LSTM-based coal mine gas concentration time sequence prediction method and system

Abstract

The invention relates to the technical field of coal mine gas concentration prediction, in particular to a coal mine gas concentration time sequence prediction method and system based on LSTM. The method comprises the following steps: collecting multi-source data such as a mining machine pose time sequence, a micro-seismic monitoring event sequence and roadway environment monitoring, constructing a damage-permeability evolution equation by combining geological information, inverting a coal-rock mass dynamic stress damage field, calculating a gas dynamic permeability field and a gushing source intensity field, and finally predicting the gushing source intensity field by combining a space-time prediction model with historical data. And predicting future gas concentration distribution. And real-time space positioning of the coal mine gas emission source is realized.

Description

Technical Field

[0001] This invention relates to the field of coal mine gas concentration prediction technology, and in particular to a time-series prediction method and system for coal mine gas concentration based on LSTM. Background Technology

[0002] Among the existing methods for predicting coal mine gas concentration time series, there is a technology that uses microseismic data to invert the stress field and predict the total amount of gas emission from the working face through empirical formulas, such as the technology disclosed in patent application CN114810213A. However, this type of technology has disadvantages such as the inability to spatially locate the source of gas emission, poor decision support capability, and disconnection from real-time mining operations. Summary of the Invention

[0003] This invention addresses the shortcomings of existing technologies, such as the inability to spatially locate the source of gas outbursts, poor decision support capabilities, and disconnection from real-time mining operations, by providing a time-series prediction method and system for coal mine gas concentration based on LSTM.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a time-series prediction method for coal mine gas concentration based on LSTM, including: Collect multi-source monitoring data of the target mining face, wherein the multi-source monitoring data includes at least mining machinery position and timing data, microseismic monitoring event sequences, and roadway environmental monitoring data; Obtain geological information of the target mining face, and construct a damage-permeability evolution equation for the target mining face based on the geological information; Based on the time-series data of the mining machinery's pose and the sequence of microseismic monitoring events, the dynamic stress damage field of the coal and rock mass with the current time as the reference is obtained by inverting the geological information. Based on the dynamic stress damage field of the coal and rock mass, and combined with the preset damage-permeability evolution equation, the dynamic permeability field of gas is calculated. Based on the gas dynamic permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal body gas content distribution updated according to the mining progress, the gas dynamic outburst source strength field at the current moment is obtained by solving the preset source strength calculation rules. Using the gas dynamic emission source intensity field as input, and combining historical gas concentration time series data, a spatiotemporal prediction model is used to predict the gas concentration distribution in future periods.

[0005] Optionally, multi-source monitoring data of the target mining face is collected. This multi-source monitoring data includes at least the timing data of mining machinery pose, microseismic monitoring event sequences, and roadway environmental monitoring data, including: Positioning and attitude sensors installed on the mining machinery are used to collect the timing data of the mining machinery's posture at a first sampling frequency. The timing data of the mining machinery's posture includes at least three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed. By deploying a microseismic sensor array on the target mining face and surrounding rock, a sequence of microseismic monitoring events is collected at a second sampling frequency. Each microseismic monitoring event in the sequence includes at least the occurrence time, spatial coordinates, and released energy. Environmental monitoring data of the tunnel is collected at a third sampling frequency by environmental sensors installed in the tunnel. The environmental monitoring data of the tunnel includes at least gas concentration, gas pressure and wind speed. A unified spatiotemporal coordinate system is established, and the time sequence data of the mining machinery pose, the sequence of microseismic monitoring events, and the roadway environmental monitoring data are time-stamped and spatially registered to form synchronized multi-source monitoring data.

[0006] Optionally, the geological information of the target mining face is obtained, and a damage-permeability evolution equation for the target mining face is constructed based on the geological information, including: Obtain geological information and coal and rock mass test data of the target mining face; The basic original permeability value of the coal and rock mass is obtained based on the geological information. Based on the coal and rock mass test data, the coupling coefficient was obtained by fitting. The damage coefficient of the target mining face is extracted from the dynamic stress damage field of the coal and rock mass; Based on the original permeability baseline value, the coupling coefficient, and the damage coefficient, the damage-permeability evolution equation is constructed.

[0007] Optionally, based on the time-series data of the mining machinery's pose and the sequence of microseismic monitoring events, a dynamic stress-damage field of the coal and rock mass with the current time as a reference is obtained by inverting the geological information, including: An initial geomechanical numerical model of the target mining face is established based on the geological information; The timing data of the mining machinery's pose is converted into dynamic boundary load inputs for the initial geomechanical numerical model; The spatial coordinates and energy release distribution of the microseismic monitoring event sequence within the current time window are used as the observation data input to the data assimilation algorithm; Using the data assimilation algorithm, the initial geomechanical numerical model is updated with the observation data, so that the theoretical microseismic energy field output by the initial geomechanical numerical model matches the observation data, and the dynamic stress damage field of the coal and rock mass containing stress values ​​and damage coefficients at various points in space is inverted and output.

[0008] Optionally, based on the dynamic stress damage field of the coal and rock mass and combined with a preset damage-permeability evolution equation, the dynamic permeability field of the gas is calculated, including: Substitute the damage coefficient at each spatial location in the dynamic stress damage field of the coal and rock mass into the damage-permeability evolution equation; The dynamic permeability of gas at each spatial point is calculated in parallel to generate a dynamic permeability field of gas that is spatially isomorphic to the dynamic stress damage field.

[0009] Optionally, based on the dynamic gas permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal gas content distribution updated according to the mining progress, the dynamic gas outburst source strength field at the current moment is obtained by solving a preset source strength calculation rule, including: Based on the geological information of the target mining face and the roadway environment monitoring data, the comprehensive conductivity coefficient of multiple spatial coordinate points of the target mining face is obtained; The dynamic permeability of multiple spatial coordinate points of the target mining face is obtained from the gas dynamic permeability field; Based on the roadway environmental monitoring data, the absolute values ​​of gas pressure gradients at multiple spatial coordinate points of the target mining face are obtained; Based on the time sequence data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated. Based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content of multiple spatial coordinate points of the target mining face, the dynamic gas outburst source strength of multiple spatial coordinate points is calculated. The dynamic gas emission source strength field is obtained based on multiple dynamic gas emission source strengths.

[0010] The dynamic gas emission source strength is calculated based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content at multiple spatial coordinate points of the target mining face, including: A preset source strength calculation mathematical model is established, in which the dynamic gas outburst source strength is positively correlated with the comprehensive conductivity coefficient, the dynamic permeability, the absolute value of the gas pressure gradient, and the residual gas content.

[0011] Among them, based on the temporal data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated, including: Based on the time sequence data of the mining machinery's position, combined with the geological information of the target mining face, the coal body of the target mining face is divided into three dynamic regions in three-dimensional space: the mined-out area, the current mining pressure relief influence area, and the original undisturbed area. For mined-out areas, the residual gas content is set to zero; For the current mining and depressurization impact zone, based on the time elapsed since the start of mining-induced depressurization and the spatial distance between the current mining and depressurization impact zone and the mining face, the real-time residual gas content of the current mining and depressurization impact zone is calculated using a preset dynamic gas content decay function. The dynamic gas content decay function includes a time decay factor and a spatial decay factor. For the original undisturbed area, the original gas content value determined by the geological survey report is used as the residual gas content.

[0012] Optionally, using the gas dynamic emission source intensity field as input, and combining historical gas concentration time-series data, a spatiotemporal prediction model is used to predict the gas concentration distribution for future periods, including: A spatiotemporal prediction model is constructed, which is a hybrid model composed of a long short-term memory network and a convolutional neural network; Construct a spatiotemporal input feature vector, which includes historical gas concentration time series data and dynamic gas outburst source intensity field; The spatiotemporal input feature vector is reorganized into a feature matrix with spatial dimension, which is then input into a convolutional neural network module to obtain a high-level spatial feature map. The advanced spatial feature map is expanded along the time dimension and input into the long short-term memory network module, outputting a sequence of predicted continuous gas concentration values ​​for each monitoring point within a preset future time period.

[0013] Secondly, this invention provides a time-series prediction system for coal mine gas concentration based on LSTM, comprising: The multi-source monitoring data acquisition module is used to collect multi-source monitoring data of the target mining face. The multi-source monitoring data includes at least mining machinery position and timing data, microseismic monitoring event sequences, and roadway environmental monitoring data. The permeability equation construction module is used to obtain geological information of the target mining face and construct the damage-permeability evolution equation of the target mining face based on the geological information. The stress damage distribution inference module is used to obtain the dynamic stress damage field of the coal and rock mass based on the current time, according to the geological information, based on the time series data of the mining machinery's pose and the sequence of microseismic monitoring events. The dynamic permeability distribution calculation module is used to calculate the dynamic permeability field of gas based on the dynamic stress damage field of the coal and rock mass and in combination with the preset damage-permeability evolution equation. The dynamic emission source strength calculation module is used to calculate the dynamic emission source strength field of gas at the current moment based on the dynamic gas permeability field, the gas pressure data in the roadway environment monitoring data, and the coal body gas content distribution updated according to the mining progress, through a preset source strength calculation rule. The gas concentration distribution prediction module is used to predict the gas concentration distribution in the future period by taking the gas dynamic emission source strength field as input, combining historical gas concentration time series data, and using a spatiotemporal prediction model.

[0014] By implementing this invention, it is possible to collect multi-source monitoring data of the target mining face. The multi-source monitoring data includes at least the temporal data of the mining machinery position, the sequence of microseismic monitoring events, and the roadway environmental monitoring data. This provides comprehensive and synchronous basic data support for subsequent steps such as inverting the dynamic stress damage field of coal and rock mass and calculating the dynamic permeability field of gas, ensuring the spatiotemporal consistency of the data.

[0015] By implementing this invention, it is possible to obtain geological information of the target mining face and construct a damage-permeability evolution equation for the target mining face based on the geological information; establish a quantitative relationship between the degree of damage to coal and rock mass and gas permeability, and provide a theoretical model basis for deriving the dynamic permeability field of gas from the dynamic stress damage field of coal and rock mass.

[0016] By implementing this invention, it is possible to obtain the dynamic stress damage field of coal and rock mass based on the current time by inverting the geological information according to the time sequence data of the mining machinery posture and the microseismic monitoring event sequence; to realize the real-time and spatial characterization of the dynamic stress damage state of coal and rock mass, and to accurately reflect the dynamic impact of mining activities on coal and rock mass.

[0017] By implementing this invention, it is possible to calculate the dynamic permeability field of gas based on the dynamic stress damage field of the coal and rock mass and in combination with the preset damage-permeability evolution equation; thus realizing the spatial dynamic distribution calculation of gas permeability and overcoming the shortcomings of traditional methods that have a single and static permeability value.

[0018] By implementing this invention, it is possible to obtain the dynamic gas emission source strength field at the current moment by solving the gas dynamic permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal body gas content distribution updated according to the mining progress, through a preset source strength calculation rule; thus realizing the real-time spatial positioning of the gas emission source and clarifying the differences in gas emission intensity in different areas.

[0019] By implementing this invention, it is possible to use the dynamic gas emission source strength field as input, combined with historical gas concentration time series data, and utilize a spatiotemporal prediction model to predict the gas concentration distribution in future periods, thereby improving the predictability and accuracy of gas concentration prediction and providing precise decision-making basis for gas prevention and control.

[0020] In summary, by implementing this invention, real-time spatial positioning of coal mine gas outburst sources can be achieved, outputting advanced gas concentration prediction results with high spatiotemporal resolution. At the same time, the physical interpretability of the prediction process is enhanced, significantly improving the decision support capability for coal mine gas prevention and control. Attached Figure Description

[0021] Figure 1 A flowchart illustrating the time-series prediction method for coal mine gas concentration based on LSTM provided by this invention; Figure 2 A schematic diagram of the structure of the LSTM-based coal mine gas concentration time series prediction system provided by the present invention.

[0022] In the attached diagram, the components represented by each number are as follows: The module includes: multi-source monitoring data acquisition module 11, permeability equation construction module 12, stress damage distribution deduction module 13, dynamic permeability distribution calculation module 14, dynamic outflow source strength calculation module 15, and gas concentration distribution prediction module 16. Detailed Implementation

[0023] Example 1, as Figure 1 As shown, this invention provides a method and system for time-series prediction of coal mine gas concentration based on LSTM, including: S100: Collect multi-source monitoring data of the target mining face, wherein the multi-source monitoring data includes at least mining machinery position and timing data, microseismic monitoring event sequence, and roadway environment monitoring data; S200: Obtain geological information of the target mining face and construct a damage-permeability evolution equation for the target mining face based on the geological information; S300: Based on the time-series data of the mining machinery's pose and the sequence of microseismic monitoring events, the dynamic stress damage field of the coal and rock mass with the current time as the reference is obtained by inverting the geological information. S400: Based on the dynamic stress damage field of the coal and rock mass, and combined with the preset damage-permeability evolution equation, the dynamic permeability field of the gas is calculated. S500: Based on the gas dynamic permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal body gas content distribution updated according to the mining progress, the gas dynamic outburst source strength field at the current moment is obtained by solving the preset source strength calculation rules. S600: Using the gas dynamic emission source strength field as input, combined with historical gas concentration time series data, a spatiotemporal prediction model is used to predict the gas concentration distribution in future periods.

[0024] In step S100 of this application embodiment, multi-source monitoring data of the target mining face is collected. The multi-source monitoring data includes at least mining machinery pose time-series data, microseismic monitoring event sequences, and roadway environmental monitoring data, including: Positioning and attitude sensors installed on the mining machinery are used to collect the timing data of the mining machinery's posture at a first sampling frequency. The timing data of the mining machinery's posture includes at least three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed. By deploying a microseismic sensor array on the target mining face and surrounding rock, a sequence of microseismic monitoring events is collected at a second sampling frequency. Each microseismic monitoring event in the sequence includes at least the occurrence time, spatial coordinates, and released energy. Environmental monitoring data of the tunnel is collected at a third sampling frequency by environmental sensors installed in the tunnel. The environmental monitoring data of the tunnel includes at least gas concentration, gas pressure and wind speed. A unified spatiotemporal coordinate system is established, and the time sequence data of the mining machinery pose, the sequence of microseismic monitoring events, and the roadway environmental monitoring data are time-stamped and spatially registered to form synchronized multi-source monitoring data.

[0025] In this embodiment of the application, the core purpose of step S100 is to obtain multi-dimensional and synchronized basic monitoring data of the target mining face, so as to provide a unified spatiotemporal reference data source for subsequent inversion of the dynamic stress damage field of coal and rock mass, calculation of the dynamic gas outburst source strength field and prediction of gas concentration distribution.

[0026] To achieve the above objectives, it is first necessary to collect the timing data of the mining machinery's posture at a first sampling frequency using positioning sensors and attitude sensors installed on the mining machinery. The timing data of the mining machinery's posture includes at least three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed. Specifically, positioning sensors and attitude sensors need to be installed on the mining machinery. Positioning sensors are responsible for capturing spatial position information, while attitude sensors are used to record the machinery's operating attitude parameters.

[0027] The initial sampling frequency needs to be determined and adjusted according to the operating speed and conditions of the mining machinery. For example, setting the initial sampling frequency to 10Hz can accurately capture the dynamic changes of the machinery and avoid data omissions or redundancy.

[0028] The system collects three core parameters: three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed. The three-dimensional spatial coordinates determine the real-time position of the mining machinery underground, for example, (X: 1250m, Y: 820m, Z: -560m). The cutting drum rotation speed reflects the intensity of the mining operation, for example, 30 r / min. The travel speed reflects the pace of mining progress, for example, 0.5 m / min.

[0029] Next, a microseismic monitoring event sequence is acquired at a second sampling frequency using a microseismic sensor array deployed on the target mining face and surrounding rock. Each microseismic monitoring event in the sequence includes at least the occurrence time, spatial coordinates, and released energy. Specifically, microseismic sensor arrays can be deployed at certain intervals on the working face, the sides of the roadways, and deep within the surrounding rock of the target mining face. This array-based deployment ensures comprehensive capture of microseismic signals generated by coal and rock mass fracturing.

[0030] Next, a second sampling frequency is determined, which should match the signal characteristics of coal and rock mass fractures. For example, setting the second sampling frequency to 5Hz can effectively capture microseismic events generated by minor fractures in the coal and rock mass, while avoiding interference from invalid signals.

[0031] Each microseismic monitoring event requires the collection of three core pieces of information: occurrence time, spatial coordinates, and released energy. The occurrence time is used to record the event sequence, such as 2025-06-15 10:23:45. The spatial coordinates are used to locate the microseismic event's position, such as (X:1252m, Y:821m, Z:-562m). The released energy reflects the degree of damage to the coal and rock mass, such as 10... 4 J.

[0032] The sequence of microseismic monitoring events is used as the input data for the data assimilation algorithm, which drives the initial geomechanical numerical model to update its state. This is the key basis for inverting the dynamic stress damage field of coal and rock mass.

[0033] Then, environmental monitoring data of the tunnel is collected at a third sampling frequency by environmental sensors arranged in the tunnel. The tunnel environmental monitoring data includes at least gas concentration, gas pressure and wind speed. This involves deploying environmental sensors at different cross-sections and heights within the tunnel. The environmental sensors must cover key areas such as the vicinity of the mining face, return airway, and intake airway.

[0034] Determine a third sampling frequency, based on the principle of stable acquisition of environmental parameters. For example, setting the third sampling frequency to 1Hz ensures data continuity without generating excessive data volume.

[0035] The key parameters collected are three types: gas concentration, gas pressure, and wind speed. Gas concentration directly reflects the current gas state in the roadway, for example, 0.3%. Gas pressure is a key indicator for calculating the gas emission source strength, for example, 0.8 MPa. Wind speed affects the diffusion pattern of gas in the roadway, for example, 1.2 m / s.

[0036] Finally, a unified spatiotemporal coordinate system is established, and the time-series data of the mining machinery pose, the sequence of microseismic monitoring events, and the roadway environmental monitoring data are time-stamped and spatially registered to form synchronized multi-source monitoring data.

[0037] A three-dimensional spatiotemporal coordinate system is constructed using a fixed reference point underground as the origin. The coordinate system must cover the entire target mining face and the surrounding rock area to ensure that all monitoring data can be mapped to the same spatial frame.

[0038] The time-series data of the mining machinery's pose, the sequence of microseismic monitoring events, and the roadway environmental monitoring data, after being timestamped and spatially registered, are integrated to form synchronized multi-source monitoring data, constituting a multi-source monitoring dataset. This multi-source monitoring dataset serves as the fundamental data source for all subsequent analyses and calculations.

[0039] The specific methods for timestamp alignment and spatial registration of the above data are existing technologies and will not be elaborated here.

[0040] In step S200 of this application embodiment, the geological information of the target mining face is obtained, and a damage-permeability evolution equation of the target mining face is constructed based on the geological information, including: Obtain geological information and coal and rock mass test data of the target mining face: The basic original permeability value of the coal and rock mass is obtained based on the geological information. Based on the coal and rock mass test data, the coupling coefficient was obtained by fitting. The damage coefficient of the target mining face is extracted from the dynamic stress damage field of the coal and rock mass; Based on the original permeability baseline value, the coupling coefficient, and the damage coefficient, the damage-permeability evolution equation is constructed.

[0041] In this embodiment of the application, the purpose of step S200 is to construct the damage-permeability evolution equation of the coal and rock mass at the target mining face, establish a quantitative relationship between the degree of damage to the coal and rock mass and the gas permeability, and provide mathematical model support for subsequent calculation of the dynamic permeability field of gas.

[0042] To achieve the above objectives, it is first necessary to obtain geological information and coal and rock mass test data of the target mining face: Specifically, it is necessary to use methods such as underground geological drilling, tunnel surrounding rock logging, and geophysical exploration to collect comprehensive geological information about the target mining face. This geological information includes coal seam thickness, coal seam depth, geological structure distribution, coal and rock mass permeability data, coal and rock composition, and the magnitude and direction of original geostress.

[0043] The coal and rock mass test data were collected through both laboratory experiments and in-situ field tests. Laboratory experiments included triaxial compression tests and permeability tests on coal and rock samples at different stress levels, recording the damage and deformation of the samples and the corresponding permeability values ​​during stress changes.

[0044] In-situ testing can be conducted by installing monitoring boreholes in the surrounding rock of the mining face to collect stress and permeability data of the coal and rock mass at different mining stages in real time. These test data are the key basis for fitting the coupling coefficient.

[0045] Then, based on the geological information, the original permeability baseline value of the coal and rock mass is obtained; The original baseline permeability value refers to the benchmark value of gas permeability in coal and rock masses that have not been disturbed by mining and are in their original stress state. Specifically, it is necessary to select the permeability data of coal and rock masses in areas unaffected by mining from the acquired geological information and determine this as the original baseline permeability value. This value will vary depending on the type of coal and rock.

[0046] For example, if the target mining face is an anthracite coal seam, based on the original coal and rock mass test data in the geological exploration report, its original permeability baseline value is determined to be 1.5 × 10⁻⁶. -18 m 2 .

[0047] Next, the coupling coefficient is obtained by fitting the test data of the coal and rock mass; The coupling coefficient is a core parameter reflecting the sensitivity of the degree of damage to coal and rock mass to permeability. The larger the coupling coefficient, the more significant the impact of changes in the degree of damage to coal and rock mass on permeability.

[0048] Specifically, based on the collected coal and rock mass test data, mathematical methods such as linear regression and nonlinear fitting are needed to establish a functional relationship between the degree of damage to the coal and rock mass and the rate of change in permeability. Then, the specific value of the coupling coefficient is calculated using this functional relationship.

[0049] For example, by performing nonlinear fitting on the test data of coal and rock samples from the target mining face, the coupling coefficient of the coal and rock mass at the mining face was found to be 0.92.

[0050] Furthermore, the damage coefficient of the target mining face is extracted from the dynamic stress damage field of the coal and rock mass; The damage coefficient is an index describing the degree of damage to coal and rock masses after mining disturbance, with a value ranging from 0 to 1. The closer the value is to 0, the lower the degree of damage to the coal and rock mass. The closer the value is to 1, the more severe the damage to the coal and rock mass.

[0051] The dynamic stress-damage field of the coal and rock mass is the result obtained through subsequent data assimilation algorithms. This dynamic stress-damage field includes the stress values ​​and corresponding damage coefficients at various spatial locations on the target mining face.

[0052] Damage coefficients for different regions of the target mining face are extracted from the dynamic stress damage field of the coal and rock mass.

[0053] For example, the damage coefficient of the original undisturbed area is 0.05, the damage coefficient of the current mining and depressurization affected area is 0.65, and the damage coefficient of the mined-out area is 0.98.

[0054] The mined-out area, the current mining and depressurization impact area, and the original undisturbed area are obtained by dividing the coal body of the target mining face in three-dimensional space based on the timing data of the mining machinery's position and the geological information of the target mining face. The specific division process will be described in detail in the subsequent step S500.

[0055] Finally, based on the original permeability baseline value, the coupling coefficient, and the damage coefficient, the damage-permeability evolution equation is constructed.

[0056] That is, taking the original permeability as the baseline, and using the coupling coefficient and damage coefficient as the core variables, a quantitative mathematical relationship is constructed among the three, namely the damage-permeability evolution equation.

[0057] For example, the basic form of the damage-permeability evolution equation can be: Gas dynamic permeability = original permeability base value × (1 + coupling coefficient × damage coefficient).

[0058] Example calculation using specific parameters: The original permeability baseline value is 1.5 × 10⁻⁶. -18 m 2 The coupling coefficient is 0.92, and the damage coefficient of the current mining and depressurization impact zone is 0.65. The dynamic gas permeability of this area is 1.5 × 10⁻⁶. -18 m 2 ×(1+0.92×0.65)≈2.39×10 -18 m 2 .

[0059] This damage-permeability evolution equation can be used to calculate the dynamic permeability of gas at various spatial locations on the target mining face in parallel, providing a calculation basis for generating a dynamic gas permeability field.

[0060] In step S300 of this application embodiment, based on the temporal data of the mining machinery's pose and the sequence of microseismic monitoring events, the dynamic stress damage field of the coal and rock mass with the current time as the reference is obtained by inverting the geological information, including: An initial geomechanical numerical model of the target mining face is established based on the geological information; The timing data of the mining machinery's pose is converted into dynamic boundary load inputs for the initial geomechanical numerical model; The spatial coordinates and energy release distribution of the microseismic monitoring event sequence within the current time window are used as the observation data input to the data assimilation algorithm; Using the data assimilation algorithm, the initial geomechanical numerical model is updated with the observation data, so that the theoretical microseismic energy field output by the initial geomechanical numerical model matches the observation data, and the dynamic stress damage field of the coal and rock mass containing stress values ​​and damage coefficients at various points in space is inverted and output.

[0061] In this embodiment, the purpose of step S300 is to invert and generate a dynamic stress-damage field of the coal and rock mass based on the current moment. This dynamic stress-damage field of the rock mass includes the stress values ​​and damage coefficients of each spatial point on the target mining face, which can accurately reflect the real-time damage state of the coal and rock mass under mining disturbance, and provide data support for subsequent calculation of the dynamic permeability field of gas.

[0062] To achieve the above objectives, it is first necessary to establish an initial geomechanical numerical model of the target mining face based on the geological information. This involves collecting geological information about the target mining face. This information includes coal seam depth, geological structure, physical and mechanical parameters of the coal and rock mass, and the magnitude and direction of the original geostress.

[0063] Then, using geotechnical engineering numerical simulation software, such as UDEC, a three-dimensional geomechanical numerical model of the target mining face is constructed.

[0064] This geomechanical numerical model divides the space into fine grids, covering the mining face, roadways, and surrounding rock areas. It can reflect the mechanical response characteristics of coal and rock masses under their original stress state.

[0065] For example, a mining face has a burial depth of 680m and an original ground stress of 18MPa. Based on these data, an initial geomechanical numerical model can simulate the initial stress value of each grid point in the coal and rock mass when it is not disturbed by mining, which is approximately 17.5-18.5MPa.

[0066] Next, the timing data of the mining machinery's pose is converted into dynamic boundary load inputs for the initial geomechanical numerical model; This involves extracting key parameters from the timing data of the mining machinery's pose. These key parameters include three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed.

[0067] The temporal data of these mining machinery positions are converted into dynamic boundary loads for the initial geomechanical numerical model. Specifically, the cutting action of the machinery generates impact loads on the coal and rock mass, and the movement of the machinery alters the boundary conditions of the mining face. The load boundaries and geometric boundaries of the initial geomechanical numerical model are updated in real time to simulate the dynamic disturbance of the coal and rock mass during the mining process.

[0068] For example, the mining machinery travels at a speed of 0.5 m / min, and the cutting drum rotates at a speed of 30 r / min. After inputting these data into the initial geomechanical numerical model, the stress concentration zone of the coal and rock mass in the cutting area can be simulated, and the stress value will instantly rise to 25-30 MPa.

[0069] Furthermore, the spatial coordinates and energy distribution of the microseismic monitoring event sequence within the current time window are used as input data assimilation algorithms for the observation data; This involves capturing the sequence of microseismic monitoring events within the current time window. The time window can be set according to the mining schedule, such as 10 minutes.

[0070] Extract core information from each microseismic event in the microseismic monitoring event sequence. This information includes spatial coordinates, released energy, etc.

[0071] This information is used as input to the data assimilation algorithm, which integrates numerical models with field monitoring data.

[0072] Example: Within the current time window, 5 microseismic events were detected. One of the events has spatial coordinates (X:1252m, Y:821m, Z:-562m) and releases 10 energy. 4 J. These data will be used as true values ​​to calibrate the calculation results of the numerical model.

[0073] Finally, the data assimilation algorithm is used to drive the initial geomechanical numerical model to update its state using the observation data, so that the theoretical microseismic energy field output by the initial geomechanical numerical model matches the observation data, and the dynamic stress damage field of the coal and rock mass containing stress values ​​and damage coefficients at each point in space is inverted and output.

[0074] First, choose a suitable data assimilation algorithm. Commonly used algorithms include Kalman filtering and ensemble Kalman filtering.

[0075] Driven by observational data, the initial geomechanical numerical model is updated. The data assimilation algorithm continuously adjusts the coal and rock mass mechanical parameters in the model. The observational data is collected by an array of microseismic sensors deployed at the mining face and surrounding rock, including the spatial coordinates and energy distribution of microseismic monitoring event sequences.

[0076] The goal of adjusting the data assimilation algorithm is to ensure that the theoretical microseismic energy field output by the initial geomechanical numerical model is highly matched with the microseismic energy field observed in the field.

[0077] When the error between the two reaches a preset threshold, such as less than 5%, the result output by the initial geomechanical numerical model is the dynamic stress damage field of the coal and rock mass at the current moment.

[0078] This damage field includes the stress values ​​and damage coefficients at each spatial point. For example, after inversion, the stress value in the mining-induced stress relief zone drops to 8-12 MPa, and the damage coefficient is 0.65. The stress value in the mined-out area is close to 0, and the damage coefficient is 0.98.

[0079] In step S400 of this application embodiment, the dynamic gas permeability field is calculated based on the dynamic stress damage field of the coal and rock mass and in conjunction with a preset damage-permeability evolution equation, including: Substitute the damage coefficient at each spatial location in the dynamic stress damage field of the coal and rock mass into the damage-permeability evolution equation; The dynamic permeability of gas at each spatial point is calculated in parallel to generate a dynamic permeability field of gas that is spatially isomorphic to the dynamic stress damage field.

[0080] In this embodiment of the application, the purpose of step S400 is to generate a gas dynamic permeability field that is spatially isomorphic to the dynamic stress damage field of the coal and rock mass, establish a spatial correspondence between the damage state of the coal and rock mass and the gas permeability under mining disturbance, and provide key parameter support for subsequent calculation of the gas dynamic emission source strength field.

[0081] To achieve the above objectives, it is first necessary to substitute the damage coefficient at each spatial location point in the dynamic stress damage field of the coal and rock mass into the damage-permeability evolution equation. This involves extracting the damage coefficient corresponding to each spatial location point of the target mining face from the dynamically stress-damage field of the coal and rock mass that has already been inverted. The damage coefficient varies in different regions. For example, the damage coefficient of the original undisturbed area is 0.05, the damage coefficient of the currently mined and depressurized area is 0.65, and the damage coefficient of the mined-out area is 0.98.

[0082] Next, the dynamic permeability of gas corresponding to each spatial point is calculated in parallel to generate a dynamic permeability field of gas that is spatially isomorphic to the dynamic stress damage field. The damage coefficient at each spatial location is substituted into the preset damage-permeability evolution equation. Parallel computation is used to simultaneously solve for the dynamic gas permeability at each spatial location. For example, the initial base permeability value is 1.5 × 10⁻⁶. -18 m 2 The coupling coefficient is 0.92. The current dynamic gas permeability in the mining and depressurization affected area is 1.5 × 10⁻⁶. -18m 2 ×(1+0.92×0.65)≈2.39×10 -18 m 2 .

[0083] The calculation results of all spatial locations are integrated to generate a dynamic gas permeability field that is completely consistent with the spatial structure of the dynamic stress damage field of the coal and rock mass.

[0084] In step S500 of this embodiment, based on the dynamic gas permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal gas content distribution updated according to the mining progress, the dynamic gas outburst source strength field at the current moment is obtained by solving a preset source strength calculation rule, including: Based on the geological information of the target mining face and the roadway environment monitoring data, the comprehensive conductivity coefficient of multiple spatial coordinate points of the target mining face is obtained; The dynamic permeability of multiple spatial coordinate points of the target mining face is obtained from the gas dynamic permeability field; Based on the roadway environmental monitoring data, the absolute values ​​of gas pressure gradients at multiple spatial coordinate points of the target mining face are obtained; Based on the time sequence data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated. Based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content of multiple spatial coordinate points of the target mining face, the dynamic gas outburst source strength of multiple spatial coordinate points is calculated. The dynamic gas emission source strength field is obtained based on multiple dynamic gas emission source strengths.

[0085] In this embodiment of the application, the purpose of step S500 is to solve for and generate the dynamic gas emission source intensity field at the current moment, realize the spatial accurate quantification of gas emission intensity, input core parameters for subsequent spatiotemporal prediction models, and support the accurate prediction of gas concentration distribution.

[0086] To achieve the above objectives, it is first necessary to obtain the comprehensive conductivity coefficient of multiple spatial coordinate points of the target mining face based on the geological information of the target mining face and the roadway environmental monitoring data. This involves calculating the comprehensive conductivity coefficient at multiple spatial coordinate points by combining geological information of the target mining face and roadway environmental monitoring data. The comprehensive conductivity coefficient reflects the flow and conduction capacity of gas in coal and rock fractures, and its value is related to the degree of coal seam fracture development, pore structure, and roadway ventilation conditions. Specifically, it is necessary to calculate the comprehensive conductivity coefficient at multiple spatial coordinate points by combining geological information of the target mining face and roadway environmental monitoring data.

[0087] For example, the overall conductivity of a fractured area in a certain mining face is 0.08m. 2 / (MPa 2 •d), the coefficient for the intact coal and rock region is 0.01m. 2 / (MPa 2 •d).

[0088] Then, the dynamic permeability of multiple spatial coordinate points of the target mining face is obtained from the gas dynamic permeability field; Dynamic permeability is a direct reflection of the damage state of coal and rock mass. The higher the dynamic permeability value, the easier it is for gas to penetrate.

[0089] Specifically, the corresponding values ​​of multiple spatial coordinate points of the target mining face can be extracted from the generated dynamic gas permeability field.

[0090] For example, the coal and rock mass in the current mining and stress relief area is severely damaged, with a dynamic permeability of 2.39 × 10⁻⁶. -18 m 2 The original undisturbed area suffered minimal damage, and the dynamic permeability remained at its original value of 1.5 × 10⁻⁶. -18 m 2 .

[0091] Next, based on the roadway environmental monitoring data, the absolute values ​​of the gas pressure gradient at multiple spatial coordinate points of the target mining face are obtained; The absolute value of the gas pressure gradient is the driving force of gas flow; the larger the value, the stronger the gas outburst.

[0092] Specifically, based on the gas pressure data collected by the tunnel environment sensors, the ratio of the pressure difference to the spatial distance at each spatial point on the target mining face can be calculated to obtain the absolute value of the pressure gradient.

[0093] For example, the gas pressure near the mining face changes drastically, with an absolute pressure gradient of 0.5 MPa / m; while the pressure is stable in areas far from the working face, with an absolute pressure gradient of only 0.05 MPa / m.

[0094] Furthermore, based on the timing data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated. In step S500 of this embodiment, the residual gas content at multiple spatial coordinate points of the target mining face is calculated based on the timing data of the mining machinery's position and the geological information of the target mining face, including: Based on the time sequence data of the mining machinery's position, combined with the geological information of the target mining face, the coal body of the target mining face is divided into three dynamic regions in three-dimensional space: the mined-out area, the current mining pressure relief influence area, and the original undisturbed area. For mined-out areas, the residual gas content is set to zero; For the current mining and depressurization impact zone, based on the time elapsed since the start of mining-induced depressurization and the spatial distance between the current mining and depressurization impact zone and the mining face, the real-time residual gas content of the current mining and depressurization impact zone is calculated using a preset dynamic gas content decay function. The dynamic gas content decay function includes a time decay factor and a spatial decay factor. For the original undisturbed area, the original gas content value determined by the geological survey report is used as the residual gas content.

[0095] In step S500 of this application embodiment, the purpose of the above step is to accurately calculate the residual gas content at each spatial coordinate point of the target mining face, clarify the remaining gas reserves in different mining disturbance areas, and provide key basic parameters for subsequent solution of the dynamic gas outburst source strength.

[0096] To achieve the above objectives, the first step is to divide the coal body of the target mining face into three dynamic regions in three-dimensional space based on the timing data of the mining machinery's position and the geological information of the target mining face: the mined-out area, the current mining pressure relief influence area, and the original undisturbed area. This involves extracting the core parameters of the mining machinery's position and timing data, including three-dimensional spatial coordinates, cutting drum rotation speed, travel speed, and cutting depth. These parameters reflect the real-time mining range and progress.

[0097] By combining geological information of the target mining face, including coal seam thickness, coal seam dip angle, geological structure distribution, and surrounding rock properties, the spatial distribution characteristics and mechanical properties of the coal body are clarified.

[0098] Based on the above data, in the three-dimensional geomechanical numerical model, the coal body is divided into three dynamic regions according to the degree of mining disturbance: the mined-out area, the current mining pressure relief influence area, and the original undisturbed area.

[0099] For example, the real-time three-dimensional coordinates of the mining machinery are (X:1250m, Y:820m, Z:-560m), the travel speed is 0.5m / min, and the cutting depth is 6m. Based on the geological information of a coal seam thickness of 5.2m, the area 15m behind the machinery is designated as the mined-out area, the area 0-10m in front of the machinery is the current mining pressure relief influence zone, and the area beyond 10m in front of the machinery is the original undisturbed area.

[0100] The second step is to set the residual gas content to zero for mined-out areas. In mined-out areas, the coal seam is completely freed from its original stress state, and the gas has been largely released. Therefore, the residual gas content at all spatial coordinate points in that area is directly set to zero.

[0101] For example, the residual gas content at any spatial point in a mined-out area, such as (X:1245m, Y:820m, Z:-560m), is 0.

[0102] The third step is to calculate the real-time residual gas content of the current mining and depressurization affected area based on the time elapsed since the start of depressurization and the spatial distance between the current mining and depressurization affected area and the mining face, using a preset dynamic decay function of gas content. The dynamic decay function of gas content includes a time decay factor and a spatial decay factor. First, it is necessary to determine two key parameters required for the calculation: one is the time elapsed since the start of the mining-induced depressurization in the area, which can be estimated by the timestamps of the mining machinery's position and timing data; the other is the spatial distance between each spatial point in the area and the mining face, which can be calculated by using a three-dimensional coordinate system to determine the straight-line distance between the two points.

[0103] Then, the preset dynamic decay function for gas content is invoked. For example, the formula for the dynamic decay function for gas content can be expressed as: Residual gas content = Original gas content × e^(-Time decay factor × Decompression time - Spatial decay factor × Spatial distance). Among them, the time decay factor reflects the decay law of gas release rate over time, and the spatial decay factor reflects the change law of gas release degree with distance from the working face. The values ​​of the two factors need to be determined in combination with coal and rock test data.

[0104] Substitute the parameters to calculate the real-time residual gas content at each spatial point in the area.

[0105] For example, a spatial point is located in the current mining depressurization impact zone, 4m from the working face, and is affected by the depressurization for 3 hours. The original gas content is known to be 8.5 m³ / t, and the time decay factor is set to 0.03h. -1 The spatial attenuation factor is 0.12m. -1 Substituting the values ​​into the function, we get: Residual gas content = 8.5 × e^(-0.03 × 3 - 0.12 × 4) ≈ 8.5 × e^(-0.09 - 0.48) ≈ 8.5 × 0.58 ≈ 4.93 m³ 3 / t.

[0106] The fourth step is to use the original gas content value determined by the geological survey report as the residual gas content for the original undisturbed area.

[0107] The original undisturbed area was not affected by the mechanical disturbance and stress relief of mining activities. The coal and rock mass was in the original stress state, with low degree of fracture development and stable gas occurrence, and no obvious gas escape occurred.

[0108] Specifically, the original gas content value determined in the previous geological survey report is directly used as the residual gas content of all spatial coordinate points in the area.

[0109] For example, a geological survey report, through borehole sampling tests, determined that the original gas content at the target mining face was 8.5 m³. 3 / t. The residual gas content at any spatial point (X: 1265m, Y: 820m, Z: -560m) within the original undisturbed zone is taken as 8.5m. 3 / t.

[0110] Furthermore, in step S500 of this application embodiment, based on the comprehensive conductivity coefficient, the dynamic permeability, the absolute value of the gas pressure gradient, and the residual gas content at multiple spatial coordinate points of the target mining face, the dynamic gas emission source strength at multiple spatial coordinate points needs to be calculated; including: A preset source strength calculation mathematical model is established, in which the dynamic gas outburst source strength is positively correlated with the comprehensive conductivity coefficient, the dynamic permeability, the absolute value of the gas pressure gradient, and the residual gas content.

[0111] Specifically, a source strength calculation mathematical model needs to be constructed, which includes four core parameters: comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content.

[0112] For example, the basic form of the mathematical model for calculating gas source strength can be expressed as: Dynamic gas emission source strength = k × Comprehensive conductivity coefficient × Dynamic permeability × Absolute value of gas pressure gradient × Residual gas content. Where k is a correction coefficient, which needs to be calibrated based on on-site conditions and coal and rock characteristics, and its value range can be set to 0.01-0.05.

[0113] Assume the overall conductivity of a spatial point in a certain mining pressure relief influence zone is 0.08m. 2 / (MPa 2 •d), the dynamic permeability is 2.39 × 10 -18 m 2 The absolute value of the gas pressure gradient is 0.5 MPa / m, and the residual gas content is 2.1 m³. 3 / t. Set the correction factor k to 0.03.

[0114] Substituting the parameters into the model for calculation: Dynamic gas emission source strength = 0.03 × 0.08 × 2.39 × 10 -18 m 2 ×0.5×2.1≈0.06×10 -18 m 2 / (min・m 2 ).

[0115] Using the same method, calculate the dynamic gas outburst source strength at all spatial coordinate points of the target mining face.

[0116] The dynamic gas emission source strength field is obtained based on multiple dynamic gas emission source strengths.

[0117] This involves summarizing the calculated gas dynamic emission source strength results for all spatial coordinate points. The source strength values ​​at each point are then mapped to their corresponding three-dimensional spatial locations to generate a gas dynamic emission source strength field that is spatially isomorphic to the target mining face.

[0118] The dynamic gas emission source strength field can intuitively show the strength distribution of gas emission within the mining face. For example, the source strength value is high in the pressure relief influence area and low in the original undisturbed area.

[0119] In step S600 of this embodiment, the gas dynamic emission source intensity field is used as input, and combined with historical gas concentration time series data, a spatiotemporal prediction model is used to predict the gas concentration distribution for future periods, including: A spatiotemporal prediction model is constructed, which is a hybrid model composed of a long short-term memory network and a convolutional neural network; Construct a spatiotemporal input feature vector, which includes historical gas concentration time series data and dynamic gas outburst source intensity field; The spatiotemporal input feature vector is reorganized into a feature matrix with spatial dimension, which is then input into a convolutional neural network module to obtain a high-level spatial feature map. The advanced spatial feature map is expanded along the time dimension and input into the long short-term memory network module, outputting a sequence of predicted continuous gas concentration values ​​for each monitoring point within a preset future time period.

[0120] In this embodiment of the application, the purpose of step S600 is to accurately predict the gas concentration distribution of the target mining face in the future period based on the dynamic gas outburst source strength field and historical gas concentration time series data, and to provide data support for the advanced prevention and control of coal mine gas and safety decision-making.

[0121] To achieve the above objectives, it is first necessary to build a spatiotemporal prediction model, which is a hybrid model composed of a long short-term memory network and a convolutional neural network. Among them, the convolutional neural network module is responsible for extracting the spatial correlation features of gas data, and the long short-term memory network module is responsible for capturing the temporal evolution of gas concentration.

[0122] A convolutional neural network module includes an input layer, convolutional layers, activation functions, pooling layers, and flattening layers.

[0123] The dimensions of the input layer match the size of the feature matrix, for example, set to (100, 100, 2). 100×100 corresponds to the spatial grid of the mining face, and 2 corresponds to the two types of features: dynamic gas outburst source strength and historical gas concentration.

[0124] The convolutional layers consist of two 2D convolutions. The first layer has a 3×3 kernel size, 32 kernels, a stride of 1, and the same padding. The second layer has a 3×3 kernel size, 64 kernels, a stride of 1, and the same padding.

[0125] The ReLU activation function is chosen to avoid the gradient vanishing problem.

[0126] Max pooling was used as the pooling layer, with a 2×2 kernel size and a stride of 2. The 64×50×50 feature map was reduced in dimensionality to preserve key spatial features.

[0127] Flattening layers are used to transform pooled 3D feature maps into 1D vectors. For example, a 64×50×50 feature map can be transformed into a 160,000-dimensional vector to be fed into the input layer of a Long Short-Term Memory (LSTM) network.

[0128] The Long Short-Term Memory (LSTM) network module includes a hidden layer, a Dropout layer, a fully connected layer, and an output layer.

[0129] The hidden layers consist of two long short-term memory (LSTM) networks, with 128 neurons in each layer. Tanh is used as the activation function, and sigmoid is used as the gating function.

[0130] A dropout layer with a coefficient of 0.2 is added between the two long short-term memory (LSM) network layers to prevent overfitting.

[0131] Set up a single fully connected layer with the number of neurons matching the number of monitoring points. For example, set the number of neurons to 120, corresponding to 120 monitoring points on the mining face.

[0132] The output layer uses a linear activation function, and the output dimension is (prediction duration / temporal resolution, 120). For example, if the prediction duration is 6 hours and the temporal resolution is 10 minutes, the output dimension is (36, 120).

[0133] The mean squared error loss function was used for both the Long Short-Term Memory network and the Convolutional Neural Network. The Adam optimizer was selected as the optimizer, the learning rate was set to 0.001, and the training batch size was set to 32.

[0134] The training data for the hybrid model consists of historical monitoring data from the target mining face. This may include multi-source monitoring data from the past 6-12 months, dynamic gas emission source intensity field data, and historical gas concentration time-series data. The training, validation, and test sets are divided in a 7:2:1 ratio and used for training the hybrid model.

[0135] When the loss value on the validation set no longer decreases for 10 consecutive rounds, and the difference between the loss values ​​on the training set and the validation set is less than 0.0001, the model is considered to have converged, training is terminated, and the hybrid model is obtained.

[0136] Next, a spatiotemporal input feature vector is constructed, which includes historical gas concentration time series data and dynamic gas outburst source intensity field; For example, select gas concentration time series data from the past 48 hours, with a time resolution of 10 minutes, totaling 288 time points. Extract the average concentration value at each time point, and combine it with the gas dynamic emission source strength field data of the current time at a 100×100 grid to construct a feature vector with both spatiotemporal attributes, which serves as the spatiotemporal input feature vector.

[0137] Then, the spatiotemporal input feature vector is reorganized into a feature matrix with spatial dimensions, input into the convolutional neural network module, and output to obtain a high-level spatial feature map; The spatiotemporal input feature vectors are then reorganized into a feature matrix according to the spatial grid of the mining face. For example, source strength data and historical average gas concentration data are integrated to form a 100×100×2 feature matrix.

[0138] The feature matrix is ​​input into the convolutional neural network module, and after convolution, activation, and pooling operations, it outputs a high-level spatial feature map. For example, it outputs a 64×50×50 feature map, which can highlight the spatial clustering characteristics of high gas outburst areas.

[0139] Finally, the advanced spatial feature map is expanded along the time dimension and input into the long short-term memory network module to output a sequence of predicted continuous gas concentration values ​​for each monitoring point within a preset future time period.

[0140] The high-level spatial feature map is transformed into a one-dimensional temporal vector through a flattening layer, and then expanded along the time dimension before being input into the Long Short-Term Memory (LSTM) network module. The LSM network module learns the temporal patterns of historical data through a gating mechanism and outputs a sequence of predicted concentration values ​​for a preset future time period.

[0141] For example, with a prediction duration of 6 hours and a time resolution of 10 minutes, the Long Short-Term Memory Network outputs the predicted gas concentration values ​​for 36 future time points from 120 monitoring points, forming a prediction sequence of (36, 120).

[0142] Finally, the predicted value sequence output by the Long Short-Term Memory Network is mapped back to the three-dimensional grid of the mining face according to spatial coordinates, and a heat map of gas concentration distribution at different future times is generated to intuitively present the spatiotemporal change trend of high-concentration risk areas.

[0143] Example 2, as Figure 2 As shown, based on the same inventive concept as the LSTM-based coal mine gas concentration time-series prediction method provided in Embodiment 1, this embodiment of the invention also provides an LSTM-based coal mine gas concentration time-series prediction system, including: The multi-source monitoring data acquisition module 11 is used to collect multi-source monitoring data of the target mining face. The multi-source monitoring data includes at least mining machinery posture time sequence data, microseismic monitoring event sequence and roadway environment monitoring data. The permeability equation construction module 12 is used to obtain geological information of the target mining face and construct the damage-permeability evolution equation of the target mining face based on the geological information. The stress damage distribution inference module 13 is used to obtain the dynamic stress damage field of the coal and rock mass based on the current time based on the time series data of the mining machinery pose and the sequence of microseismic monitoring events and the geological information. The dynamic permeability distribution calculation module 14 is used to calculate the dynamic permeability field of gas based on the dynamic stress damage field of the coal and rock mass and in combination with the preset damage-permeability evolution equation. The dynamic emission source strength calculation module 15 is used to calculate the dynamic emission source strength field of gas at the current moment based on the gas dynamic permeability field, the gas pressure data in the roadway environment monitoring data, and the coal body gas content distribution updated according to the mining progress, through a preset source strength calculation rule. The gas concentration distribution prediction module 16 is used to predict the gas concentration distribution in the future period by taking the gas dynamic emission source strength field as input, combining historical gas concentration time series data, and using a spatiotemporal prediction model.

[0144] Furthermore, the multi-source monitoring data acquisition module 11 includes the following execution steps: Positioning and attitude sensors installed on the mining machinery are used to collect the timing data of the mining machinery's posture at a first sampling frequency. The timing data of the mining machinery's posture includes at least three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed. By deploying a microseismic sensor array on the target mining face and surrounding rock, a sequence of microseismic monitoring events is collected at a second sampling frequency. Each microseismic monitoring event in the sequence includes at least the occurrence time, spatial coordinates, and released energy. Environmental monitoring data of the tunnel is collected at a third sampling frequency by environmental sensors installed in the tunnel. The environmental monitoring data of the tunnel includes at least gas concentration, gas pressure and wind speed. A unified spatiotemporal coordinate system is established, and the time sequence data of the mining machinery pose, the sequence of microseismic monitoring events, and the roadway environmental monitoring data are time-stamped and spatially registered to form synchronized multi-source monitoring data.

[0145] Furthermore, the permeability equation construction module 12 includes the following execution steps: Obtain geological information and coal and rock mass test data of the target mining face; The basic original permeability value of the coal and rock mass is obtained based on the geological information. Based on the coal and rock mass test data, the coupling coefficient was obtained by fitting. The damage coefficient of the target mining face is extracted from the dynamic stress damage field of the coal and rock mass; Based on the original permeability baseline value, the coupling coefficient, and the damage coefficient, the damage-permeability evolution equation is constructed.

[0146] Furthermore, the stress damage distribution simulation module 13 includes the following execution steps: An initial geomechanical numerical model of the target mining face is established based on the geological information; The timing data of the mining machinery's pose is converted into dynamic boundary load inputs for the initial geomechanical numerical model; The spatial coordinates and energy release distribution of the microseismic monitoring event sequence within the current time window are used as the observation data input to the data assimilation algorithm; Using the data assimilation algorithm, the initial geomechanical numerical model is updated with the observation data, so that the theoretical microseismic energy field output by the initial geomechanical numerical model matches the observation data, and the dynamic stress damage field of the coal and rock mass containing stress values ​​and damage coefficients at various points in space is inverted and output.

[0147] Furthermore, the dynamic permeability distribution calculation module 14 includes the following execution steps: Substitute the damage coefficient at each spatial location in the dynamic stress damage field of the coal and rock mass into the damage-permeability evolution equation; The dynamic permeability of gas at each spatial point is calculated in parallel to generate a dynamic permeability field of gas that is spatially isomorphic to the dynamic stress damage field.

[0148] Furthermore, the dynamic outflow source strength calculation module 15 includes the following execution steps: Based on the geological information of the target mining face and the roadway environment monitoring data, the comprehensive conductivity coefficient of multiple spatial coordinate points of the target mining face is obtained; The dynamic permeability of multiple spatial coordinate points of the target mining face is obtained from the gas dynamic permeability field; Based on the roadway environmental monitoring data, the absolute values ​​of gas pressure gradients at multiple spatial coordinate points of the target mining face are obtained; Based on the time sequence data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated. Based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content of multiple spatial coordinate points of the target mining face, the dynamic gas outburst source strength of multiple spatial coordinate points is calculated. The dynamic gas emission source strength field is obtained based on multiple dynamic gas emission source strengths.

[0149] The dynamic gas emission source strength is calculated based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content at multiple spatial coordinate points of the target mining face, including: A preset source strength calculation mathematical model is established, in which the dynamic gas outburst source strength is positively correlated with the comprehensive conductivity coefficient, the dynamic permeability, the absolute value of the gas pressure gradient, and the residual gas content.

[0150] Among them, based on the temporal data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated, including: Based on the time sequence data of the mining machinery's position, combined with the geological information of the target mining face, the coal body of the target mining face is divided into three dynamic regions in three-dimensional space: the mined-out area, the current mining pressure relief influence area, and the original undisturbed area. For mined-out areas, the residual gas content is set to zero; For the current mining and depressurization impact zone, based on the time elapsed since the start of mining-induced depressurization and the spatial distance between the current mining and depressurization impact zone and the mining face, the real-time residual gas content of the current mining and depressurization impact zone is calculated using a preset dynamic gas content decay function. The dynamic gas content decay function includes a time decay factor and a spatial decay factor. For the original undisturbed area, the original gas content value determined by the geological survey report is used as the residual gas content.

[0151] Furthermore, the gas concentration distribution prediction module 16 includes the following execution steps: A spatiotemporal prediction model is constructed, which is a hybrid model composed of a long short-term memory network and a convolutional neural network; Construct a spatiotemporal input feature vector, which includes historical gas concentration time series data and dynamic gas outburst source intensity field; The spatiotemporal input feature vector is reorganized into a feature matrix with spatial dimension, which is then input into a convolutional neural network module to obtain a high-level spatial feature map. The advanced spatial feature map is expanded along the time dimension and input into the long short-term memory network module, outputting a sequence of predicted continuous gas concentration values ​​for each monitoring point within a preset future time period.

Claims

1. A time-series prediction method for coal mine gas concentration based on LSTM, characterized in that, The method includes: Collect multi-source monitoring data of the target mining face, wherein the multi-source monitoring data includes at least mining machinery position and timing data, microseismic monitoring event sequences, and roadway environmental monitoring data; Obtain geological information of the target mining face, and construct a damage-permeability evolution equation for the target mining face based on the geological information; Based on the time-series data of the mining machinery's pose and the sequence of microseismic monitoring events, the dynamic stress damage field of the coal and rock mass with the current time as the reference is obtained by inverting the geological information. Based on the dynamic stress damage field of the coal and rock mass, and combined with the preset damage-permeability evolution equation, the dynamic permeability field of gas is calculated. Based on the gas dynamic permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal body gas content distribution updated according to the mining progress, the gas dynamic outburst source strength field at the current moment is obtained by solving the preset source strength calculation rules. Using the gas dynamic emission source intensity field as input, and combining historical gas concentration time series data, a spatiotemporal prediction model is used to predict the gas concentration distribution in future periods.

2. The LSTM-based time-series prediction method for coal mine gas concentration according to claim 1, characterized in that, Collect multi-source monitoring data of the target mining face. The multi-source monitoring data includes at least the temporal data of the mining machinery's position, microseismic monitoring event sequences, and roadway environmental monitoring data, including: Positioning and attitude sensors installed on the mining machinery are used to collect the timing data of the mining machinery's posture at a first sampling frequency. The timing data of the mining machinery's posture includes at least three-dimensional spatial coordinates, cutting drum rotation speed, and travel speed. By deploying a microseismic sensor array on the target mining face and surrounding rock, a sequence of microseismic monitoring events is collected at a second sampling frequency. Each microseismic monitoring event in the sequence includes at least the occurrence time, spatial coordinates, and released energy. Environmental monitoring data of the tunnel is collected at a third sampling frequency by environmental sensors installed in the tunnel. The environmental monitoring data of the tunnel includes at least gas concentration, gas pressure and wind speed. A unified spatiotemporal coordinate system is established, and the time sequence data of the mining machinery pose, the sequence of microseismic monitoring events, and the roadway environmental monitoring data are time-stamped and spatially registered to form synchronized multi-source monitoring data.

3. The time-series prediction method for coal mine gas concentration based on LSTM according to claim 1, characterized in that, Obtain geological information of the target mining face, and construct a damage-permeability evolution equation for the target mining face based on the geological information, including: Obtain geological information and coal and rock mass test data of the target mining face; The basic original permeability value of the coal and rock mass is obtained based on the geological information. Based on the coal and rock mass test data, the coupling coefficient was obtained by fitting. The damage coefficient of the target mining face is extracted from the dynamic stress damage field of the coal and rock mass; Based on the original permeability baseline value, the coupling coefficient, and the damage coefficient, the damage-permeability evolution equation is constructed.

4. The LSTM-based time-series prediction method for coal mine gas concentration according to claim 1, characterized in that, Based on the time-series data of the mining machinery's pose and the sequence of microseismic monitoring events, the dynamic stress-damage field of the coal and rock mass, with the current time as the baseline, is obtained by inversion using the geological information, including: An initial geomechanical numerical model of the target mining face is established based on the geological information; The timing data of the mining machinery's pose is converted into dynamic boundary load inputs for the initial geomechanical numerical model; The spatial coordinates and energy release distribution of the microseismic monitoring event sequence within the current time window are used as the observation data input to the data assimilation algorithm; Using the data assimilation algorithm, the initial geomechanical numerical model is updated with the observation data, so that the theoretical microseismic energy field output by the initial geomechanical numerical model matches the observation data, and the dynamic stress damage field of the coal and rock mass containing stress values ​​and damage coefficients at various points in space is inverted and output.

5. The LSTM-based time-series prediction method for coal mine gas concentration according to claim 1, characterized in that, Based on the dynamic stress damage field of the coal and rock mass, and combined with the preset damage-permeability evolution equation, the dynamic permeability field of gas is calculated, including: Substitute the damage coefficient at each spatial location in the dynamic stress damage field of the coal and rock mass into the damage-permeability evolution equation; The dynamic permeability of gas at each spatial point is calculated in parallel to generate a dynamic permeability field of gas that is spatially isomorphic to the dynamic stress damage field.

6. The LSTM-based time-series prediction method for coal mine gas concentration according to claim 1, characterized in that, Based on the dynamic gas permeability field, the gas pressure data in the roadway environmental monitoring data, and the coal seam gas content distribution updated according to the mining progress, the dynamic gas outburst source strength field at the current moment is obtained by solving a preset source strength calculation rule, including: Based on the geological information of the target mining face and the roadway environment monitoring data, the comprehensive conductivity coefficient of multiple spatial coordinate points of the target mining face is obtained; The dynamic permeability of multiple spatial coordinate points of the target mining face is obtained from the gas dynamic permeability field; Based on the roadway environmental monitoring data, the absolute values ​​of gas pressure gradients at multiple spatial coordinate points of the target mining face are obtained; Based on the time sequence data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face is calculated. Based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content of multiple spatial coordinate points of the target mining face, the dynamic gas outburst source strength of multiple spatial coordinate points is calculated. The dynamic gas emission source strength field is obtained based on multiple dynamic gas emission source strengths.

7. The LSTM-based time-series prediction method for coal mine gas concentration according to claim 6, characterized in that, Based on the comprehensive conductivity coefficient, dynamic permeability, absolute value of gas pressure gradient, and residual gas content at multiple spatial coordinate points of the target mining face, the dynamic gas emission source strength is calculated, including: A preset source strength calculation mathematical model is established, in which the dynamic gas outburst source strength is positively correlated with the comprehensive conductivity coefficient, the dynamic permeability, the absolute value of the gas pressure gradient, and the residual gas content.

8. The LSTM-based time-series prediction method for coal mine gas concentration according to claim 6, characterized in that, Based on the temporal data of the mining machinery's position and the geological information of the target mining face, the residual gas content at multiple spatial coordinate points of the target mining face was calculated, including: Based on the time sequence data of the mining machinery's position, combined with the geological information of the target mining face, the coal body of the target mining face is divided into three dynamic regions in three-dimensional space: the mined-out area, the current mining pressure relief influence area, and the original undisturbed area. For mined-out areas, the residual gas content is set to zero; For the current mining and depressurization impact zone, based on the time elapsed since the start of mining-induced depressurization and the spatial distance between the current mining and depressurization impact zone and the mining face, the real-time residual gas content of the current mining and depressurization impact zone is calculated using a preset dynamic gas content decay function. The dynamic gas content decay function includes a time decay factor and a spatial decay factor. For the original undisturbed area, the original gas content value determined by the geological survey report is used as the residual gas content.

9. The time-series prediction method for coal mine gas concentration based on LSTM according to claim 1, characterized in that, Using the gas dynamic emission source intensity field as input, and combining historical gas concentration time series data, a spatiotemporal prediction model is used to predict the gas concentration distribution for future periods, including: A spatiotemporal prediction model is constructed, which is a hybrid model composed of a long short-term memory network and a convolutional neural network; Construct a spatiotemporal input feature vector, which includes historical gas concentration time series data and dynamic gas outburst source intensity field; The spatiotemporal input feature vector is reorganized into a feature matrix with spatial dimension, which is then input into a convolutional neural network module to obtain a high-level spatial feature map. The advanced spatial feature map is expanded along the time dimension and input into the long short-term memory network module, outputting a sequence of predicted continuous gas concentration values ​​for each monitoring point within a preset future time period.

10. A time-series prediction system for coal mine gas concentration based on LSTM, characterized in that, The system is used to implement the LSTM-based time-series prediction method for coal mine gas concentration as described in any one of claims 1-9, and the system comprises: The multi-source monitoring data acquisition module is used to collect multi-source monitoring data of the target mining face. The multi-source monitoring data includes at least mining machinery position and timing data, microseismic monitoring event sequences, and roadway environmental monitoring data. The permeability equation construction module is used to obtain geological information of the target mining face and construct the damage-permeability evolution equation of the target mining face based on the geological information. The stress damage distribution inference module is used to obtain the dynamic stress damage field of the coal and rock mass based on the current time, according to the geological information, based on the time series data of the mining machinery's pose and the sequence of microseismic monitoring events. The dynamic permeability distribution calculation module is used to calculate the dynamic permeability field of gas based on the dynamic stress damage field of the coal and rock mass and in combination with the preset damage-permeability evolution equation. The dynamic emission source strength calculation module is used to calculate the dynamic emission source strength field of gas at the current moment based on the dynamic gas permeability field, the gas pressure data in the roadway environment monitoring data, and the coal body gas content distribution updated according to the mining progress, through a preset source strength calculation rule. The gas concentration distribution prediction module is used to predict the gas concentration distribution in the future period by taking the gas dynamic emission source strength field as input, combining historical gas concentration time series data, and using a spatiotemporal prediction model.

Images