Coal spontaneous combustion temperature prediction method based on graph convolutional neural network

Abstract

The present application provides a kind of coal spontaneous combustion temperature prediction method based on graph convolutional neural network, it is related to coal spontaneous combustion temperature prediction technical field, the present application is established three-dimensional coordinate system by the region where the coal to be predicted, divides solid grid, and judges whether grid is exposed area, the temperature distribution function of internal temperature change with depth is calculated by oxygen diffusion model and heat model, the heat characteristics of the grid are obtained according to temperature distribution function, the heat pointing value is calculated by the heat characteristics of surrounding grid, the heat accumulation center point and heat accumulation area are judged, the average change rate of dangerous gas in roadway area is judged as the dangerous gas concentration change characteristics of all heat accumulation center points in the roadway, the heat accumulation center point is taken as node, the coal spontaneous combustion temperature prediction model is constructed by graph convolutional neural network, the real-time maximum temperature and confidence of coal in each grid are output, and the grid length is optimized.

Description

Technical Field

[0001] This invention relates to the field of coal spontaneous combustion temperature prediction technology, specifically a method for predicting coal spontaneous combustion temperature based on graph convolutional neural networks. Background Technology

[0002] Spontaneous combustion of coal is a major safety hazard in coal mine production, which can easily lead to accidents such as fires and gas explosions. Traditional monitoring methods mostly rely on point measurements with limited sensors or empirical models based on a single type of physical parameter, which are difficult to fully reflect the complex heat accumulation and gas diffusion process inside the coal.

[0003] In existing methods for predicting coal spontaneous combustion, convolutional neural network models based on gas concentration monitoring have been widely used. However, these models often rely solely on local gas concentration data for spatial feature extraction, failing to fully consider the dynamic impact of mine ventilation systems on gas transport and diffusion. Ventilation airflow significantly alters the distribution and accumulation of hazardous gases and oxygen. Traditional methods do not incorporate ventilation network structure and airflow direction as key factors into the modeling process, making it difficult to accurately reflect the true changes in gas concentration within the roadway. Furthermore, due to the lack of systematic coupling of multi-source heterogeneous data (such as coal physical properties, oxygen diffusion, ventilation airflow, and heat conduction) and the impact of dynamic changes in roadway gases on the spontaneous combustion process, especially the lack of effective modeling of the spatiotemporal correlation of multiple factors, existing models cannot accurately capture the synergistic effect of heat accumulation trends and hazardous gas concentration changes. Consequently, these models exhibit lag and low robustness in predicting coal spontaneous combustion temperatures, making it difficult to achieve effective early warning of potential spontaneous combustion areas.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the spontaneous combustion temperature of coal based on graph convolutional neural networks, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for predicting the spontaneous combustion temperature of coal based on graph convolutional neural networks, comprising the following steps:

[0008] Step 1: Establish a three-dimensional coordinate system for the area where the coal is to be predicted, obtain the coordinate data of the coal in the coordinate system, and divide the area where the coal is located into a three-dimensional grid with the same length interval. Determine whether the grid is an exposed area by the exposed area of ​​the coal in each grid.

[0009] Step 2: For each grid identified as an exposed area, obtain the parameters of the coal and the oxygen concentration within the grid to construct an oxygen diffusion model, obtain the wind speed, air temperature and coal surface temperature on the coal surface, and construct a coal heat model based on the coal parameters. Calculate the temperature distribution function of the coal's internal temperature as a function of depth based on the oxygen diffusion model and the heat model.

[0010] Step 3: Calculate the cumulative heat, surface temperature difference and average temperature at the center of the coal exposed area in each grid by obtaining the coal surface temperature and temperature distribution function at the center of the grid. Use these as the heat characteristics at the center of the grid. Calculate the heat direction value by using the heat characteristics at the center of the grid and the surrounding grids to determine the heat accumulation direction. Determine the heat accumulation center point and heat accumulation area based on the heat accumulation direction.

[0011] Step 4: Obtain the concentration of hazardous gases at the intersection of coal mine roadways and the dimensions of each roadway, and determine the average rate of change of hazardous gases in the roadway area as the characteristic of hazardous gas concentration change at all heat accumulation centers in the roadway.

[0012] Step 5: Using the heat accumulation center point as the node and the distance between nodes as the edge, the model input data is formed by constructing node features and edge features. A coal spontaneous combustion temperature prediction model is constructed through a graph convolutional neural network, outputting the real-time maximum temperature and confidence level of coal in each grid, and optimizing the grid length.

[0013] Furthermore, the method for determining whether a grid is an exposed area by the exposed coal area in each grid is to set an exposed area threshold. When the exposed coal area in a grid is greater than the threshold, the grid is determined to be an exposed area.

[0014] Furthermore, the parameters of the coal include coal type, coal pore size, coal depth, oxidation reaction heat release rate, thermal conductivity, coal density, and specific heat capacity;

[0015] The process of constructing the oxygen diffusion model is as follows:

[0016]

[0017] in, In order to be in Coal Depth at All Times Oxygen concentration at that location. The oxygen diffusion coefficient depends on the type of coal. For wind speed, Let be the oxidation efficiency constant of coal. These are depth and time variables, respectively.

[0018] Furthermore, the process of constructing the coal calorific value model is as follows:

[0019]

[0020] in, The exothermic rate of the oxidation reaction, This is the enthalpy of the oxidation reaction.

[0021] Furthermore, the specific process of the temperature distribution function of the coal's internal temperature changing with depth is as follows:

[0022]

[0023] in, In order to be in Coal Depth at All Times The surface temperature of the coal at that location The thermal conductivity of coal, For the density of coal, The specific heat capacity of coal, This refers to the surface temperature of the coal.

[0024] Furthermore, the method for calculating the cumulative heat, surface temperature difference, and average temperature at the center point by obtaining the coal surface temperature and temperature distribution function at the center of the exposed coal area in each grid is as follows:

[0025] The temperature analysis depth is set according to the temperature distribution function, and the cumulative heat at the center point is obtained by integrating the depth.

[0026] The average temperature is calculated based on the ratio of accumulated heat to the depth of temperature analysis.

[0027] The surface temperature difference includes the maximum surface temperature difference and the minimum surface temperature difference. Based on the temperature distribution function and the temperature analysis depth, the maximum and minimum temperatures distributed inside the center point are calculated. By subtracting these from the surface temperature, the maximum and minimum surface temperature differences are obtained.

[0028] Furthermore, the method for calculating the heat direction value by comparing the heat characteristics of the grid with those of the surrounding grid centers, determining the heat accumulation direction, and determining the heat accumulation center point and heat accumulation area based on the heat accumulation direction is as follows: The heat characteristics of the grid are subtracted from those of the grids in eight directions around the exposed surface (front, back, left, right, left front, right front, left back, and right back) to obtain the heat characteristic difference. Then, a weight is assigned to each heat characteristic difference to calculate the heat direction value. The grid with the smallest heat direction value among the grids is selected as the heat accumulation direction. The grid whose heat accumulation direction points to the endpoint is selected as the heat accumulation center point. The grid with the smallest heat characteristic value among adjacent heat accumulation center points is selected as the area boundary point. The heat accumulation center point at the boundary of the coal area is selected with the coal boundary as the boundary of the heat accumulation area. The heat accumulation area is formed by connecting the area boundary points around the heat accumulation center point with the boundary of the heat accumulation area.

[0029] Furthermore, the types of hazardous gases include CO, CO2, CH4, C2H6, and C2H4;

[0030] The dimensions of a tunnel include its length, width, and height;

[0031] The method for determining gas variables in a roadway area based on the concentration of hazardous gases, air velocity, and the dimensions of each roadway is as follows:

[0032] The volume of the roadway is calculated based on its dimensions. The sampling interval is set with the same time length. The concentration of hazardous gas is collected based on the concentration of hazardous gas at adjacent time intervals. The concentration increment of hazardous gas in the roadway is calculated based on the concentration of hazardous gas at both ends of the roadway. The concentration change of hazardous gas is calculated based on the duration of the sampling interval and the concentration increment of hazardous gas. The average rate of change of hazardous gas in the roadway is calculated based on the volume of the roadway and the concentration change of hazardous gas.

[0033] Furthermore, the steps of constructing a coal spontaneous combustion temperature prediction model using a graph convolutional neural network include: model construction, model training, and data output;

[0034] The model building process is as follows:

[0035] The node features are the grid center coordinates of the heat accumulation area, the characteristics of hazardous gas concentration changes, the heat accumulation area, the heat characteristics of each grid within the heat accumulation area, and the coal type.

[0036] The ventilation direction and the distance between nodes are used as edge features;

[0037] Graph convolutional neural networks include an input and adjacency matrix layer, a graph convolutional layer, an edge feature fusion layer, a global graph pooling layer, and a fully connected regression layer.

[0038] The model training process is as follows:

[0039] Based on historical input data, sampling is performed at the same time intervals. The input data of the previous moment is used as the input of the model, and the real-time highest temperature of coal in the grid at the next moment is used as the output of the model to train the coal spontaneous combustion temperature prediction model.

[0040] The data output process is as follows:

[0041] The real-time input data is used as input to the coal spontaneous combustion temperature prediction model to predict the real-time maximum temperature of coal in the grid at the next moment.

[0042] Furthermore, the method for optimizing the grid length is as follows:

[0043] The objective function is set by weighted summation based on the confidence grid number of the real-time temperature output by the trained model. The grid length is then optimized using the sparrow optimization algorithm, which includes discoverers and followers.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention establishes a three-dimensional coordinate system and divides a three-dimensional grid in the area where the coal to be predicted is located, and determines whether the grid is an exposed area. It calculates the temperature distribution function of the internal temperature of the coal with depth through an oxygen diffusion model and a heat model. Based on the temperature distribution function, it calculates the cumulative heat, surface temperature difference and average temperature of the center point as the heat characteristics of the grid. It calculates the heat direction value through the heat characteristics of the surrounding grids to determine the heat accumulation center point and heat accumulation area. It judges the average change rate of hazardous gases in the roadway area as the hazardous gas concentration change characteristics of all heat accumulation center points in the roadway. With the heat accumulation center point as the node, it constructs a coal spontaneous combustion temperature prediction model through a graph convolutional neural network, outputs the real-time maximum temperature and confidence level of the coal in each grid, and optimizes the grid length.

[0045] This invention achieves high-precision prediction and dynamic optimization of coal spontaneous combustion temperature by combining three-dimensional mesh modeling, multi-physics coupled calculation, and graph convolutional neural networks. First, through three-dimensional mesh generation and exposed area identification, the heat accumulation center and diffusion path are accurately located, overcoming the shortcomings of traditional methods in representing spatial structures. Second, by integrating oxygen diffusion models, heat transfer models, and roadway gas change characteristics, node and edge features reflecting the multi-factor coupling mechanism of coal spontaneous combustion are constructed, improving the physical interpretability of the model input.

[0046] By introducing the dynamic change characteristics of gas concentration under the action of the ventilation system and constructing node and edge features based on the graph structure, the accuracy and reliability of coal spontaneous combustion temperature prediction are significantly improved. By constructing edge features using the ventilation direction, wind speed and gas concentration change rate of the roadway, the model's ability to represent the gas transport process is enhanced.

[0047] Finally, a graph convolutional neural network is used to capture the complex nonlinear relationships between nodes, thereby achieving collaborative prediction of heat accumulation areas and gas hazard changes. By optimizing the grid granularity, computational efficiency and prediction accuracy are further balanced. This method significantly improves the real-time performance and confidence of temperature prediction, provides a scientific basis for early warning of spontaneous combustion in coal mines, and effectively supports safety management decisions. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0049] Figure 2 This is a schematic diagram illustrating the determination of the heat accumulation center point and heat accumulation area according to the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0051] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0052] Example:

[0053] Please see Figure 1-2 The present invention provides a technical solution:

[0054] A method for predicting the spontaneous combustion temperature of coal based on graph convolutional neural networks, comprising the following steps:

[0055] Step 1: Establish a three-dimensional coordinate system for the area where the coal is to be predicted, obtain the coordinate data of the coal in the coordinate system, and divide the area where the coal is located into a three-dimensional grid with the same length interval. Determine whether the grid is an exposed area by the exposed area of ​​the coal in each grid.

[0056] The coal-bearing area refers to the region where coal seams are distributed underground, as well as areas where broken coal bodies (such as remnants of coal in goafs) or coal surfaces exposed to air (such as tunnel walls) are deposited. These areas possess the material and necessary conditions for spontaneous combustion of coal. In this embodiment, the method for determining whether a grid is an exposed area based on the exposed coal area in each grid is as follows: a threshold for exposed area is set; when the exposed coal area within a grid exceeds this threshold, the grid is determined to be an exposed area.

[0057] This step achieves refined digital reconstruction of the coal spatial structure by establishing a three-dimensional coordinate system and dividing it into three-dimensional meshes. Its core advantage lies in discretizing continuous coal regions into computable standardized units and effectively identifying key areas of coal-oxygen contact by determining the exposed area threshold. This spatial discretization method provides a unified spatial benchmark for subsequent physical field modeling and graph structure construction, fundamentally solving the prediction bias problem caused by spatial ambiguity in traditional methods.

[0058] This step transforms the complex three-dimensional spatial features of coal into a quantifiable topological structure through gridding, providing a node foundation containing spatial relationships for graph neural networks. Then, by identifying exposed areas, it accurately locates potential areas where oxidation reactions occur, significantly improving the computational efficiency and accuracy of subsequent oxygen diffusion and heat models, and laying a reliable spatial data foundation for the entire prediction system.

[0059] The exposed coal area refers to the area of ​​coal in contact with air. Using this area as a criterion is crucial because spontaneous combustion of coal requires the fundamental condition of coal-oxygen contact, and the exposed area directly determines the effective range of the oxidation reaction. This method, by setting scientific thresholds, can accurately identify key areas with genuine spontaneous combustion risk, effectively avoiding the inclusion of areas completely covered by rock strata or invalid spaces in the calculation model, significantly improving the accuracy of subsequent oxidation reaction calculations. In practice, coal surface elevation data can be obtained through 3D laser scanning or photogrammetry, and then the actual exposed area within each grid cell can be calculated using grid projection and surface integration. This quantitative judgment method ensures both the accuracy of spatial analysis and good engineering feasibility.

[0060] Step 2: For each grid identified as an exposed area, obtain the parameters of the coal and the oxygen concentration within the grid to construct an oxygen diffusion model, obtain the wind speed, air temperature and coal surface temperature at the coal surface, and construct a coal calorific model based on the coal parameters. Calculate the temperature distribution function of the coal's internal temperature as a function of depth based on the oxygen diffusion model and the calorific model.

[0061] In this embodiment, the parameters of the coal include coal type, coal pore size, coal depth, oxidation reaction exothermic rate, thermal conductivity, coal density, and specific heat capacity.

[0062] The process of constructing the oxygen diffusion model is as follows:

[0063]

[0064] in, In order to be in Coal Depth at All Times Oxygen concentration at that location. The oxygen diffusion coefficient depends on the type of coal. For wind speed, Let be the oxidation efficiency constant of coal. These are depth and time variables, respectively.

[0065] Partial differential equations describe oxygen concentration At coal depth and time The rate of change in oxygen, which consists of three key terms, determines the dynamic distribution of oxygen:

[0066] diffusion term This describes the spontaneous diffusion of oxygen molecules from a region of high concentration to a region of low concentration; the diffusion rate is determined by the diffusion coefficient. Control, and The size of the pores depends primarily on the type of coal and the size of its pores (porosity, permeability). Larger pores generally indicate better connectivity. The higher the value, the easier it is for oxygen to penetrate into the coal body. This enables quantitative modeling of the influence of coal's physical properties on its oxygen transport capacity, explaining why different coal types exhibit different regional spontaneous combustion tendencies.

[0067] Convection term Described due to wind speed The presence of airflow causes overall oxygen migration (convection). Airflow carries oxygen, forcibly pushing it into or extracting it from the pores of the coal seam, significantly accelerating the oxygen transport process, with an impact far greater than simple molecular diffusion. This is achieved by introducing the strong influence of external environmental factors (ventilation) on oxygen supply. This allows the model to respond to changes in the underground ventilation system and predict how airflow alters oxygen distribution in hazardous areas—a crucial point often overlooked by traditional methods.

[0068] Consumption items Based on the principles of chemical reaction kinetics, the rate at which oxygen is consumed in the oxidation reaction between coal and oxygen is described, and this rate of consumption is related to the current oxygen concentration. It is directly proportional, and the proportionality constant is the oxidation efficiency constant. This comprehensively reflects the chemical activity of coal; different types of coal have different... The values ​​also differ, quantifying the contribution of coal's chemical properties to oxygen consumption. It transforms the abstract concept of chemical reactions into a calculable mathematical term, explaining why some coals are more prone to spontaneous combustion under the same oxygen supply.

[0069] This model dynamically simulates the entire process of how oxygen enters the coal seam from the roadway space and is continuously consumed during its movement by coupling diffusion, convection and reaction. The output is an accurate oxygen concentration distribution that varies with time and depth.

[0070] In this embodiment, the process of constructing the coal calorific value model is as follows:

[0071]

[0072] in, The exothermic rate of the oxidation reaction, This is the enthalpy of the oxidation reaction.

[0073] The exothermic rate of the oxidation reaction depends on the enthalpy Δφ of the coal oxidation reaction, which represents the heat released when a unit mass of coal undergoes an oxidation reaction. During coal spontaneous combustion, coal reacts with oxygen to release heat. Using this formula, the exothermic rate of the coal oxidation reaction at different depths and times can be dynamically calculated using oxygen concentration. The oxidation efficiency constant reflects the rate of coal oxidation. Different types of coal have different oxidation characteristics; some coals may oxidize more readily and release more heat, while others may oxidize more slowly and release less heat.

[0074] In this embodiment, the specific process of the temperature distribution function of the coal's internal temperature changing with depth is as follows:

[0075]

[0076] in, In order to be in Coal Depth at All Times The surface temperature of the coal at that location The thermal conductivity of coal, For the density of coal, The specific heat capacity of coal, This refers to the surface temperature of the coal.

[0077] This partial differential equation describes the coal temperature. The rate of change with depth and time, where the heat conduction term This is the second derivative of the internal temperature of coal along the depth direction γ, representing the spatial distribution of coal temperature. This term describes the diffusion or propagation of heat within the coal. Specifically, heat propagates from high-temperature regions to low-temperature regions; the greater the temperature gradient, the faster the heat conduction rate. The thermal conductivity γ determines the efficiency of heat conduction. The higher the thermal conductivity of coal, the faster the heat propagates, and the temperature differences within the coal will quickly tend towards equilibrium. This refers to the temperature of the coal surface, which is closely related to the external environment and the heat release during the spontaneous combustion process. As a boundary condition, This affects the internal temperature changes of coal. For example, if the surface temperature is high, heat will be conducted from the surface inward, altering the temperature distribution of the coal. This equation allows for the dynamic simulation of temperature changes in coal at different depths and time points. This is of great significance for assessing the risk of spontaneous combustion in coal mines, exploring for coal gas, and designing mine ventilation systems. By calculating temperature changes at different depths of coal in real time, the thermal response of coal can be accurately predicted.

[0078] The core advantage of employing a multiphysics coupled modeling method lies in establishing an accurate mathematical model of oxygen diffusion and heat transfer, thereby enabling refined numerical simulation of the oxidation reaction and heat conduction processes within coal. By introducing an oxygen diffusion model and a heat model based on Fourier's law of heat conduction, this method accurately characterizes the oxygen concentration distribution and temperature variation patterns at different depths, effectively overcoming the limitations of traditional methods that rely solely on surface monitoring data and cannot reflect internal reaction processes.

[0079] The temperature distribution function calculated using the physical model provides the neural network with node features of clear physical meaning, significantly enhancing the reliability of feature engineering. It transforms the complex chemical reaction process of coal spontaneous combustion into a quantifiable mathematical expression, providing accurate supervisory signals for neural network learning. By establishing a temperature distribution in the depth dimension, it provides a complete data foundation for subsequent thermal feature calculations, enabling the neural network to simultaneously acquire information on surface and internal temperature changes, greatly improving the model's ability to identify potential spontaneous combustion zones. This deep integration of the physical model and the neural network maintains the accuracy of the physical process while leveraging the advantages of neural networks in handling complex nonlinear relationships, ultimately forming a novel paradigm for predicting coal spontaneous combustion temperature that combines mechanistic reliability with predictive accuracy.

[0080] Step 3: Calculate the cumulative heat, surface temperature difference, and average temperature at the center of the exposed coal area in each grid by obtaining the coal surface temperature and temperature distribution function. This information serves as the heat characteristic of the center of the grid. Calculate the heat direction value using the heat characteristics of the grid and the centers of surrounding grids to determine the direction of heat accumulation. Based on the direction of heat accumulation, determine the heat accumulation center point and the heat accumulation area.

[0081] This step transforms the physical data (temperature distribution function) obtained in the previous step into dynamic features that can characterize the spatial evolution trend of coal spontaneous combustion, and intelligently identifies potential dangerous core areas based on these features. By introducing concepts such as "heat orientation value" and "heat accumulation direction", this step simulates the process of heat conduction and accumulation in space, achieving a key leap from static physical quantity calculation to dynamic risk identification.

[0082] By extracting multi-dimensional thermal characteristics such as cumulative heat, average temperature, and surface temperature difference through deep integration, the method can comprehensively reflect the thermal stability and energy accumulation degree within the grid cell, which is more predictive than a single surface temperature value. Then, by calculating and weighting the thermal characteristic differences with the centers of the eight neighboring grids (i.e., the center of the grid space), the method can accurately determine the macroscopic flow trend of heat, thus pointing out the final destination of heat convergence (the heat accumulation center point) like a "compass," and delineating continuous heat accumulation areas accordingly.

[0083] This step plays a crucial role in structuring and focusing the prediction methods of graph convolutional neural networks. It transforms uniform grid data into non-uniform graph structure data, providing a naturally suitable input topology for subsequent graph convolutional neural networks, greatly improving the network's processing efficiency and the specificity of feature extraction. Secondly, by identifying heat accumulation areas, it helps the network focus its learning on the key areas most likely to spontaneously combust, filtering out a large amount of irrelevant background grid noise. This is equivalent to providing the network with an "attention mechanism," thereby significantly improving the model's prediction accuracy, convergence speed, and the reliability of the final warning for dangerous areas.

[0084] The center of the exposed coal area refers to a representative center point selected for simplified calculations within a three-dimensional grid cell that is determined to be exposed. This point is not necessarily the geometric center, but rather refers to the core location that represents the reaction between the coal surface and oxygen and the exchange of heat within that grid.

[0085] In this embodiment, the method for calculating the cumulative heat, surface temperature difference, and average temperature at the center point by obtaining the coal surface temperature and temperature distribution function at the center of the exposed coal area in each grid is as follows:

[0086] The temperature analysis depth is set according to the temperature distribution function, and the cumulative heat at the center point is obtained by integrating the depth.

[0087] The average temperature is calculated based on the ratio of accumulated heat to the depth of temperature analysis.

[0088] The surface temperature difference includes the maximum surface temperature difference and the minimum surface temperature difference. Based on the temperature distribution function and the temperature analysis depth, the maximum and minimum temperatures distributed inside the center point are calculated. By subtracting these from the surface temperature, the maximum and minimum surface temperature differences are obtained.

[0089] Cumulative heat reflects heat capacity and energy accumulation. While surface temperature may fluctuate temporarily due to ventilation, cumulative heat reflects the overall energy level within the coal seam. An area with high cumulative heat, even if its surface temperature is temporarily low, signifies that it has accumulated enormous energy internally, possessing the "capital" to undergo violent oxidation reactions and rapid temperature increases—a potential "energy reservoir." Compared to the volatile surface temperature, cumulative heat is an integral quantity, insensitive to short-term surface fluctuations, and can more stably indicate the long-term thermal hazard level of a region.

[0090] Average temperature provides a single, comprehensive indicator to summarize the overall thermal state from the surface to a certain depth, avoiding potential misjudgments that may arise from relying solely on surface temperature (such as situations where the surface dissipates heat quickly but the interior is hot). By comparing average temperatures across different grids, it is possible to quickly assess which region has a more dangerous overall thermal environment, providing an intuitive basis for risk ranking.

[0091] Surface temperature difference reveals internal thermal activity and conductivity. A large and positive maximum surface temperature difference indicates the existence of a "thermal core" with a temperature much higher than the surface. This means that the oxidation reaction inside the coal is very intense, and at the same time, the coal has poor thermal conductivity, so heat cannot be easily dissipated to the surface, resulting in a "smoldering" state. This is a very strong warning signal that spontaneous combustion is about to occur. A small or negative surface temperature difference indicates that heat conduction from the inside to the surface is good, there is no abnormal heat accumulation inside, or the surface is hotter due to oxidation or the external environment.

[0092] The magnitude of the temperature difference directly reflects the balance between the rate of heat release from the internal reaction and the rate of heat dissipation from the external environment. A continuously increasing temperature difference indicates that heat release from the reaction is dominating, and the risk is rising sharply.

[0093] In this embodiment, the method for calculating the heat direction value by comparing the heat characteristics of the grid with those of the surrounding grid centers, determining the heat accumulation direction, and determining the heat accumulation center point and heat accumulation area based on the heat accumulation direction is as follows: The heat characteristics of the grid are subtracted from those of the grids in eight directions around the exposed surface (front, back, left, right, left front, right front, left back, and right back) to obtain the heat characteristic difference. Then, a weight is assigned to each heat characteristic difference to calculate the heat direction value. The grid with the smallest heat direction value is selected. The direction of the line connecting the center points of two grids is taken as the heat accumulation direction. The grid whose heat accumulation direction points to the endpoint is taken as the heat accumulation center point. The grid with the smallest heat characteristic value between adjacent heat accumulation center points is taken as the area boundary point. The heat accumulation center point at the boundary of the coal area is taken as the boundary of the heat accumulation area, with the coal boundary as the boundary of the heat accumulation area. The heat accumulation area is formed by connecting the area boundary points around the heat accumulation center point with the boundary of the heat accumulation area.

[0094] By calculating heat orientation values ​​and determining the direction and region of heat accumulation, its core advantage lies in achieving a transformation from discrete grid temperature data to continuous spatial thermodynamic trend analysis, providing dynamic and forward-looking risk insights for coal spontaneous combustion prediction. This method simulates the natural laws of heat transfer and accumulation in complex coal environments, enabling precise location of the true core danger, rather than merely identifying current temperature high points.

[0095] By introducing the vector concept of heat direction value, the intensity and direction of heat transfer between grids are effectively quantified, revealing hidden heat flow paths and confluence centers. This provides early warning of "potential hotspots" that may not currently be at their highest temperature but are in a phase of rapid heat accumulation. Then, by judging the direction of heat accumulation, the endpoint of the heat flow (the heat accumulation center) is automatically tracked. This avoids the overemphasis on isolated high-temperature points that may result from judging solely by temperature. Instead, it intelligently identifies dangerous cores that are thermodynamically stable and continuously attract surrounding heat. By delineating heat accumulation areas, the influence range of each dangerous center is clarified, providing a precise spatial basis for the formulation of differentiated prevention and control measures, making resource allocation more targeted.

[0096] Step 4: Obtain the concentration of hazardous gases at the intersection of coal mine roadways and the dimensions of each roadway, and determine the average rate of change of hazardous gases in the roadway area as the characteristic of hazardous gas concentration change at all heat accumulation centers in the roadway.

[0097] In this embodiment, the types of hazardous gases include CO, CO2, CH4, C2H6, and C2H4;

[0098] The dimensions of a tunnel include its length, width, and height;

[0099] The method for determining gas variables in a roadway area based on the concentration of hazardous gases, air velocity, and the dimensions of each roadway is as follows:

[0100] The volume of the roadway is calculated based on its dimensions. The sampling interval is set with the same time length. The concentration of hazardous gas is collected based on the concentration of hazardous gas at adjacent time intervals. The concentration increment of hazardous gas in the roadway is calculated based on the concentration of hazardous gas at both ends of the roadway. The concentration change of hazardous gas is calculated based on the duration of the sampling interval and the concentration increment of hazardous gas. The average rate of change of hazardous gas in the roadway is calculated based on the volume of the roadway and the concentration change of hazardous gas.

[0101] CO is the first and most stable indicator gas produced during the low-temperature oxidation stage of coal (from room temperature to about 70°C). Its appearance precedes open flames and even smoke, making it the earliest warning signal and providing the longest warning time. The presence of ethylene (C2H4) and ethane (C2H6) is a key turning point, usually indicating that the coal temperature has risen to a certain level (above 110-130°C), hydrocarbons begin pyrolysis, and the oxidation reaction enters an accelerated phase. By analyzing the CH4 value, it is possible to determine whether the increased CO concentration originates from coal oxidation (increased ratio) or is simply due to dilution from gas outbursts (unchanged ratio), greatly reducing the false alarm rate. Simultaneously, the CO2 value is a very important criterion; this ratio changes systematically with increasing temperature, providing a more reliable indication of the oxidation process than using CO concentration alone.

[0102] This step incorporates the dynamic rate of change of hazardous gases as a key feature into the coal spontaneous combustion prediction model, overcoming the limitation of traditional methods that only use static gas concentration values. By quantifying the changing trends of gas concentrations within the roadway space, this step captures indirect but highly sensitive evidence of the intensity of the coal oxidation reaction, providing crucial time-dimensional information for the prediction model.

[0103] By calculating the average rate of change of hazardous gases, the intensification of coal oxidation reactions can be detected earlier and more sensitively. Even before a significant increase in temperature, the concentration change rates of key gases such as CO and C2H4 will show anomalies first, providing a valuable early warning window. This calculation also fully considers the influence of roadway size and ventilation. By combining roadway volume (size) and concentration increment, the calculated rate of change is a standardized indicator, eliminating interference from differences in space size and ventilation conditions across different roadways. This ensures the comparability of data from different locations and accurately reflects the source strength of gas generation.

[0104] Step 5: Using the heat accumulation center point as the node and the distance between nodes as the edge, the model input data is formed by constructing node features and edge features. A coal spontaneous combustion temperature prediction model is constructed through a graph convolutional neural network, outputting the real-time maximum temperature and confidence level of coal in each grid, and optimizing the grid length.

[0105] This step deeply integrates graph structure learning with physical mechanism models, constructing a graph convolutional neural network (GCN) framework specifically tailored for the problem of coal spontaneous combustion prediction. It successfully solves the inherent problems of spatial non-uniformity, multi-source heterogeneous data fusion, and computational efficiency optimization in coal spontaneous combustion prediction.

[0106] By abstracting physically interconnected heat accumulation centers as graph nodes and constructing edge features using node distances and ventilation directions, this model expresses the complex topological relationships and transmission effects of ventilation dynamics in underground spaces. This allows the neural network to "understand" how heat and gas propagate between various hazardous areas under specific ventilation conditions. Then, by fusing node features (heat, gas, coal quality) and edge features (ventilation, distance), the GCN model can simultaneously learn the state of a node and its interactions with neighboring nodes, thus achieving a systematic, rather than isolated, assessment of the spontaneous combustion process. Finally, by introducing a sparrow optimization algorithm to adaptively optimize the grid length, a dynamic balance between computational accuracy and efficiency is achieved, ensuring the feasibility of the method in large and complex mines.

[0107] This step unifies the encoding of the multi-source information (spatial grid, physical model, thermal characteristics, and gas changes) generated in steps 1 to 4, transforming it into structured data that graph neural networks can efficiently process. Through its powerful neighborhood information aggregation capability, the GCN model can accurately capture the key physical law that the temperature change of a node (hazardous area) depends not only on its own state but also on the influence of heat and gas transport from upwind nodes (represented by ventilation edge features). Ultimately, this GNN model, built upon physical mechanisms, can output prediction accuracy and confidence levels far exceeding traditional methods. Through self-optimization of the grid scale, it forms a complete closed-loop system that combines predictive intelligence with computational economy, providing coal mines with an unprecedented, accurate, and reliable spontaneous combustion early warning solution.

[0108] In this embodiment, the step of constructing a coal spontaneous combustion temperature prediction model using a graph convolutional neural network includes: model construction, model training, and data output;

[0109] The model building process is as follows:

[0110] The node features are the grid center coordinates of the heat accumulation area, the characteristics of hazardous gas concentration changes, the heat accumulation area, the heat characteristics of each grid within the heat accumulation area, and the coal type.

[0111] The ventilation direction and the distance between nodes are used as edge features;

[0112] Graph convolutional neural networks include an input and adjacency matrix layer, a graph convolutional layer, an edge feature fusion layer, a global graph pooling layer, and a fully connected regression layer.

[0113] The input layer and the adjacency matrix layer are responsible for receiving and structuring the input data. Each heat accumulation center point is defined as a node in the graph, and its features (such as coordinates, gas concentration, heat characteristics, etc.) are used as node feature inputs. At the same time, an adjacency matrix is ​​constructed based on the spatial location and ventilation direction between nodes to define the connection relationship between nodes, and the ventilation direction and distance are used as edge features. In this way, the risk points and their complex relationships in the physical space are transformed into graph structure data, providing a foundation for subsequent calculations.

[0114] The graph convolutional layer is the core of the model. It aggregates the feature information of each node's neighboring nodes through the connection relationship defined by the adjacency matrix, thereby capturing the local spatial dependencies between nodes. For example, how a temperature rise at a point will affect its downwind direction or the state of neighboring points. This layer, through multiple stacks, enables each node to receive information from more distant neighbors, thereby simulating the propagation process of heat and gas in the tunnel network.

[0115] The role of the edge feature fusion layer is to deeply integrate edge features (such as wind force in the ventilation direction and precise distance between nodes) into the feature extraction process. Through attention mechanisms or weighted aggregation, it enables the model to distinguish between "strong connections" (such as node pairs with strong airflow and close proximity) and "weak connections" (such as node pairs with no ventilation relationship or far distance), thereby more accurately simulating the dominant influence of ventilation dynamics on heat and gas transport processes.

[0116] The role of the global graph pooling layer is to integrate the node features of the entire graph (i.e., the risk network composed of all heat accumulation centers) after all nodes have fully exchanged information and extracted features. It generates a generalized feature vector of fixed size that can represent the global state of the entire coal mine risk area through operations such as summation, mean, or attention pooling.

[0117] Finally, the fully connected regression layer receives this global feature vector and maps it to the final prediction output through a series of fully connected neural network layers. This output is the real-time maximum temperature and confidence level of coal in each grid, thus completing the regression task from complex multi-source heterogeneous data to specific predicted values.

[0118] The model training process is as follows:

[0119] Based on historical input data, sampling is performed at the same time intervals. The input data of the previous moment is used as the input of the model, and the real-time highest temperature of coal in the grid at the next moment is used as the output of the model to train the coal spontaneous combustion temperature prediction model.

[0120] The data output process is as follows:

[0121] The real-time input data is used as input to the coal spontaneous combustion temperature prediction model to predict the real-time maximum temperature of coal in the grid at the next moment.

[0122] By using the grid center coordinates of heat accumulation areas, hazardous gas concentration variation characteristics, heat accumulation area identifiers, grid thermal characteristics, and coal type as node features, and ventilation direction and node spacing as edge features, this method significantly improves the performance of coal spontaneous combustion prediction models through multi-dimensional data fusion and physical relationship quantification. Node features comprehensively describe the essential attributes of each hazardous area; spatial coordinates provide location information to capture geographical relevance; gas concentration variation characteristics dynamically reflect the intensity of oxidation reactions, providing early warning; thermal characteristics (such as cumulative heat and temperature difference) directly quantify the thermodynamic state; and coal type defines the intrinsic properties of the material (such as reactivity and thermal conductivity), enabling the model to distinguish the spontaneous combustion behavior of different coal types. These features together constitute an "information complex" reflecting the state of each node, providing a solid data foundation for prediction.

[0123] Edge features precisely define the interaction mechanisms between nodes. The ventilation direction introduces asymmetric influence relationships, ensuring the model can simulate the transfer paths of heat and gas along the airflow direction (such as the significant influence of upwind nodes on downwind nodes). Node distance quantifies the interaction strength, conforming to the physical law of "stronger near, weaker far." This edge feature design enables graph convolutional networks to simulate the real propagation process of disasters underground, rather than simply performing topological aggregation. This feature construction method transforms the physical rules of the coal mine environment (ventilation, heat transfer, chemical reactions) into learnable relational priors for graph neural networks, greatly enhancing the model's accuracy, generalization ability, and interpretability, enabling it to extract hazardous evolution patterns that conform to mechanical laws from the data.

[0124] In this embodiment, the method for optimizing the grid length is as follows:

[0125] The objective function is set by weighted summation based on the confidence grid number of the real-time temperature output by the trained model. The grid length is then optimized using the sparrow optimization algorithm, which includes a discoverer and a follower.

[0126] The core purpose of optimizing grid length is to achieve an optimal balance between computational efficiency, sensor usage costs, and prediction accuracy. This method dynamically evaluates the impact of different grid division accuracies on the reliability of prediction results using model confidence as a key indicator. A larger number of grids with high confidence indicates more reliable overall prediction results at the current grid scale. This evaluation criterion is transformed into a quantifiable objective function through a weighted summation method, aiming to maximize the range of high-confidence prediction regions. The Sparrow Optimization algorithm, with its powerful global search capability and efficient convergence characteristics, automatically searches for the optimal grid length solution for this objective function in a complex parameter space, avoiding the blindness and inefficiency of traditional manual trial-and-error parameter tuning.

[0127] This optimization process plays a crucial adaptive adjustment role in the overall prediction method. It ensures that the grid partitioning is neither too coarse (leading to loss of detail and decreased prediction accuracy) nor too fine (resulting in huge computational overhead and reduced real-time performance), and instead of increasing the number of sensors used, it autonomously determines the most suitable spatial resolution based on the model's performance in actual predictions, taking into account the current coal characteristics and data conditions. Ultimately, this allows the entire graph convolutional neural network prediction system to significantly improve computational efficiency and engineering practicality while maintaining high-precision early warning, forming a self-optimizing closed loop.

[0128] The Sparrow Optimization Algorithm simulates the foraging and anti-predation behavior of a sparrow population (including discoverers and followers). It iteratively searches for the optimal grid length through swarm intelligence: discoverers explore all possible global solutions, while followers explore local solutions around high-quality ones, quickly converging to the grid length solution that maximizes the objective function. The essence of this optimization process is to enable the system to self-adjust its spatial awareness granularity, finding the optimal balance between being too coarse (fast but with low confidence and poor accuracy) and too fine (slow and potentially overfitting). This allows for the automatic optimization of computational efficiency and modeling accuracy while ensuring the reliability of prediction results, forming an intelligent closed loop that runs through data input, model prediction, and parameter adjustment.

[0129] All the above formulas use dimensionless numerical values ​​for calculation, and the numerical values ​​substituted into the formulas are all in the International System of Units (SI). The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0130] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0132] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A coal spontaneous combustion temperature prediction method based on a graph convolutional neural network, characterized in that, The specific steps include: Step 1: A three-dimensional coordinate system is established for the area where the coal to be predicted is located, and coordinate data of the coal in the coordinate system are obtained, and the area where the coal is located is divided into a three-dimensional grid according to the same length interval, and whether the grid is a bare area is judged by the exposed area of the coal in each grid; Step 2: For each grid judged as a bare area, the parameters of the coal in the grid, the oxygen concentration are obtained to construct an oxygen diffusion model, the wind speed, air temperature and coal surface temperature on the coal surface are obtained, and a coal heat model is constructed according to the parameters of the coal, and a temperature distribution function of the internal temperature of the coal changing with depth is calculated according to the oxygen diffusion model and the heat model; Step 3: The surface temperature of the coal at the center of the exposed area in each grid and the temperature distribution function are used to calculate the cumulative heat, the surface temperature difference and the average temperature of the center point, which are taken as the heat characteristics of the center of the grid, the heat pointing value is calculated by the heat characteristics of the center of the grid and the surrounding grids, the heat aggregation direction is judged, and the heat aggregation center point and the heat aggregation area are judged according to the heat aggregation direction; Step 4: The concentration of the dangerous gas at the intersection of the coal mine roadway and the size of each roadway are obtained, and the average change rate of the dangerous gas in the roadway area is judged as the dangerous gas concentration change characteristic of all heat aggregation center points in the roadway; Step 5: The heat aggregation center point is taken as a node, and the distance between the nodes is taken as an edge, the node characteristics and the edge characteristics are constructed to form the model input data, the coal spontaneous combustion temperature prediction model is constructed by the graph convolutional neural network, and the real-time maximum temperature and the confidence of the coal in each grid are output, and the grid length is optimized.

2. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 1, characterized in that: The method for judging whether the grid is a bare area by the exposed area of the coal in each grid is: a bare area threshold is set, when the exposed area of the coal in the grid is greater than the threshold, the grid is judged as a bare area.

3. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 1, characterized in that: The parameters of the coal include the type of the coal, the size of the void of the coal, the depth of the coal, the heat release rate of the oxidation reaction, the thermal conductivity, the density of the coal, and the specific heat capacity; The construction process of the oxygen diffusion model is: wherein is the oxygen concentration at the depth at the time instant t, at the depth is the oxygen diffusion coefficient, which depends on the type of coal, is the wind speed, is the oxidation efficiency constant of the coal, are the depth variable and the time variable, respectively.

4. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 3, characterized in that: The construction process of the coal heat model is: wherein, is the heat release rate of the oxidation reaction, is the enthalpy of the oxidation reaction.

5. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 4, characterized in that: The specific process of the temperature distribution function of the internal temperature of the coal changing with depth is: wherein, is the temperature of the coal surface at a depth of is the temperature of the coal surface at a depth of is the temperature of the coal surface at a depth of is the thermal conductivity of the coal, is the density of the coal, is the specific heat capacity of the coal, is the temperature of the coal surface.

6. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 1, characterized in that: The method for calculating the cumulative heat, the surface temperature difference and the average temperature of the center point by the surface temperature of the coal at the center of the exposed area in each grid and the temperature distribution function is: The cumulative heat of the center point is obtained by integrating the depth according to the temperature distribution function; The average temperature is obtained according to the proportion of the cumulative heat and the temperature analysis depth; The surface temperature difference includes the maximum surface temperature difference and the minimum surface temperature difference, the maximum temperature and the minimum temperature distributed at the center point are calculated according to the temperature distribution function and the temperature analysis depth, and the maximum surface temperature difference and the minimum surface temperature difference are obtained by subtracting the surface temperature.

7. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 1, characterized in that: The method for calculating the heat pointing value through the heat features of the grid and the surrounding grid center, judging the heat gathering direction, judging the heat gathering center point and the heat gathering area according to the heat gathering direction is as follows: the heat features of the grid and the surrounding 8 grids in front, back, left, right, left front, right front, left rear and right rear directions on the exposed layer are subtracted to obtain the heat feature difference, then the heat pointing value is calculated by assigning a weight to each heat feature difference, the position of the grid with the minimum heat pointing value between the grids is selected as the heat gathering direction, the grid at the end of the heat gathering direction is taken as the heat gathering center point, the position of the grid with the minimum heat feature between the adjacent heat gathering center points is taken as the area boundary point, the heat gathering center point at the area boundary of the coal is taken as the heat gathering area boundary, and the heat gathering area is formed by connecting the area boundary points around the heat gathering center point and the heat gathering area boundary.

8. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 1, characterized in that: The dangerous gas types include CO, CO2, CH4, C2H6 and C2H4; The size of the roadway includes the length, width and height of the roadway; According to the concentration of the dangerous gas, the air flow rate and the size of each roadway, the method for judging the gas variable in the roadway area is as follows: The volume of the roadway is calculated according to the size of the roadway, the collection interval is set through the same time length, the concentration of the dangerous gas collected in the adjacent time intervals is obtained, the concentration increment of the dangerous gas in the roadway is calculated according to the concentration of the dangerous gas at both ends of the roadway, the concentration change of the dangerous gas is calculated according to the time length of the collection interval and the concentration increment of the dangerous gas, and the average change rate of the dangerous gas in the roadway is calculated according to the volume of the roadway and the concentration change of the dangerous gas.

9. The coal spontaneous combustion temperature prediction method based on the graph convolutional neural network according to claim 1, characterized in that: The steps of constructing the coal spontaneous combustion temperature prediction model through the graph convolutional neural network include model construction, model training and data output. The process of model construction is as follows: The grid center coordinates of the heat gathering area, the dangerous gas concentration change feature, the heat gathering area, the heat feature of each grid in the heat gathering area and the coal type are taken as the node features; The ventilation direction and the distance between the nodes are taken as the edge features; The graph convolutional neural network includes an input and adjacency matrix layer, a graph convolutional layer, an edge feature fusion layer, a global graph pooling layer and a fully connected regression layer; The process of model training is as follows: According to the historical input data, the input data at the previous moment is taken as the input of the model, the real-time maximum temperature of the coal in the grid at the next moment is taken as the output of the model, and the coal spontaneous combustion temperature prediction model is trained through the same time interval sampling; The process of data output is as follows: According to the real-time input data as the input of the coal spontaneous combustion temperature prediction model, the real-time maximum temperature of the coal in the grid at the next moment is predicted.

10. The coal spontaneous combustion temperature prediction method based on a graph convolutional neural network according to claim 1, characterized in that: The method for optimizing the grid length is as follows: The confidence grid number of the real-time temperature output by the trained model is set as the target function through the weight assignment summation method, and the grid length is optimized through the sparrow optimization algorithm, wherein the sparrow optimization algorithm includes a discoverer and a follower.

Images