Intelligent forecasting method for underground water condition of tunnel

Abstract

The invention discloses a tunnel underground water condition intelligent forecasting method, which comprises the following steps: firstly, obtaining an apparent resistivity data matrix from a transient electromagnetic detection result, and carrying out data standardization; secondly, in order to improve data quality, apparent resistivity change trend characteristics are extracted, and abnormal values are eliminated; then, considering expert experience, fully utilizing prior information for detecting underground water conditions of the construction surfaces, and respectively establishing an underground water condition forecasting model according to data of different construction surfaces; and finally, extracting result information of the forecasting section to finish underground water condition forecasting. According to the method, the rule of detection data in the construction process is fully excavated, and when the underground water condition is forecasted, the detection data is accurately utilized according to different types of detection construction surfaces, so that the accuracy of the tunnel underground water condition forecasting model is remarkably improved, the safe construction operation of a tunnel is effectively guided, and the construction quality and safety are ensured.

Description

Technical Field

[0001] The invention relates to an intelligent prediction method for tunnel groundwater conditions taking into account apparent resistivity trend characteristics and construction surface geological information, and belongs to the field of tunnel advanced geological prediction. Background Art

[0002] Tunnel construction is a key task in the modernization of my country's comprehensive transportation system. Tunnel construction is highly uncertain and blind, and is prone to water inrush disasters. Forecasting the groundwater conditions ahead of tunnel excavation and preventing and controlling water inrush disasters are of great significance to ensuring the safety of tunnel construction.

[0003] At present, deep and long tunnel projects are concentrated in the western mountainous areas and karst areas with complex terrain and geological conditions. The detection data has multi-source heterogeneous characteristics. The conventional prediction method is to use apparent resistivity combined with geological information to predict groundwater conditions, but this approach is too dependent on the professionalism of the relevant personnel, which is costly and inefficient. There is also a method of intelligent prediction through neural network training models, but due to the complex geology and multi-source data, the prediction accuracy is not high. Therefore, a new intelligent prediction method for tunnel groundwater conditions is needed, which can analyze the distribution law of detection data, mine the effective features in the data, and accurately predict the groundwater conditions. Summary of the invention

[0004] The purpose of the present invention is to propose an intelligent prediction method for tunnel groundwater conditions, which takes into account the apparent resistivity trend characteristics and the geological information of the construction surface, and establishes a groundwater prediction model framework for the tunnel excavation process for different types of construction surfaces, so as to solve the problem of insufficient accuracy of the existing intelligent prediction models and prediction methods for tunnel groundwater conditions.

[0005] In order to achieve the above technical objectives, the technical solutions adopted by the present invention are as follows:

[0006] A method for intelligently predicting groundwater conditions in a tunnel comprises the following steps:

[0007] S1. Obtain the apparent resistivity data matrix from the transient electromagnetic detection results and perform data standardization;

[0008] S2, extract the apparent resistivity change trend characteristics and remove abnormal values;

[0009] S3, using the prior information of groundwater conditions on the detection application surface, establish groundwater condition prediction sub-models according to the data of different application surfaces;

[0010] S4. Extract the result information of the section to be predicted, match the corresponding groundwater condition prediction sub-model to complete the intelligent prediction of groundwater condition.

[0011] Furthermore, the apparent resistivity data matrix R obtained in step S1 is as follows:

[0012]

[0013] Its row vector and column vector represent different spatial dimensions, the row vector represents the apparent resistivity of the section perpendicular to the detection direction, and the column vector represents the apparent resistivity of the section parallel to the detection direction; where m and n are the number of rows and columns of the matrix R, respectively, representing the number of different detection points. ij It is the element in the i-th row and j-th column of the matrix R, representing the apparent resistivity value at the intersection of the i-th vertical section and the j-th parallel section.

[0014] Further, step S2 is specifically as follows:

[0015] S2.1. Define the apparent resistivity change trend characteristic matrix as T, and the calculation method of T is that each element t ij It is equal to the corresponding element in the original matrix R minus the element r in the same column of the first row of R. 1j , the calculation formula is:

[0016]

[0017] S2.2. When there is a situation where the mechanism does not match the data distribution, a data screening mechanism is added to calculate the value of the corresponding column of the apparent resistivity matrix of the tunnel face. When half of the values ​​are greater than the threshold, the column is removed. The conditions are met:

[0018] F(T i ,θ)≥0.5 (3)

[0019] In the formula, F(T i .,θ) is used to evaluate the proportion of apparent resistivity exceeding the threshold T in the i-th row of the matrix T. If this proportion is greater than half of the elements in the row, the row will be eliminated;

[0020] Specifically, F can be expressed as:

[0021]

[0022] Further, step S3 is specifically as follows:

[0023] S3.1. Based on the groundwater conditions on the working surface and the apparent resistivity range of the detection section, the groundwater conditions in the detection section are defined as:

[0024]

[0025] Where I is the groundwater condition of the construction surface, D is dry, L is linear water discharge, W is wet, and G is strand water discharge. Then I∈{D, L, W, G}, R max is the maximum value of the matrix R;

[0026] S3.2. Classify the dataset according to the geological differences of the transient electromagnetic detection application surface, and divide the dataset into four categories according to the standards of no water, wet, linear water and strand water;

[0027] S3.3, Assume that there is a training sample set {(x1, y1), (x2, y2), ..., (x n ,y n )}, where each x i is a row vector of a detection matrix, y i ∈{-1, 1} is the corresponding label, -1 means "wet", 1 means "water out in strands", in order to separate the data of different categories, we only need to find w and b such that:

[0028] y i (w·x i +b)≥1,i=1,...,n. (6)

[0029] On this basis, we need to minimize the norm of w and introduce the slack variable ξ i ≥0, a certain amount of misclassification is allowed, so the optimization problem becomes:

[0030]

[0031] In the formula, C is a preset positive penalty parameter used to control the penalty intensity of the error;

[0032] S3.4, extract four types of samples, namely, waterless, wet, linear water discharge and strand water discharge, to generate training sets, respectively use the support vector machine method to fit the relationship between data and labels, and establish a groundwater status prediction sub-model;

[0033] S3.5. Use the test set for verification. The verification indicators of the test set are as follows;

[0034]

[0035] Where N is the total number of groundwater condition categories; M is the total sample size of the test set; TP i is the number of correctly predicted categories for the i-th category; FP i is the number of non-i-th class that is incorrectly predicted as i-th class; FN i is the number of i-th class that is incorrectly predicted as non-i-th class.

[0036] The beneficial effects of the present invention are:

[0037] (1) The intelligent prediction method for tunnel groundwater conditions provided by the present invention takes into account the apparent resistivity trend characteristics and the geological information of the construction surface at the same time. By establishing different models by layering the construction surface, data errors caused by different construction surfaces can be eliminated as much as possible. Each model can focus on processing the apparent resistivity data corresponding to a single construction surface, thereby improving data utilization and prediction accuracy.

[0038] (2) The intelligent prediction method for tunnel groundwater conditions provided by the present invention can effectively eliminate the influence caused by differences in instrument parameters by extracting the change trend of apparent resistivity without relying on specific absolute values, and can obtain better prediction results.

[0039] (3) The intelligent prediction method for tunnel groundwater conditions provided by the present invention uses training set and validation set data to establish a groundwater condition prediction model. Then, the prediction effect of the model is verified using the test set data, which is conducive to the application of the present invention in actual production, can more accurately identify instances of each category, has good generalization ability, and can adapt to different data distributions and category characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a diagram of the architecture of the intelligent prediction model for tunnel groundwater conditions in an embodiment of the present invention.

[0041] Figure 2 1 is a result diagram of the extracted transient electromagnetic features before and after the outliers are removed in an embodiment of the present invention, wherein Figure (a) represents: before the outliers of apparent resistivity are removed; Figure (b) represents: after the outliers of apparent resistivity are removed.

[0042] Figure 3 This is the test result compared with manual judgment in the embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the technical solutions of the present invention are now described in detail with reference to the accompanying drawings and embodiments.

[0044] The present invention provides an intelligent prediction method for tunnel groundwater conditions, which takes into account the apparent resistivity trend characteristics and the geological information of the construction surface. First, a method is constructed as follows: Figure 1 The structure of the intelligent prediction model for tunnel groundwater conditions shown in Figure 1 is divided into two stages. The first stage is to extract the trend characteristics of the apparent resistivity of the detection data and Figure 2The outliers are removed as shown in the figure. The second stage is to statistically analyze the detection data of different types of construction surfaces according to the differences in groundwater conditions of the construction surfaces, and establish knowledge-driven models and data-driven models. Then, the apparent resistivity data is divided into training sets and test sets; finally, a groundwater condition prediction model is established, and the test set data is used to verify the effectiveness of the method.

[0045] Please refer to Figure 1 , Figure 1 : is a schematic diagram of an intelligent prediction method for tunnel groundwater conditions in an embodiment of the present invention that considers apparent resistivity trend characteristics and construction surface geological information, specifically comprising:

[0046] (1) Read the transient electromagnetic results profile file in the specified format to form an apparent resistivity matrix.

[0047] (2) When there is a situation where the mechanism does not match the data distribution, a data screening mechanism is added to extract the vectors corresponding to the labeled tunnel face mileage in the apparent resistivity matrix. When half of the values ​​of the vectors are greater than the threshold, the column is removed. Figure 2 It is a comparison chart before and after removing outliers;

[0048] (3) Subtract the first column vector from the apparent resistivity matrix to obtain the apparent resistivity change trend characteristics.

[0049] (4) All tunnel faces were subjected to mileage registration, feature matrix extraction, and outlier screening operations according to the above methods to form a data set. The data set was then classified according to the geological differences of the transient electromagnetic detection application surface. The data set was divided into four categories according to the standards of waterless, wet, linear water discharge, and strand water discharge.

[0050] (5) For different types of geological conditions at each construction surface, intelligent groundwater condition prediction models are established. The training process mainly uses the support vector machine method to establish an advanced geological prediction model, while the testing process is used to test the prediction effect of the model. In order to prove the effectiveness of the proposed method in tunnel groundwater condition prediction, the results of manual prediction are used for comparison and compared with the other four methods.

[0051] Example:

[0052] A method for intelligently predicting groundwater conditions in a tunnel comprises the following steps:

[0053] Step S1, normalizing the apparent resistivity data;

[0054] 1) Define the matrix after normalization of the detection data as R, where the row vector and column vector represent different spatial dimensions respectively. The row vector represents the apparent resistivity of the section perpendicular to the detection direction, and the column vector represents the apparent resistivity of the section parallel to the detection direction;

[0055]

[0056] Where m and n are the number of rows and columns of the matrix R, respectively, representing the number of different detection points. ij It is the element in the i-th row and j-th column of the matrix R, representing the apparent resistivity value at the intersection of the i-th vertical section and the j-th parallel section.

[0057] S2, extract the apparent resistivity change trend characteristics and remove abnormal values;

[0058] S2.1. Define the apparent resistivity change trend characteristic matrix T, and the calculation method of T is that each element t ij It is equal to the corresponding element in the original matrix R minus the element r in the same column of the first row of R. 1j , the calculation formula is:

[0059]

[0060] S2.2. When there is a situation where the mechanism does not match the data distribution, a data screening mechanism is added to calculate the value of the corresponding column of the apparent resistivity matrix of the tunnel face. When half of the values ​​are greater than the threshold, the column is removed. The conditions are met:

[0061] F(T i. ,θ)≥0.5 (3)

[0062] In the formula, F(T i· ,θ) is used to evaluate the proportion of apparent resistivity exceeding the threshold T in the i-th row of the matrix T. If this proportion is greater than half of the elements in the row, the row will be eliminated.

[0063] Specifically, F can be expressed as:

[0064]

[0065] S3, using the prior information of groundwater conditions on the detection application surface, establish groundwater condition prediction sub-models according to the data of different application surfaces;

[0066] S3.1, Submatrix extraction: When the maximum value of the apparent resistivity of each forecast section is below 2000, it is considered that the groundwater condition in the detection section is the same as the groundwater condition of the working surface. When the groundwater condition of the working surface is strand-shaped water discharge, there will only be a mixture of multiple groundwater conditions. The groundwater condition in the detection section can be defined as follows based on the groundwater condition of the working surface and the apparent resistivity range of the detection section:

[0067]

[0068] Where I is the groundwater condition of the construction surface, D is dry, L is linear water discharge, W is wet, and G is strand water discharge. Then I∈{D, L, W, G}, R max is the maximum value of the matrix R, and S is the possible groundwater condition in the detection section;

[0069] S3.2. Classify the dataset according to the geological differences of the transient electromagnetic detection application surface. According to the standards of no water, wet, linear water and strand water, the dataset is divided into four categories:

[0070] S3.3, Assume that there is a training sample set {(x1, y1), (x2, (y2), ..., (x n ,y n )}, where each x i is a row vector of a detection matrix, y i ∈{-1, 1} is the corresponding label, -1 means "wet", 1 means "water out in strands", in order to separate the data of different categories, we only need to find w and b such that:

[0071] y i (w·x i +b)≥1,i=1,...,n. (6)

[0072] On this basis, we need to minimize the norm of w and introduce the slack variable ξ i ≥0, allowing a certain amount of misclassification. Therefore, the optimization problem becomes:

[0073]

[0074] Where C is a preset positive penalty parameter used to control the penalty intensity of the error.

[0075] S3.4, extract four types of samples, namely, waterless, wet, linear water discharge and strand water discharge, to generate training sets, respectively use support vector machine method to fit the relationship between data and labels, and establish groundwater condition prediction model;

[0076] S3.5. Use the test set for verification. The verification indicators of the test set are as follows;

[0077]

[0078] Where N is the total number of groundwater condition categories; M is the total sample size of the test set; TP i is the number of correctly predicted categories for the i-th category; FP i is the number of non-i-th class that is incorrectly predicted as i-th class; FN i is the number of i-th class that is incorrectly predicted as non-i-th class.

[0079] S4. Extract the result information of the section to be predicted, match the corresponding groundwater condition prediction sub-model to complete the intelligent prediction of groundwater condition.

[0080] To demonstrate the effectiveness of the proposed method in predicting groundwater conditions in tunnels, the results of manual prediction are used for comparison and compared with four other methods.

[0081] Table 1 Test accuracy

[0082]

[0083]

[0084] Figure 3 Table 1 and Table 2 are the test results of the test set, using horizontal comparison. Figure 3 Table 1 is the test results of the intelligent prediction method for tunnel groundwater conditions and the manual prediction method proposed in the present invention, and Table 2 is the test results of the intelligent prediction method for tunnel groundwater conditions proposed in the present invention and other machine learning methods. The test results show that the model accuracy and generalization ability of the method proposed in the present invention are greatly improved, which is conducive to the application of the present invention in actual production.

[0085] Figure 3 Table 1 and Table 2 are the test results of the test set, using horizontal comparison. Figure 3 Table 1 is the test results of the intelligent prediction method for tunnel groundwater conditions and the manual prediction method proposed in the present invention, and Table 2 is the test results of the intelligent prediction method for tunnel groundwater conditions proposed in the present invention and other machine learning methods. The test results show that the model accuracy and generalization ability of the method proposed in the present invention are greatly improved, which is conducive to the application of the present invention in actual production.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligent prediction of tunnel groundwater conditions, characterized in that: The steps include: S1. Obtain the apparent resistivity data matrix from the transient electromagnetic detection results and perform data standardization; S2, extract the apparent resistivity change trend characteristics and remove abnormal values; S3, using the prior information of groundwater conditions on the detection application surface, establish groundwater condition prediction sub-models according to the data of different application surfaces; S4. Extract the result information of the section to be predicted, match the corresponding groundwater condition prediction sub-model to complete the intelligent prediction of groundwater condition.

2. The intelligent prediction method for tunnel groundwater conditions according to claim 1 is characterized in that: The apparent resistivity data matrix R obtained in step S1 is as follows: Its row vector and column vector represent different spatial dimensions, the row vector represents the apparent resistivity of the section perpendicular to the detection direction, and the column vector represents the apparent resistivity of the section parallel to the detection direction; where m and n are the number of rows and columns of the matrix R, respectively, representing the number of different detection points. ij It is the element in the i-th row and j-th column of the matrix R, representing the apparent resistivity value at the intersection of the i-th vertical section and the j-th parallel section.

3. The intelligent prediction method for tunnel groundwater conditions according to claim 2 is characterized in that: Step S2 is specifically as follows: S2.

1. Define the apparent resistivity change trend characteristic matrix as T, and the calculation method of T is that each element t ij It is equal to the corresponding element in the original matrix R minus the element r in the same column of the first row of R. 1j , the calculation formula is: S2.

2. When there is a situation where the mechanism does not match the data distribution, a data screening mechanism is added to calculate the value of the corresponding column of the apparent resistivity matrix of the tunnel face. When half of the values ​​are greater than the threshold, the column is removed. The conditions are met: F(T i .,θ)≥0.5 (3) In the formula, F(T i .,θ) is used to evaluate the proportion of apparent resistivity exceeding the threshold T in the i-th row of the matrix T. If this proportion is greater than half of the elements in the row, the row will be eliminated; Specifically, F can be expressed as:

4. The intelligent prediction method for tunnel groundwater conditions according to claim 3 is characterized in that: Step S3 is as follows: S3.

1. According to the groundwater condition of the working surface and the apparent resistivity range of the detection section, the groundwater condition in the detection section is defined as: Where I is the groundwater condition of the construction surface, D is dry, L is linear water discharge, W is wet, and G is strand water discharge. Then I∈{D, L, W, G}, R max is the maximum value of the matrix R; S3.

2. Classify the dataset according to the geological differences of the transient electromagnetic detection application surface, and divide the dataset into four categories according to the standards of no water, wet, linear water and strand water; S3.3, Assume that there is a training sample set {(x1, y1), (x2, y2), ..., (x n ,y n )}, where each x i is a row vector of a detection matrix, y i ∈{-1, 1} is the corresponding label, -1 means "wet", 1 means "water out in strands", in order to separate the data of different categories, we only need to find w and b such that: y i (w·x i +b)≥1,i=1,...,n. (6) On this basis, we need to minimize the norm of w and introduce the slack variable ξ i ≥0, a certain amount of misclassification is allowed, so the optimization problem becomes: In the formula, C is a preset positive penalty parameter used to control the penalty intensity of the error; S3.4, extract four types of samples, namely, waterless, wet, linear water discharge and strand water discharge, to generate training sets, respectively use the support vector machine method to fit the relationship between data and labels, and establish a groundwater status prediction sub-model; S3.

5. Use the test set for verification. The verification indicators of the test set are as follows; Where N is the total number of groundwater condition categories; M is the total sample size of the test set; TP i is the number of correctly predicted categories for the i-th category; FP i is the number of non-i-th class that is incorrectly predicted as i-th class; FN i is the number of i-th class that is incorrectly predicted as non-i-th class.

Images