A coal mine pressure multi-source data fusion intelligent early warning platform and method

Abstract

The present application provides a kind of coal mine pressure multi-source data fusion intelligent early warning platform and method, belong to coal mine pressure multi-source data fusion early warning technical field;To solve the technical problems that multi-source data is difficult to collaborative analysis in existing mine pressure monitoring, early warning linkage is insufficient, by arranging multi-source sensor and collecting hydraulic support pressure data, borehole stress data, roof separation data, surrounding rock displacement data and anchor rod / anchor cable stress data;Analysis and standardization processing are carried out to the collected data, including time alignment, abnormal value repair, missing value filling and filtering processing based on space-time weighting;Multi-source fusion analysis is carried out to the preprocessed data, including mine pressure change trend prediction, rock burst risk prediction and surrounding rock stability analysis;According to multi-index combination determination rule and space neighborhood propagation mechanism, hierarchical linkage early warning is carried out;Based on monitoring early warning platform, monitoring data, analysis results and early warning information are cooperatively displayed;The present application is applied to coal mine pressure early warning.

Description

Technical Field

[0001] This invention provides an intelligent early warning platform and method for coal mine pressure multi-source data fusion, belonging to the field of coal mine pressure multi-source data fusion early warning technology. Background Technology

[0002] With the advancement of intelligent coal mine construction, mine pressure monitoring technology has gradually evolved from manual observation and single-instrument analysis to multi-sensor online data acquisition, centralized transmission, and platform-based processing. Existing research and engineering applications show that joint monitoring of information such as hydraulic support working resistance, borehole stress, surrounding rock displacement, roof delamination, and anchor bolt / cable loads helps to understand the manifestation patterns of mine pressure at the working face, the deformation and failure process of the surrounding rock, and the characteristics of stress evolution. Simultaneously, platform-based technologies combining data acquisition, data processing, comprehensive identification, and visualization can improve the timeliness and accuracy of identifying mine pressure anomalies and rockburst hazards.

[0003] However, current mine pressure monitoring methods still have the following shortcomings: First, different monitoring systems have heterogeneous sources, and the data formats, sampling frequencies, and time scales are not uniform, making it difficult to standardize and collaboratively utilize multi-source data; second, the analysis of hydraulic support pressure, borehole stress, and surrounding rock deformation is usually independent, making it difficult to form a unified and comprehensive judgment result for the working face; third, existing early warning methods mostly use fixed thresholds or single indicator criteria, which do not adequately consider the temporal synchronization, spatial propagation, and multi-indicator coupling relationship of anomalies, making it easy to have false alarms, missed alarms, or delayed early warnings; fourth, existing platforms focus more on displaying monitoring results and lack a closed-loop mechanism that combines on-site handling suggestions, false alarm feedback, and model optimization, making it difficult to continuously improve the early warning effect.

[0004] Therefore, there is an urgent need to develop a coal mine pressure multi-source data fusion intelligent early warning platform and method to uniformly collect, standardize, analyze, and dynamically visualize multi-source monitoring data of the working face, and iteratively optimize the model and early warning strategy based on feedback information to meet the needs of coal mine working face safety monitoring and disaster early warning. Summary of the Invention

[0005] To address the technical problems existing in the background art, the present invention adopts the following technical solution: A coal mine pressure multi-source data fusion intelligent early warning platform is provided, comprising a multi-source data acquisition module, a data preprocessing module, a multi-source fusion analysis module, a hierarchical linkage early warning module, a dynamic visualization interaction module, and a feedback optimization module, wherein:

[0006] The multi-source data acquisition module is used to acquire mine pressure monitoring data, including hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data.

[0007] The data preprocessing module is used to parse and standardize the mine pressure monitoring data, including time alignment, outlier repair, missing value filling, and spatiotemporal weighted filtering.

[0008] The multi-source fusion analysis module is used for collaborative analysis of preprocessed data, including a support pressure prediction submodule, an advanced support stress prediction submodule, and a surrounding rock stability analysis submodule, wherein:

[0009] The support pressure prediction submodule is used to build a deep learning model based on time series data to predict the trend of mine pressure changes.

[0010] The advanced support stress prediction submodule is used to build a risk prediction model based on stress data to identify rockburst risks;

[0011] The surrounding rock stability analysis submodule is used to perform surrounding rock stability analysis based on roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable stress data, and outputs surrounding rock stability evaluation results and abnormal area distribution information;

[0012] The hierarchical linkage early warning module combines and judges the abnormal status of multiple monitoring indicators, determines the early warning level based on the number of abnormal indicators, the degree of abnormality and the time synchronization, and dynamically adjusts the early warning threshold of the neighboring monitoring points based on the spatial neighborhood propagation mechanism when the early warning is triggered.

[0013] The dynamic visualization interaction module is used to dynamically display monitoring data, analysis results, and early warning information;

[0014] The feedback optimization module is used to establish a feedback loop based on the confirmation of the early warning result and the labeling of false alarms, and to use an online incremental learning strategy to locally update the model parameters corresponding to the early warning error samples.

[0015] The multi-source data acquisition module is communicatively connected to the data preprocessing module. The output of the data preprocessing module is connected to the input of the multi-source fusion analysis module. The multi-source fusion analysis module is connected to the hierarchical linkage early warning module and the dynamic visualization interaction module, respectively. The feedback optimization module is connected to the multi-source fusion analysis module and the hierarchical linkage early warning module.

[0016] A method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure includes the following steps:

[0017] Step S1: Deploy multi-source sensors and collect hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data;

[0018] Step S2 involves parsing and standardizing the collected data, including time alignment, outlier repair, missing value imputation, and spatiotemporal weighted filtering.

[0019] Step S3: Perform multi-source fusion analysis on the preprocessed data, including prediction of mine pressure change trends, prediction of rockburst risk, and analysis of surrounding rock stability.

[0020] Step S4: Construct a hierarchical linkage early warning mechanism based on the results of multi-source data fusion: trigger a level 1 early warning when a single indicator exceeds the threshold, trigger a level 2 early warning when two or more indicators at the same location exceed the threshold, and trigger a level 3 early warning when multi-source data synchronization is abnormal.

[0021] The tiered linkage early warning system determines the warning level by comprehensively considering the risk value, the number of abnormal indicators, and the time synchronization. When a level 2 or higher warning is triggered, a spatial neighborhood propagation mechanism is introduced to adjust the threshold of monitoring points within a certain range of the warning point in order to simulate the stress disturbance propagation effect.

[0022] Step S5: Receive the raw monitoring data from step S1, the analysis results from step S3, and the early warning information from step S4; construct a dynamic visualization monitoring and early warning platform; display the raw monitoring data, analysis results, and early warning information in a multi-view linkage; and retrieve corresponding safety handling suggestions according to the early warning level.

[0023] Step S6: Establish the data interaction relationship between the feedback optimization module, the multi-source fusion analysis module, and the hierarchical linkage early warning module. Based on the early warning confirmation results and false alarm labeling results, establish a feedback closed loop and use an online incremental learning strategy to locally update the model parameters corresponding to the early warning error samples.

[0024] The specific method of step S2 includes:

[0025] Step S2-1: Parse the raw data into data containing timestamps, device IDs, monitoring values, and data quality identifiers, and perform quality assessment on the raw sampling points;

[0026] Step S2-2: Perform unified time alignment on monitoring data with different sampling frequencies;

[0027] Step S2-3: Identify and repair outliers in the time-aligned monitoring sequence;

[0028] Step S2-4: Perform spatiotemporal joint imputation on missing monitoring values;

[0029] Step S2-5: Perform spatiotemporal weighted filtering on the monitoring data after missing values ​​are filled to obtain the final standardized monitoring values.

[0030] The specific method for step S2-2 is as follows:

[0031] Let the i-th type of monitoring data be at time i. The original monitoring value is At the time of unified alignment The alignment value at the location is denoted as The calculation formula is:

[0032] ;

[0033] in, This represents the data quality coefficient corresponding to the original sampling points, and ; For the first The time weighting parameter of the monitoring data, and ; To ensure uniform alignment within the window The number of valid sampling points for this type of monitoring data;

[0034] The length of the unified alignment window is determined based on the minimum sampling period or preset sampling period of various monitoring data. When there are no valid sampling points in the unified alignment window, the monitoring value corresponding to that moment is marked as a missing value, and the missing value is filled in step S2-4.

[0035] The specific method for step S2-3 is as follows:

[0036] Let the first Class of monitoring data in the first The alignment value at each alignment moment is In length of Calculating the local median within a sliding window The expression is:

[0037] ;

[0038] in, ,and ;

[0039] If the following formula is satisfied:

[0040] ;

[0041] Then determine These are outliers, where... For the first The threshold for judging anomalies in monitoring data is determined based on historical sample statistical results or on-site calibration results.

[0042] For the outliers identified, a joint repair method using temporal neighborhood and local statistics is employed, resulting in repaired values. The calculation formula is:

[0043] ;

[0044] in, For temporal neighborhood repair weights, and ;

[0045] when When the data is located at the boundary of the sequence and there is only data from one adjacent time point, the mean of the time neighborhood is replaced by the data from one adjacent time point; the sliding window length adopts an odd number of windows and is set according to the sampling frequency and fluctuation characteristics of different monitoring indicators.

[0046] The specific method for step S2-4 is as follows:

[0047] Let the first The monitoring point at the 1st The missing value imputation result at each alignment time point is The calculation formula is:

[0048] ;

[0049] in, For time neighborhood compensation weights, and ; For monitoring points A set of spatial neighborhood monitoring points; For monitoring points With monitoring points The spatial distance between them, and ; It is the spatial distance decay exponent, and ; For neighborhood monitoring points In the Post-alignment monitoring values ​​at each alignment moment;

[0050] When the k-th alignment time is located at the boundary of the monitoring sequence, or when there are only valid monitoring values ​​of one-sided adjacent time, the time neighborhood mean term is replaced by the valid monitoring values ​​of the one-sided adjacent time.

[0051] When the spatial neighborhood set When the value is empty, only the time neighborhood item is used to fill in the missing value; when there are no valid values ​​in both the time neighborhood and the spatial neighborhood, the monitoring value is left empty and marked as a low confidence state, to be updated at a later time.

[0052] The specific method for step S2-5 is as follows:

[0053] Final standardized monitoring value The calculation formula is:

[0054] ;

[0055] in, The weighting coefficient for the monitored value at the current moment. Let be the weighting coefficient of the time smoothing term, and satisfy . , , The weight of the spatial weighting term is ; For the first The filtered value of each monitoring point at the previous alignment time; the meanings of the other symbols are the same as before;

[0056] When the spatial neighborhood set If it is empty, Then for and Normalize proportionally; if This will make the final standardized monitoring value .

[0057] The specific method of step S3 includes:

[0058] Step S3-1: Construct a deep learning prediction model based on the hydraulic support pressure time series data. The deep learning prediction model includes network units for extracting local features and attention units for extracting global temporal features. Train the model with the goal of minimizing the prediction error. Construct anomaly judgment rules and set early warning thresholds. Output the trend value of mine pressure change or anomaly probability within a preset time window in the future.

[0059] Step S3-2: Construct a stress-push distance relationship curve based on borehole stress gauge data, and identify the location of peak stress and the range of stress concentration area by combining elastoplastic theory; construct a rockburst risk prediction model using data augmentation algorithm and machine learning classification model; construct anomaly judgment rules and set early warning thresholds, and output the rockburst risk level or risk probability.

[0060] Step S3-3: Construct a rock deformation characteristic index system based on roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable stress data; generate a dynamic heat map using the roadway plan as the base map; establish deformation-advance distance curves and deformation-time curves respectively; extract surrounding rock deformation characteristic parameters such as delamination amount, displacement rate, and stress change rate; construct anomaly judgment rules and set early warning thresholds; and output the surrounding rock stability evaluation results and anomaly area distribution information.

[0061] In the hierarchical linkage early warning mechanism in step S4, different early warning levels are determined by a comprehensive assessment of multiple indicators, including the number of indicators, the degree of anomaly, and time synchronization. To eliminate the influence of differences in the dimensions of different monitoring indicators, the degree of anomaly of each monitoring indicator is first normalized.

[0062] Let the first The normalized anomaly level of each monitoring indicator at the current moment is: ,and Then the overall risk value Calculate using the following formula:

[0063] ;

[0064] in, The normalized version The degree of abnormality of each monitoring indicator, and meets the following requirements. ; For the first The weight coefficients corresponding to each monitoring indicator, and satisfying the following conditions: , ; This is the time synchronization coefficient, and ; Let n be the number of monitoring indicators currently in an abnormal state, and n be the total number of monitoring indicators participating in the current comprehensive judgment. ; , This is a correction factor;

[0065] Among them, the time synchronization coefficient Used to characterize multiple monitoring indicators within a preset synchronization time window The degree of synchronization anomaly within; when the abnormal indicator is in When internal height synchronization occurs, Take the larger value; when abnormal indicators appear over a period of time, Take the smaller value;

[0066] The weighting coefficient and correction factor , Obtained through calibration using historical mine pressure samples;

[0067] Based on the comprehensive risk value The warning level is determined by comparing the result with the preset level threshold.

[0068] In the spatial neighborhood propagation mechanism in step S4, a spatial neighborhood model is constructed with the warning point as the center. The warning threshold of each monitoring point within the neighborhood is dynamically adjusted according to its spatial distance from the warning point, and the adjustment range of the threshold is negatively correlated with the spatial distance.

[0069] Let the baseline warning threshold be , No. The spatial distance between each neighboring monitoring point and the early warning point is The dynamic early warning threshold for that monitoring point is... satisfy:

[0070] ;

[0071] in, Let be the distance attenuation coefficient, and ; This is the threshold adjustment coefficient, and ; ;

[0072] The closer a monitoring point is to a warning point, the lower its corresponding dynamic warning threshold; as the distance between monitoring points increases, the dynamic warning threshold gradually approaches the baseline warning threshold. ;

[0073] The parameters and Obtained through calibration using historical mining pressure propagation samples.

[0074] The beneficial effects of this invention compared to existing technologies are as follows: This invention provides a coal mine pressure multi-source data fusion intelligent early warning platform. This platform collects hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data. After time alignment, anomaly repair, missing data filling, and spatiotemporal weighted filtering, it performs mine pressure change trend prediction, rockburst risk identification, and surrounding rock stability evaluation. It also combines multi-index combination judgment rules and spatial neighborhood propagation mechanisms to achieve hierarchical linkage early warning, realizing dynamic display and feedback optimization of monitoring data, analysis results, and early warning information. This invention achieves deep fusion and intelligent analysis of multi-source mine pressure data by integrating multi-source monitoring data from different sensors, improving the accuracy and timeliness of early warning, constructing a hierarchical early warning system from single indicators to multi-source linkage, enhancing the comprehensive disaster assessment capability, and improving the intuitiveness and interactivity of monitoring information through dynamic visualization. It also supports closed-loop management of early warning and push of safety disposal suggestions, which is conducive to improving the comprehensive application effect of safety monitoring and early warning in coal mine working faces. Attached Figure Description

[0075] The present invention will be further described below with reference to the accompanying drawings:

[0076] Figure 1 This is a flowchart illustrating the steps of the intelligent early warning method for coal mine pressure multi-source data fusion according to the present invention.

[0077] Figure 2 This is a schematic diagram of the structure of the intelligent early warning platform for multi-source data fusion of coal mine pressure according to the present invention;

[0078] Figure 3 This is a diagram illustrating the layout of the measuring stations at the working face in an embodiment of the present invention.

[0079] Figure 4 This is a diagram illustrating the data transmission effect of the platform in an embodiment of the present invention. Detailed Implementation

[0080] like Figure 2As shown, this invention provides a coal mine pressure multi-source data fusion intelligent early warning platform, including a multi-source data acquisition module, a data preprocessing module, a multi-source fusion analysis module, a hierarchical linkage early warning module, a dynamic visualization interaction module, and a feedback optimization module, wherein:

[0081] The multi-source data acquisition module is used to acquire mine pressure monitoring data, including hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data.

[0082] The data preprocessing module is used to parse and standardize the mine pressure monitoring data, including time alignment, outlier repair, missing value filling, and spatiotemporal weighted filtering.

[0083] The multi-source fusion analysis module is used for collaborative analysis of preprocessed data, including a support pressure prediction submodule, an advanced support stress prediction submodule, and a surrounding rock stability analysis submodule, wherein:

[0084] The support pressure prediction submodule is used to build a deep learning model based on time series data to predict the trend of mine pressure changes.

[0085] The advanced support stress prediction submodule is used to build a risk prediction model based on stress data to identify rockburst risks;

[0086] The surrounding rock stability analysis submodule is used to perform surrounding rock stability analysis based on roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable stress data, and outputs surrounding rock stability evaluation results and abnormal area distribution information;

[0087] The hierarchical linkage early warning module combines and judges the abnormal status of multiple monitoring indicators, determines the early warning level based on the number of abnormal indicators, the degree of abnormality and the time synchronization, and dynamically adjusts the early warning threshold of the neighboring monitoring points based on the spatial neighborhood propagation mechanism when the early warning is triggered.

[0088] The dynamic visualization interaction module is used to dynamically display monitoring data, analysis results, and early warning information;

[0089] The feedback optimization module is used to establish a feedback loop based on the confirmation of the early warning result and the labeling of false alarms, and to use an online incremental learning strategy to locally update the model parameters corresponding to the early warning error samples.

[0090] The multi-source data acquisition module is communicatively connected to the data preprocessing module. The output of the data preprocessing module is connected to the input of the multi-source fusion analysis module. The multi-source fusion analysis module is connected to the hierarchical linkage early warning module and the dynamic visualization interaction module, respectively. The feedback optimization module is connected to the multi-source fusion analysis module and the hierarchical linkage early warning module.

[0091] like Figure 1 As shown, based on the aforementioned early warning platform, this invention also provides an intelligent early warning method for coal mine pressure multi-source data fusion, the specific scheme of which is as follows:

[0092] Step S1: Deploy multi-source sensors and collect hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data;

[0093] Furthermore, step S1 includes the following sub-steps:

[0094] Step S1-1: Install pressure sensors, borehole stress gauges, top plate delamination gauges, anchor bolt / anchor cable force gauges, and displacement monitoring devices, and complete the connection configuration of the corresponding acquisition nodes.

[0095] Steps S1-2: After completing the hardware connection, perform single-point debugging and group debugging to ensure that the data collected by each monitoring device can be stably transmitted to the data processing unit.

[0096] Step S2 involves parsing and standardizing the collected data, including time alignment, outlier repair, missing value imputation, and spatiotemporal weighted filtering.

[0097] Furthermore, step S2 includes the following sub-steps:

[0098] Step S2-1: Parse the raw data into data containing timestamps, device IDs, monitoring values, and data quality identifiers; based on the sensor's online status, communication integrity, range consistency, and physical range constraints, evaluate the quality of the raw sampling points, and mark invalid points as those with abnormal timestamps, duplicate sampling, out-of-range values, invalid values ​​caused by communication interruptions, and obviously distorted code data; the invalid points do not directly participate in the subsequent time alignment weighted calculation, but their time positions can be retained for subsequent missing value determination and filling.

[0099] Step S2-2: Perform unified time alignment on monitoring data with different sampling frequencies. Let the i-th type of monitoring data be at time... The original monitoring value is At the time of unified alignment The alignment value at the location is denoted as The calculation formula is as follows:

[0100] ;

[0101] in, This represents the data quality coefficient corresponding to the original sampling points, and ; For the first The time weighting parameter of the monitoring data, and ; To ensure uniform alignment within the window The number of valid sampling points for this type of monitoring data;

[0102] The length of the unified alignment window is determined based on the minimum sampling period or preset sampling period of various types of monitoring data.

[0103] By using the above method, sampling points that are closer to the unified alignment time and have higher data quality are given greater weight, thereby obtaining alignment monitoring values ​​at the unified time node;

[0104] When there are no valid sampling points within the unified alignment window, the monitoring value corresponding to that moment is marked as a missing value, and the process proceeds to step S2-4 to fill in the missing value.

[0105] Step S2-3 involves outlier identification and repair in the time-aligned monitoring sequence. Let the first... Class of monitoring data in the first The alignment value at each alignment moment is In length of Calculating the local median within a sliding window Its expression is:

[0106] ;

[0107] in, ,and ;

[0108] If the following formula is satisfied:

[0109] ;

[0110] Then determine These are outliers, where... For the first The threshold for judging anomalies in monitoring data is determined based on historical sample statistical results or on-site calibration results.

[0111] For the outliers identified, a joint repair method using temporal neighborhood and local statistics is employed, resulting in repaired values. Calculate using the following formula:

[0112] ;

[0113] in, For temporal neighborhood repair weights, and ;

[0114] when When the time neighborhood mean is replaced by the data of only one adjacent time when it is located at the boundary of the sequence;

[0115] The sliding window length is an odd number of windows, and is set according to the sampling frequency and fluctuation characteristics of different monitoring indicators.

[0116] By using the above method, the continuity of the monitoring sequence can be maintained by utilizing data from adjacent time points, while the impact of abnormal spikes on subsequent analysis results can be reduced by using local medians.

[0117] Step S2-4: Perform spatiotemporal joint imputation on missing monitoring values. Let the first... The monitoring point at the 1st The missing value imputation result at each alignment time point is Then it is calculated using the following formula:

[0118] ;

[0119] in, For time neighborhood compensation weights, and ; For monitoring points A set of spatial neighborhood monitoring points; For monitoring points With monitoring points The spatial distance between them, and ; It is the spatial distance decay exponent, and ; For neighborhood monitoring points In the Post-alignment monitoring values ​​at each alignment moment;

[0120] When the k-th alignment time is located at the boundary of the monitoring sequence, or when there are only valid monitoring values ​​of one-sided adjacent time, the time neighborhood mean term is replaced by the valid monitoring values ​​of the one-sided adjacent time.

[0121] When the spatial neighborhood set When the value is empty, only the time neighborhood item is used to fill in the missing value; when there are no valid values ​​in both the time neighborhood and the spatial neighborhood, the monitoring value is left empty and marked as a low confidence state, to be updated at subsequent times.

[0122] By using the above methods, both temporal and spatial proximity information are utilized to compensate for missing data, thereby improving the stability and reliability of the missing value imputation results.

[0123] Step S2-5: Perform spatiotemporal weighted filtering on the monitoring data after missing value imputation to obtain the final standardized monitoring values. The calculation formula is as follows:

[0124] ;

[0125] in, The weighting coefficient for the monitored value at the current moment. Let be the weighting coefficient of the time smoothing term, and satisfy . , , The weight of the spatial weighting term is ; For the first The filtered value of each monitoring point at the previous alignment time; the meanings of the other symbols are the same as before.

[0126] When the spatial neighborhood set If it is empty, Then for and Normalize proportionally; if This will make the final standardized monitoring value .

[0127] By employing the above methods, random fluctuations are suppressed in the time dimension and local disturbances are reduced in the spatial dimension, thereby obtaining standardized monitoring data suitable for subsequent multi-source fusion analysis.

[0128] After processing through steps S2-2 to S2-5, a standardized mining pressure monitoring dataset under a unified time benchmark is obtained and stored in a database for subsequent prediction of mining pressure change trends, prediction of rockburst risk, and analysis of surrounding rock stability.

[0129] Wherein, the data quality coefficient The time weight parameter is determined comprehensively based on the online status, communication integrity, range consistency, and physical range consistency of the original sampling points; Temporal Neighborhood Repair Weight Temporal neighborhood compensation weight Spatial distance attenuation index The filter weight coefficients are determined based on historical mine pressure samples, on-site monitoring patterns, and parameter calibration results.

[0130] Step S3: Perform multi-source fusion analysis on the preprocessed data, including prediction of mine pressure change trends, prediction of rockburst risk, and analysis of surrounding rock stability.

[0131] Furthermore, step S3 includes the following sub-steps:

[0132] Step S3-1: Construct a mine pressure change trend prediction model based on hydraulic support pressure time series data; slice the preprocessed hydraulic support pressure monitoring sequence according to a preset time window to construct input samples, and extract time series features characterizing the mine pressure change state; the prediction model includes at least a local time series feature extraction unit, a global correlation feature extraction unit, and a prediction output unit; train the prediction model using historical hydraulic support pressure samples; determine anomalies based on a preset threshold, and output the mine pressure change trend value and anomaly probability within the future preset time window.

[0133] Step S3-2: Construct a rockburst risk prediction model based on borehole stress gauge data; correlate borehole stress monitoring values ​​with working face advance positions to establish a "stress-advance distance" relationship curve, and extract one or more features from peak stress, peak stress location, stress gradient, stress concentration zone width, and stress change rate as risk identification features; perform data augmentation on historical risk samples, and input the augmented features into a machine learning classification model to output the rockburst risk probability and risk level.

[0134] Step S3-3: Construct a rock deformation characteristic index system based on roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable stress data; generate a dynamic heat map using the roadway plan as the base map, and establish "deformation amount-advance distance" curves and "deformation amount-time" curves respectively; extract surrounding rock deformation characteristic parameters such as delamination amount, displacement rate, and stress change rate to form a rock stability evaluation feature set; determine anomalies based on preset thresholds, and output the rock stability state level and anomaly area distribution information.

[0135] Step S4: Implement graded linkage early warning based on multi-indicator combination judgment rules and spatial neighborhood propagation mechanism. The early warning level is determined based on the number of abnormal indicators, the degree of abnormality, and the time synchronization. When a level II or higher early warning is triggered, the early warning threshold of the neighborhood monitoring point is dynamically adjusted.

[0136] Furthermore, step S4 includes the following sub-steps:

[0137] Step S4-1: Construct a hierarchical linkage early warning mechanism based on the results of multi-source data fusion: trigger a level 1 early warning when a single indicator exceeds the threshold; trigger a level 2 early warning when two or more indicators at the same location exceed the threshold; and trigger a level 3 early warning when multi-source data synchronization is abnormal.

[0138] Step S4-2, the graded linkage early warning determines the early warning level by comprehensively considering the risk value, the number of abnormal indicators and time synchronization; when a level II or higher early warning is triggered, a spatial neighborhood propagation mechanism is introduced to adjust the threshold of monitoring points within a certain range of the early warning point in order to simulate the stress disturbance propagation effect.

[0139] Furthermore, in the hierarchical linkage early warning mechanism described in step S4-1, different early warning levels are determined comprehensively based on multiple indicators, including the number of indicators, the degree of anomaly, and time synchronization. To eliminate the influence of differences in the dimensions of different monitoring indicators, the degree of anomaly of each monitoring indicator is first normalized; let the first... The normalized anomaly level of each monitoring indicator at the current moment is: ,and Then the overall risk value Calculate using the following formula:

[0140] ;

[0141] in, The normalized version The degree of abnormality of each monitoring indicator, and meets the following requirements. ; For the first The weight coefficients corresponding to each monitoring indicator, and satisfying the following conditions: , ; This is the time synchronization coefficient, and ; Let n be the number of monitoring indicators currently in an abnormal state, and n be the total number of monitoring indicators participating in the current comprehensive judgment. ; , This is a correction factor;

[0142] Among them, the time synchronization coefficient Used to characterize multiple monitoring indicators within a preset synchronization time window The degree of synchronization anomaly within; when the abnormal indicator is in When internal height synchronization occurs, Take the larger value; when abnormal indicators appear over a period of time, Take the smaller value;

[0143] The weighting coefficient and correction factor , Obtained through calibration using historical mine pressure samples;

[0144] Based on the comprehensive risk value The warning level is determined by comparing the result with the preset level threshold;

[0145] Furthermore, in the spatial neighborhood propagation mechanism described in step S4-2, a spatial neighborhood model is constructed centered on the warning point. The warning threshold of each monitoring point within the neighborhood is dynamically adjusted based on its spatial distance from the warning point, and the adjustment range of the threshold is negatively correlated with the spatial distance. Let the baseline warning threshold be... , No. The spatial distance between each neighboring monitoring point and the early warning point is The dynamic early warning threshold for that monitoring point is... satisfy:

[0146] ;

[0147] in, Let be the distance attenuation coefficient, and ; This is the threshold adjustment coefficient, and ; ;

[0148] As shown in the above formula, the closer the monitoring point is to the warning point, the lower its corresponding dynamic warning threshold; as the distance between the monitoring points increases, the dynamic warning threshold gradually approaches the baseline warning threshold. ;

[0149] The parameters and Obtained through calibration using historical mine pressure propagation samples;

[0150] The above methods can improve the sensitivity to potential abnormal expansion processes in the vicinity of the warning point.

[0151] Step S5: Construct a dynamic visualization monitoring and early warning platform to collaboratively display monitoring data, analysis results, and early warning information;

[0152] Furthermore, step S5 includes the following sub-steps:

[0153] Step S5-1: Receive the raw monitoring data from step S1, the analysis results from step S3, and the early warning information from step S4, and construct a dynamic visualization monitoring and early warning platform.

[0154] Step S5-2: Display the raw monitoring data, analysis results and early warning information in multiple views, and retrieve corresponding safety handling suggestions according to the early warning level.

[0155] Step S6: Establish a feedback loop through early warning confirmation and false alarm labeling, and iteratively optimize the model;

[0156] Furthermore, step S6 includes the following sub-steps:

[0157] Step S6-1: Establish the data interaction relationship between the feedback optimization module, the multi-source fusion analysis module, and the hierarchical linkage early warning module. The feedback optimization module receives the original monitoring data, preprocessed standardized data, multi-source fusion analysis results, early warning level, on-site handling records, and manual review results corresponding to the early warning trigger time; and marks the feedback samples as real early warning samples, false alarm samples, or samples to be confirmed based on the manual review results.

[0158] Step S6-2: For the marked real warning samples and false alarm samples, construct an incremental update sample set; based on the incremental update sample set, locally update the prediction model parameters, classification boundary parameters, and warning threshold parameters in the multi-source fusion analysis module; after the model or threshold is updated, use reserved verification samples or the most recent measured samples to verify the update results. When the updated results meet the preset accuracy requirements, the updated parameters are used for subsequent real-time warnings; otherwise, the parameters before the update are retained.

[0159] To better understand the above-mentioned objectives, features and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0160] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0161] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and specific examples.

[0162] Example: The 10103 longwall face of a certain mine;

[0163] like Figures 2 to 4 As shown, the intelligent early warning platform and method for coal mine pressure multi-source data fusion provided by the present invention is carried out according to the following steps:

[0164] Step 1: Based on the collected mine data, the 10103 longwall mining face has a dip length of 220m and a strike length of 1519m. The average thickness of the No. 10 coal seam is 3.18m, containing 1-3 layers of interbedded rock. It employs a strike-bound longwall fully mechanized longwall mining method with one-pass full-height extraction. The goaf is managed using the complete caving method. The transport roadway of the working face is excavated along the coal seam roof, with a rectangular cross-section, a net width of 4.7m, and a net height of 3.5m, using a combined anchor-mesh-beam-cable support system. Its roof is mainly composed of mudstone, sandy mudstone, and multiple layers of limestone.

[0165] Step 2, Station Layout: 147 hydraulic supports are arranged on the 10103 working face, each equipped with one pressure sensor. A fixed measuring point is set up for every 10 supports, resulting in 15 measuring lines. These lines are located on supports #6, #16, #26, #36, #46, #56, #66, #76, #86, #96, #106, #116, #126, #136, and #146. The transport roadway is 1200m long. In the 750m section, one set of monitoring stations is set up every 50m. Each set of stations includes one roof separation meter, one roof anchor cable dynamometer, one roof anchor bolt dynamometer, one roadway surface displacement monitoring point, and one anchor bolt dynamometer on each side. In the transport roadway, 100m from the working face cut, borehole stress gauges with lengths of 3m, 5m, 9m, and 12m are placed every 2m on the outer side, and two borehole stress gauges with a length of 5m are placed every 2m on the inner side. The working face monitoring station layout is as follows. Figure 3 As shown.

[0166] Step 3: After completing the station setup, collect hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data. The collected raw data is then uniformly parsed into data records containing timestamps, equipment identifiers, monitoring values, and data quality identifiers. Subsequently, the data undergoes time alignment, outlier repair, missing value imputation, and spatiotemporal weighted filtering to obtain standardized data suitable for subsequent analysis. This step corresponds to the multi-source data acquisition and data preprocessing process in the method of this invention.

[0167] Step 4: Based on the preprocessed multi-source data, conduct mine pressure change trend prediction, rockburst risk identification, and surrounding rock stability evaluation, and output abnormal area distribution information. Specifically, a mine pressure change trend prediction model is constructed for hydraulic support pressure time series data. The preprocessed hydraulic support pressure monitoring sequence is sliced ​​according to a preset time window to construct input samples, and temporal features characterizing the mine pressure change state are extracted. The prediction model includes a local temporal feature extraction unit, a global correlation feature extraction unit, and a prediction output unit. The prediction model is trained using historical hydraulic support pressure samples. Anomalies are determined based on preset thresholds, and the mine pressure change trend value and anomaly probability within the future preset time window are output. A rockburst risk prediction model is constructed for borehole stress gauge data. The borehole stress monitoring values ​​are correlated with the working face advancement position to establish a "stress-advance distance" relationship curve, and peak stress, peak stress location, stress concentration zone width, and stress change rate are extracted as risk identification features. Historical risk samples are data augmented, and the augmented features are input into a machine learning classification model to output the rockburst risk probability and risk level. A rock deformation characteristic index system is constructed based on roof delamination data, surrounding rock displacement data, and anchor bolt / cable stress data. A dynamic heat map is generated using the roadway plan as the base map, and "deformation-advance distance" curves and "deformation-time" curves are established. Rock deformation characteristic parameters such as delamination amount, displacement rate, and stress change rate are extracted to form a rock stability evaluation feature set. Anomalies are determined based on preset thresholds, and the rock stability state level and anomaly area distribution information are output. This step corresponds to the multi-source fusion analysis process in the method of this invention.

[0168] Step 5: Based on the multi-source analysis results, construct a hierarchical linkage early warning mechanism. A Level 1 early warning is triggered when a single indicator anomaly occurs at a monitoring point on the underground working face; a Level 2 early warning is triggered when two or more indicators at the same location are anomaly; and a Level 3 early warning is triggered when multiple monitoring indicators show synchronous anomalies. When a Level 2 or higher early warning is triggered, the threshold values ​​of neighboring monitoring points are dynamically adjusted using a spatial neighborhood propagation mechanism to simulate the stress disturbance propagation effect, thereby improving the ability to identify the process of local anomalies spreading to the neighborhood. This step corresponds to the hierarchical linkage early warning process in the method of this invention.

[0169] Step Six: In this embodiment, a dynamic visualization monitoring and early warning platform is constructed to collaboratively display raw monitoring data, analysis results, and early warning information. As one implementation method, the platform front-end uses a visualization framework to achieve graphical interactive display, while the back-end provides data management, analysis result retrieval, and early warning information services. When an operator clicks on a corresponding monitoring point, the real-time data, historical curves, analysis results, and current early warning status of that point can be displayed in conjunction with the warning level, and corresponding safety handling suggestions can be retrieved based on the warning level. This step corresponds to the dynamic visualization interaction process in the method of this invention; platform data transmission is as follows... Figure 4 As shown.

[0170] Step 7: Further, the on-site confirmation information of the early warning results and the false alarm labeling information are transmitted back to the platform to establish a feedback loop. Based on the feedback samples, the model parameters and early warning threshold parameters are locally updated to achieve continuous optimization of the early warning model and early warning strategy. Specifically, the feedback optimization module receives the original monitoring data, preprocessed standardized data, multi-source fusion analysis results, early warning level, on-site handling records, and manual review results corresponding to the early warning trigger time. Based on the manual review results, the feedback samples are marked as real early warning samples, false alarm samples, or samples awaiting confirmation. For the marked real early warning samples and false alarm samples, an incremental update sample set is constructed. Based on the incremental update sample set, the prediction model parameters, classification boundary parameters, and early warning threshold parameters in the multi-source fusion analysis module are locally updated. After the model or threshold is updated, the update results are verified using reserved verification samples or the most recent measured samples. When the updated results meet the preset accuracy requirements, the updated parameters are used for subsequent real-time early warnings; otherwise, the parameters before the update are retained. This step corresponds to the feedback optimization process in the method of this invention.

[0171] Step 8: Field application results show that, combined with measured data from the operation of the 10103 longwall face, the hierarchical linkage early warning method of this invention can identify abnormal working conditions 1 to 12 hours in advance, with an early warning accuracy rate of up to 88%. On-site, timely corresponding measures are taken based on the early warning information. Compared with conventional methods, this method improves the prediction accuracy of the periodic pressure step distance of the working face by 21% and reduces the support failure rate during the periodic pressure step by 23%. This early warning method provides practical technical support for safe production and on-site scheduling decisions throughout the entire longwall face mining process of the 10103 working face.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A coal mine pressure multi-source data fusion intelligent early warning platform, characterized in that, It includes a multi-source data acquisition module, a data preprocessing module, a multi-source fusion analysis module, a hierarchical linkage early warning module, a dynamic visualization interaction module, and a feedback optimization module, among which: The multi-source data acquisition module is used to acquire mine pressure monitoring data, including hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data. The data preprocessing module is used to parse and standardize the mine pressure monitoring data, including time alignment, outlier repair, missing value filling, and spatiotemporal weighted filtering. The multi-source fusion analysis module is used for collaborative analysis of preprocessed data, including a support pressure prediction submodule, an advanced support stress prediction submodule, and a surrounding rock stability analysis submodule, wherein: The support pressure prediction submodule is used to build a deep learning model based on time series data to predict the trend of mine pressure changes. The advanced support stress prediction submodule is used to build a risk prediction model based on stress data to identify rockburst risks; The surrounding rock stability analysis submodule is used to perform surrounding rock stability analysis based on roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable stress data, and outputs surrounding rock stability evaluation results and abnormal area distribution information; The hierarchical linkage early warning module combines and judges the abnormal status of multiple monitoring indicators, determines the early warning level based on the number of abnormal indicators, the degree of abnormality and the time synchronization, and dynamically adjusts the early warning threshold of the neighboring monitoring points based on the spatial neighborhood propagation mechanism when the early warning is triggered. The dynamic visualization interaction module is used to dynamically display monitoring data, analysis results, and early warning information; The feedback optimization module is used to establish a feedback loop based on the confirmation of the early warning result and the labeling of false alarms, and to use an online incremental learning strategy to locally update the model parameters corresponding to the early warning error samples. The multi-source data acquisition module is communicatively connected to the data preprocessing module. The output of the data preprocessing module is connected to the input of the multi-source fusion analysis module. The multi-source fusion analysis module is connected to the hierarchical linkage early warning module and the dynamic visualization interaction module, respectively. The feedback optimization module is connected to the multi-source fusion analysis module and the hierarchical linkage early warning module.

2. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 1, characterized in that, Includes the following steps: Step S1: Deploy multi-source sensors and collect hydraulic support pressure data, borehole stress data, roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable force data; Step S2 involves parsing and standardizing the collected data, including time alignment, outlier repair, missing value imputation, and spatiotemporal weighted filtering. Step S3: Perform multi-source fusion analysis on the preprocessed data, including prediction of mine pressure change trends, prediction of rockburst risk, and analysis of surrounding rock stability. Step S4: Construct a hierarchical linkage early warning mechanism based on the results of multi-source data fusion: trigger a level 1 early warning when a single indicator exceeds the threshold, trigger a level 2 early warning when two or more indicators at the same location exceed the threshold, and trigger a level 3 early warning when multi-source data synchronization is abnormal. The tiered linkage early warning system determines the warning level by comprehensively considering the risk value, the number of abnormal indicators, and the time synchronization. When a level 2 or higher warning is triggered, a spatial neighborhood propagation mechanism is introduced to adjust the threshold of monitoring points within a certain range of the warning point in order to simulate the stress disturbance propagation effect. Step S5: Receive the raw monitoring data from step S1, the analysis results from step S3, and the early warning information from step S4; construct a dynamic visualization monitoring and early warning platform; display the raw monitoring data, analysis results, and early warning information in a multi-view linkage; and retrieve corresponding safety handling suggestions according to the early warning level. Step S6: Establish the data interaction relationship between the feedback optimization module, the multi-source fusion analysis module, and the hierarchical linkage early warning module. Based on the early warning confirmation results and false alarm labeling results, establish a feedback closed loop and use an online incremental learning strategy to locally update the model parameters corresponding to the early warning error samples.

3. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 2, characterized in that: The specific method of step S2 includes: Step S2-1: Parse the raw data into data containing timestamps, device IDs, monitoring values, and data quality identifiers, and perform quality assessment on the raw sampling points; Step S2-2: Perform unified time alignment on monitoring data with different sampling frequencies; Step S2-3: Identify and repair outliers in the time-aligned monitoring sequence; Step S2-4: Perform spatiotemporal joint imputation on missing monitoring values; Step S2-5: Perform spatiotemporal weighted filtering on the monitoring data after missing values ​​are filled to obtain the final standardized monitoring values.

4. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure as described in claim 3, characterized in that: The specific method for step S2-2 is as follows: Let the i-th type of monitoring data be at time i. The original monitoring value is At the time of unified alignment The alignment value at the location is denoted as The calculation formula is: ； in, This represents the data quality coefficient corresponding to the original sampling points, and ; For the first The time weighting parameter of the monitoring data, and ; To ensure uniform alignment within the window The number of valid sampling points for this type of monitoring data; The length of the unified alignment window is determined based on the minimum sampling period or preset sampling period of various monitoring data. When there are no valid sampling points in the unified alignment window, the monitoring value corresponding to that moment is marked as a missing value, and the missing value is filled in step S2-4.

5. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 4, characterized in that: The specific method for step S2-3 is as follows: Let the first Class of monitoring data in the first The alignment value at each alignment moment is In length of Calculating the local median within a sliding window The expression is: ； in, ,and ; If the following formula is satisfied: ； Then determine These are outliers, where... For the first The threshold for judging anomalies in monitoring data is determined based on historical sample statistical results or on-site calibration results. For the outliers identified, a joint repair method using temporal neighborhood and local statistics is employed, resulting in repaired values. The calculation formula is: ； in, For temporal neighborhood repair weights, and ; when When the data is located at the boundary of the sequence and there is only data from one adjacent time point, the mean of the time neighborhood is replaced by the data from the one adjacent time point. The sliding window length is an odd number of windows, and is set according to the sampling frequency and fluctuation characteristics of different monitoring indicators.

6. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 5, characterized in that: The specific method for step S2-4 is as follows: Let the first The monitoring point at the 1st The missing value imputation result at each alignment time point is The calculation formula is: ； in, For time neighborhood compensation weights, and ; For monitoring points A set of spatial neighborhood monitoring points; For monitoring points With monitoring points The spatial distance between them, and ; It is the spatial distance decay exponent, and ; For neighborhood monitoring points In the Post-alignment monitoring values ​​at each alignment moment; When the k-th alignment time is located at the boundary of the monitoring sequence, or when there are only valid monitoring values ​​of one-sided adjacent time, the time neighborhood mean term is replaced by the valid monitoring values ​​of the one-sided adjacent time. When the spatial neighborhood set When the value is empty, only the time neighborhood item is used to fill in the missing value; when there are no valid values ​​in both the time neighborhood and the spatial neighborhood, the monitoring value is left empty and marked as a low confidence state, to be updated at a later time.

7. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 6, characterized in that: The specific method for step S2-5 is as follows: Final standardized monitoring value The calculation formula is: ； in, The weighting coefficient for the monitored value at the current moment. Let be the weighting coefficient of the time smoothing term, and satisfy . , , The weight of the spatial weighting term is ; For the first The filtered value of each monitoring point at the previous alignment time; the meanings of the other symbols are the same as before; When the spatial neighborhood set If it is empty, Then for and Normalize proportionally; if This will make the final standardized monitoring value .

8. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 7, characterized in that: The specific method of step S3 includes: Step S3-1: Construct a deep learning prediction model based on the hydraulic support pressure time series data. The deep learning prediction model includes network units for extracting local features and attention units for extracting global temporal features. Train the model with the goal of minimizing the prediction error. Construct anomaly judgment rules and set early warning thresholds. Output the trend value of mine pressure change or anomaly probability within a preset time window in the future. Step S3-2: Construct a stress-push distance relationship curve based on borehole stress gauge data, and identify the location of peak stress and the range of stress concentration area by combining elastoplastic theory; construct a rockburst risk prediction model using data augmentation algorithm and machine learning classification model; construct anomaly judgment rules and set early warning thresholds, and output the rockburst risk level or risk probability. Step S3-3: Construct a rock deformation characteristic index system based on roof delamination data, surrounding rock displacement data, and anchor bolt / anchor cable stress data; generate a dynamic heat map using the roadway plan as the base map; establish deformation-advance distance curves and deformation-time curves respectively; extract surrounding rock deformation characteristic parameters such as delamination amount, displacement rate, and stress change rate; construct anomaly judgment rules and set early warning thresholds; and output the surrounding rock stability evaluation results and anomaly area distribution information.

9. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 8, characterized in that: In the hierarchical linkage early warning mechanism in step S4, different early warning levels are determined by a comprehensive assessment of multiple indicators, including the number of indicators, the degree of anomaly, and time synchronization. To eliminate the influence of differences in the dimensions of different monitoring indicators, the degree of anomaly of each monitoring indicator is first normalized. Let the first The normalized anomaly level of each monitoring indicator at the current moment is: ,and Then the overall risk value Calculate using the following formula: ； in, The normalized version The degree of abnormality of each monitoring indicator, and meets the following requirements. ; For the first The weight coefficients corresponding to each monitoring indicator, and satisfying the following conditions: , ; This is the time synchronization coefficient, and ; Let n be the number of monitoring indicators currently in an abnormal state, and n be the total number of monitoring indicators participating in the current comprehensive judgment. ; , This is a correction factor; Among them, the time synchronization coefficient Used to characterize multiple monitoring indicators within a preset synchronization time window The degree of synchronization anomaly within; when the abnormal indicator is in When internal height synchronization occurs, Take the larger value; when abnormal indicators appear over a period of time, Take the smaller value; The weighting coefficient and correction factor , Obtained through calibration using historical mine pressure samples; Based on the comprehensive risk value The warning level is determined by comparing the result with the preset level threshold.

10. The method for early warning using a multi-source data fusion intelligent early warning platform for coal mine pressure according to claim 9, characterized in that: In the spatial neighborhood propagation mechanism in step S4, a spatial neighborhood model is constructed with the warning point as the center. The warning threshold of each monitoring point within the neighborhood is dynamically adjusted according to its spatial distance from the warning point, and the adjustment range of the threshold is negatively correlated with the spatial distance. Let the baseline warning threshold be , No. The spatial distance between each neighboring monitoring point and the early warning point is The dynamic early warning threshold for that monitoring point is... satisfy: ； in, Let be the distance attenuation coefficient, and ; This is the threshold adjustment coefficient, and ; ; The closer a monitoring point is to a warning point, the lower its corresponding dynamic warning threshold; as the distance between monitoring points increases, the dynamic warning threshold gradually approaches the baseline warning threshold. ; The parameters and Obtained through calibration using historical mining pressure propagation samples.

Images