A test method for simulating high altitude forest fire spread

Abstract

The application discloses a test method for simulating high-altitude forest fire spreading, and belongs to the technical field of forest fire prevention and disaster prevention and control. In view of the problem of insufficient simulation precision of the coupling scene of high-altitude hypoxia, low temperature, steep terrain and strong wind in the prior art, the application realizes precise simulation of high-altitude forest fire spreading behavior by means of single-factor decoupling and multi-parameter coupling experiment, and by regulating and controlling core parameters such as oxygen concentration, environmental temperature, slope, wind speed and wind direction, in combination with distributed multi-physical field data acquisition, deep neural network modeling and progressive closed-loop feedback control, so as to provide scientific decision support for high-altitude forest fire prevention and control.

Description

Technical Field

[0001] This invention relates to the field of forest fire prevention and disaster control technology, specifically to an experimental method for simulating the spread of forest fires at high altitudes. Background Technology

[0002] Forest fires are a type of natural disaster characterized by their suddenness, destructiveness, and difficulty in handling and responding to them. Forest fires in high-altitude mountainous areas, in particular, exhibit unique combustion behaviors, unpredictable spread patterns, and great difficulty in firefighting due to the extremely complex environment of low oxygen, low temperature, steep terrain, and strong convective winds. In-depth research on the spread mechanism of forest fires in high-altitude environments has important theoretical and practical significance for fire prevention and emergency rescue decision-making in high-altitude forest areas.

[0003] Currently, domestic and international research on the spread of forest fires mainly falls into two mainstream categories:

[0004] The first category consists of real-time simulation schemes based on remote sensing and ground data, such as the invention patent with publication number CN115099073A. This scheme acquires fire, geographical, fuel, and climate information of the target forest through remote sensing satellites and ground communication equipment, and combines meteorological, topographical, and fuel factors to achieve numerical simulation of forest fire spread. However, this scheme can only achieve numerical extrapolation of forest fires under normal altitude and atmospheric conditions. It completely fails to consider the core impacts of low oxygen partial pressure and extreme low temperature in high-altitude areas on the combustion characteristics of combustibles. It lacks controllable physical experimental support, cannot quantify the independent effects and coupling mechanisms of various environmental parameters on forest fire spread, and the simulation results rely on historical statistical data, resulting in extremely poor prediction accuracy for special high-altitude scenarios.

[0005] The second category is forest fire spread simulation schemes based on cellular automata, such as the invention patent with publication number CN114781169A. This scheme achieves visualized simulation of forest fire spread by dividing the cellular space, designing multi-state transition rules, and combining parameters such as wind speed, temperature, slope, and combustible material type. However, this scheme is also a pure numerical simulation method, which is only applicable to conventional altitude scenarios. It completely ignores the decisive influence of high-altitude low-oxygen and low-temperature environments on forest fire spread, cannot reproduce the coupling effect of wind field and slope topography in high-altitude mountains, and lacks measured physical experimental data to calibrate the model. The model's generalization ability and practical applicability are seriously insufficient.

[0006] In summary, all existing technologies have core shortcomings: ① They are all purely numerical simulation / simulation schemes, lacking controllable physical experimental methods for the spread of forest fires at high altitudes, and cannot obtain measured physical data to support mechanism research and model calibration; ② They cannot achieve decoupling experiments for single influencing factors and coupling experiments for multiple factors, and cannot quantify the independent contribution and synergistic effect of each parameter on the spread of forest fires; ③ They are insufficient in simulating the coupling effect of high-altitude mountain slopes and wind fields, and the simulation results deviate significantly from the actual fire scene. Summary of the Invention

[0007] The purpose of this invention is to provide an experimental method for simulating the spread of forest fires at high altitudes, aiming to optimize the fire spread model, improve simulation accuracy, and provide more scientific and reliable decision support for forest fire prevention and control in high-altitude areas.

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

[0009] An experimental method for simulating the spread of forest fires at high altitudes includes the following steps:

[0010] (a) Set initial experimental conditions, including combustible material distribution parameters, terrain slope control range, environmental gas component ratio, environmental temperature, environmental humidity, and air pressure baseline value; among which, the oxygen concentration in the environmental gas components is adjusted according to the target high-altitude environmental parameters, and the low oxygen partial pressure environment corresponding to an altitude of 2000m to 5500m is equivalently simulated by reducing the oxygen volume fraction (the actual oxygen volume fraction in the atmosphere is maintained at about 21%, and the low oxygen partial pressure is equivalently achieved by reducing the total pressure or volume fraction; this method adopts the method of reducing the volume fraction).

[0011] (b) Oxygen concentration regulation experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant, and the oxygen concentration was dynamically regulated to record the flame combustion and spread characteristic parameters under different oxygen concentrations; then, a multi-parameter coupling experiment was carried out, and the oxygen concentration was orthogonally combined with the parameters of ambient temperature, slope and wind speed to quantify the influence weight of low oxygen environment on fire spread.

[0012] (c) Environmental temperature control experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant, and the environmental temperature was dynamically controlled to record the flame combustion and spread characteristic parameters at different temperatures; then, a multi-parameter coupling experiment was carried out to quantify the coupling mechanism between the low-temperature environment and the other parameters.

[0013] (d) Slope control experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant, and the slope was controlled within the range of 0° to 60°. The dynamic characteristics of flame propagation along the slope were recorded under different slopes. Then, a multi-parameter coupling experiment was carried out, combining all factors of slope and wind speed and direction parameters to quantify the synergistic effect of slope-wind field coupling on fire spread.

[0014] (e) Wind speed and direction control experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant. The wind speed was controlled in the range of 0 to 30 m / s and the wind direction was controlled in the range of 0° to 360°. The fire spread characteristics under different wind field conditions were recorded. Then, a multi-parameter coupling experiment was carried out. A four-factor orthogonal experiment of wind speed-wind direction-slope-oxygen concentration was carried out to quantify the critical conditions for fire spread under multi-parameter coupling at high altitude.

[0015] (f) Using time series data acquisition technology, monitor and record in real time the flame front position, temperature field distribution, flue gas concentration, heat release rate, flame flow field characteristics, and free radical concentration changes during the above experiments.

[0016] Furthermore, in the single-factor controlled variable experiment in step (b), a single closed-loop control system was used to regulate the oxygen concentration, with each adjustment interval being a gradual decrease of 2%, and the oxygen regulation accuracy being ±0.5%. A high-speed photography system was used to observe the flame propagation behavior at a sampling rate of 1000 frames / second, and the change in flame free radical concentration was simultaneously determined by spectral analysis.

[0017] Furthermore, in the single-factor controlled variable experiment in step (c), an environmental cavity temperature field uniformity calibration experiment is first carried out to ensure that the temperature field uniformity of the experimental area is ≤±1℃; the ambient temperature is adjusted at a cooling rate of 5℃ / min, with a temperature control accuracy of ±0.5℃ and an adjustment range of -20℃~30℃. A test node is set for every 10℃ decrease. After the temperature of each node stabilizes for 30 minutes, no less than 3 parallel combustion experiments are carried out, and the average value is taken as the valid data.

[0018] Furthermore, in the single-factor controlled variable experiment of step (d), the levelness of the slope platform is first calibrated to ensure that the flatness error of the slope is ≤0.2mm / m at a slope of 0°; the slope is controlled within the range of 0° to 60° using an adjustable slope experimental platform with a control accuracy of ±0.1°; core gradient nodes are set at 0°, 15°, 30°, 45°, and 60°, and no less than 3 parallel combustion experiments are carried out after each slope node is stabilized; the characteristics of the flame propagation flow field are measured simultaneously using particle image velocimetry.

[0019] Furthermore, the adjustable slope test platform has a rectangular box structure with a detachable grooved combustion bed inside. The combustion bed has an overall width of 1.5m and a length of 4.0m, with ceramic fiber side plates with a height of 10cm on both sides. The bottom plate has evenly distributed ventilation holes, and the surface is covered with stainless steel fine wire mesh with a mesh diameter of 0.9mm and an opening area of ​​85%.

[0020] Furthermore, in the single-factor controlled variable experiment of step (e), a wind field calibration experiment is first carried out. A three-dimensional hot-wire anemometer is used to calibrate the wind field of the experimental slope point by point to ensure that the wind field uniformity of the slope is ≤±5% and the turbulence intensity is ≤5%. An attached airflow field parallel to the slope is generated by a slope-adaptive modular wind field generation system. The system can adjust the pitch angle and rotation direction of the wind outlet synchronously with the slope angle to achieve continuous control of wind speed from 0 to 30 m / s and wind direction from 0° to 360°. Gradient nodes are set for wind speed in 1 m / s increments and full-range scanning nodes are set for wind direction in 15° increments. After each node is stable, no less than 3 parallel combustion experiments are carried out.

[0021] Furthermore, in step (f), a distributed sensor network is used to collect data, including a thermocouple array, a gas sampling probe and a particle image velocimetry system, with a data acquisition frequency of 1kHz; simultaneously, a high-speed camera is used to capture the flame morphology and the position of the fire spread front from multiple angles on the side and rear of the experimental platform, with a shooting frame rate of not less than 500 frames / second.

[0022] Furthermore, the present invention also includes a machine learning optimization step, in which the flame front position, temperature field data, flow field characteristics, flue gas concentration, and heat release rate data collected in step (f) are input into a deep neural network for training; the input parameter matrix of the deep neural network is... ,in Representing the set of real numbers, N is the number of training samples. The 12 input parameters are, in order: oxygen concentration, ambient temperature, slope angle, wind speed, wind direction angle, combustible type, combustible moisture content, ambient humidity, air pressure, combustible density, combustion time, and initial ignition power; the output parameter matrix is... The output results are flame propagation speed, flame front x coordinate, and flame front y coordinate (where x and y are planar coordinates with the ignition point as the origin, used to describe the position of the fire front at the next moment).

[0023] Furthermore, the deep neural network has a structure of 12-dimensional input layer → 5 hidden layers → 3-dimensional output layer, with 128 neurons in each hidden layer and ReLU activation function. Mean squared error is used as the loss function, and Adam optimizer is used for backpropagation weight update. The parameter transfer learning method is used to adapt to 8 typical high-altitude vegetation types, and each target vegetation type has a dedicated dataset of no less than 300 sets during the fine-tuning stage.

[0024] Furthermore, a global dynamic feedback mechanism is introduced during the multi-parameter coupling experiment. The triggering logic of this mechanism is as follows: when the measured value of the flame propagation speed deviates from the model prediction value, the single-factor exclusive closed-loop control corresponding to the dominant factor of the deviation is triggered first; if the deviation is still >10% after single-factor correction, the global dynamic feedback mechanism is triggered, and the oxygen concentration, ambient temperature, wind speed and slope parameters are adjusted in linkage through the PID controller according to the parameter sensitivity weights (the weights of oxygen concentration, slope and wind speed are 0.35, 0.28 and 0.22 respectively, and the remaining parameters are sorted according to actual sensitivity) until the deviation is stably controlled within 10%.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) This invention constructs a complete experimental method for the spread of high-altitude forest fires, consisting of "single-factor decoupling experiment - multi-parameter coupling experiment - multi-physical field synchronous acquisition". It can effectively simulate the physical spread of forest fires in the entire range of 2000m to 5500m at high altitudes, overcoming the limitation of existing technologies that can only conduct conventional altitude numerical simulations. By systematically covering the four core influencing factors of oxygen concentration, ambient temperature, slope, wind speed and direction, this invention not only quantifies the independent effect of a single parameter, but also reveals the synergistic coupling mechanism of multiple parameters. It fills the industry gap in physical experimental methods for high-altitude forest fires, provides a feasible experimental means and high-fidelity real-measured data support for the prevention and control of high-altitude forest fires, and solves the core problems of existing pure numerical simulations that lack real-measured data support and have large prediction deviations.

[0027] (2) This invention utilizes a wide-range, high-precision programmable temperature control system (-20℃ to 30℃) combined with temperature field uniformity calibration and multiple parallel repeated experiments to systematically quantify the combustion inhibition mechanism of low temperature. Experimental data show that for every 10℃ decrease in ambient temperature, the flame propagation speed decreases by 15% to 20%, corresponding to a 25% decrease in the combustion reaction rate constant, while the activation energy remains constant. This effectively reproduces the combustion inhibition effect in extreme low-temperature environments at high altitudes, solves the prediction bias problem caused by neglecting low temperatures in traditional models, and ensures that the prediction error of the propagation speed under low-temperature conditions is ≤8%.

[0028] (3) This invention uses a high-precision adjustable slope test platform of 0° to 60° and a grooved combustion bed structure, combined with particle image velocimetry (PIV) technology, to realize the in-situ measurement of flame flow field characteristics in the entire slope range. This can reveal the physical mechanism of the positive correlation between buoyancy-induced pressure gradient and slope, and improve the accuracy of fire path prediction in complex terrain.

[0029] (4) This invention adopts a slope-adaptive modular wind field generation system. Through an integrated design that tilts and rotates synchronously with the slope, it achieves precise control of wind speeds from 0 to 30 m / s and wind directions from 0° to 360°, completely solving the problem of mismatch between traditional horizontal wind fields and inclined slopes. Experimental verification shows that the flame propagation speed reaches a peak of 2.0 m / s under downhill wind speed of 15 m / s, and the speed decreases by 60% under headwind conditions. The system successfully captures key phenomena such as flame lift and a 60% increase in plume entrainment coefficient caused by high wind speeds (>25 m / s), improving the prediction spatial resolution of extreme wind fields at high altitudes by 3 times and well reproducing the mountain wind field-topography coupling effect.

[0030] (5) This invention constructs a 5-layer deep neural network model with 12 inputs and 3 outputs, introduces a parameter transfer learning framework, and adapts to 8 typical high-altitude vegetation types. The model is trained based on 3000 sets of measured data. The prediction determination coefficient R² for flame propagation speed reaches 0.96, the average Euclidean distance error for predicting the flame front position is less than 0.15m, the overlap rate of the actual fire scene over 72 hours reaches 82%, and the prediction error is consistently below 9%. This reduces the response time of traditional models by 40%, realizes accurate simulation of the "leapfrog spread" behavior at high altitudes, and solves the problems of weak generalization ability and insufficient adaptability to high-altitude scenarios of existing models.

[0031] (6) This invention introduces a progressive dynamic feedback mechanism of "single-factor dedicated closed loop - global multi-parameter linkage". It prioritizes correcting the dominant deviation term through single-factor closed loop. If the correction is ineffective, it links and adjusts the multi-parameter combination to ensure that the parameter error between the experimental conditions and the actual high-altitude environment is controlled within ±2%. This mechanism forms a collaborative closed loop with various control systems, acquisition systems and prediction models, and constructs a full-process adaptive system of "simulation-acquisition-prediction-correction". The overall prediction accuracy and scenario adaptability are significantly better than existing pure numerical simulation technology, providing scientific and reliable decision support for the prevention and control of forest fires at high altitudes. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the overall process of the present invention.

[0033] Figure 2 This is an overall system block diagram of the test system of the present invention.

[0034] Figure 3 This is a front view of the experimental platform in the adjustable slope test platform and wind field generation system.

[0035] Figure 4 This is a side view of the experimental platform in the adjustable slope test platform and wind field generation system.

[0036] The accompanying figure is labeled as follows:

[0037] 1 - Experimental bench, 2 - Air duct, 3 - Reducer, 4 - Flow stabilizer, 5 - Gantry frame, 6 - Honeycomb screen, 7 - Combustion bed, 101 - Right side stainless steel plate, 102 - Left side fireproof glass plate, 103 - Top fireproof glass plate. Detailed Implementation

[0038] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0039] Example

[0040] This embodiment provides an experimental method for simulating the spread of high-altitude forest fires, aiming to address technical problems in existing forest fire simulation experiments, such as insufficient coupling ability between wind field and slope, low accuracy in simulating extreme environmental parameters at high altitudes, and insufficient quantification of multi-factor coupling mechanisms. This embodiment enables overall slope adjustment of the experimental space and provides a stable, uniform, and adjustable airflow environment under varying slope conditions, thus providing a more realistic experimental means for studying the spread mechanism of high-altitude mountain forest fires. The overall process of the experimental method in this embodiment is as follows: Figure 1 As shown, the test system is as follows Figure 2 As shown. Among them, Figure 1 The process is as follows: initial condition setting, single-factor parameter control experiment, multi-parameter coupling control experiment, multi-physics field data synchronous acquisition, data preprocessing and feature extraction, deep neural network model training and optimization, fire spread prediction and result output, and synchronous setting of real-time monitoring data to the closed-loop branch of dynamic feedback control. Figure 2 The test system is based on a closed and controllable environment simulation chamber and integrates seven core functional subsystems: oxygen concentration control subsystem, environmental temperature and humidity control subsystem, adjustable slope test subsystem, slope-adaptive modular wind field subsystem, multi-physics field synchronous acquisition subsystem, dynamic feedback control subsystem, and machine learning simulation prediction subsystem. Each subsystem is connected to the dynamic feedback control subsystem to achieve coordinated control.

[0041] Figure 3 , 4 The structure of the experimental platform in the adjustable slope test platform and wind field generation system is as follows: In this embodiment, the core components and connections of the experimental platform are as follows: the experimental platform 1 is rigidly connected to the air duct 2, the reducing pipe 3, and the flow stabilizing pipe 4; the gantry frame 5 is connected to the bottom plate lifting lugs of the experimental platform 1 via an electric hoist; the honeycomb screen 6 is installed inside the flow stabilizing pipe 4; the combustion bed 7 is fixed to the bottom plate of the experimental platform 1; the right stainless steel plate 101, the left fireproof glass plate 102, and the top fireproof glass plate 103 respectively constitute the side wall and top of the experimental platform 1. Figure 3 , 4 As shown, the following will provide a detailed explanation of the specific structure, connection relationship and complete experimental procedure of each component.

[0042] I. Experimental Preparation Stage

[0043] Before conducting the experiment, a scenario that matches the characteristics of high altitude must first be selected as the research object. The core tests of the experiment are completed in a closed and controllable environment simulation chamber. At the same time, real high-altitude mountainous areas with an altitude of ≥2000 meters can be selected as field verification sites to collect measured parameters of the on-site environment, combustibles, and terrain. The parameters are then reproduced 1:1 in the simulation chamber to ensure that the error between the simulated conditions and the actual high-altitude environmental parameters is ≤±2%. The vegetation types are mainly coniferous forests and shrubs. These types of vegetation are highly flammable and evenly distributed, which can effectively simulate the characteristics of forest fires in high-altitude areas.

[0044] Combustible material distribution parameters in the experimental area were obtained using airborne / ground-based 3D laser scanning technology. The specific implementation process is as follows:

[0045] (1) Selection and parameter settings of scanning equipment: A ground-based 3D laser scanner (range range 0.5~300m, range accuracy ±2mm, point cloud sampling frequency 1 million points / second, scanning angle resolution 0.001°) was used in conjunction with an airborne lidar (point cloud density ≥20 points / m). 2 (with an elevation accuracy of ±5cm), a full-coverage scan of the target experimental area was performed to obtain the original three-dimensional point cloud data;

[0046] (2) Point cloud data preprocessing: Outliers are removed by statistical filtering and noise is removed by radius filtering. The ICP algorithm is combined to achieve multi-site point cloud data registration with a registration error of ≤5mm, and a complete three-dimensional point cloud model of the experimental area is generated.

[0047] (3) Combustible material parameter inversion and extraction: Based on the geometric features and reflection intensity features of point cloud data, combined with the random forest classification algorithm, the classification and identification of combustible and non-combustible materials are realized with a classification accuracy of ≥95%; at the same time, structural parameters such as combustible material type, number of trees, crown width, tree height, branch height, and canopy closure are extracted, and the horizontal density of combustible material (unit: kg / m³) is calculated. 2 ) and vertical stratification density (stratified at 0.5m height intervals, unit: kg / m³) 3 );

[0048] (4) Simultaneous calibration of combustible material moisture content: The drying method was used to sample and calibrate the living branches, leaves and dead matter of typical vegetation in the scanning area. The sampling points were evenly distributed in a grid method, with each 100 m² sample point being a sampler. 2 At least one sampling point was used to establish an inversion model of vegetation reflectance intensity and moisture content, with an inversion accuracy of ±1.5%, and finally, spatial distribution data of combustible moisture content in the whole region was obtained.

[0049] (5) Construction and coupling of real-time environmental parameter correction model: Based on the environmental parameters (oxygen concentration, ambient temperature, ambient humidity, air pressure, wind speed and direction) measured on-site for 48 consecutive hours, a real-time environmental parameter correction model is constructed. The model expression is as follows:

[0050]

[0051] In the formula: The i-th characteristic parameter of the combustible material after correction at time t; These are the initial calibration values ​​for the characteristic parameters of combustible materials; , , , These represent the differences between the ambient temperature, oxygen volume fraction, relative humidity, and air pressure at time t and the initial calibration values. , , , These are the correction coefficients for the corresponding environmental parameters, obtained through pre-experimental calibration. The corrected combustible parameters and environmental parameters are spatiotemporally registered to ensure that the parameter errors between the experimental conditions and the actual high-altitude environment are controlled within ±2%.

[0052] The experimental area was divided into several 10m × 10m grid cells. Based on the combustible parameters obtained from the aforementioned scans, a combustible distribution heatmap for each grid cell was generated and spatially registered with the initial condition database for dynamic modeling in subsequent experiments. Before the experiment began, a portable weather station equipped with differential GPS was used to continuously monitor environmental parameters for 48 hours, including temperature, humidity, air pressure, solar radiation intensity, wind speed, and wind direction. A high-precision gas analyzer was used to measure the oxygen volume fraction in the air at a resolution of 0.1ppm, completing the on-site calibration of the real-time environmental parameter correction model.

[0053] Before the experiment, the combustion bed needs to be set up: a grooved combustion bed with a width of 1.5 m and a length of 4.0 m is selected and installed in the T-slot of the experimental platform base plate using positioning pins, and the fixing bolts are tightened. The selected fuel is evenly spread in the combustion bed, with the fuel layer thickness controlled at 4 cm. After spreading, a fine iron wire mesh with a mesh diameter of 0.9 mm is placed over the fuel surface and fixed to the side plates of the combustion bed with stainless steel pressure strips around the perimeter, ensuring that the wire mesh is flat and in contact with the fuel surface to prevent fuel from slipping during the slope experiment. The combustion bed base plate is made of 304 stainless steel plate with a thickness of 2 mm, and the surface is evenly distributed with vent holes of 3 mm in diameter, with a spacing of 50 mm and a diamond arrangement. The two side plates are made of ceramic fiber plate with a thickness of 10 mm and a thermal conductivity of less than 0.2 W / (m・K), effectively isolating heat transfer.

[0054] Based on the pre-set measurement points, thermocouple trees and heat flow meters were arranged inside the experimental platform: the bases of the thermocouple trees were fixed to the base of the experimental platform using magnetic mounts, and the heat flow meters were fixed in predetermined positions using dedicated brackets. All sensors were connected to the data acquisition system, the signal was checked for normality, and channel calibration was performed.

[0055] II. Oxygen Concentration Regulation Experiment (corresponding to) Figure 1 In the single-factor parameter regulation experiment, through Figure 2 (Implementation of the oxygen concentration regulation subsystem)

[0056] The first step of the experiment was to establish a basic model and fine-tune the oxygen fraction. The oxygen fraction adjustment range was set from 21% to 10%. In the single-factor controlled variable experiment, a single closed-loop control system (inner loop only) was used to achieve precise control of the oxygen concentration. This system consisted of a high-precision two-way gas mixing unit (oxygen and nitrogen range 0–100 SLM, control accuracy ±0.2% FS), a magnetic pressure oxygen sensor (measurement range 0–25% VOL, accuracy ±0.1% VOL, response time ≤100 ms), a PID controller (sampling frequency 10 kHz, control cycle 10 ms), a high-speed electromagnetic switching valve (response time ≤5 ms), a stainless steel mixing tank (with turbulence structure, volume 50 L, pressure ≥1.0 MPa), and an online mass spectrometer (gas component resolution 0.1 ppm, sampling frequency 10 Hz). The inner closed loop used the real-time concentration detection values ​​from the oxygen sensor and the online mass spectrometer as feedback, and adjusted the opening of the high-speed electromagnetic valve through the PID controller to achieve rapid, accurate, and stable oxygen concentration.

[0057] During the experiment, the oxygen fraction was initially set at 21% under standard atmospheric conditions, gradually decreasing to 15% in 2% increments, and then further reduced to 10% using a gradient descent method. After each adjustment of the oxygen fraction, the environmental parameters were stabilized for 5 minutes. Once the oxygen concentration fluctuation was ≤±0.5%, the experiment was initiated. Each oxygen concentration gradient experiment was repeated at least three times, and the average value was used as the valid data. After each adjustment, flame propagation behavior was observed using a high-speed photography system at a sampling rate of 1000 frames per second, and the changes in flame free radical concentration were simultaneously measured using spectral analysis.

[0058] Experimental data shows that when the oxygen fraction decreases from 21% to 15%, the flame propagation speed decreases by an average of about 30%, and the OH radical concentration decreases by 45%. When the oxygen fraction continues to decrease to 10%, the flame propagation speed decreases further, and the decrease is greater (approximately 55% compared to the value at 15%). At this point, the CO / CO2 ratio increases significantly, indicating a decrease in combustion efficiency. This systematic change process was numerically simulated using a multiphysics coupling model. The results show that in a low-oxygen environment, the combustion reaction rate decays exponentially with oxygen concentration, leading to a decrease in flame temperature of 200–350°C, thus significantly affecting the speed and direction of fire spread. In the experiment, an infrared thermal imager equipped with automatic tracking function was used to monitor the flame front position in real time at the sub-millimeter level, and the highest temperature in the flame center region and the temperature gradient changes in the edge region were recorded using a temperature field reconstruction algorithm.

[0059] III. Environmental Temperature Control Experiment (corresponding to) Figure 1 In the single-factor parameter regulation experiment, through Figure 2 (Implementation of the environmental temperature and humidity control subsystem)

[0060] The second step of the experiment was to quantitatively analyze the effect of ambient temperature on combustion characteristics and fire spread. Other initial conditions, such as oxygen concentration, combustible material parameters, slope angle, and wind speed, were kept constant, and only the ambient temperature parameter was changed to eliminate interference from other variables on the experimental results.

[0061] The ambient temperature control range is set from -20℃ to 30℃. A high-precision programmable temperature control system is used to achieve a temperature control accuracy of ±0.5℃. First, a temperature field uniformity calibration experiment is carried out in the ambient cavity to ensure that the temperature field uniformity of the experimental area is ≤±1℃. The ambient temperature is gradually reduced at a cooling rate of 5℃ / min. A test node is set every 10℃ reduction. After the temperature of each node stabilizes for 30 minutes, no less than 3 parallel combustion experiments are carried out, and the average value is taken as the valid data.

[0062] During the experiment, a high-speed photography system was used to record changes in flame propagation speed at a sampling rate of 1000 frames per second. A thermocouple array was used to simultaneously monitor the flame temperature field distribution. Spectroscopic analysis was employed to determine changes in the combustion reaction rate. Simultaneously, a gas sampling probe was used to analyze the CO / CO2 ratio to quantify combustion efficiency. Experimental data showed that for every 10°C decrease in ambient temperature, the flame propagation speed decreased by 15%–20%, corresponding to a 25% decrease in the combustion reaction rate constant, while the activation energy remained constant. Combustion efficiency decreased significantly under low-temperature conditions, and the CO / CO2 ratio increased by 30%–40%. This process was numerically simulated using a multiphysics coupling model, verifying the inhibitory mechanism of low temperature on the combustion reaction and providing fundamental data support for simulating fire spread in high-altitude, low-temperature environments.

[0063] IV. Slope Adjustment Experiment (corresponding to) Figure 1 In the single-factor parameter regulation experiment, through Figure 2 (Implementation of the adjustable slope experiment subsystem)

[0064] The third step of the experiment was to quantitatively analyze the dynamic impact of slope angle on fire spread. Three representative slope angles, 15°, 30°, and 45°, were selected to construct an adjustable slope experimental platform. This platform was a rectangular box structure with effective internal dimensions of 6m long, 2m wide, and 2m high. It consisted of a base plate, a fireproof glass panel on the left, a stainless steel panel on the right, a fireproof glass panel on the top, and a front panel, forming an experimental chamber that was closed at the front and open at the rear. The base plate was made of 12mm thick steel plate, with a rust-proof treatment and a high-temperature resistant coating. The lower surface was welded with reinforcing ribs and four symmetrically arranged lifting lugs. The upper surface had T-slots spaced 100mm apart for fixing the combustion bed. Slope adjustment was achieved through a gantry crane and two synchronously controlled electric hoists. The electric hoists had a rated lifting capacity of 5 tons and a lifting speed of 0.5m / min. They were connected to the base plate lifting lugs via steel wire ropes and, in conjunction with a dual-axis tilt sensor, achieved closed-loop precise adjustment of the slope from 0° to 60°, with an angle control accuracy of ±0.1°.

[0065] During the experiment, all other parameters were kept constant, and only the slope angle was changed to eliminate the interference of other variables on the experimental results. First, the levelness of the slope platform was calibrated to ensure that the flatness error of the slope was ≤0.2mm / m at a slope of 0°. Core gradient nodes were set at 0°, 15°, 30°, 45°, and 60°. After each slope node stabilized, no less than three parallel combustion experiments were carried out, and the flow field characteristics of flame propagation along the slope were measured simultaneously using particle image velocimetry (PIV).

[0066] Experimental results show that under windless conditions, the flame propagation speed along a 15° slope reaches 0.5 m / s, with a turbulence intensity of 12%. In contrast, on a 45° slope, the propagation speed rapidly increases to 1.2 m / s, and the turbulence intensity increases to 35%. This phenomenon, as demonstrated by computational fluid dynamics (CFD) simulations, indicates that with increasing slope, the buoyancy-induced longitudinal pressure gradient strengthens, leading to a more significant effect of gravity driving the flame. The convective heat transfer coefficient of the hot airflow moving upwards along the slope increases by 2-3 times, thereby accelerating the spread of fire in steep terrain. The experiment also used oxygen calorimetry to determine the heat release rate at different slopes. The results show that the heat release rate increases non-linearly from 0° to 45°, with the peak heat release rate at a 45° slope being approximately 2.0 times that at 0°, verifying the promoting effect of steep terrain on fire spread.

[0067] V. Wind speed and direction control experiment (corresponding to) Figure 1 In the single-factor parameter regulation experiment, through Figure 2 (Implementation of the slope-adaptive modular wind farm subsystem)

[0068] The fourth step of the experiment is to study the multi-dimensional coupling effects of wind speed and wind direction angle. A precisely controlled airflow field is generated using a slope-adaptive modular wind field generation system. This system is integrated into a sealed, controllable environment simulation chamber and rigidly connected coaxially to an adjustable slope experimental platform. It can tilt and rotate synchronously with the slope, ensuring that the airflow always flows parallel to the slope at any gradient. The core of the system consists of four parts: a fan unit, an outlet connection channel, a double-layer rectification unit, and a rotation adjustment mechanism.

[0069] Fan unit: 2 to 4 axial flow fans are arranged in parallel. Each fan is equipped with a variable frequency speed controller. The fan speed is continuously adjustable from 0 to 30 m / s through the frequency converter, with a control accuracy of ±0.2 m / s. The fan inlet is equipped with a protective net and inlet guide vanes. The outlet is sealed to the outlet connection channel through a high-temperature resistant silicone flexible connector. The flexible connector is 200 mm long and is used to isolate the mechanical vibration of the fan to avoid the vibration affecting the stability of the flame pattern and the accuracy of data acquisition.

[0070] The air outlet connection channel is rigidly connected along the airflow direction by a duct, a reducer, and a flow stabilizer. The entire structure is airtightly connected to the front panel of the experimental platform via a flange and a high-temperature resistant rubber sealing strip, and can tilt / rotate synchronously with the experimental platform. The duct is a rectangular converging pipe matched to the fan outlet, merging the outlet airflow from multiple fans into a single uniform flow field. The reducer is a rectangular cross-section gradually expanding / contracting pipe with a diffusion angle controlled within 15°, achieving a smooth transition of airflow from the fan outlet section to the experimental platform inlet section, avoiding airflow separation and eddy currents. The flow stabilizer has the same cross-sectional dimensions as the experimental platform inlet section, a length of 1.0m, and is used to stabilize the airflow velocity and provide installation space for the rectifier unit.

[0071] Double-layer rectifier unit: Installed inside the flow stabilizer tube, with a honeycomb screen and a fine stainless steel screen arranged sequentially along the airflow direction; the honeycomb screen is composed of hexagonal honeycomb channels with a side-to-side distance of 20mm and a thickness of 200mm along the airflow direction, used to break up large-scale eddies and eliminate transverse airflow pulsations; the fine stainless steel screen uses 80-mesh 304 stainless steel wire with a mesh diameter of 0.18mm, used to further eliminate residual small-scale turbulence, ultimately achieving an inlet airflow turbulence intensity ≤5% and a wind field uniformity ≤±5% in the experimental area; removable covers are installed at the top and bottom of the flow stabilizer tube for easy disassembly, cleaning, and maintenance of the rectifier unit;

[0072] Rotation adjustment mechanism: The coaxial rotation platform driven by a servo motor can drive the entire wind field generation system to rotate continuously from 0° to 360° along the central axis of the experimental platform, with an angle control accuracy of ±0.5°, realizing continuous adjustment across the entire wind direction range. It can be linked with the slope adjustment device to complete the accurate simulation of coupling conditions with different slopes and wind directions.

[0073] During the experiment, a wind field calibration experiment was first carried out. A three-dimensional hot-wire anemometer was used to calibrate the wind field on the experimental slope point by point to ensure that the wind field uniformity on the slope is ≤±5% and the turbulence intensity is ≤5%. Gradient nodes were set for wind speed in 1 m / s increments and full-range scanning nodes were set for wind direction in 15° increments. After each node was stabilized for 5 minutes, no less than 3 parallel combustion experiments were carried out to establish a three-dimensional wind speed vector field model. The flow field structure was monitored in real time using a particle image velocimeter (PIV).

[0074] Experimental data reveals that at a wind speed of 15 m / s and with the wind direction aligned with the slope, the flame propagation speed reaches a peak of 2.0 m / s, at which point the vortex shedding frequency at the flame front resonates with the wind speed. When the wind direction is against the slope, the propagation speed drops to 0.8 m / s, and boundary layer separation is significantly enhanced. Notably, after the wind speed exceeds 25 m / s, the flame propagation speed plateaus, at which point the increased flame lift height leads to a weakening of surface heat feedback. Monitoring using laser-induced fluorescence (LIF) technology shows that under high wind speeds, the plume entrainment coefficient increases by 60%, resulting in the rapid dissipation of heat and smoke, and the emergence of a discrete, jumping flame propagation pattern at the flame front, with smoke concentration reduced to 35% of the baseline value. This phenomenon was reproduced using large eddy simulation (LES) to demonstrate the heat loss mechanism caused by enhanced turbulent mixing.

[0075] VI. Multiphysics Data Acquisition and Database Construction (corresponding to) Figure 1 In the multiphysics data synchronous acquisition stage, through Figure 2 (The multiphysics synchronous acquisition subsystem in the middle is completed)

[0076] The fifth step of the experiment involved constructing a high-resolution time-series database, employing a distributed sensor network to collect key variable data during the fire spread process. The system deployed a temperature sensing array consisting of 200 thermocouples, 50 gas sampling probes, and a 128-channel particle image velocimetry system. The data acquisition frequency was set to 1 kHz, and time-frequency analysis was used to capture the transient characteristics of flame dynamics.

[0077] Analysis revealed that the flame front expanded exponentially over time, reaching an exponent of 1.5 under high wind speeds, indicating a strong convection-dominated propagation process. Temperature field distribution showed that the highest temperature in the flame center reached 800–950℃, following an Arrhenius-type temperature distribution pattern, while the temperature gradient at the edge reached 150℃ / cm. Flue gas concentration, measured using laser absorption spectroscopy, exhibited a spatial distribution conforming to a Gaussian diffusion model, but showed a bimodal distribution under controlled wind conditions. To ensure data reliability, the system employed a triple redundancy design; all sensors were NIST-calibrated with traceability, and the data acquisition unit was equipped with a real-time anomaly detection algorithm, ensuring measurement uncertainty was less than 1.5%.

[0078] High-speed cameras were simultaneously installed on the left and rear sides of the experimental platform. The left high-speed camera filmed through the left fireproof glass panel, with its lens axis at a 90° angle to the center line of the combustion bed. It was used to record the lateral shape of the flame, the flame tilt angle, the flame height, and the horizontal position of the fire spread front. The rear high-speed camera filmed through the rear opening of the experimental platform, with its lens axis at a 45° angle to the center line of the combustion bed. It was used to record the overall structure of the flame, the outline of the fire line, and the interaction between the flame and the incoming flow from a rearward perspective. The frame rate was set to 500 frames per second.

[0079] VII. Machine Learning Model Training and Optimization (corresponding to) Figure 1 In the data preprocessing and feature extraction, deep neural network model training and optimization stages, through Figure 2 (Completed by the machine learning simulation prediction subsystem)

[0080] The sixth step of the experiment involves using machine learning algorithms to extract features and optimize the model from the massive experimental data. A deep neural network model with 12 input parameters and 5 hidden layers (128 neurons per layer, ReLU activation function) is established, and the training dataset contains 3000 sets of experimental data. The specific algorithm and implementation process are as follows:

[0081] (1) Data preprocessing and normalization

[0082] The input and output parameters are normalized using a min-max method, mapping all parameters to the [0,1] interval. The normalization formula is as follows:

[0083]

[0084] In the formula: Here are the normalized parameter values, and x is the original parameter value. , These are the minimum and maximum values ​​of the parameter in the training set, respectively.

[0085] (2) Input parameter description

[0086] The input layer has 12 nodes, corresponding to: oxygen concentration, ambient temperature, slope angle, wind speed, wind direction angle, combustible material type (using integer codes 1-8, representing alpine pine, Yunnan pine, Sichuan spruce, Minjiang fir, alpine oak, rhododendron thicket, caragana thicket, and alpine meadow, respectively), combustible material moisture content, ambient humidity, air pressure, combustible material density, combustion time, and initial ignition power. All parameters are numerical and require no additional coding.

[0087] (3) Dataset partitioning

[0088] The 3000 sets of experimental data were divided into training, validation and test sets in a ratio of 7:2:1. The training set was used for updating model weights, the validation set was used to monitor model overfitting, and the test set was used to verify the model's generalization ability.

[0089] (4) Network forward propagation computation

[0090] The network structure consists of a 12-dimensional input layer → 5 hidden layers → a 3-dimensional output layer, with each hidden layer containing 128 neurons; the propagation formula from the input layer to the first hidden layer is as follows:

[0091]

[0092] The propagation formula from the l-th hidden layer to the (l+1)-th hidden layer (l=1,2,3,4):

[0093]

[0094] The propagation formula from the 5th hidden layer to the output layer is as follows:

[0095]

[0096] In the formula: W1~W6 are the weight matrices of each layer, and b1~b6 are the bias vectors of each layer; The network predicts the output values ​​(the three components correspond to the flame propagation speed, the x-coordinate of the flame front, and the y-coordinate of the flame front, respectively); the ReLU activation function expression is:

[0097]

[0098] (4) Loss function and backpropagation optimization

[0099] Using mean squared error (MSE) as the loss function, the expression is:

[0100]

[0101] In the formula: Y i This represents the measured value of the i-th sample (including flame propagation speed, x-coordinate, and y-coordinate). Let be the predicted value of the i-th sample; backpropagation weight update is performed using the Adam optimizer, with the learning rate set to 0.001, the weight decay coefficient set to 1e-5, the batch size set to 32, and the training iterations set to 2000. An early stopping mechanism is triggered when the validation set loss does not decrease for 50 consecutive iterations.

[0102] (5) Transfer learning adaptation framework

[0103] To adapt to eight typical high-altitude vegetation types—alpine pine, Yunnan pine, Sichuan spruce, Minjiang fir, alpine oak, rhododendron shrubland, Caragana shrubland, and alpine meadow—a parameter transfer learning method was adopted. First, the basic model was pre-trained using 3000 sets of experimental data across all types. The weight parameters of the first four hidden layers of the pre-trained model were frozen, and only the weights and biases of the fifth hidden layer and the output layer were fine-tuned. During the fine-tuning stage, a dedicated dataset for each target vegetation type was used, with no less than 300 samples for each target vegetation type. The fine-tuning iteration rounds were set to 500 rounds to achieve rapid adaptation to different vegetation types.

[0104] Validation results show that the model achieves a coefficient of determination (R²) of 0.96 for predicting flame propagation speed and an average Euclidean distance error of less than 0.15 m for predicting the flame front location. In a real-world case study, the model successfully reproduced the spread of a high-altitude forest fire, with an 82% overlap between the predicted boundary and the actual fire scene over 72 hours. Sensitivity analysis revealed that the parameter weights for oxygen concentration, slope, and wind speed were 0.35, 0.28, and 0.22, respectively, confirming the importance of the multi-parameter coupling mechanism. To further improve generalizability, a transfer learning framework was introduced into the model, and the prediction error remained below 9% in validation with eight typical vegetation types.

[0105] VIII. Implementation of the Global Dynamic Feedback Mechanism (corresponding to) Figure 1 The multi-parameter coupled regulation experiment and dynamic feedback closed-loop control loop in the process, through Figure 2 (Implementation of the dynamic feedback control subsystem in the middle)

[0106] In the multi-parameter coupling experiment, a global dynamic feedback mechanism is introduced. The triggering logic and execution order of this mechanism are as follows:

[0107] (1) Applicable scenarios: This mechanism is only activated in the stage of multi-parameter coupling experiments (such as orthogonal experiments and full factor coupling experiments). In the stage of single-factor experiments, the dual closed-loop control system of the corresponding parameter is activated first and this mechanism is not triggered.

[0108] (2) Progressive triggering sequence: During the coupling experiment, when the measured value of flame propagation speed deviates from the model prediction value, the single-parameter exclusive closed-loop control corresponding to the dominant factor of the deviation is triggered first (for example, when oxygen concentration is the dominant factor of the deviation, the oxygen dual closed-loop control system is activated first for correction); if the deviation between the measured value and the model prediction value is still >10% after the single-parameter closed-loop correction is completed, the global dynamic feedback mechanism is triggered.

[0109] (3) Linkage adjustment rules: After triggering, the PID controller adjusts the combined parameters of oxygen fraction ±1%, ambient temperature ±2℃, wind speed ±0.5m / s and slope angle ±0.5° according to the priority of parameter sensitivity weights (oxygen concentration weight 0.35, slope weight 0.28, wind speed weight 0.22, and other parameters sorted according to actual sensitivity) in a linkage manner. Only one core parameter is adjusted at a time. After adjustment, wait for the environmental parameters to stabilize for 3 minutes before restarting the experiment. If the deviation after adjusting a single parameter is still not reduced to within 10%, the next parameter is adjusted according to priority until the deviation is stably controlled within 10%, thus realizing global adaptive control in multi-parameter coupled scenarios.

[0110] This invention constructs a high-altitude fire simulation system with autonomous optimization capabilities, which can accurately reproduce the unique "leapfrog spread" and "three-dimensional combustion" phenomena of high-altitude fires, providing key technical support for establishing an intelligent fire early warning system and formulating precise fire extinguishing strategies.

[0111] The above embodiments are merely preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but which still solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A test method for simulating the spread of forest fires at high altitudes, characterized in that, Includes the following steps: (a) Set initial experimental conditions, including combustible material distribution parameters, terrain slope control range, environmental gas component ratio, environmental temperature, environmental humidity, and air pressure reference value; The oxygen concentration in the environmental gas components is adjusted according to the target high-altitude environmental parameters, and the low oxygen partial pressure environment corresponding to an altitude of 2000m to 5500m is simulated by reducing the oxygen volume fraction. (b) Oxygen concentration regulation experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant, and the oxygen concentration was dynamically regulated to record the flame combustion and spread characteristic parameters under different oxygen concentrations; then, a multi-parameter coupling experiment was carried out, and the oxygen concentration was orthogonally combined with the parameters of ambient temperature, slope and wind speed to quantify the influence weight of low oxygen environment on fire spread. (c) Environmental temperature control experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant, and the environmental temperature was dynamically controlled to record the flame combustion and spread characteristic parameters at different temperatures; then, a multi-parameter coupling experiment was carried out to quantify the coupling mechanism between the low-temperature environment and the other parameters. (d) Slope control experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant, and the slope was controlled within the range of 0° to 60°. The dynamic characteristics of flame propagation along the slope were recorded under different slopes. Then, a multi-parameter coupling experiment was carried out, combining all factors of slope and wind speed and direction parameters to quantify the synergistic effect of slope-wind field coupling on fire spread. (e) Wind speed and direction control experiment: First, a single-factor controlled variable experiment was carried out, keeping the other initial conditions constant. The wind speed was controlled in the range of 0 to 30 m / s and the wind direction was controlled in the range of 0° to 360°. The fire spread characteristics under different wind field conditions were recorded. Further multi-parameter coupling experiments were conducted, including an orthogonal experiment on four factors: wind speed, wind direction, slope, and oxygen concentration, to quantify the critical conditions for fire spread under multi-parameter coupling at high altitudes. (f) Using time series data acquisition technology, the flame front position, temperature field distribution, flue gas concentration, heat release rate, flame flow field characteristics, and free radical concentration changes are monitored and recorded in real time during each experiment.

2. The test method according to claim 1, characterized in that: In the single-factor controlled variable experiment in step (b), a single closed-loop control system was used to regulate the oxygen concentration. The oxygen concentration was gradually reduced by 2% each time, with an oxygen regulation accuracy of ±0.5%. A high-speed photography system was used to observe the flame propagation behavior at a sampling rate of 1000 frames / second, and the flame free radical concentration change was measured simultaneously using spectral analysis.

3. The test method according to claim 2, characterized in that: In the single-factor controlled variable experiment in step (c), the uniformity of the ambient cavity temperature field was first calibrated to ensure that the uniformity of the temperature field in the experimental area was ≤±1℃. The ambient temperature was adjusted at a cooling rate of 5℃ / min, with a temperature control accuracy of ±0.5℃ and an adjustment range of -20℃ to 30℃. A test node was set for every 10℃ decrease. After the temperature of each node stabilized for 30 minutes, no less than 3 parallel combustion experiments were carried out, and the average value was taken as the valid data.

4. The test method according to claim 3, characterized in that: In the single-factor controlled variable experiment in step (d), the levelness of the slope platform was first calibrated to ensure that the flatness error of the slope was ≤0.2mm / m at a slope of 0°. The slope was controlled within the range of 0° to 60° using an adjustable slope experimental platform with a control accuracy of ±0.1°. Core gradient nodes were set at 0°, 15°, 30°, 45°, and 60°. After each slope node stabilized, no less than three parallel combustion experiments were conducted. Simultaneously, particle image velocimetry was used to determine the characteristics of the flame propagation flow field.

5. The test method according to claim 4, characterized in that: The adjustable slope test platform has a rectangular box structure with a detachable grooved combustion bed inside. The combustion bed is 1.5m wide and 4.0m long, with ceramic fiber side plates 10cm high on both sides. The bottom plate has evenly distributed ventilation holes, and the surface is covered with stainless steel fine wire mesh with a mesh diameter of 0.9mm and an opening area of ​​85%.

6. The test method according to claim 4 or 5, characterized in that: In the single-factor controlled variable experiment in step (e), a wind field calibration experiment was first carried out. A three-dimensional hot-wire anemometer was used to calibrate the wind field of the experimental slope point by point to ensure that the wind field uniformity of the slope is ≤±5% and the turbulence intensity is ≤5%. An attached airflow field parallel to the slope was generated by a slope-adaptive modular wind field generation system. The system can adjust the pitch angle and rotation direction of the wind outlet synchronously with the slope angle to achieve continuous control of wind speed from 0 to 30 m / s and wind direction from 0° to 360°. Gradient nodes were set for wind speed in 1 m / s increments and full-range scanning nodes were set for wind direction in 15° increments. After each node stabilized, no less than 3 parallel combustion experiments were carried out.

7. The test method according to claim 6, characterized in that: In step (f), a distributed sensor network is used to collect data, including a thermocouple array, a gas sampling probe and a particle image velocimetry system, with a data acquisition frequency of 1kHz; at the same time, a high-speed camera is used to capture the flame shape and the position of the fire spread front from multiple angles from the side and rear of the experimental platform, with a shooting frame rate of not less than 500 frames / second.

8. The test method according to claim 1, 2, 3, 4, 5, or 7, characterized in that: It also includes a machine learning optimization step, in which the flame front location, temperature field data, flow field characteristics, flue gas concentration, and heat release rate data collected in step (f) are input into a deep neural network for training; the input parameter matrix of the deep neural network is... ,in Representing the set of real numbers, N is the number of training samples. The 12 input parameters are, in order: oxygen concentration, ambient temperature, slope angle, wind speed, wind direction angle, combustible type, combustible moisture content, ambient humidity, air pressure, combustible density, combustion time, and initial ignition power; the output parameter matrix is... The output results are, in order, the flame propagation speed, the x-coordinate of the flame front, and the y-coordinate of the flame front.

9. The test method according to claim 8, characterized in that: The deep neural network has a structure of 12-dimensional input layer → 5 hidden layers → 3-dimensional output layer. Each hidden layer has 128 neurons, and the activation function is ReLU. The mean squared error is used as the loss function, and the Adam optimizer is used for backpropagation weight update. The parameter transfer learning method is used to adapt to 8 typical high-altitude vegetation types.

10. The test method according to claim 9, characterized in that: In the multi-parameter coupling experiment, a global dynamic feedback mechanism is introduced. The triggering logic of this mechanism is as follows: when the measured value of flame propagation speed deviates from the model prediction value, the single-factor closed-loop control corresponding to the dominant factor of the deviation is triggered first; if the deviation is still >10% after single-factor correction, the global dynamic feedback mechanism is triggered, and the oxygen concentration, ambient temperature, wind speed and slope parameters are adjusted in linkage according to the parameter sensitivity weight through the PID controller until the deviation is stably controlled within 10%.

Images