CN111636891B - Real-time shield attitude prediction system and construction method of prediction model - Google Patents
Real-time shield attitude prediction system and construction method of prediction model Download PDFInfo
- Publication number
- CN111636891B CN111636891B CN202010514756.3A CN202010514756A CN111636891B CN 111636891 B CN111636891 B CN 111636891B CN 202010514756 A CN202010514756 A CN 202010514756A CN 111636891 B CN111636891 B CN 111636891B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- shield
- database
- subsystem
- 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
- 238000010276 construction Methods 0.000 title claims abstract description 87
- 238000000034 method Methods 0.000 claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 37
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000004458 analytical method Methods 0.000 claims abstract description 15
- 230000005540 biological transmission Effects 0.000 claims abstract description 8
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 7
- 230000003993 interaction Effects 0.000 claims abstract description 7
- 238000005096 rolling process Methods 0.000 claims description 28
- 238000013527 convolutional neural network Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 230000002452 interceptive effect Effects 0.000 claims description 4
- 230000005641 tunneling Effects 0.000 abstract description 31
- 230000036544 posture Effects 0.000 description 27
- 230000006870 function Effects 0.000 description 14
- 230000002159 abnormal effect Effects 0.000 description 9
- 238000009412 basement excavation Methods 0.000 description 9
- 238000013461 design Methods 0.000 description 7
- 238000007726 management method Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000006073 displacement reaction Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 210000004027 cell Anatomy 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 239000002699 waste material Substances 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 102100029469 WD repeat and HMG-box DNA-binding protein 1 Human genes 0.000 description 1
- 101710097421 WD repeat and HMG-box DNA-binding protein 1 Proteins 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000012047 cause and effect analysis Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH OR ROCK 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 OR ROCK 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
- E21D9/093—Control of the driving shield, e.g. of the hydraulic advancing cylinders
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH OR ROCK 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 OR ROCK 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
Landscapes
- Engineering & Computer Science (AREA)
- Mining & Mineral Resources (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geochemistry & Mineralogy (AREA)
- Geology (AREA)
- Environmental & Geological Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a shield attitude real-time prediction system and a construction method of a prediction model, wherein the system comprises: the system comprises an interaction subsystem, a database subsystem and a model library subsystem, wherein the interaction subsystem is used for mutual bidirectional data transmission and comprises the following components: the method is used for front-end display, operation and transmission of the results of model analysis and data query; the database subsystem: the method is used for storing and managing various data in the shield project construction process, and comprises the following steps: the system comprises a source database, a data acquisition module, a decision support database, a data directory module and a query module; a model library subsystem: carrying out modeling analysis and providing analysis results, comprising: the system comprises a model library, a model catalogue module and a training and decision platform. The invention can predict the tunneling attitude in real time through the real-time parameters in the shield tunneling process, thereby effectively improving the decision speed and the engineering quality of a construction site.
Description
Technical Field
The invention relates to a system and a modeling method for shield construction projects, in particular to a shield attitude real-time prediction system and a prediction model construction method.
Background
In a shield construction project site, the difficult point and key point of the decision service is the management and control of shield tunneling, wherein the tunneling attitude of shield equipment is a key link influencing the construction quality, safety and progress. At present, scholars at home and abroad study the following aspects of the shield attitude control decision problem:
(1) paying attention to daily monitoring of shield postures of a construction site and posture adjustment technology after deviation, establishing a posture adjustment model and a posture control model, including control of a tunneling track, posture adjustment of an initial section and the like.
(2) The generation mechanism of the shield attitude risk is concerned, and the action and the reaction of the tunneling attitude and other influence factors are deduced by analyzing equipment components, a construction process and an external environment, so that the attitude prejudgment and decision making on a construction site are assisted.
(3) The measurement calculation of the tunneling attitude of the shield tunneling machine is used as an entry point, and an object and a method of the measurement calculation are researched to achieve the purposes of controlling and adjusting the attitude, including calculation of an attitude angle and a coordinate system, comparison and optimization of a manual measurement system and an automatic measurement system and the like.
(4) The method focuses on discussing the influence factors of the tunneling attitude of the shield tunneling machine and other shield subsystems to help a primary administrator of a construction site to carry out attitude prejudgment and control decision, including control of a jack system, a duct piece and the like.
However, the control decision research on the shield attitude is more important than the following construction aspects, the artificial experience is combined, the post-cause and effect analysis and the influence factor research are carried out on the abnormal attitude, the control model is established, and a complete set of analysis method suitable for various shield models and various geological environments cannot be provided in the aspect of the advance prediction of the tunneling attitude. The shield construction site still needs to depend on expert knowledge to carry out prejudgment and decision making, so that economic loss and social influence are easily caused by abnormal shield posture, equipment damage, progress lag and accident occurrence due to artificial low-efficiency decision making, and the industrial requirements of efficient prevention and control in advance and posture risk reduction cannot be completely met.
Meanwhile, with the current research, there are also problems in the following respects:
(1) the waste of structured data. With the gradual convergence of information science and the shield field, the sensing technology is mature and applied to data acquisition work of a construction site. However, at present, a large amount of accumulated shield project data is not sufficiently applied to the management decision work of a construction site supporting large-scale complex engineering, and although some students use the algorithm of a support vector machine in machine learning to perform posture rectification research, the method is more suitable for learning of small samples, and a scientific data processing method suitable for large samples is urgently needed for a high-dimensional and huge construction data set which is collected in seconds and is used for shield projects.
(2) The utilization of unstructured data is low. The traditional large complex engineering projects such as the shield also contain a large amount of unstructured information which are important reference bases for management decisions in construction sites, including geological survey reports, design drawings, conference documents, construction logs and the like. The unstructured data format is not normalized and scattered, so that the speed and quality of multi-thread service decision of a manager can be greatly reduced.
(3) In the aspect of selection of a prediction method, the traditional shallow network model is mostly used for researching the tunneling attitude of the shield equipment at present, but the models are easy to show poor training effect and fall into the condition of local optimum when facing a complex objective function.
(4) In the aspect of constructing a machine learning model, at present, a shallow neural network structure is mainly used for researching the tunneling attitude of shield equipment, and when an input feature set is constructed, the importance of selecting individual important parameters by means of domain knowledge is emphasized, and the work of feature identification, sequence data modeling, prediction and the like is carried out, so that the final result is greatly influenced by artificial subjective experience and knowledge, and the final accuracy of the model is low because only the individual important parameters are selected when the model is constructed.
Disclosure of Invention
The invention provides a shield attitude real-time prediction system and a construction method of a prediction model, aiming at the problem of prediction of a tunneling attitude in the management and control of the current shield tunneling process, the state of the tunneling attitude of a shield tunneling machine at the future moment is predicted through non-attitude data so as to improve the decision speed and the engineering quality of a construction site and solve the problems of low-efficiency manual decision and data knowledge waste existing at present.
The invention relates to a shield attitude real-time prediction system, which comprises an interaction subsystem, a database subsystem and a model library subsystem, wherein the interaction subsystem, the database subsystem and the model library subsystem are used for mutual bidirectional data transmission,
the interaction subsystem: the method is used for front-end display, operation and transmission of the results of model analysis and data query;
the database subsystem: the system is used for storing and managing various data in the shield project construction process; comprises the following steps: the system comprises a source database for storing data related to a decision target, which is analyzed and taken from the environment of a decision support system, a data acquisition module for acquiring the data of the source database, a data directory module for data definition, data type description and data source description, an inquiry module for data retrieval and reading, and interpreting and responding to data requests from other subsystems through the data directory module, and a decision support database for receiving the output of the data acquisition module, wherein the decision support database stores data generated inside the system and externally acquired project data, and the data directory module and the inquiry module are respectively in bidirectional connection with the decision support database;
a model library subsystem: modeling analysis and providing analysis results aiming at different service ranges of a construction site; comprises the following steps: the model library is used for storing models for realizing various decisions, the model catalog module is used for managing and calling various models, and the training and decision platform is used for completing modeling analysis and providing decision results.
The prediction system can predict the attitude state of the shield machine in a future period of time according to the parameters collected in the running process of the shield machine, and comprises the prediction of various attitude parameters such as a rolling angle, a pitch angle, a horizontal trend, a vertical trend and the like in the shield attitude.
Further, the source database of the database subsystem comprises an internal database and an external database, wherein the internal database is used for storing data strongly related to the shield equipment and comprises shield equipment information and data generated by equipment in construction operation; the external database is used for storing underground environment information, ground environment information and data of personnel work affair records.
Furthermore, the model library of the model library subsystem comprises a rolling angle prediction model library, a pitch angle prediction model library, a horizontal trend prediction model library, a vertical trend prediction model library, a shield head posture prediction model library and a shield tail posture prediction model library.
The invention also provides a construction method of the shield attitude real-time prediction model for the prediction system, wherein the shield attitude real-time prediction model is based on a one-dimensional convolutional neural network, and the construction steps comprise:
carrying out layer stacking: after a multivariate time sequence sample data set enters a convolution layer and a pooling layer of a one-dimensional convolution neural network, capturing the characteristics of the time sequence sample data through convolution operation from top to bottom, and performing maximum pooling operation for dimension reduction on the premise of ensuring that the characteristics are not changed;
defining the distance between the predicted value and the true value under the current model through a loss function, namely a loss value, and measuring the matching degree of the predicted value and the expected result;
updating the internal weight in the hidden layer of the model by using the loss value through an optimizer to reduce the loss value after the model training to the minimum;
setting iteration times of training after the training set is input into the model, so that the output of the model after the iterative training of corresponding times is stable;
setting batch hyper-parameters in gradient descent for controlling the number of samples before updating the internal parameters of the model;
and calling corresponding historical data resources from the database subsystem to perform data preprocessing, and training the model through the preprocessed data until the model is trained.
According to the prediction model, aiming at different attitude parameters, after data are trained according to the method, the prediction model of specific parameters can be trained. And various attitude parameters such as a rolling angle, a pitch angle, a horizontal trend, a vertical trend and the like in the shield attitude can be predicted through the prediction model.
Furthermore, in the process of training the model, in the fine tuning stage of the model, the iteration times are searched within a preset range in a grid searching mode, and the iteration times when the model training effect can reach a stable state and overfitting does not occur are taken as the optimal values of the iteration times.
Specifically, the preprocessing of data by using historical data resources in the database subsystem includes:
selecting corresponding data from historical data resources according to the shield project construction requirement to construct a data space of a shield project construction operation state, and determining data contents required to be included for solving the shield attitude problem through the data space;
extracting parameters corresponding to data contents required for solving the shield attitude problem from a decision support database of a database subsystem;
taking the parameter name in the parameters as a column name and the parameter value as a cell value, and constructing a two-dimensional data table integrating construction data according to the construction parameter recording time sequence;
adding an artificial label column in the two-dimensional data table according to the recording time information and/or the ring number information of shield construction in the two-dimensional data table, so that the two-dimensional data table comprises construction data and artificial labels;
and determining the interval of the prediction time and the corresponding parameter variable according to the data of the historical construction state, and finally forming a final data set format through data standardization.
The shield attitude real-time prediction system and the construction method of the prediction model can predict the tunneling attitude in real time through real-time parameters in the shield tunneling process, and effectively improve the decision speed and the engineering quality of a construction site.
The present invention will be described in further detail with reference to the following examples. This should not be understood as limiting the scope of the above-described subject matter of the present invention to the following examples. Various substitutions and alterations according to the general knowledge and conventional practice in the art are intended to be included within the scope of the present invention without departing from the technical spirit of the present invention as described above.
Drawings
FIG. 1 is a block diagram of a shield attitude real-time prediction system of the present invention.
FIG. 2 is a flow chart of a method for constructing a shield attitude real-time prediction model according to the present invention.
Fig. 3 is a schematic structural diagram of a shield rolling angle prediction model based on a one-dimensional convolutional neural network in the embodiment.
Fig. 4 is a diagram illustrating an example of displacement and deletion of roll angle data.
Fig. 5 is a schematic diagram of the roll angle pose tag division.
FIG. 6 is a diagram illustrating a resulting data set structure according to an embodiment.
Detailed Description
As shown in fig. 1, the shield attitude real-time prediction system of the present invention has an interactive subsystem, a database subsystem and a model library subsystem, wherein the interactive subsystem, the database subsystem and the model library subsystem are used for mutual bidirectional data transmission,
the interaction subsystem: the method is used for front-end display, operation and transmission of the results of model analysis and data query.
The database subsystem: the system is used for storing and managing massive and multi-source data sets generated and accumulated in the shield project construction process; comprises the following steps:
(1) a source database: data related to the decision-making objective parsed from the environment in which the decision support system is located is stored. The system comprises an internal database and an external database, wherein the internal database is used for storing data strongly related to shield equipment, including shield equipment information and data generated by equipment in construction operation, such as shield machine design data, construction parameter records, fault records and the like; the external database is used for storing underground environment information, ground environment information and data of personnel work affair records, such as tunnel design data, geological condition data, construction logs and the like.
(2) A data acquisition module: and analyzing the data in the source database through four operations of data display, description, subset and aggregation.
(3) Decision support database: and receiving the output of the data acquisition module. Data generated internally by the system and project data collected externally are stored in a decision support database. The data generated inside the system comprises system running script data, model historical decision data and the like. When the decision support database is constructed, the external structured and unstructured data are collected, formatted and cleaned. And based on the environment of various relational databases, data storage is carried out so as to enhance the capability of large-data-volume system management expansion and ensure the stability of the system.
(4) A data directory module: for data definition, data type description and data source description.
(5) The query module: for data retrieval and reading, and for interpreting and responding to data requests from other subsystems through the data directory module.
The data directory module and the query module are respectively in bidirectional connection with the decision support database.
A model library subsystem: modeling analysis and providing analysis results aiming at different service ranges of a construction site; comprises the following steps: the model library is used for storing models for realizing various decisions, the model catalog module is used for managing and calling various models, and the training and decision platform is used for completing modeling analysis and providing decision results. In the construction site of the shield project, a decision maker needs to make decision behaviors aiming at different service ranges, and in the process, one or more preset corresponding models need to be called in a model library for decision analysis. The method specifically comprises the following steps:
I. rolling angle prediction model library: aiming at the problem that a shield machine body rotates clockwise/anticlockwise around a central axis in excavation and tunneling, the identification and prediction of abnormal rolling angle postures at the future moment are realized, and a decision maker is assisted to avoid the risk of the rolling angle postures in advance.
II, a pitch angle prediction model library: aiming at the problems that a shield tunneling machine tunneling line deviates from a preset tunnel line and shield head raising/lowering occurs in the excavation tunneling process, the identification and prediction of an abnormal pitch angle attitude at a future moment are realized, and a decision maker is assisted to avoid the risk of the pitch angle attitude in advance.
Library of horizontal trend prediction models: aiming at the problem that the shield machine deviates from the central line of the designed tunnel in the horizontal direction during excavation and tunneling, the identification and prediction of the abnormal horizontal trend at the future moment are realized, and a decision maker is assisted to avoid the horizontal trend risk in advance.
Vertical trending prediction model library: aiming at the problem that the shield machine deviates from the central line of the designed tunnel in the vertical direction during excavation and tunneling, the identification and prediction of abnormal vertical trend at the future moment are realized, and a decision maker is assisted to avoid the risk of the vertical trend in advance.
V, shield head attitude prediction model base: aiming at the relative deviation condition of the shield head and the central axis in excavation and excavation, the recognition and prediction of the abnormal shield head posture at the future moment are realized, and a decision maker is assisted to avoid the risk of the shield head posture in advance.
VI, a shield tail attitude prediction model library: aiming at the relative deviation condition of the shield tail and the central axis in excavation and excavation, the identification and prediction of the abnormal shield tail posture at the future moment are realized, and a decision maker is assisted to avoid the risk of the shield tail posture in advance.
The prediction system can predict the attitude state of the shield machine in a future period of time according to the parameters collected in the running process of the shield machine, and comprises the prediction of various attitude parameters such as a rolling angle, a pitch angle, a horizontal trend, a vertical trend and the like in the shield attitude.
As shown in fig. 2, the method for constructing the real-time shield posture prediction model for the prediction system of the present invention is based on a one-dimensional convolutional neural network, and includes:
carrying out layer stacking: after a multivariate time sequence sample data set enters a convolution layer and a pooling layer of a one-dimensional convolution neural network, capturing the characteristics of the time sequence sample data through convolution operation from top to bottom, and performing maximum pooling operation for dimension reduction on the premise of ensuring that the characteristics are not changed;
defining the distance between the predicted value and the true value under the current model through a loss function, namely a loss value, and measuring the matching degree of the predicted value and the expected result;
updating the internal weight in the hidden layer of the model by using the loss value through an optimizer to reduce the loss value after the model training to the minimum;
and setting the iteration times of training after the training set is input into the model, so that the output of the model after the iterative training of corresponding times is stable. In the process of training the model, in the fine tuning stage of the model, the iteration times are searched within a preset range in a grid searching mode, and the iteration times when the model training effect can reach a stable state and overfitting does not occur are taken as the optimal values of the iteration times;
setting batch hyper-parameters in gradient descent for controlling the number of samples before updating the internal parameters of the model;
and calling corresponding historical data resources from the database subsystem to perform data preprocessing, and training the model through the preprocessed data until the model is trained. Wherein the data preprocessing comprises the following steps:
selecting corresponding data from historical data resources according to the shield project construction requirement to construct a data space of a shield project construction operation state, and determining data contents required to be included for solving the shield attitude problem through the data space;
extracting parameters corresponding to data contents required for solving the shield attitude problem from a decision support database of a database subsystem;
taking the parameter name in the parameters as a column name and the parameter value as a cell value, and constructing a two-dimensional data table integrating construction data according to the construction parameter recording time sequence;
adding an artificial label column in the two-dimensional data table according to the recording time information and/or the ring number information of shield construction in the two-dimensional data table, so that the two-dimensional data table comprises construction data and artificial labels;
and determining the interval of the prediction time and the corresponding parameter variable according to the data of the historical construction state, and finally forming a final data set format through data standardization.
The invention can realize that:
(1) the method helps a site manager to respond in time and avoid hidden danger risks in the construction process in advance, reduces the decision difficulty of the site manager, further relieves the multithreading working pressure of the site manager, and reduces quality, safety and progress risks caused by low efficiency and errors of decision.
(2) The accumulated shield project data resources are applied to decision-making problems of a construction site, a scheme is provided for solving practical problems of data waste, low knowledge utilization efficiency and the like in similar projects, more intelligent industry data analysis is realized, and the response speed of management decision-making behaviors can be further improved.
(3) The method reduces the over-strong dependence of the shield project construction site on expert knowledge, improves the problems of low site decision efficiency, uncontrollable decision quality and the like, and is beneficial to promoting informatization and intellectualization of management decision.
The following further describes the implementation of the shield rolling angle attitude prediction model as an example:
firstly, constructing a model framework:
when the problem of predicting the excavation attitude of the shield tunneling machine is solved, a model design of the rolling angle attitude state is carried out by adopting a one-dimensional Convolutional Neural network (1D Convolutional Neural Networks,1D CNN), and the structure is shown in FIG. 3.
(1) Layer stack structure
The logical framework structure of the designed 1D CNN model is represented as follows:
[Conv1D,MaxPooling1D]*3→Dropout→Flatten→Dense
[ Conv1D, Max Point 1D ]. sup.3 shows that the depth of the neural network is changed and deepened by repeated stacking, including one-dimensional convolution layer and one-dimensional maximum pooling layer, which is one of the concentrated expressions of the difference of each network model. After the multivariate time sequence sample data set enters Conv1D, capturing features through convolution operation from top to bottom, and performing maximum pooling operation of dimension reduction on the premise of ensuring that local features are not changed. Wherein: step size (Stride) of Conv1D and MaxPooling1D is set to 1, Padding value (Padding) is set to 0; the window size of MaxPooling is set to 2; the window (i.e., size) of the convolution kernel in Conv1D is set to 1, and the number is 256; the stacking times are set to be 3, namely the one-dimensional convolution layer and the one-dimensional maximum pooling layer are stacked for 3 times; the activation function selects the ReLU function to turn all non-positive values to 0, speeding up model convergence.
Dropout: the method is a common regularization function, and can randomly disable the output node of the alpha probability in the training process, namely, the output node is not subjected to feature learning, so that the overfitting phenomenon that the model is excellent in performance on a training set but large in prediction error in a test set can be effectively avoided. In the model of the present embodiment, α is set to 0.5.
Flatten: and the multi-dimensional characteristic information is used for converting the received multi-dimensional characteristic information into a one-dimensional vector format and sending the one-dimensional vector format to the next layer.
Dense: in the Dense layer, all neurons are connected with all neurons in the previous layer for operation, so that the parameter quantity and the calculation quantity in the layer are huge. The model of this example includes two layers of sense: the parameter size in Dense1 is 512, the activation function is a ReLU function, and the function of mapping the learned distributed feature representation to the sample mark space is realized; dense2 is called a classifier layer, so that 1D CNN finally outputs a multi-dimensional vector, each dimension of the vector represents different rolling angle posture states and plays a role in classification, the classification quantity is set according to the prediction requirement of the rolling angle posture, for example, the rolling angle is 9 classification, the parameter size in Dense2 is 9, the activation function is a Softmax function, the probability of each dimension in the output vector is mapped to the range between (0 and 1), and the sum of the probabilities is 1.
(2) Loss function, optimizer, iteration count, batch
The loss function defines the distance between the predicted value and the true value under the current network model and is used for measuring the matching degree of the predicted value and the expected result of the network, and in the model training process, the minimization of the loss value is one of the model training targets.
The role of the optimizer is to update the network with the loss values. And selecting a multi-classification cross entropy (Category Cross entropy) loss function and an Adam optimizer according to the shield project data characteristics and a multi-classification output target.
The number of iterations refers to the number of times the training set is input to the network for training. In the model fine-tuning stage, the iteration times are searched within a certain range in a grid searching mode, and a better value-taking state is obtained when the model training effect reaches a stable state and overfitting does not occur. The number of iterations of the model in this example is 200.
Batch is one of the hyper-parameters in gradient descent, mainly used to control the number of samples before the intra-model parameter update, and generally includes Batch gradient descent (Batch GD), random gradient descent (SGD), and Mini-Batch gradient descent (Mini-Batch GD). While the batch size affects the convergence effect and speed of the model. By combining the experimental environment and the data characteristics, the model of the embodiment selects the Mini-Batch GD mode to perform Batch search, and the Batch size parameter is 64.
In summary, the 1D CNN model structure and parameter combinations suitable for roll angle attitude prediction in the shield project construction site are as follows (in roll angle):
where X in Dense2 is the number of classifications, set to "9" in this example, as shown in FIG. 5.
Second, model training data preprocessing
When the attitude prediction model is trained, historical data resources are required to be called from a database subsystem, and data preprocessing operation is required to be carried out. The method comprises the following steps:
(1) data space S for constructing shield project construction operation state
And constructing a data space S of the shield project construction operation state according to three dimensions of shield equipment conditions, ground line environment and underground natural environment, wherein the shield equipment can show different rolling angle states under different S. Let S ═ be (S1, S2, S3, S4, S5), resulting from 5 parts acting together:
s1 represents the construction parameter data collected by the construction site sensor and generated by the operation of the shield equipment;
s2 shows the basic situation of design and configuration of shield equipment for the construction site tunneling operation;
s3 represents the alarm data of abnormal state of the construction shield equipment;
s4 shows the natural environment of the rock and soil when the construction shield equipment is driven and passed through underground;
s5 represents project-related information during the reconnaissance and design phases.
(2) Integrated construction data table
By constructing the data space S, the data content required to be included in the rolling angle posture problem of the shield tunneling machine can be further definitely solved. In the decision support database, for S1-S5, corresponding target parameters are extracted, as exemplified in table 1.
Table 1:
after determining the parameters to be extracted from the decision support database, taking the parameter names of the construction parameters as column names and the parameter values as cell values, and constructing a two-dimensional data table according to the construction parameter recording time sequence.
Then, adding a label column in the data table according to the column of the current ring number and the recording time respectively, wherein the label column comprises: a geology type column (geography), a warning status column (risk), a design grade column (gradient), and a along-the-line risk source column (risk source). The labels and processing are shown in table 2.
Table 2:
(3) determination of model prediction time interval and displacement of roll angle data
The recording time of the construction parameters is collected in seconds, but the conditions of sensor failure and data interruption and loss caused by network blockage need to be considered. In order to implement predictive modeling of such sequence data and achieve the purpose of reflecting the rolling angle posture at the future time from the historical construction state, the following operations are required, as shown in fig. 4:
a) and (5) mapping relation. The input data and output data are represented by X and Y, respectively, X (t) → Y (t + s), where:
x represents a sample data set (comprising construction parameters and an artificial label column) after integrated processing, Y represents a column of roll angle posture data, t represents the recording time of input data, and s represents the time interval after t. The formula as a whole represents that at each time t, X corresponds to Y at time t + s.
b) And (6) displacement processing. In order to express the mapping relationship in the above formula in the two-dimensional data table, the roll angle pose data Y at each time t + s needs to be translated backwards (expressed as upward translation on the data table level) to the corresponding time t position, which is also a commonly used data processing method for the sequence prediction problem.
c) And (5) deleting operation. After the displacement processing, for sample data without time t + s, deleting corresponding time t data, so that each row of data in the processed data set comprises time t X and time t + s Y.
(4) Scroll angle data processing into range tags
The roll angle is a real number with a sign, and has a unit of mm/m (millimeter/meter) or degrees (since the former unit is more in practical use, the former unit is adopted in the embodiment). Wherein: smaller values indicate smaller amplitudes of rotation; the sign is "+" which indicates that the shield machine body rotates rightwards, namely rotates forwards when facing the tunneling direction; the symbol of negative indicates that the shield machine body rotates leftwards, namely rotates negatively when facing the tunneling direction. In the construction process, under the ordinary condition, field managers and operation technicians can adjust values of other construction parameters and the positive and negative rotation states of the cutter head, so that the current rolling angle is as close as possible to 0mm/m, and the condition that samples are unevenly distributed is caused. The large difference in the number of samples easily causes the gravity center of the classifier to be biased to the learning of large samples in the training process, so that small samples are difficult to identify, and therefore, the rolling angle data needs to be subjected to label processing to balance the data set.
In general, the positive/negative rotation amplitude of a rolling angle is required to be ensured not to exceed 5mm/m in shield project construction tunneling, and the state of a risk accident is realized when the amplitude exceeds 20 mm/m. Let Y be D × L, wherein: d represents the rotation direction of the rolling angle, the D belongs to {0,1, -1}, 0 refers to the posture state of the rolling angle without adjustment, 1 refers to forward rotation, and-1 refers to reverse rotation; l represents the degree of rotation of the roll angle, L ∈ {1,2,3,4}, with larger numbers representing larger rotational forces. By different D and L combinations, 9 roll angle label results were generated, as shown in fig. 5: the label "0" indicates that the shield equipment is in a normal rolling angle posture state; the label "1", the label "2" and the label "3" indicate that the shield equipment is in a positive rotation state to be adjusted, the label "-1", the label "-2" and the label "-3" indicate that the shield equipment is in a negative rotation state to be adjusted, the steering and the speed of the cutter head need to be adjusted in time, and the numerical value of the label is prevented from increasing; the label "4" and the label "-4" indicate that the shield equipment is in an early warning rolling angle posture state, a construction site manager and an operation technician are required to quickly adjust to avoid the risk of the rolling angle posture, a shutdown and other measures are required under the situation of serious situation, and an emergency expert conference discussion problem is called for.
By establishing a rolling angle label rule, the Y-column rolling angle data is replaced by corresponding rolling angle posture labels, and finally, the data with uneven distribution and large difference is balanced to ensure that small samples can be effectively identified and learned.
(5) Data normalization
The input data X includes construction parameter data and manual label data, wherein the construction parameter range has thousand times difference, for example, the cutter torque range is [0,8000] KN.m, and the cutter rotation speed range is [0,4] r/min. Too large a dimension difference will result in too slow a model training speed. In order to eliminate the influence of dimension among data, before the input data X is used as the input of the deep learning model, z-score standardization processing needs to be performed on the construction parameter data in the input data X column by column through formula (1) to balance comparability among data indexes, and the format of the finally formed data set is shown in fig. 6.
Wherein the content of the first and second substances,
the capacity (number of lines) of the input data X is denoted by m;
the number of the construction parameters in the input data X is represented by n;
Aiindicates the ith construction parameter data column, a, in the input data XijRepresents the jth row of data, A, in the ith construction parameter data columni＝ai1,ai2,ai3,…,aij，i＝1,2,3…n，j＝1,2,3…m；
BiIs represented by AiStandardized construction parameter data set, bijData representing the jth row in the ith construction parameter data column after normalization;
μiis represented by AiThe average value of the columns is,
σiis represented by AiThe standard deviation of the columns is determined,
Claims (3)
1. the shield attitude real-time prediction system is characterized by comprising an interactive subsystem, a database subsystem and a model library subsystem, wherein the interactive subsystem, the database subsystem and the model library subsystem are used for mutual bidirectional data transmission,
the interaction subsystem: the method is used for front-end display, operation and transmission of the results of model analysis and data query;
the database subsystem: the shield project construction system comprises a source database for storing data related to a decision target extracted from the environment where a decision support system is located, a data acquisition module for acquiring the data of the source database, a data directory module for data definition, data type description and data source description, a query module for data retrieval and reading, interpreting and responding data requests from other subsystems through the data directory module, and a decision support database for receiving the output of the data acquisition module, wherein data generated inside the system and project data acquired outside the system are stored in the decision support database, and the data directory module and the query module are respectively in bidirectional connection with the decision support database;
a model library subsystem: the system comprises a model library, a model catalog module and a training and decision platform, wherein the model library is used for storing models for realizing various decisions, the model catalog module is used for managing and calling various models, and the training and decision platform is used for completing modeling analysis and providing decision results;
the source database of the database subsystem comprises an internal database and an external database, wherein the internal database is used for storing data strongly related to the shield equipment, and the external database is used for storing underground environment information, ground environment information and data of personnel work affair records;
the model libraries of the model library subsystem comprise a rolling angle prediction model library, a pitch angle prediction model library, a horizontal trend prediction model library, a vertical trend prediction model library, a shield head attitude prediction model library and a shield tail attitude prediction model library.
2. The method for constructing the shield attitude real-time prediction system according to claim 1, wherein the shield attitude real-time prediction model is based on a one-dimensional convolutional neural network, and the construction step comprises the following steps:
carrying out layer stacking: after a multivariate time sequence sample data set enters a convolution layer and a pooling layer of a one-dimensional convolution neural network, capturing the characteristics of the time sequence sample data through convolution operation from top to bottom, and performing maximum pooling operation for dimension reduction on the premise of ensuring that the characteristics are not changed;
defining the distance between the predicted value and the true value under the current model through a loss function, namely a loss value, and measuring the matching degree of the predicted value and the expected result;
updating the internal weight in the hidden layer of the model by using the loss value through an optimizer to reduce the loss value after the model training to the minimum;
setting iteration times of training after the training set is input into the model, so that the output of the model after the iterative training of corresponding times is stable;
setting batch hyper-parameters in gradient descent for controlling the number of samples before updating the internal parameters of the model;
calling corresponding historical data resources from a database subsystem to perform data preprocessing, and training the model through the preprocessed data until the model is trained;
the data preprocessing by historical data resources in the database subsystem comprises:
selecting corresponding data from historical data resources according to the shield project construction requirement to construct a data space of a shield project construction operation state, and determining data contents required to be included for solving the shield attitude problem through the data space;
extracting parameters corresponding to data contents required for solving the shield attitude problem from a decision support database of a database subsystem;
taking the parameter name in the parameters as a column name, taking the parameter value as a cell value, and constructing a two-dimensional data table integrating construction data according to the construction parameter recording time sequence;
adding an artificial label column in the two-dimensional data table according to the recording time information and/or the ring number information of shield construction in the two-dimensional data table, so that the two-dimensional data table comprises construction data and artificial labels;
and determining the interval of the prediction time and the corresponding parameter variable according to the data of the historical construction state, and finally forming a final data set format through data standardization.
3. The method for constructing the shield attitude real-time prediction system according to claim 2, wherein in the process of training the model, in the fine tuning stage of the model, the iteration times are searched within a preset range in a grid search mode, and the iteration times when the model training effect can reach a stable state and overfitting does not occur are taken as the optimal values of the iteration times.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010514756.3A CN111636891B (en) | 2020-06-08 | 2020-06-08 | Real-time shield attitude prediction system and construction method of prediction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010514756.3A CN111636891B (en) | 2020-06-08 | 2020-06-08 | Real-time shield attitude prediction system and construction method of prediction model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111636891A CN111636891A (en) | 2020-09-08 |
| CN111636891B true CN111636891B (en) | 2022-04-08 |
Family
ID=72329817
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010514756.3A Active CN111636891B (en) | 2020-06-08 | 2020-06-08 | Real-time shield attitude prediction system and construction method of prediction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111636891B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN112796768B (en) * | 2021-03-08 | 2022-07-19 | 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 |
| CN115358494B (en) * | 2022-10-20 | 2023-01-06 | 中电建铁路建设投资集团有限公司 | Danger early warning method for subway shield underpass construction |
| CN115450642B (en) * | 2022-11-08 | 2023-02-17 | 西南交通大学 | Shield attitude control method and system based on big data |
| CN115879642A (en) * | 2023-01-06 | 2023-03-31 | 浩鲸云计算科技股份有限公司 | Construction split prediction-based real-time sensing method for broadband installation progress |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105909270B (en) * | 2016-04-19 | 2018-07-06 | 上海建科工程咨询有限公司 | A kind of shield machine axis control system |
| CN106481344B (en) * | 2016-10-25 | 2018-07-06 | 华中科技大学 | A kind of balancing earth-pressure shielding machine tunnel piercing parameter intelligent control method |
| CN107503757B (en) * | 2017-09-20 | 2019-04-05 | 中建交通建设集团有限公司 | A kind of shield tunnel intelligence construction auxiliary system and application method based on big data technology |
| CN108868805B (en) * | 2018-06-08 | 2019-11-15 | 西安电子科技大学 | Shield method for correcting error based on statistical analysis in conjunction with XGboost |
| CN108846521A (en) * | 2018-06-22 | 2018-11-20 | 西安电子科技大学 | Shield-tunneling construction unfavorable geology type prediction method based on Xgboost |
| CN108868807B (en) * | 2018-09-07 | 2019-11-08 | 上海隧道工程有限公司 | The intelligent control method of shield driving correction |
| CN109779649A (en) * | 2019-02-13 | 2019-05-21 | 上海大学 | Shield driving axis real-time deviation correcting system and method based on big data |
| CN110195592B (en) * | 2019-04-30 | 2021-02-05 | 华中科技大学 | Intelligent shield tunneling pose prediction method and system based on hybrid deep learning |
| CN110533065A (en) * | 2019-07-18 | 2019-12-03 | 西安电子科技大学 | Based on the shield attitude prediction technique from coding characteristic and deep learning regression model |
-
2020
- 2020-06-08 CN CN202010514756.3A patent/CN111636891B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111636891A (en) | 2020-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111636891B (en) | Real-time shield attitude prediction system and construction method of prediction model | |
| CN109523316A (en) | The automation modeling method of commerce services model | |
| CN108255854A (en) | Fuzzy neural network data digging method based on particle group optimizing | |
| CN111815054A (en) | Industrial steam heat supply network short-term load prediction method based on big data | |
| CN115470962A (en) | LightGBM-based enterprise confidence loss risk prediction model construction method | |
| CN110196886A (en) | The multi-source heterogeneous big data correlating method of agricultural non-point source pollution and the big data supervising platform for using this method | |
| CN116861331A (en) | Expert model decision-fused data identification method and system | |
| Hongliang et al. | Management and instant query of distributed oil and gas production dynamic data | |
| CN113268370B (en) | Root cause alarm analysis method, system, equipment and storage medium | |
| CN114219096A (en) | Training method and device of machine learning algorithm model and storage medium | |
| Yamin et al. | Research on matching method for case retrieval process in CBR based on FCM | |
| CN116070973A (en) | Knowledge graph-based digital factory management index intelligent recommendation method | |
| Wang et al. | A Novel Multi‐Input AlexNet Prediction Model for Oil and Gas Production | |
| CN111738563B (en) | Method for detecting process control capacity of aerospace production site | |
| Cao | Design and optimization of a decision support system for sports training based on data mining technology | |
| Yang et al. | Prediction of criminal tendency of high-risk personnel based on combination of principal component analysis and support vector machine | |
| CN113177040A (en) | Full-process big data cleaning and analyzing method for aluminum/copper plate strip production | |
| CN113269380A (en) | Return-to-school scheme estimation method for epidemic situation prevention and control | |
| CN113064924A (en) | Nuclear power big data experience retrieval and pushing method | |
| Duan et al. | Data mining technology for structural health monitoring | |
| Wang et al. | A Prediction and evaluation model analysis of enterprise economic management mode based on neural network strategy | |
| Sun et al. | An artificial intelligence-based real-time monitoring framework for time series | |
| CN117077005B (en) | Optimization method and system for urban micro-update potential | |
| Liu et al. | Wind power prediction based on meteorological data visualization | |
| CN117828280B (en) | Intelligent fire information acquisition and management method based on Internet of things |
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 |