A fire spread trend prediction method based on multi-model fusion and related equipment

Abstract

A fire spread trend prediction method based on multi-model fusion and related equipment relate to the field of fire simulation. The method comprises: identifying the burning area in the target forest and the physical characteristic data of each burning area, and inversely deducing the burning data of the corresponding cell in the preset cell space model; through a preset fire spread prediction model, simulating to generate the first fire spread data of the fire field at each prediction time and the first burning data of each cell; after obtaining the latest satellite data, identifying the actual burning data of each cell within the monitoring time of the latest satellite data; based on the deviation of the actual burning data of each cell and the corresponding first burning data, correcting the simulation parameters and model input parameters in the preset fire spread prediction model, and re-predicting to obtain the second fire spread data at each prediction time and the second burning data of each cell. The implementation of the above technical solution improves the accuracy of fire spread trend prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fire field simulation, and in particular to a fire spread trend prediction method based on multi-model fusion and related equipment. BACKGROUND

[0002] Forest fires are one of the most destructive natural disasters worldwide, not only seriously threatening people's life and property safety, but also causing long-term and far-reaching negative impacts on forest ecosystems, climate and environment, and biodiversity. In order to effectively organize rescue forces, scientifically develop evacuation plans and minimize losses, it is crucial to quickly and accurately predict the trend of fire spread.

[0003] Currently, fire spread prediction is usually achieved by constructing a digital simulation environment representing the real forest area. This method usually discretizes the complex forest geographical space, and inputs multi-source heterogeneous data such as topography, surface fuel distribution and real-time weather conditions as the basic parameters of the model. On this basis, the model generates the spatio-temporal dynamic evolution process of the fire field in the future period of time according to the physical mechanism and empirical law of fire behavior, with the actually monitored fire point information as the initial condition, through the calculation of the propagation and expansion of the fire line in space, in order to derive the prediction results of the fire field range and spread path.

[0004] However, fire spread is an uncertain dynamic process, and various data in the spread will be inaccurate due to real-time mutations of external environmental factors (such as wind field, temperature and humidity), and influences of measurement errors and timeliness of initial input data. The existing prediction model usually uses fixed simulation parameters, which is easy to mismatch with the dynamic evolution process in the actual fire field, leading to the accumulation of the deviation between the simulation state of the model and the real situation of the fire field over time, and further leading to the deviation between the prediction results and the actual results, affecting the accuracy of fire spread trend prediction. SUMMARY

[0005] The present application provides a fire spread trend prediction method based on multi-model fusion and related equipment, which can improve the accuracy of fire spread trend prediction.

[0006] In a first aspect, the application provides a fire spread trend prediction method, which comprises: identifying a burning area in a target forest and physical characteristic data of each burning area based on satellite data of the target forest, the physical characteristic data including thermodynamic characteristics, spectroscopic characteristics, and structural change characteristics; obtaining, by a preset combustion process inversion algorithm, burning data of a corresponding cell in a preset cell space model of the target forest based on the physical characteristic data of each burning area, the burning data including combustion intensity and combustible fuel combustion state; simulating, by a preset fire spread prediction model, a state transition process of each cell over a future time sequence based on the burning data of each cell in the preset cell space model, pre-acquired meteorological data, and ground surface data, to generate first fire spread data of a fire scene at each prediction time and first burning data of each cell, the preset fire spread prediction model including a fire spread speed model and a cellular automaton model; after obtaining the latest satellite data, identifying actual burning data of each cell within a monitoring time of the latest satellite data; based on a deviation between the actual burning data of each cell and the corresponding first burning data, correcting simulation parameters and model input parameters in the preset fire spread prediction model to obtain a corrected fire spread prediction model and corrected model input parameters; and based on the corrected model input parameters, re-predicting, by the corrected fire spread prediction model, second fire spread data at each prediction time and second burning data of each cell.

[0007] The above technical solution identifies fire burning data from satellite data, uses a preset fire spread prediction model to perform initial prediction of the fire based on the fire burning data to generate fire spread data, and after obtaining the latest satellite data each time, corrects simulation parameters and input parameters of the model based on actual burning data in the latest satellite data, and uses the corrected model and parameters to re-predict the fire spread data. This solution forms a closed loop of “satellite monitoring-data inversion-initial prediction-dynamic correction-re-prediction”, which enables the prediction model and model input data to be calibrated in real time based on the real dynamic evolution of the fire scene, thereby suppressing the accumulation of prediction errors over time and improving the accuracy of fire spread trend prediction.

[0008] In some embodiments of the first aspect, based on the combustion data of each cell in the preset cell space model, the pre-acquired meteorological data and surface data, the state transition process of each cell in the future time sequence is simulated through a preset fire spread prediction model to generate the first fire spread data of the fire field at each prediction time and the first combustion data of each cell, specifically comprising: generating a three-dimensional fire-induced disturbance field based on the combustion data of each cell in the preset cell space model; based on the three-dimensional fire-induced disturbance field, coupling and correcting the physical data of each combustible in the pre-acquired meteorological data and surface data respectively to generate corrected meteorological data and corrected surface data at the current prediction time; based on the combustion data of each cell, the corrected meteorological data and the corrected surface data, the state transition process of each cell in the future time sequence is simulated through the preset fire spread prediction model to generate the first fire spread data of the fire field at the current prediction time and the first combustion data of each cell; the first combustion data of each cell, the corrected meteorological data and the corrected surface data at the last prediction time adjacent to each prediction time are taken as the combustion data, the meteorological data and the surface data of each cell at each prediction time respectively to iteratively generate the first fire spread data at each prediction time after the current prediction time and the first combustion data of each cell.

[0009] By using the above technical solution, the three-dimensional fire-induced disturbance field generated according to the cell combustion data is used to correct the meteorological data and the physical data of the combustible to obtain corrected data, and then the cell state transition is simulated through the preset model to generate the first fire spread and combustion data at the current prediction time by combining the cell combustion data and the corrected data. Finally, the data at the last prediction time is taken as the initial value to iteratively generate the corresponding data at each subsequent prediction time. This scheme takes into account the dynamic influence of the fire itself on the local meteorology (such as wind field, temperature) and adjacent combustible (such as preheating and drying), quantifies the real-time changes (such as wind speed and direction adjustment, temperature and humidity fluctuation) of the local microclimate caused by the thermal plume, smoke and other factors generated by the fire through the three-dimensional fire-induced disturbance field, and the preheating and drying effect (such as water content reduction) on the surrounding combustible, so that the corrected meteorological data and surface data are no longer static initial values, but dynamic parameters reflecting the interaction between the fire and the environment. At the same time, the iterative mechanism makes the input data at each prediction time point be affected by the actual influence of the fire field at the last stage, ensuring that the entire prediction process is always based on dynamic parameters matching the real-time state of the fire field, avoiding the input data inaccuracy problem caused by fixed use of initial environmental data and ignoring the reaction of the fire on the surrounding environment in traditional prediction, thereby reducing the simulation deviation caused by the dynamic disconnection between the environment and the fire, and improving the accuracy of the fire spread trend prediction.

[0010] In some embodiments of the first aspect, after the step of generating the first fire spread data of the fire field at the current prediction time and the first combustion data of each cell by simulating the state transition process of each cell in the future time sequence based on the combustion data of each cell, the corrected meteorological data, and the corrected ground surface data through the preset fire spread prediction model, the method further comprises: identifying the firebrand source cell and the firebrand physical characteristics corresponding to the firebrand source based on the combustion intensity and the combustible burning state in the first combustion data of each cell; identifying the state change cell, the change data, and the change probability at the current prediction time according to the preset firebrand generation rule based on the firebrand source cell, the firebrand physical characteristics, the corrected meteorological data, and the corrected ground surface data; and correcting the first fire spread data at the current prediction time and the first combustion data of each cell based on the state change cell, the change data, and the change probability.

[0011] By identifying the firebrand source cell and the corresponding firebrand physical characteristics, and combining the firebrand source cell, the firebrand physical characteristics, the corrected meteorological data (wind field), and the corrected ground surface data (combustible attribute), the preset firebrand generation rule is used to simulate the spread and ignition of firebrands, identify the cells that may change due to firebrands, the change data, and the change probability, and correct the fire spread data and the combustion data at the current prediction time according to these data. This scheme identifies the firebrand source and the firebrand physical characteristics based on the combustion data, simulates the whole process of firebrand generation, propagation, and ignition of new areas according to the preset rule by combining the corrected meteorological data (such as wind field driving firebrand trajectory) and ground surface data (such as combustible ignition response), quantifies the information and probability of the state change cell, and then corrects the fire spread and combustion data. Therefore, the prediction not only includes the conventional continuous spread, but also covers the breakthrough new fire points caused by firebrands, avoiding the problems of small fire field range and path distortion caused by the lack of firebrand effect. Especially in the scene of strong wind and other conditions that are easy to cause firebrands, the scheme can more realistically reflect the complex evolution of the fire field, thereby improving the comprehensiveness and accuracy of the fire spread trend prediction.

[0012] In some embodiments of the first aspect, after the step of re-predicting the second fire spread data and the second combustion data of each cell at each prediction time by correcting the fire spread prediction model based on the corrected model input parameters, the method further comprises: if the prediction deviation reason at the monitoring time includes a new fire point, identifying a feasibility probability of the new fire point being formed by a firebrand based on a preset firebrand generation rule, the actual meteorological data and the actual surface data at the monitoring time; if the feasibility probability is lower than a preset firebrand threshold, determining whether the new fire point belongs to a smoldering fire relapse based on a preset smoldering fire relapse rule and historical fire data before the monitoring time; if yes, identifying a relapse cell and a corresponding relapse time that first satisfy the preset smoldering fire relapse rule after the monitoring time based on the preset smoldering fire relapse rule, the second fire spread data and the second combustion data of each cell; and re-predicting the second fire spread data and the second combustion data of each cell at each prediction time after the relapse time based on the relapse cell and the corresponding relapse time to generate third fire spread data and third combustion data of each cell at each prediction time.

[0013] With the above technical solution, when it is identified that the prediction deviation includes a new fire point, it is determined whether the new fire point is a smoldering fire relapse, and in the case of smoldering fire relapse, a relapse cell and a corresponding relapse time that first satisfy the relapse rule after the monitoring time are identified, and finally, the fire spread data and the combustion data are re-predicted based on these information. This solution solves the problem that the traditional model does not consider the smoldering fire relapse which is a hidden and delayed fire source. By distinguishing the causes of the new fire point (firebrand / smoldering fire), the potential fire point and the relapse time of the smoldering fire relapse are included in the prediction system, so that the re-generated data can reflect the subsequent spread caused by such hidden fire source, avoid the underestimation of the fire range and the distortion of the evolution path caused by the omission, and make the prediction results cover both continuous spread and non-continuous new fire point caused by smoldering fire relapse, thereby improving the prediction integrity and accuracy of the complex fire field evolution process.

[0014] In some embodiments in combination with the first aspect, in some embodiments, based on the preset smoldering fire rekindling rule, the second fire spread data and the second combustion data of each cell, a rekindling cell and a corresponding rekindling time that first satisfy the preset smoldering fire rekindling rule after the monitoring time are identified, specifically comprising: based on the preset smoldering fire rekindling rule, extracting the rekindling reference feature from the ground surface data of the new fire point; based on the second combustion data and the ground surface data of each cell at each prediction time from the fire starting time to the prediction cutoff time, identifying the cells similar to the rekindling reference feature in all cells to obtain a high-risk rekindling cell set; based on the second combustion data of each high-risk rekindling cell at each prediction time, calculating the smoldering fire data of each high-risk rekindling cell at each prediction time after the monitoring time; based on the smoldering fire data of each high-risk rekindling cell, identifying the rekindling cell and the corresponding rekindling time that first satisfy the preset smoldering fire rekindling rule through the preset smoldering fire rekindling rule.

[0015] By using the above technical solution, by identifying the high-risk rekindling cell, combining the combustion data of the high-risk rekindling cell and its adjacent cells in the historical fire field and the current fire field, calculating the smoldering fire data of each high-risk rekindling cell at the monitoring time, and identifying the cell and the corresponding rekindling time that first satisfy the rekindling condition according to the preset smoldering fire rekindling rule and the smoldering fire data. The scheme physically simulates the accumulation and future evolution of underground latent heat, changes the smoldering fire rekindling from a post-event passive identification phenomenon to a future event that can be actively and quantitatively predicted, so that the potential new fire source information can be brought into the overall prediction framework of fire spread in advance, avoiding the problem that the prediction result deviates seriously from the actual situation due to the inability to predict the new fire point in the burned-out area, and improving the comprehensiveness and accuracy of the fire spread trend prediction.

[0016] In some embodiments in combination with the first aspect, after the step of identifying the burning area in the target forest and the physical feature data of each burning area based on the satellite data of the target forest, the method further comprises: if there is a first burning area with a non-empty structural change feature, generating the theoretical combustion feature of each first burning area based on the structural change feature and the ground surface data of each first burning area; calculating the theoretical thermodynamic feature and the theoretical spectroscopy feature of each first burning area based on the theoretical combustion feature and the ground surface data of each first burning area; generating the first combustion feature of each first burning area based on the thermodynamic feature, the meteorological data and the theoretical thermodynamic feature of each first burning area; generating the second combustion feature of each first burning area based on the spectroscopy feature, the meteorological data and the theoretical spectroscopy feature of each first burning area; and correcting the physical feature data of each first burning area based on the theoretical combustion feature, the first combustion feature and the second combustion feature of each first burning area.

[0017] According to the technical scheme, for the first combustion area with the structural change characteristic not being empty, the theoretical combustion characteristic is generated by combining the structural change characteristic and the surface data, and then the theoretical thermodynamic and spectroscopic characteristics are calculated according to the theoretical combustion characteristic, and then the first and second combustion characteristics are generated by combining the actual thermodynamic / spectroscopic characteristics and the meteorological data, and finally the physical characteristics are corrected by using the theoretical combustion characteristic and the first and second combustion characteristics. According to the scheme, the thermodynamic and spectroscopic characteristics which are easily interfered by smoke and other factors are cross-checked and corrected by taking the stable structural change characteristic as a physical reference, thereby establishing a self-consistent correction mechanism among multiple data sources, effectively solving the measurement distortion or uncertainty problem caused by smoke shielding, atmospheric interference or sensor saturation of a single remote sensing data source (such as only thermal radiation or spectral index), ensuring that the description of the combustion state has higher accuracy and reliability, and further improving the accuracy of the fire spread trend prediction.

[0018] In combination with some embodiments of the first aspect, in some embodiments, after the step of identifying the combustion areas in the target forest and the physical characteristic data of each combustion area based on the satellite data of the target forest, the method further includes: if there is a second combustion area with the structural change characteristic being empty and the thermodynamic and spectroscopic characteristics being not empty, generating candidate combustion characteristics by reverse physical deduction based on the thermodynamic and spectroscopic characteristics, the meteorological data and the surface data of the second combustion area, the thermodynamic and spectroscopic characteristics being characteristic data that can be collected by a satellite under the influence of the candidate combustion characteristics and the meteorological data; calculating the compliance of the candidate combustion characteristics with the combustion characteristics of each combustible in the second combustion area based on the candidate combustion characteristics, the meteorological data and the surface data of the second combustion area; if the compliance is lower than a preset compliance threshold, searching for a first target combustion characteristic set matched with the thermodynamic characteristics and the meteorological data of the second combustion area and a second target combustion characteristic set matched with the spectroscopic characteristics and the meteorological data of the second combustion area based on a preset combustion characteristic library; selecting the combustion characteristic with the highest compliance with the combustion characteristics of each combustible in the second combustion area in the first target combustion characteristic set and the second target combustion characteristic set as a hypothetical combustion characteristic; and correcting the second combustion area and the corresponding physical characteristics based on the hypothetical combustion characteristic.

[0019] The second combustion region has no structural change characteristics, but has thermodynamic and spectroscopic characteristics. The candidate combustion characteristics are generated by reverse physical deduction. If the candidate combustion characteristics do not match the actual combustion characteristics, the feature set corresponding to the thermodynamic characteristics and the spectroscopic characteristics is matched from the preset combustion feature library, the combustion characteristics with the highest conformity to the combustible characteristics are selected as the assumed combustion characteristics, and the second combustion region and the corresponding physical characteristics are finally corrected by using the characteristics. The scheme uses known data to match and derive the assumed combustion characteristics with the highest physical credibility, ensures that the combustion characteristics input into the subsequent spread model have the highest physical credibility, avoids the situation that the blind deduction does not match the actual combustion, thereby providing more accurate initial conditions for the entire prediction process, and further improving the accuracy of the fire spread trend prediction.

[0020] In a second aspect, an embodiment of the present application provides a server, comprising: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is configured to store computer program codes, the computer program codes comprising computer instructions, and the one or more processors are configured to invoke the computer instructions to enable the server to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, comprising instructions, when the instructions are executed on a server, the server is enabled to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0022] In a fourth aspect, the present application provides a computer program product, when the computer program product is executed on a server, the server is enabled to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0023] It can be understood that the server provided in the second aspect, the storage medium provided in the third aspect, and the computer program product provided in the fourth aspect are all used to execute the method provided in the present application. Therefore, the beneficial effects that can be achieved are referred to the beneficial effects in the corresponding method, which will not be described here.

[0024] The one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0025] 1. After obtaining the latest satellite data each time, the application corrects the simulation parameters and input parameters of the model according to the actual combustion data in the latest satellite data, and re-predicts the fire spread data by using the corrected model and parameters. Through the construction of a closed loop of "satellite monitoring-data inversion-initial prediction-dynamic correction-re-prediction", the prediction model and model input data can be calibrated in real time according to the real dynamic evolution of the fire field, thereby suppressing the accumulation of prediction errors over time and improving the accuracy of fire spread trend prediction.

[0026] 2. The application quantifies the dynamic influence of fire on local micro-meteorology and physical properties of combustible materials through a three-dimensional fire-induced disturbance field, and combines an iterative mechanism to ensure that the prediction input data matches the real-time state of the fire field. At the same time, the generation, propagation and ignition process of flying fire are included, covering the scenario of jump spread, and by distinguishing the causes of new fire points, potential fire sources and rekindling time of rekindling fire are identified, solving the problem of ignoring the interaction between fire and environment and missing the calculation of jump / implicit fire spread in traditional models, and further improving the comprehensiveness and accuracy of fire spread prediction.

[0027] 3. By taking stable structural change characteristics as the physical benchmark, the application cross-checks and corrects the thermodynamic and spectroscopic characteristics that are easily disturbed by factors such as smoke, effectively addressing the measurement distortion or uncertainty problems caused by smoke shielding, atmospheric interference or sensor saturation of a single remote sensing data source, ensuring that the description of the combustion state has higher accuracy and reliability, and thus improving the accuracy of fire spread trend prediction. At the same time, when the observation data is incomplete or there are inaccurate feature data that cannot be identified, the highest physically credible assumed combustion characteristics are derived by matching known data, ensuring that the combustion characteristics input into the subsequent spread model have the highest physical credibility, avoiding the inaccuracy caused by blind derivation, and thus providing more accurate initial conditions for the entire prediction process, and thus improving the accuracy of fire spread trend prediction. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a flowchart of the fire spread trend prediction method in the embodiment of the application;

[0029] Figure 2 is another flowchart of the fire spread trend prediction method in the embodiment of the application;

[0030] Figure 3 is an exemplary hardware structure diagram of a server in the embodiment of the application. DETAILED DESCRIPTION

[0031] The terminology used in the following description of the embodiments herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the description of the embodiments and the appended claims herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It also will be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0032] Hereinafter, the terms "first", "second" are used only for the purpose of description, and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise specified.

[0033] In the related art, fire spread prediction is usually achieved by constructing a digital simulation environment representing a real forest area. This method usually discretizes the complex forest geographical space, and comprehensively inputs topography, surface fuel distribution, and real-time meteorological conditions and other multi-source heterogeneous data as the basic parameters of the model. On this basis, the model generates the spatio-temporal dynamic evolution process of the fire field in the future period of time according to the physical mechanism and empirical law of fire behavior, with the actually monitored fire point information as the initial condition, simulates the propagation and expansion of the fire line in space, and generates the spatio-temporal dynamic evolution process of the fire field in the future period of time in a time step by time step manner, so as to obtain the prediction results of the fire field range and spread path.

[0034] However, fire spread is an uncertain dynamic process, and various data in the spread will be inaccurate due to real-time mutations of external environmental factors (such as wind field, temperature and humidity), and influences of measurement errors and timeliness of initial input data. The existing prediction model usually uses fixed simulation parameters, which is easy to mismatch with the dynamic evolution process in the actual fire field, causing the deviation between the simulation state of the model and the real situation of the fire field to accumulate over time, and further causing the deviation between the prediction results and the actual results, affecting the accuracy of the fire spread trend prediction.

[0035] However, fire spread is an uncertain dynamic process, and various data in the spread will be inaccurate due to real-time mutations of external environmental factors (such as wind field, temperature and humidity), and influences of measurement errors and timeliness of initial input data. The existing prediction model usually uses fixed simulation parameters, which is easy to mismatch with the dynamic evolution process in the actual fire field, causing the deviation between the simulation state of the model and the real situation of the fire field to accumulate over time, and further causing the deviation between the prediction results and the actual results, affecting the accuracy of the fire spread trend prediction.

[0035] However, fire spread is an uncertain dynamic process, and various data in the spread will be inaccurate due to real-time mutations of external environmental factors (such as wind field, temperature and humidity), and influences of measurement errors and timeliness of initial input data. The existing prediction model usually uses fixed simulation parameters, which is easy to mismatch with the dynamic evolution process in the actual fire field, causing the deviation between the simulation state of the model and the real situation of the fire field to accumulate over time, and further causing the deviation between the prediction results and the actual results, affecting the accuracy of the fire spread trend prediction.

[0036] The method of the embodiment of the application will be described below in combination with Figure 1

[0037] Please refer to Figure 1 , a flowchart of the fire spread trend prediction method in the embodiment of the application.

[0038] S101, based on satellite data of the target forest, identify the burning area in the target forest and the physical characteristic data of each burning area.

[0039] The physical characteristic data includes thermodynamic characteristics, spectroscopic characteristics and structural change characteristics.

[0040] Specifically, the server first connects to the satellite data distribution platform to obtain the latest multi-source satellite data covering the target forest range in the same time window, including optical satellite data, thermal infrared satellite data, microwave radar satellite data and laser radar satellite data, etc.

[0041] Then, based on the multi-source satellite data, identify the burning area in the target forest. Through the thermal infrared data, according to the statistical range of the ground temperature of the target forest in the same period without fire, screen out potential thermal anomaly pixels that exceed the range and have continuous heat radiation; through the optical data, set a threshold value based on the normal vegetation burning index distribution before the fire, screen out potential burning pixels whose index is lower than the threshold value; through the microwave radar data, set a threshold value according to the natural fluctuation of the backscattering coefficient in the non-burning period, screen out potential burning pixels whose coefficient drops below the threshold value. Then use the density clustering algorithm to determine the clustering radius according to the satellite resolution, set the minimum number of clustered pixels according to the minimum size of the burning area, aggregate the spatially adjacent and feature-consistent potential pixels into multiple continuous burning areas (one pixel corresponds to one area), eliminate isolated pixels, and form a set of burning areas with unique identification and boundaries.

[0042] ​Then, the physical characteristic data of each combustion area is extracted. For the thermodynamic characteristics, the brightness temperature of the corresponding pixel is extracted from the thermal infrared data, combined with the radiation calibration parameters determined by the satellite sensor factory calibration and periodic on-orbit verification to convert into thermal radiation intensity, and then a functional relationship model between thermal radiation intensity and fire radiation power is established based on the thermal radiation transmission law, and combined with the pixel area and thermal radiation intensity, the fire radiation power is calculated; for the spectral characteristics, the combustion index (such as the normalized combustion index, the combination mode is determined according to the spectral characteristics of vegetation combustion) of the corresponding pixel is calculated from the optical data, and the difference combustion index is obtained by comparing the pre-fire data to quantify the degree of vegetation damage, and the reflectivity in the short-wave infrared band is extracted to reflect the characteristics of the combustion residues; for the structural change characteristics, when there is laser radar data, the pre-fire and current digital canopy height models are compared to calculate the canopy height loss value and the canopy coverage change rate, and when there is no laser radar data, the optical stereo image pair (obtained from the optical satellite data) is used to reconstruct the three-dimensional point cloud model by matching the same named points in the stereo image pair with different viewing angles, and the vegetation height change and the biomass loss are calculated.

[0043] In addition, due to the influence of smoke, cloud cover or sensor accuracy during the acquisition process of satellite data, the extraction of physical characteristics of part of the combustion area may be biased, at the same time, satellite data is collected by multiple satellites, and the satellite data obtained in the same time window may not be complete, and thus the physical characteristics of all combustion areas cannot be completely and accurately extracted, affecting the comprehensiveness and accuracy of the subsequent fire spread prediction, in order to complete the incomplete satellite data at the same time and ensure the integrity and reliability of the extraction of the physical characteristics of the combustion area, when the key physical characteristics (such as the structural change characteristics) exist, the following steps can be performed:

[0044] If the first combustion area with a non-empty structural change feature exists, first, based on the structural change feature and the ground data of each first combustion area, the theoretical combustion characteristics of each first combustion area are generated. The server first calls the target forest ground database (including basic attributes such as fuel type, density, heat value, moisture content, and carbon content), synchronously extracts the structural change feature of the first combustion area (such as the canopy height loss rate and the biomass loss amount); then, the exposed area of the combustible material is calculated by combining the basic combustible material area of the first combustion area (determined by the combustible material density and the area) and the structural change feature; then, according to the exposed area of the combustible material, the heat value of the combustible material, and the combustion efficiency coefficient of the corresponding type of combustible material, the preliminary theoretical combustion intensity is calculated, and then the theoretical combustion intensity is corrected by combining the moisture content of the combustible material and the moisture content correction coefficient to obtain the corrected theoretical combustion intensity; then, based on the carbon content of the combustible material, the incomplete combustion proportion coefficient associated with the corrected theoretical combustion intensity, the smoke generation coefficient of the corresponding combustible material, and the total mass of the combustible material (determined by the combustible material density and the area), the total amount of smoke generated by combustion is calculated, and then the near-surface smoke concentration is obtained by combining the corrected theoretical combustion intensity and the smoke diffusion height coefficient; finally, the theoretical burnout time and the theoretical combustible material consumption rate are derived by the total mass of the combustible material, the exposed area of the combustible material, the corrected theoretical combustion intensity, and the heat value of the combustible material, and finally the theoretical combustion characteristics including the corrected theoretical combustion intensity, the theoretical burnout time, the theoretical combustible material consumption rate, the total smoke generation amount, and the near-surface smoke concentration are output.

[0045] Secondly, based on the theoretical combustion characteristics and the ground data of each first combustion area, the theoretical thermodynamic characteristics and the theoretical spectroscopy characteristics of each first combustion area are calculated. For the theoretical thermodynamic characteristics, the corrected theoretical combustion intensity is multiplied by the heat radiation efficiency of the combustible material (obtained from the combustion characteristics measurement report published by the forestry research institution according to the type of combustible material, which reflects the proportion of heat energy released by combustion to radiant energy), to obtain the theoretical fire radiation power (directly reflecting the intensity of heat radiation released by combustion), and then the theoretical fire radiation power is converted to the theoretical brightness temperature according to the blackbody radiation law (the general law of heat radiation and temperature correlation in the field) and the radiation characteristics of the thermal infrared band; for the theoretical spectroscopy characteristics, the original spectrum attribute of the combustible material (reflecting the basic spectrum reflection law before combustion) is taken as the reference, the damage degree of combustion to vegetation is determined by combining the corrected theoretical combustion intensity (the higher the intensity, the more serious the damage, and the more significant the spectrum change), and then the spectrum response coefficient of the combustible material after combustion corresponding to the type of residual material after combustion is matched (obtained from the combustible material spectrum database constructed by satellite observation and ground measurement, which quantifies the change amplitude of spectrum reflectivity under different combustion damage degrees), and then the theoretical shortwave infrared reflectivity and the theoretical combustion index (such as the normalized combustion index) are obtained by calculating the ratio or difference of the reflectivity of a specific band (such as the reflectivity relationship between the shortwave infrared band and the near-infrared band).

[0046] Third, based on the thermodynamic characteristics of each first combustion area, meteorological data and theoretical thermodynamic characteristics, the first combustion characteristics of each first combustion area are generated. The server first acquires regional meteorological data completely synchronized with the satellite data acquisition time, including atmospheric temperature vertical profile, humidity, air pressure, and aerosol optical thickness calculated from smoke concentration and other parameters, and constructs a real-time atmospheric radiation transmission path model from the ground combustion area to the satellite sensor. Subsequently, the server takes the theoretical thermodynamic characteristics (representing the true radiation of the ground surface without any atmospheric interference) as the initial input radiation source of the model, and uses an atmospheric radiation transmission correction algorithm to simulate the attenuation of the initial radiation energy due to the absorption and scattering of specific components (such as smoke and water vapor) when passing through the atmosphere, thereby simulating the attenuation of the initial radiation energy due to absorption and scattering when passing through the constructed atmospheric path. The server compares the satellite end radiation value obtained after simulation with the actual observed thermodynamic characteristics, and takes the deviation between the two as the objective function, iteratively optimizes the key attenuation factors (such as smoke optical thickness, water vapor content, etc.) in the atmospheric path model, until the difference between the simulation result and the actual observation value converges to a minimum threshold set based on the sensor noise level. Finally, the attenuation factors related to the combustion characteristics in the optimized physical attenuation factor combination that achieves dynamic balance between the theoretical value and the actual observation value are quantified as the first combustion characteristics of the first combustion area, representing the comprehensive attenuation effect of smoke and water vapor and other products generated by combustion on the true thermal radiation signal of the fire point on the satellite observation path.

[0047] Fourthly, based on the spectral characteristics of each first combustion area and meteorological data and theoretical spectral characteristics, the second combustion characteristics of each first combustion area are generated. The server first obtains the regional meteorological data that accurately matches the satellite optical data collection time point, including atmospheric visibility, total column water vapor, and aerosol optical thickness as the initial estimate of smoke concentration, and constructs a two-way atmospheric radiation transmission path model covering from the sun to the ground and then to the satellite sensor. Subsequently, the server inputs the theoretical spectral characteristics (characterizing the true reflectance spectrum of the ground surface without atmospheric influence) as the ground boundary condition of the model, and uses mature atmospheric correction algorithms to forward simulate the signal changes caused by smoke aerosol scattering and gas molecule absorption in the process of solar radiation penetrating through the atmosphere, reflecting on the ground, and then penetrating out of the atmosphere to the sensor. The server compares the simulated satellite apparent reflectance with the actually observed spectral characteristics at each key waveband through inverse calculation, and takes the total deviation of all wavebands as the objective function, iteratively optimizes the attenuation factors (especially the aerosol optical thickness and its scattering phase function) in the atmospheric path model dominated by combustion activities, until the similarity of the simulated spectrum and the actually observed spectrum reaches an optimal matching threshold preset based on the sensor calibration accuracy. Finally, this set of optimized aerosol physical parameter combinations that can best explain the difference between the theoretical and observed spectra is quantified as the second combustion characteristics of the first combustion area by mapping the aerosol physical parameters (such as optical thickness, scattering phase function) to the calculation rules of the quantitative indicators representing the shielding effect of the smoke plume, and the data related to combustion in the combination is quantified as the second combustion characteristics of the first combustion area, representing the shielding and distortion degree of the combustion smoke plume to the ground spectral information.

[0048] Finally, based on the theoretical combustion characteristics of each first combustion area, the first combustion characteristics and the second combustion characteristics, the physical characteristic data of each first combustion area is corrected. If the physical characteristic data of the first combustion area is complete, the server first determines the fusion weight of the theoretical combustion characteristics, the first combustion characteristics and the second combustion characteristics based on the matching degree of each characteristic in the historical combustion data and the ground measured value (the weight of the theoretical combustion characteristics is the highest due to the anchoring physical source, and the weights of the first and second combustion characteristics are inversely set according to the inversion deviation rate, the lower the deviation rate, the greater the weight, and the parameters are determined by statistical historical correction effect), and the three types of characteristics are fused under the least square optimization framework (to realize the internal contradiction between the characteristics by minimizing), to generate comprehensive combustion characteristics with integrity and consistency. Subsequently, the physical characteristics are corrected based on the comprehensive combustion characteristics: if the physical characteristics of the first combustion area are complete, the thermal radiation attenuation parameters in the comprehensive combustion characteristics are substituted into the mature thermal radiation transfer inverse model in the field to compensate the initial thermodynamic characteristics in reverse to restore the true thermal radiation signal; the aerosol scattering parameters in the comprehensive combustion characteristics are substituted into the spectral unmixing inverse model to strip the spectral distortion of the smoke plume to restore the true spectral characteristics; at the same time, the biomass consumption law in the comprehensive combustion characteristics is used as a benchmark to verify and fine-tune the original structural change characteristics, and finally the corrected physical characteristics are integrated. If the physical characteristics of the first combustion area are not complete, the missing physical characteristic data is first completed based on the comprehensive combustion characteristics (such as using the theoretical thermal radiation parameters in the comprehensive characteristics to complete the missing thermodynamic characteristics, and using the theoretical spectral parameters to complete the missing spectral characteristics), and then the correction process of the complete physical characteristics is performed to generate the final corrected physical characteristic set, which replaces the initial extracted physical characteristic data.

[0049] When the key physical characteristics are missing, but other observation data (such as thermodynamic and spectral characteristics) exist, the following steps can be performed:

[0050] If there is a second combustion area with no structural change feature and non-empty thermodynamic feature and spectroscopy feature, first, based on the thermodynamic feature, spectroscopy feature, meteorological data and surface data of the second combustion area, the candidate combustion feature is generated through reverse physical deduction. The thermodynamic feature and spectroscopy feature are the feature data that can be collected by the satellite under the influence of the candidate combustion feature and meteorological data. The server first determines the physically reasonable atmospheric interference parameter iteration range according to the meteorological data; in each iteration, a set of specific atmospheric parameters (including smoke concentration and water vapor content) is selected, the thermal radiation attenuation coefficient table (corresponding to the thermal infrared band attenuation coefficient of different parameters) in the authoritative atmospheric physics manual is used to calculate the attenuation amount of thermal radiation from the ground to the satellite, and the satellite observed thermal dynamic feature is superimposed with the attenuation amount to obtain the real thermal radiation signal of the ground; at the same time, the spectral scattering coefficient table (corresponding to the scattering coefficient of different parameters in the short-wave infrared band) in the manual is used to calculate the scattering loss amount of the spectral signal, and the satellite observed spectroscopy feature is superimposed with the loss amount to obtain the real spectral signal of the ground; then, combined with the surface data, the real thermal radiation signal is used to divide the combustible material thermal radiation efficiency (obtained from the combustible material combustion measurement report published by the forestry scientific research institution) to preliminarily obtain the combustion intensity; then, according to the “combustion index-combustion intensity correlation table” (based on the combustion experiment statistics of different combustible material types), the preliminary combustion intensity is calibrated by the real combustion index; the real short-wave infrared reflectivity is matched with the combustion residue type (such as ash, unburned wood), the combustible material consumption rate is adjusted (obtained by dividing the calibrated combustion intensity by the calorific value of the combustible material) by referring to the “residue type-consumption rate correction table” (constructed based on the residue sampling and consumption rate observation data), and the smoke generation amount is corrected combined with the residue type (the correlation coefficient comes from the smoke generation measurement statistics of the same type of combustible material combustion); after traversing all atmospheric parameter combinations, the calibrated combustion intensity, adjusted combustible material consumption rate and corrected smoke generation amount corresponding to each group are collected to form a candidate combustion feature set, so that the satellite observation results can match the current thermodynamic and spectroscopy features after the interference of the corresponding atmospheric parameters.

[0051] Secondly, based on the candidate combustion characteristics of the second combustion area, meteorological data and surface data, the compliance of the candidate combustion characteristics with the combustion characteristics of each combustible in the second combustion area is calculated. The server first extracts the standard combustion characteristics (including theoretical combustion intensity range, standard combustible consumption rate, typical smoke generation amount, data taken from different combustible combustion measurement reports published by forestry scientific research institutions) corresponding to each combustible type in the second combustion area from the surface data; then determines the characteristic correction coefficient (the coefficient is generated based on the deviation rate of the actual combustion characteristics of this type of combustible from the standard characteristics under the same meteorological conditions in the same period in history, which is used to eliminate the influence of meteorology on combustion characteristics), adjusts the standard combustion characteristics with the correction coefficient to obtain the actual combustion characteristics benchmark of the combustible under the current meteorological conditions; then for each set of candidate combustion characteristics, the single parameter matching degree with the benchmark is calculated; finally, according to the evaluation standard of the importance of each parameter in the field of fire combustion (such as the highest weight of the combustion intensity on the combustion state), the weight of each parameter matching degree is determined, and the comprehensive compliance of each set of candidate combustion characteristics with the actual combustion characteristics of each combustible is calculated by weighted summation, and the higher the value, the higher the matching degree of the candidate characteristics with the real combustion state of the combustible.

[0052] Thirdly, if the compliance exceeds the preset compliance threshold, the physical characteristics of each first combustion area are modified based on the candidate combustion characteristics with the highest compliance. First, based on the parameters such as combustion intensity and smoke generation amount contained in the candidate combustion characteristics with the highest compliance, the thermodynamic characteristics and spectroscopic characteristics initially identified are modified to eliminate the interference of atmosphere and smoke. Then, based on the combustible consumption rate and total consumption amount in the characteristics, the structural change characteristics initially missing are physically estimated and completed, the deduced total consumption amount of combustible is directly related to the biomass loss amount, and combined with the prior knowledge of the combustible type, distribution density and vertical structure of the region obtained from the surface data, the key indicators such as crown height loss value and crown coverage change rate are further calculated, so as to generate the structural change characteristics.

[0053] Fourthly, if none of the compliances exceeds the preset compliance threshold, based on the preset combustion characteristic library, the first target combustion characteristic set matching the thermodynamic characteristics and meteorological data of the second combustion area is found, and the second target combustion characteristic set matching the spectroscopic characteristics and meteorological data of the second combustion area is found. The preset combustion characteristic library is a multi-dimensional knowledge base constructed by comprehensive analysis and calibration of a large number of controlled combustion experiment data and historical wild fire cases, which stores the internal correlation between the thermodynamic characteristics, spectroscopic characteristics that can be collected under the influence of satellite under various combustion behaviors and the actual physical characteristics that can be collected under the influence of satellite under different combustible types and meteorological conditions.

[0054] Fifth, the highest degree of compliance of the combustion characteristics in the first target combustion characteristic set and the second target combustion characteristic set with the combustion characteristics of each combustible in the second combustion area is selected as the assumed combustion characteristic. The server first combines all the characteristics in the first target combustion characteristic set and the second target combustion characteristic set into a candidate set, and then for each characteristic in the set, the same method as calculating the degree of compliance of the candidate combustion characteristic is used to calculate the single parameter matching degree of the combustion intensity, combustible consumption rate, and smoke generation amount in the characteristic with the benchmark, in combination with the actual combustion characteristic benchmark (standard combustion characteristic after meteorological correction) of each combustible in the second combustion area, and the comprehensive compliance is obtained by weighting and summing the parameter weights (the combustion intensity weight is the highest) determined in the fire combustion field. Finally, the degrees of compliance of all characteristics with the comprehensive compliance greater than the preset compliance threshold are compared, and the characteristic with the highest value is selected as the assumed combustion characteristic, ensuring that the characteristic has the optimal matching degree with the actual combustion characteristics of the combustibles in the second combustion area.

[0055] Finally, based on the assumed combustion characteristic, the physical characteristics of the second combustion area are corrected. Based on the actual physical characteristics corresponding to the assumed combustion characteristic in the preset combustion characteristic library, the physical characteristics of the second combustion area are replaced.

[0056] S102, based on the physical characteristic data of each combustion area, the combustion data of the corresponding cell in the preset cell space model of the target forest is obtained by reversing the preset combustion process algorithm.

[0057] The combustion data includes combustion intensity and combustible combustion state.

[0058] Specifically, the server first acquires the preset cell space model of the target forest constructed in advance, each cell is associated with a unique spatial coordinate and basic combustible attribute (type, initial biomass, layer proportion, etc., data obtained from the ground database); at the same time, the preset combustion process inversion algorithm is acquired, which is based on the energy conservation law and combustible combustion dynamics law as the theoretical basis, by establishing a physical mapping model (such as a physical model based on radiation transfer model and combustion dynamics equation), and combining historical data containing different combustion state records for training and calibration, thereby realizing the reverse deduction from remote sensing physical characteristics to combustion data (such as heat release rate, combustion efficiency, etc.); the core parameters of the algorithm can be set according to the measured reports of forestry research institutions and fire investigation specifications.

[0059] Then, for each combustion area, the server first matches the physical characteristic data of the area, such as thermodynamic, spectroscopy, structural change, etc., to the corresponding cell in the preset cell space model according to the spatial coordinates. If the spatial resolution of the satellite is low, resulting in a single satellite pixel covering multiple cells in the preset cell space model in space, and the single and homogeneous physical characteristic data of the pixel cannot distinguish which of the covered cells are at the real fire front position and which have not yet burned, thereby causing significant uncertainty about the initial state of the fire field, the sub-pixel fire line reconstruction mechanism is started to solve the spatial scale mismatch problem. First, a thermodynamic decomposition model is used to decompose the mixed thermal radiation signal observed by the cell into a combustion part and a background part, thereby estimating the fire pit area fraction of the actual combustion area in the cell. Subsequently, the spatial neighborhood combustion state of the cell, higher resolution terrain data, and refined combustible material distribution map are comprehensively analyzed to reconstruct the most likely spatial position and form of the fire pit area in the cell under high-dimensional information constraints, forming a sub-pixel scale fire line.

[0060] Subsequently, by means of a preset combustion process inversion algorithm, the server substitutes the total fire radiation power in the thermodynamic characteristics into the "radiation power-burning intensity conversion model", which combines the blackbody radiation law and the combustible thermal radiation efficiency to calculate the total burning intensity. This total burning intensity will no longer be evenly assigned to the entire cell, but will be concentrated on the sub-cell representing the fire line according to the results of the sub-pixel fire line reconstruction, and will be numerically amplified according to the area ratio to ensure energy conservation, while the remaining areas in the cell will be marked as unburned or burned out. In determining the combustion state of the combustible material, the server combines the biomass loss rate of different vegetation layers in the structural change characteristics and the change amount of the combustion index in the spectroscopy characteristics to determine the combustion degree of the tree canopy layer, shrub layer, and herb layer, respectively; and then calls the three-dimensional posture change data in the structural change characteristics, and refers to the upright baseline threshold set in the forestry standard to determine whether the combustible material is in an upright, inclined, or collapsed combustion posture.

[0061] Optionally, for the combustion area where physical characteristic data is missing, the server will start a spatial interpolation completion mechanism, i.e., based on the adjacent cells around the data missing cell in the area that have complete physical characteristics, a weighted average algorithm that combines the inverse of spatial distance and the similarity of basic combustible properties is used to estimate the missing physical characteristic data, and then the inversion process with complete data is performed. If the adjacent cells still have missing physical characteristic data, the historical data that best matches the existing physical characteristic data, meteorological data, and surface data of the area is found in the historical combustion data, and based on the historical combustion data, the missing physical characteristic data in the area is completed.

[0062] Finally, the server integrates the burning intensity of each cell with the combustible burning state into the burning data, and associates the burning data to the corresponding position of the preset cell space model.

[0063] S103, based on the burning data of each cell in the preset cell space model, the pre-acquired meteorological data and the ground surface data, simulating the state transition process of each cell in the future time sequence through the preset fire spread prediction model, generating the first fire spread data of the fire field at each prediction time and the first burning data of each cell.

[0064] The preset fire spread prediction model includes a fire spread speed model and a cellular automaton model.

[0065] Specifically, the server first calls the fire spread speed model (based on the law of conservation of energy and the dynamics of combustible burning, the core parameters are taken from the measured reports of forestry research institutions and fire model specifications), and inputs the burning data, meteorological data and ground surface data of each cell into the model. After receiving the input data, the model integrates the combustible burning state as a key parameter into the calculation: if the cell crown layer is in a severe burning state, its heat radiation release intensity is higher, and the model increases the potential spread speed of the cell to the surrounding according to the burning degree corresponding coefficient; if the combustible material of the cell has been completely burned (the burning state is burned out), the model sets its potential spread speed to zero; if the burning posture is inclined or collapsed, because the contact area of the plant increases, the model corrects the direction weight of the spread speed according to the degree of vegetation collapse (such as additional increase of the spread probability of the collapse direction). At the same time, the model couples the influence of environmental factors such as wind speed (correcting the heat level transmission efficiency to affect the spread direction), slope (adjusting the uphill spread acceleration in a proportional relationship), aspect (correcting the moisture content parameter of the combustible material according to the difference in solar radiation), etc. on the spread speed, and finally calculates the spread speed of each cell to the surrounding cells.

[0066] Then, the cellular automaton model is called, and the spread speed of each cell to the surrounding, the burning data of each cell, the ground surface data and the meteorological data are input into the model. After receiving the input data, the model first defines the core state of the cell as unburned, burning and burned out, and refines the state attributes based on the layered burning of the combustible material and the posture.

[0067] Subsequently, the model constructs state transition rules. For the burning cells, first, the ignition contribution weight is calculated according to the layered combustion degree, then the direction weight determined by the spread speed is combined, if the combustion posture is inclined, the ignition probability of the cell in the same height direction of the inclined direction is additionally increased, if it is collapsed, the ignition probability of the cells in the grass layer covered by the collapsed posture is mainly increased, and the ignition probability of the cells in the upper crown layer is reduced; for the unburned cells, first, the corresponding ignition threshold is matched according to the layered flammable material type in the ground data, then the cumulative amount of thermal radiation of each layer of the adjacent burning cells is calculated, if the cumulative amount of a layer exceeds the threshold, the cell is first converted to a lightly burning state, and when multiple layers reach the threshold, the cells are burned synchronously according to the priority of “crown layer > shrub layer > grass layer”, and the initial posture is straight by default; for the burned-out cells, the ignition probability of each layer is set to zero, and the cells do not participate in state transition.

[0068] At the same time, the model combines the non-combustible area in the ground data to block the spread interaction between the cells in the non-combustible area and the surrounding cells during state transition, and the cells in the non-combustible area always remain in the state of “all layers unburned + no posture attribute”; and according to the layered initial biomass in the ground data and the current layered combustion intensity, the model calculates the combustion duration of each layer (the crown layer has the largest initial biomass, so even if the heavy burning duration is longer than that of the shrub layer and the grass layer, the grass layer has the smallest initial biomass, and the lightly burning can quickly burn out), when the burning time of a layer reaches the corresponding threshold, the layer is immediately converted to a burned-out state, and the remaining unburned layers continue to participate in the burning.

[0069] Then, the model divides the prediction period into equal time steps, and in each time step: first, the direction weight of the spread speed and the ignition probability coefficient are fine-tuned according to the latest weather data; then, the layered combustion state and the combustion posture of each cell are updated; finally, the state transition rules are executed to complete the iteration of the cell state.

[0070] After all the time steps are iterated, the model integrates the data of each cell to generate the output at each preset prediction time node: the first fire spread data includes the coordinates of the fire field boundary, the spread direction vector, and the spread speed gradient; the first combustion data includes the layered combustion state of each cell, the real-time combustion posture, the layered remaining biomass (the initial biomass minus the burned biomass), and the combustion phase identifier (determined according to the proportion of the remaining biomass and the combustion degree as “initial / vigorous / decaying combustion”).

[0071] S104, after obtaining the latest satellite data, identifying the actual combustion data of each cell in the monitoring time of the latest satellite data.

[0072] Specifically, the server interfaces with the satellite data distribution platform to obtain the latest satellite data in the same time window covering the target forest range in real time, and after obtaining the latest satellite data, the steps of S101-S102 are executed to obtain the actual combustion data of each cell in the monitoring time of the latest satellite data.

[0073] S105, based on the actual burning data of each cell and the corresponding first burning data, correct the simulation parameters and model input parameters in the preset fire spread prediction model to obtain a corrected fire spread prediction model and corrected model input parameters.

[0074] Specifically, the deviation of the actual burning data of each cell from the first burning data is calculated: for the burning intensity, the relative error formula (the absolute value of the difference between the actual value and the first burning data value divided by the actual value) is used to calculate the deviation; for the combustible state, the deviation of the burning degree is quantified in layers (the deviation value is determined according to the level difference of "severe > moderate > mild > burned out"), and the deviation of the burning posture is counted (the proportion of cells with inconsistent actual posture and predicted posture).

[0075] For the simulation parameters, if the burning intensity deviation exceeds the standard, adjust the "burning degree corresponding coefficient" in the fire spread speed model, and the adjustment amount is referenced from the historical deviation-parameter adjustment data set (this data set is generated by arranging the corresponding relationship between "deviation value-optimal parameter correction amount" in past fire cases, and the reliability is ensured by cross-validation); if the deviation of the spreading direction caused by the burning posture is significant, use the gradient descent algorithm to correct the "direction weight coefficient of inclined / collapsed posture" in the cellular automaton model, and the iteration step is set according to the parameter sensitivity analysis result (the step is smaller for parameters with high sensitivity); if the ignition probability deviation is large, the ignition threshold of the cellular automaton is corrected according to the cumulative amount of thermal radiation of the actual burning cells.

[0076] For the model input parameters, if the actual burning intensity is higher than the first burning data, combine the "burning intensity-wind speed" correlation model (constructed by referring to the influence of wind speed on thermal radiation release in forestry combustion experiments) to correct the wind speed input value, and adjust the humidity input value according to the physical law that "burning intensity is negatively correlated with humidity"; if the remaining biomass deviation is large, the initial biomass of the combustible material in the ground surface data is corrected by calculating the remaining biomass after actual burning (calculated from the actual burning intensity and the burning duration), and the correction formula is determined according to the mass consumption model of combustible burning (initial biomass = remaining biomass + burned amount, and the burned amount is calculated by integrating the actual burning intensity). The corrected initial biomass needs to meet the reasonable range of biomass of the combustible material in the ground surface database (the range is taken from the forestry resource survey specification).

[0077] During the correction process, the server applies physical constraints on the parameters (such as the wind speed correction value does not exceed the fluctuation range of the observation data of the same period meteorological station); after correction, the corrected parameters are substituted into the model to verify the simulation combined with the corrected input parameters, if the verification deviation is reduced to the acceptable range of the field, the corrected fire spread prediction model and the corrected model input parameters are determined; if the deviation is still excessive, the deviation calculation and parameter correction process is repeated until the accuracy requirement is met.

[0078] S106, re-predicting second fire spread data of each prediction time and second combustion data of each cell based on the corrected model input parameters by correcting the fire spread prediction model.

[0079] Specifically, based on the corrected model input parameters, the steps of S103 are executed to re-predict the second fire spread data of each prediction time after the monitoring time and the second combustion data of each cell. At the same time, according to the actual combustion data of each cell of each monitoring time before the monitoring time, the fire spread data of each prediction time before the monitoring time and the combustion data of each cell are corrected to obtain the second fire spread data of each prediction time before the monitoring time and the second combustion data of each cell.

[0080] In the embodiments of the present application, by taking the stable structural change feature as the physical reference, the thermodynamic and spectroscopic features susceptible to interference factors such as smoke are cross-checked and corrected, effectively dealing with the measurement distortion or uncertainty problem caused by smoke shielding, atmospheric interference or sensor saturation of a single remote sensing data source (such as relying only on thermal radiation or spectral index). At the same time, when the observation data is incomplete or there are inaccurate feature data that cannot be identified, the most physically reliable assumed combustion feature is derived by using known data matching, ensuring that the combustion feature input into the subsequent spread model has the highest physical reliability, avoiding the inaccuracy caused by blind derivation, thereby providing more accurate initial conditions for the entire prediction process, and further improving the accuracy of fire spread trend prediction. After the initial prediction according to the initial conditions, the simulation parameters and input parameters of the model are corrected according to the actual combustion data in the latest satellite data after each time the latest satellite data is obtained, and the fire spread data is re-predicted by using the corrected model and parameters, so that the prediction model and model input data can be calibrated in real time according to the real dynamic evolution of the fire site, improving the accuracy of fire spread trend prediction.

[0081] The method of the embodiments of the present application will be further described below in combination with Figure 2 .

[0082] Please refer to Figure 2 , which is another flowchart of the fire spread trend prediction method in the embodiments of the present application.

[0083] S201, identifying a burning area in the target forest and physical feature data of each burning area based on satellite data of the target forest.

[0084] S202, inversely deducing the combustion data of the corresponding cell in the preset cell space model of the target forest by a preset combustion process inversion algorithm based on the physical feature data of each burning area.

[0085] Steps S201-S202 are the same as Figure 1The steps S101-S102 in the illustrated embodiment are similar, and reference can be made to the description in steps S101-S102, which will not be repeated here.

[0086] S203, based on the combustion data of each cell in the preset cell space model, a three-dimensional fire-induced disturbance field is generated.

[0087] Specifically, the server first expands and constructs a three-dimensional atmospheric grid model covering the entire fire field area based on the cell space model of the target forest. The model is divided into multiple levels in the vertical direction according to a preset resolution, extending from the ground to the top of the atmospheric boundary layer. Subsequently, the server quantifies the combustion intensity data of each burning cell into the heat flux and water vapor flux on the ground surface upward according to the thermodynamic conversion relationship. Then, using the buoyancy-driven flow model in computational fluid dynamics, the server simulates how these fluxes generate rising thermal plumes in the three-dimensional grid and calculates the vertical gradient changes in local atmospheric temperature, pressure and humidity caused thereby. At the same time, the server simulates the injection, diffusion and transport process of aerosol particles generated by combustion under the driving of the thermal plume according to the smoke generation rate implied by the combustion state of each cell of combustible material, thereby generating an aerosol concentration distribution in the three-dimensional grid. Finally, the server integrates the temperature disturbance value, pressure disturbance value, humidity disturbance value, aerosol concentration and vertical and horizontal wind speed component disturbance caused by thermal convection at each grid point to form a dynamically evolving three-dimensional fire-induced disturbance field, which depicts the real-time influence of the fire on the local microclimate environment.

[0088] S204, based on the three-dimensional fire-induced disturbance field, the meteorological data and the physical data of each combustible material in the pre-acquired ground surface data are respectively coupled and corrected to generate corrected meteorological data and corrected ground surface data at the current prediction time.

[0089] Specifically, for the meteorological data, the server superimposes the disturbance quantities such as wind speed, temperature and humidity at each grid point in the three-dimensional fire-induced disturbance field with the pre-acquired background meteorological data corresponding to the spatial position to generate corrected meteorological data affected by the fire. For the physical data of each combustible material in the ground surface data, based on the combustion intensity of the burning cell, the server applies a thermal radiation transfer model to calculate the received radiation heat of the surrounding unburned cells; then inputs this heat into a combustible material drying model established based on the principle of energy balance, which dynamically calculates the rate of water loss of the combustible material due to baking in combination with the corrected near-surface meteorological data (such as higher temperature and lower humidity), and finally obtains the corrected moisture content of the combustible material preheated and dried by the fire field environment.

[0090] S205, based on the combustion data of each cell, the corrected meteorological data and the corrected ground surface data, simulating the state transition process of each cell in the future time sequence through a preset fire spread prediction model, generating first fire spread data of the fire field at the current prediction time and first combustion data of each cell.

[0091] Specifically, the implementation process in S103 is performed, and only the state transition process of each cell from the last prediction time (if the current prediction time is the initial prediction time, the last prediction time is the prediction start time) to the current prediction time is predicted, and first fire spread data of the fire field at the current prediction time and first combustion data of each cell are generated.

[0092] S206, based on the combustion intensity and the combustible material combustion state in the first combustion data of each cell, identifying the fire source cell and the corresponding fire source physical characteristics of the fire.

[0093] Specifically, the server traverses all cells whose state is "burning" at the current prediction time, and based on the combustion data of the cell, a fire generation rule established according to the principles of combustion dynamics and aerodynamics is used for screening to identify fire source cells that have the ability to produce fire. The core condition of this rule is that the combustion intensity of the cell must exceed a critical threshold, and the combustible material combustion state is "heavy crown layer burning", because only this high-intensity crown fire can produce a powerful upward airflow to lift the burning debris. The critical intensity threshold is preset according to different combustible material types (such as pine trees, eucalyptus trees) and corresponding experimental data of fuel branch breakage and peeling. When a fire source cell is identified, the server matches and extracts a set of quantitative fire physical characteristics from a preset fire physical characteristic library according to the combustible material type and real-time combustion intensity of the cell. The characteristic library is constructed from a large number of wind tunnel experiments and field fire observation data, and the query results include the characteristic particle size distribution, mass, shape coefficient and initial combustion temperature of the fire (i.e. the burning debris).

[0094] S207, based on the fire source cell, the fire physical characteristics, the corrected meteorological data and the corrected ground surface data, identifying the state change cell, the change data and the change probability at the current prediction time according to the preset fire generation rule.

[0095] Specifically, for each firebrand source cell, a firebrand trajectory simulation process based on a Lagrangian particle tracking model is performed. The model takes the extracted firebrand physical characteristics (mass, size, etc.) as particle attributes, and the geographical coordinates and tree canopy height of the firebrand source cell as the initial release point. The model calculates the flight trajectory of the firebrand particle in three-dimensional space according to Newton's law of motion, coupled with real-time corrected three-dimensional wind field data (corrected meteorological data from S204), as well as gravity and air resistance. During the flight process, the model also synchronously calculates the particle's combustion extinction process, i.e., the decay of particle temperature over time. When a particle trajectory intersects with the ground, the server records the landing cell and determines whether ignition can be successfully achieved based on a pre-set ignition probability model. The ignition probability model is a function that takes the remaining temperature of the particle at landing and the corrected fuel characteristics (such as moisture content, fuel type) of the landing cell as input, and outputs an ignition success rate between zero and one. The server simulates the trajectory of a large number of particles for each firebrand source cell by the Monte Carlo method, and finally counts all successful ignition events to identify the cells where ignition occurs as state-changing cells, with the changed data being the state changing from "unburned" to "burning", and the changed probability being the total probability of successful ignition of the cell by particles from all firebrand sources.

[0096] S208, based on the state-changing cells, the changed data, and the changed probability, correcting the first fire spread data and the first burning data of each cell at the current prediction time.

[0097] Specifically, for each state-changing cell, the state of the cell in the first burning data is updated from "unburned" to "burning" according to the changed data. At the same time, the changed probability of the cell is added as a new attribute to the corresponding first burning data. Then, the first fire spread data is corrected, and the newly added burning cells are included in the calculation range of the fire field boundary, and the total area, perimeter, and main vector of the spread direction of the fire field are recalculated and updated to ensure that the "jumping" spread caused by firebrands can be accurately reflected in the fire field situation output at the current prediction time.

[0098] S209, taking the first burning data of each cell, the corrected meteorological data, and the corrected ground surface data in the last prediction time adjacent to each prediction time as the burning data, meteorological data, and ground surface data of each cell in each prediction time, respectively, to iteratively generate the first fire spread data and the first burning data of each cell at each prediction time after the current prediction time.

[0099] Specifically, at the beginning of each new prediction time step, the server takes the "first combustion data" generated at the end of the last prediction time step and corrected by the fire whirl, as the initial "combustion data" of the current time step. At the same time, it updates the background meteorological field according to the latest weather forecast data, and executes the procedures of S203 and S204 again to generate a new three-dimensional fire-induced disturbance field based on the new combustion data, and to couple to generate the "corrected meteorological data" and "corrected ground surface data" of the current time step. Subsequently, the server repeats the whole process of S205 to S208, i.e. uses these updated inputs to perform the spread simulation by the preset fire spread prediction model, and performs the fire whirl identification and correction, so as to generate the "first fire spread data" and "first combustion data" of the current time step. This cycle is repeated until the entire preset prediction time is covered, and a complete time series of fire spread considering the dynamic feedback of fire and atmosphere and the effect of fire whirl is constructed.

[0100] S210, after obtaining the latest satellite data, identifying the actual combustion data of each cell in the monitoring time of the latest satellite data.

[0101] S211, based on the deviation of the actual combustion data of each cell and the corresponding first combustion data, correcting the simulation parameters and model input parameters in the preset fire spread prediction model to obtain a corrected fire spread prediction model and corrected model input parameters.

[0102] S212, based on the corrected model input parameters, re-predicting the second fire spread data of each prediction time and the second combustion data of each cell by the corrected fire spread prediction model.

[0103] Steps S210-S212 are similar to steps S104-S106 in the embodiment shown in Figure 1 The steps S104-S106 in the embodiment shown in

[0104] S213, if the prediction deviation reason of the monitoring time includes a new fire point, identifying the feasibility probability of the new fire point being formed by the fire whirl based on the preset fire whirl generation rule, the actual meteorological data and the actual ground surface data of the monitoring time.

[0105] Specifically, the server first identifies whether there is a new fire point in the monitoring time: the actual combustion data of the monitoring time and the first combustion data of the last prediction time (i.e., the last prediction time before the monitoring time) are called, and are compared one by one according to the cell space coordinates. If a cell is in the first combustion data and is in the actual combustion data, it is preliminarily marked as a candidate new fire point. Then, the shortest spatial distance between the candidate new fire point and the burning cell of the last prediction time is calculated, and the preset normal fire spread speed threshold (the threshold is taken from the measured report of the maximum natural spread speed of different combustible types published by forestry research institutions, and the maximum possible spread distance is calculated according to the time interval between the monitoring time and the last prediction time) is combined. If the candidate new fire point is within the maximum possible spread distance, it is determined to be a fire point formed by natural spread and is excluded.

[0106] If the prediction deviation reason of the monitoring time includes a new fire point, the server selects a cell with a combustion intensity exceeding a critical threshold (determined according to the combustible type around the new fire point and the forestry branch experiment data) and a heavy burning crown layer from the actual combustion data in a time window corresponding to the typical burnout duration of the fire particle at the monitoring time before the monitoring time, as a potential fire source. From the preset fire particle characteristic library (constructed by wind tunnel experiments and field observations), the physical characteristics (particle size, mass, initial temperature, etc.) of the fire particle are matched, combined with the actual three-dimensional wind field, temperature, and humidity data at the monitoring time, and the Lagrangian particle tracking model is used to simulate the trajectory and temperature decay (humidity corrected heat loss) of the particle from the potential fire source to the new fire point. Through multiple simulations by the Monte Carlo method, the proportion of the number of times that the particle landing temperature is higher than the ignition point of the combustible material of the new fire point (adjusted according to the type and water content) is calculated, and then the feasibility probability is obtained by correcting the proportion according to the blocking ratio of the non-combustible area between the fire source and the new fire point and whether the particle flight exceeds the burnout time.

[0107] S214、If the feasibility probability is lower than the preset fire threshold, determine whether the new fire point belongs to a smoldering fire.

[0108] If the feasibility probability is lower than the preset fire threshold, determine whether the new fire point belongs to a smoldering fire based on the preset smoldering fire rules and the historical fire data before the monitoring time. If yes, execute the steps of S215; if no, execute the steps of S210.

[0109] Specifically, the server first queries the historical fire database including the current fire to determine whether the cell where the new fire point is located is recorded as a burned-out area in a fire (including the current fire) before the monitoring time.

[0110] If the cell has no fire history, all the historical burned-out cells in the neighborhood region corresponding to the new fire cell are identified, the neighborhood region being preset according to the known maximum horizontal spreading rate of the smoldering fire and the time interval under the soil type and organic matter conditions of the region. If there is no historical burned-out cell in the neighborhood region, it is determined that the new fire point does not belong to smoldering fire rekindling.

[0111] If there is a fire history or there is a burned-out cell in the neighborhood region, the surface data of the cell is retrieved, and it is determined whether it belongs to smoldering fire rekindling based on a preset smoldering fire rekindling rule. The rule is a multi-dimensional condition determination set based on surface physical properties, which is constructed in advance based on combustion science research and statistical analysis of a large number of historical smoldering fire rekindling cases, and defines a plurality of surface physical parameters directly related to smoldering fire potential, such as soil organic matter content and humus layer thickness, and sets clear critical thresholds for each parameter. If the cell meets the preset smoldering fire rekindling rule, it is determined to belong to smoldering fire rekindling.

[0112] S215, based on the preset smoldering fire rekindling rule, the second fire spreading data and the second burning data of each cell, identifying the rekindling cell and the corresponding rekindling time that first meet the preset smoldering fire rekindling rule after the monitoring time.

[0113] Specifically, first, based on the preset smoldering fire rekindling rule, the rekindling reference features are extracted from the surface data of the new fire point. The server uses the preset smoldering fire rekindling rule as a template to extract specific numerical values corresponding to all parameters in the rule from the surface data of the new fire point cell that has been confirmed to be smoldering fire rekindling. These numerical values and corresponding parameters are integrated into a multi-dimensional feature vector to obtain the rekindling reference features.

[0114] Secondly, based on the second burning data and the surface data of each cell from the fire starting time to the predicted cut-off time, the cells similar to the rekindling reference features in all cells are identified to obtain a high-risk rekindling cell set. All cells in the preset cell space model are traversed, and cells with a fire history or a burned-out cell in the neighborhood region and a current state of burned-out are selected as candidate cells. For each candidate cell, the surface physical properties corresponding to each cell are extracted from the surface data to construct a physical feature vector. Then, a preset multi-dimensional vector similarity measurement algorithm is used to calculate the similarity score between the physical feature vector of the candidate cell and the rekindling reference feature vector. If the score exceeds a preset similarity threshold (the threshold is set according to the statistical distribution of the surface features of historical rekindling cases and non-rekindling cases, and is designed to ensure that the selected cells are highly similar to the confirmed rekindling points in terms of surface conditions), the cell is added to the high-risk rekindling cell set.

[0115] Third, based on the second combustion data of each high-risk rekindling cell at each prediction time, the smoldering data of each high-risk rekindling cell at each prediction time after the monitoring time is calculated. For each high-risk rekindling cell, the historical fire records of the cell and each adjacent cell in the spatial neighborhood are queried, and for each historical fire, the combustion intensity and duration data of the cell and the adjacent cells are extracted. Then, these historical combustion data are converted into initial heat through a preset energy conversion function based on the type of combustible and the combustion efficiency. At the same time, according to the spatial adjacency relationship between the cells and the predetermined soil thermal conductivity, the lateral heat flux generated by the combustion of adjacent cells is calculated and superimposed on the initial heat of the target cell. Then, through a preset time decay algorithm, with the interpolated historical meteorological data as input, the initial heat is calculated in time steps, taking into account the dissipation due to heat conduction to the deep soil and the energy loss caused by the change in water content due to surface evaporation and precipitation infiltration, to calculate the underground residual latent heat and the baseline of the combustible dryness degree accumulated by all historical fires before the current fire occurs.

[0116] Then, the server extracts the second combustion data of the cell and the adjacent cells in the current fire, uses the same energy conversion function and lateral heat flux calculation method to quantify the instantaneous strong heat disturbance directly applied to the high-risk rekindling cell and its neighborhood by the current fire, superimposes this strong heat disturbance on the previously calculated historical cumulative latent heat and dryness degree baseline to obtain the initial smoldering data at the monitoring time point. Finally, taking the initial smoldering data as the starting point, the time decay algorithm is called again to drive the interpolated meteorological data at each future prediction time point, and the underground temperature and moisture content evolution sequence of the cell at each prediction time after the monitoring time is calculated in time steps, which constitutes the smoldering data of the cell.

[0117] Finally, based on the smoldering data of each high-risk rekindling cell, the rekindling cell that first meets the preset smoldering rekindling rule and the corresponding rekindling time are identified through a preset smoldering rekindling rule. Starting from the first prediction time point after the monitoring time, the smoldering data of each high-risk rekindling cell at this time is compared with the corresponding threshold in the preset smoldering rekindling rule to determine whether all the constraint conditions defined in the rule are met. Once a cell is identified as meeting the condition combination at a prediction time point, the cell is taken as the rekindling cell, the prediction time point is taken as the corresponding rekindling time, and the subsequent time of the cell is checked.

[0118] S216, based on the rekindling cell and the corresponding rekindling time, the second fire spread data at each prediction time after the rekindling time and the second combustion data of each cell are re-predicted to generate the third fire spread data at each prediction time and the third combustion data of each cell.

[0119] Specifically, based on the smoldering cell in the smoldering time, the corresponding preset combustion data is searched as the combustion data of the smoldering cell, and then the combustion data of each cell in the predicted time corresponding to the smoldering time is updated according to the combustion data of the smoldering cell.

[0120] Then, based on the updated combustion data of each cell, the steps of S203-S209 are performed to obtain the third fire spread data of each predicted time and the third combustion data of each cell. In the prediction process, before each model simulation, the smoldering cell identification process in S215 is added, and the combustion data of each cell is dynamically updated according to the identification result, and then the model simulation prediction is performed.

[0121] In the embodiments of the present application, the dynamic influence of fire on local micro-meteorology and physical properties of combustible materials is quantified by three-dimensional fire-induced disturbance field, and an iterative mechanism is combined to ensure that the prediction input data matches the real-time state of the fire field. At the same time, the generation, propagation and ignition process of flying fire are included, covering the jump spread scenario, and by distinguishing the causes of new fire points, the potential fire sources of smoldering and the smoldering time are identified, solving the problem of ignoring the interaction between fire and environment and missing the jump / implicit fire spread in traditional models, and further improving the comprehensiveness and accuracy of fire spread prediction.

[0122] The fire spread trend prediction method in the embodiments of the present application is described above, and the server in the embodiments of the present application will be described in detail in combination with the above fire spread trend prediction method.

[0123] Please refer to Figure 3 , which is an exemplary hardware structure diagram of the server in the embodiments of the present application.

[0124] In some embodiments, the server 300 includes a computer device, which can be a terminal device. The computer device includes a processor 301, a memory 302, a communication module 303, an input device 304, and an output device 305 connected by a system bus. The processor 301 of the computer device is used to provide calculation and control capability. The memory 302 of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database is used to store data. The communication module 303 of the computer device is used to transmit satellite data and the like. The input device 304 of the computer device is used to receive satellite data, user instructions and the like. The output device 305 of the computer device is used to display fire spread prediction results and the like. The computer program is executed by the processor 301 to implement the fire spread trend prediction method in the embodiments of the present application.

[0125] Those skilled in the art can understand, Figure 3The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0126] In some embodiments of the present application, a computer readable storage medium is provided, including instructions which, when executed on the server 300, can cause the server 300 to perform the fire spread trend prediction method in the embodiments of the present application.

[0127] In some embodiments of the present application, a computer program product is also provided, which, when executed on the server 300, causes the server 300 to perform the fire spread trend prediction method in the embodiments of the present application.

[0128] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the foregoing embodiments of the present application have been described in detail, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0129] In the above-described embodiments, according to the context, the term "when" can be interpreted as meaning "if" or "after" or "in response to determining" or "in response to detecting". Similarly, according to the context, the phrase "upon determining" or "if detecting (the stated condition or event)" can be interpreted as meaning "if determining" or "in response to determining" or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)".

[0130] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk) and the like.

[0131] Those of ordinary skill in the art understand that all or part of the processes in the above embodiments can be implemented by a computer program to instruct the relevant hardware, which can be stored in a computer readable storage medium. The program can include the processes of the above method embodiments when executed. The aforementioned storage medium includes ROM or random access memory (RAM), magnetic disk or optical disk, and various media that can store program codes.

Claims

1. A method for predicting the spread trend of a fire, characterized in that, include: Based on satellite data of the target forest, the burning areas in the target forest and the physical characteristic data of each burning area are identified. The physical characteristic data includes thermodynamic characteristics, spectroscopic characteristics and structural change characteristics. Based on the physical characteristic data of each combustion zone, the combustion data of the corresponding cells in the preset cellular space model of the target forest are obtained by reverse deduction through a preset combustion process inversion algorithm. The combustion data includes combustion intensity and combustible combustion state. Based on the combustion data of each cell in the preset cellular space model, the pre-acquired meteorological data and surface data, the preset fire spread prediction model is used to simulate the state transition process of each cell in the future time series, and generate the first fire spread data of the fire at each predicted time and the first combustion data of each cell. The preset fire spread prediction model includes a fire spread rate model and a cellular automata model. After acquiring the latest satellite data, identify the actual combustion data of each cell within the monitoring time of the latest satellite data; Based on the deviation between the actual combustion data of each cell and the corresponding first combustion data, the simulation parameters and model input parameters in the preset fire spread prediction model are corrected to obtain the corrected fire spread prediction model and the corrected model input parameters. Based on the input parameters of the modified model, the second fire spread data and the second combustion data of each cell are re-predicted at each prediction time through the modified fire spread prediction model.

2. The method according to claim 1, characterized in that, Based on the combustion data of each cell in the preset cellular space model, pre-acquired meteorological data, and surface data, a preset fire spread prediction model is used to simulate the state transition process of each cell in the future time series, generating the first fire spread data of the fire scene and the first combustion data of each cell at each predicted time, specifically including: Based on the combustion data of each cell in the preset cellular space model, a three-dimensional fire-induced disturbance field is generated. Based on the three-dimensional fire-induced disturbance field, the physical data of each combustible in the pre-acquired meteorological data and surface data are coupled and corrected to generate corrected meteorological data and corrected surface data for the current prediction time. Based on the combustion data of each cell, the corrected meteorological data, and the corrected surface data, a preset fire spread prediction model is used to simulate the state transition process of each cell in the future time series, generating the first fire spread data of the fire at the current prediction time and the first combustion data of each cell. The first fire spread data, corrected meteorological data, and corrected surface data of each cell in the previous prediction time adjacent to each prediction time are used as the fire data, meteorological data, and surface data of each cell in each prediction time, respectively. The first fire spread data and the first fire data of each cell in each prediction time after the current prediction time are generated iteratively.

3. The method according to claim 2, characterized in that, After the step of generating the first fire spread data and the first combustion data of each cell at the current prediction time by simulating the state transition process of each cell in the future time series using a preset fire spread prediction model based on the combustion data of each cell, the corrected meteorological data, and the corrected surface data, the method further includes: Based on the combustion intensity and combustible state in the first combustion data of each cell, the flying fire source cells and the corresponding flying fire sources are identified as having flying fire physical characteristics. Based on the fire source cell, the fire physical characteristics, the corrected meteorological data, and the corrected surface data, the state change cell, change data, and change probability for the current prediction time are identified according to the preset fire generation rules. Based on the state-changing cells, the change data, and the change probability, the first fire spread data and the first combustion data of each cell at the current prediction time are corrected.

4. The method according to claim 1, characterized in that, After the step of re-predicting the second fire spread data and the second combustion data of each cell at each prediction time based on the input parameters of the modified model and through the modified fire spread prediction model, the method further includes: If the reasons for the prediction deviation of the monitoring time include the addition of new fire points, the feasibility probability of the new fire points being formed by flying fire is identified based on the preset flying fire generation rules, the actual meteorological data and the actual surface data of the monitoring time. If the probability of feasibility is lower than the preset flying fire threshold, based on the preset smoldering reignition rule and the historical fire data before the monitoring time, it is determined whether the new fire point belongs to smoldering reignition. If so, based on the preset smoldering reignition rule, the second fire spread data, and the second combustion data of each cell, identify the reignition cell that first meets the preset smoldering reignition rule after the monitoring time and the corresponding reignition time; Based on the reignition cell and the corresponding reignition time, the second fire spread data and the second combustion data of each cell at each predicted time after the reignition time are re-predicted to generate the third fire spread data and the third combustion data of each cell at each predicted time.

5. The method according to claim 4, characterized in that, The process of identifying reignited cells and their corresponding reignition times that first meet the preset smoldering reignition rules after the monitoring time, based on the preset smoldering reignition rules, the second fire spread data, and the second combustion data of each cell, specifically includes: Based on the preset rules for smoldering fire reignition, reignition baseline features are extracted from the surface data of the newly added fire points; Based on the second combustion data and surface data of each cell from the fire start time to the predicted end time, cells that are similar to the reignition benchmark characteristics are identified among all cells to obtain a high-risk reignition cell set. Based on the second combustion data of each high-risk re-ignition cell in the high-risk re-ignition cell set at each predicted time, calculate the smoldering data of each high-risk re-ignition cell at each predicted time after the monitoring time; Based on the smoldering data of each high-risk re-ignition cell, the re-ignition cells that first meet the preset smoldering re-ignition rules and their corresponding re-ignition times are identified by using preset smoldering re-ignition rules.

6. The method according to claim 1, characterized in that, Following the step of identifying burn zones and their physical characteristics based on satellite data of the target forest, the method further includes: If there is a first combustion zone with non-empty structural change characteristics, theoretical combustion characteristics of each first combustion zone are generated based on the structural change characteristics and surface data of each first combustion zone. Based on the theoretical combustion characteristics and surface data of each first combustion zone, the theoretical thermodynamic characteristics and theoretical spectroscopic characteristics of each first combustion zone are calculated. Based on the thermodynamic characteristics, meteorological data, and theoretical thermodynamic characteristics of each first combustion zone, the first combustion characteristics of each first combustion zone are generated. Based on the spectroscopic characteristics and meteorological data of each first combustion zone and the theoretical spectroscopic characteristics, second combustion characteristics of each first combustion zone are generated; Based on the theoretical combustion characteristics, first combustion characteristics, and second combustion characteristics of each first combustion zone, the physical characteristic data of each first combustion zone are corrected respectively.

7. The method according to claim 6, characterized in that, Following the step of identifying burn zones and their physical characteristics based on satellite data of the target forest, the method further includes: If there is a second combustion region where the structural change features are empty but the thermodynamic and spectroscopic features are not empty, candidate combustion features are generated based on the thermodynamic features, spectroscopic features, meteorological data, and surface data of the second combustion region through reverse physics deduction. The candidate combustion features refer to combustion features that are generated on the surface and under the atmospheric conditions described by the meteorological data, and can be collected by satellite sensors and characterized as the thermodynamic and spectroscopic features. Based on the candidate combustion characteristics, meteorological data, and surface data of the second combustion zone, the degree of conformity between the candidate combustion characteristics and the combustion characteristics of each combustible in the second combustion zone is calculated. If the compliance degree is lower than the preset compliance threshold, based on the preset combustion feature library, a first target combustion feature set that matches the thermodynamic features and meteorological data of the second combustion region, and a second target combustion feature set that matches the spectroscopic features and meteorological data of the second combustion region are searched. The combustion feature with the highest degree of conformity between the first target combustion feature set and the second target combustion feature set and the combustion characteristics of each combustible in the second combustion region is selected as the hypothetical combustion feature, and the degree of conformity of the hypothetical combustion feature is greater than the preset conformity threshold. Based on the assumed combustion characteristics, the second combustion zone and its corresponding physical characteristics are modified.

8. A server, characterized in that, include: One or more processors and memory; The memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the server to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium storing computer instructions, characterized in that, When the computer instructions are executed on the server, the server causes the server to perform the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the server, the server performs the method as described in any one of claims 1-7.

Images