Gas concentration prediction method based on micro-seismic monitoring data

Abstract

The invention discloses a gas concentration prediction method based on micro-seismic monitoring data, and belongs to the technical field of coal mine safety monitoring. Comprising the steps that micro-seismic sensors and gas sensors which are distributed in a net shape are arranged in a mining area, and data are collected in real time and subjected to denoising and standardization processing; the method comprises the following steps of: constructing a gas concentration prediction model fusing a long-short-term memory network and a convolutional neural network by extracting multi-dimensional characteristic parameters such as a microseismic energy gradient, a seismic source aggregation degree and a gas concentration change rate, and dynamically distributing spatial-temporal characteristic weights by utilizing an attention mechanism; a safety threshold value is dynamically adjusted based on coal seam permeability, mining depth and geological parameters, score calculation of comprehensive danger evaluation indexes is combined, and three-level early warning of underground sound-light alarm, ventilation regulation and control and remote expert cooperation is triggered; and meanwhile, through periodic model calibration and parameter retraining, the prediction precision is improved. According to the invention, space-time collaborative perception and self-adaptive early warning of the gas risk are realized, and the real-time performance and reliability of coal mine safety prevention and control are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine safety monitoring, and in particular to a gas concentration prediction method based on microseismic monitoring data. Background Art

[0002] As a by-product resource of coal seams, gas control has always been a major problem that needs to be faced and solved during the coal mining process. Due to the complexity and uncertainty of underground coal mining, how to accurately predict the gas concentration in the tunnel and timely detect potential dangers is an urgent problem to be solved in the field of coal mine safety.

[0003] At present, with the in-depth study of gas disasters, gas concentration prediction methods have achieved fruitful results. Many patents on gas concentration prediction have involved various methods and technologies such as machine learning, numerical simulation, and comprehensive data analysis, but the accuracy of gas concentration prediction is still not high.

[0004] As coal mining progresses deeper, disturbances to the coal body become increasingly frequent. Existing gas concentration prediction methods struggle to comprehensively consider the correlations between factors influencing gas hazards, such as stress, gas outburst dynamics, and dynamic load disturbances, which are associated with the rapid increase in the intensity of coal-rock dynamic hazards. Consequently, various monitoring and early warning methods are ineffective, and it is difficult to integrate different early warning methods to achieve comprehensive disaster warnings.

[0005] Therefore, it is necessary to provide a gas concentration prediction method based on microseismic monitoring data to solve the problems mentioned in the above background technology. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting gas concentration based on microseismic monitoring data, comprising:

[0007] Step 1: Within the coal mining area, several monitoring units are deployed based on the geological structure and mining area layout. The monitoring units are used to collect microseismic event data and gas concentration data within the mining area in real time;

[0008] Step 2: De-noising, standardization and integrity processing of microseismic event data and gas concentration data;

[0009] Step 3: extracting multidimensional characteristic parameters related to the gas concentration change trend from the microseismic event data according to the gas concentration change trend, and constructing a mining area gas concentration prediction model using a neural network based on the multidimensional characteristic parameters;

[0010] Step 4: Establish a comprehensive risk assessment index with microseismic energy gradient, source concentration, and gas concentration change rate as input features;

[0011] Step 5: Predicting gas concentration based on the comprehensive risk assessment index. When the gas concentration exceeds a dynamic threshold, triggering a three-level warning system consisting of underground sound and light alarms, mine-level ventilation control, and remote expert collaboration.

[0012] Step 6: Periodically calibrate the gas concentration prediction model through the model optimization module. The model optimization module is based on the error index between the downhole measured data and the prediction results. When the error continuously exceeds the preset threshold, it automatically triggers the retraining of the gas concentration prediction model parameters.

[0013] Preferably, the monitoring unit includes a plurality of microseismic sensors and gas sensors, and the microseismic sensors and gas sensors are distributed in a grid-like manner in the mining area.

[0014] Preferably, the multi-dimensional feature parameter extraction includes:

[0015] The concentration of earthquake sources is calculated by spatial clustering algorithm to quantify the density of earthquake sources within a unit volume;

[0016] The microseismic energy gradient is calculated based on the time window sliding calculation method, which represents the energy change rate per unit time;

[0017] Combined with the historical data of gas concentration, the gas concentration change rate is calculated using the sliding window statistical method.

[0018] Preferably, the gas concentration prediction model construction method includes:

[0019] Long short-term memory network is used to process time series data of gas concentration;

[0020] The spatial distribution characteristics of microseismic events are extracted through convolutional neural networks;

[0021] The spatiotemporal features are fused through a dynamic weight allocation mechanism, which includes assigning dynamic weights to the spatial features of microseismic events and the temporal features of gas concentration, and outputting a predicted value of gas concentration.

[0022] Preferably, the method for determining the dynamic threshold includes:

[0023] Dynamically adjust the threshold range based on coal seam permeability, mining depth and geological structure parameters;

[0024] When the microseismic energy gradient continues to grow and exceeds the preset ratio, the gas concentration safety threshold is automatically lowered.

[0025] Preferably, the multi-level warning includes:

[0026] Level 1 warning: The underground substation terminal analyzes minute-level data in real time. When the predicted concentration is ≥ 80% of the dynamic threshold and persists for 3 minutes, an audible and visual alarm and a local power outage are triggered.

[0027] Level 2 warning: The mine-level monitoring center generates an expert diagnosis report based on hourly data. When the calculated score of the comprehensive risk assessment index is ≥70 points and lasts for 1 hour, a ventilation adjustment instruction is issued;

[0028] For the third-level warning, the remote expert center receives data via the Internet. When the calculated score of the comprehensive risk assessment index is ≥90 points, the remote expert center initiates emergency decision support.

[0029] Preferably, the spatial clustering algorithm adopts a density clustering method to identify the spatial correlation between high-density earthquake source areas and gas-rich areas.

[0030] Preferably, the dynamic weight mechanism is implemented through an attention mechanism, which assigns dynamic weights to the spatial characteristics of microseismic events and the temporal characteristics of gas concentration respectively.

[0031] Compared with the existing technology, the present invention provides a method for predicting gas concentration based on microseismic monitoring data, which has the following beneficial effects:

[0032] In the present invention, by analyzing the spatial distribution characteristics of microseismic events and the temporal characteristics of gas concentration, and combining long-short-term memory networks, convolutional neural networks, and attention mechanisms to dynamically allocate weights, the limitations of traditional single-indicator monitoring are overcome and the accuracy of gas concentration prediction is improved; at the same time, the safety threshold is adjusted in real time based on coal seam permeability, mining depth, and microseismic energy gradient to ensure the accuracy and reliability of multi-level early warning; and the gas concentration prediction model is periodically calibrated through an optimization module to ensure the stability of the gas concentration prediction error and avoid performance degradation of the traditional gas concentration prediction model due to environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Schematic diagram of the overall structure of a gas concentration prediction method based on microseismic monitoring data. DETAILED DESCRIPTION

[0034] See also Figure 1 The present invention provides a method for predicting gas concentration based on microseismic monitoring data, comprising:

[0035] Step 1: In the coal mining area, a number of monitoring units are deployed based on the geological structure and mining area layout. The monitoring units are used to collect microseismic event data and gas concentration data in the mining area in real time; the monitoring units include a number of microseismic sensors and gas sensors, and the microseismic sensors and gas sensors are distributed in a grid in the mining area.

[0036] Specifically, microseismic sensors, with their high sensitivity and wide-band response, can accurately capture weak vibration signals generated by microseismic events (such as rock fractures, earthquakes, and human activity). Gas sensors are deployed in locations where gas is likely to accumulate and emerge, such as coal mining faces, return air lanes, and at the edges of goafs, to ensure the representativeness and accuracy of the collected gas concentration data. The collected microseismic event data and gas concentration data are transmitted in real time to the data processing center via wired / wireless networks, providing a reliable basis for subsequent data analysis.

[0037] Step 2: De-noising, standardization and integrity processing of microseismic event data and gas concentration data;

[0038] Specifically, after receiving data collected by the monitoring units, the data processing center performs denoising on the microseismic event data to eliminate invalid microseismic signals caused by factors such as environmental interference. The microseismic data is then standardized to ensure that different parameters have a unified dimension and numerical range, facilitating subsequent data analysis and model calculations. Missing value addition and outlier correction are performed on gas concentration data to ensure data integrity and accuracy.

[0039] Step 3: extracting multidimensional characteristic parameters related to the gas concentration change trend from the microseismic event data according to the gas concentration change trend, and constructing a mining area gas concentration prediction model using a neural network based on the multidimensional characteristic parameters;

[0040] The multidimensional feature parameter extraction includes:

[0041] The spatial clustering algorithm is used to calculate the concentration of earthquake sources and quantify the density of earthquake sources within a unit volume. The spatial clustering algorithm uses a density clustering method to identify the spatial correlation between high-density earthquake source areas and gas-rich areas;

[0042] Specifically, the location data of all microseismic events collected by the data processing center are marked as three-dimensional coordinate points. A "search circle" with a radius of 50 meters is drawn around each microseismic point. If there are at least five microseismic points within the "search circle", these microseismic points are considered to belong to a "cluster area". The classification results are output: isolated events, single microseismic points not covered by any search circle (such as sporadic vibrations); local cluster areas: density ≥ 0.2 / m 3 Areas with a density of more than 0.2 microseismic points per cubic meter; large-scale active zones: density ≥ 0.5 microseismic points / m 3 areas (such as intensive vibration zones, which may correspond to gas-rich areas).

[0043] The microseismic energy gradient is calculated based on the time window sliding calculation method, which represents the energy change rate per unit time. Specifically, based on the microseismic event data collected by the data processing center, the total energy of all microseismic events is counted every 10 minutes, using the formula:

[0044] ΔE=(total energy of current window - total energy of previous window) / total energy of previous window×100%, calculate the energy change rate per unit time.

[0045] If the rate of change in three consecutive time windows exceeds 15%, it is determined to be an “energy anomaly” (such as rapid rock fracture).

[0046] Combined with the historical data of gas concentration, the gas concentration change rate is calculated using the sliding window statistical method.

[0047] Specifically, based on the gas concentration data collected by the data processing center, the maximum, minimum and average values of the gas concentration are counted every 15 minutes. According to the maximum, minimum and average values of the gas concentration in a sliding window, the formula is used:

[0048] R = (maximum value - minimum value) / average value × 100%, calculate the rate of change of gas concentration per unit time.

[0049] If the gas concentration change rate exceeds 30%, it will be marked as "gas concentration abnormality" (such as a crack suddenly breaking through and releasing gas).

[0050] In some embodiments, during the multidimensional feature parameter extraction phase, multiple features, such as the frequency, b-value, and s-value of the vibration event, as well as the peak, mean, and variance of the magnitude over time, can be extracted and combined with geological characteristics to construct a gas concentration prediction model for the mining area using a neural network to assess its impact on the likelihood of gas outburst. By extracting and combining multidimensional features, a feature vector is constructed that comprehensively and accurately reflects the underlying patterns of gas concentration changes, providing rich and effective input information for the gas concentration prediction model.

[0051] Furthermore, the gas concentration prediction model construction method includes:

[0052] Long short-term memory network is used to process the time series data of gas concentration and learn the gas concentration data in time series;

[0053] The spatial distribution characteristics of microseismic events are extracted through convolutional neural networks;

[0054] Specifically, the spatial position and energy data of microseismic sensors are converted into a "vibration density heat map" to automatically detect high-density earthquake source areas and extract the spatial distribution characteristics of microseismic events.

[0055] The spatiotemporal features are fused through a dynamic weight allocation mechanism, which includes assigning dynamic weights to the spatial features of microseismic events and the temporal features of gas concentration, and outputting a predicted value of gas concentration.

[0056] The attention mechanism is introduced to dynamically assign weights: First, the importance scores of the temporal features of gas concentration and the spatial features of microseismic events are calculated separately. Weights are then dynamically assigned based on the importance scores. Finally, the weighted features are integrated to output the final gas concentration prediction. For example, if a region experiences frequent microseismic events but stable gas concentration, the spatial features are weighted more heavily, while the temporal features are weighted less heavily. If a region experiences continuously rising gas concentration but normal microseismic activity, the temporal features are weighted more heavily, while the spatial features are weighted less heavily.

[0057] Step 4: Establish a comprehensive risk assessment index with microseismic energy gradient, source concentration, and gas concentration change rate as input features;

[0058] Specifically, the calculation formula of the comprehensive risk assessment index Z is:

[0059] Z=0.4·ΔE+0.3C+0.3R

[0060] Among them, ΔE is the energy change rate, C is the source concentration, and R is the gas concentration change rate.

[0061] Step 5: Predicting gas concentration based on the comprehensive risk assessment index. When the gas concentration exceeds a dynamic threshold, triggering a three-level warning system consisting of underground sound and light alarms, mine-level ventilation control, and remote expert collaboration.

[0062] Furthermore, the method for determining the dynamic threshold includes:

[0063] Dynamically adjust the threshold range based on coal seam permeability, mining depth and geological structure parameters;

[0064] When the microseismic energy gradient continuously increases beyond the preset ratio, the gas concentration safety threshold is automatically lowered. The basic safety threshold calculation formula is:

[0065] T 基础 =k1·P+k2·D+k3·G

[0066] Among them, P is the coal seam permeability (mD), D is the mining depth (m), G is the fault development index, k1 = 0.2, k2 = 0.05, k3 = 0.3.

[0067] Specifically, the trigger condition for dynamic threshold down-regulation is that when the ΔE of three consecutive time windows increases by more than 15%, the safety threshold is adjusted to T = 0.8·T 基础 .

[0068] The multi-level warning includes:

[0069] Level 1 warning: The underground substation terminal analyzes minute-level data in real time. When the predicted concentration is ≥ 80% of the dynamic threshold and persists for 3 minutes, an audible and visual alarm and a local power outage are triggered.

[0070] Level 2 warning: The mine-level monitoring center generates an expert diagnosis report based on hourly data. When the calculated score of the comprehensive risk assessment index is ≥70 points and lasts for 1 hour, a ventilation adjustment instruction is issued;

[0071] For the third-level warning, the remote expert center receives data via the Internet. When the calculated score of the comprehensive risk assessment index is ≥90 points, the remote expert center initiates emergency decision support.

[0072] Specifically, the multi-level early warning mechanism issues early warnings based on the analysis results of long-term or short-term indicators. When the predicted gas concentration exceeds the dynamic safety threshold, it can send early warning signals to coal mine managers, front-line workers and relevant safety supervision departments through various means such as sound and light alarm devices, text message notifications, and internal communication systems of coal mines, detailing key information such as the location where the gas concentration exceeds the limit, the expected time of exceeding the limit, and the possible degree of danger.

[0073] The first level of real-time warning consists of sensors and underground substation terminals, which are mainly used for real-time data collection. The substation terminals have four levels of analysis software, which collect data by minute, hour, shift, and day, determine changes in gas concentration, and upload real-time data and analysis results. When coal seam characteristics change, real-time alarms and power outages are triggered, and real-time data and analysis results are uploaded. At the same time, control instructions from the monitoring and analysis center are received.

[0074] The mine-level monitoring center is the second level, equipped with a real-time gas concentration tracking and analysis expert diagnosis system. This system completes data processing, summarization, analysis, judgment, early warning, establishment of expert database, output of various reports, comparison and analysis of real-time analysis results underground, issues various voice prompts or control instructions, uploads real-time data and analysis results, and receives instructions from the third-level remote expert analysis center.

[0075] The remote expert center is the third level, which uploads real-time data and analysis results from the mine-level monitoring center server through the Internet. Gas disaster prevention and control experts and researchers perform some auxiliary analysis and processing, and promptly discover technical deficiencies in the real-time tracking and analysis system, make timely modifications, and provide technical support for on-site analysis, processing, and early warning.

[0076] Step 6: Periodically calibrate the gas concentration prediction model through the model optimization module. The model optimization module is based on the error index between the downhole measured data and the prediction results. When the error continuously exceeds the preset threshold, it automatically triggers the retraining of the gas concentration prediction model parameters.

[0077] Specifically, the retraining of model parameters (temporal feature weights and spatial feature weights, etc.) includes the following process:

[0078] The system calculates the average error between the daily predicted gas concentration and the actual gas measurement. If the error exceeds 0.5% for five consecutive days or if a major geological change (such as new excavation) is detected, retraining is automatically initiated. Then, valid records are screened from the last 30 days of data, outliers are removed, and incremental training is performed on the data, which is divided chronologically into a training set (the first 25 days) and a validation set (the last 5 days). When the validation set error is ≤0.4% and the warning response time is ≤3 minutes, the old prediction model is replaced. If the error continues to fall below the target, the old model is retained and technical personnel are notified to conduct an inspection.

[0079] The above is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A gas concentration prediction method based on microseismic monitoring data, characterized in that: include: Step 1: Within the coal mining area, several monitoring units are deployed based on the geological structure and mining area layout. The monitoring units are used to collect real-time microseismic event data and gas concentration data within the mining area; Step 2: De-noising, standardization and integrity processing of microseismic event data and gas concentration data; Step 3: extracting multidimensional characteristic parameters related to the gas concentration change trend from the microseismic event data according to the gas concentration change trend, and constructing a mining area gas concentration prediction model using a neural network based on the multidimensional characteristic parameters; Step 4: Establish a comprehensive risk assessment index with microseismic energy gradient, source concentration, and gas concentration change rate as input features; Step 5: Predicting gas concentration based on the comprehensive risk assessment index. When the gas concentration exceeds a dynamic threshold, triggering a three-level warning system consisting of underground sound and light alarms, mine-level ventilation control, and remote expert collaboration. Step 6: Periodically calibrate the gas concentration prediction model through the model optimization module. The model optimization module is based on the error index between the downhole measured data and the prediction results. When the error continuously exceeds the preset threshold, the gas concentration prediction model parameter retraining is automatically triggered.

2. The method for predicting gas concentration based on microseismic monitoring data according to claim 1, characterized in that: The monitoring unit includes a plurality of microseismic sensors and gas sensors, which are distributed in a network shape in the mining area.

3. The method for predicting gas concentration based on microseismic monitoring data according to claim 1, characterized in that: The multidimensional feature parameter extraction includes: The concentration of earthquake sources is calculated by spatial clustering algorithm to quantify the density of earthquake sources within a unit volume; The microseismic energy gradient is calculated based on the time window sliding calculation method, which represents the energy change rate per unit time; Combined with the historical data of gas concentration, the gas concentration change rate is calculated using the sliding window statistical method.

4. The method for predicting gas concentration based on microseismic monitoring data according to claim 1, characterized in that: The gas concentration prediction model construction method includes: Long short-term memory network is used to process time series data of gas concentration; The spatial distribution characteristics of microseismic events are extracted through convolutional neural networks; The spatiotemporal features are fused through a dynamic weight allocation mechanism, which includes assigning dynamic weights to the spatial features of microseismic events and the temporal features of gas concentration, and outputting a predicted value of gas concentration.

5. The method for predicting gas concentration based on microseismic monitoring data according to claim 1, characterized in that: The method for determining the dynamic threshold includes: Dynamically adjust the threshold range based on coal seam permeability, mining depth and geological structure parameters; When the microseismic energy gradient continues to grow and exceeds the preset ratio, the gas concentration safety threshold is automatically lowered.

6. The method for predicting gas concentration based on microseismic monitoring data according to claim 1, characterized in that: The multi-level warning includes: Level 1 warning: The underground substation terminal analyzes minute-level data in real time. When the predicted concentration is ≥ 80% of the dynamic threshold and persists for 3 minutes, an audible and visual alarm and a local power outage are triggered. Level 2 warning: The mine-level monitoring center generates an expert diagnosis report based on hourly data. When the calculated score of the comprehensive risk assessment index is ≥70 points and lasts for 1 hour, a ventilation adjustment instruction is issued; For the third-level warning, the remote expert center receives data via the Internet. When the calculated score of the comprehensive risk assessment index is ≥90 points, the remote expert center initiates emergency decision support.

7. The method for predicting gas concentration based on microseismic monitoring data according to claim 3, characterized in that: The spatial clustering algorithm adopts a density clustering method to identify the spatial correlation between high-density earthquake source areas and gas-rich areas.

8. The method for predicting gas concentration based on microseismic monitoring data according to claim 4, characterized in that: The dynamic weight mechanism is implemented through an attention mechanism, which assigns dynamic weights to the spatial characteristics of microseismic events and the temporal characteristics of gas concentration respectively.

Images