Method and equipment for training intelligent prediction model of tunnel surrounding rock strength of full face tunnel boring machine

Abstract

The invention provides a full face tunnel boring machine tunnel surrounding rock strength intelligent prediction model training method and equipment. A model capable of effectively predicting coal mine tunnel TBM tunneling tunnel surrounding rock point load strength can be trained. The method comprises the following steps: acquiring first original data corresponding to a target heading machine; removing data of the target heading machine in a stop section and an empty pushing section in the first original data to obtain second original data; processing an abnormal value in the second original data to obtain target operation data; feature parameters in the target operation data are extracted; performing normalization processing on the feature parameters to obtain a model training data set; a Stacking algorithm model is constructed according to a base learner and a meta learner, the base learner comprises a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner; and training the Stacking algorithm model based on the model training data set so as to obtain the prediction model for the surrounding rock strength of the tunneling roadway of the full-face tunneling machine.

Description

Technical Field

[0001] The present invention relates to the field of tunnel excavation, and in particular to a method and equipment for training an intelligent prediction model of surrounding rock strength of a full-face tunnel boring machine. Background Art

[0002] Tunnel Boring Machine (TBM) excavation is an efficient, safe and environmentally friendly tunnel construction method. TBM construction is sensitive to geological conditions. The point load strength of the surrounding rock is a key indicator for measuring the excavability of the surrounding rock and determining the tunnel support parameters. It has important practical significance for optimizing excavation parameters and guiding tunnel support design. However, with the increasing complexity of the underground stratum environment, the current acquisition of surrounding rock strength information mainly relies on static rock mechanics experiments and geological exploration. This method has obvious lag and cannot be obtained in real time. Therefore, there is an urgent need for a technology to predict the point load strength of rock in TBM excavation tunnels. This technology will be able to quickly and in real time identify rock strength, provide accurate rock strength information for underground excavation operations, and effectively cope with the working environment of TBM under complex stratum conditions.

[0003] In recent years, with the surge in data volume and the development of artificial intelligence, various machine learning methods have gradually been introduced into the field of rock mass perception. The surrounding rock perception model is mainly established from two directions: image and excavation parameter analysis. In terms of images, deep learning is used to extract and segment the features of rock or hyperspectral images to achieve rock state recognition. Although image methods have achieved good results in rock mass recognition, due to the complex TBM excavation conditions in coal mine tunnels, it is difficult to collect clear rock mass images. Therefore, it is not suitable for rock state perception in coal mine tunnels. Summary of the invention

[0004] The embodiment of the present invention provides a method and device for training an intelligent prediction model of surrounding rock strength of a full-section tunnel boring machine, which can train a model that can effectively predict the point load strength of surrounding rock in a TBM tunneling tunnel of a coal mine.

[0005] The first aspect of the present invention provides a method for training an intelligent prediction model of surrounding rock strength of a full-face tunnel boring machine, comprising:

[0006] Acquire first original data corresponding to the target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load strength data;

[0007] Eliminate the data of the target tunneling machine in the shutdown section and the idle pushing section in the first original data to obtain the second original data;

[0008] Processing the abnormal values ​​in the second original data to obtain target operating data;

[0009] Extracting characteristic parameters from the target operation data, wherein the characteristic parameters include cutterhead thrust, cutterhead torque, penetration, cutterhead rotation speed, cutterhead vibration acceleration, and rock mass geomechanical property parameters;

[0010] Normalizing the characteristic parameters to obtain a model training data set;

[0011] A Stacking algorithm model is constructed according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner;

[0012] The Stacking algorithm model is trained based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0013] The second aspect of the present invention provides a full-face tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device, comprising:

[0014] An acquisition module, used to acquire first original data corresponding to a target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load intensity data;

[0015] A removal module, used for removing the data of the target tunneling machine in the stop section and the idle pushing section in the first original data, so as to obtain the second original data;

[0016] an abnormal value processing module, used for processing abnormal values ​​in the second original data to obtain target operation data;

[0017] A feature extraction module is used to extract feature parameters from the target operation data, wherein the feature parameters include cutter head thrust, cutter head torque, penetration, cutter head rotation speed, cutter head vibration acceleration and rock mass geomechanical property parameters;

[0018] A normalization module, used to normalize the feature parameters to obtain a model training data set;

[0019] A construction module, used to construct a Stacking algorithm model according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner;

[0020] A training module is used to train the Stacking algorithm model based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0021] The third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the processor is used to implement the steps of the intelligent prediction model training method for the surrounding rock strength of the full-section tunnel boring machine as described in the first aspect above when executing the computer management program stored in the memory.

[0022] The fourth aspect of the present invention provides a computer-readable storage medium having a computer management program stored thereon. When the computer management program is executed by a processor, the steps of the intelligent prediction model training method for the surrounding rock strength of a full-section tunnel boring machine as described in the first aspect above are implemented.

[0023] In summary, it can be seen that in the embodiment provided by the present invention, relevant data in the excavation process are obtained by the on-site collection method, and invalid data and abnormal data in the original data are eliminated; based on the rock-machine relationship analysis, the input characteristics of the model are determined, and three machine learning algorithms, namely gradient boosting decision tree (GBDT), random forest (RF), and support vector regression (SVR), are used as base learners, and the linear regression (LR) algorithm is used as a meta-learner to train a prediction model based on the Stacking ensemble algorithm. Compared with the prediction model of a single machine learning algorithm, the goodness of fit is the highest, the error index value is the smallest, and it has higher prediction accuracy, and can train a model that effectively predicts the load strength of surrounding rock points in coal mine tunnels TBM excavation tunnels. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of a flow chart of a method for training an intelligent prediction model for surrounding rock strength of a full-face tunnel boring machine provided in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of raw data provided by an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of complete on-site excavation data provided by an embodiment of the present invention;

[0027] Figure 4 A schematic diagram of a set of excavation cycle data provided by an embodiment of the present invention;

[0028] Figure 5 A schematic diagram of the structure of a box plot provided in an embodiment of the present invention;

[0029] Figure 6 A schematic diagram of the rock-machine interaction relationship provided by an embodiment of the present invention;

[0030] Figure 7 A schematic diagram of an input feature sampling interval provided in an embodiment of the present invention;

[0031] Figure 8A schematic diagram of the principle of the Stacking integration algorithm provided by an embodiment of the present invention;

[0032] Figure 9a A schematic diagram of the prediction results of the GBDT model provided in an embodiment of the present invention;

[0033] Figure 9b A schematic diagram of the RF model prediction results provided by an embodiment of the present invention;

[0034] Fig.9c A schematic diagram of the prediction results of the SVR model provided in an embodiment of the present invention;

[0035] Figure 9d A schematic diagram of the prediction results of the Stacking model provided in an embodiment of the present invention;

[0036] Fig.10 A virtual structural schematic diagram of a full-face tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device provided by an embodiment of the present invention;

[0037] Fig.11 A schematic diagram of the hardware structure of a full-face tunnel boring machine surrounding rock strength intelligent prediction model training device provided by an embodiment of the present invention;

[0038] Fig.12 A schematic diagram of an electronic device according to an embodiment of the present invention;

[0039] Fig.13 A schematic diagram of an embodiment of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0041] The tunneling parameter identification method has lower requirements on the site environment and higher practicality. For example, after collecting TBM on-site tunneling data, a rock mass excavability grade perception model based on random forest (RF) was established, which can accurately predict the surrounding rock related characteristics. The quantum genetic algorithm (QGA) and radial basis neural network (RBF) were combined to achieve advanced identification of surrounding rock categories. Based on the degree neural network (DNN), 10 tunneling parameters related to surrounding rock conditions were used as input features to establish a surrounding rock state perception model. Using 5 selected characteristic input parameters, a long short-term memory (LSTM) model was constructed to identify the crushing state of tunnel surrounding rock. The back propagation (BP) neural network was used to establish a nonlinear relationship between rock mass state and input parameters. Using various parameters in the TBM tunneling process as input features, a prediction model for surrounding rock state was established, which has high accuracy and practicality. However, a single machine learning algorithm often fails to fully utilize the information of the data, and its prediction effect may vary greatly for different data distributions and feature combinations.

[0042] In view of this, the present invention provides a full-section tunnel boring machine tunnel surrounding rock strength intelligent prediction model training method to fuse multiple models, combine different machine learning algorithms, so that the multiple algorithms can be coordinated and optimized, thereby improving the prediction accuracy of the model.

[0043] The following describes the method for training an intelligent prediction model for the surrounding rock strength of a full-section tunnel boring machine provided by the present invention from the perspective of a training device for an intelligent prediction model for the surrounding rock strength of a full-section tunnel boring machine. The training device for an intelligent prediction model for the surrounding rock strength of a full-section tunnel boring machine can be a server or a service unit in a server, without specific limitation.

[0044] See also Figure 1 , Figure 1 A flow chart of a method for training a surrounding rock strength intelligent prediction model for a full-section tunnel boring machine provided in an embodiment of the present invention, wherein the method comprises:

[0045] 101. Obtain first original data corresponding to a target tunnel boring machine.

[0046] In this embodiment, the full-section tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device can obtain the first original data corresponding to the target tunnel boring machine, and the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data and surrounding rock point load strength data. Among them, the full-section tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device exports data through the airborne system to obtain 164 types of tunnel boring machine operation mechanical electro-hydraulic parameter data including the tunnel boring machine cutterhead cutting status, tunnel boring machine walking status, anchor drilling rig drilling status and belt conveyor operation status in real time, with a sampling frequency of 1Hz, please refer to Figure 2 , Figure 2 Schematic diagram of the raw data provided by the embodiment of the present invention. Since the change in surrounding rock strength is only closely related to the cutting state of the cutter disc, the redundant data is eliminated and the cutting state parameter data is retained. The vibration of the cutter disc during rock breaking can also reflect the changes in the rock mass in front to a certain extent, so the cutter disc vibration acceleration is also used as the main parameter for collection, and the sampling frequency is 100Hz. The point load test is used to collect the point load strength data state of the rock mass. Table 1 is the collection result of the point load data:

[0047] Table 1

[0048]

[0049]

[0050] 102. Eliminate data in which the target tunnel boring machine is in a shutdown section and an idle pushing section from the first original data to obtain second original data.

[0051] In this embodiment, generally speaking, the operation process of the TBM excavation cycle can be regarded as the "excavation-stop-excavation" stage, which is repeated back and forth. Therefore, the original data collected on site also includes data of the two stages of excavation and stop (such as Figure 3 Among them, the excavation section data can be divided into the empty push section, the rising section and the steady state section (as shown in Figure 4 As shown). In the idle pushing stage, the cutter head speed increases and approaches the face rock mass; in the ascending stage, the cutter head contacts the face, the cutter gradually penetrates into the rock mass, and the values ​​of torque, propulsion force, etc. gradually increase to the set rock breaking parameter requirements; in the steady-state stage, the cutter head rotates to break the rock, and various values ​​tend to be stable, maintaining a small range of fluctuations around a value. The key to establishing a surrounding rock strength prediction model is to analyze the relationship between the rock-machine interaction data generated during the contact between the TBM's cutter head and the coal rock mass. Therefore, it is necessary to eliminate the shutdown section and the idle pushing section data in the original data, and retain the excavation data in the ascending stage and the steady-state stage. Specifically, the first sub-data of the target tunnel boring machine in the shutdown section and the second sub-data of the target tunnel boring machine in the excavation stage in the first original data can be determined by the following formula:

[0052] S = f(F)·f(T)·f(RPM);

[0053] Among them, F is the cutter head thrust, T is the cutter head torque, RPM is the cutter head speed, S is the discriminant function,

[0054] And according to the second sub-data, the third sub-data of the target tunnel boring machine in the empty push section is determined, that is, the first tunneling data in the second sub-data is determined, and the target time is determined based on the time of the first tunneling data and the preset time. Finally, the data between the time of the first tunneling data and the target time is determined as the third sub-data. In order to delete the empty push section data in the tunneling section, by comparing the large amount of original data collected, it is found that the data in this stage are all within 40s before the tunneling section data (here the preset time length is 40s for explanation, of course, it can also be adjusted according to the actual situation, and is not limited to the specific), such as Figure 4 Therefore, the data points within the first 40 seconds of the excavation stage are deleted by manual processing. The first sub-data and the third sub-data are removed from the first original data to obtain the second original data.

[0055] 103. Process the abnormal values ​​in the second original data to obtain target operating data.

[0056] In this embodiment, during the TBM excavation process, due to the harsh working environment and equipment failure, some abnormal data will exist in the original data collected on site. These abnormal data will not only interfere with the subsequent model training, resulting in a decrease in model accuracy, but also increase the amount of calculation and increase the complexity of model learning. Therefore, it is necessary to remove the abnormal data, specifically:

[0057] Based on the box plot, the upper box whisker and lower box whisker corresponding to the second original data are determined by the following formula:

[0058] UEV=Q3+1.5·IQR;

[0059] LEV = Q1-1.5·IQR;

[0060] Among them, UEV is the upper box whisker, LEV is the lower box whisker, Q3 is the upper quartile corresponding to the box plot, Q1 is the lower quartile corresponding to the box plot, and IQR is the lower quartile difference corresponding to the box plot.

[0061] Determine the data value that is not between the upper box whisker and the lower box whisker in the second original data as an abnormal value;

[0062] Eliminate outliers in the second original data to obtain third operating data;

[0063] The missing values ​​in the third operating data are supplemented based on linear interpolation to obtain target operating data.

[0064] See also Figure 5 , Figure 5The schematic diagram of the box plot composition structure provided by the embodiment of the present invention, the full-section tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device can determine the upper box whiskers and the lower box whiskers, and determine the data values ​​in the second original data that are not between the upper box whiskers and the lower box whiskers as outliers. For outliers in the original data, due to the traditional outlier removal method 3σ criterion is also called the Laida criterion, this method needs to assume in advance that the original data conforms to the normal distribution, but the actual data often does not conform to this rule, and this method uses the mean and standard deviation as the standard for outlier judgment, which is susceptible to interference from outliers. Compared with the 3σ criterion, the box plot outlier removal method uses quartiles and interquartile ranges as the basis for outlier judgment. These statistics have a certain resistance to outliers, and this method does not need to assume that the data obeys a normal distribution. Therefore, the box plot method is used to remove outliers in the original data. The box plot consists of three main parts: box body, box whiskers and outliers. The box is composed of the upper quartile, median and lower quartile, and the whiskers are divided into upper whiskers and lower whiskers. The values ​​that are not between the upper whiskers and the lower whiskers in the box plot are determined as outliers and removed. Finally, the missing values ​​are supplemented by linear interpolation.

[0065] 104. Extract characteristic parameters from the target operation data.

[0066] In this embodiment, during the TBM excavation process, the cutterhead is in direct contact with the rock mass of the excavation face, and the parameters of its cutting part are directly affected by the state of the rock mass in front. As the cutting process proceeds, the working face forms a circular crushing section under the continuous action of the roller cutter breaking the rock. When the state of the cutting surrounding rock changes, the existence of the surrounding rock crushing section not only affects the torque T and the speed RPM of the cutting part, but also causes the direction of the reaction force of the working face on the roller cutter to change dynamically. Therefore, the cutterhead cutting thrust F should be included in the category of key parameters of the main input. As the cutterhead rotates continuously, the roller cutters periodically break the surrounding rock. In a rock breaking cycle, the maximum impact state occurs at the beginning of the contact between the roller cutter and the rock mass. At this time, the rock mass is compressed, resulting in a rapid increase in the load of the cutting part. When the stress reaches its own material limit, the rock mass is crushed and the cutting part is unloaded. The continuous step crushing process causes the acceleration to change rapidly, and the vibration signal A and penetration signal REV generated thereby should also be included in the category of key parameters reflecting the state of the rock mass. Therefore, the cutterhead thrust F, cutterhead torque T, penetration REV, cutterhead speed RPM, and cutterhead vibration acceleration A are determined as the main input characteristics of the model. Figure 6 shown.

[0067] For the steady-state phase of the tunneling cycle, the reflection of the TBM equipment status parameters on the geological conditions of the tunnel face is mainly reflected in the mean and fluctuation range of the tunneling parameters within the interval. Figure 7It can be seen that in the steady-state stage, the excavation parameters fluctuate around a certain value, and the data in the first 50 seconds can basically reflect the law and characteristics of its subsequent changes. Considering that in the actual excavation process of the tunnel, it is more beneficial to the excavation process to identify the strength value of the surrounding rock mass in front as early as possible. Therefore, the present invention selects the mean value of the first 50 seconds of the steady-state stage of TBM excavation (F m , T m , Rev. m , RPM m , A m ) and standard deviation (F n , T n , Rev. n , RPM n , A n ) as the input parameters of the model, and the final selected input parameters include 10 eigenvalues.

[0068] 105. Normalize the feature parameters to obtain a model training data set.

[0069] In this embodiment, the full-section roadheader tunnel surrounding rock strength intelligent prediction model training device can perform normalization processing on the characteristic parameters (the purpose of the normalization processing is to eliminate the dimension and scale differences of the data and improve the stability and accuracy of the model) to obtain a model training data set. Specifically, the characteristic parameters can be normalized by the following formula to obtain a model training data set:

[0070]

[0071] Wherein, x* is the characteristic parameter after normalization, x is the characteristic parameter, and xmax and xmin are the maximum value and minimum value of the characteristic parameter, respectively.

[0072] It should be noted that in this embodiment, the tunneling cycle corresponding to the target operation data is 524. The data preprocessing method is used to process five tunneling state parameters of the tunnel boring machine (cutterhead thrust F, cutterhead torque T, penetration REV, cutterhead speed RPM, cutterhead vibration acceleration A) and one rock mass geomechanical property parameter (point load intensity value IS) to form a rock-machine interaction database containing the mean and standard deviation, with a total of 524 groups of data, which provides a data basis for the subsequent learning and verification of the algorithm model.

[0073] 106. Construct a Stacking algorithm model based on the base learner and meta learner.

[0074] In this embodiment, the Stacking ensemble algorithm is used to build the model. The Stacking ensemble algorithm is a hierarchical algorithm that uses the outputs of multiple different base learners in the first layer as new input features, and then performs comprehensive learning through a meta-learner in the second layer, thereby optimizing and integrating the base learners and improving the prediction accuracy of the prediction model. The principle of the Stacking ensemble algorithm is as follows: Figure 8 shown.

[0075] Base learners and meta-learners are important components of the Stacking ensemble algorithm, and they directly affect the performance and effect of the ensemble model. The first layer of the Stacking algorithm model should select different types of base learners to increase the diversity of the model. The second layer of meta-learners should select simple models to prevent overfitting. Since the geomechanical parameters of coal rock mass are continuous variables, the above-mentioned mutual feedback relationship is a regression problem between multiple independent variables and a dependent variable. Therefore, the present invention uses a variety of machine learning algorithms that can be used for regression analysis as the base learners of the first layer of the Stacking model, including GBDT (gradient boosting decision tree), RF (random forest) and SVR (support vector regression algorithm). GBDT (gradient boosting decision tree) is an ensemble learning algorithm based on the boosting algorithm, which uses multiple regression trees as base learners and uses a series of optimization techniques to accelerate the training of the model and improve the accuracy of the model. The RF (random forest) algorithm is an ensemble learning method composed of multiple decision trees, which votes or averages the prediction results of all decision trees to improve the accuracy of the prediction model. SVR (support vector regression algorithm) is a supervised learning method for regression problems, which can avoid overfitting and can better handle nonlinear regression problems. These base learners have different learning capabilities and characteristics, and can learn and analyze data from different angles and levels, thereby improving the diversity and complementarity of the integrated model. In the second layer, LR (linear regression algorithm) is selected as the meta-learner. The linear regression model is a simple and effective model used to fit the linear relationship between features and labels in the training set to avoid overfitting.

[0076] 107. The Stacking algorithm model is trained based on the model training data set to obtain a prediction model for the surrounding rock strength of the full-section tunnel boring machine.

[0077] In this embodiment, after constructing the Stacking algorithm model, the full-section tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device can train the Stacking algorithm model based on the model training data set to obtain the full-section tunnel boring machine tunnel surrounding rock strength prediction model.

[0078] In one embodiment, training the Stacking algorithm model based on the model training data set to obtain a full-face roadheader tunnel surrounding rock strength prediction model includes:

[0079] Divide the model training data into training set and test set;

[0080] Divide the training set into multiple training subsets;

[0081] The first training subset is determined as a validation set, and the other training subsets except the first training subset among the multiple training subsets are used as training sets, and each regression tree in the multiple regression trees is trained by a K-fold cross-validation method to obtain multiple first prediction results corresponding to each regression tree;

[0082] Based on each test data in the test set, each trained regression tree is tested respectively to obtain a second prediction result corresponding to each regression tree;

[0083] A plurality of first prediction results are input into a meta-learner for training, and the trained meta-learner is tested based on the second prediction result to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0084] In this embodiment, the model training data set (524 groups of data) is divided into a training set and a test set according to a ratio of 5:1, and 437 groups of training data and 87 groups of test data are obtained. The training data is used to train the Stacking integrated model, and the test data is used to verify the prediction effect of the model. In order to verify that the Stacking integrated model has a good prediction effect on the point load strength on different lithologies, the selection of test data ensures that the data of the three lithologies of sandy mudstone, fine sandstone, and mudstone are consistent, all of which are 29 groups. This can avoid the model prediction deviation caused by uneven distribution of lithology.

[0085] The training device for the intelligent prediction model of surrounding rock strength of the full-face tunnel boring machine can divide the training set into multiple training subsets (for example, 4 or 5, not specifically limited, and 4 are used as an example for explanation here), train1, train2, train3, train4. And select the base learner, which assumes that three regression trees, GBDT (gradient boosting decision tree), RF (random forest) and SVR (support vector regression algorithm), are selected as the base learners.

[0086] For example, GBDT: use train1, train2, train3 and train4 as validation sets, and the remaining three sets as training sets, perform k-fold cross validation to train the model; then make predictions on the test set. This will result in 4 prediction results trained by the GBDT algorithm on the training set and 1 prediction value B1 on the test set. These 4 prediction results are overlapped and merged vertically to obtain A1. The above is explained using GBDT as an example. The RF algorithm and SVR algorithm are similar to GBDT, and the details will not be repeated here.

[0087] After the three regression trees in the base learner are trained, the predicted values ​​of the three regression trees on the training set are used as three features A1, A2 and A3 respectively, and trained using a meta-learner (LR model is used as the meta-learner for illustration) to establish an LR model. The trained LR model is used to make predictions on the values ​​of the three features (B1, B2, B3) constructed by the predicted values ​​of the three base models on the test set before, and the final predicted category or probability is obtained.

[0088] It should be noted that in machine learning research, the performance of the model is often affected by the hyperparameters. Therefore, the enumeration method is used in the present invention to test and tune the hyperparameters of the Stacking algorithm model multiple times, and finally determine the hyperparameters of the Stacking algorithm model. At the same time, in order to evaluate the performance and effect of the prediction model, the mean square error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), goodness of fit (R 2 ) as the model performance evaluation criteria. The calculation formulas of these evaluation indicators are as follows:

[0089]

[0090] It should also be noted that in order to evaluate the prediction effect of the surrounding rock strength prediction model of the full-section tunnel boring machine, three control models were established using single GBDT, RF and SVR respectively. The training process and data used in the model are exactly the same. The hyperparameter setting of the model uses enumeration method to take multiple values ​​and select the optimal value. The prediction accuracy of different models is discussed through the evaluation criteria to evaluate the prediction effect of the model.

[0091] The comparison between the predicted results and actual values ​​of the surrounding rock point load strength of the Stacking integrated model (i.e., the full-face TBM tunneling tunneling surrounding rock strength prediction model) and the other three control models on 87 groups of test samples is shown in Figure 9. From the analysis of the figure, it can be seen that the prediction of the surrounding rock point load strength by different models has a good degree of fit and explanation. Among them, the Stacking model has the best prediction effect. The error between the model's prediction results and the true value is small in most test samples, and only a few individual samples have errors with the true value. The GBDT model and the RF model are followed. The prediction results of these two models are also close to the true value, but there are large deviations in some samples. The SVR model has the worst effect, and the difference between the model's prediction results and the true value is large, indicating that the model has a weak ability to predict the surrounding rock point load strength.

[0092] The prediction results of the model are calculated using the evaluation index, and the calculation results are shown in Table 3. From the data in the analysis table, it can be seen that the Stacking integrated model has the largest goodness of fit R2 value on the test set, which is 0.939, which is 0.03, 0.092 and 0.148 higher than the three single models of GBDT, RF and SVR, which are 0.909, 0.847 and 0.791, respectively, and the Stacking integrated model has the smallest values ​​in the remaining MSE, MAE and MAPE indicators. For a model, the closer its R2 is to 1, the smaller the MSE, MAE and MAPE are, the better the model prediction ability is. This shows that the model proposed in the present invention has higher prediction accuracy, and the Stacking integrated model is used to predict the point load strength of the surrounding rock during the TBM excavation process of the coal mine tunnel, which can better establish a feedback relationship model that characterizes the rock-machine interaction during the tunnel excavation process, and also provide a basis for excavation control and tunnel support parameter adjustment.

[0093] Table 3

[0094] Model Name MSE MAE <![CDATA[R 2 ]]> MAPE GBD 0.105 0.252 0.909 0.167 RF 0.177 0.304 0.847 0.204 SVR 0.241 0.383 0.791 0.282 Stacking 0.071 0.186 0.939 0.105

[0095] Table 3

[0096] In summary, it can be seen that in the embodiment provided by the present invention, relevant data in the excavation process are obtained by the on-site collection method, and invalid data and abnormal data in the original data are eliminated; based on the rock-machine relationship analysis, the input characteristics of the model are determined, and three machine learning algorithms, namely gradient boosting decision tree (GBDT), random forest (RF), and support vector regression (SVR), are used as base learners, and the linear regression (LR) algorithm is used as a meta-learner to train a prediction model based on the Stacking ensemble algorithm. Compared with the prediction model of a single machine learning algorithm, the goodness of fit is the highest, the error index value is the smallest, and it has higher prediction accuracy, which can effectively predict the load strength of the surrounding rock points in the TBM excavation tunnel of the coal mine tunnel.

[0097] The above describes an embodiment of the present invention from the perspective of a training method for an intelligent prediction model for surrounding rock strength in a full-section tunnel boring machine. The following describes an embodiment of the present invention from the perspective of a training device for an intelligent prediction model for surrounding rock strength in a full-section tunnel boring machine.

[0098] See also Fig.10 , a virtual structural diagram of a full-section tunnel boring machine roadway surrounding rock strength intelligent prediction model training device in an embodiment of the present invention, the full-section tunnel boring machine roadway surrounding rock strength intelligent prediction model training device 1000 comprises:

[0099] An acquisition module 1001 is used to acquire first original data corresponding to a target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load intensity data;

[0100] A removal module 1002 is used to remove the data of the target tunnel boring machine in the stop section and the idle pushing section in the first original data to obtain the second original data;

[0101] An outlier processing module 1003 is used to process outliers in the second original data to obtain target operating data;

[0102] A feature extraction module 1004 is used to extract feature parameters from the target operation data, wherein the feature parameters include cutter head thrust, cutter head torque, penetration, cutter head rotation speed, cutter head vibration acceleration and rock mass geomechanical property parameters;

[0103] A normalization module 1005 is used to perform normalization processing on the characteristic parameters to obtain a model training data set;

[0104] A construction module 1006 is used to construct a Stacking algorithm model according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner;

[0105] The training module 1007 is used to train the Stacking algorithm model based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0106] In a possible design, the rejection module 1002 is specifically used for:

[0107] The first sub-data indicating that the target tunnel boring machine is in the shutdown stage and the second sub-data indicating that the target tunnel boring machine is in the excavation stage in the first original data are determined by the following formula:

[0108] S = f(F)·f(T)·f(RPM);

[0109] Among them, F is the cutter head thrust, T is the cutter head torque, RPM is the cutter head speed, S is the discriminant function,

[0110] Determine third sub-data indicating that the target tunnel boring machine is in an empty pushing section according to the second sub-data;

[0111] The first sub-data and the third sub-data are removed from the first original data to obtain the second original data.

[0112] In a possible design, the rejection module 1002 is further specifically used for:

[0113] Determining the first excavation data in the second sub-data;

[0114] Determine the target time based on the time of the first excavation data and the preset duration;

[0115] The data between the time of the first excavation data and the target time is determined as the third sub-data.

[0116] In a possible design, the outlier processing module 1003 is specifically used for:

[0117] Based on the box plot, the upper box whisker and the lower box whisker corresponding to the second original data are determined by the following formula:

[0118] UEV=Q3+1.5·IQR;

[0119] LEV = Q1-1.5·IQR;

[0120] Among them, UEV is the upper box whisker, LEV is the lower box whisker, Q3 is the upper quartile corresponding to the box plot, Q1 is the lower quartile corresponding to the box plot, and IQR is the lower quartile difference corresponding to the box plot.

[0121]

[0122] Determine the data value in the second original data that is not between the upper box whisker and the lower box whisker as the abnormal value;

[0123] Eliminating the abnormal value in the second original data to obtain third operating data;

[0124] The missing values ​​in the third operating data are supplemented based on linear interpolation to obtain the target operating data.

[0125] In a possible design, the model training module 1007 is specifically used for:

[0126] Dividing the model training data into a training set and a test set;

[0127] Dividing the training set into a plurality of training subsets;

[0128] Determine the first training subset as a validation set, and use the other training subsets among the multiple training subsets except the first training subset as training sets, and train each regression tree in the multiple regression trees by a K-fold cross-validation method to obtain multiple first prediction results corresponding to each regression tree;

[0129] Testing each of the trained regression trees based on each test data in the test set to obtain a second prediction result corresponding to each regression tree;

[0130] The multiple first prediction results are input into the meta-learner for training, and the trained meta-learner is tested based on the second prediction result to obtain the surrounding rock strength prediction model of the full-face roadheader tunnel.

[0131] In a possible design, the normalization module 1005 is specifically used for:

[0132] The characteristic parameters are normalized by the following formula:

[0133]

[0134] Among them, x* is the characteristic parameter after normalization, x is the characteristic parameter, and xmax and xmin are the maximum value and minimum value of the characteristic parameter respectively.

[0135] above Fig.10The intelligent prediction model training device for the surrounding rock strength of the full-section roadheader tunnel in the embodiment of the present invention is described from the perspective of modular functional entities. The following is a detailed description of the intelligent prediction model training device for the surrounding rock strength of the full-section roadheader tunnel in the embodiment of the present invention from the perspective of hardware processing. Please refer to Fig.10 , a schematic diagram of an embodiment of a full-section tunnel boring machine roadway surrounding rock strength intelligent prediction model training device 1000 in an embodiment of the present invention, the full-section tunnel boring machine roadway surrounding rock strength intelligent prediction model training device 1000 comprises:

[0136] An input device 1001, an output device 1002, a processor 1003 and a memory 1004 (the number of the processor 1003 can be one or more, Fig.10 In some embodiments of the present invention, the input device 1001, the output device 1002, the processor 1003 and the memory 1004 may be connected via a communication bus or other means, wherein: Fig.10 The communication bus connection is taken as an example.

[0137] Wherein, by calling the operation instruction stored in the memory 1004, the processor 1003 is used to perform the following steps:

[0138] Acquire first original data corresponding to the target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load strength data;

[0139] Eliminate the data of the target tunneling machine in the shutdown section and the idle pushing section in the first original data to obtain the second original data;

[0140] Processing the abnormal values ​​in the second original data to obtain target operating data;

[0141] Extracting characteristic parameters from the target operation data, wherein the characteristic parameters include cutterhead thrust, cutterhead torque, penetration, cutterhead rotation speed, cutterhead vibration acceleration, and rock mass geomechanical property parameters;

[0142] Normalizing the characteristic parameters to obtain a model training data set;

[0143] A Stacking algorithm model is constructed according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner;

[0144] The Stacking algorithm model is trained based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0145] By calling the operation instructions stored in the memory 1004, the processor 1003 is also used to execute Figure 1 Any method in the corresponding embodiment.

[0146] See also Fig.11 , Fig.11 A schematic diagram of an electronic device according to an embodiment of the present invention.

[0147] like Fig.11 As shown, an embodiment of the present invention provides an electronic device, including a memory 1110, a processor 1120, and a computer program 1111 stored in the memory 1110 and executable on the processor 1120. When the processor 1120 executes the computer program 1111, the following steps are implemented:

[0148] Acquire first original data corresponding to the target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load strength data;

[0149] Eliminate the data of the target tunneling machine in the shutdown section and the idle pushing section in the first original data to obtain the second original data;

[0150] Processing the abnormal values ​​in the second original data to obtain target operating data;

[0151] Extracting characteristic parameters from the target operation data, wherein the characteristic parameters include cutterhead thrust, cutterhead torque, penetration, cutterhead rotation speed, cutterhead vibration acceleration, and rock mass geomechanical property parameters;

[0152] Normalizing the characteristic parameters to obtain a model training data set;

[0153] A Stacking algorithm model is constructed according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner;

[0154] The Stacking algorithm model is trained based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0155] In a specific implementation process, when the processor 1120 executes the computer program 1111, it can achieve Figure 1 Any implementation manner in the corresponding embodiments.

[0156] Since the electronic device introduced in this embodiment is a device used to implement a full-section tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device in the embodiment of the present invention, based on the method introduced in the embodiment of the present invention, technical personnel in this field can understand the specific implementation mode of the electronic device of this embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiment of the present invention will not be introduced in detail here. As long as the equipment used by technical personnel in this field to implement the method in the embodiment of the present invention is within the scope of protection of the present invention.

[0157] See also Fig.12 , Fig.12 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention.

[0158] like Fig.12 As shown, the embodiment of the present invention further provides a computer-readable storage medium 1200, on which a computer program 1211 is stored. When the computer program 1211 is executed by a processor, the following steps are implemented:

[0159] Acquire first original data corresponding to the target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load strength data;

[0160] Eliminate the data of the target tunneling machine in the shutdown section and the idle pushing section in the first original data to obtain the second original data;

[0161] Processing the abnormal values ​​in the second original data to obtain target operating data;

[0162] Extracting characteristic parameters from the target operation data, wherein the characteristic parameters include cutterhead thrust, cutterhead torque, penetration, cutterhead rotation speed, cutterhead vibration acceleration, and rock mass geomechanical property parameters;

[0163] Normalizing the characteristic parameters to obtain a model training data set;

[0164] A Stacking algorithm model is constructed according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner;

[0165] The Stacking algorithm model is trained based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

[0166] In the specific implementation process, the computer program 1211 is executed by the processor to achieve Figure 1 Any implementation manner in the corresponding embodiments.

[0167] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0168] The embodiment of the present invention also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device executes the following Figure 1 The process in the corresponding embodiment.

[0169] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for training an intelligent prediction model for surrounding rock strength of a full-face tunnel boring machine, characterized in that: include: Acquire first original data corresponding to the target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load strength data; Eliminate the data of the target tunneling machine in the shutdown section and the idle pushing section in the first original data to obtain the second original data; Processing the abnormal values ​​in the second original data to obtain target operating data; Extracting characteristic parameters from the target operation data, wherein the characteristic parameters include cutterhead thrust, cutterhead torque, penetration, cutterhead rotation speed, cutterhead vibration acceleration, and rock mass geomechanical property parameters; Normalizing the characteristic parameters to obtain a model training data set; A Stacking algorithm model is constructed according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner; The Stacking algorithm model is trained based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

2. The method according to claim 1, characterized in that The step of eliminating the data of the target tunnel boring machine being in the stop section and the idle pushing section in the first original data to obtain the second original data includes: The first sub-data indicating that the target tunnel boring machine is in the shutdown stage and the second sub-data indicating that the target tunnel boring machine is in the excavation stage in the first original data are determined by the following formula: S = f(F)·f(T)·f(RPM); Among them, F is the cutter head thrust, T is the cutter head torque, RPM is the cutter head speed, S is the discriminant function, Determine third sub-data indicating that the target tunnel boring machine is in an empty pushing section according to the second sub-data; The first sub-data and the third sub-data are removed from the first original data to obtain the second original data.

3. The method according to claim 2, characterized in that The third sub-data for determining that the target tunnel boring machine is in the empty push section according to the second sub-data includes: Determining the first excavation data in the second sub-data; Determine the target time based on the time of the first excavation data and the preset duration; The data between the time of the first excavation data and the target time is determined as the third sub-data.

4. The method according to claim 1, characterized in that: The processing of the abnormal value in the second original data to obtain the target operation data includes: Based on the box plot, the upper box whisker and the lower box whisker corresponding to the second original data are determined by the following formula: UEV=Q3+1.5·IQR; LEV = Q1-1.5·IQR; Among them, UEV is the upper box whisker, LEV is the lower box whisker, Q3 is the upper quartile corresponding to the box plot, Q1 is the lower quartile corresponding to the box plot, and IQR is the lower quartile difference corresponding to the box plot. Determine the data value in the second original data that is not between the upper box whisker and the lower box whisker as the abnormal value; Eliminating the abnormal value in the second original data to obtain third operating data; The missing values ​​in the third operating data are supplemented based on linear interpolation to obtain the target operating data.

5. The method according to any one of claims 1 to 4, characterized in that The method of training the Stacking algorithm model based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel includes: Dividing the model training data into a training set and a test set; Dividing the training set into a plurality of training subsets; Determine the first training subset as a validation set, and use the other training subsets among the multiple training subsets except the first training subset as training sets, and train each regression tree in the multiple regression trees by a K-fold cross-validation method to obtain multiple first prediction results corresponding to each regression tree; Testing each of the trained regression trees based on each test data in the test set to obtain a second prediction result corresponding to each regression tree; The multiple first prediction results are input into the meta-learner for training, and the trained meta-learner is tested based on the second prediction result to obtain the surrounding rock strength prediction model of the full-face roadheader tunnel.

6. The method according to any one of claims 1 to 4, characterized in that The normalizing the feature parameters to obtain a model training data set includes: The characteristic parameters are normalized by the following formula: Among them, x* is the characteristic parameter after normalization, x is the characteristic parameter, and xmax and xmin are the maximum value and minimum value of the characteristic parameter respectively.

7. A full-face tunnel boring machine tunnel surrounding rock strength intelligent prediction model training device, characterized in that: include: An acquisition module, used to acquire first original data corresponding to a target tunnel boring machine, wherein the first original data includes tunnel boring machine operation status data, cutterhead vibration acceleration data, and surrounding rock point load intensity data; A removal module, used for removing the data of the target tunneling machine in the stop section and the idle pushing section in the first original data, so as to obtain the second original data; an abnormal value processing module, used for processing abnormal values ​​in the second original data to obtain target operation data; A feature extraction module is used to extract feature parameters from the target operation data, wherein the feature parameters include cutter head thrust, cutter head torque, penetration, cutter head rotation speed, cutter head vibration acceleration and rock mass geomechanical property parameters; A normalization module, used to normalize the feature parameters to obtain a model training data set; A construction module, used to construct a Stacking algorithm model according to a base learner and a meta learner, wherein the base learner includes a plurality of regression trees, and the output result of the base learner is used as the input of the meta learner; A training module is used to train the Stacking algorithm model based on the model training data set to obtain a surrounding rock strength prediction model for a full-face roadheader tunnel.

8. The device according to claim 7, characterized in that The rejection module is specifically used for: The first sub-data indicating that the target tunnel boring machine is in the shutdown stage and the second sub-data indicating that the target tunnel boring machine is in the excavation stage in the first original data are determined by the following formula: S = f(F)f(T)f(RPM); Among them, F is the cutter head thrust, T is the cutter head torque, RPM is the cutter head speed, S is the discriminant function, Determine third sub-data indicating that the target tunnel boring machine is in an empty pushing section according to the second sub-data; The first sub-data and the third sub-data are removed from the first original data to obtain the second original data.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to implement the intelligent prediction model training method for the surrounding rock strength of a full-section tunnel boring machine as described in any one of claims 1 to 6 when executing the computer management program stored in the memory.

10. A computer-readable storage medium having a computer management program stored thereon, characterized in that: When the computer management program is executed by the processor, the intelligent prediction model training method for the surrounding rock strength of the full-section tunnel boring machine as described in any one of claims 1 to 6 is implemented.

Images