Coal mine fault control degree evaluation method and system

Abstract

The invention discloses a coal mine fault control degree evaluation method and system, and relates to the technical field of fault identification reliability evaluation. The method comprises the following steps: fusing a mining area feature set with a label set, and establishing a training set and a to-be-predicted set; constructing a support vector machine dichotomy model and training the support vector machine dichotomy model; inputting the test set into the trained model, outputting a posterior probability value P1 of each sample point, and determining a dynamic threshold according to the posterior probability value; mapping the posterior probability into the fault point category by using the dynamically determined threshold value, and evaluating the quality of the fault point; and evaluating the fault risk grade of an area with densely distributed fault points in the mining area through the proportions of the breakpoints with different qualities. According to the method, objectivity and high efficiency of AI can be maintained, and reliability quantitative evaluation capability of artificial experience can be recovered and even exceeded.

Description

Technical Field

[0001] This invention relates to the field of fault identification reliability evaluation technology, and in particular to a method and system for evaluating the degree of fault control in coal mines. Background Technology

[0002] In the field of mineral extraction, accurately identifying geological structures such as faults is crucial for ensuring safety and optimizing design. Currently, mainstream fault identification and reliability assessment methods are mainly based on two types of technologies: The first type is traditional seismic interpretation technology, which relies on interpreters manually identifying and tracking faults by comprehensively analyzing seismic profiles and various attribute volumes. For reliability assessment, traditional methods rely on the interpreter's comprehensive judgment, which is highly dependent on the interpreter's personal experience and subjective judgment, making it difficult to quantify and standardize, resulting in poor reproducibility, difficulty in passing on the evaluation results, and low efficiency. The second type is conventional artificial intelligence fault identification technology, which, to overcome the drawbacks of manual interpretation, begins to utilize machine learning algorithms for automatic fault identification. This type of method achieves a leap in efficiency by learning patterns from seismic attributes. However, existing AI solutions, in pursuit of high recognition accuracy, generally go to the other extreme—"binary" simplification. They typically only output a label of "fault" or "non-fault." This approach has the following drawbacks: it loses uncertain information, cannot achieve reliability classification, and cannot meet the management needs of mine safety production for quantitative risk classification. This reveals a prominent contradiction in the current state of technology: traditional methods offer qualitative evaluation but are subjective and inefficient; emerging AI methods are efficient and objective but lack quantitative evaluation. Therefore, for those skilled in the art, there is an urgent need for a new method that can maintain the objectivity and efficiency of AI while restoring or even surpassing the reliability and quantitative evaluation capabilities of human experience in mine safety production. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for evaluating the degree of fault control in coal mines, which can maintain the objectivity and efficiency of AI algorithms, and restore or even surpass the reliability and quantitative evaluation capabilities of human experience, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following solution: On one hand, it provides a method for evaluating the degree of fault control in coal mines, the specific steps of which include the following: Extract seismic attribute values ​​and construct a feature set for the entire region; A tag set is created by statistically analyzing the exposure information from boreholes and roadways within the mining area. This tag set is then fused with the overall feature set for the entire area to establish a training set A and a prediction set B. The number of fault sample points in training set A is [number missing]. The number of non-fault sample points is ; Construct a support vector machine binary classification model, divide the training set into a first training set and a first test set, train the support vector machine binary classification model using the first training set, and use the first test set to test the support vector machine binary classification model until the training accuracy reaches a preset threshold, and output the trained support vector machine binary classification model. Input the first test set into the trained support vector machine binary classification model, use the support vector machine posterior probability model output, and record the posterior probability value P1 of each sample point in the first test set output by the model. Sort the posterior probability values ​​P1 and record the nth value. The size of each value and the +1 value size ; The prediction set B is used to make predictions through a trained support vector machine binary classification model, and the output of the support vector machine posterior probability model is used to record the posterior probability value P2 of each sample point in the prediction set output by the model. The quality of the fault points is evaluated based on the probability value P2, and the reliability of the fault points is classified. The risk level of faults in the mining area is evaluated by the proportion of faults of different qualities.

[0005] Preferably, the specific steps for extracting seismic attribute values ​​and constructing a feature set for the entire region are as follows: Obtain the raw seismic data volume of the target strata in the mining area; The location of the seismic reflection wave phase axis of the target layer is determined based on well logging data, and the target layer of the original seismic data volume is tracked to obtain the stratigraphic information of the target strata. Seismic attribute values ​​are extracted along the target layer of the original seismic data volume to construct the feature set of the entire region.

[0006] Preferably, the seismic attribute values ​​include variance, curvature, and instantaneous amplitude.

[0007] The preferred mathematical expression for evaluating and judging the quality of breakpoints is: ; in, This indicates the breakpoint evaluation result. A Indicates a reliable fault point. B Indicates a relatively reliable fault point. C Indicates an unreliable fault point. D Indicates a non-fault point.

[0008] The preferred mathematical expression for evaluating the fault risk level within a mining area is: ; in, This indicates the fault assessment result: 2 represents a high-risk fault area, 1 represents a medium-risk fault area, and 0 represents a low-risk fault area. a This indicates the number of Class A breakpoints. b This indicates the number of Class B breakpoints. c This indicates the number of Class C breakpoints.

[0009] On the other hand, a coal mine fault control degree evaluation system is provided, including a feature extraction module, a dataset construction module, a model training module, a posterior probability quantification output module, a dynamic threshold determination module, a prediction module, a fault point reliability classification module, and a fault risk level rating module. The feature extraction module is used to extract seismic attribute values ​​and construct a feature set for the entire region; The dataset construction module is used to collect exposed information from boreholes and tunnels within the mining area to create a tag set. This tag set is then fused with the overall feature set to establish a training set A and a prediction set B. The number of fault sample points in training set A is... The number of non-fault sample points is ; The model training module is used to construct a support vector machine binary classification model, divide the training set into a first training set and a first test set, train the support vector machine binary classification model using the first training set, and use the first test set to test the support vector machine binary classification model until the training accuracy reaches a preset threshold, and output the trained support vector machine binary classification model. The posterior probability quantization output module is used to input the first test set into the trained support vector machine binary classification model, use the support vector machine posterior probability model output, and record the posterior probability value P1 of each sample point in the first test set output by the model. The dynamic threshold determination module is used to sort the posterior probability values ​​P1 and record the... The size of each value and the +1 value size ; The prediction module is used to predict the set B to be predicted using a trained support vector machine binary classification model, and to record the posterior probability value P2 of each sample point in the set to be predicted output by the support vector machine posterior probability model. The fault point reliability classification module is used to evaluate the quality of fault points based on the probability value P2 and classify the reliability of fault points. The fault risk level rating module is used to evaluate the fault risk level in the mining area by the proportion of faults of different qualities.

[0010] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: It successfully resolved the core contradiction that "traditional methods have evaluations but are subjective and inefficient, while conventional AI methods are efficient and objective but lack evaluations," and constructed a complete, data-driven reliability evaluation closed loop. By employing the posterior probability model output of a support vector machine, what is obtained is not a simple "yes / no" label, but a continuous probability value P between 0 and 1. This value quantitatively expresses the model's confidence level that each sample point is a "fault," providing a continuous and quantifiable basis for subsequent fine-grained classification. The method of dynamically determining the grading threshold range ensures the adaptability of the grading standard to the specific data characteristics of the mining area and avoids the blindness of manually setting the threshold. Extending the value of the grading results to the macro-level decision-making process, and quantitatively evaluating the fault risk level of the entire region based on the spatial proportion of faults at different reliability levels, allows geological identification results to directly and clearly guide engineering planning and risk prevention, greatly enhancing the practical value of the method. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram showing the drilling, tunnel layout, and fault exposure within the study area of ​​this invention. Figure 3 This is a graph showing the results of the fault control degree assessment based on the SVM posterior probability model of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] The purpose of this invention is to provide a method for evaluating the degree of fault control in coal mines, such as... Figure 1 As shown, the specific steps include the following: S1. Extract seismic attribute values ​​and construct a feature set for the entire region; S2. Collect data from boreholes and roadways within the mining area to create a label set (faults are labeled 1, non-faults are labeled 0). Merge this label set with the overall feature set for the entire area to establish a training set A and a prediction set B. The number of fault sample points in training set A is... The number of non-fault sample points is ; S3. Construct a support vector machine binary classification model. Take 80% of the training set A for training the support vector machine binary classification model and 20% for testing the support vector machine binary classification model until the training accuracy exceeds 98%. Then, the model training is terminated and the trained support vector machine binary classification model is output. S4. Input the test set into the trained support vector machine binary classification model, use the output of the support vector machine posterior probability model, and record the posterior probability value P1 of each sample point in the first test set output by the model. S5. Arrange the posterior probability values ​​P1 in ascending order and record the nth probability value. The size of each value and the +1 value size ; S6. Use the set to be predicted B to make predictions using the trained support vector machine binary classification model, and use the output of the support vector machine posterior probability model to record the posterior probability value P2 of each sample point in the set to be predicted output by the model. S7. Evaluate the quality of the fault points based on the probability value P2 and classify the reliability of the fault points. S8. Evaluate the fault risk level in the mining area by the proportion of faults of different qualities.

[0015] Furthermore, the specific steps for extracting seismic attribute values ​​from S1 and constructing a feature set for the entire region are as follows: S11. Obtain the original seismic data volume of the target strata in the mining area; S12. Determine the position of the seismic reflection wave phase axis of the target layer based on the well logging data, and track the target layer of the original seismic data volume to obtain the layer information of the target strata. Extract seismic attribute values ​​such as variance, curvature, and instantaneous amplitude along the target layer of the original seismic data volume to construct a feature set for the entire region.

[0016] Furthermore, in S7, the quality of the fault points is evaluated based on the probability value P2, classifying them into four categories: reliable fault points (A), relatively reliable fault points (B), unreliable fault points (C), and non-fault points (D). The mathematical expression for the evaluation and discrimination is as follows: ; in, This indicates the breakpoint evaluation result. A Indicates a reliable fault point. B Indicates a relatively reliable fault point.C Indicates an unreliable fault point. D Indicates a non-fault point.

[0017] Furthermore, in S8, the fault risk level of densely distributed fault points within the mining area is evaluated based on the proportion of faults of different qualities, classifying them into high-risk, medium-risk, and low-risk fault zones. The mathematical expression is as follows: ; in, This indicates the fault assessment result: 2 represents a high-risk fault area, 1 represents a medium-risk fault area, and 0 represents a low-risk fault area. a This indicates the number of Class A breakpoints. b This indicates the number of Class B breakpoints. c This indicates the number of Class C breakpoints.

[0018] In this invention, S4 and S6 form the basis of the grading. By employing the posterior probability model output of a support vector machine, what is obtained is not a simple "yes / no" label, but a continuous probability value P between 0 and 1. This value quantitatively expresses the model's confidence level that each sample point is a "fault," providing a continuous and quantifiable basis for subsequent fine-grained grading.

[0019] Furthermore, S5 is key to achieving scientific classification. This invention does not subjectively set a fixed probability threshold, but innovatively utilizes the known number of non-tomographic samples in the training set. In the probability sequence output by the model to the test set, locate the first... The and the first +1 probability value This is equivalent to allowing the model to calibrate the judgment criteria based on its own "learning experience" (training set distribution), thereby dynamically determining the grading threshold range. This method ensures the adaptability of the grading criteria to the specific data characteristics of the mining area, avoiding the blindness of manually setting thresholds.

[0020] Based on this, S7 completed the reliability evaluation of a single fault point. This was achieved using the dynamically determined thresholds described above. The posterior probability P is mapped to four distinct categories: reliable (A), relatively reliable (B), unreliable (C), and non-fault point (D). This completely changes the limitation of traditional methods that can only provide binary judgments, and enables a refined classification of the recognition results.

[0021] Ultimately, S8 extends the value of the grading results to the macro-level decision-making. Based on the spatial proportion of fault points (A, B, C) at different reliability levels, it quantitatively evaluates the fault risk level (high, medium, low) of the entire region. This allows the geological identification results to directly and clearly guide engineering planning and risk prevention, greatly enhancing the practical value of the method.

[0022] Furthermore, the method of this invention is used to quantitatively grade and evaluate the reliability of faults in specific research areas. A specific embodiment is as follows: Study Area Overview: The study area features significant elevation differences, steep strata dips, and complex geological conditions, classifying it as a typical karst landform. The main mineable coal seams in this mining area are No. 2, No. 3, No. 7+8, and No. 9 coal seams, with a mineable depth of up to 200m and the longest coal roadway exceeding 1500m. The target coal seam for this study is No. 2 coal seam within the mining area.

[0023] The study area is dominated by normal faults, with a small number of reverse faults. The drilling, tunnel layout, and fault exposure within the study area are as follows: Figure 2 As shown, information was collected from 11 wells, 4 roadways, and 18 exposed faults. Table 1 lists the displacement, dip angle, and extension length of the exposed faults. To facilitate the research, the study area was divided into sections. Section A mainly consists of small faults, with displacements not exceeding 5m, and all faults are normal (including F2, F3, F4, F5, F6, and F7). Section B uses faults with larger displacements, all exceeding 20m, and all are normal (including F8 and F9); F10 and F11 were not included in the study due to their small displacement. Section C has diverse fault types, including faults with different displacements and properties (including F16, F17, and F18). Therefore, a higher-quality dataset can be constructed for Section C based on Sections A and B.

[0024]

[0025] Table 1 Dataset construction and model training: The target coal seam was calibrated using well logging information collected within the study area, and the stratigraphic position of coal seam #2 was determined using seismic interpretation software. The seismic data for the work area has a pixel size of 5m × 10m, while the pixel size in the interpretation software is 5m × 5m, totaling 50,451 data points within the work area. A 10ms time window was used to extract four attributes along the coal seam: variance, instantaneous phase, instantaneous frequency, and chaotic volume, constructing an attribute dataset for the study area. This dataset served as input features for the subsequent construction of the support vector machine model.

[0026] Simultaneously, the fault information exposed at drilling sites and tunnels was statistically analyzed, with "existence" represented by "1" and "non-existence" by "0". These data points were matched with their attribute values ​​based on their coordinate locations to construct a dataset with known labels. Among them, there were 3724 known fault points and 11835 known non-fault points.

[0027] Using known data to train SVM models, the PSO algorithm was used to help optimize the main parameters C and g of the SVM models. The optimal parameter values ​​and their corresponding optimal fitness values ​​for each model are shown in Table 2.

[0028]

[0029] Table 2 The SVM model built on the region A dataset reached an optimal fitness value of 98.0066% after 14 generations and remained unchanged. At this point, parameter C was 1.1224 and parameter g was 16.3124.

[0030] The test set data is fed into the trained model. The support vector machine posterior probability model is then used to output the posterior probability value P1 for each sample point in the test set. Furthermore, the posterior probability values ​​P1 output by the model are arranged in ascending order, and the magnitude of the 11835th value is recorded. The size of the 11836th value .

[0031] SVM binary classification model fault prediction results: All data in regions A, B, and C are organized into a training set to train the support vector machine model, and then used to predict the entire study area.

[0032] SVM posterior probability model fault point reliability evaluation results: Based on binary fault identification, the posterior probability output of the support vector machine model is further performed. The set to be predicted is predicted using the trained model, and the posterior probability model output of the support vector machine is used. The posterior probability value P of each sample point in the set to be predicted is recorded. The quality of the fault points is evaluated based on the probability value P, and they are divided into four categories: reliable fault points (red area), relatively reliable fault points (red area), unreliable fault points (red area), and non-fault points (red area). The results are as follows: Figure 3 As shown, the posterior probability model outputs richer information and can better analyze the likelihood of a fault existing.

[0033] On the other hand, a coal mine fault control degree evaluation system is provided, including a feature extraction module, a dataset construction module, a model training module, a posterior probability quantification output module, a dynamic threshold determination module, a prediction module, a fault point reliability classification module, and a fault risk level rating module. The feature extraction module is used to extract seismic attribute values ​​and construct a feature set for the entire region. The dataset construction module is used to statistically analyze the exposed information from boreholes and tunnels within the mining area to create a label set. This label set is then fused with the overall feature set to establish a training set A and a prediction set B. The number of fault sample points in training set A is [number missing]. The number of non-fault sample points is ; The model training module is used to construct a support vector machine binary classification model. The training set is divided into a first training set and a first test set. The support vector machine binary classification model is trained using the first training set and the first test set is used to test the support vector machine binary classification model until the training accuracy reaches a preset threshold. The trained support vector machine binary classification model is then output. The posterior probability quantization output module is used to input the first test set into the trained support vector machine binary classification model, use the support vector machine posterior probability model output, and record the posterior probability value P1 of each sample point in the first test set output by the model. The dynamic threshold determination module is used to sort the posterior probability values ​​P1 and record the threshold values. The size of each value and the +1 value size ; The prediction module is used to predict the set B to be predicted using a trained support vector machine binary classification model, and to record the posterior probability values ​​P2 of each sample point in the set to be predicted output by the support vector machine posterior probability model. The fault point reliability classification module is used to evaluate the quality of fault points based on the probability value P2 and classify the reliability of fault points. The fault risk rating module is used to evaluate the fault risk level in the mining area by the proportion of faults of different qualities.

[0034] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for evaluating the degree of fault control in coal mines, characterized in that, The specific steps include the following: Extract seismic attribute values ​​and construct a feature set for the entire region; A tag set is created by statistically analyzing the exposure information from boreholes and roadways within the mining area. This tag set is then fused with the overall feature set for the entire area to establish a training set A and a prediction set B. The number of fault sample points in training set A is [number missing]. The number of non-fault sample points is ; Construct a support vector machine binary classification model, divide the training set into a first training set and a first test set, train the support vector machine binary classification model using the first training set, and use the first test set to test the support vector machine binary classification model until the training accuracy reaches a preset threshold, and output the trained support vector machine binary classification model. Input the first test set into the trained support vector machine binary classification model, use the support vector machine posterior probability model output, and record the posterior probability value P1 of each sample point in the first test set output by the model. Sort the posterior probability values ​​P1 and record the nth value. The size of each value and the +1 value size ; The prediction set B is used to make predictions through a trained support vector machine binary classification model, and the output of the support vector machine posterior probability model is used to record the posterior probability value P2 of each sample point in the prediction set output by the model. The quality of the fault points is evaluated based on the probability value P2, and the reliability of the fault points is classified. The risk level of faults in the mining area is evaluated by the proportion of faults of different qualities.

2. The method for evaluating the degree of fault control in coal mines according to claim 1, characterized in that, The specific steps for extracting seismic attribute values ​​and constructing a feature set for the entire region are as follows: Obtain the raw seismic data volume of the target strata in the mining area; The location of the seismic reflection wave phase axis of the target layer is determined based on well logging data, and the target layer of the original seismic data volume is tracked to obtain the stratigraphic information of the target strata. Seismic attribute values ​​are extracted along the target layer of the original seismic data volume to construct the feature set of the entire region.

3. The method for evaluating the degree of fault control in coal mines according to claim 1, characterized in that, The seismic attribute values ​​include variance, curvature, and instantaneous amplitude.

4. The method for evaluating the degree of fault control in coal mines according to claim 1, characterized in that, The mathematical expression for evaluating and judging the quality of breakpoints is: ； in, This indicates the breakpoint evaluation result. A Indicates a reliable fault point. B Indicates a relatively reliable fault point. C Indicates an unreliable fault point. D Indicates a non-fault point.

5. The method for evaluating the degree of fault control in coal mines according to claim 4, characterized in that, The mathematical expression for evaluating the fault risk level within a mining area is: ； in, This indicates the fault assessment result: 2 represents a high-risk fault area, 1 represents a medium-risk fault area, and 0 represents a low-risk fault area. a This indicates the number of Class A breakpoints. b This indicates the number of Class B breakpoints. c This indicates the number of Class C breakpoints.

6. A coal mine fault control degree evaluation system, characterized in that, It includes a feature extraction module, a dataset construction module, a model training module, a posterior probability quantization output module, a dynamic threshold determination module, a prediction module, a fault point reliability classification module, and a fault risk level rating module. The feature extraction module is used to extract seismic attribute values ​​and construct a feature set for the entire region; The dataset construction module is used to collect exposed information from boreholes and tunnels within the mining area to create a tag set. This tag set is then fused with the overall feature set to establish a training set A and a prediction set B. The number of fault sample points in training set A is... The number of non-fault sample points is ; The model training module is used to construct a support vector machine binary classification model, divide the training set into a first training set and a first test set, train the support vector machine binary classification model using the first training set, and use the first test set to test the support vector machine binary classification model until the training accuracy reaches a preset threshold, and output the trained support vector machine binary classification model. The posterior probability quantization output module is used to input the first test set into the trained support vector machine binary classification model, use the support vector machine posterior probability model output, and record the posterior probability value P1 of each sample point in the first test set output by the model. The dynamic threshold determination module is used to sort the posterior probability values ​​P1 and record the... The size of each value and the +1 value size ; The prediction module is used to predict the set B to be predicted using a trained support vector machine binary classification model, and to record the posterior probability value P2 of each sample point in the set to be predicted output by the support vector machine posterior probability model. The fault point reliability classification module is used to evaluate the quality of fault points based on the probability value P2 and classify the reliability of fault points. The fault risk level rating module is used to evaluate the fault risk level in the mining area by the proportion of faults of different qualities.

Images