Mine hidden fire source safety monitoring method based on multi-dimensional feature space calibration

Abstract

This application discloses a method for monitoring the safety of hidden fire sources in mines based on multi-dimensional feature space calibration. This method integrates surface thermal radiation, surrounding rock deformation rate, and ecological load parameters to reconstruct a multi-dimensional disaster-causing characteristic field. Through a central benchmark mapping mechanism, the core of the disaster is calibrated to a unified physical benchmark unit, thereby eliminating the monitoring scale drift problem caused by seasonal environmental fluctuations and background differences. The method in this application combines an energy spatial diffusion physical constraint model to analyze the radial evolution trajectory of disaster intensity, elevating calibration from numerical calculation to a physical representation of the deep fire source conduction mechanism. Simultaneously, it employs an adaptive threshold strategy based on residual statistics, using the convergence of the physical evolution trajectory as a criterion. This allows for accurate identification and removal of false anomaly signals from complex mining backgrounds caused by production activities such as surface reconstruction, compaction, and drilling operations, effectively addressing the industry pain points of inaccurate location of hidden disaster-causing factors and the susceptibility to false alarms.

Description

Technical Field

[0001] This application relates to the field of coal mine disaster prevention and control technology, specifically to a method for monitoring the safety of hidden fire sources in mines based on multi-dimensional feature space calibration. Background Technology

[0002] Underground coal fires pose a significant and hidden threat to the inherent safety of mining areas. They not only cause substantial losses of coal resources but also trigger thermal stress fractures in overlying strata, surface subsidence, and severe damage to the ecological environment. Because coal fires often occur in underground goafs and blind tunnels, they are highly concealed and have long evolution cycles. Therefore, employing multi-source detection technology to achieve dynamic identification of fire sources and calibration of fire intensity is a crucial prerequisite for the implementation of precise grouting, cooling, and sealing fire prevention and extinguishing projects. Current industry monitoring has shifted from single-indicator to multi-dimensional integration; by coupling surface thermal radiation with surrounding rock deformation fields, macroscopic identification of large-scale fire zones can be initially achieved.

[0003] However, existing technologies still face significant bottlenecks in refined calibration and complex mining environments. One challenge is cross-physical field coupling calibration. Surface temperature (LST) and surface deformation rate (InSAR) belong to thermodynamics and solid mechanics, respectively, and their dimensions are not uniform. Furthermore, surface thermal signals are significantly affected by diurnal and seasonal climate fluctuations, while deformation is controlled by geological conditions and burial depth. This makes it impossible to compare the disaster intensity of different time periods and regions under a unified standard, which seriously affects the dynamic optimization of fire extinguishing plans.

[0004] Second, there is a lack of physical mechanism constraints. The heat conduction and stress diffusion of coal fires exhibit a point-source radial continuous attenuation characteristic. However, existing models are mostly based on isolated pixel calculations and lack an overall description of the spatial evolution trend of the damage field. The results are fragmented and discontinuous, making it difficult to accurately invert the fire source center, influence radius and diffusion gradient, and the accuracy of fire zone boundary delineation is insufficient.

[0005] Third, the ability to resist interference is weak. Human disturbances such as soil covering and stockpiling in the mining area, mechanical operations, and surface remediation can easily create "false thermal anomalies". Without physical convergence constraints, these anomalies can be easily misjudged as coal fire signals, causing grouting boreholes to deviate from the real fire source, and even delaying treatment and causing secondary disasters such as gas explosions and surface collapses. Summary of the Invention

[0006] The purpose of this application is to provide a method for safety monitoring of hidden fire sources in mines based on multidimensional feature space calibration, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this application provides a method for safety monitoring of concealed fire sources in mines based on multidimensional feature space calibration, comprising the following steps: S1. Obtain the surface thermal radiation intensity, surface deformation rate and ecological load parameters of the coalfield fire area, and construct a multi-dimensional damage characteristic field of thermo-mechanical-biological based on spatial location alignment to characterize the evolution intensity of hidden fire sources. S2. Based on the multidimensional damage feature field, locate the surface thermal anomaly mapping center, and use the benchmark mapping mechanism to confirm the disaster intensity at the center location as the physical benchmark unit, and generate a damage intensity distribution field to eliminate the dimensional differences between different fire zones. S3. Taking the confirmed thermal anomaly center as the origin, construct a radial spatial physical constraint model based on the physical law of energy spatial diffusion, and quantitatively analyze the natural evolution trajectory of disaster-causing damage intensity as spatial distance increases. S4. Using the convergence characteristics of the physical constraint model, identify and eliminate false abnormal signal interference generated by non-coal fire sources, and identify the influence domain and damage intensity of hidden fire sources in the mine.

[0008] Preferably, in step S1, the process of constructing the multidimensional damage feature field includes: S11. Employ resampling technology to uniformly align data grids with different physical resolutions; S12, Obtained by introducing a geographic detector q The statistical values, as physical contribution coefficients, are used to construct a feature field reflecting the destructive capacity of the fire zone through a linear weighted fusion method.

[0009] Preferred, q The formula for calculating the statistical value is: ; in, k =1,2,3…, For the discretization levels of each monitoring indicator, and Layers Compared with the total number of pixel samples in the entire region, and Layers The variance of the environmental damage value of the entire region q The higher the statistical value, the stronger the indicative power of that physical dimension for the evolution of concealed fire sources, and according to... q Statistical values ​​are used to construct a weighted coupled feature field. : ; in, For coordinates The weighted coupled feature field values ​​at the location are used to locate potential hazards. This represents the selected set of monitoring indicators. This represents the corresponding set of regions or levels. The representative weighting coefficient is calculated using the above. qThe values ​​are normalized; the better the performance, the better the metric. q The larger, It is the standardized one-dimensional physical distribution function after Gaussian smoothing.

[0010] Preferably, in step S2, the processing logic for the reference mapping is as follows: ; in, d The Euclidean distance from the pixel to the center of the thermal anomaly. k The physical attenuation control coefficient reflects the hindering effect of coal mine geological conditions on heat conduction. At that time, the disaster intensity was forcibly mapped to the benchmark unit to establish the physical starting point of the entire field evolution; The baseline mapping mechanism then performs a scale transformation on the original intensity using a quadratic mapping operator to calculate the catastrophic calibration index for each pixel: ; in, λ is the Euclidean distance from the pixel to the center of the thermal anomaly, and λ is the preset spatial attenuation adjustment coefficient. Through the quadratic mapping operator, the disaster intensity at the mapping core location is forcibly confirmed as the baseline unit 1, and the remaining pixels are mapped to a standardized intensity value between 0 and 1 according to their physical distance from the core and energy evolution logic.

[0011] Preferably, in step S3, the process of constructing the radial space physical constraint model includes: S31. Establish a radial buffer zone that reflects the thermal influence domain of the fire source, with the center of the thermal anomaly as the origin of the coordinate system. S32. Within the step interval, through pixel masking, calculate the physical average value of the damage intensity in the spatial radial direction; S33. The spatial distance-intensity sequence is physically modeled and fitted using an exponential decay trajectory model, expressed as follows: ； Indicates the corresponding radial distance Theoretical predicted value of damage intensity at the location, The intensity intercept represents the potential extreme value of destruction at the core location. The spatial attenuation coefficient characterizes the rate attenuation of damage effects as energy diffuses. S34. Evaluate the fitting accuracy, used to quantitatively verify the physical control law of thermal radiation and surrounding rock stress field diffusion on the intensity of environmental damage.

[0012] Preferably, in step S4, the process of eliminating false abnormal signals from non-coal fire sources includes: S41. Extract the physical deviation between the actual intensity values ​​of each point in the radial space and the predicted trajectory of the physical evolution model, and generate a residual sequence. S42. An adaptive statistical discrimination mechanism is introduced to analyze the distribution characteristics of the residual response sequence. Combined with the statistical characteristics of background noise in the mining area, a multi-criteria noise identification threshold for identifying human interference is automatically constructed. S43. Identify detection points whose residual response values ​​exceed the threshold as outlier interference points, where the interference points correspond to non-target source impulse noise or pseudo-anomaly signals generated by mining production activities. S44. The high-precision disaster intensity distribution field after physical constraint purification serves as the positioning basis for mine hidden danger management projects and is used for the calibration of fire extinguishing grouting areas.

[0013] Preferably, the parameters of the above exponential decay trajectory model are optimized by minimizing the objective loss function: ; in, Let be the target loss function, and let be the target loss function, which aims to improve the loss by adjusting the parameters. a and b To minimize the function value, For the model parameters that need to be optimized, For the first The average of the calibration index observations within each radial annulus. The total number of radial rings This is a theoretical prediction model that describes the process of energy or index decaying exponentially with increasing distance, where... Represents the natural constant. These are the distance variables and constant terms in the model.

[0014] Preferably, the method for constructing the above residual sequence is as follows: ; in, For the actual calibrated intensity of the pixel, The standard deviation is calculated based on the statistical distribution characteristics of the full-field residuals by introducing an adaptive statistical discrimination mechanism to represent the theoretical predictions of the physical evolution model at the corresponding distances. And construct a dynamic noise immunity threshold. : ; in, The mean of the residuals for the entire game. This is the confidence level adjustment coefficient. The total number of pixel samples in the detection area. Representing the The residual value of each pixel The threshold can be adaptively adjusted according to the signal-to-noise ratio level of a specific image.

[0015] Preferably, the above-mentioned physical constraint purification is achieved by utilizing physical constraint functions. Identify all pixels in the field: ; If pixel residual Exceeding the threshold If the location does not follow the spatial continuity and exponential decay law of energy release from a concealed fire source, then it is determined that the location does not follow the spatial continuity and exponential decay law of energy release from a concealed fire source.

[0016] According to the technical solution of this application, a readable storage medium is also provided, on which a computer program is stored. When the program is executed by a processor, it implements the steps of the mine concealed fire source safety monitoring method based on multi-dimensional feature space calibration as described above.

[0017] Compared with the prior art, the beneficial effects of this application are: This application integrates surface thermal radiation, surrounding rock deformation rate and ecological load parameters to reconstruct a multidimensional disaster-causing characteristic field. Through a central benchmark mapping mechanism, the core of the disaster is calibrated to a unified physical benchmark unit, thereby eliminating the monitoring scale drift problem caused by seasonal environmental fluctuations and background differences, and ensuring the horizontal comparability of long-term monitoring data.

[0018] This application combines an energy spatial diffusion physical constraint model to analyze the radial evolution trajectory of disaster intensity, elevating calibration from numerical calculation to the physical characterization of deep fire source conduction mechanisms. This enhances the physical robustness and scientific explanatory power of the detection results. Simultaneously, it employs an adaptive threshold strategy based on residual statistics, using the convergence of the physical evolution trajectory as a criterion constraint. This strategy can accurately identify and remove false anomaly signals from complex mining backgrounds caused by production activities such as surface reconstruction, compaction, and drilling. It effectively solves the problems of inaccurate location of hidden disaster-causing factors and the high likelihood of false alarms. This provides a high-precision quantitative calibration basis for fire prevention and extinguishing projects such as mine grouting cutoff, and has significant engineering application value while ensuring the inherent safety of mining areas. Attached Figure Description

[0019] Figure 1 This is a flowchart of a mine concealed fire source safety monitoring method based on multidimensional feature space calibration according to an embodiment of this application; Figure 2 This is a comparison diagram of the results before and after coal fire damage calibration and purification based on simulated coal fire data according to an embodiment of this application; Figure 3 This is a diagram showing the adaptive spatial threshold distribution field results according to an embodiment of this application; Figure 4 This is a graph showing the fitting curve results of the radial evolution law according to an embodiment of this application; Figure 5 This is a diagram showing the final damage level calibration result according to an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] Please see Figure 1 According to an embodiment of this application, a method for safety monitoring of concealed fire sources in mines based on multidimensional feature space calibration is provided, including the following steps S1-S4.

[0022] In step S1, heterogeneous data reflecting different physical characteristics of the disaster source within the coalfield fire area are acquired through a multi-dimensional detection platform. These include surface thermal radiation intensity (LST) characterizing energy release intensity, surface deformation rate characterizing the mechanical evolution of overlying strata, and ecological load parameters (RSEI) characterizing surface environmental load parameters. Resampling technology is used to align data grids with different physical resolutions. Spatial interpolation is then used to resample deformation data and ecological load data into a geographic grid consistent with the thermal radiation image, ensuring that all feature fields are fully spatiotemporally aligned at the pixel scale.

[0023] To quantitatively analyze the synergistic contribution of multidimensional "thermal-mechanical-biological" indicators to environmental damage, data obtained from a geographic detector were introduced. q The statistical values, used as physical contribution coefficients, are used to construct a feature field reflecting the fire zone's destructive capacity through a linear weighted fusion method. q The formula for calculating the statistical value is: (Formula 1); in, k =1, 2, 3…, For the discretization levels of each monitoring indicator, and Layers Compared with the total number of pixel samples in the entire region, and Layers The variance of the environmental damage value of the entire region q The higher the statistical value, the stronger the indicative power of that physical dimension for the evolution of concealed fire sources, and according to... q Statistical values ​​are used to construct a weighted coupled feature field. : (Formula 2); in, For coordinates The weighted coupled feature field values ​​at the location are used to locate potential hazards. This represents the selected set of monitoring indicators. This represents the corresponding set of regions or levels. The representative weighting coefficient is calculated using the above. q The values ​​are normalized; the better the performance, the better the metric. q The larger, It is the standardized one-dimensional physical distribution function after Gaussian smoothing.

[0024] By using this explanatory power index as a weighting coefficient, multidimensional weighted feature coupling is performed, thereby transforming discrete physical quantities into a feature distribution field that characterizes the overall degree of damage. At the same time, a spatially continuous Gaussian smoothing operator is introduced, and a specific spatial convolution kernel radius and standard deviation are set to perform low-pass filtering on the coupled feature field. Its physical significance lies in stripping away local non-stationary noise through spatial weighted averaging and suppressing instantaneous background disturbances, thereby smoothing the distribution characteristics of feature gradients and providing a smooth and continuous energy distribution field foundation for subsequent high-precision core driving source locking and feature mapping.

[0025] In step S2, the center of the surface thermal anomaly is located based on the multidimensional damage feature field. The disaster intensity at the center location is confirmed as the physical reference unit using the reference mapping mechanism, and a damage intensity distribution field is generated to eliminate the dimensional differences between different fire zones.

[0026] By using a mechanism indexing algorithm to retrieve feature intensity peak pixels, these are defined as the thermal mapping core coordinates of a concealed fire source. Using these coordinates as the physical origin, a radial distance matrix covering the entire domain is constructed using a spatial distance transformation operator, transforming the chaotic pixel distribution into a geometric distance field centered on the causative core. d The processing logic for the reference mapping is as follows: (Formula 3); in, The Euclidean distance from the pixel to the center of the thermal anomaly. The physical attenuation control coefficient reflects the hindering effect of coal mine geological conditions on heat conduction. At that time, the disaster intensity was forcibly mapped to the benchmark unit to establish the physical starting point of the entire field evolution; The baseline mapping mechanism then performs a scale transformation on the original intensity using a quadratic mapping operator to calculate the catastrophic calibration index for each pixel: (Formula 4); in, is the Euclidean distance from the pixel to the center of the thermal anomaly, and is the preset spatial attenuation adjustment coefficient, the value of which is adaptively adjusted according to the geological conditions of the coal mine. For a pixel ( x,y The original intensity value at the location (such as thermal infrared radiation intensity, brightness temperature, etc.). thermal anomaly center The original intensity value at the location. Through a secondary mapping operator, the disaster intensity at the mapping core location is forcibly confirmed as the reference unit 1, and the remaining pixels are mapped to a standardized intensity value between 0 and 1 according to their physical distance from the core and the energy evolution logic. This mapping method physically constructs a phase reference system without background, so that no matter how the external environmental parameters (such as seasonal changes in temperature) shift, the identification benchmark is always anchored to the core area that is most severely affected by the disaster.

[0027] By employing a benchmark mapping mechanism strategy, the problem of inconsistent calibration benchmarks caused by seasonal background fluctuations in safety monitoring is addressed. The secondary mapping operator can perform scale transformation on the original intensity and calculate the disaster-causing calibration index for each pixel.

[0028] In step S3, a radial spatial physical constraint model is constructed based on the physical law of energy spatial diffusion, with the thermal anomaly center as the origin, to quantitatively analyze the natural evolution trajectory of disaster-causing damage intensity as spatial distance increases. Step S3 further includes steps S31-S34: S31. Establish a radial buffer zone that reflects the thermal influence domain of the fire source, with the center of the thermal anomaly as the origin of the coordinate system. S32. Within the step interval, through pixel masking, calculate the physical average value of the damage intensity in the spatial radial direction; S33. The spatial distance-intensity sequence is physically modeled and fitted using an exponential decay trajectory model, expressed as follows: (Formula 5); Indicates the corresponding radial distance Theoretical predicted value of damage intensity at the location, The intensity intercept represents the potential extreme value of destruction at the core location. The spatial attenuation coefficient characterizes the rate attenuation of damage effects as energy diffuses. S34. Evaluate the fitting accuracy, used to quantitatively verify the physical control law of thermal radiation and surrounding rock stress field diffusion on the intensity of environmental damage.

[0029] The parameters of the exponentially decaying trajectory model are optimized by minimizing the objective loss function: (Formula 6); in, Let be the objective loss function, and let be the objective to minimize the function value by adjusting parameters a and b. For the model parameters that need to be optimized, For the first The average of the calibration index observations within each radial annulus. The total number of radial rings This is a theoretical prediction model that describes the process of energy or index decaying exponentially with increasing distance, where... Represents the natural constant. These are the distance variables and constant terms in the model.

[0030] In the optimization process, a goodness-of-fit threshold was set for robustness experiments to ensure that the evolutionary model could accurately map the real thermodynamic diffusion logic of coal fires. This modeling process not only quantitatively characterizes the physical trajectory under the interaction of coal fire thermal radiation and surface stress field, but more importantly, the fitted evolution curve represents the "theoretical damage benchmark" under ideal driving conditions, such as... Figure 4 As shown, this provides a consistent comparison criterion with physical logic for subsequent identification of abnormal man-made engineering signals (such as backfilling and construction interference).

[0031] In step S4, the physical evolution trajectory constructed in step S3 is used as a convergence constraint criterion to identify and eliminate false abnormal signal interference generated by non-coal fire sources, and to identify the influence domain and damage intensity of hidden fire sources in the mine.

[0032] First, the physical deviation between the actual intensity values ​​at each point in the radial space and the predicted trajectory of the physical evolution model is extracted to generate a residual sequence.

[0033] The method for constructing the residual sequence is as follows: (Formula 7); in, For the actual calibrated intensity of the pixel, The standard deviation is calculated based on the statistical distribution characteristics of the full-field residuals by introducing an adaptive statistical discrimination mechanism to represent the theoretical predictions of the physical evolution model at the corresponding distances. And construct a dynamic noise immunity threshold. : (Formula 8); The mean of the residuals for the entire game. This is the confidence level adjustment coefficient. The total number of pixel samples in the detection area. Representing the The residual value of each pixel The threshold can be adaptively adjusted according to the signal-to-noise ratio level of a specific image, without the need for manual subjective preset.

[0034] Secondly, detection points whose residual response values ​​exceed the threshold are identified as outlier interference points. These interference points correspond to non-target source pulse noise or pseudo-abnormal signals generated by mining production activities. The high-precision disaster intensity distribution field, after being purified by physical constraints, serves as the positioning basis for mine hazard management projects and is used for calibration of the fire extinguishing grouting area.

[0035] Physical constraint purification is achieved by utilizing physical constraint functions. Identify all pixels in the field: (Formula 9); If pixel residual Exceeding the threshold If the location does not follow the spatial continuity and exponential decay law of energy release from a concealed fire source, and deviates significantly from the physical evolution trajectory, it is usually precisely corresponding to artificial pseudo-anomaly signals generated by surface reconstruction zone, earthwork backfilling and compaction zone, temporary operation road, or fire prevention and extinguishing drilling construction in actual mining engineering. Although these interference sources are similar to the real disaster-causing signals in terms of thermal radiation intensity or terrain features, their spatial distribution does not have physical convergence characteristics.

[0036] To quantitatively verify the engineering effectiveness of the method described in this application, this embodiment conducted a comparative experiment by simulating typical human interference scenarios in mining areas, and compared the identification accuracy of traditional calibration methods with the optimized method of this application.

[0037] like Figure 2 As shown in the left figure, a production activity noise with high-intensity thermo-mechanical characteristics (such as simulating large-scale backfilling or construction heat sources) was simulated within the detection area through physical injection. When processed using traditional multi-dimensional feature fusion calibration methods, such as... Figure 2 As shown in the left figure, the results show that traditional algorithms, in the absence of spatial physical and logical constraints, cannot distinguish between man-made interference and real disaster-causing signals, resulting in a significant "pseudo-damaged" closed area in the evaluation field, which seriously misleads the qualitative and location of the disaster.

[0038] like Figure 2 As shown in the right figure, based on the results of this embodiment, noise caused by human activities can be extracted more easily, and the RMSE is significantly reduced from the original 0.117 to 0.065, demonstrating the superior performance of this algorithm. In comparison, based on the physical convergence constraint method described in this application, the system can accurately capture the residual characteristics between the human-caused noise signal and the natural diffusion law of concealed fire sources, such as... Figures 3-4 As shown, the adaptive threshold mechanism achieves complete removal of artificial interference regions. Comparative experimental data show that the root mean square error (RMSE) of the proposed method is significantly reduced from 0.117 in the traditional method to 0.065, and the model accuracy is improved by 44.4%.

[0039] like Figure 5 As shown, by accurately removing the outlier noise through this physical constraint mechanism, the essential purification of the disaster-causing characteristic field is achieved, and the final output is a high-precision monitoring calibration and classification that conforms to the logic of deep energy evolution. It can classify the damage level into undamaged, mild, moderate and moderate, providing a physically substantial decision-making basis for mine hazard management projects.

[0040] In summary, the safety monitoring method according to the embodiments of this application reconstructs a multi-dimensional disaster-causing characteristic field by integrating surface thermal radiation, surrounding rock deformation rate and ecological load parameters, and uses a central benchmark mapping mechanism to calibrate the disaster core as a physical benchmark unit, fundamentally eliminating the problem of monitoring scale drift caused by seasonal environmental fluctuations and background differences, and ensuring the horizontal comparability of calibration dimensions during long-term governance.

[0041] By combining the physical constraint model of energy space diffusion to analyze the radial evolution trajectory of disaster intensity, the calibration process is elevated from simple numerical calculation to a physical characterization of the deep fire source conduction mechanism, which significantly enhances the physical robustness and scientific explanatory power of the detection results.

[0042] Meanwhile, the adaptive threshold strategy based on residual statistics, using the convergence of physical evolution trajectories as a judgment constraint, can accurately identify and remove false anomaly signals generated by production activities such as surface reconstruction, overburden compaction, and drilling construction from complex mining backgrounds. This effectively solves the industry pain point of inaccurate positioning of hidden disaster-causing factors and the high risk of false alarms. It provides a high-precision quantitative calibration basis for fire prevention and extinguishing projects such as mine grouting cut-off, and has significant engineering application value while ensuring the inherent safety of mining areas.

[0043] All parts not covered in this application are the same as or can be implemented using existing technology. Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for safety monitoring of concealed ignition sources in mines based on multidimensional feature space calibration, characterized in that, Includes the following steps: S1. Obtain surface thermal radiation intensity, surface deformation rate, and ecological load parameters of the coalfield fire area. Construct a multi-dimensional thermal-mechanical-biological damage characteristic field characterizing the evolution intensity of concealed fire sources based on spatial location alignment. This includes incorporating data obtained from geographic detectors. q The statistical values, used as physical contribution coefficients, are used to construct a feature field reflecting the fire zone's destructive capacity through a linear weighted fusion method. ; S2. Based on the multidimensional damage feature field, locate the surface thermal anomaly mapping center, and use the benchmark mapping mechanism to confirm the disaster intensity at the center location as the physical benchmark unit, generating a damage intensity distribution field to eliminate dimensional differences between different fire zones. The benchmark mapping processing logic is as follows: ； in, d The Euclidean distance from the pixel to the center of the thermal anomaly. k The physical attenuation control coefficient reflects the hindering effect of coal mine geological conditions on heat conduction. At that time, the disaster intensity was forcibly mapped to the benchmark unit to establish the physical starting point of the entire field evolution; The baseline mapping mechanism then performs a scale transformation on the original intensity using a quadratic mapping operator to calculate the catastrophic calibration index for each pixel: ； in, The Euclidean distance from the pixel to the center of the thermal anomaly. As a preset spatial attenuation adjustment coefficient, the disaster intensity at the core location is forcibly confirmed as the baseline unit 1 through a secondary mapping operator, and the remaining pixels are mapped to a standardized intensity value between 0 and 1 according to their physical distance from the core and energy evolution logic. S3. Using the confirmed thermal anomaly center as the origin, construct a radial spatial physical constraint model based on the physical law of energy spatial diffusion, and quantitatively analyze the natural evolution trajectory of disaster-causing damage intensity with increasing spatial distance. The process of constructing the radial spatial physical constraint model includes: S31. Establish a radial buffer zone that reflects the thermal influence domain of the fire source, with the center of the thermal anomaly as the origin of the coordinate system. S32. Within the step interval, through pixel masking, calculate the physical average value of the damage intensity in the spatial radial direction; S33. The spatial distance-intensity sequence is physically modeled and fitted using an exponential decay trajectory model, expressed as follows: ； Indicates the corresponding radial distance Theoretical predicted value of damage intensity at the location, The intensity intercept represents the potential extreme value of destruction at the core location. The spatial attenuation coefficient characterizes the rate attenuation of damage effects as energy diffuses. S34. Evaluate the fitting accuracy, used to quantitatively verify the physical control law of thermal radiation and surrounding rock stress field diffusion on the intensity of environmental damage; S4. Utilizing the convergence characteristics of the physical constraint model, identify and eliminate false anomaly signals from non-coal fire sources, and identify the influence domain and damage intensity of hidden fire sources in the mine. The process of eliminating false anomaly signals from non-coal fire sources includes: S41. Extract the physical deviation between the actual intensity values ​​at each point in the radial space and the predicted trajectory of the physical evolution model, and generate a residual sequence. ; S42. An adaptive statistical discrimination mechanism is introduced to analyze the distribution characteristics of the residual response sequence. Combined with the statistical characteristics of background noise in the mining area, a multi-criteria noise identification threshold for identifying human interference is automatically constructed. S43. Identify detection points whose residual response values ​​exceed the threshold as outlier interference points, where the interference points correspond to non-target source impulse noise or pseudo-anomaly signals generated by mining production activities. S44. The high-precision disaster intensity distribution field after physical constraint purification serves as the positioning basis for mine hidden danger management projects and is used for the calibration of fire extinguishing grouting areas.

2. The method for monitoring hidden fire sources in mines based on multi-dimensional feature space calibration according to claim 1, characterized in that, The q The formula for calculating the statistical value is: ； in, k =1,2,3…, For the discretization levels of each monitoring indicator, and Layers Compared with the total number of pixel samples in the entire region, and Layers The variance of the environmental damage value of the entire region q The higher the statistical value, the stronger the indicative power of that physical dimension for the evolution of concealed fire sources, and according to... q Statistical values ​​are used to construct a weighted coupled feature field. : ； in, For coordinates The weighted coupled feature field values ​​at the location are used to locate potential hazards. This represents the selected set of monitoring indicators. This represents the corresponding set of regions or levels. The representative weighting coefficient is calculated using the above. q The values ​​are normalized; the better the performance, the better the metric. q The larger, It is the standardized one-dimensional physical distribution function after Gaussian smoothing.

3. The method for monitoring hidden fire sources in mines based on multi-dimensional feature space calibration according to claim 1, characterized in that, The parameters of the exponentially decaying trajectory model are optimized by minimizing the objective loss function: ； in, Let be the target loss function, and let be the target loss function, which aims to improve the loss by adjusting the parameters. a and b To minimize the function value, For the model parameters that need to be optimized, For the first The average of the calibration index observations within each radial annulus. This represents the total number of radial annular bands. This is a theoretical prediction model that describes the process of energy or index decaying exponentially with increasing distance, where... Represents the natural constant. These are the distance variables and constant terms in the model.

4. The method for safety monitoring of concealed fire sources in mines based on multi-dimensional feature space calibration according to claim 1, characterized in that, The method for constructing the residual sequence is as follows: ； in, For the actual calibrated intensity of the pixel, The standard deviation is calculated based on the statistical distribution characteristics of the full-field residuals by introducing an adaptive statistical discrimination mechanism to represent the theoretical predictions of the physical evolution model at the corresponding distances. And construct a dynamic noise immunity threshold. : ； in, The mean of the residuals for the entire game. This is the confidence level adjustment coefficient. The total number of pixel samples in the detection area. Representing the The residual value of each pixel The threshold can be adaptively adjusted according to the signal-to-noise ratio level of a specific image.

5. The method for monitoring hidden fire sources in mines based on multi-dimensional feature space calibration according to claim 1, characterized in that, The physical constraint purification is achieved by utilizing physical constraint functions. Identify all pixels in the field: ； If pixel residual Exceeding the threshold If the location does not follow the spatial continuity and exponential decay law of energy release from a concealed fire source, then it is determined that the location does not follow the spatial continuity and exponential decay law of energy release from a concealed fire source.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the steps of the mine concealed fire source safety monitoring method based on multidimensional feature space calibration as described in any one of claims 1 to 5.

Images