Multi-source ground pressure danger level quantitative evaluation method based on XGBoost algorithm

Abstract

The invention discloses an XGBoost algorithm-based multi-source ground pressure danger level quantitative evaluation method, which comprises the following steps of: performing automatic weight distribution and calculation on all indexes influencing a ground pressure danger level through an XGBoost algorithm, determining ground pressure key indexes suitable for a current mine, arranging different detection points in a to-be-monitored area, correspondingly collecting the ground pressure key indexes, and determining a multi-source ground pressure danger level according to the ground pressure key indexes. The method comprises the following steps of: calculating a ground pressure index, assigning other ground pressure key indexes except the ground pressure index through an XGBoost algorithm, quantitatively calculating a ground pressure danger level in combination with a principal stress difference value monitored in real time, grading a quantized value, and realizing visual intelligent grading early warning in combination with colors. According to the method, visual comprehensive early warning based on time history and region multi-index and multi-parameter fusion is achieved, the problem that due to the fact that internal correlation of all indexes is complex and the weight is difficult to determine, the accuracy of ground pressure disaster forecasting is poor is effectively solved, and the technical problems that an existing ground pressure monitoring technology is single in monitoring index and high in false alarm rate are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of detection and early warning of ground pressure dynamic disasters in mining, and in particular to a quantitative evaluation method for multi-source ground pressure hazard levels based on an XGBoost algorithm. Background Art

[0002] Mine ground pressure dynamic disaster is a highly complex and nonlinear dynamic instability phenomenon. There are many influencing factors that induce ground pressure disasters, such as lithology, rock mass quality grade, degree of fracture development, fracture occurrence distribution, degree of mining disturbance, etc. At present, the specific mechanism of multi-source ground pressure disasters has not been clarified, and a unified theoretical standard has not yet been formed.

[0003] Ground pressure monitoring and early warning are the prerequisites for effective prevention and control of ground pressure, and are also the key means to avoid disasters. There are many ground pressure monitoring methods, which are divided into regional monitoring and local monitoring. Regional monitoring can adopt microseismic monitoring methods, and local monitoring mainly includes electromagnetic radiation monitoring methods, geoacoustic monitoring methods, stress monitoring methods, drilling cuttings methods, and vibration wave CT monitoring. However, most monitoring methods reflect the physical and mechanical properties of coal and rock, the magnitude of load stress, and the role of disturbance dynamic loads. They only reflect the characteristics of one aspect, and most methods can only monitor but not prevent. It is difficult to extend them to non-coal rock mining. They are not suitable for situations where there are many indicators and complex weights of multi-source ground pressure disasters, and cannot accurately reflect the precursor information of ground pressure disasters.

[0004] The current single-indicator prediction and early warning method for mine ground pressure disasters is the fundamental reason for the serious lack of accuracy in forecasting mine ground pressure disasters and the high false alarm rate. The existing ground pressure monitoring technology is developing in the direction of regionalization, continuous online, and intelligent networking. However, due to the many factors affecting mine ground pressure, people's understanding of the laws of ground pressure manifestation is insufficient. At present, the accuracy of ground pressure disaster risk identification and monitoring and early warning cannot meet production needs. Therefore, how to reasonably, effectively, scientifically and reliably carry out mine ground pressure monitoring and control is a major issue and problem that needs to be solved urgently in mines, especially non-coal mines. Summary of the invention

[0005] The purpose of the present invention is to propose a quantitative evaluation method for multi-source ground pressure hazard level based on the XGBoost algorithm to solve the problem that most of the existing ground pressure monitoring technologies are based on coal and rock characteristics, and the monitoring indicators are single and cannot be applied to the monitoring and early warning of multi-source ground pressure disasters in non-coal mines.

[0006] The present invention is achieved through the following technical solutions:

[0007] The present invention proposes a quantitative evaluation method for multi-source ground pressure hazard level based on XGBoost algorithm, which includes the following steps:

[0008] S1, screening of key ground pressure indicators: combining the mine type and production practice, collecting ground pressure indicator data that affect the ground pressure hazard level in the sample area as sample data, taking the ground pressure indicator as the input indicator and the ground pressure hazard level as the output indicator, using the XGBoost algorithm to obtain the importance ranking and training model of each ground pressure indicator, and then determining that the key indicators for evaluating the ground pressure hazard level are five categories: rock mass grade, section index, disturbance coefficient, ground pressure index and support index, and generating the assignment model of each key ground pressure indicator and the ground pressure hazard level quantification model;

[0009] S2, deploy detection points and obtain key indicators of ground pressure: select representative detection points in the area to be monitored, deploy three-dimensional stress monitoring equipment at the detection points, and collect ground pressure stress data of the detection points in real time; record the physical coordinates of the detection points, and simultaneously collect the rock mass quality, tunnel / chamber section geometry parameters, rock mass disturbance status, and support status data of the detection points;

[0010] S3, assigning values ​​to key indicators of ground pressure: according to the association between the physical coordinates of the detection points and the three-dimensional model of the mine, the data collected at each detection point is input into the key indicator assignment model of ground pressure generated in step S1, and the rock mass grade, section index, disturbance coefficient and support index are assigned and calculated to obtain the fixed indicator value of each corresponding point;

[0011] S4, obtaining the ground pressure index value: the real-time ground pressure stress data obtained in step S2 are processed by the hole wall strain method, and the magnitude, azimuth and inclination data of the maximum principal stress, intermediate principal stress and minimum principal stress with time-history characteristics at each detection point are calculated and obtained, and the difference between the maximum principal stress and the minimum principal stress is further calculated as the ground pressure index value.

[0012] S5, quantitatively calculate the ground pressure hazard level: according to the fixed index values ​​of rock mass grade, section index, disturbance coefficient and support index obtained in S3 and the ground pressure index value obtained in step S4, input them into the ground pressure hazard level quantification model generated in step S1 for AI calculation, take the ground pressure hazard level as the output index, and constrain the output result within the interval [1,5), obtain the quantified value of the ground pressure hazard level at each detection point, and output the ground pressure hazard level evaluation table;

[0013] S6, determine the ground pressure danger level and implement color warning: divide the ground pressure danger level into four levels, and assign different color warning signals to each level, specifically: level one [1.000, 1.999], level two [2.000, 2.999], level three [3.000, 3.999], level four [4.000, 4.999], corresponding to blue, yellow, orange and red color warning signals respectively; combined with the location coordinates of each detection point, generate a visualization graph of the ground pressure danger level and output the corresponding warning measures to complete the intelligent graded warning.

[0014] Based on the above technical scheme, the key ground pressure indicators that affect the ground pressure danger level are effectively screened out from numerous ground pressure indicators through the XGBoost algorithm, and a comprehensive early warning method based on regional and time-course multi-indicator and multi-parameter fusion is realized. The problem of poor accuracy in ground pressure disaster forecasting caused by the complex inherent correlation of each indicator and the difficulty in determining the weight is effectively avoided, and the technical difficulties of existing ground pressure monitoring technology with single monitoring indicators and high false alarm rate are solved.

[0015] Furthermore, the step of screening the key ground pressure indicators in step S1 includes:

[0016] S101, based on the actual mining status of the mine, select the sample area, divide the factors affecting ground pressure into internal determining factors, external induced parameters and aggravating factors to determine the ground pressure index, and comprehensively collect ground pressure index data that affect the ground pressure danger level in the sample area;

[0017] S102, using the collected ground pressure index data as a model input index, using the ground pressure danger level as a model output index, performing data normalization processing and model training using an XGBoost algorithm, and obtaining a ground pressure danger level quantitative model;

[0018] S103, calling the feature_importances attribute in the trained XGBoost algorithm model to obtain the importance ranking value of each ground pressure indicator;

[0019] S104, performing cluster analysis based on the importance ranking values ​​obtained in S103, dividing the key indicators for evaluating the ground pressure danger degree into five categories: rock mass grade, section index, disturbance data, ground pressure index, and support index, and obtaining the value assignment model for each key indicator;

[0020] Furthermore, the data normalization processing method in step S102 is: respectively obtaining the feature times weight, feature average gain value Gain and feature average coverage rate Cover of each ground pressure index used as a segmentation sample in the XGBoost algorithm model, as shown in specific calculation formulas (1) to (3):

[0021] weight=|X| (1)

[0022]

[0023]

[0024] Among them, X is the set of ground pressure indicators classified into each leaf node of the XGBoost algorithm model; Gain x is the gain of each node, cover x is the number of samples at each node.

[0025] Furthermore, the three-dimensional stress monitoring equipment in step S2 includes a three-dimensional stress probe, a data acquisition box and a data transmission base station. The data transmission base station is arranged according to the mining field, the three-dimensional stress probe and the data acquisition box are arranged at the detection point, the data acquisition box is connected to the data transmission base station through a communication cable, the three-dimensional stress probe is arranged in the borehole and connected to the data acquisition box, and three groups of strain rosettes with an angle of 120° to each other are arranged on the three-dimensional stress probe, and the strain rosette includes a plurality of strain gauges for collecting stress data.

[0026] Furthermore, the borehole has a diameter of 40 mm and a depth of 1.0 to 1.2 m. The borehole is inclined downward by 1 to 2 degrees relative to the horizontal position. The purpose of this design is to make the three-dimensional stress probe fit more closely with the borehole, so as to improve the sensitivity of the strain gauge in capturing stress changes, thereby improving the accuracy of the detection data.

[0027] Furthermore, the rock mass grade, section index, disturbance coefficient and support index in step S3 are assigned as follows:

[0028] Rock mass grade: determined by the RMR method, a specific value constrained within the interval [1,5];

[0029] Section index: With section height, section span and section area as input values, the weight is automatically calculated according to the XGBoost algorithm to generate a comprehensive index value;

[0030] Disturbance coefficient: It is selected according to the degree of influence shown near the monitoring point after blasting. The degree of influence is divided into five levels. Combined with the proportion of unfilled goaf in the detection point area, the XGBoost algorithm assigns a specific value in the interval of [0,5]. The classification standard is as follows: Level 1 [0,1], relatively light, the rock mass is intact and there are no cracks; Level 2 (1,2], medium, the rock mass is relatively intact, and there are basically no obvious cracks; Level 3 (2,3], relatively serious, obvious cracks appear, and the width is less than 1 cm; Level 4 (3,4], serious, the rock mass is relatively broken, and the crack width is greater than 1 cm; Level 5 (4,5], very serious, the rock mass is broken and prone to spalling and roof collapse;

[0031] Support index: According to the support type and specific support parameters, the weight is automatically calculated according to the XGBoost algorithm to generate a comprehensive index value. The support types are divided into no support, sprayed support, anchor mesh support, steel-concrete support, and steel support.

[0032] Furthermore, the magnitude, azimuth and inclination data of the maximum principal stress, intermediate principal stress and minimum principal stress with time-history characteristics obtained in step S4 are excluded from abnormal points according to the SPC abnormality judgment principle before use to avoid interference from sudden and non-continuous accidental factors such as blasting disturbances, and generate time-history curves of principal stress and principal stress difference, and radar maps of principal stress inclination and azimuth to assist in judging the dangerous level of ground pressure.

[0033] The multi-source ground pressure hazard level quantitative evaluation method based on the XGBoost algorithm provided by the present invention is combined with the mine type and actual situation. First, all indicators affecting the ground pressure hazard level are automatically weighted and calculated through the XGBoost algorithm in big data and artificial intelligence technology to determine the key ground pressure indicators suitable for the current mine. Secondly, different detection points are arranged in the monitored area to collect the corresponding key ground pressure indicators, and other key ground pressure indicators except the ground pressure index are assigned values ​​within a certain constraint range through the XGBoost algorithm. Finally, the ground pressure hazard level is quantitatively calculated in combination with the principal stress value and principal stress difference monitored in real time, and the quantitative value of the ground pressure hazard level is graded. The intelligent graded warning is completed in combination with color, and a visual comprehensive warning based on multi-indicator and multi-parameter fusion of time course and region is realized, which effectively avoids the problem of poor accuracy of ground pressure disaster forecasting caused by the complex internal correlation of each indicator and the difficulty in determining the weight, and solves the technical difficulties of the existing ground pressure monitoring technology with a single monitoring indicator and a high false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0035] Figure 1 It is the visual early warning information interface of the present invention;

[0036] Figure 2 It is a schematic diagram of the arrangement of the strain rosette of the three-dimensional stress probe of the present invention;

[0037] Figure 3 It is a drilling schematic diagram of the present invention;

[0038] Figure 4 It is the ground pressure key index input interface diagram of the present invention;

[0039] Figure 5 A schematic diagram showing the visualization of the ground pressure hazard level of the present invention; DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto.

[0041] The present invention provides a quantitative evaluation method for multi-source ground pressure hazard level based on XGBoost algorithm, such as Figure 1 As shown in the figure, a visual warning information interface intelligently generated according to the method is shown. The specific steps for generating the interface are as follows:

[0042] S1, screening of key ground pressure indicators: combining the mine type and production practice, collecting ground pressure indicator data that affect the ground pressure hazard level in the sample area as sample data, taking the ground pressure indicator as the input indicator and the ground pressure hazard level as the output indicator, using the XGBoost algorithm to obtain the importance ranking and training model of each ground pressure indicator, and then determining that the key indicators for evaluating the ground pressure hazard level are five categories: rock mass grade, section index, disturbance coefficient, ground pressure index and support index, and generating the assignment model of each key ground pressure indicator and the ground pressure hazard level quantification model;

[0043] Among them, the specific steps for screening key indicators of ground pressure include:

[0044] S101, combined with the actual mining status of the mine, select the sample area, divide the factors affecting ground pressure into internal determining factors, external induced parameters and aggravating factors to determine the ground pressure index, and comprehensively collect ground pressure index data that affect the ground pressure hazard level in the sample area; specifically, the internal determining factors mainly include the original rock stress conditions, rock mass quality grade, rock hardness, rock mass structural surface conditions and characteristics, etc.; the external induced factors mainly include the magnitude and direction distribution of secondary stress, the degree of rock movement deformation, the response characteristics of the rock mass disturbed by mining, the degree of surrounding rock fragmentation, the excavation and blasting method, the mining process and sequence, etc.; the aggravating factors include the quality of engineering construction, the exposed volume of the goaf, the favorableness of the occurrence, the mining rate, and the support method and effect, etc.

[0045] S102, using the collected ground pressure index data as the model input index, using the ground pressure danger level as the model output index, and performing data normalization processing and model training using the XGBoost algorithm to obtain a quantitative model of the ground pressure danger level; wherein the data normalization processing method is to obtain the feature times weight, feature average gain value Gain and feature average coverage rate Cover of each ground pressure index used as a segmentation sample in the XGBoost algorithm model, respectively, as shown in the specific calculation formulas (1) to (3):

[0046] weight=|X| (1)

[0047]

[0048] X is the set of ground pressure indicators classified into each leaf node of the XGBoost algorithm model; Gain x is the gain of each node, cover x is the number of samples at each node.

[0049] S103, calling the feature_importances attribute in the trained XGBoost algorithm model to obtain the importance ranking value of each ground pressure indicator;

[0050] S104, performing cluster analysis based on the importance ranking values ​​obtained in S103, dividing the key indicators for evaluating the ground pressure danger degree into five categories: rock mass grade, section index, disturbance data, ground pressure index, and support index, and obtaining the value assignment model for each key indicator;

[0051] S2, deploy detection points and obtain ground pressure key indicator data: select representative detection points in the area to be monitored, deploy three-dimensional stress monitoring equipment at the detection points, and collect ground pressure stress data of the detection points in real time; record the physical coordinates of the detection points, and simultaneously collect the rock mass quality, tunnel / chamber section geometric parameters, rock mass disturbance state, and support situation data of the detection points; wherein the three-dimensional stress monitoring equipment includes a three-dimensional stress probe, a data acquisition box and a data transmission base station, the data transmission base station is deployed according to the mining field, the three-dimensional stress probe and the data acquisition box are deployed at the detection point, the data acquisition box is connected to the data transmission base station through a communication cable, the three-dimensional stress probe is deployed in a borehole with a hole diameter of 40mm and a hole depth of 1.0-1.2m and is connected to the data acquisition box, such as Figure 2 As shown, the three-dimensional stress probe is provided with three groups of strain rosettes with an angle of 120° to each other, and the strain rosette includes a plurality of strain gauges for collecting stress data; Figure 3 As shown, the borehole is drilled at a downward inclination of 1 to 2 degrees relative to the horizontal position, so that the three-dimensional stress probe fits more closely with the borehole, thereby improving the sensitivity of the strain gauge in capturing stress changes, thereby improving the accuracy of the detection data.

[0052] S3, ground pressure key indicator assignment: according to the detection point entity coordinates and the mine three-dimensional model associated, such as Figure 4 As shown, the data collected at each detection point is input into the assignment model of the ground pressure key indicators generated in step S1, and the four types of ground pressure key indicators except the ground pressure index are assigned and calculated to obtain the fixed indicator value of each corresponding point; the assignment method and standard of each indicator are as follows:

[0053] (1) Rock mass grade: determined by the RMR method, which is a specific value constrained in the interval [1,5];

[0054] (2) Section index: Using section height, section span, and section area as input values, the XGBoost algorithm automatically calculates weights and generates a comprehensive index value;

[0055] (3) Disturbance coefficient: The disturbance coefficient is selected according to the degree of influence shown near the monitoring point after blasting. The degree of influence is divided into five levels. Combined with the proportion of unfilled goaf in the detection point area, the XGBoost algorithm assigns a specific value in the interval [0,5]. The classification standard is as follows: Level 1, relatively light, the rock mass is intact and has no cracks; Level 2, medium, the rock mass is relatively intact and basically has no obvious cracks; Level 3, relatively serious, obvious cracks appear, and the width is less than 1 cm; Level 4, severe, the rock mass is relatively broken, and the crack width is greater than 1 cm; Level 5, very severe, the rock mass is broken and prone to spalling and roof collapse;

[0056] (4) Support index: Based on the support type and specific support parameters, the weight is automatically calculated and the comprehensive index value is generated according to the XGBoost algorithm. The support types are divided into no support, shotcrete support, anchor mesh support, steel-concrete support, and steel support.

[0057] S4, obtaining the ground pressure index value: the real-time ground pressure stress data obtained in step S2 is processed by the hole wall strain method. Preferably, the collected stress data is excluded from abnormal points according to the SPC abnormality judgment principle to avoid interference from sudden, non-continuous accidental factors such as blasting disturbances, and then the maximum principal stress, intermediate principal stress and minimum principal stress with time-course characteristics at each detection point are calculated and obtained. The difference between the maximum principal stress and the minimum principal stress is further calculated as the ground pressure index value. At the same time, a time-course curve diagram of the principal stress and the principal stress difference, and a radar diagram of the principal stress inclination and azimuth are generated to assist in judging the ground pressure danger level, such as Figure 1 shown.

[0058] S5, quantitative calculation of ground pressure hazard level: according to the fixed index values ​​of rock mass grade, section index, disturbance coefficient and support index obtained in S3 and the ground pressure index value obtained in step S4, they are input into the ground pressure hazard level quantification model generated in step S1 for AI calculation, and the ground pressure hazard level is used as the output index. The output result is constrained within the interval [1,5), and the ground pressure hazard level quantification value at each detection point is obtained, and the ground pressure hazard level evaluation table is output. Table 1 shows the ground pressure hazard level evaluation of some detection points:

[0059] Table 1 Earth pressure hazard level evaluation table

[0060] Probe No. Rock mass grade Section Index Perturbation coefficient Ground pressure index Support index Hazard Level YL01 3.048 4.195 0.650 50.000 0.300 2.112 YL02 3.047 4.195 0.650 50.320 0.300 1.136 YL03 3.091 4.195 0.650 49.660 0.300 2.695 YL04 3.103 4.195 0.650 48.332 0.300 1.876 YL05 3.188 4.195 0.650 49.623 0.300 3.012 YL06 3.485 4.195 0.650 52.360 0.300 1.136 YL07 3.105 4.195 0.650 52.000 0.300 2.555

[0061] S6, determine the ground pressure danger level and implement color warning: divide the ground pressure danger level into four levels, and assign different color warning signals to each level, specifically: level 1 [1.000, 1.999], level 2 [2.000, 2.999], level 3 [3.000, 3.999], level 4 [4.000, 4.999], corresponding to blue, yellow, orange, and red color warning signals respectively; combined with the position coordinates of each detection point, such as Figure 5 As shown, a visualization graph of the ground pressure danger level is generated to output the corresponding early warning and disposal measures, thus completing the intelligent graded early warning.

[0062] The treatment measures for the voltage danger levels in various regions are shown in Table 2:

[0063]

[0064] The present invention uses the XGBoost algorithm to automatically assign and calculate the weights of all indicators that affect the ground pressure hazard level, determines and assigns key ground pressure indicators suitable for the current mine, quantitatively calculates the ground pressure hazard level in combination with real-time monitored stress data, and completes intelligent graded warning in combination with color, thereby realizing a visual comprehensive warning based on time-course and regional multi-indicator and multi-parameter fusion, effectively avoiding the problem of poor accuracy in ground pressure disaster forecasting due to the complex inherent correlation of various indicators and the difficulty in determining weights, and solving the technical difficulties of existing ground pressure monitoring technology with a single monitoring indicator and a high false alarm rate.

[0065] The above description is only a preferred embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any slight modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the present invention.

Claims

1. A quantitative evaluation method for multi-source ground pressure hazard level based on XGBoost algorithm, characterized by: The steps include: S1, screening of key ground pressure indicators: combining the mine type and production practice, collecting ground pressure indicator data that affect the ground pressure hazard level in the sample area as sample data, taking the ground pressure indicator as the input indicator and the ground pressure hazard level as the output indicator, using the XGBoost algorithm to obtain the importance ranking and training model of each ground pressure indicator, and then determining that the key indicators for evaluating the ground pressure hazard level are five categories: rock mass grade, section index, disturbance coefficient, ground pressure index and support index, and generating the assignment model of each key ground pressure indicator and the ground pressure hazard level quantification model; S2, deploy detection points and obtain key indicators of ground pressure: select representative detection points in the area to be monitored, deploy three-dimensional stress monitoring equipment at the detection points, and collect ground pressure stress data of the detection points in real time; record the physical coordinates of the detection points, and simultaneously collect the rock mass quality, tunnel / chamber section geometry parameters, rock mass disturbance status, and support status data of the detection points; S3, assigning values ​​to key indicators of ground pressure: according to the association between the physical coordinates of the detection points and the three-dimensional model of the mine, the data collected at each detection point is input into the key indicator assignment model of ground pressure generated in step S1, and the rock mass grade, section index, disturbance coefficient and support index are assigned and calculated to obtain the fixed indicator value of each corresponding point; S4, obtaining the ground pressure index value: the real-time ground pressure stress data obtained in step S2 are processed by the hole wall strain method, and the magnitude, azimuth and inclination data of the maximum principal stress, intermediate principal stress and minimum principal stress with time-history characteristics at each detection point are calculated and obtained, and the difference between the maximum principal stress and the minimum principal stress is further calculated as the ground pressure index value. S5, quantitatively calculate the ground pressure hazard level: according to the fixed index values ​​of rock mass grade, section index, disturbance coefficient and support index obtained in S3 and the ground pressure index value obtained in step S4, input them into the ground pressure hazard level quantification model generated in step S1 for AI calculation, take the ground pressure hazard level as the output index, and constrain the output result within the interval [1,5), obtain the quantified value of the ground pressure hazard level at each detection point, and output the ground pressure hazard level evaluation table; S6, determine the ground pressure danger level and implement color warning: divide the ground pressure danger level into four levels, and assign different color warning signals to each level, specifically: level one [1.000, 1.999], level two [2.000, 2.999], level three [3.000, 3.999], level four [4.000, 4.999], corresponding to blue, yellow, orange and red color warning signals respectively; combined with the location coordinates of each detection point, generate a visualization graph of the ground pressure danger level and output the corresponding warning measures to complete the intelligent graded warning.

2. The multi-source ground pressure hazard level quantitative evaluation method based on the XGBoost algorithm according to claim 1 is characterized in that: The ground pressure key indicator screening step of step S1 includes: S101, based on the actual mining status of the mine, select the sample area, divide the factors affecting ground pressure into internal determining factors, external induced parameters and aggravating factors to determine the ground pressure index, and comprehensively collect ground pressure index data that affect the ground pressure danger level in the sample area; S102, using the collected ground pressure index data as a model input index, using the ground pressure danger level as a model output index, performing data normalization processing and model training using an XGBoost algorithm, and obtaining a ground pressure danger level quantitative model; S103, calling the feature_importances attribute in the trained XGBoost algorithm model to obtain the importance ranking value of each ground pressure indicator; S104, performing cluster analysis based on the importance ranking values ​​obtained in S103, dividing the key indicators for evaluating the danger level of ground pressure into five categories: rock mass grade, section index, disturbance data, ground pressure index, and support index, and obtaining the value assignment model for each key indicator.

3. The method for quantitatively evaluating the multi-source ground pressure hazard level based on the XGBoost algorithm according to claim 2 is characterized in that: The data normalization processing method in step S102 is: respectively obtain the feature times weight, feature average gain value Gain and feature average coverage rate Cover of each ground pressure index used as a segmentation sample in the XGBoost algorithm model, as shown in the specific calculation formulas (1) to (3): weight=|X| (1) Among them, X is the set of ground pressure indicators classified into each leaf node of the XGBoost algorithm model; Gain x is the gain of each node, cover x is the number of samples at each node.

4. The method for quantitatively evaluating the multi-source ground pressure hazard level based on the XGBoost algorithm according to claim 1 is characterized in that: The three-dimensional stress monitoring equipment in step S2 includes a three-dimensional stress probe, a data acquisition box and a data transmission base station. The data transmission base station is arranged according to the mining field. The three-dimensional stress probe and the data acquisition box are arranged at the detection point. The data acquisition box is connected to the data transmission base station through a communication cable. The three-dimensional stress probe is arranged in the borehole and connected to the data acquisition box. Three groups of strain rosettes with an angle of 120° to each other are arranged on the three-dimensional stress probe. The strain rosette includes a plurality of strain gauges for collecting stress data.

5. The method for quantitatively evaluating the multi-source ground pressure hazard level based on the XGBoost algorithm according to claim 4 is characterized in that: The borehole has a diameter of 40 mm and a depth of 1.0 to 1.2 m, and the borehole is inclined downward by 1 to 2 degrees relative to the horizontal position.

6. The multi-source ground pressure hazard level quantitative evaluation method based on XGBoost algorithm according to claim 1 is characterized in that: The assignment criteria for rock mass grade, section index, disturbance coefficient and support index in step S3 are: Rock mass grade: determined by the RMR method, a specific value constrained within the interval [1,5]; Section index: With section height, section span and section area as input values, the weight is automatically calculated according to the XGBoost algorithm to generate a comprehensive index value; Disturbance coefficient: It is selected according to the degree of influence shown near the monitoring point after blasting. The degree of influence is divided into five levels. Combined with the proportion of unfilled goaf in the detection point area, the XGBoost algorithm assigns a specific value in the range of [0,5]. The specific classification standard is: Level 1, lighter, complete rock mass, no cracks; Level 2: moderate, the rock mass is relatively intact, with basically no obvious cracks; Level 3: more serious, with obvious cracks, with a width of less than 1 cm; Level 4: severe, the rock mass is relatively broken, with crack widths greater than 1 cm; Level 5: very severe, the rock mass is broken and prone to spalling and roof collapse; Support index: According to the support type and specific support parameters, the weight is automatically calculated according to the XGBoost algorithm to generate a comprehensive index value. The support types are divided into no support, sprayed support, anchor mesh support, steel-concrete support, and steel support.

7. The multi-source ground pressure hazard level quantitative evaluation method based on XGBoost algorithm according to claim 1 is characterized in that: The magnitude, azimuth and inclination data of the maximum principal stress, intermediate principal stress and minimum principal stress with time-history characteristics obtained in step S4 are used to eliminate abnormal points according to the SPC abnormality judgment principle before use, and generate time-history curves of principal stress and principal stress difference, and radar charts of principal stress inclination and azimuth to assist in judging the dangerous level of ground pressure.

Images