method, system, equipment and medium for slurry balance shield | |
| CN113111467B (en) | Tunneling parameter prediction method, medium and equipment based on hybrid neural network | |
| Xue et al. | Gas well performance prediction using deep learning jointly driven by decline curve analysis model and production data | |
| CN116361624B (en) | Error feedback-based large-range ground subsidence prediction method and system | |
| Liu et al. | Autonomous intelligent control of earth pressure balance shield machine based on deep reinforcement learning | |
| Yu et al. | Traffic prediction method based on RBF neural network with improved artificial bee colony algorithm |
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 |