Rail transit material combustion smoke density dynamic monitoring method and system

Abstract

The present application provides a rail transit material combustion smoke density dynamic monitoring method and system, relating to the technical field of rail transit, comprising collecting smoke concentration data of multiple sensors in a combustion test box, using a three-dimensional space interpolation algorithm of adaptive grid refinement for three-dimensional reconstruction, combining material combustion characteristic parameters to construct a combustion kinetics model, fusing high-precision three-dimensional distribution map and the kinetics model, constructing a multi-scale diffusion prediction model based on graph neural network, generating smoke density distribution and diffusion path prediction data at future time, so as to realize accurate monitoring and early warning of rail transit carriage smoke density.

Description

Technical Field

[0001] This invention relates to rail transit technology, and more particularly to a method and system for dynamic monitoring of smoke density from the combustion of rail transit materials. Background Technology

[0002] With the rapid development of urban rail transit, the safety of rail transit vehicles has become an increasingly important concern. Inside rail transit vehicles, various decorative materials and equipment, once ignited, can produce large amounts of toxic fumes, seriously threatening passenger safety. Monitoring the smoke density of rail transit materials is a crucial indicator for assessing their safety performance and is essential for preventing fires and developing effective evacuation strategies. Traditional methods for detecting the smoke density of rail transit materials primarily rely on static testing, using optical equipment within a standard test chamber to measure the smoke density produced by material combustion and assess the material's safety performance. However, with advancements in computer and sensor technologies, dynamically monitoring the smoke density of rail transit materials and predicting their diffusion paths has become possible, which is of significant value in improving the safety level of rail transit.

[0003] First, traditional monitoring of smoke density from combustion of rail transit materials typically uses sensors at single points or a small number of fixed locations for data collection. This fails to fully reflect the three-dimensional distribution of smoke within the combustion test chamber or carriage, resulting in inaccurate monitoring results and making it difficult to provide comprehensive data support for carriage safety assessment.

[0004] Secondly, most existing monitoring methods are limited to static data collection and simple data analysis, lacking the ability to predict the dynamic process of smoke diffusion and unable to predict the diffusion path and future concentration distribution of smoke in real time. This makes it difficult to make effective evacuation planning and ventilation control decisions when a fire occurs.

[0005] Finally, existing technologies typically fail to effectively combine the combustion characteristics of materials with smoke diffusion models, and lack comprehensive consideration of key parameters such as heat release rate and mass loss rate during material combustion. This results in the inability to accurately simulate and predict smoke generation and diffusion behavior in complex environments, limiting the accuracy and practicality of fire early warning systems. Summary of the Invention

[0006] This invention provides a method and system for dynamic monitoring of combustion smoke density of rail transit materials, which can solve the problems in the prior art.

[0007] A first aspect of the present invention provides a method for dynamic monitoring of the smoke density of combustion in rail transit materials, comprising:

[0008] Smoke concentration data are acquired from multiple smoke concentration sensors inside the rail transit material combustion test chamber, wherein the multiple smoke concentration sensors are distributed in a three-dimensional grid pattern inside the rail transit material combustion test chamber;

[0009] Based on the position coordinate information of the multiple smoke concentration sensors and the corresponding smoke concentration data, an adaptive mesh refinement three-dimensional spatial interpolation algorithm is used to reconstruct the smoke density distribution in the rail transit material combustion test chamber in three dimensions. When the smoke concentration data is higher than the preset concentration threshold, the mesh density is automatically increased; when the smoke concentration data is lower than the preset concentration threshold, the mesh density is automatically reduced to obtain a high-precision three-dimensional distribution map.

[0010] Collect material combustion characteristic parameters inside the carriage, including the material's heat release rate, mass loss rate, and combustion product composition data; establish a material combustion kinetic model based on the aforementioned material combustion characteristic parameters;

[0011] The high-precision three-dimensional distribution map is deeply integrated with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural network. The multi-scale diffusion prediction model includes a macro-diffusion prediction layer and a micro-diffusion prediction layer.

[0012] Based on the output of the multi-scale diffusion prediction model, predictive data of smoke density distribution and diffusion path in the carriage at future times are generated to guide fire evacuation and ventilation control in rail transit carriages.

[0013] The three-dimensional reconstruction of the smoke density distribution in the combustion test chamber of the rail transit materials using an adaptive mesh refinement three-dimensional spatial interpolation algorithm includes:

[0014] Smoke density data collected by multiple smoke concentration sensors in a rail transit material combustion test chamber are obtained. The multiple smoke concentration sensors are distributed in three dimensions in the rail transit material combustion test chamber. The position information of the multiple smoke concentration sensors is mapped to a three-dimensional coordinate system.

[0015] The spatial gradient value of smoke density in the combustion test chamber of the rail transit materials is calculated. The ratio of the spatial gradient value to the maximum smoke density at the current moment is multiplied by the initial grid size to obtain the grid refinement criterion. The spatial grid of the combustion test chamber of the rail transit materials is adaptively refined according to the grid refinement criterion.

[0016] A three-dimensional spatial interpolation model is established based on the adaptively refined mesh structure. The interpolation weight coefficient is determined according to the local mesh characteristics in the rail transit material combustion test chamber. The interpolation weight coefficient is composed of the confidence coefficient of the smoke concentration sensor and the distance attenuation function based on the local mesh size.

[0017] The interpolation weight coefficients are input into the three-dimensional spatial interpolation model to calculate the smoke density values ​​of the grid nodes in the rail transit material combustion test chamber, thereby obtaining the three-dimensional reconstruction results of the smoke density distribution in the rail transit material combustion test chamber.

[0018] Establishing a material combustion dynamics model based on the aforementioned material combustion characteristic parameters includes:

[0019] Thermogravimetric data of the material are collected at different heating rates. The mass loss rate at each temperature point is calculated based on the thermogravimetric data. The peak temperature point is determined according to the mass loss rate. The peak temperature point is used as the characteristic temperature of material decomposition. The combustion process of the material is divided into multiple sub-reaction stages based on the characteristic temperature.

[0020] For each of the sub-reaction stages, multiple characteristic conversion rate points and their corresponding temperature data are selected. The mass loss degree of each stage is calculated based on the characteristic conversion rate points. The mass loss degree, the temperature data, and the heating rate are substituted into the Flynn-Woll-Ozawa equation.

[0021] Logarithmic transformation is performed on the Flynn-Woll-Ozawa equation to transform the nonlinear equation into a linear relationship between activation energy and frequency factor, establishing a set of parametric equations. The initial kinetic parameters of each sub-reaction are obtained by solving the set of parametric equations using the least squares method.

[0022] A parameter optimization space is constructed based on the initial kinetic parameters. The parameter optimization space is then input into a genetic algorithm optimization model, with the root mean square error between the experimental curve and the calculated curve as the optimization objective. Based on the optimization objective, parameter optimization calculations are performed through selection, crossover, and mutation operations of the genetic algorithm optimization model to obtain the optimal activation energy, frequency factor, and reaction order. A material combustion kinetic model is then established based on the optimal activation energy, the frequency factor, and the reaction order.

[0023] A logarithmic transformation is performed on the Flynn-Woll-Ozawa equations to convert the nonlinear equations into a linear relationship between activation energy and frequency factor, establishing a set of parametric equations. The initial kinetic parameters for each sub-reaction are obtained by solving the set of parametric equations using the least squares method:

[0024] The temperature-dependent reaction rate in the Flynn-Wohl-Ozawa equation is obtained. The Flynn-Wohl-Ozawa equation includes reaction conversion, frequency factor, activation energy, and reaction order. The Flynn-Wohl-Ozawa equation characterizes the exponential function relationship between the temperature-dependent reaction rate and the reaction conversion, frequency factor, activation energy, and reaction order. The exponential function relationship is then subjected to a logarithmic transformation to obtain a linearized equation.

[0025] A set of parametric equations is established based on the linearized equations. A target data sequence is constructed by combining multiple sets of logarithmic reaction rate values. A coefficient data sequence is constructed by combining the corresponding reciprocal temperature values ​​with the logarithmic conversion rate values. The set of parametric equations characterizes the linear relationship between the target data sequence and the coefficient data sequence.

[0026] The parametric equations are solved using the least squares method. The sum of squared errors between the experimental and fitted values ​​of the target data sequence is used as the objective function. The optimal parameter sequence that minimizes the objective function is obtained, and the optimal estimated value is obtained. The kinetic parameters are recovered from the optimal estimated value to obtain the initial kinetic parameters of each sub-reaction.

[0027] The high-precision three-dimensional distribution map is deeply integrated with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural networks, including:

[0028] The high-precision three-dimensional distribution map is divided into a grid structure, the physical parameters of the nodes are measured, the physical parameters are substituted into the material combustion kinetics model to calculate the node reaction rate, the node feature vector is constructed, and the edge feature vector is constructed based on the transmission parameters calculated from adjacent nodes.

[0029] Design a dual-feature fusion network, using a first weight matrix and a second weight matrix to map the node feature vector and the edge feature vector respectively, and obtain a micro-macro feature mapping matrix through an activation function, and calculate the evolution trend of the node state based on the micro-macro feature mapping matrix;

[0030] The evolution trend is input into a multi-level prediction network to calculate the prediction error between the local reaction rate and the overall transport field. When the prediction error is less than a preset error threshold, the material combustion diffusion prediction model is output.

[0031] Based on the output of the multi-scale diffusion prediction model, predictive data on smoke density distribution and diffusion path within the carriage at future times are generated to guide fire evacuation and ventilation control in rail transit carriages, including:

[0032] Determine the temporal and spatial characteristics of the output of the multi-scale diffusion prediction model, and fuse the temporal and spatial characteristics to obtain spatiotemporal enhanced features;

[0033] The spatiotemporal enhancement features are input into a causal convolutional network, and temporal features are extracted using dilated convolution operations. The temporal features are then passed between layers using a residual connection structure. The temporal correlation coefficient of the passed features is calculated, and the optimal number of network layers and dilation rate are determined based on the temporal correlation coefficient to generate a feature sequence with long-range dependencies.

[0034] A dual-branch prediction network is constructed based on the feature sequence. The density prediction branch generates smoke density distribution prediction data for future times inside the carriage, and the path prediction branch generates diffusion path prediction data for future times. The prediction error between the smoke density distribution prediction data and the diffusion path prediction data is calculated, and the weight coefficients of the prediction branches are dynamically adjusted according to the prediction error.

[0035] The reliability of the predicted data after weight adjustment is evaluated using a probabilistic prediction model. The confidence interval and calibration error of the reliability of the predicted data are calculated. The parameters of the probabilistic prediction model are iteratively optimized based on the confidence interval and the calibration error, and a prediction result with guaranteed reliability is output.

[0036] The safety level of each area of ​​the carriage is calculated based on the prediction results. The optimal evacuation route is planned based on the safety level. The smoke concentration deviation of each area is calculated using the smoke density distribution prediction data. Ventilation control instructions are generated based on the smoke concentration deviation.

[0037] The reliability of the weighted prediction data is evaluated using a probabilistic prediction model. The confidence interval and calibration error of the predicted data reliability are calculated. Based on the confidence interval and calibration error, the parameters of the probabilistic prediction model are iteratively optimized, and the reliable prediction results are output, including:

[0038] Construct a weight matrix for the prediction parameter association network, perform association analysis on the prediction parameters based on the prediction parameter association network and the weight matrix, set the error threshold between the predicted value and the true value as a reliability constraint, select the optimal parameter combination according to the weight matrix and the reliability constraint, calculate the probability deviation between the predicted value and the true value based on the optimal parameter combination, and establish the reliability probability distribution of the prediction result.

[0039] A dynamic confidence interval is constructed based on the reliability probability distribution and the weight matrix. A calibration error calculation model is constructed using the dynamic confidence interval and the reliability probability distribution. The uncertainty state of the prediction system is input into the calibration error calculation model to obtain the dynamic calibration error in the prediction process.

[0040] A knowledge modulation matrix is ​​constructed using the weight matrix. The dynamic calibration error is weighted using the knowledge modulation matrix. An adaptive learning rate is calculated based on the weighted dynamic calibration error. The parameters of the prediction model are iteratively optimized to obtain the optimal prediction parameters. Based on the optimal prediction parameters and the weight matrix, a prediction calculation is performed to output a prediction result with guaranteed reliability.

[0041] A second aspect of the present invention provides a dynamic monitoring system for the smoke density of combustion in rail transit materials, comprising:

[0042] The first unit is used to acquire smoke concentration data collected by multiple smoke concentration sensors in the rail transit material combustion test chamber. The multiple smoke concentration sensors are distributed in a three-dimensional grid pattern in the rail transit material combustion test chamber.

[0043] The second unit is used to reconstruct the smoke density distribution in the rail transit material combustion test chamber in three dimensions based on the position coordinate information of the multiple smoke concentration sensors and the corresponding smoke concentration data, using an adaptive mesh refinement three-dimensional spatial interpolation algorithm. When the smoke concentration data is higher than a preset concentration threshold, the mesh density is automatically increased; when the smoke concentration data is lower than the preset concentration threshold, the mesh density is automatically decreased to obtain a high-precision three-dimensional distribution map.

[0044] The third unit is used to collect material combustion characteristic parameters inside the carriage, including the material's heat release rate, mass loss rate, and combustion product composition data; and to establish a material combustion kinetic model based on the aforementioned material combustion characteristic parameters.

[0045] The fourth unit is used to deeply integrate the high-precision three-dimensional distribution map with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural networks. The multi-scale diffusion prediction model includes a macro-diffusion prediction layer and a micro-diffusion prediction layer.

[0046] The fifth unit is used to generate smoke density distribution prediction data and diffusion path prediction data for future moments inside the carriage based on the output results of the multi-scale diffusion prediction model, so as to guide the fire evacuation and ventilation control of the rail transit carriage.

[0047] A third aspect of the present invention provides an electronic device, comprising:

[0048] processor;

[0049] Memory used to store processor-executable instructions;

[0050] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0051] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0052] The beneficial effects of this application are as follows:

[0053] This invention solves the technical problems of low spatial resolution and inability to reflect the uneven distribution of actual smoke by arranging multiple smoke concentration sensors in a three-dimensional grid inside a combustion test chamber for rail transit materials and using an adaptive grid refinement three-dimensional spatial interpolation algorithm. This significantly improves the accuracy and spatial resolution of smoke density monitoring.

[0054] This invention deeply integrates high-precision three-dimensional distribution maps with material combustion dynamics models to construct a multi-scale diffusion prediction model based on graph neural networks. This model can predict smoke diffusion behavior from both macroscopic and microscopic levels, significantly improving the accuracy of smoke diffusion path and density prediction. Compared with traditional methods, the prediction accuracy is improved by 25%-35%, and the prediction time is reduced by 3-5 minutes.

[0055] This invention enables real-time monitoring and prediction of smoke density distribution in rail transit carriages, providing a scientific basis for fire evacuation and ventilation control in the event of a fire in the carriage. It can effectively reduce the risk of casualties in fire accidents, improve the safety level of rail transit, and has significant social value and application prospects. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the dynamic monitoring method for combustion smoke density of rail transit materials according to an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram illustrating the real-time reconstruction error under changes in smoke density during combustion, according to an embodiment of the present invention.

[0058] Figure 3 This is a flowchart illustrating the linearization solution of the Flynn-Wall-Ozawa equations for obtaining dynamic parameters according to an embodiment of the present invention.

[0059] Figure 4 This is a flowchart illustrating the material combustion and diffusion prediction process based on dual feature fusion, as described in an embodiment of the present invention.

[0060] Figure 5 This is a schematic diagram comparing the prediction accuracy under different training sample sizes in an embodiment of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0063] Figure 1 This is a flowchart illustrating the dynamic monitoring method for combustion smoke density of rail transit materials according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0064] Smoke concentration data are acquired from multiple smoke concentration sensors inside the rail transit material combustion test chamber, wherein the multiple smoke concentration sensors are distributed in a three-dimensional grid pattern inside the rail transit material combustion test chamber;

[0065] Based on the position coordinate information of the multiple smoke concentration sensors and the corresponding smoke concentration data, an adaptive mesh refinement three-dimensional spatial interpolation algorithm is used to reconstruct the smoke density distribution in the rail transit material combustion test chamber in three dimensions. When the smoke concentration data is higher than the preset concentration threshold, the mesh density is automatically increased; when the smoke concentration data is lower than the preset concentration threshold, the mesh density is automatically reduced to obtain a high-precision three-dimensional distribution map.

[0066] Collect material combustion characteristic parameters inside the carriage, including the material's heat release rate, mass loss rate, and combustion product composition data; establish a material combustion kinetic model based on the aforementioned material combustion characteristic parameters;

[0067] The high-precision three-dimensional distribution map is deeply integrated with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural network. The multi-scale diffusion prediction model includes a macro-diffusion prediction layer and a micro-diffusion prediction layer.

[0068] Based on the output of the multi-scale diffusion prediction model, predictive data of smoke density distribution and diffusion path in the carriage at future times are generated to guide fire evacuation and ventilation control in rail transit carriages.

[0069] In one optional implementation, the three-dimensional reconstruction of the smoke density distribution within the rail transit material combustion test chamber using an adaptive mesh refinement three-dimensional spatial interpolation algorithm includes:

[0070] Smoke density data collected by multiple smoke concentration sensors in a rail transit material combustion test chamber are obtained. The multiple smoke concentration sensors are distributed in three dimensions in the rail transit material combustion test chamber. The position information of the multiple smoke concentration sensors is mapped to a three-dimensional coordinate system.

[0071] The spatial gradient value of smoke density in the combustion test chamber of the rail transit materials is calculated. The ratio of the spatial gradient value to the maximum smoke density at the current moment is multiplied by the initial grid size to obtain the grid refinement criterion. The spatial grid of the combustion test chamber of the rail transit materials is adaptively refined according to the grid refinement criterion.

[0072] A three-dimensional spatial interpolation model is established based on the adaptively refined mesh structure. The interpolation weight coefficient is determined according to the local mesh characteristics in the rail transit material combustion test chamber. The interpolation weight coefficient is composed of the confidence coefficient of the smoke concentration sensor and the distance attenuation function based on the local mesh size.

[0073] The interpolation weight coefficients are input into the three-dimensional spatial interpolation model to calculate the smoke density values ​​of the grid nodes in the rail transit material combustion test chamber, thereby obtaining the three-dimensional reconstruction results of the smoke density distribution in the rail transit material combustion test chamber.

[0074] Several smoke concentration sensors were installed inside a combustion test chamber for rail transit materials, arranged in a three-dimensional configuration. The chamber measures 2 meters × 1.5 meters × 2.5 meters and contains 32 smoke concentration sensors. Each sensor has a unique three-dimensional coordinate identifier; for example, the first sensor's coordinates are (0.2, 0.3, 0.4) meters, the second sensor's coordinates are (0.2, 0.3, 1.2) meters, and so on, until the coordinate mapping of all 32 sensors is completed. This coordinate data is stored in the system database as the basis for subsequent calculations.

[0075] The system acquires the smoke density readings of each smoke concentration sensor at a specific time t. For example, at t=30 seconds, the first sensor reads 0.15 g / m³, the second sensor reads 0.21 g / m³, and so on, acquiring the density readings of all 32 sensors. After collecting this data, the system associates it with the corresponding sensor location information to form a data point set {(xi, yi, zi, di)}, where (xi, yi, zi) represents the three-dimensional coordinates of the i-th sensor, and di represents the smoke density value measured by that sensor.

[0076] The space of the rail transit material combustion test chamber was initially divided into grids with an initial grid size of 0.2 m × 0.2 m × 0.2 m. This resulted in a total of 1040 initial grid cells (10 × 8 × 13). For each grid cell, the system estimated the smoke density gradient at the center point of that cell based on surrounding sensor data.

[0077] When calculating the spatial gradient of smoke density, the system considers the estimated smoke density at the center points of the six adjacent grid cells (top, bottom, left, right, front, and back) for each grid cell. Assuming the center point of a grid cell has coordinates (x, y, z) and an estimated density value of d, the gradient in the x-direction can be obtained by dividing the difference between the density values ​​of this point and its adjacent points at x+0.2 and x-0.2 locations by the distance (0.4 meters). Similarly, the gradients in the y and z directions are calculated. The square root of the sum of the squares of the gradient values ​​in the three directions yields the spatial gradient magnitude at that point.

[0078] For each grid cell, the grid refinement criterion is calculated. Assuming the current maximum smoke density is dmax = 0.35 g / m³, and the spatial gradient of a certain grid cell is g = 0.25 g / m³ / m, then the grid refinement criterion value for that cell is (g / dmax) × 0.2 = (0.25 / 0.35) × 0.2 ≈ 0.143. The system sets a threshold of 0.1; when the refinement criterion value exceeds this threshold, the grid cell is refined.

[0079] During the mesh refinement process, mesh cells whose criterion values ​​exceed a threshold are divided into 8 sub-mesh cells (2×2×2). For example, if the original mesh cell size is 0.2m×0.2m×0.2m, after refinement, the size of each sub-mesh cell becomes 0.1m×0.1m×0.1m. In this example, approximately 150 initial mesh cells are refined, increasing the total number of meshes in the system from 1040 to approximately 2240.

[0080] Based on the mesh refinement results, the system establishes a three-dimensional spatial interpolation model. For each mesh node where smoke density needs to be estimated, the system first identifies the sensors that affect it. The influence radius is set to 0.5 meters, meaning that sensor data within 0.5 meters of the target node will be included in the calculation. Weights are assigned to the selected sensor data, with the weighting coefficient consisting of two parts:

[0081] The sensor reliability coefficient is determined based on the sensor's historical stability and calibration accuracy. For example, the reliability of the first sensor is 0.95, the second is 0.92, and so on. The distance attenuation function adopts an exponential decay form, with the effect decreasing as the distance increases. For example, if the distance between the sensor and the target node is r meters, the distance attenuation factor can be expressed as exp(-3r / h), where h is the local mesh size. This attenuation factor differs for refined regions (h=0.1 meters) and non-refined regions (h=0.2 meters).

[0082] The final interpolation weight coefficient is obtained by multiplying the sensor confidence coefficient by the distance attenuation factor. For example, if a sensor has a confidence coefficient of 0.95, is 0.15 meters away from the target node, and is located in a region with a grid size of 0.1 meters, then its weight coefficient is 0.95×exp(-3×0.15 / 0.1)≈0.25.

[0083] The smoke density data from each sensor are weighted and averaged using the aforementioned weighting coefficients to calculate the estimated smoke density value for the target node. This calculation process is repeated for all grid nodes to obtain the final three-dimensional reconstruction result of the smoke density distribution throughout the entire test chamber space.

[0084] Through the above implementation methods, the system can generate high-resolution three-dimensional visualization results of smoke density distribution, especially in areas with large smoke density gradients (such as around fire sources), providing more refined reconstruction results, thus providing more accurate analytical basis for the study of combustion safety of rail transit materials.

[0085] Figure 2 This is a schematic diagram illustrating the real-time reconstruction error under changes in smoke density during combustion, as described in an embodiment of the present invention.

[0086] This figure illustrates the trend of prediction error (RMSE) over time for three different meshing methods (this technical solution, the fixed mesh method, and the standard adaptive method) during material combustion. The horizontal axis represents combustion time, ranging from 0 to 60 seconds; the vertical axis represents reconstruction error, ranging from 0 to 0.20. The figure shows that all three methods exhibit a bimodal characteristic, but the error amplitude and fluctuation degree differ significantly. The fixed mesh method (circled marker) shows the largest error fluctuation, with two peaks at 20 and 45 seconds, reaching a maximum error of approximately 0.19. The standard adaptive method (square marker) performs well, with two peaks appearing around 20 and 45 seconds, and a maximum error of approximately 0.13. This technical solution (triangle marker) performs best, with the smoothest overall error curve; the errors at the two peaks are only approximately 0.07 and 0.08, respectively, maintaining the lowest prediction error throughout the entire combustion process. Especially in the initial stage (0-10 seconds) and later stage (50-60 seconds) of combustion, the error of this technical solution remains at a low level of 0.02-0.03, demonstrating excellent predictive stability and accuracy.

[0087] In one optional implementation, establishing a material combustion kinetics model based on the material combustion characteristic parameters includes:

[0088] Thermogravimetric data of the material are collected at different heating rates. The mass loss rate at each temperature point is calculated based on the thermogravimetric data. The peak temperature point is determined according to the mass loss rate. The peak temperature point is used as the characteristic temperature of material decomposition. The combustion process of the material is divided into multiple sub-reaction stages based on the characteristic temperature.

[0089] For each of the sub-reaction stages, multiple characteristic conversion rate points and their corresponding temperature data are selected. The mass loss degree of each stage is calculated based on the characteristic conversion rate points. The mass loss degree, the temperature data, and the heating rate are substituted into the Flynn-Woll-Ozawa equation.

[0090] Logarithmic transformation is performed on the Flynn-Woll-Ozawa equation to transform the nonlinear equation into a linear relationship between activation energy and frequency factor, establishing a set of parametric equations. The initial kinetic parameters of each sub-reaction are obtained by solving the set of parametric equations using the least squares method.

[0091] A parameter optimization space is constructed based on the initial kinetic parameters. The parameter optimization space is then input into a genetic algorithm optimization model, with the root mean square error between the experimental curve and the calculated curve as the optimization objective. Based on the optimization objective, parameter optimization calculations are performed through selection, crossover, and mutation operations of the genetic algorithm optimization model to obtain the optimal activation energy, frequency factor, and reaction order. A material combustion kinetic model is then established based on the optimal activation energy, the frequency factor, and the reaction order.

[0092] In establishing a combustion kinetics model based on the material's combustion characteristic parameters, the first step is to collect thermogravimetric data. A thermogravimetric analyzer is used to measure the mass change of the target material at different heating rates (e.g., 5℃ / min, 10℃ / min, 15℃ / min, and 20℃ / min), obtaining temperature-mass curves. By processing the collected thermogravimetric data, the mass loss rate at each temperature point is calculated, i.e., the rate of change of mass with temperature.

[0093] By analyzing the mass loss rate curves, peak temperature points at various heating rates can be identified. For example, at a heating rate of 10℃ / min, a polymer material may exhibit mass loss rate peaks at 350℃, 450℃, and 550℃. These peak temperature points are determined as characteristic temperatures of the material's thermal decomposition, used to divide the entire combustion process into multiple sub-reaction stages. For instance, the combustion process can be divided into three stages: the first stage is from room temperature to the first characteristic temperature; the second stage is from the first characteristic temperature to the second characteristic temperature; and the third stage is from the second characteristic temperature to the end of combustion.

[0094] For each predefined sub-reaction stage, multiple characteristic conversion rate points (e.g., 0.1, 0.2, 0.3, ..., 0.9) and their corresponding temperature data are selected. Taking a certain wood material as an example, in the first stage reaction, when the conversion rate is 0.2, the corresponding temperatures at heating rates of 5℃ / min, 10℃ / min, and 20℃ / min are 210℃, 225℃, and 240℃, respectively.

[0095] Based on these characteristic conversion rate points, the mass loss degree at each stage is calculated. The mass loss degree is defined as the ratio of the mass lost at a certain moment in that stage to the total mass lost in that stage. The mass loss degree, temperature data, and heating rate are substituted into the Flynn-Woll-Ozawa equation, which describes the relationship between conversion rate and temperature and time during the thermal decomposition of materials.

[0096] To simplify the solution process, a logarithmic transformation is performed on the Flynn-Woll-Ozawa equations, converting the nonlinear equations containing exponential relationships into a linear expression relating activation energy and frequency factor. This transformation simplifies the complex nonlinear problem into a linear one, facilitating subsequent solution. A system of parametric equations incorporating data from multiple characteristic conversion rate points is then established for the transformed linear equations.

[0097] The system of parametric equations was solved using the least squares method to obtain the initial kinetic parameters for each sub-reaction stage, including activation energy, frequency factor, and reaction order. Taking a certain plastic material as an example, the initial activation energy of the first-stage reaction was calculated to be 120 kJ / mol, the frequency factor to be 10^13 s^-1, and the reaction order to be 1.2 using the least squares method.

[0098] After obtaining the initial kinetic parameters, a parameter optimization space is constructed for further optimization. The parameter optimization space is a multi-dimensional search space centered on the initial kinetic parameters, floating upwards and downwards within a certain range (e.g., activation energy ±20 kJ / mol, frequency factor ±1 order of magnitude, reaction order ±0.5). This parameter optimization space is input into the genetic algorithm optimization model, with the root mean square error (RMSE) between the experimentally measured thermogravimetric curve and the curve predicted by the computational model serving as the optimization objective. The RMS error is calculated as the square root of the sum of the squares of the mass differences between the two curves at corresponding temperature points, divided by the number of data points.

[0099] In the optimization process of genetic algorithms, parameter optimization is performed through selection, crossover, and mutation operations. Specifically, the selection operation selects high-quality individuals using a roulette wheel method based on the individual's fitness (i.e., the reciprocal of the error value); the crossover operation swaps the parameters of the selected individuals with a certain probability (e.g., 0.8); and the mutation operation randomly fine-tunes the individual parameters with a lower probability (e.g., 0.1).

[0100] After multiple generations of evolution (e.g., 100 generations), the algorithm converges to the optimal solution or a state close to the optimal solution. Taking a certain composite material as an example, after optimization by the genetic algorithm, the optimal activation energy of the second-stage reaction is 155 kJ / mol, the optimal frequency factor is 10^14.5 s^-1, and the optimal reaction order is 1.5. At this point, the root mean square error between the calculated curve and the experimental curve is reduced to 0.003.

[0101] Based on the obtained optimal activation energy, frequency factor, and reaction order, a complete material combustion kinetic model was established. This model can describe the decomposition rate and mass loss of materials under different temperature conditions, and thus predict the combustion behavior of materials in actual fire environments. Model validation shows that the established kinetic model can accurately predict the thermal decomposition process of materials within a heating rate range of 5℃ / min to 20℃ / min, with a prediction error of less than 5%, meeting the requirements of engineering applications.

[0102] In one optional implementation, the Flynn-Woll-Ozawa equation is logarithmically transformed to convert the nonlinear equation into a linear relationship between activation energy and frequency factor, establishing a set of parametric equations. The initial kinetic parameters of each sub-reaction are obtained by solving the set of parametric equations using the least squares method, including:

[0103] The temperature-dependent reaction rate in the Flynn-Wohl-Ozawa equation is obtained. The Flynn-Wohl-Ozawa equation includes reaction conversion, frequency factor, activation energy, and reaction order. The Flynn-Wohl-Ozawa equation characterizes the exponential function relationship between the temperature-dependent reaction rate and the reaction conversion, frequency factor, activation energy, and reaction order. The exponential function relationship is then subjected to a logarithmic transformation to obtain a linearized equation.

[0104] A set of parametric equations is established based on the linearized equations. A target data sequence is constructed by combining multiple sets of logarithmic reaction rate values. A coefficient data sequence is constructed by combining the corresponding reciprocal temperature values ​​with the logarithmic conversion rate values. The set of parametric equations characterizes the linear relationship between the target data sequence and the coefficient data sequence.

[0105] The parametric equations are solved using the least squares method. The sum of squared errors between the experimental and fitted values ​​of the target data sequence is used as the objective function. The optimal parameter sequence that minimizes the objective function is obtained, and the optimal estimated value is obtained. The kinetic parameters are recovered from the optimal estimated value to obtain the initial kinetic parameters of each sub-reaction.

[0106] In fuel kinetic analysis, to obtain more accurate initial kinetic parameters, this embodiment provides a parameter solution method based on the Flynn-Wol-Ozawa equation. The Flynn-Wol-Ozawa equation is a commonly used equation to describe the pyrolysis process of solid fuels, which expresses the exponential functional relationship between temperature, reaction rate, reaction conversion rate, frequency factor, activation energy, and reaction order.

[0107] In this embodiment, temperature reaction rate data in the Flynn-Woll-Ozawa equation are first obtained. Through thermogravimetric analysis, mass change data of the sample at different temperatures can be obtained. Based on the mass change curves, the reaction conversion rate α and the corresponding reaction rate dα / dt at different temperature points can be calculated. For example, for a coal sample subjected to pyrolysis at a heating rate of 10℃ / min, mass change data ranging from 200℃ to 900℃ can be obtained.

[0108] Based on the mass change data, the reaction conversion rate α was calculated to be 0.15 and the reaction rate dα / dt was 0.002 min^-1 at 350℃; the reaction conversion rate α was 0.38 and the reaction rate dα / dt was 0.005 min^-1 at 450℃; and the reaction conversion rate α was 0.72 and the reaction rate dα / dt was 0.009 min^-1 at 550℃.

[0109] A logarithmic transformation of the Flynn-Wall-Ozawa equation converts the exponential function relationship into a linear one. The Flynn-Wall-Ozawa equation describes the relationship between the reaction rate dα / dt and the reaction conversion α, the frequency factor A, the activation energy E, and the reaction order n. By taking the logarithmic transformation, a linearized equation is obtained, in which the logarithmic value of the reaction rate has a linear relationship with the reciprocal of the temperature and the logarithmic value of the reaction conversion.

[0110] For example, for the above experimental data, at a temperature of 350℃ (i.e., 623K), the logarithm of the reaction rate dα / dt is -6.2146, the reciprocal of the temperature 1 / T is 0.00161, and the logarithm of the reaction conversion α is -1.8971; at a temperature of 450℃ (i.e., 723K), the logarithm of the reaction rate is -5.2983, the reciprocal of the temperature is 0.00138, and the logarithm of the reaction conversion is -0.9676; at a temperature of 550℃ (i.e., 823K), the logarithm of the reaction rate is -4.7105, the reciprocal of the temperature is 0.00122, and the logarithm of the reaction conversion is -0.3285.

[0111] A set of parametric equations is established based on the linearized equations. A target data sequence is constructed from multiple sets of logarithmic reaction rate values, and a coefficient data sequence is constructed from the corresponding reciprocal values ​​of temperature and the logarithmic values ​​of conversion rates. For example, for the three sets of data mentioned above, the target data sequence is [-6.2146, -5.2983, -4.7105], and the coefficient data sequence includes the reciprocal values ​​of temperature [0.00161, 0.00138, 0.00122] and the logarithmic values ​​of reaction conversion rates [-1.8971, -0.9676, -0.3285]. The parametric equations characterize the linear relationship between the target data sequence and the coefficient data sequence. The parameters to be determined include coefficients related to activation energy, reaction order, and frequency factor.

[0112] The least squares method is used to solve the parametric equations. The sum of squared errors between the experimental and fitted values ​​of the target data sequence is used as the objective function. The optimal parameter sequence that minimizes the objective function is then obtained, yielding the best estimate. For example, for the above data, the least squares method yields a coefficient of -15000 related to the activation energy, a coefficient of 0.8 related to the reaction order, and a coefficient of 16.2 related to the frequency factor.

[0113] Kinetic parameters were recovered from the optimal estimates to obtain the initial kinetic parameters for each sub-reaction. Based on the coefficient related to the activation energy, the activation energy E was calculated to be 124.7 kJ / mol; based on the coefficient related to the reaction order, the reaction order n was calculated to be 0.8; and based on the coefficient related to the frequency factor, the frequency factor A was calculated to be 10^7 min^-1. These parameter values ​​can be used as initial values ​​for subsequent kinetic model optimization.

[0114] To verify the accuracy of the parameter solution, the obtained initial kinetic parameters can be substituted into the Flynn-Woll-Ozawa equation to calculate the theoretical reaction rate, which is then compared with the experimentally measured rate. For example, for a temperature of 350℃ and a reaction conversion rate of 0.15, the calculated theoretical reaction rate is 0.0019 min⁻¹, close to the experimental value of 0.002 min⁻¹; for a temperature of 450℃ and a reaction conversion rate of 0.38, the calculated theoretical reaction rate is 0.0048 min⁻¹, close to the experimental value of 0.005 min⁻¹; and for a temperature of 550℃ and a reaction conversion rate of 0.72, the calculated theoretical reaction rate is 0.0087 min⁻¹, close to the experimental value of 0.009 min⁻¹. The calculation results show that the initial kinetic parameters obtained by this method have good accuracy.

[0115] The above method can effectively transform the nonlinear Flynn-Woll-Ozawa equation into a linear relationship. The parametric equations can then be solved using the least squares method to obtain the initial kinetic parameters of each sub-reaction, providing basic data support for subsequent fuel pyrolysis kinetic analysis.

[0116] Figure 3 The flowchart for solving the dynamic parameters of the Flynn-Wall-Ozawa equation linearization according to an embodiment of the present invention is as follows:

[0117] This figure details the process of solving the kinetic parameters based on the Flynn-Wol-Ozawa equations, presenting a clear top-down workflow. First, experimental data on temperature and reaction rate are acquired, and preliminary modeling is performed using the fundamental Flynn-Wol-Ozawa equation da / dt = A(1-α)^n exp(-E / RT). To simplify the calculation, the equation is logarithmically transformed into a linear form ln(da / dt) = ln A + n ln(1-α) - E / RT. Subsequently, a set of parametric equations is constructed, with ln(da / dt) set as the target data and 1 / T and ln(1-α) as coefficients. Based on these data, the least squares method is used to minimize the sum of squared errors, thereby obtaining the optimal estimate. Finally, through a kinetic parameter recovery step, three key kinetic parameters of the reaction system are obtained: the frequency factor A, the activation energy E, and the reaction order n. This systematic parameter solution method not only ensures the accuracy of the calculation results but also provides a clear mathematical approach, offering important theoretical support for the study of material combustion kinetics.

[0118] In one optional implementation, the high-precision three-dimensional distribution map is deeply integrated with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on a graph neural network, including:

[0119] The high-precision three-dimensional distribution map is divided into a grid structure, the physical parameters of the nodes are measured, the physical parameters are substituted into the material combustion kinetics model to calculate the node reaction rate, the node feature vector is constructed, and the edge feature vector is constructed based on the transmission parameters calculated from adjacent nodes.

[0120] Design a dual-feature fusion network, using a first weight matrix and a second weight matrix to map the node feature vector and the edge feature vector respectively, and obtain a micro-macro feature mapping matrix through an activation function, and calculate the evolution trend of the node state based on the micro-macro feature mapping matrix;

[0121] The evolution trend is input into a multi-level prediction network to calculate the prediction error between the local reaction rate and the overall transport field. When the prediction error is less than a preset error threshold, the material combustion diffusion prediction model is output.

[0122] In the mesh generation and feature construction stage, the system first divides the high-precision 3D distribution map into a regular mesh structure. Taking the combustion process of a certain material as an example, a sample with dimensions of 100mm×50mm×30mm is divided into cubic mesh units of 0.5mm×0.5mm×0.5mm, resulting in a total of 2,000,000 mesh nodes. For each node, its physical parameters are measured, including key physical quantities such as temperature, density, pressure, and oxygen concentration.

[0123] For example, node (25, 30, 42) has a measured temperature of 453 K, a density of 1.2 g / cm³, a pressure of 101.3 kPa, and an oxygen concentration of 21%. Substituting these physical parameters into the material combustion kinetics model, the calculated reaction rate for this node is 0.015 mol / (m³·s). Based on this, a feature vector of length 12 is constructed for this node, containing the aforementioned physical parameters and reaction rate. Based on the mass and energy exchange patterns between adjacent nodes, transport parameters are calculated, such as the thermal conductivity coefficient of 0.024 W / (m·K) and the mass diffusion coefficient of 2.88 × 10⁻⁻⁻⁴ W / (m·K). 5 m² / s, constructing edge feature vectors with a dimension of 8.

[0124] In the design phase of the dual-feature fusion network, the system employs two weight matrices to map node feature vectors and edge feature vectors. The first weight matrix, with a size of 12×64, is used to map node feature vectors; the second weight matrix, with a size of 8×64, is used to map edge feature vectors. The mapping results are then subjected to a nonlinear transformation using the ReLU activation function to obtain a micro-to-macro feature mapping matrix with a size of 64 times the number of nodes.

[0125] Taking node (25, 30, 42) in the above example as an example, after processing by the feature fusion network, its first five feature mapping values ​​are [0.82, 0.56, 0.93, 0.47, 0.71]. Based on this mapping matrix, graph convolution operations are applied to calculate the evolution trend of the node state. Specifically, for each node, the feature information of its neighboring nodes is aggregated to update its own state. After processing by three graph convolutional layers, the predicted evolution trend of node (25, 30, 42) is a temperature rise rate of 15 K / s and a reaction rate increase of 0.003 mol / (m³·s), etc.

[0126] In the multi-layer prediction network construction phase, the system inputs the evolution trend into a prediction network consisting of four fully connected layers, with 128, 256, 128, and 64 neurons in each layer, respectively. Iterative predictions are performed with a time step of 0.1 s, calculating the prediction error between the local reaction rate and the overall transport field. The prediction network outputs the physical parameters and reaction rates of each node at time t+1, which are compared with actual observations. In the test case, after 10 iterations of training, the average prediction error decreased to 0.82%, less than the preset 1% error threshold, at which point the final material combustion diffusion prediction model is output.

[0127] In a specific application example, the combustion diffusion process of a polyurethane foam material was predicted. The initial ignition point was set at the center of the sample surface at a temperature of 873 K. The constructed prediction model was applied to simulate the combustion diffusion process for 100 seconds, recording the state changes every 0.1 seconds. The model successfully predicted the phenomenon of the flame spreading outwards at an average speed of 0.8 mm / s, and the maximum deviation of the temperature distribution from the experimental measurement was 5.4%, accurately capturing the changes in the internal temperature gradient of the material and the movement trajectory of the combustion reaction zone.

[0128] To further improve model accuracy, an adaptive mesh refinement mechanism was designed. In the combustion front region, the mesh was further refined to 0.2mm × 0.2mm × 0.2mm to more accurately capture the dramatic changes in the reaction front. Through dynamic adjustment of the mesh generation, the model prediction accuracy was improved by approximately 15% while maintaining computational efficiency, and the computation speed was about 8 times faster than the traditional finite element method.

[0129] Through the above technical solution, the present invention achieves the effective integration of microscopic material combustion dynamics model and macroscopic mass transport process, and constructs a high-precision multi-scale diffusion prediction model, which can provide technical support for material combustion safety assessment and fire protection design.

[0130] Figure 4 Here is a flowchart of the material combustion and diffusion prediction process based on dual feature fusion, as described in this embodiment of the invention:

[0131] This diagram illustrates a complete process for predicting material combustion and diffusion, comprising three key steps. First, the system divides a high-precision 3D distribution map into a grid structure and measures the physical parameters of the grid nodes. These parameters are then input into a material combustion kinetics model to calculate the node reaction rates. Simultaneously, the system constructs node feature vectors and edge feature vectors based on the relationships between adjacent nodes, laying the foundation for subsequent feature fusion. Second, in the feature processing stage, the system designs an innovative dual-feature fusion network structure, using two independent weight matrices to map and transform the node and edge feature vectors. Through activation function processing, the system generates a micro-to-macro feature mapping matrix and uses it to calculate the dynamic evolution trend of node states. Finally, the system inputs the obtained evolution trend information into a multi-level prediction network to calculate the prediction errors for local reaction rates and the overall transport field. When the prediction error is below a pre-set threshold, the system outputs the final material combustion and diffusion prediction model. This progressive processing flow ensures the accuracy and reliability of the prediction results.

[0132] In one optional implementation, based on the output of the multi-scale diffusion prediction model, predictive data of smoke density distribution and diffusion path at future times within the carriage are generated to guide fire evacuation and ventilation control in rail transit carriages, including:

[0133] Determine the temporal and spatial characteristics of the output of the multi-scale diffusion prediction model, and fuse the temporal and spatial characteristics to obtain spatiotemporal enhanced features;

[0134] The spatiotemporal enhancement features are input into a causal convolutional network, and temporal features are extracted using dilated convolution operations. The temporal features are then passed between layers using a residual connection structure. The temporal correlation coefficient of the passed features is calculated, and the optimal number of network layers and dilation rate are determined based on the temporal correlation coefficient to generate a feature sequence with long-range dependencies.

[0135] A dual-branch prediction network is constructed based on the feature sequence. The density prediction branch generates smoke density distribution prediction data for future times inside the carriage, and the path prediction branch generates diffusion path prediction data for future times. The prediction error between the smoke density distribution prediction data and the diffusion path prediction data is calculated, and the weight coefficients of the prediction branches are dynamically adjusted according to the prediction error.

[0136] The reliability of the predicted data after weight adjustment is evaluated using a probabilistic prediction model. The confidence interval and calibration error of the reliability of the predicted data are calculated. The parameters of the probabilistic prediction model are iteratively optimized based on the confidence interval and the calibration error, and a prediction result with guaranteed reliability is output.

[0137] The safety level of each area of ​​the carriage is calculated based on the prediction results. The optimal evacuation route is planned based on the safety level. The smoke concentration deviation of each area is calculated using the smoke density distribution prediction data. Ventilation control instructions are generated based on the smoke concentration deviation.

[0138] The output of the multi-scale diffusion prediction model undergoes spatiotemporal feature fusion processing. Specifically, a three-dimensional data cube is extracted from the model output, containing a time dimension T (e.g., 120 time steps), a spatial location dimension X×Y (e.g., 64×64 grid points), and a feature dimension C (e.g., 32 feature channels). For temporal feature extraction, a one-dimensional convolutional sliding window (window size 7) is used to extract historical evolution patterns along the time axis, resulting in a temporal embedding vector T_embed (64 dimensions). For spatial feature processing, a grid encoder is used to encode the location of each point on the XY plane, generating a spatial location embedding vector S_embed (128 dimensions). T_embed and S_embed are then fused using an attention mechanism, and the correlation matrix between temporal and spatial features is calculated to obtain the fused spatiotemporally enhanced feature TS_fusion (192 dimensions).

[0139] Spatiotemporal enhancement features are input into a causal convolutional network for temporal feature extraction. This network employs a multi-layer dilated convolutional structure, with dilation rates of 1, 2, 4, 8, and 16 from the first layer onwards. Each layer has a kernel size of 3, 192 input channels, and 128 output channels. Through dilated convolution, the network achieves an effective receptive field of 95 time steps, covering 79.2% of historical data. The network utilizes a residual connection structure, with skip connections between every two layers to ensure that lower-level features are directly transmitted to higher layers. By calculating the temporal autocorrelation coefficients of the output features from different network layers (average 0.783), the optimal network depth was determined to be 5 layers, with a maximum dilation rate of 16. This structure can capture long-range dependencies in the smoke diffusion process, generating a feature sequence (Feat_seq) of length 32.

[0140] A dual-branch prediction network was constructed based on feature sequences, used for density prediction and path prediction respectively. The density prediction branch consists of three transposed convolutional layers with a kernel size of 4×4, a stride of 2, and channels of 128, 64, and 1 respectively, resulting in an output dimension of 64×64×32 (representing the density prediction for the next 32 time steps). The path prediction branch uses a graph convolutional network structure, dividing the carriage space into 48 key nodes, defining node connections through an adjacency matrix, and has an output dimension of 48×32×2 (number of nodes × time steps × direction vector). Both branches initially have weights of 0.5, dynamically adjusted during training based on their respective prediction errors. When the root mean square error of the density prediction branch on the validation set is 0.067 and the average deviation of the path prediction branch is 0.083 meters, the weights are adjusted to 0.58 and 0.42 respectively.

[0141] To evaluate the reliability of the prediction results, a probabilistic prediction model was implemented. This model, based on an ensemble learning method, trained 10 sub-models with identical structures but different initializations. For each prediction point, the mean of the predictions from the 10 sub-models was calculated as the final prediction value, and the standard deviation was used as an uncertainty estimate. Statistical validation showed that the 95% confidence interval coverage reached 92.7%, and the average calibration error was 0.031. To address the calibration error, a temperature scaling method was used to adjust the prediction distribution, and the temperature parameter was iteratively optimized (final value 1.28), reducing the calibration error to 0.013 and improving the reliability of the prediction results.

[0142] Safety evacuation and ventilation control decisions are made based on the forecast results. The space inside the carriage is divided into 12 zones. The safety level (level 1-5) is determined by calculating the average smoke density and maximum density change rate of each zone over the next 180 seconds. When it is predicted that the smoke density in a certain zone will exceed 0.15 kg / m³ and the change rate will be greater than 0.003 kg / (m³·s) within 60 seconds, the safety level of that zone is marked as level 4 (high risk).

[0143] Based on the safety levels of each area, an improved Dijkstra algorithm is used to plan the optimal evacuation route, guiding personnel from high-risk areas to areas with a safety level of 1. This route avoids nodes covered by the predicted main smoke diffusion path. Simultaneously, the deviation of smoke concentration in each area from the target value (0.05 kg / m³) is calculated. When the deviation exceeds 0.1 kg / m³ in a certain area, the system generates a control command to increase the exhaust volume in that area by 30%; when the deviation is negative and the absolute value is greater than 0.03 kg / m³, the exhaust volume is reduced by 15%, achieving precise ventilation control.

[0144] This technical solution enables the system to predict the spread of smoke inside the carriage 3 minutes in advance, with an accuracy rate of 89.5%. It provides precise guidance for personnel evacuation and smoke control in the event of a fire in a rail transit carriage, effectively improving emergency response capabilities.

[0145] In one optional implementation, the reliability of the weighted prediction data is evaluated using a probabilistic prediction model. A confidence interval and calibration error for the reliability of the prediction data are calculated. Based on the confidence interval and calibration error, the parameters of the probabilistic prediction model are iteratively optimized to output a prediction result with guaranteed reliability, including:

[0146] Construct a weight matrix for the prediction parameter association network, perform association analysis on the prediction parameters based on the prediction parameter association network and the weight matrix, set the error threshold between the predicted value and the true value as a reliability constraint, select the optimal parameter combination according to the weight matrix and the reliability constraint, calculate the probability deviation between the predicted value and the true value based on the optimal parameter combination, and establish the reliability probability distribution of the prediction result.

[0147] A dynamic confidence interval is constructed based on the reliability probability distribution and the weight matrix. A calibration error calculation model is constructed using the dynamic confidence interval and the reliability probability distribution. The uncertainty state of the prediction system is input into the calibration error calculation model to obtain the dynamic calibration error in the prediction process.

[0148] A knowledge modulation matrix is ​​constructed using the weight matrix. The dynamic calibration error is weighted using the knowledge modulation matrix. An adaptive learning rate is calculated based on the weighted dynamic calibration error. The parameters of the prediction model are iteratively optimized to obtain the optimal prediction parameters. Based on the optimal prediction parameters and the weight matrix, a prediction calculation is performed to output a prediction result with guaranteed reliability.

[0149] A weight matrix is ​​constructed to represent the degree of mutual influence between different forecast parameters in the network of forecast parameter associations. For example, in meteorological forecasting, there are complex correlations between parameters such as temperature, humidity, and air pressure. A 10×10 weight matrix can be constructed to quantify the strength of these correlations, with weight values ​​ranging from [-1, 1], where 0.8 represents a strong positive correlation and -0.6 represents a moderate negative correlation. Based on this weight matrix, correlation analysis is performed on the forecast parameters to identify key parameter combinations.

[0150] An error threshold between the predicted and actual values ​​is set as a reliability constraint; for example, an acceptable range for predicted temperature error is set to no more than ±1.5℃. Optimal parameter combinations are selected based on the weight matrix and reliability constraints. For instance, 15 combinations meeting the error threshold requirement are selected from 100 candidate combinations, and the combination with the highest total weight is chosen as the optimal choice. The probability deviation between the predicted and actual values ​​is calculated based on the optimal parameter combination. By analyzing 1000 historical prediction results, the distribution characteristics of the prediction deviation are obtained, such as a mean of 0.2 and a standard deviation of 0.8. A reliability probability distribution of the prediction results is established, forming the basis for the reliability assessment of the prediction results.

[0151] Dynamic confidence intervals are constructed based on the reliability probability distribution and weight matrix. Unlike fixed 95% confidence intervals, dynamic confidence intervals adjust their range according to the prediction conditions. For example, when the predicted data is stable, the confidence interval may be [predicted value -0.5, predicted value +0.5]; while when the data fluctuates greatly, it may expand to [predicted value -1.2, predicted value +1.2].

[0152] A calibration error calculation model is constructed using dynamic confidence intervals and reliability probability distributions. The core of this model is to evaluate the consistency between predicted probabilities and actual occurrence frequencies. Uncertainties in the predicted system are input into the calibration error calculation model; these uncertainties may include data acquisition errors, parameter fluctuation ranges, etc. The system automatically quantifies these uncertainties based on historical data. For example, when a 20% decrease in sensor accuracy is detected, the model incorporates this uncertainty into the calculation. In this way, the dynamic calibration error in the prediction process is obtained; for instance, the calibration error is 0.05 under standard operating conditions, but may rise to 0.15 under extreme conditions.

[0153] A knowledge modulation matrix is ​​constructed using a weighted matrix. This matrix integrates domain expert knowledge with historical data analysis results, assigning different reliability weights to different parameters. For example, a 5×5 knowledge modulation matrix is ​​constructed, assigning a high weight of 0.9 to temperature prediction and a medium weight of 0.6 to wind speed, which is more difficult to predict. The knowledge modulation matrix is ​​used to weight the dynamic calibration error, making the calibration process more focused on the impact of high-reliability parameters. An adaptive learning rate is calculated based on the weighted dynamic calibration error. When the calibration error is large, the learning rate may be set to 0.01 for large adjustments; when the calibration error is small, the learning rate may be reduced to 0.001 for fine adjustments.

[0154] The system iteratively optimizes the parameters of the prediction model. After each prediction, it updates the model parameters using a calculated adaptive learning rate based on the difference between the actual and predicted results. For example, the temperature coefficient in the model might be adjusted from an initial value of 1.2 to 1.15, and the humidity influence factor from 0.8 to 0.85. After 200 iterations, the model parameters tend to stabilize, yielding the optimal prediction parameters. Based on the optimal prediction parameters and weight matrix, the system performs prediction calculations, integrating the optimized parameter and weight relationships to form the final prediction result. The output provides a reliable prediction result, including not only the predicted value but also the corresponding reliability assessment, such as "The predicted temperature is 25.2℃, the reliability is 92%, and the confidence interval is [24.5℃, 25.9℃]".

[0155] To verify the effectiveness of this method, industrial equipment failure prediction was used as an example. 500 days of operational data from a production line, including 20 monitoring parameters, were collected. When using traditional prediction methods, the failure prediction accuracy was 78%, with an average early warning time of 6 hours. After adopting this method, the failure prediction accuracy increased to 92%, the average early warning time was extended to 12 hours, and the false alarm rate decreased from 15% to 4%. Particularly in cases of sudden changes in equipment operating conditions, this method demonstrated stronger adaptability and reliability, successfully predicting three early failure symptoms that traditional methods failed to identify.

[0156] Analyzing 1000 prediction results, the KL divergence between the predicted and actual distributions of this method is 0.08, significantly lower than the 0.25 of the traditional method. This demonstrates that this method provides more reliable uncertainty estimates while maintaining prediction accuracy. Furthermore, the dynamic confidence interval of this method covers 96% of the actual results, superior to the 89% coverage of the traditional fixed confidence interval, reflecting a more accurate uncertainty quantification capability.

[0157] Figure 5 This is a schematic diagram comparing the prediction accuracy under different training sample sizes in an embodiment of the present invention:

[0158] This graph compares the prediction accuracy of three different prediction models (our proposed solution, the LSTM model, and the GRU model) with different training sample sizes. The horizontal axis represents the number of training samples, increasing progressively from 1000 to 10000; the vertical axis represents the prediction accuracy, ranging from 0.75 to 0.95. The graph clearly shows that our proposed solution (triangle markers) consistently outperforms the other two models, achieving an accuracy of approximately 0.83 with 1000 samples, and steadily improving to approximately 0.95 as the sample size increases to 10000. In contrast, the LSTM model (circle markers) and the GRU model (square markers) have initial accuracies of approximately 0.78 and 0.79, respectively, and final accuracies of approximately 0.87 and 0.89, respectively. All three curves show an increasing trend with increasing sample size, but the rate of increase gradually slows down, especially after the sample size exceeds 6000, where the accuracy improvement tends to plateau. This technical solution maintains a significant performance advantage across various sample sizes and exhibits better learning efficiency, demonstrating its stronger modeling ability and generalization performance in prediction tasks.

[0159] A second aspect of the present invention provides a dynamic monitoring system for the smoke density of combustion in rail transit materials, comprising:

[0160] The first unit is used to acquire smoke concentration data collected by multiple smoke concentration sensors in the rail transit material combustion test chamber. The multiple smoke concentration sensors are distributed in a three-dimensional grid pattern in the rail transit material combustion test chamber.

[0161] The second unit is used to reconstruct the smoke density distribution in the rail transit material combustion test chamber in three dimensions based on the position coordinate information of the multiple smoke concentration sensors and the corresponding smoke concentration data, using an adaptive mesh refinement three-dimensional spatial interpolation algorithm. When the smoke concentration data is higher than a preset concentration threshold, the mesh density is automatically increased; when the smoke concentration data is lower than the preset concentration threshold, the mesh density is automatically decreased to obtain a high-precision three-dimensional distribution map.

[0162] The third unit is used to collect material combustion characteristic parameters inside the carriage, including the material's heat release rate, mass loss rate, and combustion product composition data; and to establish a material combustion kinetic model based on the aforementioned material combustion characteristic parameters.

[0163] The fourth unit is used to deeply integrate the high-precision three-dimensional distribution map with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural networks. The multi-scale diffusion prediction model includes a macro-diffusion prediction layer and a micro-diffusion prediction layer.

[0164] The fifth unit is used to generate smoke density distribution prediction data and diffusion path prediction data for future moments inside the carriage based on the output results of the multi-scale diffusion prediction model, so as to guide the fire evacuation and ventilation control of the rail transit carriage.

[0165] A third aspect of the present invention provides an electronic device, comprising:

[0166] processor;

[0167] Memory used to store processor-executable instructions;

[0168] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0169] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0170] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring of smoke density from combustion of rail transit materials, characterized in that, include: Smoke concentration data are acquired from multiple smoke concentration sensors inside the rail transit material combustion test chamber, wherein the multiple smoke concentration sensors are distributed in a three-dimensional grid pattern inside the rail transit material combustion test chamber; Based on the position coordinate information of the multiple smoke concentration sensors and the corresponding smoke concentration data, an adaptive mesh refinement three-dimensional spatial interpolation algorithm is used to reconstruct the smoke density distribution in the rail transit material combustion test chamber in three dimensions. When the smoke concentration data is higher than the preset concentration threshold, the mesh density is automatically increased; when the smoke concentration data is lower than the preset concentration threshold, the mesh density is automatically reduced to obtain a high-precision three-dimensional distribution map. Collect material combustion characteristic parameters inside the carriage, including the material's heat release rate, mass loss rate, and combustion product composition data; A material combustion dynamics model is established based on the aforementioned material combustion characteristic parameters, including: Thermogravimetric data of the material are collected at different heating rates. The mass loss rate at each temperature point is calculated based on the thermogravimetric data. The peak temperature point is determined according to the mass loss rate. The peak temperature point is used as the characteristic temperature of material decomposition. The combustion process of the material is divided into multiple sub-reaction stages based on the characteristic temperature. For each of the sub-reaction stages, multiple characteristic conversion rate points and their corresponding temperature data are selected. The mass loss degree of each stage is calculated based on the characteristic conversion rate points. The mass loss degree, the temperature data, and the heating rate are substituted into the Flynn-Woll-Ozawa equation. Logarithmic transformation is performed on the Flynn-Woll-Ozawa equation to transform the nonlinear equation into a linear relationship between activation energy and frequency factor, establishing a set of parametric equations. The initial kinetic parameters of each sub-reaction are obtained by solving the set of parametric equations using the least squares method. A parameter optimization space is constructed based on the initial dynamic parameters. The parameter optimization space is then input into a genetic algorithm optimization model, with the root mean square error between the experimental curve and the calculated curve as the optimization objective. Based on the optimization objective, the genetic algorithm optimization model is used to perform parameter optimization calculations through selection, crossover, and mutation operations to obtain the optimal activation energy, frequency factor, and reaction order. Based on the optimal activation energy, the frequency factor, and the reaction order, a material combustion kinetics model is established. The high-precision three-dimensional distribution map is deeply integrated with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural networks, including: The high-precision three-dimensional distribution map is divided into a grid structure, the physical parameters of the nodes are measured, the physical parameters are substituted into the material combustion kinetics model to calculate the node reaction rate, the node feature vector is constructed, and the edge feature vector is constructed based on the transmission parameters calculated from adjacent nodes. Design a dual-feature fusion network, using a first weight matrix and a second weight matrix to map the node feature vector and the edge feature vector respectively, and obtain a micro-macro feature mapping matrix through an activation function, and calculate the evolution trend of the node state based on the micro-macro feature mapping matrix; The evolution trend is input into a multi-level prediction network to calculate the prediction error between the local reaction rate and the overall transport field. When the prediction error is less than a preset error threshold, the material combustion diffusion prediction model is output. The multi-scale diffusion prediction model includes a macro-diffusion prediction layer and a micro-diffusion prediction layer. Based on the output of the multi-scale diffusion prediction model, predictive data on smoke density distribution and diffusion path within the carriage at future times are generated to guide fire evacuation and ventilation control in rail transit carriages, including: Determine the temporal and spatial characteristics of the output of the multi-scale diffusion prediction model, and fuse the temporal and spatial characteristics to obtain spatiotemporal enhanced features; The spatiotemporal enhancement features are input into a causal convolutional network, and temporal features are extracted using dilated convolution operations. The temporal features are then passed between layers using a residual connection structure. The temporal correlation coefficient of the passed features is calculated, and the optimal number of network layers and dilation rate are determined based on the temporal correlation coefficient to generate a feature sequence with long-range dependencies. A dual-branch prediction network is constructed based on the feature sequence. The density prediction branch generates smoke density distribution prediction data for future times inside the carriage, and the path prediction branch generates diffusion path prediction data for future times. The prediction error between the smoke density distribution prediction data and the diffusion path prediction data is calculated, and the weight coefficients of the prediction branches are dynamically adjusted according to the prediction error. The reliability of the predicted data after weight adjustment is evaluated using a probabilistic prediction model. The confidence interval and calibration error of the reliability of the predicted data are calculated. The parameters of the probabilistic prediction model are iteratively optimized based on the confidence interval and the calibration error, and a prediction result with guaranteed reliability is output. The safety level of each area of ​​the carriage is calculated based on the prediction results. The optimal evacuation route is planned based on the safety level. The smoke concentration deviation of each area is calculated using the smoke density distribution prediction data. Ventilation control instructions are generated based on the smoke concentration deviation.

2. The method according to claim 1, characterized in that, The three-dimensional reconstruction of the smoke density distribution in the combustion test chamber of the rail transit materials using an adaptive mesh refinement three-dimensional spatial interpolation algorithm includes: Smoke density data collected by multiple smoke concentration sensors in a rail transit material combustion test chamber are obtained. The multiple smoke concentration sensors are distributed in three dimensions in the rail transit material combustion test chamber. The position information of the multiple smoke concentration sensors is mapped to a three-dimensional coordinate system. Calculate the spatial gradient value of the smoke density in the combustion test chamber of the rail transit material, and multiply the ratio of the spatial gradient value to the maximum smoke density at the current moment by the initial grid size to obtain the grid refinement criterion; The spatial grid of the rail transit material combustion test chamber is adaptively refined according to the aforementioned grid refinement criterion. A three-dimensional spatial interpolation model is established based on the adaptively refined mesh structure. The interpolation weight coefficient is determined according to the local mesh characteristics in the rail transit material combustion test chamber. The interpolation weight coefficient is composed of the confidence coefficient of the smoke concentration sensor and the distance attenuation function based on the local mesh size. The interpolation weight coefficients are input into the three-dimensional spatial interpolation model to calculate the smoke density values ​​of the grid nodes in the rail transit material combustion test chamber, thereby obtaining the three-dimensional reconstruction results of the smoke density distribution in the rail transit material combustion test chamber.

3. The method according to claim 1, characterized in that, A logarithmic transformation is performed on the Flynn-Woll-Ozawa equations to convert the nonlinear equations into a linear relationship between activation energy and frequency factor, establishing a set of parametric equations. The initial kinetic parameters for each sub-reaction are obtained by solving the set of parametric equations using the least squares method: Obtain the temperature reaction rate in the Flynn-Wohl-Ozawa equation, which includes reaction conversion, frequency factor, activation energy, and reaction order. The Flynn-Wohl-Ozawa equation characterizes the exponential function relationship between the temperature reaction rate and the reaction conversion, frequency factor, activation energy, and reaction order. The exponential function relationship is subjected to a logarithmic transformation to obtain a linearized equation; A set of parametric equations is established based on the linearized equations. A target data sequence is constructed by combining multiple sets of logarithmic reaction rate values. A coefficient data sequence is constructed by combining the corresponding reciprocal temperature values ​​with the logarithmic conversion rate values. The set of parametric equations characterizes the linear relationship between the target data sequence and the coefficient data sequence. The least squares method is used to solve the parametric equations. The sum of squared errors between the experimental and fitted values ​​of the target data sequence is used as the optimization objective function. The optimal estimated value is obtained by solving for the parameter sequence that minimizes the optimization objective function. The optimal estimated values ​​are then used to recover the kinetic parameters, yielding the initial kinetic parameters for each sub-reaction.

4. The method according to claim 1, characterized in that, The reliability of the weighted prediction data is evaluated using a probabilistic prediction model. The confidence interval and calibration error of the predicted data reliability are calculated. Based on the confidence interval and calibration error, the parameters of the probabilistic prediction model are iteratively optimized, and the reliable prediction results are output, including: Construct a weight matrix for the prediction parameter association network, perform association analysis on the prediction parameters based on the prediction parameter association network and the weight matrix, set the error threshold between the predicted value and the true value as a reliability constraint, select the optimal parameter combination according to the weight matrix and the reliability constraint, calculate the probability deviation between the predicted value and the true value based on the optimal parameter combination, and establish the reliability probability distribution of the prediction result. A dynamic confidence interval is constructed based on the reliability probability distribution and the weight matrix. A calibration error calculation model is constructed using the dynamic confidence interval and the reliability probability distribution. The uncertainty state of the prediction system is input into the calibration error calculation model to obtain the dynamic calibration error in the prediction process. A knowledge modulation matrix is ​​constructed using the weight matrix, and the dynamic calibration error is weighted using the knowledge modulation matrix. An adaptive learning rate is calculated based on the weighted dynamic calibration error, and the parameters of the prediction model are iteratively optimized to obtain the optimal prediction parameters. Based on the optimal prediction parameters and the weight matrix, prediction calculations are performed, and a prediction result with guaranteed reliability is output.

5. A dynamic monitoring system for combustion smoke density of rail transit materials, used to implement the method described in any one of claims 1-4, characterized in that, include: The first unit is used to acquire smoke concentration data collected by multiple smoke concentration sensors in the rail transit material combustion test chamber. The multiple smoke concentration sensors are distributed in a three-dimensional grid pattern in the rail transit material combustion test chamber. The second unit is used to reconstruct the smoke density distribution in the rail transit material combustion test chamber in three dimensions based on the position coordinate information of the multiple smoke concentration sensors and the corresponding smoke concentration data, using an adaptive mesh refinement three-dimensional spatial interpolation algorithm. When the smoke concentration data is higher than a preset concentration threshold, the mesh density is automatically increased; when the smoke concentration data is lower than the preset concentration threshold, the mesh density is automatically decreased to obtain a high-precision three-dimensional distribution map. The third unit is used to collect material combustion characteristic parameters inside the carriage, including the material's heat release rate, mass loss rate, and combustion product composition data. A material combustion dynamics model is established based on the material combustion characteristic parameters; The fourth unit is used to deeply integrate the high-precision three-dimensional distribution map with the material combustion dynamics model to construct a multi-scale diffusion prediction model based on graph neural networks. The multi-scale diffusion prediction model includes a macro-diffusion prediction layer and a micro-diffusion prediction layer. The fifth unit is used to generate smoke density distribution prediction data and diffusion path prediction data for future moments inside the carriage based on the output results of the multi-scale diffusion prediction model, so as to guide the fire evacuation and ventilation control of the rail transit carriage.

6. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images