A surrounding rock monitoring index effectiveness evaluation method

Abstract

The present application relates to a kind of surrounding rock monitoring index effectiveness evaluation method, comprising: obtaining the various monitoring index data of the section of tunnel, according to the mutation of surrounding rock displacement, the deformation process of surrounding rock is divided into several damage stages;For each section, respectively establish two-dimensional table for storing section monitoring data;With one or more monitoring index column in two-dimensional table as data feature, damage stage column as data label input machine learning model, data feature is classified according to data label, obtain the prediction accuracy matrix of surrounding rock damage determination, and obtain the significance score of each data feature, determine the effectiveness of monitoring index.This method can evaluate the effectiveness of the index of monitoring surrounding rock information under different scenarios / working conditions, so as to obtain the effectiveness of different types, different position data, to determine which type of monitoring data can better reflect rock mass damage, so as to guide the early stage of safety monitoring work, data analysis in middle period and safety decision in later period.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring of surrounding rock in deep underground engineering, specifically a method for evaluating the effectiveness of surrounding rock monitoring information. Background Technology

[0002] The most critical issue in on-site monitoring of underground engineering projects is determining which information to use as the primary reference: It is desirable for the monitoring information to directly reflect the mechanical properties of the surrounding rock and support structure. As underground engineering progresses towards deeper rock masses, engineering design and construction become increasingly complex, often leading to phenomena that defy empirical understanding. For example, for conventional shallow tunnels, monitoring indicator A may be more effective than indicator B in reflecting surrounding rock damage; however, for deep tunnels, the opposite conclusion may be reached. Therefore, it is essential to accurately identify which surrounding rock monitoring indicator best reflects the current state of surrounding rock damage in order to obtain more predictive information and comprehensive on-site evaluation data.

[0003] Currently, there is a lack of a scheme for evaluating the effectiveness of surrounding rock monitoring data. We need to know the effectiveness of data from different types and locations of measuring points, and we need to determine which type of monitoring data can better reflect rock mass damage, so as to guide the early-stage measuring point layout, mid-stage data analysis, and late-stage safety decision-making in safety monitoring work. Summary of the Invention

[0004] The purpose of this invention is to overcome the defects of the existing technology and provide a method for evaluating the effectiveness of surrounding rock monitoring indicators, thereby providing technical support for data analysis of surrounding rock stability analysis in underground engineering construction.

[0005] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0006] A method for evaluating the effectiveness of surrounding rock monitoring indicators, characterized in that it includes:

[0007] Acquire various monitoring index data of a certain section of the tunnel, wherein the monitoring index data includes at least the surrounding rock deformation data obtained through displacement monitoring;

[0008] The deformation process of the surrounding rock is divided into several damage stages based on the abrupt changes in the surrounding rock displacement.

[0009] A two-dimensional table is created for each cross section to store the cross section monitoring data. The attributes of the two-dimensional table include monitoring time, monitoring index and damage stage.

[0010] One or more monitoring index columns in a two-dimensional table are used as data features, and the damage stage column is used as data labels. The data features are then input into a pre-trained machine learning model. The machine learning model fits the relationship between the data features and the data labels, classifies the data features according to the data labels, obtains the prediction accuracy matrix for surrounding rock damage determination, and obtains the significance score of each data feature to determine the effectiveness of the monitoring indicators.

[0011] In a preferred embodiment, the various monitoring index data are measured data or simulated data obtained through simulation experiments.

[0012] As a preferred implementation method, displacement monitoring curves and acoustic emission microcrack development curves / images are plotted for each measuring point. The displacement monitoring curves are used as the primary method, supplemented by the acoustic emission microcrack development curves / images. Based on the abrupt changes in surrounding rock displacement, the deformation process of the surrounding rock is divided into several damage stages; the number of damage stages is 2-6. Surrounding rocks typically have stable and unstable modes. Stage division is based on existing methods, for example:

[0013] For cases where the surrounding rock is stable, it can be divided into two stages based on the inherent laws governing deformation and stability: Stage I is the deformation stage; Stage II is the stability stage.

[0014] For situations where the surrounding rock stabilizes and then gradually becomes unstable, it can be divided into the following stages based on the acoustic emission damage process and displacement mutation criteria: Stage I: Deformation Stage; Stage II: Stability Stage; Stage III: Instability Stage; Stage IV: Instability Stage; ...

[0015] When the surrounding rock cannot be stabilized, it can be divided into: Damage Stage I, Damage Stage II, Damage Stage III, etc.

[0016] The damage stages should not be defined too many times; this application divides them into 2 to 6 stages.

[0017] As a preferred implementation method, when collecting monitoring index data, there should be no less than two monitoring lines for cross-section monitoring, and at least one monitoring line should be arranged at the top of the tunnel and the sidewalls. When collecting monitoring index data, the length of the monitoring line for cross-section monitoring should be greater than one tunnel diameter, or determined according to the scale of the project.

[0018] As a preferred implementation method, when collecting monitoring index data, there should be no fewer than 3 measuring points on each measuring line, and the depth of the first measuring point should be less than 0.5m.

[0019] In a preferred embodiment, the monitoring index data also includes surrounding rock stress data obtained through stress monitoring; when collecting the monitoring index data, each measuring point simultaneously monitors horizontal and vertical displacement, as well as radial and tangential surrounding rock stress.

[0020] As a preferred embodiment, the monitoring index data also includes surrounding rock volume expansion rate data, which is determined based on the following formula:

[0021]

[0022] In the formula, The average volume expansion rate between two measuring points i and i+1 is denoted as the surrounding rock volume expansion rate. , These are the distances from measuring points i and i+1 to the center point of the tunnel, respectively. , These are the displacement distances of measuring points i and i+1, respectively. The surrounding rock volume expansion rate is a new surrounding rock deformation index used to describe the deformation of the surrounding rock mass between two measuring points, which can overcome the limitations of existing displacement indices.

[0023] As a preferred implementation method, a multi-class logistic regression algorithm is used for classification.

[0024] As a preferred implementation method, the same type of indicators are first calculated using a machine learning model. Then, the combination of indicators with the best significance among each type of indicators is selected for calculation to obtain the prediction accuracy matrix for each rock damage determination. The input data features corresponding to the model with high prediction accuracy are determined, and the significance scores of each data feature are obtained to determine the effectiveness of the monitoring indicators.

[0025] As a preferred implementation, during model computation, 50-80% of the data is used for model training, and the remaining data is used for accuracy testing. The data used for model training is then processed by shuffling and input into the machine learning model.

[0026] This invention utilizes machine learning technology to conduct experiments on surrounding rock damage assessment, thereby evaluating the effectiveness of surrounding rock monitoring indicators. The method of this invention can evaluate the effectiveness of indicators for monitoring surrounding rock information under different scenarios / working conditions, thus revealing the validity of data from different types and locations of measuring points. This allows for the determination of which type of monitoring data better reflects rock mass damage, guiding the initial measurement point layout, mid-term data analysis, and subsequent safety decision-making in safety monitoring work. Attached Figure Description

[0027] Figure 1 It is a two-dimensional tabular diagram.

[0028] Figure 2 It is a simulation calculation of the working condition design diagram.

[0029] Figure 3 These are acoustic emission monitoring images after the monitoring period has ended.

[0030] Figure 4 This is a visualization curve of monitoring data for operating condition 4.

[0031] Figure 5(a) shows the machine learning analysis results for working condition 1.

[0032] Figure 5(b) shows the machine learning analysis results for working condition 2.

[0033] Figure 5(c) shows the machine learning analysis results for working condition 3.

[0034] Figure 5(d) shows the machine learning analysis results for working condition 4.

[0035] Figure 5(e) shows the machine learning analysis results for operating condition 5.

[0036] Figure 5(f) shows the machine learning analysis results for working condition 6. Detailed Implementation

[0037] The application software and devices involved in the embodiments include:

[0038] During the design phase, if the monitoring index data is simulated data, the discrete element method PFC2D program can be used to conduct tunnel excavation monitoring simulation tests to obtain monitoring data; if the monitoring index data is measured data, relevant multi-point displacement monitoring equipment, surrounding rock stress monitoring equipment, acoustic emission monitoring equipment, etc. can be used to obtain cross-sectional monitoring index data.

[0039] ORIGIN (plotting software) and EXCEL programs can be used to visualize, edit, and store monitoring data.

[0040] Microsoft Azure machine learning program can be used in the data processing stage of machine learning.

[0041] Example 1

[0042] This embodiment combines simulation analysis to illustrate the specific steps and results of the method of the present invention.

[0043] Step 1: Collection of various monitoring data for the tunnel.

[0044] Taking a coal mine roadway in Henan Province as an example, the method and system disclosed in this invention were used to evaluate the effectiveness of surrounding rock monitoring indicators. This is a circular mining roadway in soft rock with a burial depth of 980 m and an average unit weight of 24.5 kN / m³ for the overlying rock strata. 3 The tunnel is located in mudstone strata with compressive and tensile strengths of 21.5 MPa and 1.28 MPa, respectively; elastic modulus and Poisson's ratio of 2.68 GPa and 0.23, respectively; and cohesion and internal friction angle of 2.5 MPa and 40°, respectively. Using the laboratory uniaxial results as a target, the micro-parameters of the model are adjusted through trial calculations to ensure that the uniaxial model response is basically consistent with the actual results.

[0045] A rock mass and tunnel excavation model was established. The rock mass dimensions were 18m × 18m, and the circular tunnel diameter was 3.5m. The vertical ground stress was initially set to 24MPa based on the actual engineering requirements. Three sets of corresponding horizontal ground stresses were then set: 12MPa, 24MPa, and 36MPa. The vertical ground stress was then set to 36MPa, with corresponding horizontal ground stresses of 24MPa, 36MPa, and 48MPa. A total of six calculation cases were designed, such as... Figure 2 As shown.

[0046] Two sets of measuring points were arranged around the tunnel, with three points in each set, at distances of 0.3m, 1.3m, and 2.3m from the free face of the tunnel opening, respectively, and were defined as measuring points X1, X2, and X3 on the sidewall and Y1, Y2, and Y3 on the tunnel roof. A total of four types of data were monitored: ① Acoustic emission monitoring, which recorded the number of microcracks in the surrounding rock after the tunnel excavation in real time; ② Surrounding rock stress monitoring, including radial stress (denoted as SR-measuring point) and tangential stress (denoted as ST-measuring point) at the measuring points; ③ Displacement monitoring (denoted as DISP-measuring point); ④ The volume expansion rate of the surrounding rock between measuring points 1~2, 2~3, and 1~3 in the two sets of measuring points was calculated according to formula (1) (denoted as VER-measuring point).

[0047] (1)

[0048] In the formula, The average volume expansion rate between two measuring points i and i+1 is denoted as the surrounding rock volume expansion rate. r is the distance from the measuring point to the center point of the tunnel, and u is the displacement of the measuring point.

[0049] Step 2: Divide the surrounding rock damage stages based on the monitoring results.

[0050] Figure 3 Acoustic emission microcrack monitoring images are presented after 8000 steps of surrounding rock monitoring under six different working conditions. The ORIGIN program is used for data visualization analysis. Figure 4 The displacement monitoring data and AE curve, volume expansion rate curve, radial stress and tangential stress curves for working condition 4 are given.

[0051] For each operating condition, one cross-section is selected for monitoring. The data from each cross-section is compiled into a two-dimensional table. See the format of the two-dimensional table for details. Figure 1 The column attributes of the two-dimensional table include monitoring time, index 1~x, and damage stage. Each cross-section has at least 100 records, i.e., 100 samples. The two-dimensional table is stored as a CSV file for subsequent model import. When selecting multiple cross-sections for a working condition, it is preferable to use only one two-dimensional table, adding a cross-section attribute to the column attributes to distinguish the cross-sections. Based on the acoustic emission monitoring curves and the surrounding rock displacement abrupt change criterion, the stable deformation stage of the surrounding rock under each working condition is divided into several damage stages, and the results are shown in Table 1.

[0052] Table 1. Classification of Injury Stages

[0053]

[0054] Step 3: Conduct damage assessment and evaluation experiments.

[0055] One or more monitoring indicator columns from a two-dimensional table are used as data features, and the damage stage column is used as data labels. These are input into a pre-trained machine learning model to fit the relationship between the data features and the data labels. The data features are then classified according to their labels, the significance scores of each data feature are evaluated, and a prediction accuracy matrix (confusion matrix) is obtained. During model training, the data samples (rows) are split into two distinct sets: a training set and a test set. x% of the data is used for model training, and the remaining data is used for accuracy testing. The parameter x should ideally be between 50 and 80. Finally, the data samples (rows) are shuffled before being input into the machine learning model.

[0056] The prediction accuracy matrix represents the accuracy result of surrounding rock damage assessment. A larger value (0-1) on the diagonal of the matrix indicates that the corresponding proportion of data was correctly identified, resulting in higher classification accuracy.

[0057] The higher the significance score, the better the significance of the indicator and the stronger its power.

[0058] This embodiment compares multi-class logistic regression algorithms, neural network classification, and multi-class decision forest methods, and ultimately selects the multi-class logistic regression algorithm with better accuracy for fitting.

[0059] The overall displacement index, overall VER index, and overall stress index (including radial and tangential stress indexes) for each working condition were calculated separately. The optimal results for each index (i.e., the index with the best significance among all indices) were then selected and combined into a comprehensive index for joint calculation of different types of indices. The prediction accuracy matrix and index significance score graphs were then derived, and the results are shown in Figures 5(a) to 5(f) (i.e., working conditions 1 to 6). It can be seen that:

[0060] Damage assessment accuracy analysis: The overall damage assessment results show high accuracy. The displacement and VER indices show the highest accuracy, while the stress index has moderate accuracy for surrounding rock damage assessment, but its accuracy is still generally greater than 70%. Significant errors occurred in the damage assessment of stage I in working conditions 5 and 6, possibly due to the short data sample size. In this case, the proportion of training samples (e.g., greater than 70%) or the data sample density could be increased. The accuracy ranking of damage assessment is: displacement index > VER index > tangential stress index > radial stress index.

[0061] Significance analysis of indicators: The significant indicators for condition 3 are the displacement of measuring point Y1, the volume expansion rate between measuring points Y1~Y2 and Y1~Y3, and the radial and tangential stress at measuring point Y3 on the tunnel roof. These results reflect the location and extent of damage to the tunnel roof in condition 3. Corresponding to condition 3, condition 6, with a higher ground stress level, has significant indicators for the displacement of tunnel roof Y2 and the volume expansion rate between Y2 and Y3. The results of the significant indicators for the two conditions reflect the difference in the degree of damage between the two conditions. Condition 4 is the opposite of conditions 3 and 6, with its significant indicators all pointing to the measuring points at the sidewall location. In condition 5, the significance scores of the volume expansion rate indicators for the tunnel roof and sidewall are similar, reflecting that the tunnel roof and sidewall have similar damage ranges in this condition.

Claims

1. A method for evaluating the effectiveness of a surrounding rock monitoring indicator, characterized by, include: Acquire various monitoring index data from multiple measuring points on a certain cross section of the tunnel, wherein the monitoring index data includes at least the surrounding rock deformation data obtained through displacement monitoring; The deformation process of the surrounding rock is divided into several damage stages based on the abrupt changes in the surrounding rock displacement. A two-dimensional table is created for each cross section to store the cross section monitoring data. The attributes of the two-dimensional table include monitoring time, monitoring index and damage stage. Using one or more monitoring indicator columns in a two-dimensional table as data features and damage stage columns as data labels, a pre-trained multi-class logistic regression model is input. The relationship between the data features and data labels is fitted by the multi-class logistic regression model, and the data features are classified according to the data labels to obtain the prediction accuracy matrix for surrounding rock damage assessment. The significance scores of each data feature are also obtained to determine the effectiveness of the monitoring indicators. This process includes: first, calculating the same type of indicators using the multi-class logistic regression model; then, selecting the indicator with the best significance among all types of indicators and combining them using the multi-class logistic regression model to obtain the prediction accuracy matrix for each surrounding rock damage assessment; determining the input data features corresponding to the model with high prediction accuracy; and obtaining the significance scores of each data feature to determine the effectiveness of the monitoring indicators.

2. The method of claim 1, wherein, The various monitoring indicator data are either actual measured data or simulated data obtained through simulation experiments.

3. The method of claim 1, wherein, Displacement monitoring curves and acoustic emission microcrack development curves / images were plotted for each measuring point. The displacement monitoring curves were used as the main data, and the acoustic emission microcrack development curves / images were used as supplementary data. The deformation process of the surrounding rock was divided into several damage stages according to the abrupt changes in the displacement of the surrounding rock.

4. The method according to claim 3, characterized in that, The number of damage stages is divided into 2 to 6.

5. The method of claim 1, wherein, When collecting monitoring index data, there should be no less than two monitoring lines for cross-section monitoring, with at least one monitoring line at the top of the tunnel and one at the sidewall; the length of the monitoring line for cross-section monitoring should be greater than one tunnel diameter, or determined according to the scale of the project.

6. The method of claim 1, wherein, When collecting monitoring index data, there should be no fewer than 3 measuring points on each measuring line, and the depth of the first measuring point should be less than 0.5m.

7. The method of claim 1, wherein, The monitoring index data also includes surrounding rock stress data obtained through stress monitoring; when collecting monitoring index data, each measuring point simultaneously monitors horizontal and vertical displacement, as well as radial and tangential surrounding rock stress.

8. The method of claim 1, wherein, The monitoring index data also includes surrounding rock volume expansion rate data, which is determined based on the following formula: ； In the formula, is the average volume dilatancy between two measuring points i, i+1, denoted as the surrounding rock volume dilatancy, , are the distances of measuring points i, i+1 from the tunnel center point, respectively, , are the displacement distances of measuring points i, i+1, respectively.

9. The method of claim 1, wherein, When calculating the model, 50-80% of the data is used for model training, and the remaining data is used for accuracy testing. The data used for model training is processed by randomizing and then input into the multi-class logistic regression model.

Images