Tunnel rock mass integrity intelligent forecasting method oriented to geological radar construction surface difference
Through geological radar detection methods, multi-source heterogeneous data and feature matrix are processed and convolutional neural network models are established, which solves the problem of low prediction accuracy of rock mass integrity in tunnel engineering, and achieves higher forecast accuracy and construction Security.
Patent Information
- Application Number
- CN202410989575.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-12-27
AI Technical Summary
In existing tunnel projects, due to the variable response characteristics of multi-source heterogeneous data and detection data, the intelligent forecasting accuracy of rock mass integrity is low, and it is impossible to effectively ensure the safety and efficiency of tunnel construction.
Using a geological radar detection method, by reading the relative amplitude data of the electromagnetic wave of geological radar, normalization, data space registration, feature matrix extraction and abnormal data screening, geological differences categories were divided, and convolutional neural network model was established to perform rock bodies. Integrity forecast.
It improves the accuracy of rock mass integrity forecasting, improves data quality, and ensures the safety and efficiency of tunnel construction.
Smart Images
Figure CN119200010A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of advanced geological prediction of tunnels, and in particular to a multi-model intelligent prediction method for tunnel rock mass integrity based on geological differences of working surfaces detected by geological radar. Background Art
[0002] Broken rock is a common adverse geological phenomenon in the process of tunnel excavation. If support measures are not taken in time, it often causes block falling or even landslides, water and mud inrush and other accidents. Therefore, it is an indispensable part of the tunnel construction process to carry out accurate and quantitative intelligent comprehensive prediction of the integrity of the rock mass in front of the face before tunnel excavation. However, the overall development trend of tunnel engineering at present is "deep burial, long tunnel line, high stress, strong karst, high water pressure, complex structure, and frequent disasters". A large amount of multi-source heterogeneous detection data will be used in the prediction process, and the response characteristics of the detection data to abnormal geological conditions are unpredictable, which brings challenges to the effective use of geophysical data and the intelligent prediction of rock mass integrity. How to extract effective features from unstructured data and then improve its value density is the premise for realizing the intelligent prediction of rock mass integrity. At the same time, most of the models used in the existing prediction methods do not take into account the geological differences of the construction surface, and cannot solve the problem of low prediction accuracy caused by the variable response characteristics of the detection data, thus making it difficult to ensure the safe and efficient progress of the tunnel construction process. Therefore, it is urgent to study a multi-model intelligent prediction method for tunnel rock integrity based on geological differences detected by geological radar. Summary of the invention
[0003] In order to solve the problem that the existing tunnel advance detection data has coexistence of multi-source heterogeneous data, a lot of interference data, and unclear differences in the distribution characteristics of the detection data, which leads to difficulty in extracting effective structured features and making accurate predictions ahead of the heading face, the present invention provides an intelligent prediction method for tunnel rock integrity for geological radar detection working faces.
[0004] The technical solution of the present invention is as follows:
[0005] An intelligent prediction method for tunnel rock integrity based on geological radar working surface differences includes the following steps:
[0006] S1: Read the geological radar detection results file, obtain the relative amplitude data of the geological radar electromagnetic wave, and normalize the elements in the data;
[0007] S2: Perform data space registration and geological radar feature matrix extraction for each mileage point on the tunnel face, and remove abnormal data in the feature matrix;
[0008] S3: Perform mileage registration, feature matrix extraction and outlier screening operations on all tunnel faces according to the method in S2 to form a data set, and divide the data sets into different categories according to the geological differences of the working faces detected by geological radar;
[0009] S4: According to the different geological categories of each construction surface, rock mass integrity prediction sub-models are established respectively, and the training is completed through the corresponding data set and convolutional neural network. Then the model with the highest accuracy is determined by the cross-validation method as the final advanced geological prediction multi-model;
[0010] S5: Use the final advanced geological prediction multi-model to match the appropriate rock integrity prediction sub-model according to the geological category of the construction surface of the detection data of the section to be predicted, and complete the advanced geological prediction of the actual tunnel project.
[0011] Furthermore, the relative amplitude data of the geological radar electromagnetic wave obtained in step S1 is a two-dimensional matrix as shown in the following formula (1):
[0012]
[0013] Among them, the column vector of A is the relative amplitude value corresponding to each measuring point distributed along the cross-sectional direction of the tunnel face, and its row vector is the relative amplitude value corresponding to the mileage points distributed along the mileage direction, m is the number of rows, and n is the number of columns;
[0014] By performing maximum and minimum normalization on each element of matrix A, the normalized matrix A can be obtained. norm , as shown in the following formula (2):
[0015]
[0016] where a ij_norm is the normalized matrix A norm The element in the i-th row and j-th column of ij is the element in the i-th row and j-th column of the original matrix A, a ij_min is the minimum value of matrix A, a ij_max is the maximum value of the matrix A.
[0017] Furthermore, the data space registration in step S2 is specifically as follows:
[0018] S2.1: For each mileage point on the tunnel face, the relative amplitude matrix of the geological radar electromagnetic wave obtained in S1 is matched to the relative amplitude matrix within the neighborhood of the mileage point. First, the row position corresponding to the mileage point in the electromagnetic wave relative amplitude matrix is determined. The formula is as follows:
[0019]
[0020] Among them, Lpalmface is the corresponding row of this mileage point in the relative amplitude matrix; M palmface is the mileage value where the tunnel face is located;
[0021] M workface is the mileage value where the construction face is located; T nsamp is the number of sampling points; T depth is the detection depth;
[0022] S2.2: Take the data of the L palmface th row and the subsequent fixed rows in the relative amplitude matrix, with a total of NUM _rows rows, to form the relative amplitude matrix within the neighborhood range corresponding to this mileage point of the tunnel face, denoted as B', corresponding to the neighborhood range of a certain distance before and after the tunnel face.
[0023] Furthermore, the extraction of the feature matrix in step S2 is specifically as follows:
[0024] S2.3: Perform an operation of summing the columns and taking the average on the newly formed matrix. First, calculate the number of columns N stacked to be summed each time, as shown in formula (4). Then, for each row element in the matrix B' obtained in the previous step, perform an operation of summing and taking the average every N stacked elements, to obtain the final feature matrix, denoted as B. Finally, unify the size of the relative amplitude matrix within the neighborhood range corresponding to each mileage point of the tunnel face to NUM _rows ×NUM _cols ;
[0025]
[0026] Furthermore, the elimination of abnormal data in step S2 is specifically as follows:
[0027] S2.4: Based on the feature matrix B obtained in S2.3, formulate rules for screening abnormal data, and determine the abnormal value index as the sum of row variances, denoted as Sum var , and the calculation formula is as follows, where b ij is the element in the i-th row and j-th column of the matrix B, and μ i is the average value of the i-th row;
[0028]
[0029] S2.5: Set the threshold as T. If Sum var <T, it indicates that the ground penetrating radar electromagnetic wave detection data has almost no fluctuation, which is an abnormal sample, and this tunnel face sample is eliminated.
[0030] Furthermore, the different categories of data sets in step S3 include: five categories of complete, relatively complete, relatively broken, broken, and extremely broken divided according to the integrity of the construction face.
[0031] The beneficial technical effects of the present invention are as follows:
[0032] The present invention proposes an intelligent prediction method for tunnel rock integrity based on geological radar detection of working surfaces. In terms of feature extraction, the unstructured detection data is structured and abnormal data screening rules are designed, which effectively improves the data quality. In terms of rock integrity prediction, various types of working surface prediction sub-models are established by considering the geological differences of the working surfaces, and the model is matched and the rock integrity is predicted according to the degree of fragmentation of the working surface of the detection data to be predicted, which can effectively improve the prediction accuracy and provide solid protection for the safety of life and property in tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is the overall architecture diagram of the method of the present invention.
[0034] Figure 2 Schematic diagram of the column-by-column superposition operation of feature matrices in an embodiment of the present invention.
[0035] Figure 3 It is a grayscale image of the anomaly detection results of the geological radar in the embodiment of the present invention.
[0036] Figure 4 These are the test results of three comparison methods in the embodiments of the present invention. DETAILED DESCRIPTION
[0037] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. The embodiments are only used to explain the present invention rather than to limit the present invention.
[0038] like Figure 1 As shown, the present invention provides a multi-model intelligent prediction method for tunnel rock integrity for geological radar detection working surface. First, according to the geological radar detection results data, mileage registration, feature matrix extraction, outlier removal and other operations are performed to form a relative amplitude feature matrix corresponding to each tunnel face to form a data set; then, according to the geological radar detection working surface, which is complete, relatively complete, relatively broken, broken, and extremely broken, the data set is divided into five categories, and rock integrity prediction multi-models are established respectively. For each model, the training set and validation set data are used to establish the rock integrity prediction model, and the test set data is used to test the final prediction accuracy of the model. The rock integrity prediction model is established using the cross-validation method; finally, the test set data is used for simulation verification.
[0039] Example:
[0040] A multi-model intelligent prediction method for tunnel rock mass integrity based on geological differences detected by geological radar, comprising:
[0041] S1: For the geological radar detection result file, use a specific file header format to read its detection result file and obtain the ground
[0042] The original relative amplitude matrix of the radar electromagnetic wave is obtained and normalized;
[0043] S2: For each mileage point on the tunnel face, the original relative amplitude matrix of the geological radar electromagnetic wave obtained in S1 is matched to the relative amplitude matrix within the neighborhood of the mileage point. In order to unify the size of the feature matrix, the column-by-column superposition strategy is adopted to extract the feature matrix corresponding to the mileage point;
[0044] Based on the obtained feature matrix, in order to improve data quality, abnormal data screening rules are formulated and the sum of row variances Sum is calculated. var , and set a threshold to remove abnormal data above the threshold;
[0045] S3: According to the feature matrix extraction and abnormal data screening method, the electromagnetic wave relative amplitude feature matrix is extracted for each tunnel face; since tunnel advanced geological prediction is a complex project, the characteristics presented by various types of geological radar detection data with different rock integrity levels are also complex and difficult to distinguish between categories. Therefore, the obtained data will be stratified according to the geological differences of the geological radar detection working faces, and the electromagnetic wave relative amplitude matrix and rock integrity label data of various working faces will be divided into training set, verification set and test set;
[0046] S4: Based on the convolutional neural network and cross-validation method, a multi-model for predicting tunnel rock mass integrity is established. The input parameter is the relative amplitude matrix of the geological radar, and the output is the rock mass integrity. The model is trained using the training set and the validation set, and the trained model is validated using the test set data to obtain the trained tunnel advanced geological prediction model.
[0047] S5: Use the trained tunnel rock integrity prediction multi-model to perform advanced geological prediction on the actual tunnel project.
[0048] Step S1 specifically includes the following process:
[0049] Read the DZT file of geological radar detection results, and the relative amplitude data of electromagnetic waves obtained is a two-dimensional matrix A, and the form of A is as follows:
[0050]
[0051] Among them, the column vector of A is the relative amplitude value corresponding to each measuring point distributed along the cross-sectional direction of the tunnel face, and its row vector is the relative amplitude value corresponding to the mileage points distributed along the mileage direction, m is the number of rows, and n is the number of columns;
[0052] By performing maximum and minimum normalization on each element of matrix A, the normalized matrix A can be obtained. norm , as shown below:
[0053]
[0054] where a ij_norm is the normalized matrix A norm The element in the i-th row and j-th column of ij is the element in the i-th row and j-th column of the original matrix A, a ij_min is the minimum value of matrix A, a ij_max is the maximum value of the matrix A.
[0055] Step S2 specifically includes the following process:
[0056] S2.1 For each mileage point on the tunnel face, the relative amplitude matrix of the obtained geological radar electromagnetic wave relative amplitude matrix is matched to the relative amplitude matrix within the neighborhood of the mileage point. First, the row position corresponding to the mileage point in the electromagnetic wave relative amplitude matrix is determined. The formula is as follows:
[0057]
[0058] Among them, L palmface is the corresponding row of the mileage point in the relative amplitude matrix; M palmface is the mileage value of the tunnel face;
[0059] M workface is the mileage value of the construction surface; T nsamp is the number of sampling points; T depth is the detection depth;
[0060] S2.2 Take the Lth in the relative amplitude matrix palmface Row and subsequent fixed row data, totaling NUM _rows The relative amplitude matrix in the neighborhood range corresponding to the mileage point of the tunnel face is formed, which is recorded as B', corresponding to the neighborhood range of a certain distance before and after the tunnel face;
[0061] S2.3 Perform the operation of stacking and averaging the newly formed matrix by columns. First, calculate the number of columns N that need to be stacked each time. stacked , formula as (4); then each row element in the matrix B' obtained in the previous step is stacked The elements are superimposed and averaged once to obtain the final characteristic matrix, which is recorded as B. Finally, the size of the relative amplitude matrix in the neighborhood corresponding to each mileage point on the tunnel face is unified to NUM _rows ×NUM _cols , denoted as B.
[0062]
[0063] As shown Figure 2 in the figure, the geological radar detection data is matched with the mileage of the heading face, the row position corresponding to the mileage point of the heading face in the relative amplitude matrix is calculated, and a total of 40 rows of data before and after it are taken as the feature matrix within the neighborhood range corresponding to the mileage point of the heading face, and the feature matrix is unified into 40×10 by the strategy of column-by-column superposition. The process of column-by-column superposition operation refers to Figure 2 .
[0064] S2.4 To improve the data quality, first, based on the feature matrix obtained in S2, an abnormal data screening rule is formulated, and the abnormal value index is determined as the sum of row variances, denoted as Sum var , and the calculation formula is as follows, where b ij is the element in the i-th row and j-th column of matrix B, and μ i is the mean value of the i-th row;
[0065]
[0066] S2.5 Set the threshold as T. If Sum var <T, it indicates that the geological radar electromagnetic wave detection data has almost no fluctuation, which is an abnormal sample, and the heading face sample will be excluded.
[0067] The grayscale image of the abnormal geological radar detection result refers to Figure 3 .
[0068] The specific steps of S3 include the following processes:
[0069] S3.1 According to the method of feature matrix extraction and abnormal data screening, extract the relative amplitude feature matrix of electromagnetic waves for each heading face to form a data set;
[0070] S3.2 According to the geological differences of the construction surfaces in the geological radar detection batches where each heading face sample is located, all samples are divided into five grades: complete, relatively complete, relatively broken, broken, and extremely broken according to the construction surface;
[0071] S3.3 Divide the five types of relative amplitude matrices of electromagnetic waves after division and the rock mass integrity label data into a training set, a validation set, and a test set;
[0072] The specific steps of S4 include the following processes:
[0073] S4.1 For the five types of samples with complete, relatively complete, relatively broken, broken, and extremely broken construction surfaces respectively, use the convolutional neural network and the cross-validation method to determine the model hyperparameters, and establish a multi-model for tunnel rock mass integrity prediction;
[0074] S4.2 Use the test set for verification, and the verification indicators are as follows:
[0075]
[0076] Among them, Accuracy is the accuracy, N true is the number of correctly predicted samples, N total is the total number of samples in the test set.
[0077] In order to prove the effectiveness of the proposed multi-domain feature intelligent extraction method in the prediction of tunnel rock integrity, the model was directly trained without dividing the samples according to the geological differences of the construction surface detected by geological radar, and compared with manual prediction.
[0078] These three methods were used to extract the electromagnetic wave relative amplitude matrix from the geological radar detection results of 10 real tunnels and conduct rock mass integrity prediction comparison experiments. Figure 4 The figures show the accuracy results of the three methods in 10 test tunnels. The results prove that the multi-model intelligent prediction method for tunnel rock integrity based on geological differences of the radar detection working surface proposed in the present invention has the best effect in the test results, which is conducive to the application of the present invention in actual production.
[0079] 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. An intelligent prediction method for tunnel rock mass integrity based on the difference in geological radar application surface, characterized by: The following steps are included: S1: Read the geological radar detection results file, obtain the relative amplitude data of the geological radar electromagnetic wave, and normalize the elements in the data; S2: Perform data space registration and geological radar feature matrix extraction for each mileage point on the tunnel face, and remove abnormal data in the feature matrix; S3: Perform mileage registration, feature matrix extraction and outlier screening operations on all tunnel faces according to the method in S2 to form a data set, and divide the data sets into different categories according to the geological differences of the working faces detected by geological radar; S4: According to the different geological categories of each construction surface, rock mass integrity prediction sub-models are established respectively, and the training is completed through the corresponding data set and convolutional neural network. Then the model with the highest accuracy is determined by the cross-validation method as the final advanced geological prediction multi-model; S5: Use the final advanced geological prediction multi-model to match the appropriate rock integrity prediction sub-model according to the geological category of the construction surface of the detection data of the section to be predicted, and complete the advanced geological prediction of the actual tunnel project.
2. The intelligent prediction method for tunnel rock mass integrity based on geological radar application surface differences as claimed in claim 1, characterized in that: The relative amplitude data of the geological radar electromagnetic wave obtained in step S1 is a two-dimensional matrix as shown in the following formula (1): Among them, the column vector of A is the relative amplitude value corresponding to each measuring point distributed along the cross-sectional direction of the tunnel face, and its row vector is the relative amplitude value corresponding to the mileage points distributed along the mileage direction, m is the number of rows, and n is the number of columns; By performing maximum and minimum normalization on each element of matrix A, the normalized matrix A can be obtained. norm , as shown in the following formula (2): where a ij_norm is the normalized matrix A norm The element in the i-th row and j-th column of ij is the element in the i-th row and j-th column of the original matrix A, a ij_min is the minimum value of matrix A, a ij_max is the maximum value of the matrix A.
3. The intelligent prediction method for tunnel rock mass integrity based on geological radar application surface differences as claimed in claim 2 is characterized by: The data space registration in step S2 is specifically as follows: S2.1: For each mileage point on the tunnel face, the relative amplitude matrix of the geological radar electromagnetic wave obtained in S1 is matched to the relative amplitude matrix within the neighborhood of the mileage point. First, the row position corresponding to the mileage point in the electromagnetic wave relative amplitude matrix is determined. The formula is as follows: Among them, L palmface is the corresponding row of the mileage point in the relative amplitude matrix; M palmface is the mileage value of the tunnel face; M workface is the mileage value of the construction surface; T nsamp is the number of sampling points; T depth is the detection depth; S2.2: Take the Lth relative amplitude matrix palmface Row and subsequent fixed row data, totaling NUM _rows The relative amplitude matrix in the neighborhood range corresponding to the mileage point of the tunnel face is formed, which is recorded as B', corresponding to the neighborhood range of a certain distance before and after the tunnel face.
4. The intelligent prediction method for tunnel rock mass integrity based on geological radar application surface differences as claimed in claim 3 is characterized by: The feature matrix extraction in step S2 is as follows: S2.3: Perform column-by-column stacking and mean-calculation operations on the newly formed matrix. First, calculate the number of columns N that need to be stacked each time. stacked , the formula is as follows (4), and then the elements of each row of the matrix B' obtained in the previous step are divided into stacked The elements are superimposed and averaged once to obtain the final characteristic matrix, which is recorded as B. Finally, the size of the relative amplitude matrix in the neighborhood corresponding to each mileage point on the tunnel face is unified as NUM _rows ×NUM _cols ; 5. The intelligent prediction method for tunnel rock mass integrity based on geological radar working surface differences as claimed in claim 4 is characterized by: The specific steps of removing abnormal data in step S2 are as follows: S2.4: Based on the feature matrix B obtained in S2.3, formulate abnormal data screening rules and determine the outlier index as the sum of row variances, denoted as Sum var , the calculation formula is as follows, where b ij is the element in the i-th row and j-th column of matrix B, μ i is the mean of the i-th row; S2.5: Set the threshold to T. If Sum var < T, it indicates that the ground penetrating radar electromagnetic wave detection data has almost no fluctuation, which is an abnormal sample, and the face sample will be excluded.
6. The intelligent prediction method for tunnel rock mass integrity based on geological radar application surface differences as claimed in claim 1, characterized in that: The different categories of data sets in step S3 include five categories: complete, relatively complete, relatively broken, broken, and extremely broken, divided according to the integrity of the construction surface.