GBDT-based shield construction tunnel face geologic feature identification method and system

Abstract

The invention provides a GBDT-based shield construction tunnel face geologic feature recognition method and system. The method comprises the steps of collecting relevant data of shield tunneling machine tunneling parameters and geologic feature types; according to the related data, a Pearson's correlation coefficient method is adopted to calculate correlation coefficients of the tunneling parameters and the geologic feature types; according to the related data and the correlation coefficient, based on a GBDT algorithm, establishing a geologic feature recognition model, and evaluating the performance of the geologic feature recognition model through an evaluation index; and obtaining unidentified shield tunneling machine tunneling parameters, inputting the unidentified shield tunneling machine tunneling parameters into the geological feature recognition model, and outputting a determined geological feature type. According to the method, the training model is constructed in combination with the Pearson correlation coefficient and the GBDT algorithm, and the identification accuracy is evaluated in combination with the evaluation indexes, so that the real-time geologic features of the tunnel face in the tunneling process are identified, and reference is provided for shield tunneling machine operators.

Description

Technical Field

[0001] The present application relates to the field of underground and tunnel engineering technology, and in particular, to a method and system for identifying geological features of a shield construction face based on GBDT. Background Art

[0002] With the development of urban modernization, the urbanization process is becoming more and more rapid, and the flow of people between cities is quite amazing. Due to the frequent travel of a large number of private cars, buses, taxis and other vehicles, the ground transportation system is paralyzed. The inefficient allocation and utilization of road resources and the structural defects of urban roads are difficult to solve quickly. The development of urban rail transit is the preferred solution for urban managers to solve the problem of traffic efficiency. Urban rail transit and underground rail transit have the advantages of large capacity, fast speed, development of underground space, and small construction impact. Economically developed regions such as Beijing-Tianjin-Hebei, Yangtze River Delta and Pearl River Delta have basically achieved a one-hour living circle through rail transit. In the process of underground space development, underground transportation construction dominated by subways can help alleviate the current trend of unbearable ground traffic congestion and tight land resources in large cities. Shield construction is widely used in underground space construction such as tunnels and subways due to its good safety, high construction efficiency and low environmental pollution. However, various complex geological problems are often encountered during shield excavation. Pre-estimation of shield crossing different geology can reduce disasters caused by improper shield construction operations. Therefore, predicting the geological conditions that the shield will pass through in real time in advance can provide a reasonable reference for the shield construction operation and reduce the occurrence of disasters caused by improper operation.

[0003] After searching the existing literature, it was found that the patent with patent publication number CN114492038A disclosed "a method and system for identifying geological features of shield tunneling based on fuzzy classification algorithm". It adopts an unsupervised and learning method, that is, based on the on-site shield construction tunneling parameters, a correlation analysis with the geological feature category is established by calculating the parameters and the construction parameters, thereby establishing a model for geological feature classification based on construction parameters. It is unable to identify the geological feature category in the subsequent tunneling process.

[0004] The patent with the prior art publication number CN202011641363 discloses "a method and system for geological feature detection and identification based on deep learning". This solution uses deep learning methods to classify geological feature information and generate geological feature identification information, which can perform detailed analysis of geological features and provide a basis for geological situation analysis. However, this method performs feature recognition on geological structure data, stratigraphic data, and oil and gas resource data to provide data information for oil and gas exploration. In addition, this method determines the category labels of geological information by artificial means, and human subjectivity has a great influence on the classification results of geological features, resulting in a large deviation from the actual project. In addition, compared with machine learning, deep learning has a deeper structural network and stronger feature extraction capabilities. For models based on shield parameters as input parameters, deep learning has the meaning of "overkill". Therefore, this method is not suitable for shield tunneling construction. Summary of the invention

[0005] In view of the defects in the prior art, the purpose of this application is to provide a method for identifying the geological feature categories of the tunnel face of shield construction based on GBDT. Compared with the deep learning method, this method prevents overfitting in the calculation process of the geological identification model, and the classification of geological features is based on the slag discharged by the screw machine in the shield machine during on-site shield construction and the classification obtained by direct observation of the tunnel face during the tool change process, thereby reducing the error of human classification.

[0006] The first aspect of the present application provides a method for identifying geological features of a shield construction face based on GBDT, comprising:

[0007] Collect data related to shield machine excavation parameters and geological feature types;

[0008] According to the relevant data, the correlation coefficient between the tunneling parameter and the geological feature type is calculated using the Pearson correlation coefficient method;

[0009] According to the relevant data and the correlation coefficient, a geological feature recognition model is established based on the GBDT algorithm, and the performance of the geological feature recognition model is evaluated through evaluation indicators;

[0010] Unidentified tunneling parameters of the shield machine are obtained and input into the geological feature identification model, and the geological feature type is determined by output.

[0011] Furthermore, the relevant data include: all on-site shield machine ledgers, construction environmental reports, shield machine export record data and construction geological cross-section diagrams;

[0012] The method further includes preprocessing the relevant data, wherein the preprocessing includes: removing outliers, normalizing values, and recording the geological feature type corresponding to the ring number.

[0013] Furthermore, the removal of abnormal values ​​includes: removing data abnormalities or sensor damage caused by unexpected situations during the construction process, as well as data in the startup phase and the shutdown phase after the shield machine is started;

[0014] The normalized value is a ratio of a certain parameter value minus the minimum value of the parameter to the difference between the maximum value and the minimum value of the parameter;

[0015] The formula for the normalized value is:

[0016] In the formula, x norm is the normalized value; x min is the minimum measured value; x max is the maximum measured value; x is the original value;

[0017] The recording of the geological feature type corresponding to the ring number includes:

[0018] Obtain the stratum type in the construction ring report, and record the stratum corresponding to the ring number according to the cross-section of the tunnel face with geological characteristics:

[0019] When the cross section of the stratum face is completely weathered granite, it is recorded as a soft soil stratum;

[0020] When the cross section of the formation face is full-section hard rock, it is recorded as hard rock formation;

[0021] When the cross-section of the stratum face is uneven in hardness or soft at the top and hard at the bottom, it is recorded as an uneven stratum.

[0022] Further, according to the relevant data, the correlation coefficient between the tunneling parameter and the geological feature type is calculated using the Pearson correlation coefficient method, including:

[0023] The pre-processed excavation parameters and the geological feature types are obtained, and the stratum types are coded and converted, where a soft soil stratum is assigned a value of 1, a soft and hard uneven stratum is assigned a value of 2, and a hard rock stratum is assigned a value of 3;

[0024] Calculating the Pearson correlation coefficient between each of the excavation parameters and each of the geological feature types;

[0025] Wherein, the calculation formula of the Pearson correlation coefficient is:

[0026]

[0027] In the formula, ρ s is the Pearson coefficient, R i and S i are the values ​​of the two variables, and is the average of two variables, and is the standard deviation of the two variables.

[0028] Furthermore, the geological feature recognition model is established based on the GBDT algorithm according to the relevant data and the correlation coefficient, and the performance of the geological feature recognition model is evaluated by evaluation indicators, including:

[0029] Obtaining the preprocessed excavation parameters and the correlation coefficient;

[0030] Selecting the excavation parameter with a higher correlation as an input parameter, and dividing the input parameter into a training set and a test set according to a set ratio; wherein the training set includes a plurality of training samples;

[0031] According to the training set, based on the GBDT algorithm, a geological feature recognition model is constructed by using a CART regression tree;

[0032] According to the constructed geological feature recognition model, the performance of the geological feature model is evaluated through the test set and the set evaluation indicators.

[0033] Furthermore, the geological feature recognition model is constructed according to the training set and based on the GBDT algorithm by using a CART regression tree, including:

[0034] For the given training set, let {(x1,y1),(x2,y2),...,(x N ,y N )}, set the initial prediction value of the model to the mean of the training sample labels and initialize the first weak learner F0(x);

[0035] Construct M regression trees, for each regression tree m = 1, 2, ..., M, and each of the training samples i = 1, 2, ..., N; calculate the loss function L, and select the negative gradient r of the minimum loss function in the mth regression tree m,i , as the residual of the sample under the current regression tree;

[0036] The CART regression tree is used to regress the training samples and the corresponding negative gradient (x i ,r m,i ) is fitted, and the leaf node area of ​​the mth regression tree is recorded as R m,j ; Where j = 1, 2, ..., J, J is the number of leaf nodes of the mth regression tree;

[0037] According to the leaf region j=1,2,...,J, the best fitting value in the leaf region is calculated as a m,j ,

[0038] Update the strong learner F according to the best fitting value of the leaf node and the corresponding area m (x), by adding and combining the results of all weak learners, we get the strong learner F M (x), generating the geological feature identification model.

[0039] Furthermore, after generating the geological feature recognition model, the method further includes:

[0040] Performing regularization processing on the geological feature recognition model;

[0041] The regularization process includes: defining a step size v∈(0,1], and adjusting the iterative process of the weak learner;

[0042] The subsampling ratio is used to perform regularized pruning on the CART regression tree to generate a regularized geological feature recognition model.

[0043] The sub-sampling ratio has a value range of (0, 1]. If the value is 1, all samples are used to fit the GBDT decision tree. If the value is less than 1, some samples are extracted to fit the GBDT decision tree.

[0044] The iterative process of the weak learner becomes:

[0045]

[0046] Furthermore, the first weak learner F0(x) is initialized as:

[0047]

[0048] Where L is the loss function, N is the number of samples, y i is the actual geological characteristic value, a is the initial predicted geological characteristic value, and the mean value of a can be set to the mean value of y;

[0049] The negative gradient r m,i The calculation formula is:

[0050]

[0051] In the formula, F m-1 (x) is the geological characteristic function value predicted for the m-1th time;

[0052] The calculation formula of the best fit value is:

[0053]

[0054] In the formula, a m,jis the residual value of the mth regression tree fitting;

[0055] The updated strong learner F m (x) is:

[0056]

[0057] In the formula, F m (x) is the geological characteristic function value predicted for the mth time, R m,j It refers to the set of data points corresponding to the leaf nodes obtained by dividing the feature space during the training of the mth regression tree, which means the sample data of the mth regression tree;

[0058] The strong learner F M (x) is expressed as:

[0059]

[0060] Furthermore, the geological feature recognition model constructed is used to evaluate the performance of the geological feature model through the test set and the set evaluation index, including:

[0061] The evaluation indicators set include: confusion matrix, precision, recall and f1_score;

[0062] Acquire the data of the test set, and obtain prediction results through the geological feature recognition model, wherein the prediction results include positive examples and negative examples;

[0063] According to the prediction results, calculate the model's precision, recall and f1_score;

[0064] The precision rate represents the percentage of positive examples in the test set that are actually positive examples, and is determined by the following formula:

[0065]

[0066] In the formula, TP refers to the number of samples where positive examples are identified as positive examples; FP refers to the number of samples where negative examples are identified as positive examples;

[0067] The recall rate represents the proportion of samples in the test set that are identified as positive examples among all actual positive examples, and is determined by the following formula:

[0068]

[0069] Among them: TP refers to the number of samples that are identified as positive examples; FN refers to the number of samples that are identified as negative examples.

[0070] The f1_score is the harmonic mean based on the precision and the recall, and is determined by the following formula:

[0071]

[0072] The second aspect of the present application provides a shield construction face geological feature recognition system based on GBDT, comprising:

[0073] The data collection module is used to collect all on-site shield machine records, construction environmental reports, shield machine export record data and construction geological cross-section diagrams;

[0074] A correlation calculation module, used for calculating the correlation coefficient between the tunneling parameter and the geological feature type by using the Pearson correlation coefficient method according to the relevant data;

[0075] A model training module, which establishes a geological feature recognition model based on the GBDT algorithm according to the relevant data and the correlation coefficient;

[0076] A model evaluation module is used to evaluate the performance of the geological feature recognition model constructed by using a test set and set evaluation indicators;

[0077] The comparison module is used to compare the results of predicting the test set based on the training set after model training with the geological characteristics reflected by the tunnel face section.

[0078] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0079] 1. This application combines the Pearson correlation coefficient with the GBDT algorithm to extract key information from the tunneling parameters of the shield machine, accurately identify the geological feature types of the shield construction face, improve the accuracy and efficiency of geological feature identification, reduce the subjectivity and errors of human judgment, and provide a scientific basis for geological risk prediction and tunneling strategy adjustment during shield construction, which helps to ensure construction safety and improve engineering efficiency. At the same time, the set evaluation indicators are used to evaluate the accuracy of its identification, thereby realizing the real-time identification of the geological features of the face during the tunneling process and providing a reference for shield machine operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0081] Figure 1 This is a flow chart of a method for identifying geological features of a shield construction face based on GBDT in one embodiment of the present application.

[0082] Figure 2 This is the GBDT model process in one embodiment of the present application.

[0083] Figure 3 This is a training flow chart in one embodiment of the present application.

[0084] Figure 4 This is a graph comparing the model prediction and the true label in one embodiment of the present application.

[0085] Figure 5 This is a structural diagram of a shield construction face geological feature recognition system based on GBDT in one embodiment of the present application. DETAILED DESCRIPTION

[0086] The present application is described in detail below in conjunction with specific implementation cases. The following implementation cases will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application. These all fall within the scope of protection of the present application.

[0087] Reference Figure 1 As shown, it is an implementation case of the present application: a method for identifying geological features of a shield construction face based on GBDT, comprising: S1, collecting relevant data on shield machine excavation parameters and geological feature types; S2, according to the relevant data, using the Pearson correlation coefficient method to calculate the correlation coefficient between the excavation parameters and the geological feature type; S3, based on the relevant data and the correlation coefficient, a geological feature identification model is established based on the GBDT algorithm, and the performance of the geological feature identification model is evaluated through evaluation indicators; S4, obtaining unidentified shield machine excavation parameters, inputting them into the geological feature identification model, and outputting the determined geological feature type.

[0088] This application combines the Pearson correlation coefficient with the GBDT algorithm to extract key information from the tunneling parameters of the shield machine, accurately identify the geological feature types of the shield construction face, improve the accuracy and efficiency of geological feature identification, reduce the subjectivity and error of human judgment, and provide a scientific basis for geological risk prediction and tunneling strategy adjustment during shield construction, which helps to ensure construction safety and improve engineering efficiency. At the same time, the set evaluation index is used to evaluate the accuracy of its identification, so as to realize the identification of real-time geological features of the face during the tunneling process, and provide a reference for shield machine operators.

[0089] Specifically, firstly, various parameters of the shield machine excavation process and the geological feature type data associated with the parameters are systematically collected; then, the Pearson correlation coefficient method is used to calculate the correlation between the excavation parameters and the geological features, and the strength of the association between them is quantified; based on the preprocessed data and these correlation coefficients, the gradient boosting decision tree (GBDT) algorithm is used to construct a geological feature recognition model; the model is trained through the training set, and the evaluation indicators (such as accuracy and recall rate) are used to test the performance of the model to achieve effective identification of different geological features; finally, the newly collected and unidentified shield machine excavation parameters are input into the model, and the model will output the corresponding geological feature type to achieve automatic recognition of geological features.

[0090] In some specific embodiments, the relevant data include: all on-site shield machine records, construction ring reports, shield machine export record data and construction geological cross-section maps; the method also includes preprocessing the relevant data, wherein the preprocessing includes: removing outliers, normalizing values ​​and recording the geological feature types corresponding to the ring numbers.

[0091] Specifically, removing outliers includes: eliminating data anomalies or sensor damage caused by emergencies during the construction process, as well as data from the startup phase and shutdown phase after the shield machine is turned on; the normalized value is the ratio of a parameter value minus the minimum value of the parameter to the difference between the maximum and minimum values ​​of the parameter.

[0092] By removing outliers, data caused by abnormal conditions such as emergencies, sensor damage, or shield machine startup / shutdown can be eliminated, thereby improving data quality, avoiding the impact of outliers on model training, reducing the model's sensitivity to outliers, and enhancing the model's robustness and stability; normalization processing converts the data to the same scale, so that different parameters can be directly compared and analyzed, which helps to eliminate interference caused by data differences.

[0093] Specifically, the formula for the normalized value is:

[0094]

[0095] In the formula, x norm is the normalized value; x min is the minimum measured value; x max is the maximum measured value; x is the original value.

[0096] Recording the strata corresponding to the changed numbers in the geological feature type includes: obtaining the strata type in the construction environmental report, and recording the strata corresponding to the ring number according to the cross-section of the face of the geological feature: when the cross-section of the stratum face is completely weathered granite, it is recorded as a soft soil stratum; when the cross-section of the stratum face is full-section hard rock, it is recorded as a hard rock stratum; when the cross-section of the stratum face is unevenly soft and hard or soft on top and hard on the bottom, it is recorded as an unevenly soft and hard stratum.

[0097] In some specific embodiments, based on relevant data, the Pearson correlation coefficient method is used to calculate the correlation coefficient between the excavation parameters and the geological feature type, including: obtaining the pre-processed excavation parameters and geological feature types, and encoding the formation type, assigning a value of 1 to the soft soil formation, a value of 2 to the uneven soft and hard formation, and a value of 3 to the hard rock formation; calculating the Pearson correlation coefficient between each excavation parameter and each geological feature type.

[0098] By calculating the Pearson correlation coefficient, the strength and direction of the linear relationship between the excavation parameters and the geological feature types are quantified, which facilitates a more accurate understanding of the relationship between the excavation parameters and the geological features during the model training process; identifying which excavation parameters have the greatest impact on the geological feature types, reducing unnecessary adjustments and optimization processes, and establishing a more accurate prediction model. At the same time, the stability and safety of the excavation process are improved. For example, if a certain excavation parameter has a strong correlation with a specific geological feature type, then when encountering similar geological conditions, the excavation parameter can be adjusted to optimize the excavation effect and improve construction efficiency and quality.

[0099] The calculation formula of Pearson correlation coefficient is:

[0100]

[0101] In the formula, ρ s is the Pearson coefficient, R i and S i are the values ​​of the two variables, and is the average of two variables, and is the standard deviation of the two variables.

[0102] In this application, the geological feature types are coded and assigned values, with soft soil layers assigned a value of 1, uneven soft and hard layers assigned a value of 2, and hard rock layers assigned a value of 3. The geological feature type labels are converted into numerical values ​​and the Pearson correlation coefficient is calculated to obtain the correlation between the excavation parameters and the layers.

[0103] In some specific embodiments, a geological feature recognition model is established based on the GBDT algorithm according to relevant data and correlation coefficients, and the performance of the geological feature recognition model is evaluated through evaluation indicators, including: obtaining the preprocessed excavation parameters and the correlation coefficients; selecting excavation parameters with higher correlation as input parameters, and dividing the input parameters into a training set and a test set according to a set ratio; wherein the training set contains multiple training samples; based on the training set, a geological feature recognition model is constructed through a CART regression tree based on the GBDT algorithm; based on the constructed geological feature recognition model, the performance of the geological feature model is evaluated through a test set and set evaluation indicators.

[0104] Among them, the ratio is set to 3:1, dividing the training set and the test set.

[0105] During the operation, based on the GBDT algorithm, a geological feature recognition model is constructed through the CART (Classification and Regression Trees) regression tree. GBDT is an integrated learning method that gradually constructs multiple weak learners (i.e., CART regression trees) and combines their results to form a strong learner; and uses the training set to train the GBDT model. During the training process, GBDT will continuously adjust the parameters of each CART regression tree to minimize the prediction error.

[0106] The data in the test set does not participate in the training process, which is equivalent to using new data to verify the accuracy of the model. It can truly reflect the model's ability to predict new data in practical applications.

[0107] The evaluation indicator is to evaluate the accuracy of the prediction results of the test set, which can evaluate the accuracy of the test set prediction and understand the generalization ability of the model.

[0108] By selecting excavation parameters with a high correlation with the geological feature type as input parameters, the noise and redundant information of the model can be significantly reduced, and the accuracy of geological feature recognition can be improved; the model is trained and evaluated using a training set and a test set respectively, and the generalization ability of the model is enhanced during training; a strong learner is formed by constructing multiple CART regression trees through the GBDT algorithm, which helps to understand which excavation parameters are most important for geological feature recognition.

[0109] Among them, the gradient boosted decision tree (GBDT) belongs to the iterative decision tree algorithm, which uses the CART regression tree to construct each weak learner and combines these weak learners through the additive model to obtain a strong learner. In each iteration, the gradient lost by the previous weak learner of GBDT is trained by the next weak learner, and iterates gradually in the direction of decreasing loss, and the results of all trees are accumulated as the final result.

[0110] Specifically, according to the training set, based on the GBDT algorithm, a geological feature recognition model is constructed through the CART regression tree, including: for a given training set, set to {(x1,y1),(x2,y2),...,(x N ,y N )}, by setting the loss function to L, the first weak learner F0(x) is initialized.

[0111] The training set includes multiple training samples.

[0112] First, the first weak learner F0(x) is initialized as:

[0113] Where L is the loss function, N is the number of samples, y i is the actual geological characteristic value, a is the initial predicted geological characteristic value, and the mean value of a can be set to the mean value of y;

[0114] Next, construct M regression trees. For each regression tree m = 1, 2, ..., M, and each training sample i = 1, 2, ..., N, calculate the negative gradient r of the loss function L at the mth regression tree. m,i , as the residual of the sample under the current regression tree.

[0115] In the present application, a training set includes multiple training samples, and the data included in each training sample has corresponding excavation parameters and formation types.

[0116] The GBDT algorithm uses CART regression trees for prediction. In the first layer of trees, the training samples of each tree are training sets. After fitting the negative gradient, the training set is divided into two training samples for the second layer of trees. Therefore, each regression tree has a corresponding training sample, and each training sample contains the corresponding excavation parameters and stratum type.

[0117] Among them, the GBDT algorithm stipulates that the value of the negative gradient of the loss function in the current model is used as the approximate value of the residual. Therefore, the negative gradient is equivalent to the residual, which refers to the error between the current predicted value and the actual value.

[0118] Negative gradient r m,i The calculation formula is:

[0119]

[0120] In the formula, F m-1 (x) is the geological characteristic function value predicted for the m-1th time.

[0121] Then, the CART regression tree is used to train the training samples and the corresponding negative gradient (x i ,r m,i ) is fitted, and the leaf node area of ​​the mth regression tree is recorded as R m,j ; Where j = 1, 2, ..., J, J is the number of leaf nodes of the mth regression tree.

[0122] Among them, the regression tree method is used to fit the negative gradient (i.e., residual) using the training samples. The set is expressed as (x i ,r m,i );R m,j It refers to the set of data points corresponding to the leaf nodes obtained by dividing the feature space during the training process of the mth regression tree, which means the sample data of the mth regression tree.

[0123] According to the leaf region j=1,2,...,J, the best fitting value in the leaf region is calculated as a m,j .

[0124] The best fit value is the value that minimizes the difference between the predicted result and the actual result. The value selection criteria are based on model convergence or reaching the maximum number of decision trees.

[0125] The best fit value is calculated as:

[0126]

[0127] In the formula, a m,j is the residual value of the mth regression tree fitting.

[0128] Finally, the strong learner F is updated according to the best fitting value of the leaf node and the corresponding area m (x), by adding and combining the results of all weak learners, we get the strong learner F M (x), generate a geological feature recognition model.

[0129] Update the strong learner F m (x) is:

[0130]

[0131] Where F m (x) is the geological characteristic function value predicted for the mth time, R m,j It refers to the set of data points corresponding to the leaf nodes obtained by dividing the feature space during the training process of the mth regression tree, which means the sample data of the mth regression tree.

[0132] Strong learner F M (x) is expressed as:

[0133]

[0134] In some specific embodiments, after the geological feature recognition model is generated, the method further includes: performing regularization processing on the geological feature recognition model; the regularization processing includes: defining a step size v∈(0,1], adjusting the iterative process of the weak learner; and performing regularization pruning on the CART regression tree using a subsampling ratio to generate a geological feature recognition model after regularization processing.

[0135] The subsampling ratio has a value range of (0, 1]. If the value is 1, all samples are used to fit the GBDT decision tree. If the value is less than 1, some samples are extracted to fit the GBDT decision tree.

[0136] The iterative process of the weak learner becomes:

[0137]

[0138] Specifically, in order to prevent overfitting, the gradient boosting decision tree (GBDT) usually needs to be regularized. The commonly used regularization methods for GBDT are mainly the following three:

[0139] Define the step size v∈(0,1], after adding the regularization term, the previous weak learner iteration becomes:

[0140]

[0141] For the same training set learning effect, it is necessary to increase the number of iterations of the weak learner. The step size and maximum number of iterations can be used to adjust the algorithm fitting effect. The GBDT model process is as follows Figure 2 shown.

[0142] Regularized pruning of CART regression tree: use subsampling ratio, the value is (0,1]. If the value is 1, all samples are used; if the value is less than 1, some samples are extracted for GBDT decision tree fitting. When a ratio less than 1 is selected, the variance can be reduced to prevent overfitting.

[0143] In this application, the strong learner F M (x) and the geological feature models after regularization are both geological feature recognition models in the training stage; the model after parameter adjustment through the test set is the geological feature recognition model in the testing stage, and the final model is the optimal geological feature model obtained after regularization, evaluation and parameter adjustment.

[0144] In some specific embodiments, according to the constructed geological feature recognition model, a test set and set evaluation indicators are used, wherein the set evaluation indicators include: confusion matrix, precision, recall and f1_score.

[0145] The performance evaluation of the geological feature model includes: obtaining the data of the test set, obtaining the prediction results through the geological feature recognition model, and the prediction results include positive examples and negative examples; according to the prediction results, calculating the precision, recall rate and f1_score of the model.

[0146] The evaluation indicator confusion matrix is ​​also called the algorithm possibility table or error matrix. It uses a matrix form to show the visualization effect of algorithm performance and is used as a method to evaluate model performance. Its matrix form is shown in Table 1:

[0147] Table 1

[0148]

[0149] During the test process, for example, when the actual stratum is a soft soil stratum, the result obtained through the test set is a soft soil stratum, which is a positive example, and the other two predicted strata are negative examples.

[0150] Among them, the precision rate is also called the precision rate, which indicates the proportion of positive examples in the number of positive examples identified in the test set that are actually positive examples, and is determined by the following formula:

[0151]

[0152] In the formula, TP refers to the number of samples that are identified as positive examples; FP refers to the number of samples that are identified as positive examples.

[0153] The recall rate, also known as the recall rate, indicates the proportion of samples in the test set that are identified as positive examples among all actual positive examples, which is determined by the following formula:

[0154]

[0155] Among them: TP refers to the number of samples that are identified as positive examples; FN refers to the number of samples that are identified as negative examples.

[0156] In a machine learning classification model, a high precision does not mean a high recall rate. Similarly, a high recall rate does not mean a high precision rate. However, in a better machine learning classification model, both the precision and the recall rate are relatively high, so it is necessary to consider the two evaluation indicators comprehensively.

[0157] f1_score is the harmonic mean of precision and recall, determined by the following formula,

[0158]

[0159] This application uses a GBDT (Extreme Value Boosting Tree) and its learning method to identify the geological features of the tunnel face during shield construction, and uses confusion matrix, precision, recall and f1_score evaluation indicators to evaluate its recognition accuracy, thereby realizing the real-time recognition of the geological features of the tunnel face during the excavation process and providing a reference for shield machine operators.

[0160] For example: The section from Longdong Station to Dayuan Station of the Guangzhou-Foshan Ring Line of the Pearl River Delta Intercity Rail Transit: This section is all tunnel construction, and the construction methods can be divided into dark tunnel, open tunnel (DSK28+500.00~DSK28+580.00), and shield section (DSK28+580.00~DSK31+940.00). The shield tunnel passes through the piers of the Guanghe Expressway Viaduct in the section of DSK30+250.00~DSK30+320.00.

[0161] First, collect all the shield machine records, construction environmental reports and shield machine export records on site, including construction geological cross-section diagrams, and calculate the Pearson correlation coefficient between each excavation parameter and each geological feature type. As shown in Table 2 below:

[0162]

[0163] The geological feature category of each ring is determined through the GBDT geological feature recognition model. The parameters related to shield excavation and the parameters with high correlation with the geological feature category are used as input parameters. The training set and the test set are divided according to the ratio of 3:1. The accuracy of geological identification is determined by comparison based on the geological characteristics of the face section. Otherwise, the GBDT parameters are adjusted until the requirements are met.

[0164] like Figure 3 As shown, specifically, the input parameters include 10 related parameters (tunneling parameters) and 3 types of strata. The 10 related parameters include: shield thrust, cutterhead torque, tunneling speed, penetration, upper soil bin pressure, lower soil bin pressure, FPI, TPI, SE and cutterhead speed; the 3 types of strata include: soft soil, uneven soft and hard soil, and hard rock. There are 610 sample data in total, which are divided into training set and test set according to the ratio of 3:1. The training set has 458 samples and the test set has 152 samples. The training set is used for GBDT grid search method parameter adjustment; the GBDT classification and recognition model after parameter adjustment is generated; then, the GBDT classification and recognition model after parameter adjustment is trained through training connection, and the GBDT classification and recognition model after parameter adjustment is tested with the test set to realize model training.

[0165] Then, through four evaluation indicators, for the confusion matrix, the more correct indicators, the better the model effect. Similarly, for the three indicators of precision, recall rate and f1_score, the larger the value, the better the effect. The evaluation results are shown in Table 3 and Table 4.

[0166] Table 3

[0167]

[0168] Table 4

[0169]

[0170] Finally, the results of predicting the test set based on the training set after model training were compared with the geological characteristics reflected by the tunnel face section. Figure 4 shown.

[0171] like Figure 5 As shown, the second aspect of the present application provides a shield construction face geological feature identification system based on GBDT, including: a data collection module, used to collect data of all on-site shield machine ledgers, construction ring reports, shield machine export records and construction geological cross-section diagrams; a correlation calculation module, used to calculate the correlation coefficient between the excavation parameters and the geological feature type based on the relevant data using the Pearson correlation coefficient method; a model training module, based on the relevant data and the correlation coefficient, based on the GBDT algorithm, to establish a geological feature identification model; a model evaluation module, used to evaluate the performance of the geological feature model based on the constructed geological feature identification model through a test set and set evaluation indicators; a comparison module, used to compare the results of predicting the test set based on the training set after model training with the geological features reflected in the face section.

[0172] First, the data collection module comprehensively collects and preprocesses key data such as on-site shield machine records, construction environmental reports, shield machine export records, and construction geological cross-section diagrams; then, the correlation calculation module uses the Pearson correlation coefficient method to accurately calculate the correlation coefficient between the excavation parameters and the geological feature types; the model training module selects the excavation parameters with higher correlation as input, and constructs a geological feature recognition model based on the GBDT algorithm; then, the model evaluation module objectively evaluates the model performance through the test set and preset evaluation indicators; finally, the comparison module carefully compares the model prediction results with the geological features shown in the face section to verify the accuracy of the model recognition.

[0173] This application achieves accurate identification of geological features of the shield construction face by integrating multiple modules such as data collection, correlation calculation, model training, model evaluation and comparison, improves the automation and intelligence level of geological feature identification, reduces the subjectivity and uncertainty of human judgment, and significantly improves identification efficiency and accuracy.

[0174] The above describes the specific embodiments of the present application. It should be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various modifications or variations within the scope of the claims, which does not affect the substantive content of the present application. The above preferred features can be used in any combination without conflicting with each other.

Claims

1. A method for identifying geological features of shield construction face based on GBDT, characterized in that: include: Collect data related to shield machine excavation parameters and geological feature types; According to the relevant data, the correlation coefficient between the tunneling parameter and the geological feature type is calculated using the Pearson correlation coefficient method; According to the relevant data and the correlation coefficient, a geological feature recognition model is established based on the GBDT algorithm, and the performance of the geological feature recognition model is evaluated through evaluation indicators; Unidentified tunneling parameters of the shield machine are obtained and input into the geological feature identification model, and the geological feature type is determined by output.

2. According to a method for identifying geological features of a shield construction face based on GBDT according to claim 1, it is characterized in that: The relevant data include: all on-site shield machine records, construction environmental reports, shield machine export record data and construction geological cross-section diagrams; The method further includes preprocessing the relevant data, wherein the preprocessing includes: removing outliers, normalizing values, and recording the geological feature types corresponding to the ring numbers.

3. A method for identifying geological features of a shield construction face based on GBDT according to claim 2, characterized in that: The removal of abnormal values ​​includes: removing abnormal data or sensor damage caused by unexpected situations during the construction process, as well as data in the startup phase and shutdown phase after the shield machine is started; The normalized value is a ratio of a certain parameter value minus the minimum value of the parameter to the difference between the maximum value and the minimum value of the parameter; The formula for the normalized value is: In the formula, x norm is the normalized value; x min is the minimum measured value; x max is the maximum measured value; x is the original value; The recording of the geological feature type corresponding to the ring number includes: Obtain the stratum type in the construction ring report, and record the stratum corresponding to the ring number according to the cross-section of the tunnel face with geological characteristics: When the cross section of the stratum face is completely weathered granite, it is recorded as a soft soil stratum; When the cross section of the formation face is full-section hard rock, it is recorded as hard rock formation; When the cross-section of the stratum face is uneven in hardness or soft at the top and hard at the bottom, it is recorded as an uneven stratum.

4. The method for identifying geological features of a shield construction face based on GBDT according to claim 3 is characterized in that: According to the relevant data, the correlation coefficient between the tunneling parameter and the geological feature type is calculated using the Pearson correlation coefficient method, including: The pre-processed excavation parameters and the geological feature types are obtained, and the stratum types are coded and converted, where a soft soil stratum is assigned a value of 1, a soft and hard uneven stratum is assigned a value of 2, and a hard rock stratum is assigned a value of 3; Calculating the Pearson correlation coefficient between each of the excavation parameters and each of the geological feature types; Wherein, the calculation formula of the Pearson correlation coefficient is: In the formula, ρ s is the Pearson coefficient, R i and S i are the values ​​of the two variables, and is the average of two variables, and is the standard deviation of the two variables.

5. The method for identifying geological features of a shield construction face based on GBDT according to claim 4 is characterized in that: The method of establishing a geological feature recognition model based on the GBDT algorithm according to the relevant data and the correlation coefficient, and evaluating the performance of the geological feature recognition model through evaluation indicators includes: Obtaining the preprocessed excavation parameters and the correlation coefficient; Selecting the excavation parameter with a higher correlation as an input parameter, and dividing the input parameter into a training set and a test set according to a set ratio; wherein the training set includes a plurality of training samples; According to the training set, based on the GBDT algorithm, a geological feature recognition model is constructed by using a CART regression tree; According to the constructed geological feature recognition model, the performance of the geological feature model is evaluated through the test set and the set evaluation indicators.

6. A method for identifying geological features of a shield construction face based on GBDT according to claim 5, characterized in that: According to the training set, based on the GBDT algorithm, a geological feature recognition model is constructed by a CART regression tree, including: For the given training set, let {(x1,y1),(x2,y2),...,(x N ,y N )}, set the initial prediction value of the model to the mean of the training sample labels and initialize the first weak learner F0(x); Construct M regression trees, for each regression tree m = 1, 2, ..., M, and each of the training samples i = 1, 2, ..., N; calculate the loss function L, and select the negative gradient r of the minimum loss function in the mth regression tree m,i , as the residual of the sample under the current regression tree; The CART regression tree is used to regress the training samples and the corresponding negative gradient (x i ,r m,i ) is fitted, and the leaf node area of ​​the mth regression tree is recorded as R m,j ; Where j = 1, 2, ..., J, J is the number of leaf nodes of the mth regression tree; According to the leaf region j=1,2,...,J, the best fitting value in the leaf region is calculated as a m,j , Update the strong learner F according to the best fitting value of the leaf node and the corresponding area m (x), by adding and combining the results of all weak learners, we get the strong learner F M (x), generating the geological feature identification model.

7. The method for identifying geological features of a shield construction face based on GBDT according to claim 6, characterized in that: After generating the geological feature recognition model, the method further includes: Performing regularization processing on the geological feature recognition model; The regularization process includes: defining a step size v∈(0,1], and adjusting the iterative process of the weak learner; The subsampling ratio is used to perform regularized pruning on the CART regression tree to generate a regularized geological feature recognition model. The sub-sampling ratio has a value range of (0, 1]. If the value is 1, all samples are used to fit the GBDT decision tree. If the value is less than 1, some samples are extracted to fit the GBDT decision tree. The iterative process of the weak learner becomes:

8. The method for identifying geological features of a shield construction face based on GBDT according to claim 6, characterized in that: The first weak learner F0(x) is initialized as: Where L is the loss function, N is the number of samples, y i is the actual geological characteristic value, a is the initial predicted geological characteristic value, and the mean value of a can be set to the mean value of y; The negative gradient r m,i The calculation formula is: In the formula, F m-1 (x) is the geological characteristic function value predicted for the m-1th time; The calculation formula of the best fit value is: In the formula, a m,j is the residual value of the mth regression tree fitting; The updated strong learner F m (x) is: In the formula, F m (x) is the geological characteristic function value predicted for the mth time, R m,j It refers to the set of data points corresponding to the leaf nodes obtained by dividing the feature space during the training of the mth regression tree, which means the sample data of the mth regression tree; The strong learner F M (x) is expressed as:

9. The method for identifying geological features of a shield construction face based on GBDT according to claim 5, characterized in that: The step of performing performance evaluation on the geological feature recognition model constructed by using the test set and the set evaluation index comprises: The evaluation indicators set include: confusion matrix, precision, recall and f1_score; Acquire the data of the test set, and obtain prediction results through the geological feature recognition model, wherein the prediction results include positive examples and negative examples; According to the prediction results, calculate the model's precision, recall and f1_score; The precision rate represents the percentage of positive examples in the test set that are actually positive examples, and is determined by the following formula: In the formula, TP refers to the number of samples that are identified as positive examples; FP refers to the number of samples that are identified as positive examples. The recall rate represents the proportion of samples in the test set that are identified as positive examples among all actual positive examples, and is determined by the following formula: Among them: TP refers to the number of samples that are identified as positive examples; FN refers to the number of samples that are identified as negative examples. The f1_score is the harmonic mean based on the precision and the recall, and is determined by the following formula:

10. A shield construction face geological feature recognition system based on GBDT, characterized in that: include: The data collection module is used to collect all on-site shield machine records, construction environmental reports, shield machine export record data and construction geological cross-section diagrams; A correlation calculation module, used for calculating the correlation coefficient between the tunneling parameter and the geological feature type by using the Pearson correlation coefficient method according to the relevant data; A model training module, which establishes a geological feature recognition model based on the GBDT algorithm according to the relevant data and the correlation coefficient; A model evaluation module is used to evaluate the performance of the geological feature recognition model constructed by using a test set and set evaluation indicators; The comparison module is used to compare the results of predicting the test set based on the training set after model training with the geological characteristics reflected by the tunnel face section.

Images