A coalfield fire area multi-source information singularity identification method

Abstract

This invention discloses a method for identifying singularities of multi-source information in coalfield fire zones, comprising the following steps: Step 1: Simultaneously collect four-dimensional homogeneous heterogeneous data across the entire region, along with supporting auxiliary data; process the data to obtain a standardized fusion base dataset; Step 2: Based on the standardized fusion base dataset, divide the entire region into real-world environmental clutter interference intervals; Step 3: Model a three-dimensional layered grid according to shallow exposed coal seams, shallow to medium-depth concealed coal seams, and deep pressure-bearing coal seams, and output graded fire zone singularity identification results; Step 4: For the graded fire zone singularity identification results output by the model, retrieve the original multi-source monitoring data for cross-comparison and source tracing verification. This invention achieves integrated identification of multi-source data in a unified spatiotemporal manner, dynamic background anti-interference, layered singularity quantification, and multi-source closed-loop verification, featuring high accuracy, strong anti-interference, reliable deep identification, and strong engineering applicability.

Description

Technical Field

[0001] This invention belongs to the field of coalfield fire zone assessment and identification technology, and specifically relates to a method for identifying the singularity of multi-source information in coalfield fire zones. Background Technology

[0002] The current mainstream technologies for coalfield fire zone assessment and identification in the industry mainly adopt a coupled assessment mode of single thermal infrared remote sensing inversion temperature measurement + conventional ground geophysical exploration single-point verification, and are equipped with basic threshold zoning comparison algorithms to carry out fire zone anomaly assessment. In routine operations, only single-dimensional surface thermal radiation time-series data and local shallow geothermal monitoring data are collected simultaneously. In some scenarios, simplified geomagnetic profile inspection data are overlaid to complete the basic monitoring network. Subsequently, relying on fixed global temperature thresholds and conventional amplitude thresholds, the monitoring data of the entire region is directly screened in batches to initially delineate abnormal high-temperature patches and areas of minor geomagnetic disturbances on the surface, which are directly identified as suspected coalfield fire anomalous points and contiguous fire areas. At the same time, the traditional static data splicing and adaptation method is used to complete the simple overlay and summary of multiple types of scattered monitoring data. Only local points are manually reviewed on-site to verify the authenticity of anomalous areas. Finally, the coalfield fire area identification and delineation results are output. Throughout the process, there is no specific distinction between the real disaster-causing singularity signals of the fire area and the interference clutter signals of the complex environment of the mining area. There is no dedicated multi-source data spatiotemporal registration, interference stripping, and layered singularity gradient quantification analysis and processing process. It is only suitable for conventional exploration scenarios of flat terrain and shallow exposed coal fires.

[0003] First, the compatibility of multi-source heterogeneous information is extremely poor, and the pre-identification substrate is distorted and biased. Existing technologies simply stack and splice multiple sources of data such as thermal infrared, ground temperature, geomagnetism, and surface deformation, without a unified spatiotemporal reference coordinate system, and without scaling normalization correction for the undulating terrain, slag heap accumulation, and mining subsidence of coalfields and mountains. The superimposed interference from cloud cover, solar radiation, seasonal temperature differences, and human disturbances in mining production directly leads to temporal misalignment, spatial offset, and amplitude mismatch of multi-source data, resulting in distortion of the underlying judgment substrate data. This creates hidden dangers of identification bias at the source and has extremely low adaptability to complex on-site working conditions.

[0004] Second, the lack of differentiated anti-interference stripping mechanisms results in a high proportion of false anomalies and a persistently high misjudgment rate. Existing technologies apply a fixed single threshold to screen for anomalies across the entire area, failing to accurately remove inherent false singularity interference signals from non-fire zones, such as natural thermal radiation from open-pit coalfield slag heaps, the background magnetic field of surrounding rock geology, topographic slope temperature differences, and natural settlement of shallow soil and rock. This makes it extremely easy to misjudge conventional high-temperature patches on the surface, small geomagnetic fluctuations, and non-mining-induced subsidence areas as spontaneous combustion fire zones in coalfields. At the same time, the heat from hidden smoldering fire sources deep underground cannot be effectively conducted to the surface, and weak and effective singular signals are drowned out by strong environmental clutter, frequently leading to missed fire detections and blurred and distorted fire zone boundaries.

[0005] Third, the lack of a hierarchical and quantitative analysis logic for singularities leads to a deficiency in the ability to identify deep, hidden fire zones. Existing technologies only identify visible macroscopic anomalies on the surface and have not constructed a singularity gradient hierarchical characterization model that conforms to the vertical multi-layered occurrence of coalfields and the gradual combustion evolution of fire zones. This makes it impossible to distinguish the differences between shallow surface fires, shallow and medium-shallow layer hidden spontaneous combustion, and deep latent fire hazards. It also cannot accurately match the inherent shortcomings of geophysical exploration such as transient electromagnetic shallow blind zones and high-density electrical resistivity tomography volume effects. The extraction of singularity signals from deep, weak-amplitude effective fire zones fails, and hidden fire zones and scattered small fire points on the edges are missed across the entire area, resulting in passive and delayed fire zone spread prevention and control in the later stages.

[0006] Fourth, the entire process lacks a closed-loop traceability and verification link, resulting in weak on-site adaptability and high practical costs. Current technology relies solely on manual, on-the-ground verification after anomaly identification, lacking multi-source cross-checking and singularity reverse backtracking mechanisms. The identification results lack closed-loop traceability, requiring significant manpower and material resources for verification, with long operation cycles. Furthermore, manual inspections in extreme field mining environments pose high safety risks, making it unsuitable for the urgent need for rapid, large-scale exploration operations across vast wilderness coalfields and complex mountainous coalfields. Summary of the Invention

[0007] To overcome the above technical problems, the present invention aims to provide a method for identifying singularities of multi-source information in coalfield fire areas, which realizes spatiotemporal unification of multi-source data, dynamic background anti-interference, hierarchical singularity quantification, and integrated identification of multi-source closed-loop verification. It has the characteristics of high accuracy, strong anti-interference, reliable deep identification, and strong engineering applicability.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A method for identifying singularities of multi-source information in coalfield fire zones includes the following steps;

[0010] Step 1: Differentiatedly build a pre-processing module for the fusion of multi-source information in coalfield fire areas using a unified spatiotemporal reference.

[0011] The system synchronously collects high-resolution thermal infrared remote sensing time-series images with full coverage and the same period, gridded shallow geothermal measured array data, full-domain geomagnetic vector profile data, and InSAR surface micro-deformation time-series monitoring four-dimensional homogeneous heterogeneous data. It also synchronously collects auxiliary data on mining area micro-meteorology, topography and elevation, surface cover, and working conditions at the boundary of mined-out areas. Based on the coalfield's exclusive geodetic coordinate benchmark, it unifies the spatial resolution, temporal sampling frequency, and amplitude dimension standards of all multi-source data. It completes topographic slope correction, cloud irradiance noise reduction, slag heap cover amplitude compensation, and mining-induced subsidence area distortion correction on a grid-by-grid basis, outputting a spatiotemporally homogeneous, working condition-adapted, distortion-free standardized fusion base dataset.

[0012] Step 2: Based on the standardized fusion base dataset, the background field of the in-situ normal environment of the mining area is fitted grid by grid in the partition, accurately depicting the inherent thermal background, geomagnetic background, and deformation background baseline of the non-fire zone in different blocks; synchronously linking real-time meteorological parameters and surface object specific heat capacity parameters, dynamically and adaptively correcting the background baseline fluctuation deviation, dividing the real environmental clutter interference interval of the whole domain, and providing interference benchmark constraints for subsequent singular signal purification and hierarchical identification;

[0013] Step 3: Construct a multi-level singularity gradient quantization identification and judgment model that fits the vertical occurrence of coalfields;

[0014] A three-dimensional layered grid model is constructed based on shallow exposed coal seams, shallow to medium-shallow concealed coal seams, and deep pressure-bearing coal seams. For the effective singularity signals after stripping, four-dimensional core feature vectors are extracted simultaneously, including signal amplitude gradient, temporal evolution rate, spatial contiguous extension characteristics, and vertical attenuation law. Based on the historical fire zone exploration sample database, the judgment rules are iteratively trained and classified, and the fire zones are automatically classified into three levels: Level 1 shallow surface fire high-risk fire zone, Level 2 shallow to medium-shallow concealed spontaneous combustion fire zone, and Level 3 deep latent combustion hidden fire zone. This approach specifically compensates for the technical shortcomings of conventional geophysical exploration methods, such as shallow blind spots and insufficient deep resolution. Simultaneously and accurately, the actual spatial boundaries, combustion depth, and fire spread trends of the fire zone are fitted, and the graded fire zone singularity identification results are output.

[0015] Step 4: Add a closed-loop quality control link of multi-source cross-tracing + edge point sampling verification;

[0016] For the singularity identification results of the graded fire zone output by the model, the original multi-source monitoring data is retrieved in reverse for cross-comparison and source tracing verification to check the spatiotemporal evolution continuity of singularity signals and the consistency of feature matching; only small-scale fixed-point lightweight on-site sampling verification is carried out for ambiguous points at the edge of the fire zone and points of third-level deep hidden dangers, without the need for full-area foot patrol inspection; the entire process processing log, analysis parameters, and source tracing records are automatically archived at the same time to form a standardized identification ledger that is traceable, verifiable, and reviewable.

[0017] Step 1 specifically involves:

[0018] (1) Data collection content and specific collection methods:

[0019] Multi-source monitoring data will be collected synchronously across the entire coalfield fire zone. The collection methods and parameters for each data item are specified as follows:

[0020] (11) Thermal infrared remote sensing data acquisition uses satellite thermal infrared sensors or airborne thermal infrared cameras to synchronously image the entire fire area, obtain surface thermal infrared radiation brightness values ​​and temperature inversion data, with spatial resolution controlled between 1m and 10m, and imaging time fixed between 22:00 at night and 06:00 the next day to eliminate direct solar interference. The collected content includes surface brightness temperature, radiation intensity and time series thermal anomaly sequence, covering the entire fire area and the surrounding 500m buffer zone;

[0021] (12) The gridded shallow geothermal array collects ground temperature measurement nodes in a regular grid of 10m×10m or 20m×20m to form an array monitoring network. The temperature measurement depth is divided into three levels: 0.5m, 1.0m and 2.0m. High-precision digital geothermal sensors are used for monitoring, and the sampling frequency is 1 time / 10 minutes to obtain the temperature, geothermal gradient and time series change curves of different depths.

[0022] (13) Collect geomagnetic vector profile data. Set up north-south and east-west survey lines along the coalfield to carry out high-precision geomagnetic observations on the ground. The survey line spacing is 20m to 50m and the survey point spacing is 5m to 10m. The collected parameters include total magnetic field strength, vertical component, horizontal component and magnetic anomaly gradient, which are used to identify the strange signals generated by coal fire erosion and rock magnetic variation.

[0023] (14) InSAR surface microdeformation data acquisition uses Sentinel-1 or high-resolution radar for time-series interferometry, adopts a dual-track coverage method with rising and falling rails, and a time baseline of 12 days / cycle. The acquired content includes surface subsidence rate, subsidence zone boundary and abnormal uplift zone information.

[0024] (15) Synchronously collect auxiliary working condition data, including DEM digital elevation model with a resolution of 1m to 5m, real-time meteorological data of temperature, humidity, wind speed and atmospheric radiation, surface cover data of vegetation, bare rock, slag heap and building distribution, and geological engineering data such as goaf boundary, roadway distribution and fracture distribution, to provide support for subsequent data correction:

[0025] (2) Unified spatiotemporal reference processing:

[0026] After all data acquisition was completed, a unified spatiotemporal reference was applied to the multi-source heterogeneous data. The unified spatial reference was based on the 2000 National Geodetic Coordinate System (CGCS2000), using Gauss-Kruger projection. Through control point registration and pixel-by-pixel resampling, various data types, including remote sensing, geothermal, geomagnetic, and deformation data, were unified to the same grid, the same range, and the same resolution. The unified time reference used BeiDou time synchronization as the standard, synchronizing all sensors, radars, and ground monitoring equipment to ensure accurate alignment of multi-source data from the same time and area. Unified dimensions and amplitude normalization normalized different physical quantities such as temperature, magnetic field, deformation, and radiation intensity to the [0,1] interval, eliminating identification bias caused by differences in magnitude between different physical quantities and ensuring consistency in subsequent analysis.

[0027] (3) Combined distortion correction:

[0028] After completing the spatiotemporal unification and dimensional normalization of multi-source data, four joint distortion correction processes were performed on the fused dataset to address the complex terrain, meteorological disturbances, differences in surface features, and the impact of mining in the coalfield. This aimed to eliminate data distortion caused by environmental interference and ensure the accuracy of singularity identification. First, terrain slope radiative correction was performed. Based on a high-precision DEM digital elevation model, the surface slope, aspect, and solar radiation incident angle were calculated grid-by-grid. This corrected the systematic thermal radiation deviation caused by higher temperatures on the sunny side and lower temperatures on the shady side of the slope, achieving normalization of the surface thermal infrared irradiance intensity and eliminating false anomalies caused by terrain undulations. Second, cloud cover and atmospheric noise reduction processing was conducted. Areas in the remote sensing image affected by clouds and fog were repaired using neighborhood similar pixel interpolation. Simultaneously, filtering algorithms were used to remove high-frequency interference such as atmospheric scattering and sensor noise, ensuring the continuity, smoothness, and reliability of the time-series data. Further amplitude compensation for slag heaps and land cover was conducted. Based on the land cover classification results, thermal radiation compensation coefficients were established for different land cover types, such as slag heaps, bare rock, sparse vegetation, and dense vegetation. Temperature amplitude deviations caused by differences in surface specific heat capacity and reflectivity were corrected for each type, weakening false and anomalous signals caused by land cover type. Finally, distortion correction for mining subsidence areas was carried out. Combining the distribution range of mining subsidence areas and measured data of surface subsidence, geometric distortion correction and outlier calibration were performed on the temperature, deformation, and geomagnetic data of the mining-affected areas. This eliminated data abrupt changes and shifts caused by surface subsidence and soil loosening, ensuring a stable and distortion-free data baseline across the entire region.

[0029] (4) Final output of step 1:

[0030] After synchronous acquisition, spatiotemporal unification, dimensional normalization, and joint distortion correction, the final output is a standardized multi-source fusion baseline dataset that is spatiotemporally homogeneous, has uniform resolution, is distortion-free, and adaptable to operating conditions. This provides a unified, stable, and high-quality data foundation for subsequent environmental background fitting, singular signal purification, three-dimensional hierarchical recognition, and multi-source cross-verification.

[0031] Step 2 specifically involves:

[0032] The dual-coupled discrimination mechanism of "local dynamic adaptive weak threshold screening of weak effective signals + global robust strong threshold interception of pseudo-anomaly clutter" is adopted to accurately remove all pseudo-singularity interferences such as natural high temperature of slag heap, topographic temperature difference, geomagnetic disturbance of surrounding rock, and natural rock and soil settlement, and retain the original effective singularity feature signals of pure coalfield spontaneous combustion disaster.

[0033] (1) In-situ adaptive environmental baseline fitting: Using grids as the calculation unit, historical time-series monitoring data of stable areas around each grid that are free from fire and human disturbance are selected. A dynamic environmental baseline model is constructed using a sliding window statistical fitting method to achieve in-situ, real-time, and adaptive environmental background baseline calculation. The core principle is: the arithmetic mean of the fire-free zone data represents the stable background baseline under the natural state of the grid, the standard deviation represents the random fluctuation amplitude of the environment, and k times the standard deviation is superimposed to construct a reasonable natural fluctuation range, forming an environmental baseline field that can be dynamically updated according to changes in day and night, seasons, weather, and terrain, effectively eliminating background drift and providing a true benchmark for anomaly identification. The corresponding baseline fitting formula is:

[0034] T bg =mean(T safe )+k・std(T safe )

[0035] M bg =mean(M safe )+k・std(M safe )

[0036] D bg =mean(D safe )+k・std(D safe )

[0037] In the formula: T bg M bg D bg These represent the temperature background, geomagnetic background, and deformation background of the grid, respectively; T safe M safe D safe These are the time series data of temperature, geomagnetism, and deformation for the corresponding fire-free stable zone; mean is the arithmetic mean operator used to determine the normal background level; std is the standard deviation operator used to characterize the amplitude of natural random fluctuations; k is the confidence coefficient, with a value of 2–3, used to control the confidence interval of fluctuations.

[0038] Based on this, the baseline is dynamically corrected by combining real-time meteorological parameters and surface specific heat capacity parameters, and finally the dynamic environmental baseline field of the whole region is obtained.

[0039] (2) Using the background field of the dynamic environment as a reference, the residual signal is obtained by subtracting the measured data from the background value. A strong and weak dual threshold coupling discrimination mechanism is constructed to realize the elimination of false interference and the purification of effective singular signals; a strong threshold T is set. high With weak threshold T low The discrimination rule is as follows:

[0040] When |X−X bg |<T high If the condition is deemed to be environmental interference, it will be removed.

[0041] When T low ≤|X−X bg |<T high At that time, it enters the second feature verification;

[0042] When |X−X bg |≥T low If it matches the characteristics of a fire zone, it is determined to be a valid singular signal.

[0043] In the formula: X represents the measured data from the grid, X bg The dynamic environment background value (T) obtained from the aforementioned fitting is... bg M bg D bg ),|X−X bg | represents the absolute residual between the measured value and the background value, indicating the anomaly intensity; T high A strong threshold is used to intercept strong pseudo-anomaly clutter such as high temperature in slag heaps, topographic temperature differences, and geological background; T low A weak threshold is used to sensitively capture weak and unusual signals generated by the slow oxidation of residual coal in deep, hidden fire zones.

[0044] By using dual threshold coupling discrimination, all pseudo-singularity interference from non-fire zones can be automatically removed, and the original effective singularity feature signals of pure coal fire can be accurately purified.

[0045] Step 3 specifically involves:

[0046] (1) Three-dimensional layered grid subdivision: Based on the geological conditions of the coalfield, the depth of the coal seam and the characteristics of geothermal conduction, the monitoring grid is divided into three layers along the vertical direction, corresponding to the shallow exposed coal seam, the middle and shallow concealed coal seam and the deep pressure coal seam, respectively, to realize the layered carrying and layered analysis of fire zone signals, and to provide a layered calculation carrier for the accurate identification of fires at different depths. Specifically, it is a three-dimensional grid subdivision of shallow layer 0–5m, middle layer 5–20m and deep layer below 20m;

[0047] (2) Four-dimensional singularity feature extraction: For the effective singularity signal after pseudo-interference removal, four types of core feature vectors are extracted simultaneously to construct a multi-dimensional feature system:

[0048] Amplitude gradient G: characterizes the magnitude of the spatial variation in the strength of the singular signal; Temporal evolution rate R: characterizes how fast the singular signal evolves over time; Spatial contiguous extension characteristic C: characterizes the degree of spatial connectivity and aggregation of the singular region; Vertical attenuation coefficient K: characterizes the attenuation law of the singular signal from shallow to deep.

[0049] (3) Comprehensive Singularity Index:

[0050] Based on the above four-dimensional features, a weighted and fused comprehensive singularity index calculation model is constructed to achieve a quantitative expression of the strength of singularity. The calculation formula is as follows:

[0051] SI = αG + βR + γC + δK

[0052] SI is the comprehensive singularity index, and the larger the value, the higher the probability and danger of fire disasters in the fire area; G is the amplitude gradient of the singular signal; R is the temporal evolution rate of the singular signal; C is the spatial contiguous extension feature of the singular signal; K is the vertical attenuation coefficient of the singular signal; α, β, γ, and δ are the weighting coefficients of the corresponding features, which are determined by iterative training through the historical fire area exploration sample library, satisfying α+β+γ+δ=1.

[0053] (4) Intelligent classification and determination of fire zone:

[0054] Based on the comprehensive singularity index SI, three grading thresholds SI1, SI2, and SI3 (SI1>SI2>SI3) are set to automatically classify the hazard level of fire zones. The judgment rules are as follows:

[0055] SI≥SI1: Class I shallow surface fire zone is classified as a high-risk fire zone.

[0056] SI2≤SI<SI1: Class II shallow concealed spontaneous combustion zone;

[0057] SI3≤SI<SI2: This is classified as a Class III deep latent fire hazard zone.

[0058] In the formula: SI1 is the threshold for determining open flame (Level 1); SI2 is the threshold for determining concealed spontaneous combustion (Level 2); SI3 is the threshold for determining deep latent combustion (Level 3); the classification thresholds are obtained by combining the geological conditions of the mining area and the measured data of the fire zone on site.

[0059] Step 4 specifically involves:

[0060] (1) Cross-backtracking verification of multi-source signals:

[0061] Using the fire zone identification results determined by the Singularity Index (SI) as the object, the original multi-source monitoring data after preprocessing in step 1 is retrieved in reverse, and multi-dimensional cross-source verification is carried out to verify the authenticity and reliability of the identification results. The verification content includes: verification of the temporal continuity of singular signals, verification of the spatial consistency of multi-source anomalies, and verification of the matching degree of typical coal fire evolution characteristics. By cross-verifying information from multiple dimensions, possible misjudgments from a single identification are eliminated, ensuring the authenticity and reliability of the final fire zone delineation results.

[0062] Temporal continuity verification: Singularities must remain abnormal for ≥5 consecutive days without sudden disappearance / sudden appearance, excluding transient noise;

[0063] Spatial fit verification: If the spatial overlap rate of thermal anomalies, magnetic anomalies, and deformation anomalies is ≥75%, the real fire zone is determined.

[0064] Feature matching verification: Temperature increases with depth, geomagnetic gradient remains abnormal, and deformation slowly rises / sinks, which is consistent with the spontaneous combustion law of coal;

[0065] (2) On-site sampling verification of lightweight materials:

[0066] After passing multi-source cross-referencing verification, on-site verification is only conducted for two types of uncertain locations, eliminating the need for full-area manual inspection: first, locations in boundary areas with blurred fire zone boundaries and smooth signal transitions; second, locations with weak signal amplitude and significant burial depth that pose a Level III deep latent fire hazard. The verification employs portable temperature measurement, geological observation, and anomaly point calibration methods to achieve small-scale, high-efficiency, and low-risk on-site verification, further improving the accuracy of the identification results.

[0067] (3) Closed-loop archiving and standardized ledger generation:

[0068] All data and results from the entire process are archived in a unified manner, including: raw monitoring data, spatiotemporal fusion parameters, environmental background fitting coefficients, dual threshold parameters, four-dimensional feature values, comprehensive singularity index (SI), fire zone classification results, cross-verification records, and on-site sampling review records. This automatically forms a standardized identification ledger, enabling full traceability, verification, and review of the entire process, and providing complete data support for subsequent fire zone monitoring and control.

[0069] The beneficial effects of the present invention.

[0070] Addressing the complex working conditions of coalfields and mining areas, this invention employs a unified spatiotemporal benchmark for multi-source heterogeneous monitoring information, along with a pre-processing technology for combined topographic-coverage-meteorological distortion correction. This technology is adapted to special scenarios such as coalfield goaf collapse, slag heap accumulation, and undulating mountainous terrain, achieving consistent and unbiased aggregation of four-dimensional data (thermal infrared, ground temperature, geomagnetism, and deformation), thus establishing a solid foundation for high-precision underlying data applications.

[0071] This invention fits the baseline of the normal environment in the mining area by partitioning and gridding, and uses the baseline fitting formula combined with real-time meteorological parameters and surface specific heat capacity parameters to dynamically correct the fluctuation deviation of the baseline, thus dividing the entire area into real environmental clutter interference intervals. It adopts a strong and weak dual threshold coupling discrimination mechanism, based on the residual discrimination relationship, to automatically and completely remove pseudo-singularity interference signals from non-fire areas such as natural high temperature of slag heaps, topographic temperature difference, geomagnetic disturbance of surrounding rock, and natural rock and soil settlement, accurately purifying the original effective singularity characteristic signals of coalfield spontaneous combustion disasters. It has strong anti-interference ability and can adapt to seasonal temperature differences, cloudy and sunny weather, and diurnal radiation fluctuations.

[0072] This invention constructs a multi-level singularity gradient quantization identification and judgment model that fits the vertical occurrence of coalfields. It models the three-dimensional layered grid according to shallow, mid-shallow and deep layers, and simultaneously extracts four-dimensional core feature vectors: amplitude gradient, temporal evolution rate, spatial contiguous extension characteristics, and vertical attenuation law. It adopts a comprehensive singularity index weighted model to achieve automatic classification and labeling of first-level shallow surface fire, second-level mid-shallow concealed spontaneous combustion, and third-level deep latent combustion hazards. It accurately fits the spatial boundary of the fire zone, the combustion depth and the spread trend, and makes up for the technical shortcomings of conventional geophysical exploration in shallow blind areas and insufficient deep resolution.

[0073] This invention adds a multi-source cross-tracing and edge point sampling verification closed-loop quality control link to reverse trace the identification results of graded fire zone singularity. It only conducts lightweight on-site sampling verification for points with ambiguous boundaries and third-level deep hidden danger points, eliminating the need for full-area foot patrol inspection. It simultaneously archives all process parameters and records to form a standardized identification ledger that is traceable, verifiable, and recapable, significantly reducing the workload of manual verification, shortening the operation cycle, and reducing the safety risks of field operations. It is suitable for large-scale exploration and application in large-area complex coalfields. Attached Figure Description

[0074] Figure 1 The flowchart shows the overall process for identifying singularities of multi-source information in coalfield fire zones.

[0075] Figure 2 This is a schematic diagram of spatiotemporal fusion and correction of multi-source data.

[0076] Figure 3 This is a schematic diagram of the dynamic environment background fitting and strong / weak dual threshold coupling algorithm model.

[0077] Figure 4 This is a structural diagram of a three-dimensional hierarchical singularity gradient quantization identification model.

[0078] Figure 5 This is a flowchart for multi-source cross-verification and closed-loop quality control. Detailed Implementation

[0079] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0080] A method for identifying singularities of multi-source information in coalfield fire zones includes the following steps;

[0081] Step 1: Differentiated construction of a pre-processing module for the fusion of multi-source information in coalfield fire areas (different from existing technologies that directly use data without normalization or registration);

[0082] The system synchronously collects high-resolution thermal infrared remote sensing time-series images with full coverage and the same period, gridded shallow geothermal measured array data, full-domain geomagnetic vector profile data, and InSAR surface micro-deformation time-series monitoring four-dimensional homogeneous heterogeneous data. It also synchronously collects auxiliary data on mining area micro-meteorology, topography and elevation, surface cover, and working conditions at the boundary of mined-out areas. Based on the coalfield's exclusive geodetic coordinate benchmark, it unifies the spatial resolution, temporal sampling frequency, and amplitude dimension standards of all multi-source data. It completes topographic slope correction, cloud irradiance noise reduction, slag heap cover amplitude compensation, and mining-induced subsidence area distortion correction on a grid-by-grid basis, outputting a spatiotemporally homogeneous, working condition-adapted, distortion-free standardized fusion base dataset.

[0083] Improvement effect: It completely solves the pain points of existing technologies such as misalignment, distortion and incompatibility of multi-source data, consolidates the foundation of high-precision identification of underlying data, and adapts to the complex working conditions of coalfields with all terrains and full coverage.

[0084] Step 2: The first adaptive environment background fitting + dual threshold coupling singularity pre-interference anti-interference stripping algorithm (different from the existing technology of fixed single threshold across the whole domain and no stripping of pseudo-anomalies);

[0085] Based on a standardized fusion dataset, the system fits the background field of the in-situ normal environment of the mining area in a grid-by-grid manner, accurately depicting the inherent thermal background, geomagnetic background, and deformation background baselines of different non-fire zones. Simultaneously, it links real-time meteorological parameters and surface material specific heat capacity parameters to dynamically and adaptively correct the fluctuation deviation of the background baseline, delineating the entire domain's real environmental clutter interference intervals. A dual-coupled discrimination mechanism of "local dynamic adaptive weak threshold screening for weak effective signals + global robust strong threshold interception of pseudo-anomaly clutter" is adopted to accurately eliminate all pseudo-singularity interferences such as natural high temperature of slag heaps, topographic temperature difference, geomagnetic disturbance of surrounding rock, and natural rock and soil settlement, while retaining only the original effective singularity characteristic signals of spontaneous combustion disasters in pure coalfields. Improvement effect: It reduces the false anomaly misjudgment rate from the algorithm source, simultaneously preserves effective signals of weak fires in deep areas, and completely solves the problems of misjudgment and missed detection of shallow fires in existing technologies.

[0086] Step 3: Construct a multi-level singularity gradient quantization identification and judgment model that fits the vertical occurrence of coalfields (different from existing technologies that only identify surface anomalies without hierarchical or deep analysis);

[0087] A three-dimensional layered grid model is used to model shallow exposed coal seams, shallow to medium-shallow concealed coal seams, and deep pressure-bearing coal seams. For the effective singularity signals after stripping, four-dimensional core feature vectors are simultaneously extracted: signal amplitude gradient, temporal evolution rate, spatial contiguous extension characteristics, and vertical attenuation law. Based on the historical fire zone exploration sample database, iterative training and judgment rules are used to automatically classify and label Level 1 shallow surface fire high-risk fire zones, Level 2 shallow to medium-shallow concealed spontaneous combustion fire zones, and Level 3 deep latent combustion hazard fire zones. Simultaneously, the model accurately fits the actual spatial boundaries, combustion depth, and fire spread trends of the fire zones, specifically addressing the technical shortcomings of conventional geophysical exploration methods, such as shallow blind spots and insufficient deep resolution. Improved results: Accurate identification and classification of fire zones across all depths and dimensions; precise location of concealed small fire points and deep latent combustion hazards; significantly reduced fire zone boundary delineation errors; and comprehensive fire investigation without blind spots.

[0088] Step 4: Add a multi-source cross-tracing back + edge point sampling verification closed-loop quality control link (different from the existing technology of full-domain manual review, no traceability and no quality control);

[0089] The model outputs tiered fire zone singularity identification results, which are then cross-referenced and verified using original multi-source monitoring data to check the spatiotemporal evolution continuity and feature matching consistency of the singularity signals. Only ambiguous points at the fire zone edge and third-level deep hazard points are subject to small-scale, lightweight on-site sampling verification, eliminating the need for full-area foot patrols. Simultaneously, the entire process log, analysis parameters, and tracing records are automatically archived, forming a standardized identification ledger that is traceable, verifiable, and reviewable. Improvements include: significantly reducing manual verification workload, shortening exploration operation cycles, and lowering safety risks in the field. The identification results are traceable, quality-controlled, and reusable, making them suitable for large-scale, rapid exploration scenarios in large coalfields.

[0090] Example 1: Application of singularity identification of multi-source information in the Helan Mountain Rujigou coalfield fire area:

[0091] The Rujigou mining area in Helan Mountain, Ningxia, was selected as the implementation target. This mining area is the core production area of ​​Taixi anthracite, located in the northern section of Helan Mountain. The total area of ​​the selected fire zone is 3.32 km², with an altitude of 1718~2040m. The terrain is characterized by crisscrossing valleys and exposed bedrock. There are 11 coal gangue piles, 7 mining subsidence pits, and 4 exposed coal seam sections. The deep hidden fire source is buried at a maximum depth of 280m. It is a typical complex coalfield fire zone with steep terrain, strong interference from slag piles, development of deep fire zones, and ecological sensitivity. Conventional methods have a high incidence of misjudgment and omission, and manual inspection is extremely difficult.

[0092] The method of this invention is used to accurately identify fire zones across the entire area. The implementation process is as follows:

[0093] 1) Multi-source data acquisition and spatiotemporal fusion preprocessing:

[0094] Collect the following data synchronously according to step 1:

[0095] Thermal infrared remote sensing: Landsat-8TIRS combined with UAV thermal infrared, 10m resolution, nighttime imaging from 01:00 to 04:00, covering the entire fire zone and a 500m buffer zone;

[0096] Gridded shallow ground temperature: 826 measuring points were set up in a 20m×20m grid, with monitoring depths of 0.5m, 1.0m, and 2.0m, and a sampling frequency of once every 10 minutes;

[0097] Geomagnetic vector profile: with a survey line spacing of 50m and a survey point spacing of 10m, 82 magnetic survey profiles were completed, and total field, vertical component, horizontal component and gradient data were obtained;

[0098] InSAR surface microdeformation: Sentinel-1 time series data, with a time baseline of 12 days and a deformation accuracy of 0.4 mm, to obtain the boundaries of surface subsidence and collapse;

[0099] Auxiliary data: DEM resolution 2m, real-time weather (temperature, humidity, wind speed, total radiation), surface cover, mining subsidence boundary, and fissure distribution.

[0100] Using the CGCS2000 National Geodetic Coordinate System as a unified benchmark, spatial registration, temporal alignment, and resolution normalization to a 10m×10m grid were completed. Topographic slope radiation correction, cloud cover noise reduction, slag pile cover amplitude compensation, and mining subsidence area distortion correction were carried out grid by grid to form a standardized multi-source fusion base dataset that is spatiotemporally consistent, distortion-free, and adaptable to working conditions.

[0101] 2) Dynamic environment background fitting and strong / weak dual-threshold pseudo-interference removal:

[0102] Based on a standardized dataset, using a 3×3 grid as the sliding window and a 15-day time series length, the in-situ environmental background is fitted grid-by-grid, as shown in the following formula:

[0103] T bg =mean(T safe )+k・std(T safe )

[0104] M bg =mean(M safe )+k・std(M safe )

[0105] D bg =mean(D safe )+k・std(D safe )

[0106] In the formula: T bg M bg D bgThese are the temperature background, geomagnetic background, and deformation background, respectively; T safe M safe D safe This is time series data for the fireless stable region; mean is the arithmetic mean, std is the standard deviation; k=2.5.

[0107] By combining real-time meteorological parameters with the dynamic correction of the specific heat capacity of surface objects to the baseline, a dynamic environmental baseline field for the entire region is obtained, which is uniformly denoted as X. bg (i.e., the aforementioned T) bg M bg D bg (A general term). Set dual thresholds:

[0108] Strong threshold T high Temperature 18.2℃, geomagnetic field 28nT, deformation 1.2mm / d;

[0109] Weak threshold T low Temperature 7.1℃, geomagnetic field 9nT, deformation 0.3mm / d.

[0110] Judgment rules:

[0111] |X−X bg |<T high : Determined to be environmental interference, it was removed;

[0112] T low ≤|X−X bg |<T high : Enter feature secondary verification;

[0113] |X−X bg |≥T low Furthermore, it conforms to the characteristics of coal fire evolution: thus it is determined to be a valid singular signal.

[0114] This embodiment eliminated 142 false anomalies, such as high temperature of slag heap, topographic temperature difference, magnetic disturbance of surrounding rock, and natural settlement, while retaining 83 original and effective singular signals of coal fire, thus significantly reducing the misjudgment rate from the source.

[0115] 3) Three-dimensional hierarchical singularity gradient quantization identification and fire classification:

[0116] Based on the vertical occurrence characteristics of the Helan Mountain coal seams, a three-dimensional layered grid model was constructed, consisting of shallow layers (0-5m), middle layers (5-20m), and deep layers (below 20m). Four-dimensional eigenvectors—amplitude gradient G, temporal evolution rate R, spatial connectivity C, and vertical attenuation coefficient K—were extracted from the purified effective singular signals. The comprehensive singularity index was calculated as: SI = α・G + β・R + γ・C + δ・K, where the weighting coefficients were determined through training with historical fire zone samples: α = 0.35, β = 0.25, γ = 0.20, and δ = 0.20.

[0117] Set the grading thresholds: SI1=0.75, SI2=0.50, SI3=0.25, and automatically determine the grading.

[0118] SI≥0.75: There are 3 Class I shallow surface fire high-risk fire zones with a total area of ​​0.21 km²;

[0119] 0.50≤SI<0.75: 7 Class II shallow and concealed spontaneous combustion fire zones, with a total area of ​​0.48km²;

[0120] 0.25≤SI<0.50: There are 4 Class III deep latent fire hazard areas, with a total area of ​​0.34km².

[0121] It simultaneously outputs precise spatial boundaries of the fire zone, burial depth of the combustion center, and average annual lateral spread rate (14~16m / a), effectively making up for the shortcomings of conventional geophysical exploration in shallow blind areas and insufficient resolution in deep areas.

[0122] 4) Multi-source cross-validation and lightweight sampling closed-loop quality control:

[0123] The graded identification results are verified through triple cross-source tracing: ① the anomaly signal is temporally continuous and stable (≥12 days); ② the spatial consistency of thermal, magnetic, and deformation anomalies is ≥85%; ③ the signal evolution conforms to the temperature rise law of coal spontaneous combustion. Portable temperature measurement, gas detection, and geological observation point verification are conducted only for 25 fire zone boundary ambiguities and 12 deep low signal-to-noise ratio hazard points, eliminating the need for full-area foot patrols. Simultaneously and automatically, raw data, correction parameters, background coefficients, thresholds, grading results, and verification records are archived, forming a standardized identification ledger that is traceable, verifiable, and reproducible.

[0124] Implementation effect

[0125] Fire zone identification results: 3 Class I open flames, 7 Class II concealed spontaneous combustions, and 4 Class III deep hidden dangers, with a consistency rate of 96.7% with drilling / grouting treatment verification;

[0126] Average boundary positioning error: ≤4.2m;

[0127] The false anomaly detection rate was reduced by 82.6% compared to the traditional fixed threshold method;

[0128] The detection rate of concealed fire zones at depths of ≥20m increased by 66.3%;

[0129] On-site verification workload was reduced by 71.8%, and the work cycle was shortened by 61.2%;

[0130] It can accurately match the management requirements of the Helan Mountain Ecological Protection Area, providing high-precision data support for fire zone management and ecological restoration.

[0131] The results show that the method of the present invention is highly adaptable to the complex working conditions of the Helanshan coalfield, which has large topographic relief, strong slag pile interference, deep fire zone development, and ecological sensitivity. It has high identification accuracy, strong anti-interference ability, and good engineering implementation, and has significant technical advantages and promotion and application value.

Claims

1. A method for identifying singularities of multi-source information in coalfield fire zones, characterized in that, Includes the following steps; Step 1: Collect four-dimensional homogeneous heterogeneous data synchronously across the entire region, and collect auxiliary data synchronously; based on the coalfield's exclusive geodetic coordinate benchmark, unify the spatial resolution, temporal sampling frequency, and amplitude dimension standard of all multi-source data, and complete terrain slope correction, cloud irradiance noise reduction, slag pile cover amplitude compensation, and mining-induced subsidence area distortion correction grid by grid, and output a spatiotemporally homogeneous, working condition adapted, distortion-free standardized fusion base dataset; Step 2: Based on the standardized fusion base dataset, the background field of the in-situ normal environment of the mining area is fitted grid by grid in the partition, and the inherent thermal background, geomagnetic background and deformation background baseline of the non-fire area of ​​different blocks are characterized. Synchronous linkage with real-time meteorological parameters and specific heat capacity parameters of surface objects dynamically and adaptively corrects the background baseline fluctuation deviation, delineates the clutter interference range of the real environment across the entire area, and provides interference benchmark constraints for subsequent purification and hierarchical identification of singular signals. Step 3: Model a three-dimensional layered grid according to shallow exposed coal seams, shallow and medium-shallow concealed coal seams, and deep pressure coal seams. For the effective singularity signals after stripping, simultaneously extract four-dimensional core feature vectors: signal amplitude gradient, temporal evolution rate, spatial contiguous extension characteristics, and vertical attenuation law. Based on the historical fire zone exploration sample database, iteratively train the judgment rules and automatically classify and label the first-level shallow surface fire high-risk fire zone, the second-level shallow and medium-shallow concealed spontaneous combustion fire zone, and the third-level deep latent combustion hidden fire zone. Simultaneously and accurately fit the real spatial boundary, combustion depth, and fire spread trend of the fire zone, and output the singularity identification results of the graded fire zone. Step 4: For the singularity identification results of the graded fire zone output by the model, retrieve the original multi-source monitoring data in reverse to conduct cross-comparison and source tracing verification, and check the spatiotemporal evolution continuity and feature matching consistency of the singularity signal; only conduct small-scale fixed-point lightweight on-site sampling verification for ambiguous points at the edge of the fire zone and points of third-level deep hidden dangers, without the need for full-area foot patrol inspection; simultaneously and automatically archive the whole process processing log, analysis parameters, and source tracing records to form a standardized identification ledger that is traceable, verifiable, and reviewable.

2. The method for identifying singularities of multi-source information in coalfield fire zones according to claim 1, characterized in that, In step 1, high-resolution thermal infrared remote sensing time-series images with full coverage and the same period are collected synchronously across the entire region, along with gridded shallow geothermal measured array data, global geomagnetic vector profile data, and InSAR surface micro-deformation time-series monitoring four-dimensional homogeneous heterogeneous data. Simultaneously, auxiliary data on mining area micro-meteorology, topography, elevation, surface cover, and working conditions at the boundary of mined-out areas are also collected. Based on the coalfield's exclusive geodetic coordinate benchmark, the spatial resolution, temporal sampling frequency, and amplitude dimension standards of all multi-source data are unified. Topographic slope correction, cloud irradiance noise reduction, slag heap cover amplitude compensation, and distortion correction of mining-induced subsidence areas are completed grid by grid. The result is a standardized fusion base dataset that is spatiotemporally homogeneous, adaptable to working conditions, and distortion-free.

3. The method for identifying singularities of multi-source information in coalfield fire zones according to claim 2, characterized in that, Step 1 specifically involves: (1) Multi-source monitoring data were collected synchronously across the entire coalfield fire zone. The collection methods and parameters for each data item are specified as follows; (11) Thermal infrared remote sensing data acquisition uses satellite thermal infrared sensors or airborne thermal infrared cameras to synchronously image the entire fire area, obtain surface thermal infrared radiation brightness values ​​and temperature inversion data, with spatial resolution controlled between 1m and 10m, and imaging time fixed between 22:00 at night and 06:00 the next day to eliminate direct solar interference. The collected content includes surface brightness temperature, radiation intensity and time series thermal anomaly sequence, covering the entire fire area and the surrounding 500m buffer zone; (12) The gridded shallow geothermal array collects ground temperature measurement nodes in a regular grid of 10m×10m or 20m×20m to form an array monitoring network. The temperature measurement depth is divided into three levels: 0.5m, 1.0m and 2.0m. High-precision digital geothermal sensors are used for monitoring, and the sampling frequency is 1 time / 10 minutes to obtain the temperature, geothermal gradient and time series change curves of different depths. (13) Collect geomagnetic vector profile data. Set up north-south and east-west survey lines along the coalfield to carry out high-precision geomagnetic observations on the ground. The survey line spacing is 20m to 50m and the survey point spacing is 5m to 10m. The collected parameters include total magnetic field strength, vertical component, horizontal component and magnetic anomaly gradient, which are used to identify the strange signals generated by coal fire erosion and rock magnetic variation. (14) InSAR surface microdeformation data acquisition uses Sentinel-1 or high-resolution radar for time-series interferometry, adopts a dual-track coverage method with rising and falling rails, and a time baseline of 12 days / cycle. The acquired content includes surface subsidence rate, subsidence zone boundary and abnormal uplift zone information. (15) Synchronously collect auxiliary working condition data, including DEM digital elevation model with a resolution of 1m to 5m, real-time meteorological data of temperature, humidity, wind speed and atmospheric radiation, surface cover data of vegetation, bare rock, slag heap and building distribution, and geological engineering data such as goaf boundary, roadway distribution and fracture distribution, to provide support for subsequent data correction: (2) After all data collection is completed, a unified spatiotemporal reference is implemented for the multi-source heterogeneous data. The unified spatial reference is based on the 2000 National Geodetic Coordinate System and adopts the Gauss-Kruger projection. Through control point registration and pixel-by-pixel resampling, various types of data such as remote sensing, ground temperature, geomagnetism, and deformation are unified to the same grid, the same range, and the same resolution. The unified time reference is based on BeiDou time synchronization. All sensors, radars, and ground monitoring equipment are synchronized to ensure that multi-source data in the same area at the same time can be accurately aligned. Unified dimensions and amplitude normalization unify and normalize different physical quantities such as temperature, magnetic field, deformation and radiation intensity to the [0,1] interval, eliminate the identification bias caused by the difference in magnitude between different physical quantities, and ensure the consistency of subsequent analysis. (3) After completing the spatiotemporal unification and dimensional normalization of multi-source data, four joint distortion correction processes are carried out on the fused dataset to eliminate data distortion caused by environmental interference, taking into account the complex terrain of the coalfield, meteorological disturbances, differences in surface features and mining impacts. First, terrain slope radiation correction is carried out. Based on the high-precision DEM digital elevation model, the surface slope, slope aspect and solar radiation incident angle are calculated grid by grid. The systematic thermal radiation deviation caused by the higher temperature on the sunny side of the hillside and the lower temperature on the shady side is corrected to achieve the normalization of the surface thermal infrared radiation intensity and eliminate the false anomalies caused by terrain undulation. Secondly, cloud cover and atmospheric noise reduction processing were carried out. The neighbor similar pixel interpolation method was used to repair the areas in the remote sensing image affected by clouds and fog. At the same time, filtering algorithms were used to remove high-frequency interference such as atmospheric scattering and sensor noise. Further compensation for the amplitude of slag heaps and land cover was carried out. Based on the classification results of land cover, thermal radiation compensation coefficients were established for different land cover types such as gangue heaps, bare rock, sparse vegetation, and dense vegetation. Temperature amplitude deviations caused by differences in surface specific heat capacity and reflectivity were corrected for each type, thereby weakening false and singular signals caused by land cover types. Finally, distortion correction of the mining subsidence area was carried out. Based on the distribution range of the mining subsidence area and the measured data of surface subsidence, geometric distortion correction and outlier calibration were performed on the temperature, deformation and geomagnetic data of the mining-affected area. (4) After synchronous acquisition, spatiotemporal unification, dimensional normalization and joint distortion correction, the final output is a standardized multi-source fusion base dataset with spatiotemporal homogeneity, unified resolution, no distortion and working condition adaptation, which provides a unified, stable and high-quality data foundation for subsequent environmental background fitting, singular signal purification, three-dimensional hierarchical recognition and multi-source cross-verification.

4. The method for identifying singularities of multi-source information in coalfield fire zones according to claim 1, characterized in that, Step 2 specifically involves: (1) In-situ environmental baseline adaptive fitting: Using grids as the calculation unit, historical time series monitoring data of stable areas with no fire and no human disturbance around each grid are selected, and a dynamic environmental baseline model is constructed by using the sliding window statistical fitting method to realize in-situ, real-time and adaptive environmental background baseline calculation; The arithmetic mean of the fire-free zone data represents the stable background baseline under natural grid conditions, and the standard deviation represents the amplitude of random environmental fluctuations. A reasonable natural fluctuation range is constructed by superimposing k times the standard deviation, forming an environmental background field that can be dynamically updated according to changes in day and night, seasons, weather, and topography. The corresponding background fitting formula is: T bg =mean(T safe )+k・std(T safe ) M bg =mean(M safe )+k・std(M safe ) D bg =mean(D safe )+k・std(D safe ) In the formula: T bg M bg D bg These represent the temperature background, geomagnetic background, and deformation background of the grid, respectively; T safe M safe D safe These are the time-series data of temperature, geomagnetism, and deformation for the corresponding fire-free stable zone; mean is the arithmetic mean operator used to determine the normal background level; std is the standard deviation operator used to characterize the amplitude of natural random fluctuations; k is the confidence coefficient, with a value of 2–3, used to control the confidence interval of fluctuations; Based on this, the baseline is dynamically corrected by combining real-time meteorological parameters and surface specific heat capacity parameters, and finally the dynamic environmental baseline field of the whole region is obtained. (2) Based on the background field of the dynamic environment, the residual signal is obtained by subtracting the measured data from the background value. A strong and weak dual threshold coupling discrimination mechanism is constructed to realize the elimination of false interference and the purification of effective singular signals. Set a strong threshold T high With weak threshold T low The discrimination rule is as follows: When |X−X bg |<T high If the condition is deemed to be environmental interference, it will be removed. When T low ≤|X−X bg |<T high At that time, it enters the second feature verification; When |X−X bg |≥T low If it matches the characteristics of a fire zone, it is determined to be a valid singular signal; In the formula: X represents the measured data from the grid, X bg The dynamic environment background value (T) obtained from the aforementioned fitting is... bg M bg D bg ),|X−X bg | represents the absolute residual between the measured value and the background value, indicating the intensity of the anomaly; T high A strong threshold is used to intercept strong pseudo-anomaly clutter caused by high temperatures in slag heaps, topographic temperature differences, and geological background; T low A weak threshold is used to sensitively capture weak and unusual signals generated by the slow oxidation of residual coal in deep, hidden fire zones. By using dual threshold coupling discrimination, all pseudo-singularity interference from non-fire zones can be automatically removed, and the original effective singularity feature signals of pure coal fire can be accurately purified.

5. The method for identifying singularities of multi-source information in coalfield fire zones according to claim 1, characterized in that, Step 3 specifically involves: (1) Based on the geological conditions of the coalfield, the depth of the coal seam and the characteristics of geothermal conduction, the monitoring grid is divided into three layers along the vertical direction, corresponding to the shallow exposed coal seam, the shallow and medium-depth concealed coal seam and the deep pressure coal seam, so as to realize the layered carrying and layered analysis of the fire zone signal. (2) For the effective singularity signal after pseudo-interference removal, four types of core feature vectors are extracted simultaneously to construct a multi-dimensional feature system: Amplitude Gradient G: Characterizes the magnitude of the spatial variation in the strength of the singular signal; Temporal evolution rate R: Characterizes how quickly the singular signal evolves over time; Spatial contiguous extension characteristic C: Characterizes the degree of spatial connectivity and aggregation of the singular region; Vertical attenuation coefficient K: (3) Based on the above four-dimensional features, a weighted and fused comprehensive singularity index calculation model is constructed to realize the quantitative expression of the singularity strength. The calculation formula is as follows: SI = αG + βR + γC + δK SI is the comprehensive singularity index, and the larger the value, the higher the probability and danger of fire disaster; G is the amplitude gradient of the singular signal; R is the temporal evolution rate of the singular signal; C is the spatial contiguous extension feature of the singular signal; K is the vertical attenuation coefficient of the singular signal; α, β, γ, and δ are the weighting coefficients of the corresponding features, which are determined by iterative training through the historical fire zone exploration sample database, satisfying α+β+γ+δ=1; (4) Based on the comprehensive singularity index SI, set classification thresholds SI1, SI2, and SI3 (SI1>SI2>SI3) to automatically classify the fire zone hazard level. The judgment rule is as follows: SI≥SI1: Class I shallow surface fire zone is classified as a high-risk fire zone. SI2≤SI<SI1: Class II shallow concealed spontaneous combustion zone; SI3≤SI<SI2: Class III deep latent fire hazard zone; In the formula: SI1 is the threshold for determining open flame (Level 1); SI2 is the threshold for determining concealed spontaneous combustion (Level 2); SI3 is the threshold for determining deep latent combustion (Level 3); the classification thresholds are obtained by combining the geological conditions of the mining area and the measured data of the fire zone on site.

6. The method for identifying singularities of multi-source information in coalfield fire zones according to claim 1, characterized in that, Step 4 specifically involves: (1) Taking the fire zone identification result determined by the comprehensive singularity index SI as the object, the original multi-source monitoring data after the preprocessing in step 1 is retrieved in reverse, and multi-dimensional cross-source verification is carried out to verify the authenticity and reliability of the identification result. The verification content includes: verification of the temporal continuity of singular signals, verification of the spatial consistency of multi-source anomalies, and verification of the matching degree of typical coal fire evolution characteristics; Temporal continuity verification: Singularities must remain abnormal for ≥5 consecutive days without sudden disappearance / sudden appearance, excluding transient noise; Spatial fit verification: If the spatial overlap rate of thermal anomalies, magnetic anomalies, and deformation anomalies is ≥75%, the real fire zone is determined. Feature matching verification: Temperature increases with depth, geomagnetic gradient remains abnormal, and deformation slowly rises / sinks, which is consistent with the spontaneous combustion law of coal; (2) After passing the multi-source cross-tracing verification, on-site verification is only carried out for two types of uncertain points, without the need for full-area manual inspection: one is the boundary area points with blurred fire zone boundaries and smooth signal transition; the other is the third-level deep latent fire hazard points with weak signal amplitude and large burial depth. (3) Archive all data and results in a unified manner, including: original monitoring data, spatiotemporal fusion parameters, environmental background fitting coefficients, dual threshold parameters, four-dimensional feature values, comprehensive singularity index SI, fire zone classification results, cross-verification records, and on-site sampling verification records, and automatically form a standardized identification ledger.

Images