Soft rock tunnel long-term deformation prediction method based on machine learning

Abstract

The invention belongs to the technical field of soft rock tunnel deformation prediction, and provides a soft rock tunnel long-term deformation prediction method based on machine learning, and the method comprises the steps: obtaining surrounding rock deformation, surrounding rock stress and support structure stress data in a tunnel construction process, and taking the data as a model original database; preprocessing various types of data in the original database, and establishing a transaction table after grading the various types of data according to a data change rate; an association rule is calculated and determined, and monitoring objects with high correlation with tunnel surrounding rock deformation are screened out; establishing a sample database according to the data of the class of objects; and establishing a BP-neural network prediction model for training to obtain a prediction model meeting a target value condition. According to the method, the correlation between the sample data and the target prediction value is determined, the monitoring object most related to the target value is screened out, and the data of the monitoring object is adopted as the input of the prediction model, so that the prediction rate of the model can be increased, and the prediction precision of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of soft rock tunnel deformation prediction, and in particular to a method for predicting long-term deformation of a soft rock tunnel based on machine learning. Background Art

[0002] In underground cavern construction, surrounding rock deformation is a key indicator directly related to cavern stability. Therefore, it is necessary to monitor and evaluate the surrounding rock deformation in real time during cavern construction and operation. However, if the entire process of cavern construction and operation is monitored on site, it will consume a lot of manpower and material resources, and the economic efficiency is poor. In this regard, scholars have considered using the early monitoring data to establish a mathematical model to predict the later changes of the data. Machine learning, as a discipline that can mine the correlation between data and accurately predict its later change trend, has been widely used in deformation prediction of similar underground construction projects. At present, the main machine learning methods used to predict surface settlement and surrounding rock deformation caused by underground construction include neural network and support vector machine methods. However, traditional neural network and support vector machine methods have many disadvantages in actual prediction. For example, the support vector machine method is more difficult to deal with nonlinear problems. When faced with multiple influencing factors, the neural network method often cannot identify the main factors, resulting in low calculation efficiency and low accuracy.

[0003] In this regard, in order to solve the problems existing in the prior art, technical personnel in this field have made many efforts. For example, Chinese patent application CN111412885 B proposes a method for predicting large deformation of squeezing-type surrounding rock in deep tunnels. By conducting on-site geostress tests and on-site rock mechanics tests on the stratum where the deep tunnel section of the tunnel is located, the geostress and rock mechanics parameters of the stratum are obtained, and the surrounding rock deformation prediction formula is used to predict the large deformation of the surrounding rock. However, the amount of data required to be processed is large, and the prediction efficiency is lacking. Summary of the invention

[0004] To solve the above problems, the present invention provides a long-term deformation prediction method for soft rock tunnels based on machine learning. By establishing the correlation between each sample data and the target prediction value, the monitoring objects most relevant to the target value are screened out, and the data of such monitoring objects are used as the input of the prediction model. The amount of sample data processed by the model can be significantly reduced, thereby accelerating the prediction rate of the model and improving the accuracy of the model prediction.

[0005] To achieve the above object, the present invention proposes the following technical solutions:

[0006] A method for predicting long-term deformation of soft rock tunnels based on machine learning, the method comprising the following steps:

[0007] Step S1: Conduct tests at the tunnel construction site to obtain surrounding rock deformation, surrounding rock stress and support structure stress data during tunnel construction as the original model database;

[0008] Step S2: pre-processing various types of data in the original database using a data standardization method, eliminating the dimension of the data, and grading various types of data according to the data change rate to establish a transaction table;

[0009] Step S3: According to the established transaction table, calculate and establish association rules, and screen out monitoring objects with high correlation with tunnel surrounding rock deformation;

[0010] Step S4: using the test data of the monitored object and the surrounding rock deformation test data as a training database for the prediction model;

[0011] Step S5: Establish a BP-neural network prediction model, use the training database for training, and obtain a prediction model that meets the target value conditions.

[0012] Furthermore, in step S1, when testing is carried out at the tunnel construction site, the arch settlement measuring points, convergence measuring points and surrounding rock stress measuring points are arranged on the same monitoring section, and the arch frame stress measuring points and anchor stress measuring points are arranged on the monitoring section 0.9-1.1m away from them.

[0013] Furthermore, in step S2, the specific content of preprocessing various types of data in the original database using the data standardization method is: using the maximum and minimum interval division method to process various types of data in the original database, the calculation formula is:

[0014] Furthermore, in step S2, the specific method of establishing a transaction table after grading various types of data according to the data change rate is: the data change rate is divided into level I, level II and level III according to the range of the data in [0,0.35), [0.35,0.7) and [0.7,1], which respectively represent slow change rate, medium rate and fast rate, and a data transaction table is generated according to this classification.

[0015] Furthermore, in step S3, according to the established transaction table, the specific method of calculating and establishing the association rules is: finding the frequent item sets in the database based on the FP-Growth algorithm, and then calculating the confidence and lift of each frequent item set to establish the association rules.

[0016] Furthermore, the specific method for finding frequent item sets in the database based on the FP-Growth algorithm is:

[0017] Step S3-1: Scan the entire database, count the number of occurrences of all 1-item sets, and remove items that do not meet the minimum support S requirement;

[0018] Step S3-2: Arrange all the retained frequent 1-item sets in descending order of occurrence frequency, establish a model tree with the empty set item Null as the root node, connect the frequent 1-item sets with the nodes in the model tree in descending order of occurrence frequency, and mark the number of occurrences after each item set;

[0019] Step S3-3: Search for all transaction sets prefixed with a certain item set starting from the bottom of the model tree.

[0020] Furthermore, in step S3-1, the specific method for determining the minimum support S is:

[0021] Step S3-1-1: Arrange the support numbers of each item in the transaction set in descending order, and number them 1, 2, ..., n to establish a support "support number-sequence number" sequence; where n is the number of items;

[0022] Step S3-1-2: After establishing ordered plane coordinates, a polynomial is used for curve fitting; the expression of the polynomial is: Among them, x is the sequence number; y is the support number; n is the power index of the highest order term of the polynomial to be selected;

[0023] Step S3-1-3: After solving the value when the second-order derivative of the fitting curve is 0, f(x0) is obtained, and the smallest integer not less than f(x0) is taken as the minimum support S of the frequent item selected this time.

[0024] Furthermore, in step S5, a BP-neural network prediction model is established and trained using a training database. The specific method for obtaining a prediction model that meets the target value condition is:

[0025] Step S5-1: taking a data set of monitoring quantities that are highly correlated with the target predicted quantity as the input layer X;

[0026] Step S5-2: setting the dimensions of the weight matrix W and the bias matrix b in the linear transformation, and using the Sigmoid function as the excitation function for the nonlinear transformation of the input layer;

[0027] Step S5-3: Perform linear and nonlinear transformations on the input layer X to obtain a hidden layer Z;

[0028] Step S5-4: Obtain the output layer Y after performing linear and nonlinear transformation on the hidden layer Z;

[0029] Step S5-5: Taking the measured surrounding rock deformation as the target value, by calculating the difference between the output layer Y and the target value, updating the weighted matrix and the bias matrix, and obtaining a prediction model whose output layer meets the target value condition.

[0030] Furthermore, in step S5-5, the weight matrix and the bias matrix are updated using the gradient descent method, and the update rate is controlled by defining a learning rate.

[0031] Furthermore, the size of the learning rate is determined by estimation. The specific method is: first specify a value, and then perform trial calculations in an exponential descent manner to find the optimal learning rate that has a fast calculation rate and can make the results converge.

[0032] The technical solution of the present invention has the following beneficial effects:

[0033] The present invention establishes a model original database by testing various data on the tunnel construction site, and then classifies various types of data according to the data change rate and establishes a transaction table. Then, based on the established transaction table, the association rules are calculated and established. The correlation between each sample data and the target prediction value can be established first, and the monitoring object most relevant to the target value can be screened out; then, the test data of the monitoring object and the surrounding rock deformation test data are used as the training database of the prediction model, and the established BP-neural network prediction model is used for training to obtain a prediction model that meets the target value conditions. The amount of sample data processed by the model can be significantly reduced, thereby accelerating the prediction rate of the model and improving the accuracy of the model prediction, so that the long-term deformation result of the surrounding rock after the excavation of the soft rock tunnel can be accurately predicted, which greatly reduces the cost of long-term monitoring of the tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments and in conjunction with the accompanying drawings.

[0035] Figure 1 It is a flow chart of the method for predicting long-term deformation of soft rock tunnels based on machine learning described in the present invention.

[0036] Figure 2 It is a flow chart of the method for establishing the association rule method.

[0037] Figure 3 This is a schematic diagram of the model tree creation process.

[0038] Figure 4 This is a comparison chart between the model prediction results and the measured results. DETAILED DESCRIPTION

[0039] The technical solution of the present invention is further described below in conjunction with the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention should be included in the protection scope of the present invention.

[0040] like Figure 1As shown, this embodiment also proposes a method for predicting long-term deformation of soft rock tunnels based on machine learning, comprising the following steps:

[0041] Step S1: Conduct tests at the tunnel construction site to obtain surrounding rock deformation, surrounding rock stress and support structure stress data during tunnel construction as the model original database.

[0042] Specifically, when testing is carried out at the tunnel construction site, the arch crown settlement measuring points, convergence measuring points and surrounding rock stress measuring points are arranged on the same monitoring section, and the arch frame stress measuring points and anchor stress measuring points are arranged on the monitoring section 0.9-1.1m away from them.

[0043] Step S2: Use data standardization methods to pre-process various types of data in the original database, eliminate the dimensions of the data, and classify various types of data according to the data change rate to establish a transaction table.

[0044] Specifically, when the data standardization method is used to pre-process various types of data in the original database, the maximum and minimum interval division method is used to process various types of data in the original database. The calculation formula is: After the processing is completed, the various types of data are classified according to the change rate. The data change rate is divided into level I, level II and level III according to the range of the data in [0,0.35), [0.35,0.7) and [0.7,1], which represent slow change rate, medium rate and fast rate respectively, and a data classification table is constructed, as shown in Table 1. Then, the data transaction table is generated based on this classification, and the medium and fast rates are called significant changes, and the association rule transaction table is established, as shown in Table 2; where TID represents the number of events, and one monitoring represents one event; A, B, ..., E represent various types of data; such as: arch settlement, convergence displacement, surrounding rock stress, arch frame stress and anchor stress.

[0045] Table 1

[0046] Data change rate characteristics scope Level indication slow [0,0.35) I medium [0.35,0.7) II quick [0.7,1] III

[0047] Table 2

[0048] TID Significantly changing data sets 1 A, B, C, E 2 A, B, C, D, E 3 A, B, D, E 4 A, B, D 5 A, B, D 6 A, D 7 A, D 8 D 9 D 10 D

[0049] Step S3: According to the established transaction table, calculate and establish association rules, and screen out monitoring objects with high correlation with tunnel surrounding rock deformation. Among them, according to the established transaction table, the specific method of calculating and establishing association rules is: find frequent item sets in the database based on the FP-Growth algorithm, and then calculate the confidence and lift of each frequent item set to establish association rules. Figure 2 The specific contents are as follows:

[0050] Step S3-1: Scan the entire database, count the number of occurrences of all 1-item sets, and remove items that do not meet the minimum support S requirement.

[0051] Specifically, the specific method for determining the minimum support S is: arrange the support numbers of each item in the transaction set in order from large to small, and number them 1, 2, ..., n, to establish a support "support number-sequence number" sequence; and construct a "support number-sequence number" table, as shown in Table 3; the support number is the number of times each monitoring data in the transaction table has a significant change, and the sequence number is numbered from 1 from large to small according to the support number; n is the number of items.

[0052] Table 3

[0053] Serial number 1 2 3 4 5 Support Number 9(D) 7(A) 5(B) 3(E) 2(C)

[0054] Build n-degree polynomial With the serial number as the independent variable x and the support number as the dependent variable y, fit each point in the table, where n is the power index of the highest order term of the proposed polynomial, and then solve the value when the second-order derivative of the fitting curve is 0 to find f(x0), and take the smallest integer not less than f(x0) as the minimum support S of the frequent item selected this time.

[0055] Step S3-2: Figure 3 As shown in the figure, the support number of each 1-item set is recorded, that is, the number of times each data item appears alone, and the data items with support numbers less than S are eliminated. All the retained frequent 1-item sets are arranged in descending order of frequency of occurrence, and a model tree is established with the empty set item Null as the root node. The frequent 1-item sets are connected to the nodes in the model tree in descending order according to the frequency of occurrence, and the number of occurrences is marked after each item set.

[0056] Step S3-3: Start from the bottom of the model tree and search for all transaction sets prefixed with a certain item set to obtain all frequent item sets in the transaction table, such as Then calculate the confidence and lift of each frequent item set, and take the one with high confidence and lift as the final association rule.

[0057] Step S4: The test data of the monitored object and the surrounding rock deformation test data are used as a training database for the prediction model.

[0058] Step S5: Establish a BP-neural network prediction model, use the training database for training, and obtain a prediction model that meets the target value conditions.

[0059] Specifically, all kinds of data with a high degree of correlation with the target prediction value in the established association rules are used as the input layer X of the BP-neural network prediction model, and the weight matrix W and the bias matrix b are set. At the same time, Sigmoid is taken as the excitation function of the model. After linear and nonlinear transformations are performed on the input layer X, the hidden layer Z is obtained; after linear and nonlinear transformations are performed on the hidden layer Z, the output layer Y is obtained. The measured surrounding rock deformation is used as the target value. By calculating the difference between the output layer Y and the target value, the weight matrix and the bias matrix are updated by the gradient descent method, and the update rate is controlled by defining the learning rate to obtain a prediction model in which the output layer meets the target value conditions.

[0060] Preferably, the learning rate is determined by estimation, specifically by first specifying a value of 0.1, and then performing trial calculations with values ​​of 0.01 and 0.001 in an exponentially decreasing manner to seek an optimal learning rate that has a fast calculation rate and can make the results converge.

[0061] like Figure 4 As shown, in order to verify the reliability and effectiveness of the long-term deformation prediction method for soft rock tunnels based on machine learning described in the present invention, the predicted data are compared with the data obtained from actual tests. Analysis and comparison results show that the deviation between the data predicted by the method of the present invention and the data obtained from actual tests is small. The present invention can achieve accurate prediction of the long-term deformation results of the surrounding rock after excavation of soft rock tunnels while ensuring the prediction rate.

[0062] The present invention establishes a model original database by testing various data on the tunnel construction site, and then classifies various types of data according to the data change rate and establishes a transaction table. Then, based on the established transaction table, the association rules are calculated and established. The correlation between each sample data and the target prediction value can be established first, and the monitoring object most relevant to the target value can be screened out; then, the test data of the monitoring object and the surrounding rock deformation test data are used as the training database of the prediction model, and the established BP-neural network prediction model is used for training to obtain a prediction model that meets the target value conditions. The amount of sample data processed by the model can be significantly reduced, thereby accelerating the prediction rate of the model, and at the same time improving the accuracy of the model prediction, so that the long-term deformation result of the surrounding rock after the excavation of the soft rock tunnel can be accurately predicted, which greatly reduces the cost of long-term monitoring of the tunnel, and is of great significance to the soft rock tunnel project.

[0063] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting long-term deformation of soft rock tunnels based on machine learning, characterized in that: The method comprises the following steps: Step S1: Conduct tests at the tunnel construction site to obtain surrounding rock deformation, surrounding rock stress and support structure stress data during tunnel construction as the original model database; Step S2: pre-processing various types of data in the original database using a data standardization method, eliminating the dimension of the data, and grading various types of data according to the data change rate to establish a transaction table; Step S3: According to the established transaction table, calculate and establish association rules, and screen out monitoring objects with high correlation with tunnel surrounding rock deformation; Step S4: using the test data of the monitored object and the surrounding rock deformation test data as a training database for the prediction model; Step S5: Establish a BP-neural network prediction model, use the training database for training, and obtain a prediction model that meets the target value conditions.

2. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 1 is characterized in that: In step S1, when testing is carried out at the tunnel construction site, the arch settlement measuring points, convergence measuring points and surrounding rock stress measuring points are arranged on the same monitoring section, and the arch frame stress measuring points and anchor stress measuring points are arranged on the monitoring section 0.9-1.1m away from them.

3. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 1, characterized in that: In step S2, the specific content of preprocessing various types of data in the original database using the data standardization method is: using the maximum and minimum interval division method to process various types of data in the original database, the calculation formula is:

4. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 1, characterized in that: In step S2, the specific method of establishing a transaction table after grading various types of data according to the data change rate is: the data change rate is divided into level I, level II and level III according to the range of the data in [0,0.35), [0.35,0.7) and [0.7,1], which respectively represent slow change rate, medium rate and fast rate, and a data transaction table is generated according to this classification.

5. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 1, characterized in that: In step S3, according to the established transaction table, the specific method of calculating and establishing the association rules is: finding the frequent item sets in the database based on the FP-Growth algorithm, and then calculating the confidence and lift of each frequent item set to establish the association rules.

6. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 5 is characterized in that: The specific method of finding frequent item sets in the database based on the FP-Growth algorithm is: Step S3-1: Scan the entire database, count the number of occurrences of all 1-item sets, and remove items that do not meet the minimum support S requirement; Step S3-2: Arrange all the retained frequent 1-item sets in descending order of occurrence frequency, establish a model tree with the empty set item Null as the root node, connect the frequent 1-item sets with the nodes in the model tree in descending order of occurrence frequency, and mark the number of occurrences after each item set; Step S3-3: Search for all transaction sets prefixed with a certain item set starting from the bottom of the model tree.

7. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 6, characterized in that: In step S3-1, the specific method for determining the minimum support S is: Step S3-1-1: Arrange the support numbers of each item in the transaction set in descending order, and number them 1, 2, ..., n to establish a support "support number-sequence number" sequence; where n is the number of items; Step S3-1-2: After establishing ordered plane coordinates, a polynomial is used for curve fitting; the expression of the polynomial is: Among them, x is the sequence number; y is the support number; n is the power index of the highest order term of the polynomial to be selected; Step S3-1-3: After solving the value when the second-order derivative of the fitting curve is 0, f(x0) is obtained, and the smallest integer not less than f(x0) is taken as the minimum support S of the frequent item selected this time.

8. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 1, characterized in that: In step S5, a BP-neural network prediction model is established and trained using a training database. The specific method for obtaining a prediction model that meets the target value condition is: Step S5-1: taking a data set of monitoring quantities that are highly correlated with the target predicted quantity as the input layer X; Step S5-2: setting the dimensions of the weight matrix W and the bias matrix b in the linear transformation, and using the Sigmoid function as the excitation function for the nonlinear transformation of the input layer; Step S5-3: Perform linear and nonlinear transformations on the input layer X to obtain a hidden layer Z; Step S5-4: Obtain the output layer Y after performing linear and nonlinear transformation on the hidden layer Z; Step S5-5: Taking the measured surrounding rock deformation as the target value, by calculating the difference between the output layer Y and the target value, updating the weighted matrix and the bias matrix, and obtaining a prediction model whose output layer meets the target value condition.

9. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 8, characterized in that: In step S5-5, the weight matrix and the bias matrix are updated using the gradient descent method, and the update rate is controlled by defining a learning rate.

10. The method for predicting long-term deformation of soft rock tunnels based on machine learning according to claim 9, characterized in that: The learning rate is determined by estimation. The specific method is: first specify a value, and then perform trial calculations in an exponential descent manner to find the optimal learning rate that has a fast calculation rate and can make the results converge.

Images