A gas extraction method based on dynamic characteristics and outburst early warning of gas outbursts

By acquiring gas monitoring data, extracting dynamic characteristics of gas outbursts, and using machine learning models for early warning and discrimination, a graded control strategy is generated. This solves the problems of independence and false alarms in existing gas outburst early warning systems, and realizes rapid response and intelligent control of gas extraction.

CN123109757APending Publication Date: 2026-11-17SHANXI INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611048659.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing gas outburst early warning and extraction control technologies suffer from problems such as independent early warning systems and equipment, high false alarm rates, lack of machine learning optimization and closed-loop linkage, resulting in insufficient prevention and control capabilities.

Method used

By acquiring gas monitoring data, dynamic hazard characteristics of gas outbursts are extracted, machine learning classification models are used for early warning and identification, and graded extraction and control strategies are generated to achieve coordinated control of gas extraction.

Benefits of technology

It improves the accuracy and reliability of gas outburst early warning, enables rapid response and intelligent control of gas extraction, and adapts to changes in different geological conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123109757A_ABST
    Figure CN123109757A_ABST
Patent Text Reader

Abstract

This invention relates to the field of intelligent prevention and control technology for underground gas disasters in coal mines, and particularly to a gas drainage method based on dynamic characteristics of gas emission and outburst early warning. The method includes the following steps: acquiring gas monitoring data of the gas drainage area; performing data preprocessing based on the gas monitoring data to obtain preprocessed monitoring data; extracting dynamic hazard characteristics of gas emission based on the preprocessed monitoring data, wherein the dynamic hazard characteristics of gas emission include the gas emission rate of sudden change, the temporal fluctuation amplitude of gas concentration, the slope of the gas emission trend change, and the short-term energy spectrum anomaly index; and performing control interference discrimination based on the dynamic hazard characteristics of gas emission to obtain effective feature data. This invention, based on intelligent prevention and control technology for underground gas disasters in coal mines, achieves real-time identification, accurate early warning, and automatic hierarchical linkage control of gas drainage equipment for coal and gas outburst hazards.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent prevention and control technology for underground gas disasters in coal mines, and in particular to a gas extraction method based on the dynamic characteristics of gas outbursts and outburst early warning. Background Technology

[0002] Existing technologies for early warning and drainage control based on gas outburst characteristics still have significant shortcomings: First, existing early warning systems and gas drainage equipment control units are independent of each other. After the early warning information is issued, manual control of equipment such as drainage pumps and pipeline valves is still required, resulting in a delayed response to hazards and difficulty in timely source control. Second, existing early warning models lack a mechanism to identify interference from the adjustment behavior of drainage equipment. Fluctuations in gas monitoring data caused by normal control operations such as variable frequency speed regulation of drainage pumps and adjustment of the opening of electric valves in pipelines are easily misjudged as outburst anomalies, leading to a high false alarm rate and affecting the accuracy and reliability of early warning results. Third, most existing early warning methods use fixed thresholds or analytical models for hazard determination, lacking a machine learning iterative optimization mechanism based on historical early warning cases. This makes it difficult to continuously adapt to changes in different mine geological conditions and mining conditions, and also fails to achieve closed-loop linkage between early warning determination, drainage control, and feedback optimization, resulting in insufficient proactive prevention and control capabilities and intelligence levels for gas outbursts. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a gas extraction method based on the dynamic characteristics of gas outbursts and early warning of gas outbursts, so as to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a gas extraction method based on the dynamic characteristics of gas outbursts and outburst early warning includes the following steps:

[0005] Step S1: Obtain gas monitoring data from the gas extraction area; perform data preprocessing based on the gas monitoring data to obtain preprocessed monitoring data;

[0006] Step S2: Extract dynamic hazard characteristics of gas outburst based on preprocessed monitoring data. These dynamic hazard characteristics include the gas outburst abrupt change rate, the temporal fluctuation amplitude of gas concentration, the slope of gas outburst trend change, and the short-term energy spectrum anomaly index.

[0007] Step S3: Based on the dynamic hazard characteristics of gas outbursts, control interference is judged to obtain effective feature data; the effective feature data is input into a pre-trained and converged machine learning classification model to output the early warning result;

[0008] Step S4: Generate a tiered extraction and control strategy based on the early warning results; execute coordinated control according to the tiered extraction and control strategy to obtain the gas extraction and control results.

[0009] The beneficial effects of this invention are as follows:

[0010] 1. By establishing a mapping relationship between the resistivity of magnetotelluric detection and mechanical parameters such as the elastic modulus, cohesion, and internal friction angle of rock mass, the quantitative conversion of geophysical parameters to engineering mechanical parameters is realized. This solves the problem that existing technologies mainly rely on field tests or engineering analogies to obtain rock mass mechanical parameters, resulting in long parameter acquisition cycles and insufficient assignment accuracy, thereby improving the accuracy and geological adaptability of numerical simulation parameters.

[0011] 2. By utilizing magnetotelluric inversion results to construct a normal fault electrical structure model, and combining lithological classification and physical property constraints to achieve collaborative modeling of fault geometry and physical property parameters, this approach addresses the problems of existing technologies failing to effectively couple fault electrical and mechanical structures, and numerical models failing to accurately reflect fault fracture zones and surrounding rock characteristics. This improves the precision of normal fault modeling, geological consistency, and the reliability of stress field simulation.

[0012] 3. By establishing a rapid simulation method for normal fault faulting and stress field changes based on magnetotelluric data constraints, we can achieve rapid quantitative analysis of fault structure identification, mechanical parameter inversion, and stress field evolution. This will solve the problems of low risk assessment efficiency and lack of rapid quantitative evaluation methods in existing railway crossing active normal fault projects, and improve the efficiency and scientific nature of railway engineering route selection, seismic design, and fault avoidance decisions. Attached Figure Description

[0013] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0014] Figure 1 This is a schematic diagram of the steps of a gas extraction method based on the dynamic characteristics of gas outbursts and early warning of gas outbursts according to the present invention.

[0015] Figure 2 This is a detailed flowchart illustrating step S1 of the present invention.

[0016] Figure 3 This is a schematic diagram of a tiered sampling control strategy in one embodiment;

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0019] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0020] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a gas extraction method based on the dynamic characteristics and outburst early warning of gas outbursts, the method comprising the following steps:

[0022] Step S1: Obtain gas monitoring data from the gas extraction area; perform data preprocessing based on the gas monitoring data to obtain preprocessed monitoring data;

[0023] In one embodiment, gas monitoring data of the gas drainage area is acquired. This data includes monitoring data such as gas concentration, mixed flow rate, drainage negative pressure, and ambient gas pressure. Gas concentration sensors, flow sensors, negative pressure sensors, and gas pressure sensors are deployed at key locations such as the gas drainage borehole, the main drainage pipeline, and the working face. Each sensor continuously collects monitoring data at a sampling frequency of not less than 1 Hz and transmits the data to the edge computing node via the underground industrial ring network. For example, in a certain working face, the gas concentration is 1.25%, the mixed flow rate is 18.6 m³ / min, and the drainage negative pressure is... The ambient air pressure was 100.8 kPa; the gas concentration at another sampling time was 1.32%, and the mixed flow rate was... The negative pressure of the extraction is After receiving gas monitoring data, the edge computing node first performs data preprocessing, including using a median filtering algorithm to remove instantaneous spike noise from the sensor and using a linear interpolation algorithm to compensate for data gaps caused by short-term communication interruptions or sampling omissions, thus obtaining continuous and complete preprocessed monitoring data.

[0024] Step S2: Extract dynamic hazard characteristics of gas outburst based on preprocessed monitoring data. These dynamic hazard characteristics include the gas outburst abrupt change rate, the temporal fluctuation amplitude of gas concentration, the slope of gas outburst trend change, and the short-term energy spectrum anomaly index.

[0025] In one embodiment, a fixed-length sliding time window is established based on preprocessed monitoring data, and dynamic hazard characteristics of gas outbursts are extracted within each sliding window. The gas outburst mutation rate is calculated based on the relative change in gas concentration per unit time between the current time and the previous sampling time; the temporal fluctuation amplitude of gas concentration is characterized by the coefficient of variation of the gas concentration sequence within the sliding window; when the coefficient of variation exceeds a preset threshold of 0.3, the gas concentration is determined to be in a state of violent fluctuation; the slope of the gas outburst trend is linearly fitted to the continuous gas concentration data of the past 2 to 10 minutes using the least squares method to reflect the continuous growth or decline trend of gas outbursts; the short-time energy spectrum anomaly index is calculated by analyzing the gas concentration signal spectrum using short-time Fourier transform to calculate the proportion of high-frequency disturbance energy to the total energy, which is used to identify weak abnormal fluctuations. For example, the gas outburst mutation rate calculated within a certain time window is... The concentration variation coefficient is 0.36, the trend slope is 0.015% / min, and the high-frequency energy ratio is 0.28, indicating that the current gas outburst has obvious abnormal fluctuation characteristics.

[0026] Step S3: Based on the dynamic hazard characteristics of gas outbursts, control interference is judged to obtain effective feature data; the effective feature data is input into a pre-trained and converged machine learning classification model to output the early warning result;

[0027] In one embodiment, operating parameters of the extraction equipment are acquired simultaneously, including the operating frequency of the extraction pump, the opening degree of the pipeline electric valves, and the target negative pressure setpoint of the extraction pipeline. This data is then combined with dynamic hazard characteristics of gas outbursts to perform control interference judgment. When an adjustment operation of the extraction equipment is detected within the current sliding time window, it is analyzed whether the trend of gas monitoring data changes is consistent with the theoretical change law generated by the equipment adjustment. For example, when the operating frequency of the extraction pump increases from 35Hz to 40Hz, the mixed flow rate changes from... Increase to Simultaneously, if the gas concentration decreases from 1.32% to 1.18%, this change is determined to be a data disturbance caused by normal adjustment of the extraction equipment, and all feature data in the corresponding window are discarded as interference data. If no equipment adjustment behavior is detected, or the trend of the monitored data is inconsistent with the equipment adjustment pattern, the corresponding feature is retained as valid feature data. The valid feature data is normalized, and a model input feature vector is constructed based on the gas emission mutation rate, the temporal fluctuation amplitude of gas concentration, the slope of the gas emission trend change, and the short-term energy spectrum anomaly index. This vector is then input into a pre-trained XGBoost classification model or a random forest classification model for classification prediction. For example, if the model outputs a probability of no prominent danger of 0.05, a yellow warning probability of 0.18, an orange warning probability of 0.27, and a red warning probability of 0.50, then the current warning result is determined to be a red warning.

[0028] Step S4: Generate a tiered extraction and control strategy based on the early warning results; execute coordinated control according to the tiered extraction and control strategy to obtain the gas extraction and control results.

[0029] In one embodiment, the corresponding warning level is determined based on the warning results output by the machine learning classification model, and a corresponding graded gas extraction control strategy is generated. Specifically, when the warning result is a yellow warning, a control command is generated to increase the operating frequency of the extraction pump by 5% to 15%, and the gas monitoring sampling frequency is simultaneously increased. When the warning result is an orange warning, in addition to increasing the operating frequency of the extraction pump to 70% to 90% of its rated maximum frequency, the electric valves corresponding to at least two backup extraction boreholes are automatically opened to expand the gas extraction range. When the warning result is a red warning, a command is generated to operate the extraction pump at 100% rated full load, and the mine's main ventilation fan is linked to increase the working face airflow to the maximum allowable value, while the underground emergency broadcast is activated and the power supply to non-intrinsically safe electrical equipment is cut off. For example, when the model determines a red warning, the system automatically adjusts the operating frequency of the extraction pump to 50Hz full load operation, opens the electric valves of three backup extraction boreholes, increases the main ventilation fan airflow from 2600 m³ / min to 3400 m³ / min, and simultaneously activates the underground personnel evacuation broadcast. After the coordinated control is completed, gas monitoring data is continuously collected to dynamically monitor the control effect. When the gas concentration and various hazardous characteristics gradually return to the safe range, the corresponding gas extraction control results are obtained.

[0030] Preferably, after step S4, the following steps are also included:

[0031] Dynamic feedback evaluation is conducted based on the gas extraction and control results to obtain feedback evaluation results;

[0032] In one embodiment, gas monitoring data is continuously acquired based on the gas drainage control results, and dynamic feedback evaluation is performed in conjunction with equipment execution status and early warning results. The feedback evaluation results include evaluation information such as changes in hazard level, gas concentration trends, gas emission rate mutation rate, drainage equipment operating status, control duration, and final disposal results. For example, after an orange alert is triggered, the system automatically increases the drainage pump operating frequency from 35Hz to 42Hz and opens two backup drainage boreholes. After continuous drainage for 12 minutes, the gas concentration decreases from 1.82% to 1.21%, and the gas emission rate mutation rate decreases from... Descending to If the model determines a yellow alert for three consecutive independent time windows, the feedback evaluation result is recorded as effective control; if the risk level remains at a red alert after 15 minutes of continuous control, the feedback evaluation result is recorded as the control has not completely eliminated the risk.

[0033] The historical early warning case database is updated based on the feedback and evaluation results.

[0034] In one embodiment, a standardized historical case data package is generated based on the feedback evaluation results and written into the historical early warning case database. This historical case data package includes information such as gas monitoring data before the early warning is triggered, dynamic hazard characteristics of gas outbursts, early warning results output by the machine learning classification model, operating frequency of the extraction pump, opening degree of pipeline electric valves, ventilation system control parameters, equipment execution logs, feedback evaluation results, and final handling results. For example, for a red early warning event, the database saves the gas monitoring sequence for the 30 minutes prior to the warning, the entire linkage control process, and the hazard level change curve; for false alarm events, it simultaneously saves the cause of the false alarm and the corresponding feedback evaluation results. The historical early warning case database is categorized and stored according to information such as mine area, coal seam number, early warning level, and event occurrence time, enabling unified management and rapid retrieval of case data.

[0035] The machine learning classification model is iteratively optimized based on the updated historical early warning case database to obtain the classification threshold.

[0036] In one embodiment, the machine learning classification model is periodically iteratively optimized based on the updated historical early warning case database. Model training is automatically initiated when preset update conditions are met. These update conditions include updating the model every 7 days or triggering an update when at least 50 complete handling cases are added to the historical early warning case database. During model training, new cases and historical cases are extracted from the historical early warning case database to form a training dataset. Incremental training or full retraining is performed on the XGBoost extreme gradient boosting model or the random forest ensemble learning model. After training, the updated model is evaluated for accuracy, recall, and false positive rate. When the model performance meets preset requirements, the online running model is automatically replaced, and the probability thresholds for yellow, orange, and red early warnings are recalculated based on the model's output probabilities to obtain updated classification thresholds. This allows the machine learning classification model to continuously adapt to changes in mine geological conditions and mining operations, improving the accuracy and stability of early warnings.

[0037] Preferably, step S1 includes the following steps:

[0038] Step S11: Obtain gas monitoring data in the gas extraction area;

[0039] In one embodiment, gas monitoring data of the gas drainage area is acquired. This data includes monitoring data such as gas concentration, mixed flow rate, drainage negative pressure, and ambient gas pressure. Specifically, gas concentration sensors, flow rate sensors, negative pressure sensors, and gas pressure sensors are deployed at key locations such as the gas drainage borehole, main drainage pipeline, branch drainage pipeline, and working face. Each sensor continuously collects monitoring data at a sampling frequency of not less than 1Hz and transmits the data in real time to the edge computing node via the underground industrial ring network. For example, at one sampling moment, the collected data shows a gas concentration of 1.26%, a mixed flow rate of 18.4 m³ / min, a drainage negative pressure of -18.6 kPa, and an ambient gas pressure of 100.9 kPa; at another sampling moment, the corresponding data shows a gas concentration of 1.33%, a mixed flow rate of 19.0 m³ / min, and a drainage negative pressure of -18.3 kPa. The edge computing node caches the monitoring data in chronological order, forming a continuous gas monitoring data sequence.

[0040] Step S12: Perform outlier detection on the gas monitoring data to obtain abnormal sampling points;

[0041] In one embodiment, outlier detection is performed on gas monitoring data. First, continuous sampling sequences corresponding to each monitoring parameter are extracted, and abnormal sampling points are identified according to preset anomaly judgment thresholds. Specifically, continuous change thresholds can be established for gas concentration, mixing flow rate, extraction negative pressure, and ambient air pressure. When the change between adjacent sampling points exceeds the corresponding threshold, the current sampling point is determined to be an abnormal sampling point. For example, if the gas concentration remains stable between 1.20% and 1.35% during continuous sampling, but a sampling point suddenly jumps to 4.85%, and then the next sampling point recovers to 1.28%, then the data corresponding to 4.85% is determined to be instantaneous spike noise from the sensor; similarly, if the extraction negative pressure remains continuously at a certain level... Nearby, but a certain sampling point suddenly changed to If the sampling point immediately returns to normal, it is identified as an abnormal sampling point, thus obtaining a set of abnormal sampling points.

[0042] Step S13: Median filtering is used to denoise the abnormal sampling points to obtain filtered monitoring data;

[0043] In one embodiment, a median filtering algorithm is used to denoise the identified abnormal sampling points. Specifically, a filtering window of a preset length is established centered on the abnormal sampling point. For example, two sampling points before and after the abnormal sampling point are selected to form a five-point filtering window. All monitoring data within the window are extracted and sorted according to their numerical values. The median value after sorting is used as the replacement value for the abnormal sampling point. For example, if the corresponding data in a certain gas concentration filtering window are 1.23%, 1.27%, 4.85%, 1.29%, and 1.31%, after sorting, we get 1.23%, 1.27%, 1.29%, 1.31%, and 4.85%. The median value of 1.29% is used to replace the abnormal value of 4.85%. As another example, if there is an abnormal value of 31.6 m³ / min in the mixed flow window, while the remaining data are distributed between 18.3 and 19.2 m³ / min, the median value of 18.9 m³ / min after sorting is used for replacement. Finally, filtered monitoring data is obtained, effectively eliminating the influence of instantaneous peak noise from the sensor on subsequent analysis.

[0044] Step S14: Identify missing data segments based on the continuity detection of filtered monitoring data; use linear interpolation to compensate and repair the missing data segments to obtain preprocessed monitoring data.

[0045] In one embodiment, a median filtering algorithm is used to denoise the identified abnormal sampling points. Specifically, a filtering window of a preset length is established centered on the abnormal sampling point. For example, two sampling points before and after the abnormal sampling point are selected to form a five-point filtering window. All monitoring data within the window are extracted and sorted according to their numerical values. The median value after sorting is used as the replacement value for the abnormal sampling point. For example, if the corresponding data in a certain gas concentration filtering window are 1.23%, 1.27%, 4.85%, 1.29%, and 1.31%, after sorting, we get 1.23%, 1.27%, 1.29%, 1.31%, and 4.85%. The median value of 1.29% is used to replace the abnormal value of 4.85%. As another example, if there is an abnormal value of 31.6 m³ / min in the mixed flow window, while the remaining data are distributed between 18.3 and 19.2 m³ / min, the median value of 18.9 m³ / min after sorting is used for replacement, and finally, the filtered monitoring data is obtained.

[0046] Preferably, step S12 includes the following steps:

[0047] Step S121: Extract the continuous sampling sequence from the gas monitoring data;

[0048] In one embodiment, continuous sampling sequences corresponding to gas concentration, mixed flow rate, extraction negative pressure, and ambient air pressure are extracted from gas monitoring data and arranged in chronological order to form time-series data for each monitoring parameter. Each sensor continuously uploads monitoring data at a sampling frequency of at least 1 Hz. Edge computing nodes synchronize and align the data uploaded by different sensors based on timestamps and establish continuous sampling sequences. For example, the gas concentration sequence collected over 10 seconds is 1.22%, 1.24%, 1.25%, 1.27%, 1.26%, 1.29%, 1.31%, 1.30%, 1.28%, and 1.27%, and the mixed flow rate sequence is 18.1 m³ / min, 18.3 m³ / min, 18.5 m³ / min, 18.6 m³ / min, 18.4 m³ / min, 18.7 m³ / min, 18.9 m³ / min, 18.8 m³ / min, 18.6 m³ / min, and 18.5 m³ / min.

[0049] Step S122: Select sampling points in the continuous sampling sequence that exceed the preset anomaly judgment threshold as abnormal sampling points.

[0050] In one embodiment, the variation range between adjacent sampling points is calculated based on a continuous sampling sequence and compared with a preset anomaly judgment threshold. When the variation range exceeds the preset anomaly judgment threshold, the corresponding sampling point is determined to be an abnormal sampling point. The preset anomaly judgment threshold can be set according to different monitoring parameters. For example, when the variation range of gas concentration between adjacent sampling points exceeds 0.8%, the variation range of mixed flow rate exceeds 5 m³ / min, or the variation range of extraction negative pressure exceeds 8 kPa, the corresponding sampling point is determined to be an abnormal sampling point. For example, if a continuous gas concentration sequence is 1.26%, 1.28%, 1.30%, 4.82%, 1.31%, and 1.33%, and the variation range of 4.82% compared to the previous sampling point reaches 3.52%, exceeding the gas concentration anomaly judgment threshold, the sampling point corresponding to 4.82% is determined to be an abnormal sampling point; similarly, if the extraction negative pressure is continuously maintained at a certain level... Nearby, and a certain sampling point suddenly changed If the sampling point is not found, it will also be identified as an abnormal sampling point, thus obtaining the set of abnormal sampling points.

[0051] Preferably, step S13 includes the following steps:

[0052] Step S131: Construct a filter window of a preset length centered on the abnormal sampling point;

[0053] In one embodiment, a filter window is constructed by selecting a preset number of consecutive sampling points before and after the abnormal sampling point as the center. The length of the filter window is set according to the sampling frequency and the characteristics of the monitored data changes, preferably using a 5-point window, a 7-point window, or a 9-point window. For example, when the sampling frequency is 1Hz, a 5-point window can be used, where the filter window is composed of two normal sampling points before and after the current abnormal sampling point. If the abnormal sampling point is located at the beginning or end of the sequence, the length of the filter window is supplemented by symmetrical expansion or boundary replication to ensure that each abnormal sampling point corresponds to a complete filter window.

[0054] Step S132: Extract the monitoring data of each sampling point within the filtering window, sort them according to their numerical values ​​to obtain ordered values; use the median of the ordered values ​​as the replacement value for the abnormal sampling points.

[0055] In one embodiment, monitoring data corresponding to all sampling points within the filtering window are extracted and arranged in ascending order to form an ordered numerical sequence. The value in the middle of the ordered numerical sequence is selected as the replacement value for the abnormal sampling point. For example, if the data in a certain gas concentration filtering window are 1.25%, 1.27%, 4.82%, 1.28%, and 1.29%, and after sorting, the values ​​are 1.25%, 1.27%, 1.28%, 1.29%, and 4.82%, with the median value being 1.28%, then 1.28% is used as the replacement value for the abnormal sampling point. Similarly, if the data in a certain extraction negative pressure filtering window are -18.4 kPa, -18.3 kPa, -7.6 kPa, -18.2 kPa, and -18.1 kPa, and after sorting, the values ​​are -18.4 kPa, -18.3 kPa, -18.2 kPa, -18.1 kPa, and -7.6 kPa, then -18.2 kPa is used as the replacement value for the abnormal sampling point.

[0056] Step S133: Replace the monitoring data corresponding to the abnormal sampling points with replacement values ​​to obtain filtered monitoring data.

[0057] In one embodiment, replacement values ​​are written to the corresponding abnormal sampling point locations while keeping the data from the remaining normal sampling points unchanged, thereby generating filtered monitoring data. For example, replacing the abnormal gas concentration value of 4.82% with 1.28% results in a filtered gas concentration sequence of 1.25%, 1.27%, 1.28%, 1.28%, and 1.29%; replacing the abnormal extraction negative pressure value of -7.6 kPa with -18.2 kPa restores the filtered extraction negative pressure sequence to -18.4 kPa, -18.3 kPa, -18.2 kPa, -18.2 kPa, and -18.1 kPa. Through the above replacement process, while preserving the original temporal variation trend, abnormal data caused by instantaneous sensor interference is effectively removed, providing reliable preprocessed monitoring data for subsequent continuous detection, missing data compensation, and extraction of dynamic hazard characteristics of gas outbursts.

[0058] Preferably, the formula for calculating the gas emission mutation rate in step S2 is:

[0059] ;

[0060] in, The rate of change in gas emission is used to characterize the relative change in gas concentration per unit time. Indicates the current time The corresponding gas concentration value; Indicates the time before the current moment The gas concentration value corresponding to that moment; This represents the time interval between two consecutive sampling operations, ranging from 5 to 60 seconds.

[0061] In one embodiment, continuous gas concentration time-series data is extracted from preprocessed monitoring data and sampled at preset time intervals. Obtain the gas concentration value at the current time t. and the former Gas concentration value at time .in, The settings are configured according to monitoring requirements, with a range of 5–60 seconds. For example, when the sampling frequency is 1 Hz, If the time interval is set to 10 seconds, the current gas concentration is 1.84%, and the gas concentration in the previous 10 seconds was 1.52%. These values ​​are then used as... and Participate in subsequent calculations. For example, when , , When, the calculation yields:

[0062] ;

[0063] This indicates a significant upward trend in gas concentration during this period. If the R value calculated over multiple consecutive sliding windows continues to rise, it can be considered one of the important hazardous characteristics of abnormal gas outbursts. Together with the temporal fluctuation amplitude of gas concentration, the slope of the gas outburst trend, and the short-term energy spectrum anomaly index, it constitutes the dynamic hazardous characteristics of gas outbursts.

[0064] Preferably, step S3 includes the following steps:

[0065] Step S31: Screen the dynamic hazard characteristics of gas outbursts, remove characteristic data affected by the control of extraction equipment, and obtain effective characteristic data;

[0066] In one embodiment, operating parameters of the extraction equipment, such as the operating frequency of the extraction pump, the opening degree of the electric valves in the pipeline, and the target negative pressure setpoint of the extraction pipeline, are simultaneously acquired and combined with the dynamic hazard characteristics of gas emission to perform control interference discrimination. When the control operation of the extraction equipment is detected within the current sliding time window, it is further determined whether the trend of gas monitoring data change is consistent with the theoretical change law corresponding to the equipment control; if they are consistent, it is determined that the dynamic hazard characteristics of gas emission in the current window are caused by equipment control, and the corresponding feature data is removed; if there is no equipment control behavior, or the data change trend is inconsistent with the equipment control law, the corresponding feature data is retained as valid feature data. For example, when the operating frequency of the extraction pump is increased from 40Hz to 45Hz, the mixed flow rate increases from 18.5m³ / min to 21.3m³ / min, and the gas concentration decreases from 1.62% to 1.35%, which is consistent with the normal change pattern after enhanced extraction. In this case, the characteristic data within the corresponding time window is discarded as control interference data. If the operating parameters of the extraction equipment remain unchanged, but the gas concentration suddenly increases to 2.48%, and the gas emission mutation rate and short-term energy spectrum anomaly index increase simultaneously, it is determined to be a real abnormal gas emission from the coal seam, and the corresponding characteristic data is retained as valid characteristic data.

[0067] Step S32: Normalize the effective feature data and construct the model input feature vector;

[0068] In one embodiment, the effective feature data is normalized to eliminate differences in the dimensions and numerical ranges of different features, thereby improving the stability of machine learning model training and inference. Specifically, a maximum-minimum normalization method can be used. Based on preset maximum and minimum values ​​corresponding to each hazard feature, the sudden change rate of gas emission, the temporal fluctuation amplitude of gas concentration, the slope of gas emission trend change, and the short-term energy spectrum anomaly index are converted into normalized feature values ​​between 0 and 1. Subsequently, the normalized feature values ​​are combined in a preset order to construct the model input feature vector. For example, if the four normalized features are 0.76, 0.41, 0.58, and 0.83, then the model input feature vector is constructed accordingly. , as input data for machine learning classification models.

[0069] Step S33: Input the model's input feature vector into the pre-trained, converged machine learning classification model to perform classification prediction and obtain the warning result.

[0070] In one embodiment, the model input feature vector is input into a pre-trained and converged machine learning classification model for classification prediction. The machine learning classification model employs either the XG Boost extreme gradient boosting algorithm or the random forest ensemble learning algorithm. The model calculates the probability of occurrence corresponding to each hazard level based on the input feature vector and outputs the warning level with the highest probability as the final warning result. For example, if the model outputs probabilities of no prominent danger, yellow warning, orange warning, and red warning as 0.05, 0.18, 0.63, and 0.14 respectively, then the current warning result is determined to be an orange warning; conversely, if the model outputs probabilities of 0.02, 0.08, 0.25, and 0.65 respectively, then the current warning result is determined to be a red warning. The warning result serves as the basis for subsequently generating tiered sampling and control strategies and implementing coordinated control measures.

[0071] Preferably, step S32 includes the following steps:

[0072] Step S321: Obtain the preset maximum and preset minimum values ​​of valid feature data;

[0073] In one embodiment, based on the long-term accumulated effective feature data in the historical early warning case database of the target mine, preset maximum and minimum values ​​corresponding to the gas emission mutation rate, gas concentration temporal fluctuation amplitude, gas emission trend change slope, and short-term energy spectrum anomaly index are statistically analyzed and used as benchmark parameters for subsequent normalization processing. The preset maximum and minimum values ​​can be updated periodically based on historical sample statistical results to adapt to changes in different mining areas and coal seam occurrence conditions. For example, the gas emission mutation rate obtained from historical statistics... The values ​​range from 0 to 0.08 s⁻¹, the values ​​range from 0 to 0.60, the values ​​range from -0.05 to 0.20% / min, and the values ​​range from 0 to 1.50. The above statistical values ​​are then used as the preset maximum and minimum values ​​for the corresponding characteristics.

[0074] Step S322: Normalize the effective feature data according to the preset maximum and minimum values ​​to obtain each normalized feature value;

[0075] In one embodiment, based on the preset maximum and minimum values ​​corresponding to each valid feature data, the maximum-minimum normalization method is used to calculate each normalized feature value, and the calculation formula is as follows:

[0076] ;

[0077] in, Represents the normalized eigenvalues; This represents the currently valid feature data; This represents the preset minimum value for the corresponding feature; This represents the preset maximum value for the corresponding feature. After normalization, each hazard feature is uniformly mapped to the range of 0 to 1, eliminating differences in the dimensions and numerical ranges of different features. For example, the sudden change rate of gas emission at a certain moment. for When the preset minimum value is 0 and the preset maximum value is 0.080s⁻¹, the normalized characteristic value is 0.50; when the gas concentration time-series fluctuation amplitude is 0.42 and its preset value range is 0 to 0.60, the normalized characteristic value is 0.70.

[0078] Step S323: Combine the normalized feature values ​​to obtain the model input feature vector.

[0079] In one embodiment, the normalized feature values ​​are combined to form a model input feature vector according to a preset feature arrangement order, and then input into a machine learning classification model for classification prediction. The model input feature vector is arranged in the order of gas emission mutation rate, gas concentration temporal fluctuation amplitude, gas emission trend slope, and short-term energy spectrum anomaly index. For example, after normalization, the normalized feature values ​​corresponding to each hazard feature are 0.50, 0.70, 0.42, and 0.81, respectively. The model input feature vector is then constructed accordingly. If other extended hazardous features are introduced at the same time, they will be added to the end of the model input feature vector in a preset feature order, eventually forming a feature vector that meets the input requirements of the machine learning classification model.

[0080] Preferably, the machine learning classification model in step S33 adopts the XG Boost extreme gradient boosting algorithm or the random forest ensemble learning algorithm.

[0081] In one embodiment, the machine learning classification model adopts the XG Boost extreme gradient boosting algorithm or the random forest ensemble learning algorithm; the model training dataset comes from the long-term accumulated measured gas emission data and real outburst event records of the target mine, where the positive samples are the time-series feature segments 1 hour before the outburst occurs, and the negative samples are the time-series segments randomly extracted under normal and stable extraction conditions, with a positive-to-negative sample ratio of 1:3 to 1:5.

[0082] Preferably, step S4 includes the following steps:

[0083] Step S41: Determine the warning level based on the warning results, whereby the warning levels include yellow warning, orange warning, and red warning;

[0084] In one embodiment, the probabilities of occurrence of each hazard level output by the machine learning classification model are compared, and the hazard level with the highest probability of occurrence is selected as the current warning level. The warning levels include yellow, orange, and red. A yellow warning indicates a slight anomaly in gas emission, posing a potential outburst risk; an orange warning indicates a continuously increasing anomaly in gas emission, further increasing the outburst risk; and a red warning indicates a significant anomaly in gas emission, posing a high risk of coal and gas outbursts, requiring immediate activation of the highest-level emergency control measures. The system enters the corresponding linkage control process based on the determined warning level.

[0085] Step S42: When the warning level is yellow, generate a command to adjust the operating frequency of the extraction pump and obtain yellow warning control data;

[0086] In one embodiment, when the current warning level is determined to be a yellow warning, the control system generates a pump operating frequency adjustment command according to the control strategy corresponding to the yellow warning, and sends it to the pump frequency converter controller via the SCADA system. The pump operating frequency is increased by 5% to 15% and continues to run for 10 to 30 minutes. Simultaneously, the data sampling frequency of the gas sensor is increased to perform encrypted monitoring of gas concentration and pumping parameters, forming yellow warning control data. The yellow warning control data includes information such as the target operating frequency of the pump, the adjustment duration, the sensor sampling frequency, and the execution status.

[0087] Step S43: When the warning level is orange, generate an instruction to open the backup extraction borehole electric valve and obtain orange warning control data;

[0088] In one embodiment, when the current warning level is determined to be an orange warning, the control system generates an instruction to open the electric valves of the backup drainage boreholes and simultaneously generates an instruction to enhance the operation of the drainage pumps. The SCADA system controls the operating frequency of the drainage pumps to be increased to 70%–90% of the equipment's rated maximum frequency. At the same time, it automatically opens the electric valves corresponding to at least two reserved backup drainage boreholes, expanding the gas drainage coverage area and improving gas drainage capacity, thus generating orange warning control data. The orange warning control data includes the target operating frequency of the drainage pumps, the number of backup boreholes opened, the opening status of each electric valve, and execution feedback information.

[0089] Step S44: When the warning level is red, generate a full-load operation command for the extraction pump and an emergency broadcast command, and obtain red warning control data;

[0090] In one embodiment, when the current warning level is determined to be a red alert, the control system generates a full-load operation command for the extraction pump, a linkage control command for the mine main ventilation fan, and an emergency broadcast activation command, and sends these commands to the corresponding execution equipment via the SCADA system. The extraction pump switches to 100% rated full-load operation, the mine main ventilation fan is increased to the maximum allowable air volume, the emergency broadcast system automatically plays personnel evacuation instructions, and simultaneously cuts off the power supply to non-intrinsically safe electrical equipment at the working face, forming red alert control data. The red alert control data includes the operating status of the extraction pump, the operating status of the ventilation fan, the emergency broadcast status, the power cut-off status, and feedback information from each execution device.

[0091] Step S45: Use the yellow warning control data, orange warning control data, and red warning control data as the tiered extraction control strategy; execute the linkage control according to the tiered extraction control strategy to obtain the gas extraction control results.

[0092] In one embodiment, when the current warning level is determined to be a red alert, the control system generates a full-load operation command for the extraction pump, a linkage control command for the mine main ventilation fan, and an emergency broadcast activation command, and sends these commands to the corresponding execution equipment via the SCADA system. The extraction pump switches to 100% rated full-load operation, the mine main ventilation fan is increased to the maximum allowable air volume, the emergency broadcast system automatically plays personnel evacuation instructions, and simultaneously cuts off the power supply to non-intrinsically safe electrical equipment at the working face, forming red alert control data. The red alert control data includes the operating status of the extraction pump, the operating status of the ventilation fan, the emergency broadcast status, the power cut-off status, and feedback information from each execution device.

[0093] It should be noted that you should refer to [link / reference]. Figure 3 When no significant danger is detected, maintain the current extraction parameters and conduct normal monitoring. When a yellow alert is issued, increase the operating frequency of the extraction pump by 5%–15%, increase the monitoring frequency, and continue operating for 10–30 minutes. When an orange alert is issued, increase the operating frequency of the extraction pump to 70%–90% of the rated maximum frequency, automatically open the backup extraction borehole valve, and continue operating and dynamically evaluating. When a red alert is issued, control the extraction pump to operate at 100% full load, activate maximum ventilation, initiate emergency broadcasting to organize personnel evacuation, and cut off non-intrinsically safe power supplies. Continue implementing control measures until the alert level is lowered.

[0094] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0095] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A gas extraction method based on gas emission dynamic characteristics and outburst early warning, characterized in that, Includes the following steps: Step S1: Obtain gas monitoring data in the gas extraction area; Data preprocessing is performed based on gas monitoring data to obtain preprocessed monitoring data; Step S2: Extract dynamic hazard characteristics of gas outburst based on preprocessed monitoring data. These dynamic hazard characteristics include the gas outburst abrupt change rate, the temporal fluctuation amplitude of gas concentration, the slope of gas outburst trend change, and the short-term energy spectrum anomaly index. Step S3: Based on the dynamic hazard characteristics of gas outbursts, control interference is judged to obtain effective feature data; the effective feature data is input into a pre-trained and converged machine learning classification model to output the early warning result; Step S4: Generate a tiered extraction and control strategy based on the early warning results; execute coordinated control according to the tiered extraction and control strategy to obtain the gas extraction and control results.

2. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 1, characterized in that, Step S4 is followed by: Dynamic feedback evaluation is conducted based on the gas extraction and control results to obtain feedback evaluation results; The historical early warning case database is updated based on the feedback and evaluation results. The machine learning classification model is iteratively optimized based on the updated historical early warning case database to obtain the classification threshold.

3. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain gas monitoring data in the gas extraction area; Step S12: Perform outlier detection on the gas monitoring data to obtain abnormal sampling points; Step S13: Median filtering is used to denoise the abnormal sampling points to obtain filtered monitoring data; Step S14: Identify missing data segments based on the continuity detection of filtered monitoring data; use linear interpolation to compensate and repair the missing data segments to obtain preprocessed monitoring data.

4. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 3, characterized in that, Step S12 includes the following steps: Step S121: Extract the continuous sampling sequence from the gas monitoring data; Step S122: Select sampling points in the continuous sampling sequence that exceed the preset anomaly judgment threshold as abnormal sampling points.

5. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 3, characterized in that, Step S13 includes the following steps: Step S131: Construct a filter window of a preset length centered on the abnormal sampling point; Step S132: Extract the monitoring data of each sampling point within the filtering window, sort them according to their numerical values ​​to obtain ordered values; use the median of the ordered values ​​as the replacement value for the abnormal sampling points. Step S133: Replace the monitoring data corresponding to the abnormal sampling points with replacement values ​​to obtain filtered monitoring data.

6. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 1, characterized in that, The formula for calculating the gas emission mutation rate in step S2 is: ; wherein, represents the mutation rate of gas emission amount, used to represent the relative change degree of gas concentration per unit time; represents the current time corresponding to the gas concentration value; represents the time before the current time corresponding to the gas concentration value; represents the sampling time interval of adjacent two times participating in calculation, and the value range is 5-60s.

7. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Screen the dynamic hazard characteristics of gas outbursts, remove characteristic data affected by the control of extraction equipment, and obtain effective characteristic data; Step S32: Normalize the effective feature data and construct the model input feature vector; Step S33: Input the model's input feature vector into the pre-trained, converged machine learning classification model to perform classification prediction and obtain the warning result.

8. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 7, characterized in that, Step S32 includes the following steps: Step S321: Obtain the preset maximum and preset minimum values ​​of valid feature data; Step S322: Normalize the effective feature data according to the preset maximum and minimum values ​​to obtain each normalized feature value; Step S323: Combine the normalized feature values ​​to obtain the model input feature vector.

9. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 7, characterized in that, The machine learning classification model in step S33 uses either the XGBoost extreme gradient boosting algorithm or the random forest ensemble learning algorithm.

10. The gas extraction method based on gas emission dynamic characteristics and outburst early warning according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Determine the warning level based on the warning results, whereby the warning levels include yellow warning, orange warning, and red warning; Step S42: When the warning level is yellow, generate a command to adjust the operating frequency of the extraction pump and obtain yellow warning control data; Step S43: When the warning level is orange, generate an instruction to open the backup extraction borehole electric valve and obtain orange warning control data; Step S44: When the warning level is red, generate a full-load operation command for the extraction pump and an emergency broadcast command, and obtain red warning control data; Step S45: Use the yellow warning control data, orange warning control data, and red warning control data as the tiered extraction control strategy; execute the linkage control according to the tiered extraction control strategy to obtain the gas extraction control results.