Coal spontaneous combustion time sequence early warning method based on goaf multi-modal signal

Abstract

The application discloses a coal spontaneous combustion timing early warning method based on a goaf multi-modal signal, and particularly relates to the following steps: establishing a coal mine goaf multi-modal signal monitoring system; collecting the concentration of gas in the coal mine goaf; collecting the temperature of the temperature measurement multi-mode optical fiber in the coal mine goaf; dividing the coal spontaneous combustion three zones in real time based on multi-index gas concentration; judging whether there is a coal spontaneous combustion hidden danger in the goaf at present, analyzing whether there is a coal spontaneous combustion hidden danger in the goaf by setting a threshold and a coal spontaneous combustion hidden danger discrimination model; reading the distributed optical fiber temperature measurement temperature value, calculating the next moment residual coal temperature through a goaf coal spontaneous combustion temperature early warning model; and judging whether there is a coal spontaneous combustion hidden danger in the goaf at the next moment. Through the construction of a multi-modal signal fusion dynamic monitoring system, the accuracy of the goaf coal spontaneous combustion risk identification is significantly improved. Meanwhile, through the optimized timing prediction model and multi-stage early warning mechanism, the advanced prevention and control of the coal spontaneous combustion risk is realized.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety monitoring technology, specifically relating to a method for early warning of coal spontaneous combustion based on multimodal signals from goaf areas. Background Technology

[0002] Spontaneous combustion of residual coal in goaf areas during coal mining has long threatened mine safety. Due to their enclosed space and complex ventilation, goaf areas are prone to oxidation reactions under the influence of oxygen, humidity, and heat storage conditions, leading to spontaneous combustion disasters. This results in the loss of coal and environmental resources and can even threaten the lives of underground workers. Traditional monitoring methods, such as single-point temperature sensors or bundled tube gas analysis, have limitations such as limited monitoring range and fragmented multi-source data, making it difficult to dynamically capture the spatiotemporal evolution of the heat dissipation zone, oxidation heating zone, and asphyxiation zone, leading to delayed early warnings. For example, single temperature monitoring has a small range and cannot reflect the trend of gas concentration changes, while relying solely on indicators such as CO in corner locations is easily disturbed by factors such as ventilation, residual coal quantity, and mining speed. Therefore, developing a time-series early warning technology that fuses multi-modal signals of spontaneous combustion in goaf areas is of great practical significance for improving the inherent safety level of coal mines.

[0003] Coal spontaneous combustion monitoring and early warning technologies exhibit differentiated development paths in hardware integration and algorithm innovation. Existing research focuses on the construction of multi-source data acquisition systems. A Chinese patent (application number: 202411148361.0, publication number: CN 119062398A, publication date: 2024.12.03) discloses a multi-source information intelligent monitoring and early warning method and system for coal spontaneous combustion and gas migration in goaf areas. This method collects temperature and gas concentration data from goaf areas by deploying temperature and gas sensors, and combines this with preset thresholds to determine the risk of coal spontaneous combustion. However, it relies solely on fixed temperature and gas concentration thresholds for risk assessment, failing to correct for the boundaries of the three zones (goaf, gas field, and gas field), and not incorporating parameters such as roadway width and measuring point depth to reconstruct the time-series early warning model. Therefore, it cannot determine the dangerous range of coal spontaneous combustion in goaf areas. It is necessary to comprehensively consider the impact of mining data on the development trend of coal spontaneous combustion to determine the scope of influence and future development trends, thereby effectively improving the accuracy of coal spontaneous combustion early warning. Summary of the Invention

[0004] The purpose of this invention is to provide a coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas, so as to solve the problems of single monitoring dimensions and delayed early warning in the existing technology.

[0005] The technical solution adopted in this invention is a coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas, which is implemented according to the following steps: Step 1: Establish a multi-modal signal monitoring system for coal mine goaf areas; Step 2: Collect the concentration of gases in the coal mine goaf, specifically including O2, CO, CH4, C2H4, and C2H6; collect the temperature of the temperature-measuring multimode optical fiber in the coal mine goaf. Step 3: Based on the gas concentration of multiple indicators, divide the three zones of spontaneous combustion of coal in real time; Step 4: Determine whether there is a risk of spontaneous combustion of coal in the goaf. Analyze whether there is a risk of spontaneous combustion of coal in the goaf by setting thresholds and using a coal spontaneous combustion risk discrimination model. Step 5: Read the temperature value measured by the distributed optical fiber temperature measurement, and calculate the temperature of the remaining coal at the next moment using the coal spontaneous combustion temperature early warning model for the goaf area; Step 6: Determine whether there is a risk of spontaneous combustion of coal in the goaf at the next moment.

[0006] The invention is further characterized in that, In step 1, the multimodal signal monitoring system for the goaf includes an intrinsically safe distributed fiber optic temperature measurement host installed outside the intake airway door. The intrinsically safe distributed fiber optic temperature measurement host is connected to two temperature-measuring multimode optical fibers, located near the coal pillars on both sides of the intake and return airways, passing through the lower and upper corners, and measuring the temperature of the coal pillars near both sides of the goaf at 1-meter intervals. Intrinsically safe multi-index gas wireless sensors are arranged on both sides of the coal pillars in the goaf, with a measuring point every 5 meters, to measure the concentration of various gases in the goaf. The intrinsically safe multi-index gas wireless sensors transmit signals to the wireless sensor host through a self-organizing network. The wireless sensor host is connected to a switch via a network cable, and finally the monitoring signals are transmitted to the ground industrial control computer through a 10-gigabit optical fiber.

[0007] Step 3 involves the following steps: Step 3.1: Real-time reading of the concentrations of O2 and C2H4 in the intrinsically safe multi-index gas wireless sensor; Step 3.2: Set the O2 concentration thresholds for the three zones of spontaneous combustion of coal in the goaf. Set the O2 concentration threshold a for the heat dissipation zone to 18% and the O2 concentration threshold b for the asphyxiation zone to 8%. Step 3.3: Based on the two O2 concentration thresholds set in Step 2.2, compare the O2 monitoring concentrations in the intake airway, return airway, and goaf area sequentially to obtain the location of the intrinsically safe multi-index gas wireless sensor whose actual monitoring value is closest to the threshold, i.e., the D value of the O2 concentration of 18% in the intake airway. ja The O2 concentration in the intake airway is 8% D jb The O2 concentration in the return airway is 18% D ha The O2 concentration in the return airway is 8% D hb ; Step 3.4, Determine D jb D hbCheck if the intrinsically safe multi-index gas wireless sensor at the current location can detect C2H4 gas; if it can, extend the location of another intrinsically safe multi-index gas wireless sensor deeper into the goaf, and update D. jb D hb If the location cannot be detected, proceed to step 3.4; if not, proceed to step 3.5. Step 3.5, Connect D ja With D ha D jb With D hb Draw auxiliary lines to connect the locations; Step 3.6, D ja D ha The upper and lower corners form the heat dissipation zone of the goaf, D. ja D ha D jb With D hb The area is the oxidation and heating zone of the goaf, D jb D hb The deeper goaf is considered as the asphyxiation zone, thus defining the three zones of spontaneous combustion of coal in the goaf.

[0008] Step 4 involves the following steps: Step 4.1: Determine whether C2H4 gas is present in the intrinsically safe multi-index gas wireless sensor monitored in the oxidation heating zone. If it is present, an alarm will be triggered and the emergency plan for spontaneous combustion of coal in the goaf will be executed. If it is not present, proceed to step 3.2. Step 4.2: Set the coal fire temperature threshold λ in the goaf to 60.0℃; Step 4.3: Determine whether the distributed temperature measurement fiber optic monitoring value in the oxidation heating zone exceeds the coal fire temperature threshold λ; if it exceeds, an alarm will be triggered; if it does not exceed, proceed to step 4.4. Step 4.4: Obtain the highest temperature point T1 of the distributed fiber optic temperature measurement system in the goaf and the CO concentration N1 of the nearest intrinsically safe multi-index gas wireless sensor; obtain the highest CO concentration point N2 of the intrinsically safe multi-index gas wireless sensor and the nearest fiber optic temperature measurement signal temperature point T2. Input these two sets of values ​​into the coal spontaneous combustion hazard discrimination model. The inputs to the coal spontaneous combustion hazard discrimination model are the monitored temperature, CO monitoring concentration, and lower corner temperature. The model is calculated using a quantification model, with the specific formula as follows:

[0009] Where Q is the dimensionless value output by the quantization model; T i With N i The values ​​are distributed fiber optic temperature and CO concentration at the same location, in °C and ppm. t The average temperature at the lower corner is expressed in °C. Step 4.5: Determine whether the output of the coal spontaneous combustion hazard identification model exceeds the set threshold, which is set to 1; if it exceeds, an alarm will be triggered; if it does not exceed, continue to execute steps 4.1-4.4.

[0010] Step 5 involves the following steps: Step 5.1: Read the highest point of the distributed optical fiber temperature signal in the goaf area, i.e., read the signal from step 4.4. T 1; Step 5.2: The temperature obtained in step 5.1 T 1. Lane width and burial depth, T The historical temperature data of the location after it is buried in the goaf and the corresponding burial depth of the measuring point are used as inputs to the coal spontaneous combustion temperature early warning model; the temperature of the remaining coal at the next moment is calculated.

[0011] The coal spontaneous combustion temperature early warning model is a time-series prediction model based on GRU, taking... T The temperature data point 1 is the maximum temperature point at 200°C. The feature values ​​are scaled to the [0,1] interval to form the model input data. When the input data is input into the GRU unit according to the time step, firstly, the current input x... t and the hidden state h from the previous moment t-1 The system will enter the reset gate and update gate in parallel, and calculate the gate control vector r through a fully connected layer and a sigmoid activation function, respectively. t and z t Perform the calculation; then calculate the candidate hidden state: at this point, reset the gate vector r. t with h t-1 Element-wise multiplication is performed, and data is selectively lost according to the forgetting coefficient; the modulated historical state is then compared with the current input x. t By combining a fully connected layer and a Tanh activation function, temporary candidate states h are generated. t Finally, the data converges to the final update stage: update gate z t It then acts as a dynamic mixer again, fusing the input from the previous GRU structure with the update results of the current GRU structure using a weighted average formula, and finally outputs the final hidden state h at the current time step. t The coal spontaneous combustion temperature early warning model outputs the predicted value of the goaf temperature, and uses the predicted value to determine whether the temperature will exceed the preset threshold at the next moment.

[0012] The beneficial effects of this invention are: (1) By constructing a dynamic monitoring system that integrates multimodal signals, the accuracy of identifying the risk of spontaneous combustion of coal in the goaf has been significantly improved. The distributed fiber optic temperature measurement host and the multi-index gas wireless sensor deployment scheme can simultaneously collect temperature field data and gas concentration information such as O2, CO, and C2H4, which solves the limitations of traditional single signal monitoring. The range of the "three zones" is initially divided by setting the O2 concentration threshold (18% for the heat dissipation zone and 8% for the asphyxiation zone), and the boundary is dynamically adjusted in combination with the C2H4 monitoring results, so that the definition of the oxidation heating zone can match the environmental changes in the goaf in real time, which is more in line with the actual engineering situation than the static division method. It provides a reliable basis for accurately locking the high-risk area of ​​spontaneous combustion and effectively avoids the problem of missed judgment or misjudgment.

[0013] (2) Through the optimized time-series prediction model and multi-level early warning mechanism, the risk of spontaneous combustion of coal has been prevented and controlled in advance. The parameters such as historical temperature, roadway width, and measurement point burial depth are input into the GRU time-series prediction model with Adam optimizer to accurately predict the temperature of key points at the next moment, which solves the problem of traditional early warning lag. At the same time, the presence or absence of C2H4 in the oxidation heating zone, the 60℃ temperature threshold, and the output results of the coal spontaneous combustion hazard discrimination model are combined to form a multi-level response logic, which comprehensively improves the prevention and control capabilities of spontaneous combustion fires in coal mine goaf areas. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the overall deployment of the multimodal signal monitoring system for goaf areas in the method of this invention. Figure 2 This is a detailed diagram showing the deployment of the intrinsically safe distributed optical fiber and multi-index gas wireless sensor in the method of this invention. Figure 3 This is a flowchart of the three zones of spontaneous combustion of coal in the goaf, defined based on gas concentration thresholds, in the method of this invention. Figure 4 This is a flowchart of the process for reading multimodal signals in the goaf and updating the distributed optical fiber temperature measurement dataset in the method of this invention; Figure 5 This is a flowchart illustrating the current coal spontaneous combustion hazard identification process in the goaf area using the method of this invention. Figure 6 This is a flowchart of the data input and next-moment temperature prediction process for the coal spontaneous combustion temperature early warning model in the goaf area in the method of this invention. Figure 7 This is a complete execution flowchart of the coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas in this invention. Figure 8 This is a structural diagram of the coal spontaneous combustion temperature early warning model in the method of this invention; Figure 9 This is a structural diagram of GRUCell in the method of this invention; Figure 10 This is a graph showing the variation of oxygen concentration in the return air of the goaf with burial depth; Figure 11 This is a graph showing the variation of oxygen concentration in the intake air of the goaf with burial depth; Figure 12 This is a graph showing the fitting equation for the oxygen concentration on the return air side; Figure 13 This is a graph showing the fitting equation for the oxygen concentration on the intake side; Figure 14 This is a map showing the oxygen volume fraction distribution in the goaf of the working face; Figure 15 This is a cross-sectional diagram showing the oxygen volume fraction distribution in various sections of the goaf of the working face; Figure 16 This is a map showing the distribution of the "three zones" in the goaf area of ​​the working face.

[0015] In the diagram, 1. Working face, 2. Coal pillar, 3. Intake airway, 4. Lower corner, 5. Return airway, 6. Upper corner, 7. Floor, 8. Air door, 9. Switch, 10. Wireless sensor host, 11. Intrinsically safe multi-index gas wireless sensor, 12. Intrinsically safe distributed fiber optic temperature measurement host, 13. Temperature measurement multimode fiber optic cable, 14. Goaf. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Example 1 This invention relates to a coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas, such as... Figure 1 As shown, please follow these steps: Step 1: Establish a multi-modal signal monitoring system for coal mine goaf areas; like Figure 2 As shown, the underground environment includes working face 1, coal pillar 2, intake airway 3, lower corner 4, return airway 5, upper corner 6, floor 7, air door 8, and switch 9; An intrinsically safe distributed fiber optic temperature measuring host 12 is installed outside the air door of intake airway 3. The intrinsically safe distributed fiber optic temperature measuring host 12 is connected to two temperature measuring multimode optical fibers 13. It is located near the coal pillars of intake airway 3 and return airway 5, passing through the lower corner 6 and upper corner 4, and measures the temperature of the coal pillars near both sides of the goaf 14 at a measuring point interval of 1m.

[0018] The intrinsically safe multi-index gas wireless sensor 11 is arranged on both sides of the coal pillar in the goaf 14, with a measuring point every 5m, to measure the concentration of various gases in the goaf.

[0019] The intrinsically safe distributed fiber optic temperature measurement host 12 is connected to the switch 9 in the air intake tunnel via a network cable, and finally transmits the monitoring signal to the ground industrial control computer via a 10 Gigabit fiber optic cable. like Figure 3As shown, the intrinsically safe multi-index gas wireless sensor 11 transmits signals to the wireless sensor host 10 through a self-organizing network. The wireless sensor host 10 is connected to the switch 9 via a network cable, and finally transmits the monitoring signals to the ground industrial control computer through a 10 Gigabit optical fiber.

[0020] Step 2: Collect data from the goaf area of ​​the coal mine; (1) Collect the concentration of gases in the goaf of the coal mine, including O2, CO, CH4, C2H4 and C2H6; (2) Collect the temperature of the multimode optical fiber for temperature measurement in the goaf of the coal mine; The data collected includes distributed fiber optic temperature measurement signals and wireless gas sensor signals. These signals are aggregated by the sensor host and then uploaded to a ground-based industrial control computer. The real-time collected data is stored on the ground-based industrial control computer and structured into multimodal datasets (fiber optic temperature, gas concentration) at 1-hour intervals, representing the same time-series intervals. Before being input into the GRU model, the datasets need to be normalized to ensure convergence during model computation, which is then used for subsequent data analysis and model calculations.

[0021] Step 3: Based on the multi-index gas concentrations measured by the intrinsically safe multi-index wireless gas sensor 11, the three zones of spontaneous combustion of coal are divided in real time, such as... Figure 4 As shown, the specific process is as follows: Step 3.1: Real-time reading of the concentrations of O2 and C2H4 in the intrinsically safe multi-index gas wireless sensor 11; Step 3.2: Set the O2 concentration thresholds for the three zones of spontaneous combustion of coal seam 14 in the goaf. Set the O2 concentration threshold a for the heat dissipation zone to 18% and the O2 concentration threshold b for the asphyxiation zone to 8%. Step 3.3: Based on the two O2 concentration thresholds set in Step 2.2, compare the O2 monitoring concentrations of intake airway 3, return airway 5, and goaf 14 sequentially to obtain the location of the intrinsically safe multi-index gas wireless sensor 11 whose actual monitoring value is closest to the threshold, i.e., the D value of the O2 concentration of 18% in intake airway 3. ja The O2 concentration in intake airway 3 is 8% D jb The O2 concentration in return airway 5 is 18% D ha The O2 concentration in return airway 5 is 8% D hb ; Step 3.4, Determine D jb D hb Check if the intrinsically safe multi-index gas wireless sensor 11 at the current location can detect C2H4 gas. If it can, extend the location of another intrinsically safe multi-index gas wireless sensor 11 deeper into the goaf area and update D. jb D hb If the location cannot be detected, proceed to step 3.4; if not, proceed to step 3.5. Step 3.5, Connect D ja With D ha D jb With D hb Draw auxiliary lines to connect the locations; Step 3.6, D ja D ha The area between the upper corner 6 and the lower corner 4 is the heat dissipation zone of the goaf. ja D ha D jb With D hb The area is the oxidation and heating zone of the goaf, D jb D hb The deeper goaf is considered as the asphyxiation zone, thus defining the three zones of spontaneous combustion of coal in the goaf.

[0022] Step 4: Determine whether there is a risk of spontaneous combustion of coal in the goaf 14: Based on the fiber optic temperature value and the gas concentration value of the intrinsically safe multi-index gas wireless sensor 11, analyze whether there is a risk of spontaneous combustion of coal in the goaf 14 by setting a threshold and using a coal spontaneous combustion risk discrimination model. like Figure 5 As shown, the specific process is as follows: Step 4.1: Determine whether C2H4 gas is present in the intrinsically safe multi-index gas wireless sensor 11 monitored in the oxidation heating zone. If present, an alarm is triggered and the emergency plan for spontaneous combustion of coal in the goaf is executed. If not present, proceed to step 3.2. Within the oxidative heating zone, the gas composition of the goaf is monitored by an intrinsically safe multi-index wireless gas sensor 11. When the sensor is working normally and detects that the C2H4 concentration is higher than the instrument's detection limit (i.e., the concentration is greater than 0 ppm), it is considered that C2H4 exists at that location.

[0023] Step 4.2: Set the threshold of the distributed fiber optic temperature measurement system, and set the coal fire temperature threshold λ in the goaf to 60.0℃; Step 4.3: Determine whether the distributed temperature measurement fiber optic monitoring value in the oxidation heating zone exceeds the coal fire temperature threshold λ. If it exceeds, an alarm is triggered, and the emergency plan for spontaneous combustion of coal in the goaf is executed; if it does not exceed, proceed to step 4.4. Step 4.4: Obtain the highest temperature point T1 of the distributed fiber optic temperature measurement system in the goaf and the CO concentration N1 of the nearest intrinsically safe multi-index gas wireless sensor 11; obtain the highest CO concentration point N2 of the intrinsically safe multi-index gas wireless sensor 11 and the nearest fiber optic temperature measurement signal temperature point T2. Input these two sets of values ​​into the coal spontaneous combustion hazard discrimination model. The inputs to the coal spontaneous combustion hazard discrimination model are the monitored temperature, CO monitoring concentration, and lower corner temperature. The model is calculated using a quantification model, with the specific formula as follows:

[0024] Where Q is the dimensionless value output by the quantization model; T i With N i The values ​​are distributed fiber optic temperature and CO concentration at the same location, in °C and ppm. t The average temperature of the lower corner is expressed in °C.

[0025] Step 4.5: Determine whether the output of the coal spontaneous combustion hazard identification model exceeds the set threshold, which is set to 1. If it exceeds the threshold, an alarm will be triggered, and the emergency plan for coal spontaneous combustion in the goaf will be executed; if it does not exceed the threshold, continue with steps 4.1-4.4.

[0026] Step 5: Calculate the temperature of the remaining coal at the next moment using the goaf coal spontaneous combustion temperature early warning model: Read the distributed fiber optic temperature measurement values ​​and use a deep learning model to extrapolate the temperature of the remaining coal at the same monitoring location; for example... Figure 7 As shown, the specific process is as follows: Step 5.1: Read the highest point of the distributed optical fiber temperature signal in the goaf area, i.e., read the signal from step 4.4. T 1; Step 5.2: The temperature obtained in step 5.1 T 1. Lane width and burial depth, T The historical temperature data after the location is buried in the goaf and the corresponding burial depth of the measuring point are used as inputs to the coal spontaneous combustion temperature early warning model. T 1 is used as the temperature data input, and the lane width and burial depth are used as spatial parameters. The combination of the two is then input into the temperature prediction model.

[0027] Temperature data ( T 1) is the value of the highest temperature point in the goaf, which directly affects the input of the coal spontaneous combustion temperature early warning model and determines whether the temperature exceeds the preset threshold. The two data of roadway width and burial depth are used to more accurately determine the temperature distribution in space. The coal spontaneous combustion temperature early warning model is a time series prediction model based on GRU (Gated Cyclic Unit). The data processing of the GRU temperature time series prediction model includes three main parts: input data normalization and integration, sequence construction and data standardization. The input data normalization and integration part combines the collected historical data into coal spontaneous combustion temperature characteristic values ​​using the following formula.

[0028] Eigenvalue = T 1× Lane width× Burial depth A normalized processor based on the training set, taking T The temperature data point 1 is the maximum temperature point at 200°C. The feature values ​​are scaled to the [0,1] interval using MinMaxScaler to form the model input data.

[0029] The sequence construction part mainly consists of a sliding window sequence generator, which comprises a fixed-time-step window sliding mechanism and a feature-label segmentation operation. The sliding window sequence generator is primarily controlled by two parameters: the historical time step and the prediction step. There may be an offset between the feature sequence output by the model and the label sequence of the historical data, which is used to achieve multi-step prediction tasks.

[0030] The historical time step is set to 10, the prediction step step is 1, and a sliding window with a step size of 2 is used to traverse the input data to generate a feature sequence X and a label sequence Y, where each X... i The dimension is 10×1, and each Y i The dimension is 1×1; When input data is input to the GRU unit according to time steps, such as Figure 8 and Figure 9 As shown, firstly, the current input x t and the hidden state h from the previous moment t-1 The system will enter the reset gate and update gate in parallel, and calculate the gate control vector r through a fully connected layer and a sigmoid activation function, respectively. t and z t Perform the calculation. Then calculate the candidate hidden state: at this point, reset the gate vector r. t with h t-1 Element-wise multiplication is performed, and data is selectively lost according to the forgetting coefficient; this modulated historical state is then compared with the current input x. t By combining a fully connected layer and a Tanh activation function, temporary candidate states h are generated. t Finally, the data converges to the final update stage: update gate z. t It then acts as a dynamic mixer again, fusing the input from the previous GRU structure with the update results of the current GRU structure using a weighted average formula, and finally outputs the final hidden state h at the current time step. t .

[0031] The data standardization part mainly involves inverting the predicted feature values ​​output by the prediction model into predicted temperature values. First, the predicted data is inversely normalized, and then the predicted feature values ​​calculated by inverse normalization are converted into predicted temperature values ​​using the following formula.

[0032] Predicted temperature value = Predicted characteristic value calculated by inverse normalization ÷ Lane width ÷ Burial depth The model outputs a predicted value for the temperature of the goaf, and uses the predicted value to determine whether the temperature will exceed a preset threshold at the next moment.

[0033] The training conditions are as follows: The data used for training consists of historical temperature data from the goaf area. T1) Composed of roadway width and burial depth. The model uses the Adam optimizer with a learning rate of 0.001. After training, the model can predict the temperature value of the goaf at the next moment based on the input temperature data and spatial feature parameters (such as roadway width and burial depth), and determine whether there is a risk of spontaneous combustion of coal based on the predicted value.

[0034] Step 6: Determine if there is a risk of spontaneous combustion of coal in the goaf at the next moment: Determine if the predicted temperature measured by distributed optical fiber temperature measurement at the next moment exceeds the threshold. If it does, an alarm is triggered; otherwise, continue reading the multi-modal monitoring signal of the goaf. Figure 6 As shown; The model's inputs are historical temperature data, roadway width, and burial depth. The model's output is the maximum temperature of the goaf at the next moment. This predicted value is compared with the set temperature threshold λ (60℃). If the predicted temperature value is greater than λ, it is considered that there is a risk of spontaneous combustion of coal, triggering an early warning. If it is less than λ, monitoring continues.

[0035] Example 2 To verify the effectiveness of the coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas in this invention, the following experimental design was conducted: The oxygen concentration data of the intake and return air sides collected by the buried pipes were sorted out and combined with the results obtained from the simulation; the results were verified by combining measured data, fitting rules, and simulation data. Data collection at the site: Wireless gas sensors were deployed in the goaf on both the intake and return air sides. The oxygen concentration trend with distance from the goaf is shown below. Figure 10 , Figure 11 As shown; the measured oxygen concentration threshold is consistent with the O2 concentration threshold (18% for heat dissipation zone) set in step 3 of this invention - 37m on the air inlet side (O2=18%), 26m on the air return side (O2=18%); 118m on the air inlet side (O2=7%), 96m on the air return side (O2=7%).

[0036] Figure 13 , Figure 14 The fitted curve demonstrates that the gas changes are continuous and predictable. Since the oxygen concentration-burial depth variation can be fitted, it means that the signal's spatial and temporal variations are predictable. The time-series model of the GRU in this invention utilizes this predictable signal trend to predict the temperature change at the next moment.

[0037] Example 3 Figure 15 This is a simulated oxygen volume fraction distribution map of the goaf in the working face. It can be seen that the oxygen distribution is more abundant outside and less abundant inside, which is consistent with the variation pattern measured on site.

[0038] Figure 16It is a cross-sectional diagram simulating the oxygen volume fraction distribution in various sections of the goaf of the working face, showing the changes in oxygen concentration in different sections and the way the oxygen concentration is presented in different sections is consistent; the boundaries can be obtained in different sections.

[0039] Example 4 Based on the data collected on-site, the "three zones" can be divided according to the method of this invention. The model collects on-site temperature field data and gas concentration information such as O2, CO, and C2H4, and combines them with the GRU model to predict the temperature rise trend, and the boundaries can be dynamically adjusted.

[0040] Example 5 This invention significantly improves the accuracy of identifying spontaneous combustion risks in goaf areas by constructing a dynamic monitoring system that integrates multimodal signals. It can simultaneously collect temperature field data and gas concentration information such as O2, CO, and C2H4, overcoming the limitations of traditional single-signal monitoring. By setting O2 concentration thresholds (18% for the heat dissipation zone and 8% for the asphyxiation zone), the system initially delineates the "three zones," and dynamically adjusts the boundaries based on C2H4 monitoring results, ensuring that the definition of the oxidation and heating zone matches real-time changes in the goaf environment. This multi-parameter linkage-based delineation mechanism provides a reliable basis for accurately identifying high-risk areas for spontaneous combustion, effectively avoiding missed or false alarms.

[0041] Example 6 This invention also achieves proactive prevention and control of coal spontaneous combustion risks through an optimized time-series prediction model and a multi-level early warning mechanism. By inputting parameters such as historical temperature, roadway width, and measuring point burial depth into a GRU time-series prediction model using an Adam optimizer (learning rate 0.001), the model can accurately predict the temperature of key points at the next moment, solving the problem of traditional early warning lag. Simultaneously, by combining the presence or absence of C2H4 in the oxidation heating zone, the 60℃ temperature threshold, and the output results of the coal spontaneous combustion hazard discrimination model (input temperature, CO concentration, and lower corner temperature), a multi-level response logic is formed, which can both determine the current hazard status and predict future risks. This method, through deep fusion of multimodal data and intelligent prediction, comprehensively improves the prevention and control capabilities of spontaneous combustion fires in coal mine goaf areas.

Claims

1. A method for early warning of spontaneous combustion of coal based on multimodal signals from goaf areas, characterized in that, The specific steps are as follows: Step 1: Establish a multi-modal signal monitoring system for coal mine goaf areas; Step 2: Collect the concentration of gases in the coal mine goaf, specifically including O2, CO, CH4, C2H4, and C2H6; collect the temperature of the temperature-measuring multimode optical fiber in the coal mine goaf. Step 3: Based on the gas concentration of multiple indicators, divide the three zones of spontaneous combustion of coal in real time; Step 4: Determine whether there is a risk of spontaneous combustion of coal in the goaf. Analyze whether there is a risk of spontaneous combustion of coal in the goaf by setting thresholds and using a coal spontaneous combustion risk discrimination model. Step 5: Read the temperature value measured by the distributed optical fiber temperature measurement, and calculate the temperature of the remaining coal at the next moment using the coal spontaneous combustion temperature early warning model for the goaf area; Step 6: Determine whether there is a risk of spontaneous combustion of coal in the goaf at the next moment.

2. The coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas as described in claim 1, characterized in that, In step 1, the coal mine goaf multimodal signal monitoring system includes an intrinsically safe distributed fiber optic temperature measurement host installed outside the intake airway door. The intrinsically safe distributed fiber optic temperature measurement host is connected to two temperature-measuring multimode optical fibers, close to both sides of the coal pillars in the intake and return airways, passing through the lower and upper corners, and measuring the temperature of the coal pillars near both sides in the goaf at 1-meter intervals. Intrinsically safe multi-index gas wireless sensors are arranged on both sides of the coal pillars in the goaf, with a measuring point every 5 meters, to measure the concentration of various gases in the goaf. The intrinsically safe multi-index gas wireless sensors transmit signals to the wireless sensor host through a self-organizing network. The wireless sensor host is connected to a switch via a network cable, and finally the monitoring signal is transmitted to the ground industrial control computer through a 10-gigabit optical fiber.

3. The method for early warning of spontaneous combustion of coal based on multimodal signals from goaf areas as described in claim 2, characterized in that, In step 3, the specific process is as follows: Step 3.1: Real-time reading of the concentrations of O2 and C2H4 in the intrinsically safe multi-index gas wireless sensor; Step 3.2: Set the O2 concentration thresholds for the three zones of spontaneous combustion of coal in the goaf. Set the O2 concentration threshold a for the heat dissipation zone to 18% and the O2 concentration threshold b for the asphyxiation zone to 8%. Step 3.3: Based on the two O2 concentration thresholds set in Step 2.2, compare the O2 monitoring concentrations in the intake airway, return airway, and goaf area sequentially to obtain the location of the intrinsically safe multi-index gas wireless sensor whose actual monitoring value is closest to the threshold, i.e., the D value of the O2 concentration of 18% in the intake airway. ja The O2 concentration in the intake airway is 8% D jb The O2 concentration in the return airway is 18% D ha The O2 concentration in the return airway is 8% D hb ; Step 3.4, Determine D jb D hb Check if the intrinsically safe multi-index gas wireless sensor at the current location can detect C2H4 gas; if it can, extend the location of another intrinsically safe multi-index gas wireless sensor deeper into the goaf, and update D. jb D hb If the location cannot be detected, proceed to step 3.4; if not, proceed to step 3.

5. Step 3.5, Connect D ja With D ha D jb With D hb Draw auxiliary lines to connect the locations; Step 3.6, D ja D ha The upper and lower corners form the heat dissipation zone of the goaf, D. ja D ha D jb With D hb The area is the oxidation and heating zone of the goaf, D jb D hb The deeper goaf is considered as the asphyxiation zone, thus defining the three zones of spontaneous combustion of coal in the goaf.

4. The coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas as described in claim 3, characterized in that, In step 4, the specific process is as follows: Step 4.1: Determine whether C2H4 gas is present in the intrinsically safe multi-index gas wireless sensor monitored in the oxidation heating zone. If it is present, an alarm will be triggered and the emergency plan for spontaneous combustion of coal in the goaf will be executed. If it is not present, proceed to step 3.

2. Step 4.2: Set the coal fire temperature threshold λ in the goaf to 60.0℃; Step 4.3: Determine whether the distributed temperature measurement fiber optic monitoring value in the oxidation heating zone exceeds the coal fire temperature threshold λ; if it exceeds, an alarm will be triggered; if it does not exceed, proceed to step 4.

4. Step 4.4: Obtain the highest temperature point T1 of the distributed fiber optic temperature measurement system in the goaf and the CO concentration N1 of the nearest intrinsically safe multi-index gas wireless sensor; obtain the highest CO concentration point N2 of the intrinsically safe multi-index gas wireless sensor and the nearest fiber optic temperature measurement signal temperature point T2. Input these two sets of values ​​into the coal spontaneous combustion hazard discrimination model. The inputs to the coal spontaneous combustion hazard discrimination model are the monitored temperature, CO monitoring concentration, and lower corner temperature. The model is calculated using a quantification model, with the specific formula as follows: Where Q is the dimensionless value output by the quantization model; T i With N i The values ​​are distributed fiber optic temperature and CO concentration at the same location, in °C and ppm. t The average temperature at the lower corner is expressed in °C. Step 4.5: Determine whether the output of the coal spontaneous combustion hazard identification model exceeds the set threshold, which is set to 1; if it exceeds, an alarm will be triggered; if it does not exceed, continue to execute steps 4.1-4.

4.

5. The coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas as described in claim 4, characterized in that, In step 5, the specific process is as follows: Step 5.1: Read the highest point of the distributed optical fiber temperature signal in the goaf area, i.e., read the signal from step 4.

4. T 1; Step 5.2: The temperature obtained in step 5.1 T 1. Lane width and burial depth, T The historical temperature data after the location is buried in the goaf and the corresponding burial depth of the measuring point are used as inputs to the coal spontaneous combustion temperature early warning model. Calculate the temperature of the remaining coal at the next moment.

6. The coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas as described in claim 5, characterized in that, The coal spontaneous combustion temperature early warning model is a time-series prediction model based on GRU, taking... T The temperature data point 1 is the maximum temperature point at 200°C. The feature values ​​are scaled to the [0,1] interval to form the model input data. When the input data is input into the GRU unit according to the time step, firstly, the current input x... t and the hidden state h from the previous moment t-1 The system will enter the reset gate and update gate in parallel, and calculate the gate control vector r through a fully connected layer and a sigmoid activation function, respectively. t and z t Perform the calculation; then calculate the candidate hidden state: at this point, reset the gate vector r. t with h t-1 Perform element-wise multiplication and selectively discard data according to the forgetting coefficient; The modulated historical state is then compared with the current input x. t By combining a fully connected layer and a Tanh activation function, temporary candidate states h are generated. t Finally, the data converges to the final update stage: update gate z t It then acts as a dynamic mixer again, fusing the input from the previous GRU structure with the update results of the current GRU structure using a weighted average formula, and finally outputs the final hidden state h at the current time step. t The coal spontaneous combustion temperature early warning model outputs the predicted value of the goaf temperature, and uses the predicted value to determine whether the temperature will exceed the preset threshold at the next moment.

7. The coal spontaneous combustion timing early warning method based on multimodal signals from goaf areas as described in claim 6, characterized in that, The coal spontaneous combustion temperature early warning model uses the Adam optimizer with a learning rate of 0.

001. After training, it can predict the temperature value of the goaf at the next moment based on the input temperature data and spatial feature parameters, and determine whether there is a risk of coal spontaneous combustion based on the predicted value.

8. The coal spontaneous combustion timing early warning method based on goaf multimodal signals as described in claim 7, wherein step 6 specifically involves: determining whether the predicted distributed optical fiber temperature at the next moment exceeds a threshold; if it does, an alarm is triggered; if it does not exceed the threshold, the goaf multimodal monitoring signal is read continuously.

Images