An intelligent auxiliary method for shield tunneling based on stratum classification

Abstract

An intelligent auxiliary method for shield tunneling based on stratum classification, the steps include collecting parameters of the current shield machine during the tunneling process; calculating the Pearson correlation coefficient to obtain the correlation coefficients of each parameter with the total thrust and the cutter head torque, and obtaining the feature combination; using the K-means algorithm to calculate and obtain the stratum classification results clustered with the basic combination composed of the total thrust and the cutter head torque and the feature combination as the feature set; using the cooperative training algorithm to obtain the classification of the excavated strata based on the basic combination and the feature combination under semi-supervised learning; using the machine learning algorithm to learn and train the excavated stratum categories, identify the stratum conditions with unknown excavation directions, and provide the reference range of the total thrust and the cutter head torque. The method of the present invention can be executed cyclically during the shield tunneling process, and the classification results of the excavated strata are continuously updated. It has the characteristics of practical feasibility and real-time performance, and can provide important auxiliary decision-making information for shield tunneling.

Description

Technical Field

[0001] The invention relates to an intelligent auxiliary method for shield tunneling based on stratum type, belonging to the technical field of shield tunneling. Background Art

[0002] With the development of urbanization, underground transportation systems have been vigorously developed. As a commonly used construction method for urban rail transit projects, the shield method has the characteristics of full enclosure, relatively fast excavation speed, and simultaneous excavation and assembly. Traditional shield excavation construction mainly relies on the driver's personal experience. Due to the randomness of the survey drilling, the geological conditions in front of the shield machine during excavation are unknown, which makes it difficult to control the shield machine's excavation energy consumption, excavation efficiency, and excavation posture.

[0003] In recent years, the continuous development of high-tech science and technology such as big data, artificial intelligence, intelligent monitoring, and machine learning has provided possibilities for intelligent shield tunneling. The current assisted driving strategy is mainly based on big data prediction of shield tunneling parameters. The strategy of machine assistance for shield tunneling parameter prediction generally utilizes a large number of shield tunneling parameters, synchronous grouting parameters and other features, and predicts the target parameters based on machine learning methods. Due to the lack of effective data screening and the limitations of machine learning strategies, the current method will have a large error between the predicted value and the actual value in the local range, as well as inconsistent trends. It is difficult to give a practical reference for the parameter setting of the shield machine, resulting in unreasonable or inaccurate tunneling parameter ranges provided by the intelligent assistance system. Summary of the invention

[0004] The purpose of the invention is to provide an intelligent auxiliary method for shield tunneling based on stratum category, which can provide tunneling parameters for the intelligent auxiliary system in real time and provide reasonable and accurate auxiliary decision-making information for shield tunneling.

[0005] In order to achieve the above object, the present invention provides an intelligent auxiliary method for shield tunneling based on stratum type, comprising the following steps:

[0006] S1. Collect the shield tunneling parameters and synchronous grouting parameters of the current shield machine during the tunneling process, which are divided into shield correlation research parameters consisting of total thrust and cutter head torque in the shield tunneling parameters, shield tunneling parameters excluding total thrust and cutter head torque, and synchronous grouting parameters, and identify and eliminate abnormal values ​​and extreme values ​​in the parameters;

[0007] S2. By calculating the Pearson correlation coefficient, respectively calculate the correlation coefficients between the total thrust and each shield correlation research parameter, and between the cutterhead torque and each shield correlation research parameter, respectively obtain the shield correlation research parameters above the set confidence level critical value, and select and combine the common parameters of the shield correlation research parameters with the total thrust correlation coefficient above the set confidence level critical value and the shield correlation research parameters with the cutterhead torque correlation coefficient above the set confidence level critical value to obtain a characteristic combination;

[0008] S3, using the K-means algorithm to calculate and obtain formation classification results by clustering the basic combination consisting of the total thrust and the cutter head torque and the feature combination as the feature set;

[0009] S4, using a cooperative training algorithm, adding the stratum classification results obtained in S3 to each other to form two new feature sets, and then clustering the two new feature sets respectively, and adding the new stratum classification results to the other feature set again for clustering, and so on, continuously iterating, and obtaining the classification results of the excavated strata based on the basic combination and the feature combination under semi-supervised learning, as well as the basic combination parameter range corresponding to each stratum;

[0010] S5. Use machine learning algorithms to train the stratum categories within the excavation range, identify the stratum conditions with unknown excavation directions, and obtain the reference value range of the total thrust and cutterhead torque.

[0011] Furthermore, in S1, the shield tunneling parameters of the shield machine during the tunneling process include the upper pressure, lower pressure, left pressure, right pressure, propulsion speed, upper soil pressure, cutter head torque, and excavation volume of the propulsion cylinder obtained by sensors installed on the shield machine, and the synchronous grouting parameters include grouting pressure and grouting volume; and the outliers and extreme values ​​in the parameter data are identified and eliminated by using a box plot.

[0012] Furthermore, in S2, the Pearson correlation coefficient calculation formula is as follows: (1);

[0013] In the formula, They are the sample points, sample averages, and sample standard deviations of the total thrust or cutterhead torque parameters in the shield tunneling parameters. are the sample points, sample mean values, and sample standard deviations of any parameter in the shield correlation research parameters composed of the shield tunneling parameters excluding the total thrust and cutterhead torque and the synchronous grouting parameters, respectively; n is the number of shield parameter data points after data processing;

[0014] According to formula (1), the Pearson correlation coefficients of each parameter in the total thrust and shield correlation research parameters, and each parameter in the cutterhead torque and shield correlation research parameters are calculated one by one, and the obtained Pearson correlation coefficients are sorted from high to low, and the shield correlation research parameters with correlation coefficients above the set confidence level critical value are respectively taken out, and the common parameters of the shield correlation research parameters with the total thrust correlation coefficients above the set confidence level critical value and the shield correlation research parameters with the cutterhead torque correlation coefficients above the set confidence level critical value are selected and combined to form a characteristic combination; wherein, the confidence level can be 0.1, 0.05 or 0.01, and the correlation coefficient test critical value can be found according to the correlation coefficient test table.

[0015] Furthermore, in S3, the process of using the K-means algorithm to calculate and obtain the stratum classification results clustered with the basic combination and the feature combination as the feature set is as follows:

[0016] S3.1. Select the number of clusters corresponding to the location of the slope change point through the elbow diagram, that is, determine the number of clusters for stratum classification by selecting the number of clusters corresponding to the slope change point on the elbow diagram; the smaller the number of clusters, the greater the loss; as the number of clusters increases, the loss gradually decreases, but when the number of clusters is relatively large, the reduction in loss is no longer obvious;

[0017] S3.2. Use the second-order difference method of regression coefficient to determine the slope change point of the elbow graph:

[0018] S3.2.1. The number of abscissa clusters is used to form an exploration point sequence. Starting from the exploration point, linear regression is performed on the data points including the exploration point and before and after the exploration point to obtain the regression coefficient.

[0019] S3.2.2. As the sliding window slides backward, the regression coefficients corresponding to each exploration point are obtained, and the first-order difference of the slope is obtained by making a difference between the two, which reflects the change in the slope at the exploration point; the calculation formula of the first-order difference of the slope is: (2);

[0020] In the formula, For exploration points, is the linear regression coefficient of the sliding window where the exploration point is located;

[0021] S3.2.3. Perform a second-order difference on the first-order difference sequence of the slope to reflect the magnitude of the change in the slope acceleration of adjacent exploration points. The interval where the maximum value in the second-order difference sequence is located is the location of the slope change point. The exploration point corresponding to the midpoint of the interval is taken as the slope change point. The exploration point corresponding to the slope change point is the number of selected clusters. The calculation formula for the second-order difference of the slope is: (3);

[0022] S3.3. Clustering calculation using K-means algorithm:

[0023] S3.3.1. After determining the number of clusters k, randomly select the same number of data points as the number of clusters and use them as the centroids of the clusters;

[0024] S3.3.2. Traverse the feature set, calculate the distance between each data point in the feature set and the selected centroid, and classify the centroids closest to the individual data points into the same category, thereby forming preliminary k clusters; the Euclidean distance calculation formula between each data point and the centroid is:

[0025] (4);

[0026] In the formula, is the feature set data point, is the centroid, is the dimension of the data point, They are and The qth attribute value of ;

[0027] S3.3.3. Calculate the most central data point in each category and make it the new centroid;

[0028] S3.3.4, repeat S3.3.2 and S3.3.3, and continuously redistribute the clusters to achieve a more accurate selection of the centroid, and stop the algorithm when the displacement generated when the centroid is reselected reaches the set displacement threshold to obtain the final clustering result.

[0029] Furthermore, in S4, the specific process of using the cooperative training algorithm is: selecting the number of iterations according to the number of iterations and the change of clustering evaluation indicators as the number of iterations increases; wherein the clustering evaluation indicators include silhouette index, variance ratio criterion and DBI.

[0030] Furthermore, in S5, the process of using the machine learning algorithm to learn and train the stratum categories within the excavated range is:

[0031] S5.1. According to the set ratio, the feature combination corresponding to the shield parameter data set of the excavated ring number and the clustering stratum classification result are merged into an input feature set, and divided into a training set and a test set, which are named shield parameter training set and shield parameter test set respectively;

[0032] S5.2. Constrain data of different magnitudes to the same magnitude or interval, that is, standardize the data in the training set and test set. The standardized feature scaling formula is as follows: (5);

[0033] In the formula, and are the mean and standard deviation of the shield parameter training set that need to be standardized, and are the data points of the input feature set before and after normalization;

[0034] S5.3. Establish a logistic regression model and optimize the hyperparameters through the grid search method. The hypothesis function of the logistic regression model uses the Sigmiod function: (6); (7);

[0035] In the formula, is the standardized shield parameter training set, for Elements of

[0036] The cost function of the logistic regression model is: (8); (9);

[0037] (10);

[0038] When the actual y =1, and When , the cost is zero; y = 1, and When the cost increases Become smaller and bigger; when the actual y = 0, and When , the cost is zero, when y = 0, and When the cost increases, it increases;

[0039] The gradient descent method is used to find the minimum cost function, that is: (11);

[0040] For overfitting problems, retain all features and reduce the parameter size through regularization;

[0041] The process of optimizing hyperparameters is:

[0042] By adding a regular term to the logistic regression model, the logistic regression cost function is updated to: (12);

[0043] In the formula, is the i-th data point of the standardized shield parameter training set, m is the number of data points of the standardized shield parameter training set, is the model coefficient The hyperparameter C of the logistic regression model is the regularization term in formula (12). The reciprocal of

[0044] List a set of hyperparameter values, use the grid search method to traverse all of them, train on the training set, and compare them according to the results of the logistic regression cost function in the above formula to find the best parameters and optimize the hyperparameters of the logistic regression model;

[0045] S5.4. The logistic regression model is trained using the standardized training set and clustered stratigraphic classification results, and applied to the test set to verify the evaluation indicators of the logistic regression model; the evaluation indicators include accuracy and recall rate; the accuracy rate is the ratio of the number of correct stratigraphic categories predicted by the logistic regression model to the total number of stratigraphic categories to evaluate the stratigraphic identification effect; the recall rate is the ratio of the correct number predicted for each stratigraphic category to the total number of that stratigraphic category, reflecting the stratigraphic identification effect of the logistic regression model.

[0046] The accuracy and recall rate are used for evaluation. While ensuring a high accuracy rate, it can also ensure that when there are a large number of individual categories, the high accuracy rate will not mask the low accuracy rate of other categories with a smaller number.

[0047] Furthermore, the clustering evaluation index includes silhouette index, variance ratio criterion and DBI; wherein the silhouette index is used to study the degree of separation of each category after clustering, and its principle is to compare the degree of dissimilarity between each data point and other data points in its category and the degree of dissimilarity between data points in other categories. The calculation formulas of silhouette coefficients for a single data point and the whole are: (13); (14);

[0048] In the formula, are the average value of the dissimilarity between a point in the shield parameter and other data points in the shield parameter dataset in the corresponding clustered stratum, and the minimum value of the average value of the dissimilarity between this point and data points in other stratum categories. The number of data points after adding the feature set to the stratum clustering result; the variance ratio criterion is the ratio of the distance between the cluster centers of each category after clustering to the distance between each data point within a category, and its calculation formula is: (15);

[0049] In the formula, The number of data points after adding the feature set to the stratum clustering result, For the The number of samples in a cluster, k is the number of clusters corresponding to the stratigraphic clustering, is the Euclidean distance between the cluster center of the stratum category and the overall centroid, The Euclidean distance between the shield parameter data point and the cluster center of the formation category to which it belongs, For the All data points in the clusters;

[0050] DBI is used to study the average value of the similarity between each category and its most similar category after clustering. The principle is to calculate the ratio of the distance from each point within the category to the cluster center to the distance between the centers of each category cluster. The DBI calculation formula is: (16);

[0051] In the formula, is the average distance from the shield parameter data point in a certain stratum category to the cluster center of this category, For the The distance between the centers of the first and Lth stratum category clusters.

[0052] The present invention collects shield tunneling parameters, synchronous grouting parameters and segment splicing parameters of the current shield machine during the tunneling process, and identifies and eliminates abnormal values ​​and extreme values ​​in the parameters; obtains all shield parameters whose correlation coefficients with the basic combination formed by the total thrust and the cutter head torque reach above the critical value at a set confidence level through Pearson correlation coefficient calculation, and selects common parameters to obtain feature combinations; obtains stratum classification results clustered with the basic combination and the feature combination as feature sets by using the K-means algorithm; adopts a cooperative training algorithm to add the stratum classification results obtained in S3 to each other to form two new feature sets, cluster the two new feature sets, and add the new stratum classification results to each other's feature set again for clustering, and iterates continuously in this way, and obtains the classification results of the excavated strata based on the basic combination and the feature combination under semi-supervised learning, as well as the basic combination parameter range corresponding to each stratum; uses a machine learning algorithm to learn and train the stratum categories within the excavated range, identifies the stratum conditions of unknown excavation directions, and obtains the reference value range of the total thrust and the cutter head torque. The present invention can be executed cyclically during the shield tunneling process, and the classification results of the excavated strata are continuously updated. It has the characteristics of practical feasibility and real-time performance, and provides important auxiliary decision-making information for shield tunneling. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flowchart of the present invention;

[0054] Figure 2 This is a graph showing the clustering effect changes during the iteration process in an embodiment of the present invention;

[0055] Figure 3 is a graph showing changes in clustering indicators during an iterative process in an embodiment of the present invention;

[0056] Figure 4is a graph showing changes in the evaluation index of the logistic regression model according to the number of iterations in an embodiment of the present invention;

[0057] Figure 5 It is a schematic diagram of the system structure of the present invention;

[0058] Figure 6 It is a schematic diagram of the terminal structure of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings.

[0060] like Figure 1 As shown, an intelligent auxiliary method for shield tunneling based on stratum type includes the following steps:

[0061] S1. Collect the shield tunneling parameters and synchronous grouting parameters of the current shield machine during the tunneling process, including the total thrust and cutter head torque in the shield tunneling parameters, the shield tunneling parameters excluding the total thrust and cutter head torque, and the shield correlation research parameters composed of the synchronous grouting parameters, and identify and eliminate the abnormal values ​​and extreme values ​​in the parameters;

[0062] S2. By calculating the Pearson correlation coefficient, respectively calculate the correlation coefficients between the total thrust and each shield correlation research parameter, and between the cutterhead torque and each shield correlation research parameter, respectively obtain the shield correlation research parameters above the set confidence level critical value, and select and combine the common parameters of the shield correlation research parameters with the total thrust correlation coefficient above the set confidence level critical value and the shield correlation research parameters with the cutterhead torque correlation coefficient above the set confidence level critical value to obtain a characteristic combination;

[0063] S3, using the K-means algorithm to calculate and obtain the stratum classification results clustered with the basic combination and the characteristic combination as the feature set;

[0064] S4, using a cooperative training algorithm, adding the stratum classification results obtained in S3 to each other to form two new feature sets, and then clustering the two new feature sets respectively, and adding the new stratum classification results to the other feature set again for clustering, and so on, continuously iterating, and obtaining the classification results of the excavated strata based on the basic combination and the feature combination under semi-supervised learning, as well as the basic combination parameter range corresponding to each stratum;

[0065] S5. Use machine learning algorithms to train the stratum categories within the excavation range, identify the stratum conditions with unknown excavation directions, and obtain the reference value range of the total thrust and cutterhead torque.

[0066] Embodiment: This embodiment uses various parameters of the shield tunnel of 889 rings of a shield tunnel interval and selects the shield tunnel to advance to 420 rings, that is, the method proposed by the present invention cycles to 420 rings. The specific steps are as follows:

[0067] (1) After data processing, the distribution of various parameters of the tunnel project when the shield tunneling reached the 420th ring is shown in Table 1;

[0068] Note: The pressures on the upper, lower, left and right parts of the propulsion cylinder are named D, B, C and A respectively;

[0069] (2) After the data processing is completed, the parameter data are subjected to Pearson correlation analysis and sorting. The shield correlation research parameters with correlation degrees with the total thrust and the cutter head torque above the critical value of the significant level of 0.01 are shown in Table 2. According to Table 2, the characteristic combination of the present embodiment is selected as the left side pressure of the propulsion cylinder and the excavation speed;

[0070] Note: According to the correlation coefficient test table, the critical value of the correlation coefficient in the embodiment is 0.086 at the 0.01 significance level;

[0071] (3) Using the K-means algorithm and cooperative training algorithm, the number of clusters is 5, and the formation classification results are obtained by clustering the total thrust and cutter torque and the pressure on the left side of the propulsion cylinder and the excavation speed as the feature set. The two formation classifications are added to each other to form a new feature set. The two are clustered again, and the new formation classification results are added to the other feature set. This process is repeated. The clustering effect and the changes in clustering indicators during the iteration process are shown in the figure. Figure 2 and Figure 3 As shown;

[0072] (4) The number of iterations was selected as 200. At this time, the silhouette coefficients of the basic combination and the characteristic combination were 0.203 and 0.293, respectively, and the DBIs were 1.672 and 1.305, respectively. Under semi-supervised learning, the stratum classification results within the current excavation range based on the pressure on the left side of the propulsion cylinder and the excavation speed, as well as the parameter ranges of the total thrust and cutterhead torque corresponding to each stratum were obtained, as shown in Table 3:

[0073] (5) After 200 iterations, the data set consisting of the left side pressure of the propulsion cylinder, the tunneling speed, and the clustering formation classification results was divided into a training set and a test set in a ratio of 7:3. The training set was standardized, and the features of the test set were scaled to complete the training and testing of the model. The accuracy and recall rates of the logistic regression training and testing were 78% and 78%, respectively. The specific evaluation indicators of the logistic regression model are shown in Table 4. The evaluation indicators of the logistic regression model change with the number of iterations as follows: Figure 4 As shown;

[0074] (6) The logistic regression model that has been optimized and trained is used to predict the formation category of the subsequent ring number 421, and the reference value range of the total thrust and cutter torque is given.

[0075] From the above embodiments, it can be seen that the method provided by the present invention clusters the strata by cooperative training with the basic combination of total thrust and cutter head torque through the feature combination composed of parameters that meet the requirements of the correlation with the total thrust and cutter head torque. After the selected number of cooperative training iterations, the strata categories clustered by the feature combination also have a relatively clear distinction and range for the total thrust and cutter head torque, and then the feature combination and clustering results are used to complete the training and testing of the logistic regression model, thereby realizing strata identification and prediction of the total thrust and cutter head torque range, avoiding the problems of the existing main machine-assisted shield operation methods, and having more application value for actual engineering.

[0076] like Figure 5 As shown, the present invention also provides an intelligent auxiliary system for shield tunneling based on stratum category, including a measured data acquisition module, a feature combination calculation module, a stratum clustering classification module and an intelligent auxiliary decision-making module; the measured data acquisition module is used to collect shield tunneling parameters including synchronous grouting parameters, and quickly identify and eliminate abnormal values ​​and extreme values ​​in the data; the feature combination calculation module is used to use Pearson correlation to analyze the gold mining parameters, and sort the Pearson correlation coefficients of each parameter from high to low, and select the gold mining parameter feature combination; the stratum clustering classification module is used to cluster the strata using the K-means algorithm, and adopt the cooperative training algorithm idea to realize the classification of the excavated strata categories; the intelligent auxiliary decision-making module is used to train the machine learning model according to the stratum classification results, and apply it to the intelligent assisted driving in the shield tunneling process.

[0077] like Figure 6As shown, the present invention also provides a terminal, including a memory, a processor, a display interface and an auxiliary tunneling, wherein the memory may be an external storage device of the terminal, or may be used to store application software and isolated data installed on the terminal. In this embodiment, the memory stores an intelligent auxiliary program for shield tunneling based on stratum categories, and the intelligent auxiliary program for shield tunneling based on stratum categories may be executed by the processor, thereby realizing the above-mentioned intelligent auxiliary method for shield tunneling based on stratum categories.

Claims

1. An intelligent auxiliary method for shield tunneling based on stratum classification, characterized in that: The steps include: S1. Collect the shield tunneling parameters and synchronous grouting parameters of the current shield machine during the tunneling process, including the total thrust and cutter head torque in the shield tunneling parameters, the shield tunneling parameters excluding the total thrust and cutter head torque, and the shield correlation research parameters composed of the synchronous grouting parameters, and identify and eliminate the abnormal values ​​and extreme values ​​in the parameters; S2. By calculating the Pearson correlation coefficient, respectively calculate the correlation coefficients between the total thrust and each shield correlation research parameter, and between the cutterhead torque and each shield correlation research parameter, respectively obtain the shield correlation research parameters above the set confidence level critical value, and select and combine the common parameters of the shield correlation research parameters with the total thrust correlation coefficient above the set confidence level critical value and the shield correlation research parameters with the cutterhead torque correlation coefficient above the set confidence level critical value to obtain a characteristic combination; S3. Using the K-means algorithm, the formation classification results are obtained by clustering the basic combination consisting of the total thrust and the cutter head torque and the characteristic combination as the characteristic set; the specific process is as follows: S3.1, selecting the number of clusters corresponding to the positions of the slope change points through the elbow diagram, that is, determining the number of clusters for classifying the strata by selecting the number of clusters corresponding to the slope change points on the elbow diagram; S3.

2. Use the second-order difference method of regression coefficient to determine the slope change point of the elbow graph: S3.2.

1. The number of abscissa clusters is used to form an exploration point sequence. Starting from the exploration point, linear regression is performed on the data points including the exploration point and before and after the exploration point to obtain the regression coefficient. S3.2.

2. As the sliding window slides backward, the regression coefficients corresponding to each exploration point are obtained, and the first-order difference of the slope is obtained by making a difference between the two, which reflects the change in the slope at the exploration point; the calculation formula of the first-order difference of the slope is: ； In the formula, For exploration points, is the linear regression coefficient of the sliding window where the exploration point is located; S3.2.

3. Perform a second-order difference on the first-order difference sequence of the slope to reflect the magnitude of the change in the slope acceleration of adjacent exploration points. The interval where the maximum value in the second-order difference sequence is located is the location of the slope change point. The exploration point corresponding to the midpoint of the interval is taken as the slope change point. The exploration point corresponding to the slope change point is the number of selected clusters. The calculation formula for the second-order difference of the slope is: ； S3.

3. Use K-means algorithm for clustering calculation: S3.3.

1. After determining the number of clusters k, randomly select the same number of data points as the number of clusters and use them as the centroids of the clusters; S3.3.

2. Traverse the feature set, calculate the distance between each data point in the feature set and the selected centroid, and classify the centroids closest to the individual data points into the same category, thereby forming preliminary k clusters; the Euclidean distance calculation formula between each data point and the centroid is: ； In the formula, is the feature set data point, is the centroid, is the dimension of the data point, They are No. attribute value; S3.3.

3. Calculate the most central data point in each category and make it the new centroid; S3.3.4, repeat S3.3.2 and S3.3.3, and continuously redistribute the clusters; stop the algorithm when the displacement caused by the reselection of the centroid reaches the set displacement threshold, and obtain the final clustering result; S4, using a cooperative training algorithm, adding the stratum classification results obtained in S3 to each other to form two new feature sets, and then clustering the two new feature sets respectively, and adding the new stratum classification results to the other feature set again for clustering, and so on, continuously iterating, and obtaining the classification results of the excavated strata based on the basic combination and the feature combination under semi-supervised learning, as well as the basic combination parameter range corresponding to each stratum; S5. Use machine learning algorithms to train the stratum categories within the excavation range, identify the stratum conditions with unknown excavation directions, and obtain the reference value range of the total thrust and cutterhead torque.

2. The intelligent shield tunneling assistance method based on stratum classification according to claim 1 is characterized in that: In S1, the shield tunneling parameters of the shield machine during the tunneling process include the upper pressure, lower pressure, left pressure, right pressure, propulsion speed, upper soil pressure, cutter head torque, and excavation volume of the propulsion cylinder obtained by sensors installed on the shield machine, and the synchronous grouting parameters include grouting pressure and grouting volume; the abnormal values ​​and extreme values ​​in the parameter data are identified and eliminated by using a box plot.

3. The intelligent shield tunneling assistance method based on stratum classification according to claim 1 is characterized in that: In S2, the Pearson correlation coefficient calculation formula is as follows: ； In the formula, They are the sample points, sample mean values, and sample standard deviations of the total thrust or cutter head torque parameters in the shield tunneling parameters. are the sample points, sample mean values, and sample standard deviations of any parameter in the shield correlation research parameters consisting of the shield tunneling parameters excluding the total thrust and cutterhead torque and the synchronous grouting parameters, respectively. is the number of shield parameter data points after data processing; According to the above formula, the Pearson correlation coefficient of each parameter in the total thrust and shield correlation research parameters, and each parameter in the cutterhead torque and shield correlation research parameters are calculated one by one, and the obtained Pearson correlation coefficients are sorted from high to low, and the shield correlation research parameters whose correlation coefficients are above the set confidence level critical value are taken out respectively, and the common parameters of the shield correlation research parameters whose correlation coefficients with the total thrust are above the set confidence level critical value and the shield correlation research parameters whose correlation coefficients with the cutterhead torque are above the set confidence level critical value are selected and combined to form a characteristic combination; among which, the confidence level is 0.1, 0.05 or 0.01, and the correlation coefficient test critical value is obtained according to the correlation coefficient test table.

4. The intelligent shield tunneling assistance method based on stratum classification according to claim 1 is characterized in that: In S4, the specific process of using the cooperative training algorithm is: selecting the number of iterations according to the number of iterations and the change of clustering evaluation indicators as the number of iterations increases; wherein the clustering evaluation indicators include silhouette index, variance ratio criterion and DBI.

5. The intelligent shield tunneling auxiliary method based on stratum classification according to claim 1 is characterized in that: In S5, the process of using the machine learning algorithm to learn and train the stratum categories within the excavated range is as follows: S5.

1. According to the set ratio, the feature combination corresponding to the shield parameter data set of the excavated ring number and the clustering stratum classification result are merged into an input feature set, and divided into a training set and a test set, which are named shield parameter training set and shield parameter test set respectively; S5.

2. Constrain data of different magnitudes to the same magnitude or interval, that is, standardize the data in the training set and test set. The standardized feature scaling formula is as follows: ； In the formula, are the mean and standard deviation of the shield parameter training set that need to be standardized, are the data points of the input feature set before and after normalization; S5.

3. Establish a logistic regression model and optimize the hyperparameters through the grid search method. The hypothesis function of the logistic regression model uses the Sigmiod function: ; ; In the formula, is the standardized shield parameter training set, The cost function of the logistic regression model is: ； ； ； When the actual y =1, and When , the cost is zero; y = 1, and When the cost increases Become smaller and bigger; when the actual y = 0, and When , the cost is zero, when y = 0, and When the cost increases Go big and get bigger; The gradient descent method is used to find the minimum cost function, that is: ; The process of optimizing hyperparameters is: By adding a regular term to the logistic regression model, the logistic regression cost function is updated to: ; In the formula, is the i-th data point of the standardized shield parameter training set, is the number of data points in the training set of the standardized structural parameters, is the model coefficient The hyperparameter C of the logistic regression model is the number of regular terms. The inverse of ; List a set of hyperparameter values, use the grid search method to traverse all the hyperparameter values, train on the training set, and compare them according to the results of the logistic regression cost function in the above formula to find the best parameters and optimize the hyperparameters of the logistic regression model; S5.

4. The logistic regression model is trained using the standardized training set and clustered stratigraphic classification results, and applied to the test set to verify the evaluation indicators of the logistic regression model; the evaluation indicators include accuracy and recall rate; the accuracy rate is the ratio of the number of correct stratigraphic categories predicted by the logistic regression model to the total number of stratigraphic categories to evaluate the stratigraphic identification effect; the recall rate is the ratio of the correct number predicted for each stratigraphic category to the total number of that stratigraphic category, reflecting the stratigraphic identification effect of the logistic regression model.

6. The intelligent shield tunneling assistance method based on stratum type according to claim 4 is characterized in that: The clustering evaluation index includes silhouette index, variance ratio criterion and DBI; wherein the silhouette index compares the dissimilarity of each data point with other data points in its category and the dissimilarity of data points in other categories, and the calculation formulas of the silhouette coefficient of a single data point and the whole are: ; ; In the formula, are the average value of the dissimilarity between a point in the shield parameter and other data points in the shield parameter dataset in the corresponding clustered stratum, and the minimum value of the average value of the dissimilarity between this point and data points in other stratum categories. The number of data points after adding the feature set to the stratum clustering result; the variance ratio criterion is the ratio of the distance between the cluster centers of each category after clustering to the distance between each data point within a category, and its calculation formula is: ; In the formula, The number of data points after adding the feature set to the stratum clustering result, For the The number of samples in a cluster, is the number of clusters corresponding to the stratum clustering, is the Euclidean distance between the cluster center of the stratum category and the overall centroid, is the Euclidean distance between the shield parameter data point and the cluster center of the stratum category to which it belongs, For the All data points in the clusters; DBI is the ratio of the distance from each point within the category to the cluster center to the distance between the centers of each category cluster after clustering. The calculation formula is: ; In the formula, is the average distance from the shield parameter data point in a certain stratum category to the cluster center of this category, For the and The distance between the centers of the stratigraphic class clusters.

Images