A coal mine goaf fire dynamic monitoring method and system

Abstract

The present application provides a kind of coal mine goaf fire dynamic monitoring method and system, belong to coal mine underground safety monitoring technical field, including obtaining the visible light image of monitoring goaf, real-time temperature, gas pressure, illumination intensity, infrared data and ultraviolet data;Based on multi-modal data, flame features are extracted respectively, and flame features specifically include optical features of flame, red and ultraviolet motion features, flame temperature features and flame edge pixel features;Flame features are input into a gated recurrent unit model, and the dynamic change law of key features over time is learned to obtain the classification result of whether there is a fire;If a fire occurs, calculate the dynamic risk index using real-time temperature and illumination intensity, and trigger the protection system based on the dynamic risk index.Through multi-modal perception fusion and dynamic early warning mechanism, improve the multi-dimensional data detection and dynamic early warning prevention and control of coal mine goaf fire.

Description

Technical Field

[0001] This invention belongs to the field of underground safety monitoring technology in coal mines, specifically relating to a method and system for dynamic monitoring of fires in coal mine goaf areas. Background Technology

[0002] Coal mine goafs are high-risk areas for underground fires, and their unique geological structure and environmental conditions place stringent demands on fire monitoring technology. Goafs are spaces left after coal seam mining, filled with broken coal, methane, and other flammable and explosive gases, and contain complex ventilation channels. Once a fire breaks out, the flames can spread rapidly with the help of methane and leaking air, releasing large amounts of toxic and harmful gases, easily triggering secondary disasters such as gas explosions and poisoning of personnel, with potentially disastrous consequences.

[0003] Current fire monitoring technologies applied to coal mine goaf areas have several drawbacks, such as limited data dimensions, low reliability of raw data, and significant interference from the coal mine environment. Traditional single-point sensors can only monitor temperature at local locations, failing to reflect the overall state of the goaf's three-dimensional space. The data is discrete and isolated, often issuing warnings only when the fire has already grown to a significant scale, missing the optimal response time. While gas sensors can detect fire-marking gases such as carbon monoxide, they are affected by turbulent gas flow in the goaf, and single gas indicators are susceptible to geological interference, leading to false alarms. For example, trace amounts of gas produced by the natural oxidation of residual coal in the goaf may be misinterpreted as fire signals. Furthermore, the data sources lack spatial breadth and modal diversity within the coal mine goaf, and variations in data acquisition accuracy and response speed can easily cause missed opportunities to generate critical data. This makes it difficult to meet the need for accurate early warnings, impacting the predictability of coal mine safety production. Summary of the Invention

[0004] To address the shortcomings of existing technologies in the dynamic monitoring and prevention of fires in goaf areas, this invention provides a method and system for dynamic monitoring of fires in coal mine goaf areas.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for dynamic monitoring of fires in coal mine goaf areas includes the following steps: Acquire multimodal data of the monitored goaf area, specifically including visible light images, real-time temperature, gas pressure, light intensity, infrared data, and ultraviolet data of the monitored area; Flame features are extracted based on the multimodal data. The flame features specifically include the optical features of the flame, the red and ultraviolet motion features, the flame temperature features, and the flame edge pixel features. The flame features are input into a gated recurrent unit model to capture the dynamic change pattern of the flame features over time. Based on the dynamic change pattern, a classification result indicating whether there is a fire is output. In the event of a fire, a dynamic risk index is calculated based on the real-time temperature and gas pressure, and the corresponding prevention and control system is triggered based on a preset dynamic risk index threshold.

[0006] Preferably, the gated recurrent unit model includes a hidden layer, a fully connected layer, and an output layer. The hidden layer has 16 hidden units; the fully connected layer has 64 neurons and uses ReLU as the activation function; and the output layer has 1 neuron and uses Sigmoid as the activation function.

[0007] Preferably, to acquire multimodal data of the goaf monitoring area, specifically, an infrared-visible dual-view gimbal, an ultraviolet sensor, and an IR-cut filter are deployed every 20m in the return airway of the goaf. The infrared-visible dual-view gimbal is used to acquire infrared data of the detection area in real time; the ultraviolet sensor is used to acquire ultraviolet data; the IR-cut is used in conjunction with a visible light lens, and the visible light lens filter is switched to acquire visible light images of the monitoring area in real time; gas pressure, temperature, and light intensity are acquired using a gas pressure sensor, a temperature sensor, and a light sensor, respectively.

[0008] Preferably, flame features are extracted based on the multimodal data. Specifically, the visible light image is converted to the HSV or YCbCr color space, a color threshold is set, and the optical properties of the flame in the image are extracted to obtain the flame optical feature vector. Based on continuous infrared and ultraviolet data, boundary tracking is performed to identify the motion boundary of the flame region. Then, the pixel motion direction and velocity are calculated using the optical flow method to obtain the red and ultraviolet motion features. Based on the ambient temperature and light intensity of the detection area, the radiation intensity at different wavelengths is analyzed using the dual-color thermometry method to reconstruct the temperature field distribution of the flame region and obtain the flame temperature features. Based on the visible light image of the monitoring area, pixel grayscale value analysis is performed to identify the flame edge pixels. Then, the Sobel operator is used to calculate the image gradient, extract the edge intensity and direction, and the Canny operator is used to extract fine edges to obtain the flame edge pixel features.

[0009] Preferably, the dynamic risk index threshold is as follows: when the dynamic risk index ≥ 0.7, it is considered high-risk, and high-pressure atomizing spraying and nitrogen injection are activated for coordinated prevention and control; when the dynamic risk index ≤ 0.4 and < 0.7, it is considered medium-risk, and nitrogen injection and leak sealing are triggered for prevention and control; when the dynamic risk index < 0.4, it is considered low-risk, and an audible and visual alarm is triggered.

[0010] Preferably, the dynamic risk index is calculated using the following formula: R = α(dT / dt) + β[CO] + γΔP; Where α is the temperature sensitivity coefficient, dT / dt is the rate of temperature change, β is the gas concentration weighting coefficient, γ is the pressure difference influence factor, ΔP is the gas pressure difference between roadways, and CO is the carbon monoxide concentration.

[0011] Preferably, the method further includes performing standardization, denoising, and time synchronization processing on the multimodal data sequentially to obtain preprocessed multimodal data; and performing feature extraction based on the preprocessed multimodal data.

[0012] This invention also provides a dynamic monitoring system for coal mine goaf fires, specifically comprising: The data acquisition module is used to acquire multimodal data of the monitored goaf area, specifically including visible light images, real-time temperature, gas pressure, light intensity, infrared data, and ultraviolet data of the monitored area.

[0013] The feature extraction module extracts flame features based on the multimodal data. The flame features specifically include the optical features of the flame, the movement and flickering frequency of red and ultraviolet light, and the pixel features of the flame edge. The flame features are input into a gated recurrent unit model to capture the dynamic change pattern of the flame features over time, and the classification result of whether there is a fire is output according to the dynamic change pattern.

[0014] The monitoring and prevention module is used to calculate the dynamic risk index based on the real-time temperature and gas pressure in the event of a fire, and to trigger the corresponding prevention system based on a preset dynamic risk index threshold.

[0015] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps described in the method for dynamic monitoring of coal mine goaf fires.

[0016] The present invention also provides a computer-readable storage medium storing a computer program, which, when loaded by a processor, is capable of executing the steps described in the method for dynamic monitoring of fires in coal mine goaf areas.

[0017] The method for dynamic monitoring of coal mine goaf fires provided by this invention has the following beneficial effects: This invention establishes a comprehensive, multi-dimensional data monitoring network within the three-dimensional space of the goaf by deploying various devices, ensuring the reliability of data sources in harsh mining environments. Based on the detected multi-dimensional fire scene data, feature capture is performed to obtain flame characteristics. A gated cyclic unit model is used to process these flame characteristics, capturing their dynamic changes over time to determine if a fire has occurred. A real-time dynamic risk index is calculated to construct a dynamic threshold. Based on the preset dynamic risk index threshold, the corresponding prevention and control system is triggered, adapting to the nonlinear heating process of coal and oxygen, avoiding premature or delayed alarms. This enhances the predictability of underground fires and reduces the risk of fire escalation. Attached Figure Description

[0018] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method for dynamic monitoring of fires in coal mine goaf areas according to an embodiment of the present invention.

[0020] Figure 2 This is a flowchart of a dynamic monitoring method for coal mine goaf fires according to the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0022] Example This invention provides a method for dynamic monitoring of fires in coal mine goaf areas, such as... Figure 1 As shown, the specific steps include: Step 1: Data Acquisition. Continuously collect multi-source data, covering information such as infrared, ultraviolet, and temperature.

[0023] Along the return airway, an infrared-visible dual-view gimbal, an ultraviolet sensor (GaN ultraviolet sensor, model: GT-ABC-L), and an IR-cut filter are deployed every 20m. The gimbal is used to capture infrared light with wavelengths of 3.8μm and 4.3μm in the detection area in real time, and can accurately sense the infrared radiation of specific wavelengths of the flame. The ultraviolet sensor is responsible for receiving ultraviolet radiation waves with wavelengths of 0.21μm to 0.37μm, effectively monitoring the ultraviolet characteristics of the flame. The IR-cut works with the visible light lens, and the visible light lens filter can be switched to acquire image information of the detection area in real time and obtain a comprehensive visual picture of the scene. The temperature sensor and the light sensor are used to collect the current ambient temperature and light intensity, respectively, to provide key environmental parameters for the system.

[0024] Step 2: Data Preprocessing. The data preprocessing module performs standardization, noise reduction, and time synchronization operations on the collected data in sequence.

[0025] Data acquired by the infrared-visible dual-view gimbal, ultraviolet sensor, and visible light lens were standardized, denoised, and time-synchronized. Standardization ensures a consistent data format across different sensors, facilitating subsequent analysis; denoising removes interference signals introduced during data acquisition, improving data quality; and time synchronization ensures consistency of data from all sensors over time, laying the foundation for accurate fusion analysis.

[0026] Step 3: Feature Extraction. Flame features are extracted using a specific algorithm, including: optical features of the flame, red and ultraviolet motion features, flame temperature features, and pixel features at the flame edge.

[0027] (1) The process of extracting the optical features of a flame includes: acquiring a visible light image of the monitoring area based on a visible light pan-tilt unit (analog signal converted to digital image via AD conversion). Converting the RGB image to HSV or YCbCr color space. Extracting the flame's unique optical properties such as color, brightness, and saturation by setting a color threshold or range. Obtaining the flame's optical features, which characterize the flame's visual color distribution.

[0028] (2) The process of extracting the motion features of red and ultraviolet light includes: collecting ultraviolet / infrared image sequences (analog to digital) from a red and ultraviolet light sensor; performing multi-region boundary tracking on consecutive frames of images to identify the motion boundaries of the flame region; applying optical flow methods (Lucas-Kanade or Farneback) to calculate the direction and speed of pixel motion; counting the number of grayscale value changes in a certain region in 20 consecutive frames of ultraviolet images to obtain the flicker frequency and motion trend of the flame, thus obtaining the motion features of red and ultraviolet light.

[0029] (3) The process of extracting the temperature characteristics of the flame includes: ambient temperature and light intensity collected by temperature and light sensors. A two-color thermometry method based on spectral analysis is used to calculate the temperature using the ratio of the radiation intensity of two different wavelengths. The temperature and light data are fused to reconstruct the temperature field distribution of the flame area and obtain the flame temperature characteristic map.

[0030] (4) The process of extracting the pixel features of the flame edge includes: Based on the visible light image of the monitoring area, analyzing the characteristics of pixel gray value changes: the gray value of stationary targets remains basically unchanged; the changes of moving car lights are large but short-lived and have a single trend; the gray value of flame pixels increases and decreases repeatedly. Setting a threshold to compare gray value changes and identifying flame edge pixels. Applying the Sobel operator to calculate the image gradient and detect edge intensity and direction; combining the Canny algorithm (Gaussian filtering, non-maximum suppression, and double threshold detection) to extract fine edge structures, obtaining the flame edge pixel feature map, and marking the flame contour and boundary information.

[0031] Specifically, the process involves operations such as "converting infrared / ultraviolet analog quantities into digital quantities, extracting flame optical features, and analyzing pixel grayscale value changes" to achieve preliminary fusion of visual (visible smoke texture), thermal (infrared hotspot), and gas data (indirectly associated through visual features), providing input features for the two-stream network. By constructing a YOLOv7+ResNet two-stream network, multiple collected features are input into the YOLOv7 network and the ResNet network respectively. The results of fusing the outputs of the two network models are analyzed to obtain infrared hotspot and visible smoke texture features. The extracted features are then corrected to obtain flame features.

[0032] Fiber optic acoustic wave-assisted positioning: Φ-OTDR acoustic wave sensing is embedded in distributed optical fibers for auxiliary positioning, and fiber optic acoustic wave data (detonation sound pattern) and temperature field data are integrated to achieve "acoustic wave-temperature" cross-modal fusion positioning.

[0033] Step 4: Data Processing. The extracted flame features are input into the GRU (Gated Cyclic Unit) classification model, which calculates and outputs a classification result indicating whether a fire exists. Using the extracted four types of temporal flame features as input, the GRU model learns the dynamic changes of these features over time (such as flame flicker frequency and temperature fluctuation trends), outputting a binary classification result of "fire present / no fire," achieving real-time detection of fires in goaf areas. The model is then quantized and deployed on an edge computing platform to achieve real-time fire detection. Specifically:

[0034] The input features are 4-dimensional temporal feature vectors, corresponding to the key flame features output by the feature extraction module. They are organized by time steps (each frame of data is one time step, and 20 consecutive frames constitute one temporal sample), including: X1(t): Flame optical features (digital values ​​converted from infrared to ultraviolet analog quantities, ranging from 0 to 255); X2(t): Ultraviolet motion frequency (number of grayscale value changes in a certain area in 20 consecutive frames, ranging from 0 to 50 times / frame); X3(t): Flame edge pixel ratio (number of flame edge pixels / total number of monitored pixels, ranging from 0 to 1); X4(t): Pixel grayscale fluctuation entropy (describing the degree of disorder in the "alternating increase and decrease" of flame pixel grayscale values, ranging from 0 to 2).

[0035] The GRU model consists of a hidden layer, a fully connected layer, and an output layer. The hidden layer has 16 hidden units (balancing accuracy and computational cost, adapted to the RV1126 edge chip); the fully connected layer has 64 neurons, using ReLU as the activation function; the output layer has 1 neuron (Sigmoid activation function, outputting a fire probability value of 0-1, ≥0.5 is considered "fire").

[0036] Core Gating Computation (Solving Long-Term Dependency and Gradient Problems in RNNs): Let the input at time step t be X(t) = [X1(t), X2(t), X3(t), X4(t)], and the hidden state be h(t). GRU dynamically controls the flow of information through a reset gate ((r(t))) and an update gate ((z(t))). The reset gate (determines the proportion of historical information retained): r(t) = σ (W r ·[h(t-1),X(t)]+b r ); Among them, W r To reset the gate weight matrix (dimension 16×(16+4)), b r The bias is 16×1, and σ is the sigmoid activation function (output 0-1, the closer the value is to 1, the more historical information is retained).

[0037] Update gate (determines the proportion of hidden states that are updated): z(t) = σ (W z ·[h(t-1),X(t)]+b z ); Among them, W z To update the gate weight matrix (dimension 16×(16+4)), b z As a bias (dimension 16×1), the closer (z(t)) is to 1, the more new information is updated to the hidden state.

[0038] Candidate hidden state (integrating current input with filtered historical information): ; Among them, W h The candidate state weight matrix (dimension 16×(16+4)), b h For bias (dimension 16×1). For element-wise product, (r(t)) h(t-1)) filters historical valid information.

[0039] Final hidden state (merging old and new information): .

[0040] The GRU model was trained using data containing 100,000+ samples (50,000 “fire” samples: experimental data of simulated fire in goaf areas; 50,000 “no fire” samples: environmental data of normal goaf areas).

[0041] Loss function: Binary cross-entropy loss Where y is the real label (1 = fire present, 0 = no fire), ( ( ) represents the output probability of the GRU.

[0042] Optimizer: Adam optimizer (learning rate 0.001, decay rate 0.9), iterates 100 times until the loss converges (validation set loss < 0.05).

[0043] By employing the reset and update gates of the GRU, the system focuses on learning key time points of flame characteristics (such as sudden flame flickering or temperature spikes) while ignoring irrelevant redundant information (such as noise from brief coal dust obstruction), thus addressing the problem of ambiguous long-term memory in traditional RNNs. It automatically learns key time-point information for fire prediction from images and the converted signal data. Its lightweight design and quantized deployment ensure a single-sample inference time of <20ms, effectively solving the long-term memory and backpropagation gradient problems of RNNs by ignoring irrelevant redundant information. By fusing four types of temporal features, the system achieves a fire identification accuracy of 97.6% on the test set (compared to 82.3% for traditional single-feature classification models), reducing false positives and false negatives.

[0044] Step 5: Dynamic Threshold Generation. The fire alarm threshold is updated based on real-time temperature and light intensity. Real-time temperature and light intensity are acquired through the data acquisition module. The ADC channel of the system's main control chip, RV1126, converts the output signals from the temperature and light sensors into voltage values ​​to calculate the current environmental parameters. Based on these two values ​​and a preset standard: under a preset temperature t0 and preset light intensity e0, if the output dynamic risk index R is higher than the preset fire alarm threshold, it indicates that a fire may be triggered, thus initiating a response mechanism.

[0045] Establish an entropy weight-TOPSIS risk assessment model and calculate the dynamic risk index R: R = α(dT / dt) + β[CO] + γΔP; Wherein, α is the temperature sensitivity coefficient, and α increases with high volatile coal. dT / dt is the rate of temperature change, reflecting the severity of the temperature rise; >1℃ / min indicates a risk of open flame. β is the gas concentration weighting coefficient, which is positively correlated with the gas content of the coal seam, and its range should be (0.1~0.3) based on experience. γ is the pressure difference influencing factor, which depends on the air leakage intensity in the goaf (γ tends to increase in fractured areas). ΔP is the gas pressure difference between roadways, a key parameter driving oxygen diffusion; when the pressure difference is >50Pa, the fire will spread more rapidly.

[0046] The three-level response thresholds are triggered based on the calculated dynamic risk index: R≥0.7, high-pressure atomizing spray is activated (high risk); 0.4≤R<0.7, nitrogen injection + leak sealing is triggered (medium risk); R<0.4, only audible and visual alarm is triggered (low risk).

[0047] Update the fire alarm thresholds by adjusting the specific thresholds of the three-level response thresholds according to the location and external environment. Alternatively, you can use an expert scoring method combined with AHP to calculate the weights of various environments, thereby ultimately calculating an accurate fire alarm threshold that conforms to the actual conditions and generating a dynamic fire alarm threshold that adapts to environmental changes.

[0048] Step Six: Response Mechanism, Fire Alarm. The fire probability value is compared with the updated threshold; an alarm is triggered if the threshold is exceeded. Upon alarm, the fire response process is immediately activated, transmitting fire information warnings to all downhole terminals. Simultaneously, the fire source location is precisely pinpointed, and nearby protective systems, such as nitrogen injection systems, leak-sealing devices, and sprinkler units, are activated to immediately extinguish the fire source. Coordinates are sent to personnel to confirm whether the fire has been extinguished. Specifically, after the fire alarm module determines the risk level, operations such as "activating the nitrogen injection system, leak-sealing device, and sprinkler unit" directly implement collaborative execution logic—for example, "prioritizing sealing of leaking airway (hydraulic sealing pressure ≥ 0.8 MPa), then injecting nitrogen (purity ≥ 99.2%)" corresponds to C-collaborative execution layer, I-adaptive nitrogen injection system, and J-pressure-driven leak-sealing device; "injecting foam inhibitor mixture" corresponds to K-foam-inhibitor combined spraying unit.

[0049] This invention also provides a dynamic monitoring system for coal mine goaf fires, comprising: The data acquisition module is used to acquire multimodal data of the monitored goaf area, specifically including visible light images, real-time temperature, gas pressure, light intensity, infrared data, and ultraviolet data of the monitored area.

[0050] The feature extraction module extracts flame features based on multimodal data. The flame features specifically include the optical features of the flame, the motion and flicker frequency of red and ultraviolet light, and the pixel features of the flame edge. The flame features are input into the gated recurrent unit model to capture the dynamic change pattern of the flame features over time, and output the classification result of whether there is a fire based on the dynamic change pattern.

[0051] The monitoring and prevention module is used to calculate the dynamic risk index based on real-time temperature and gas pressure in the event of a fire, and to trigger the corresponding prevention and control system based on the preset dynamic risk index threshold.

[0052] The modules in the aforementioned dynamic monitoring system for coal mine goaf fires can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0053] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in an embodiment of a method for dynamic monitoring of fires in coal mine goaf areas. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0054] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the aforementioned method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in an embodiment of a method for dynamic monitoring of fires in coal mine goaf areas. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0055] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0056] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0057] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0058] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for dynamic monitoring of fires in coal mine goaf areas, characterized in that, Includes the following steps: Acquire multimodal data of the monitored goaf area, specifically including visible light images, real-time temperature, gas pressure, light intensity, infrared data, and ultraviolet data of the monitored area; Flame features are extracted based on the multimodal data. The flame features specifically include the optical features of the flame, the red and ultraviolet motion features, the flame temperature features, and the flame edge pixel features. The flame features are input into a gated recurrent unit model to capture the dynamic change pattern of the flame features over time. Based on the dynamic change pattern, a classification result indicating whether there is a fire is output. In the event of a fire, a dynamic risk index is calculated based on the real-time temperature and gas pressure, and the corresponding prevention and control system is triggered based on a preset dynamic risk index threshold.

2. The method for dynamic monitoring of coal mine goaf fires according to claim 1, characterized in that, The gated recurrent unit model includes a hidden layer, a fully connected layer, and an output layer. The hidden layer has 16 hidden units; the fully connected layer has 64 neurons and uses ReLU as the activation function; the output layer has 1 neuron and uses Sigmoid as the activation function.

3. The method for dynamic monitoring of coal mine goaf fires according to claim 1, characterized in that, To acquire multimodal data for monitoring goaf areas, specifically, an infrared-visible dual-view gimbal, an ultraviolet sensor, and an IR-cut filter are deployed every 20m in the return airway of the goaf area. The infrared-visible dual-view gimbal is used to acquire infrared data of the detection area in real time; the ultraviolet sensor is used to acquire ultraviolet data; the IR-cut filter is used in conjunction with a visible light lens, and the visible light lens filter is switched to acquire visible light images of the monitoring area in real time; gas pressure, temperature, and light intensity are acquired using a gas pressure sensor, a temperature sensor, and a light sensor, respectively.

4. The method for dynamic monitoring of coal mine goaf fires according to claim 3, characterized in that, Flame features are extracted based on the multimodal data. Specifically, the visible light image is converted to the HSV or YCbCr color space, a color threshold is set, and the optical properties of the flame in the image are extracted to obtain the flame optical feature vector. Based on continuous infrared and ultraviolet data, boundary tracking is performed to identify the motion boundary of the flame area. Then, the optical flow method is used to calculate the pixel motion direction and speed to obtain the red and ultraviolet motion features. Based on the ambient temperature and light intensity of the detection area, the radiation intensity at different wavelengths is analyzed using the two-color thermometry method to reconstruct the temperature field distribution of the flame area and obtain the flame temperature characteristics. Pixel grayscale values ​​are analyzed based on the visible light image of the monitored area to identify the pixels at the edge of the flame. The Sobel operator is then used to calculate the image gradient, extract the edge intensity and direction, and the Canny operator is used to extract fine edges, thus obtaining the pixel features at the edge of the flame.

5. The method for dynamic monitoring of coal mine goaf fires according to claim 1, characterized in that, The specific dynamic risk index thresholds are as follows: when the dynamic risk index ≥ 0.7, it is considered a high-risk situation, and high-pressure atomized spraying and nitrogen injection are activated for coordinated prevention and control; when the dynamic risk index ≤ 0.4 and < 0.7, it is considered a medium-risk situation, and nitrogen injection and leak sealing are triggered for prevention and control; when the dynamic risk index < 0.4, it is considered a low-risk situation, and audible and visual alarms are activated.

6. The method for dynamic monitoring of coal mine goaf fires according to claim 5, characterized in that, The dynamic risk index is calculated using the following formula: R = α(dT / dt) + β[CO] + γΔP; Where α is the temperature sensitivity coefficient, dT / dt is the rate of temperature change, β is the gas concentration weighting coefficient, γ is the pressure difference influence factor, ΔP is the gas pressure difference between roadways, and CO is the carbon monoxide concentration.

7. The method for dynamic monitoring of coal mine goaf fires according to claim 3, characterized in that, It also includes performing standardization, denoising, and time synchronization processing on the multimodal data in sequence to obtain preprocessed multimodal data; and performing feature extraction based on the preprocessed multimodal data.

8. A dynamic monitoring system for fires in coal mine goaf areas, characterized in that, include: The data acquisition module is used to acquire multimodal data of the monitored goaf area, specifically including visible light images, real-time temperature, gas pressure, light intensity, infrared data, and ultraviolet data of the monitored area; The feature extraction module extracts flame features based on the multimodal data. The flame features specifically include the optical features of the flame, the motion and flicker frequency of red and ultraviolet light, and the pixel features of the flame edge. The flame features are input into the gated recurrent unit model to capture the dynamic change pattern of the flame features over time, and the classification result of whether there is a fire is output according to the dynamic change pattern. The monitoring and prevention module is used to calculate the dynamic risk index based on the real-time temperature and gas pressure in the event of a fire, and to trigger the corresponding prevention system based on a preset dynamic risk index threshold.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 7.

Images