Coal pile spontaneous combustion early warning method based on space-time gradient field analysis

Abstract

This invention discloses a method for early warning of spontaneous combustion in coal piles based on spatiotemporal gradient field analysis. The method includes: constructing a three-dimensional multimodal voxel mesh model of the coal pile containing geometric, temperature, and gas concentration attributes; calculating spatial gradient vectors of temperature and gas concentration using discrete difference; identifying surface hot spots and gas leaks based on the gradient vectors; and using a thermal-gas coupling positioning model to inversely deduce the equivalent spontaneous combustion core coordinates inside the coal pile; extracting spatiotemporal evolution characteristics of the core region, such as temperature rise rate, acceleration, and temperature-gas correlation coefficient; inputting these characteristics into a comprehensive risk assessment model to calculate a risk index and trigger a graded early warning. This invention achieves precise three-dimensional positioning of hidden fire sources inside the coal pile through multiphysics field coupling analysis, and effectively eliminates external interference such as direct sunlight by utilizing temperature rise acceleration and multi-field correlation, thus realizing early and accurate source tracing and trend prediction of spontaneous combustion disasters.

Description

Technical Field

[0001] This invention relates to the field of industrial safety monitoring and data analysis technology, and in particular to a method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis, which utilizes multimodal data fusion technology to locate internal fire sources and predict combustion trends in enclosed coal yards. Background Technology

[0002] Coal-fired power plants have enclosed coal storage areas with enormous volumes. During long-term storage, coal is highly susceptible to spontaneous combustion due to oxidation. This not only results in economic losses of fuel but also produces large amounts of toxic and harmful gases (such as carbon monoxide and sulfur dioxide), seriously threatening the lives of on-site personnel and the operational safety of related equipment. Therefore, early and accurate monitoring and warning of spontaneous combustion in coal piles are crucial.

[0003] Currently, existing coal yard safety monitoring technologies mainly include contact temperature measurement (such as insertion thermocouples) and non-contact monitoring (such as infrared thermal imaging and gas detectors). However, in practical applications, these existing technologies generally suffer from the following significant drawbacks: Lack of internal imaging and positioning capabilities: Coal is a poor conductor of heat, and the heat generated by internal spontaneous combustion has a significant lag in its conduction to the surface. While existing infrared thermal imagers can monitor surface temperatures over a wide area, by the time they capture a high-temperature "hot spot" on the surface, a large fire core has often already formed inside. Furthermore, due to the nonlinearity of heat conduction paths, the hottest spot on the surface often does not directly correspond to the vertical position of the internal ignition point. Relying solely on surface temperature data cannot accurately determine the depth and specific three-dimensional coordinates of the internal ignition point, making it difficult for fire monitors or sprinkler systems to accurately focus and extinguish the fire.

[0004] The "island effect" and spatial discrepancies of multi-source data: Existing infrared temperature measurement systems and gas monitoring systems typically operate independently. Gas sensors (monitoring CO, CH4, etc.) are usually installed on the coal shed walkway, representing sparse point measurements. Due to factors such as strong ventilation and thermal buoyancy diffusion within the coal shed, the high-concentration areas detected by the sensors often deviate from the actual leak source. Current technology lacks an effective algorithm to fuse and analyze the surface temperature field and the atmospheric gas field in a unified three-dimensional space, resulting in both inaccurate temperature measurement and uncertain gas measurement.

[0005] High false alarm rate and lack of evolutionary trend analysis: Traditional early warning logic often uses fixed threshold methods (e.g., alarming when the temperature exceeds 60℃). However, the coal yard environment is complex and affected by external factors such as direct sunlight, equipment exhaust, and operational friction, which easily leads to false positives (e.g., sunlight causes surface heating, but there is no spontaneous combustion inside). In addition, existing systems lack in-depth analysis of time-varying characteristics such as temperature rise acceleration and gas accumulation rate, making it difficult to distinguish between normal physical thermal fluctuations and malignant chemical oxidation acceleration, and failing to provide prediction functions for remaining disposal time. Summary of the Invention

[0006] This invention primarily addresses the technical problems of existing technologies, such as the inability to accurately locate the internal ignition point, low utilization of multi-source data, and high false alarm rate. It provides a coal pile spontaneous combustion early warning method based on spatiotemporal gradient field analysis, which can integrate three-dimensional spatial information, surface temperature field, and ambient gas concentration field. Through mathematical algorithms, it reverse-engineers the location of the internal spontaneous combustion core and eliminates external interference to provide accurate trend early warning.

[0007] The present invention addresses the aforementioned technical problems primarily through the following technical solution: a method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis, comprising the following steps: S1: Construct a three-dimensional multimodal data field for the coal yard: acquire real-time three-dimensional geometric point cloud data of the coal yard space, two-dimensional infrared temperature data of the coal pile surface, and multi-point gas concentration data of the environment; map the data to a unified three-dimensional coordinate system of the coal yard, and construct a three-dimensional multimodal voxel mesh model containing geometric attributes, temperature attributes, and gas concentration attributes. S2: Calculate the spatial gradient vector: Traverse the voxel mesh model and use the discrete difference algorithm to calculate the temperature gradient vector of the voxels on the coal pile surface and the gas concentration gradient vector of the voxels in the air domain, respectively. S3: Reverse Source Tracing to Locate the Equivalent Spontaneous Combustion Core: Based on the temperature gradient vector, identify the centroid of the hot spot region on the coal pile surface; based on the gas concentration gradient vector and the high-concentration region, inversely deduce the coordinates of the gas leak outlet; according to the spatial relationship between the centroid of the hot spot region and the coordinates of the gas leak outlet, use the heat-gas coupling location model to calculate the coordinates P of the equivalent spontaneous combustion core inside the coal pile. core ; S4: Extract spatiotemporal evolution features: using the equivalent spontaneous combustion core coordinates P core A spatial neighborhood query range is constructed around the center, and temperature and gas concentration sequences within a preset time window in the past are retrieved from the historical time series database; the temperature rise rate v is calculated based on the sequences. T Temperature rise acceleration a T And the temperature-gas correlation coefficient ρ; S5: Intelligent Judgment and Early Warning: The P... core vT a T The risk index is calculated using the integrated risk assessment model with input ρ, and a graded early warning signal is triggered based on the risk index.

[0008] This solution establishes a digital foundation capable of uniformly describing the physical state of a coal yard. Traditional monitoring often relies on a single infrared image (2D) or several instrument readings (1D). This invention, through step S1, maps geometric location (obtained by laser scanning), thermodynamic state (obtained by infrared), and chemical composition (obtained by gas sensors) to the same three-dimensional coordinate system. This allows a computer to analyze the physical properties of every cubic meter of space in the coal yard.

[0009] This invention goes beyond simply monitoring static values. It calculates gradient vectors in step S2 to identify the directions of fastest temperature and concentration changes; it uses reverse tracing in step S3 to resolve the spatial misalignment between surface phenomena and internal essence; and it introduces a time dimension through spatiotemporal evolution in step S4, using the rate of change to assess the severity of the situation. This analytical logic, moving from static to dynamic and from surface to internal, is key to achieving accurate early warning.

[0010] Preferably, in step S1, constructing the three-dimensional multimodal voxel mesh model specifically includes: S11: An octree data structure is used to store the spatial data of the coal yard. The coal yard space is enormous (e.g., 300 meters long and 30 meters high). If a uniform grid is used for storage, the data volume will increase cubically, making real-time computation difficult. Using an octree structure allows for sparse storage of the air domain and fine-grained storage of the coal pile surface and interior, greatly improving the retrieval efficiency of subsequent gradient calculations and forming the foundation for real-time analysis.

[0011] S12: Using a depth buffer (Z-buffer) detection algorithm, rays are emitted from the optical center of the infrared thermal imager towards the 3D geometric point cloud to remove occluded voxel nodes, ensuring that the infrared temperature data is mapped only to the visible voxels on the coal pile surface. Infrared thermal imagers are 2D projectors, subject to perspective occlusion (e.g., only the front slope of the coal pile is visible, not the back slope). Without occlusion detection, directly mapping infrared pixels onto the 3D model would result in temperature mapping errors. This step clarifies the use of the Z-buffer algorithm from computer graphics to remove occluded points, ensuring that the temperature data is accurately attached to the visible surface.

[0012] S13: Anisotropic inverse distance weighted interpolation (IDW) is used to interpolate discretely distributed gas concentration data into air domain voxels. When calculating the interpolation weights, the distance weight coefficient along the vertical Z-axis is smaller than that along the horizontal X and Y axes to simulate the upward diffusion characteristics of gas. Furthermore, when the interpolation path is blocked by a coal pile voxel, the diffusion distance is set to infinity. Anisotropic interpolation is a crucial physical correction. Ordinary inverse distance weighted interpolation assumes uniform gas diffusion in all directions (spherical). However, in reality, gases produced by spontaneous combustion (such as CO and CH4) typically have thermal buoyancy, and horizontal ventilation exists within the coal shed, making it easier for gases to diffuse upwards or downwind. Therefore, this invention assigns a smaller distance weight coefficient to the vertical direction (Z-axis) during interpolation (meaning stronger correlation and a greater influence distance in the vertical direction), thereby constructing an ellipsoidal gas concentration field that better conforms to physical laws.

[0013] Preferably, in step S2, the calculation of the spatial gradient vector specifically adopts the central difference method. For any non-empty node (i,j,k) in the mesh, the gradient vector G of its physical quantity U is calculated as follows: G=((U i+1 -U i-1 ) / (2Δx),(U j+1 -U j-1 ) / (2Δy),(U k+1 -U k-1 ) / (2Δz)); Where U represents temperature T or gas concentration C, and Δx, Δy and Δz are the side lengths of the voxel grid, respectively; before calculation, the voxel grid data is first smoothed by Gaussian filtering.

[0014] The on-site environment contains dust interference and sensor electronic noise, which would cause the gradient direction to become chaotic if used directly for differential calculation. Gaussian filtering is then applied for smoothing, which extracts the main temperature / concentration distribution trends, ensuring that the calculated gradient vector G stably points towards the center of the heat source or leakage source.

[0015] Compared to forward or backward difference methods, the central difference method has a second-order truncation error, higher accuracy, and can more accurately describe the local rate of change of the temperature field and spatial concentration field on the coal pile surface.

[0016] Preferably, in step S3, identifying the centroid of the hot spot region on the coal pile surface specifically includes: S311: Extraction of surface voxel sets with an extraction temperature T greater than the ambient reference temperature; S312: The set is divided into several independent hot spot regions using the DBSCAN clustering algorithm; S313: Calculate the weighted centroid O for each hot spot region. hotThe weight is the temperature value of the voxel.

[0017] High-temperature areas on the surface of coal yards are often irregular in shape (strip-like, spot-like) and may contain noise. Clustering algorithms such as K-Means require a preset number of clusters and are not suitable for this scenario. This invention uses DBSCAN (density-based clustering algorithm), which can automatically discover hot spot regions of arbitrary shapes and effectively remove outliers and noise, improving the robustness of identification. While a standard centroid is merely a geometric center, the weighted centroid used in this scheme considers temperature distribution, resulting in a more accurate calculated O0. hot It tends to focus on the areas with the highest temperatures, thus more accurately indicating the direction of the heat source. Based on the above process, the object of "hot spot" can be abstracted from thousands of high-temperature voxels.

[0018] Preferably, in step S3, the reverse deduction of the gas leak coordinates specifically includes: S321: Identify high-concentration air masses in the air domain voxels whose concentration values ​​exceed the background threshold, and calculate their geometric center O. gas ; S322: Along the direction of gravity, i.e., the -Z axis direction, O gas Projecting the image onto the surface of the coal pile yields projection point P. proj ; S323: Get P proj The gas concentration gradient vector at point P is used to take its horizontal component. proj By translating along the direction of this horizontal component, the corrected coordinates of the gas leak point O are obtained. leak .

[0019] Directly performing CFD inversion is computationally intensive and extremely sensitive to boundary conditions, making it difficult to implement in real-time in engineering. This invention creatively proposes a simplified "projection + correction" strategy. First, based on the principle of gravity settling / buoyancy, it is assumed that the leak point is located below (or projected above) the high-concentration gas cloud; then, fine-tuning is performed using the horizontal component of the local concentration gradient. This method is computationally fast and has sufficient engineering accuracy in most coal shed environments without strong wind interference, ensuring algorithm convergence and preventing unsolvable problems. This is a robust gas source localization algorithm.

[0020] Preferably, in step S3, the equivalent spontaneous combustion core coordinates P inside the coal pile are calculated using the thermal-gas coupling positioning model. core The formula is: P core =O hot -n×(C base +λ×d); Among them, O hot Let n be the coordinates of the centroid of the hot spot region; n is the coordinate of O. hotThe surface normal vector at point O points in the coal pile; d is the surface normal vector at point O. hot With O leak The Euclidean distance between them; C base The preset burial depth parameter is λ; λ is the air permeability conductivity coefficient, which can generally be taken as 0.4-0.8. When there are multiple hot spot areas or multiple gas leaks, the system traverses all combinations and only considers combinations where the Euclidean distance d is less than the preset correlation threshold (e.g., 15 meters). hot O leak ) to perform P core The calculation.

[0021] The equivalent spontaneous combustion core is a calculated virtual coordinate system that represents the most likely internal energy center causing the current surface temperature and gas anomalies. core The location is based on the surface hot spot centroid O hot It is obtained by extending inward along the normal n. The depth of the extension depends on two factors: one is the foundation depth parameter C. base Second, the distance d between the hot spot and the leak.

[0022] Foundation burial depth parameter C base It is an empirical physical quantity that characterizes the minimum statistical depth of an internal ignition source capable of causing a significant surface temperature rise through thermal conduction. In fact, C... base The value of C is not fixed, but closely related to the type of coal (such as lignite and bituminous coal) and the stage of spontaneous combustion (latent period, self-heating period, etc.). For example, referring to the commonly used data on the depth of spontaneous combustion heating points in closed coal yards, for lignite in the latent period, C... base A depth of 0.5-1.5 meters is acceptable; however, for the stage of intensified exothermic oxidation, C... base The distance can be 1.5-3.0 meters. In this invention, an intermediate value can be preset or dynamically matched based on historical data.

[0023] The physical logic of this deep inference algorithm lies in establishing a mapping relationship between surface features and internal states. Specifically, if the centroid of the surface hotspot O... hot and gas leak point O leak If the two are very close (i.e., the d value is small), it indicates that the internal heat and gas transport paths are approximately perpendicular, suggesting that the fire source is shallow and the gas escapes directly. If the two are far apart (i.e., the d value is large), it means that the gas flows through a long distance of gaps inside the coal pile before escaping to the surface, implying that the fire source should be deeper (i.e., a depth correction of λ×d is needed). By introducing the permeability conductivity coefficient λ, it further adapts to the different diffusion path tortuosity caused by the porosity differences of different coal types (such as compacted coal and loose coal).

[0024] Furthermore, although determining the horizontal coordinates (X,Y) is sufficient for basic aiming requirements when using a fire monitor for surface spraying, the depth information Z calculated by this invention has significant decision-making support value: 1. Basis for graded treatment: A shallower calculation depth (e.g., <1.5m) indicates that spontaneous combustion may be in the incubation period or shallow oxidation, and mechanical turning and heat dissipation are suitable; while a deeper calculation depth (e.g., >3m) indicates that high-temperature cavities or deep cores may have formed inside, and blind turning may lead to a sudden increase in oxygen supply and trigger deflagration. In this case, deep water injection rods should be inserted or isolation and compaction measures should be taken.

[0025] 2. Trend verification: If the calculated depth Z gradually becomes shallower over time, it indicates that the fire source is spreading to the surface, and the risk level needs to be increased.

[0026] Therefore, in-depth analysis is a key element in achieving accurate hierarchical early warning and scientific response in this method.

[0027] The permeability coefficient λ is introduced to accommodate the porosity differences of different coal types (such as compacted coal and loose coal).

[0028] Preferably, in step S4, the extraction of spatiotemporal evolution features specifically includes: S41: Based on the calculated P core With the center of the sphere as the center, construct a query sphere with a preset spatial radius R as the radius; S42: Retrieve the historical time-series database to obtain the maximum temperature T of all voxels falling within the sphere at each historical moment. max (t) and average concentration C avg (t); S43: To T max (t) The time series is linearly fitted using the least squares method, and the slope of the fitted line is the rate of temperature rise v. T ; S44: For adjacent time steps v T By performing differential calculations, the temperature rise acceleration a is obtained. T ; S45: Calculate T max (t) sequence and C avg The Pearson correlation coefficient of the (t) sequence is used as the temperature-gas correlation coefficient ρ.

[0029] Because P is calculated each time core Coordinates can fluctuate slightly, and directly querying historical data for a single point can lead to discontinuities in the time series. This invention introduces a spatial neighborhood radius R, aggregating voxel data within a certain range (e.g., taking the average or maximum value) as the state value at that moment, thereby achieving stable tracking of a dynamically moving virtual target.

[0030] The spontaneous combustion rate of coal oxidation increases exponentially with increasing temperature, exhibiting a sustained and significant normal acceleration (α). T ≫0); while solar exposure may exhibit positive acceleration in the initial stage of temperature rise, it is limited by the upper limit of solar radiation power and surface convection heat dissipation, and the temperature rise acceleration is usually small and has obvious periodic convergence characteristics of solar radiation. Furthermore, simple solar exposure only leads to an increase in surface temperature and does not involve the production of oxidizing gases such as CO and CH4 (i.e., correlation coefficient ρ≈0); while spontaneous combustion involves the simultaneous release of heat and gas (ρ≈1). Therefore, by jointly judging a T The magnitude and value of ρ can effectively distinguish between external physical heat sources and internal chemical heat sources.

[0031] Preferably, in step S5, the formula for calculating the risk index RI is: RI=[w1·f(T max )+w2·v T +w3·a T ]×Φ(ρ); Among them, T max f(T) represents the highest temperature in the current equivalent spontaneous combustion core region. max ) is the normalization function; w1, w2, and w3 are preset weight coefficients, and w1 + w2 + w3 = 1; Φ(ρ) is the coupling correction operator: when the correlation coefficient ρ is greater than the preset positive correlation threshold, Φ(ρ) takes a value greater than 1; when ρ is less than the preset weak correlation threshold, Φ(ρ) takes a value less than 1.

[0032] The coupling correction operator Φ(ρ) is the core logic for eliminating false alarms in this invention. ρ is the correlation coefficient between the temperature series and the gas series. When ρ≈1 (strong positive correlation), it means that the temperature increase is accompanied by an increase in gas concentration, which is a typical characteristic of coal oxidation. In this case, Φ>1, and the system amplifies the risk index, making it easier to trigger an alarm. When ρ≈0 or negative (weak correlation), it means that only the temperature increases (possibly due to sunlight) or only the gas increases (possibly due to exhaust fumes from a loader). In this case, Φ<1, and the system suppresses the risk index, thereby filtering out false positives.

[0033] Preferably, step S5 further includes a remaining disposal time prediction step based on Taylor series expansion: When the risk index is determined to reach the warning level and the temperature rise acceleration is a T When the temperature is >0, the following formula can be used to calculate the coal temperature reaching the ignition point T. ignite Required remaining time t: T ignite =T current +v T ·t+0.5×a T ·t2 ; The smallest positive real root of the equation is used as the suggested time limit for handling the situation and is output along with the warning signal.

[0034] Traditional alarms only tell the operator "the temperature is too high," while this invention calculates v... T and a T By using second-order Taylor series expansion, the time it will take to reach the ignition point can be predicted. This provides power plant operators with a clear time window, allowing them to choose between simple reactor cooling or emergency water injection to extinguish the fire based on the urgency of the remaining time (e.g., whether there are 2 hours or 20 minutes left), which has extremely high practical guiding significance.

[0035] Preferably, step S5 also includes a three-dimensional visualization interactive display step, specifically: rendering a three-dimensional geometric model of the coal pile in a three-dimensional display interface; The calculated equivalent spontaneous combustion core coordinates P core The light spheres are superimposed on the interior of the three-dimensional geometric model, and the color of the light spheres changes with the risk index. P is displayed on the surface of the coal pile core The vertical projection area along the direction of gravity serves as the target guidance area for firefighting operations; The countdown value of the remaining processing time t and the temperature rise acceleration a are displayed in real time in the sidebar of the interface. T The trend curve.

[0036] The substantial effects of this invention are: 1. Achieved precise location of internal fire sources: Through the thermal-gas coupling positioning model, it breaks through the limitation of infrared thermal imagers that can only monitor the surface, and can calculate the equivalent spontaneous combustion core coordinates inside the coal pile, providing accurate three-dimensional guidance data for the fixed-point operation of automatic fire monitors.

[0037] 2. Significantly reduced false alarm rate: Spatial gradient analysis and correlation coefficient analysis were introduced, and the coupling relationship between physical fields (whether temperature and gas change synchronously, whether the temperature rise has acceleration characteristics) was used to filter out single-dimensional interference sources such as direct sunlight and equipment exhaust.

[0038] 3. Provides forward-looking early warning information: Compared with traditional threshold alarms, this invention predicts the remaining handling time based on time series analysis, realizing the transformation from post-event alarms to pre-event trend early warnings, thus winning valuable time for disaster relief. Attached Figure Description

[0039] Figure 1 This is a flowchart of a coal pile spontaneous combustion early warning method based on spatiotemporal gradient field analysis according to the present invention. Detailed Implementation

[0040] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0041] Example: This example is based on a large-span, fully enclosed dry coal shed environment measuring 340 meters long and 102 meters wide. The on-site hardware configuration includes: 10 sets of 3D laser scanning devices installed at the top of the walkway (for acquiring geometric data), 8 sets of dual-spectrum infrared thermal imagers (for acquiring temperature data), and 20 sets of composite gas detectors (monitoring CO, CH4, etc.) arranged on both sides every 30 meters along the walkway.

[0042] like Figure 1 As shown, a method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis is described, with the following specific steps: Step S1: Construct a three-dimensional multimodal data field for the coal yard Data Acquisition and Coordinate Unification: The system collects data from the aforementioned sensors in real time. Using a pre-calibrated extrinsic parameter matrix, all local coordinate system data are uniformly transformed into a global coordinate system with the northwest corner of the coal yard as the origin.

[0043] Octree modeling: An octree data structure with a depth of 12 levels is used to store spatial data. The minimum voxel side length is set to Δx=Δy=Δz=0.5 meters.

[0044] Temperature field mapping (anti-occlusion processing): When mapping 2D infrared image values ​​to 3D voxels, a depth buffer (Z-buffer) algorithm is used. Rays are emitted from the optical center of the infrared camera to each surface voxel of the 3D geometric model. If the ray intersects with other parts of the coal pile, the voxel is determined to be occluded and no temperature value is assigned, thereby eliminating the mapping errors caused by perspective occlusion.

[0045] Concentration field construction (anisotropic interpolation): For discrete gas sensor data, an anisotropic inverse distance weighted interpolation method is used to fill the air domain voxels.

[0046] Parameter settings: Considering the buoyancy effect of the hot gases generated by spontaneous combustion, vertical diffusion is stronger than horizontal diffusion. A distance weighting coefficient k is set for the vertical direction (Z-axis). z =0.3, the distance weighting coefficient k in the horizontal direction (X, Y axis) x =k y =1.0. That is, when calculating the distance D, D = sqrt(k x (x−x i ) 2 +k y (y−y i ) 2 +k z(z−z i ) 2 This means that, at the same geometric distance, the influence of a vertical sensor on the interpolation point is approximately 3.3 times that of a horizontal sensor, thus reflecting the upward floating characteristics of the gas.

[0047] Boundary constraints: If the interpolation path passes through a voxel marked as "coal pile entity", the path is considered blocked and the weight of the path is set to 0.

[0048] Step S2: Calculate the spatial gradient vector Data smoothing: To eliminate sensor noise, the voxel grid data is first smoothed by a 3×3×3 kernel Gaussian filter.

[0049] Gradient calculation: For mesh traversal, calculate the gradient vector G of physical quantity U using the central difference method: G = ((U i+1 -U i-1 ) / (2Δx),(U j+1 -U j-1 ) / (2Δy),(U k+1 -U k-1 ) / (2Δz)); For calculating the temperature gradient G on surface voxels T For calculating the concentration gradient G of air voxels C .

[0050] Step S3: Reverse tracing to locate the equivalent spontaneous combustion core Hot spot identification: Extracting surface temperature T>T env +10℃ (T env Voxels (based on the ambient reference temperature). Clustering was performed using the DBSCAN algorithm, with a neighborhood radius of ϵ = 1.5 meters and a minimum number of points MinPts = 5. The weighted centroid O after clustering was calculated. hot .

[0051] Leakage vent reverse engineering: The system scans gas concentration data in air voxels. The background warning threshold for CO is set at 24 ppm (this value references the definition of early-stage oxidation characteristics in general coal mine safety regulations and can be dynamically adjusted based on on-site background noise), and the abnormal threshold for CH4 is set at 0.5% (or other preset values ​​compliant with safety regulations). The identification logic is as follows: extract all abnormal voxels with CO concentrations exceeding 24 ppm or CH4 concentrations exceeding 0.5%. These abnormal voxels are considered as a "complex gas cloud," and the weighted geometric center O of this cloud is calculated. gas Introducing CH4 as a co-occurrence indicator is not only because it shares a common origin with CO, but also because the characteristic of "CO + CH4 rising simultaneously" helps to eliminate false positive interference caused solely by exhaust gases from fuel vehicles (which mainly contain CO and NOx, and usually do not contain high concentrations of CH4).gas P is obtained by projecting it along the −Z axis onto the surface of the coal pile. proj And by fine-tuning based on the horizontal concentration gradient around that point, the gas leak point O is obtained. leak .

[0052] Core positioning: Calculate P using the thermo-gas coupling positioning model formula. core =O hot -n×(C base +λ×d); Parameter setting: C base (Basic burial depth parameter): In this embodiment, it is preset to 0.8 meters. This parameter characterizes the empirical minimum depth of the internal heat source from the surface when the infrared thermal imager can detect a significant temperature rise anomaly under the current thermal conductivity characteristics of the coal type. It should be noted that this parameter is not fixed. In practical applications, a coal type-self-ignition stage comparison table can be established (e.g., referring to the coal pile self-ignition heating point depth comparison data), and the C value can be adjusted according to the coal type (e.g., lignite, bituminous coal) and the estimated self-ignition stage (latent period, self-heating period). base Dynamic calibration or correction can be performed to improve the accuracy of depth estimation.

[0053] λ (air permeability conductivity coefficient): This coefficient characterizes the ratio of upward migration to outward diffusion of spontaneously combustible gases in a coal pile medium. Its value is closely related to the anisotropy of the coal pile's compaction and permeability. For stratified and compacted coal piles, due to the presence of horizontal stratification, the horizontal permeability of the gas is often greater than the vertical permeability, and the gas is prone to lateral drift. In this case, λ is taken as a smaller value (recommended range 0.5~0.8, 0.6 in this embodiment). For loosely packed coal piles, dominated by hot air pressure (chimney effect), the gas mainly rises vertically, with less lateral diffusion. In this case, λ is taken as a larger value (recommended range 1.0~1.5). In actual deployment, this parameter can be calibrated by releasing tracer gas on-site, or configured during system initialization according to the coal yard stockpiling and reclaiming process (compaction / scattering).

[0054] d is O hot With O leak The Euclidean distance.

[0055] Using this formula, the system calculated a three-dimensional coordinate P inside the coal pile. core (x,y,z) is the equivalent self-ignition core.

[0056] Step S4: Extract spatiotemporal evolution features Spatial neighborhood index: based on the calculated P core With R as the center, construct a query sphere with a radius of R = 2.0 meters (i.e., 4 times the side length of the voxel).

[0057] Time series extraction: Retrieve all records that fall within the range of the sphere within the past 30 minutes from the historical database.

[0058] Feature calculation: The maximum temperature T in this area is calculated every minute. max and average concentration C avg .

[0059] The temperature rise rate v was obtained by fitting using the least squares method. T (Unit: ℃ / min).

[0060] For v T Differentiating the derivative yields the temperature rise acceleration a T (Unit: ℃ / min) 2 ).

[0061] Calculate the Pearson correlation coefficient ρ between the temperature series and the concentration series.

[0062] Step S5: Intelligent Judgment and Early Warning Risk index calculation: using the formula RI=[w1·f(T)] max )+w2·v T +w3·a T ]×Φ(ρ); Parameter settings: Normalization function f(T) max )=(T max -T env ) / 100.

[0063] Weighting coefficients: w1=0.3 (current temperature weight), w2=0.3 (velocity weight), w3=0.4 (acceleration weight). Acceleration is given the highest weight to emphasize the concern about the worsening trend of "self-acceleration".

[0064] Coupling correction operator Φ(ρ): When ρ>0.7 (strong positive correlation, consistent with spontaneous combustion characteristics), Φ=1.5 (amplifying risk); when ρ<0.3 (weak correlation, such as sun exposure causing only temperature rise without gas), Φ=0.5 (suppressing risk); otherwise, Φ=1.0.

[0065] Tiered early warning: Level 1 Warning (Suspected): RI > 0.5. The system only records the data, without triggering an alarm.

[0066] Level II Warning (Early Stage): RI > 0.8 and a T >0.05. System locked P core The coordinates indicate "latent spontaneous combustion," suggesting a re-examination of the site.

[0067] Level 3 Alarm (Danger): T max >Tcritical Or C gas >C critical Among them, T critical This is the critical alarm temperature; this value is not a fixed number, but rather depends on the ambient temperature T. env Dynamic setting of coal type self-ignition characteristic curve, setting logic satisfies T critical =min(T absolute ,T env +ΔT rise ). T absolute The absolute auto-ignition critical temperature for a coal type (e.g., refer to the coal auto-ignition characteristics table; 70°C for lignite and 80-100°C for bituminous coal); ΔT rise For the maximum permissible relative temperature rise (e.g., set to 40-60 degrees Celsius); C gas This refers to the concentration of hazardous gases (CO or CH4); C critical The gas concentration is at a dangerous threshold. The system triggers an audible and visual alarm and automatically activates the fire monitor to spray water onto the surface area directly above the calculated equivalent spontaneous combustion core.

[0068] Remaining time prediction: When a level II or higher warning is triggered, the coal ignition point T will be used. ignite =250℃, substituting into the Taylor series formula: 250=T current +v T ·t+0.5×a T ·t 2 The solution is used to obtain t, and an alarm is issued, such as displaying "It is expected to reach the open flame ignition point in 125 minutes", to assist managers in decision-making.

[0069] The system also presents the above calculation results to the user in a 3D visualization format. Specifically, in the 3D digital twin scene of the coal yard on the monitoring screen, the system uses the calculated coordinates P... core A dynamically pulsating sphere of light is rendered inside a semi-transparent coal pile model. The color of the sphere visually reflects the risk level (e.g., yellow represents the incubation period, and red represents the active period).

[0070] At the same time, the system automatically calculates and displays the vertical projection area of ​​the light sphere on the upper surface of the coal pile (e.g., circled with a red dotted line), clearly indicating the spray target area of ​​the fire cannon.

[0071] In addition, the floating window on the right side of the interface displays a real-time countdown (e.g., "Estimated time to ignition: 125 minutes") and a temperature rise acceleration curve, helping managers to intuitively judge the trend of fire deterioration.

[0072] Through the above embodiments, the present invention can effectively distinguish the surface temperature rise caused by direct sunlight (typically ρ≈0, a) T Smaller) and true internal oxidation spontaneous combustion (ρ≈1,a)T >0), and accurately located the internal core depth in the early stage of spontaneous combustion, realizing a technological leap from post-event alarm to pre-event accurate early warning.

[0073] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0074] Although this document uses a variety of terms, the possibility of using other terms is not excluded. These terms are used merely for the convenience of describing and explaining the essence of the invention; interpreting them as any additional limitation would contradict the spirit of the invention.

Claims

1. A method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis, characterized in that, The method includes the following steps: S1: Construct a three-dimensional multimodal data field for the coal yard: acquire real-time three-dimensional geometric point cloud data of the coal yard space, two-dimensional infrared temperature data of the coal pile surface, and multi-point gas concentration data of the environment; map the data to a unified three-dimensional coordinate system of the coal yard, and construct a three-dimensional multimodal voxel mesh model containing geometric attributes, temperature attributes, and gas concentration attributes. S2: Calculate the spatial gradient vector: Traverse the voxel mesh model and use the discrete difference algorithm to calculate the temperature gradient vector of the voxels on the coal pile surface and the gas concentration gradient vector of the voxels in the air domain, respectively. S3: Reverse Source Tracing to Locate the Equivalent Spontaneous Combustion Core: Based on the temperature gradient vector, identify the centroid of the hot spot region on the coal pile surface; based on the gas concentration gradient vector and the high-concentration region, inversely deduce the coordinates of the gas leak outlet; according to the spatial relationship between the centroid of the hot spot region and the coordinates of the gas leak outlet, use the heat-gas coupling location model to calculate the coordinates P of the equivalent spontaneous combustion core inside the coal pile. core ; S4: Extract spatiotemporal evolution features: using the equivalent spontaneous combustion core coordinates P core A spatial neighborhood query range is constructed around the center, and temperature and gas concentration sequences within a preset time window in the past are retrieved from the historical time series database; the temperature rise rate v is calculated based on the sequences. T Temperature rise acceleration a T And the temperature-gas correlation coefficient ρ; S5: Intelligent Judgment and Early Warning: The P... core v T a T The risk index is calculated using the integrated risk assessment model with input ρ, and a graded early warning signal is triggered based on the risk index.

2. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, In step S1, the construction of the three-dimensional multimodal voxel mesh model specifically includes: S11: Uses an octree data structure to store coal yard spatial data; S12: Using a depth buffer detection algorithm, rays are emitted from the optical center of the infrared thermal imager to the three-dimensional geometric point cloud to remove occluded voxel nodes and map the infrared temperature data only to the visible voxels on the coal pile surface. S13: Using anisotropic inverse distance weighted interpolation, the discrete gas concentration data is interpolated into the air domain voxels; wherein, when calculating the interpolation weights, the distance weight coefficient of the vertical Z-axis is less than the distance weight coefficients of the horizontal X and Y axes; and when the interpolation path is blocked by the coal pile entity voxels, the diffusion distance is set to infinity.

3. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, In step S2, the spatial gradient vector is specifically calculated using the central difference method. For any non-empty node (i,j,k) in the mesh, the gradient vector G of its physical quantity U is calculated as follows: G = ((U i+1 -U i-1 ) / (2Δx),(U j+1 -U j-1 ) / (2Δy),(U k+1 -U k-1 ) / (2Δz)); Where U represents temperature T or gas concentration C, and Δx, Δy and Δz are the side lengths of the voxel grid, respectively; before calculation, the voxel grid data is first smoothed by Gaussian filtering.

4. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, In step S3, identifying the centroid of the hot spot region on the coal pile surface specifically includes: S311: Extraction of surface voxel sets with an extraction temperature T greater than the ambient reference temperature; S312: The set is divided into several independent hot spot regions using the DBSCAN clustering algorithm; S313: Calculate the weighted centroid O for each hot spot region. hot The weight is the temperature value of the voxel.

5. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 4, characterized in that, In step S3, the reverse deduction of the gas leak coordinates specifically includes: S321: Identify high-concentration air masses in the air domain voxels whose concentration values ​​exceed the background threshold, and calculate their geometric center O. gas ; S322: Along the direction of gravity, i.e., the -Z axis direction, O gas Projecting the image onto the surface of the coal pile yields projection point P. proj ; S323: Get P proj The gas concentration gradient vector at point P is used to take its horizontal component. proj By translating along the direction of this horizontal component, the corrected coordinates of the gas leak point O are obtained. leak .

6. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 5, characterized in that, In step S3, the equivalent spontaneous combustion core coordinates P inside the coal pile are calculated using the thermal-gas coupling positioning model. core The formula is: P core =O hot -n×(C base +λ×d); Among them, O hot Let n be the coordinates of the centroid of the hot spot region; n is the coordinate of O. hot The surface normal vector at point O points in the coal pile; d is the surface normal vector at point O. hot With O leak The Euclidean distance between them; C base λ represents the preset foundation burial depth parameter; λ is the air permeability conductivity coefficient.

7. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, In step S4, the extraction of spatiotemporal evolution features specifically includes: S41: Based on the calculated P core With the center of the sphere as the center, construct a query sphere with a preset spatial radius R as the radius; S42: Retrieve the historical time-series database to obtain the maximum temperature T of all voxels falling within the sphere at each historical moment. max (t) and average concentration C avg (t); S43: To T max (t) The time series is linearly fitted using the least squares method, and the slope of the fitted line is the rate of temperature rise v. T ; S44: For adjacent time steps v T By performing differential calculations, the temperature rise acceleration a is obtained. T ; S45: Calculate T max (t) sequence and C avg The Pearson correlation coefficient of the (t) sequence is used as the temperature-gas correlation coefficient ρ.

8. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, In step S5, the formula for calculating the risk index RI is: RI=[w1·f(T max )+w2·v T +w3·a T ]×Φ(ρ); Among them, T max f(T) represents the highest temperature in the current equivalent spontaneous combustion core region. max ) is the normalization function; w1, w2, and w3 are preset weight coefficients, and w1 + w2 + w3 = 1; Φ(ρ) is the coupling correction operator: when the correlation coefficient ρ is greater than the preset positive correlation threshold, Φ(ρ) takes a value greater than 1; when ρ is less than the preset weak correlation threshold, Φ(ρ) takes a value less than 1.

9. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, Step S5 also includes a remaining disposal time prediction step based on Taylor series expansion: When the risk index is determined to reach the warning level and the temperature rise acceleration is a T When the temperature is >0, the following formula can be used to calculate the coal temperature reaching the ignition point T. ignite Required remaining time t: T ignite =T current +v T ·t+0.5×a T ·t 2 ; The smallest positive real root of the equation is used as the suggested time limit for handling the situation and is output along with the warning signal.

10. The method for early warning of spontaneous combustion of coal piles based on spatiotemporal gradient field analysis according to claim 1, characterized in that, Step S5 also includes a 3D visualization and interactive display step, specifically: Render the three-dimensional geometric model of the coal pile in the three-dimensional display interface; The calculated equivalent spontaneous combustion core coordinates P core The light spheres are superimposed on the interior of the three-dimensional geometric model, and the color of the light spheres changes with the risk index. P is displayed on the surface of the coal pile core The vertical projection area along the direction of gravity serves as the target guidance area for firefighting operations; The countdown value of the remaining processing time t and the temperature rise acceleration a are displayed in real time in the sidebar of the interface. T The trend curve.

Images