A Smart Forest Fire Management Method and System Based on AI Analysis

Abstract

This invention relates to the field of fire management technology, specifically to a smart forest fire management method and system based on AI analysis. The method includes: constructing a target digital twin of the target forest area based on multi-source sensing information; substituting the current forest area status represented by the target digital twin into a preset fire point prediction model to calculate fire risk warning information; performing fire simulation within the target digital twin based on the fire risk warning information and / or real-time fire information to generate fire spread prediction information; and making decisions based on the fire spread prediction information to generate fire management information. This solves the technical problems of existing forest fire management methods that rely on passive perception and static situation analysis of existing smoke and fire, making it difficult to quantify and simulate the dynamic evolution of forest fire risk, resulting in delayed warnings and passive intervention.

Description

Technical Field

[0001] This invention relates to the field of fire management technology, specifically to a smart forest fire management method and system based on AI analysis. Background Technology

[0002] As an important ecosystem and natural resource, forests play a vital role in fire prevention and control, which is crucial for protecting the ecological environment, safeguarding people's lives and property, and promoting sustainable social development.

[0003] Currently, existing research has proposed and deployed various sensors and monitoring equipment to collect multi-source information such as images and meteorological data in forest areas, and used artificial intelligence models to process real-time data to achieve forest fire monitoring, early warning, and situation analysis. However, the proposed method is essentially a passive "perception-response" management approach, which relies on the detection and identification of smoke and fire to trigger subsequent processes. Its decision support heavily depends on human experience in judging the current static situation, and it cannot quantify the evolution of fire risk in forest areas before a fire occurs, nor can it conduct high-confidence simulations of the dynamic spread of fire after it occurs. This results in delayed risk warnings, resource allocation relying on experience, and difficulty in achieving precise intervention before smoke and fire appear. Summary of the Invention

[0004] To address the technical problems of existing forest fire management methods that rely on passive perception and static situation analysis of smoke and fire, making it difficult to quantify and simulate the dynamic evolution of forest fire risk and resulting in delayed early warning and passive intervention, this application provides a smart forest fire management method and system based on AI analysis.

[0005] The intelligent forest fire management method and system based on AI analysis provided in this application adopts the following technical solution: A smart forest fire management method based on AI analysis includes: Based on multi-source sensing information of the target forest area, construct a target digital twin of the target forest area; Substitute the current forest area status represented by the target digital twin into the preset fire point prediction model to calculate fire risk warning information; Based on fire risk warning information and / or real-time fire information, fire simulation is performed in the target digital twin to generate fire spread prediction information; Decisions are made based on fire spread prediction information to generate fire management information.

[0006] Furthermore, the steps based on multi-source sensing information of the target forest area include: By deploying sensor networks and remote sensing platforms in the target forest area, polarized light information, multispectral information, and thermal radiation information are collected. By suppressing interference in polarized light information and multispectral information, spectral feature information is obtained. By performing spatiotemporal registration and feature fusion of spectral feature information and thermal radiation information, multi-source sensing information is obtained.

[0007] Furthermore, based on multi-source sensing information of the target forest area, the steps for constructing a target digital twin of the target forest area include: Acquire geographic information data and historical ecological data of the target forest area, combine multi-source sensing information to perform 3D modeling and data fusion, and construct a 3D model of the forest area. A pre-defined physical mechanism model is embedded into a 3D model of the forest area to generate an initial digital twin framework; Based on the multi-source sensing information at the current moment, the state variables of the initial digital twin framework are driven and calibrated to obtain the target digital twin.

[0008] Furthermore, the steps of embedding a pre-defined physical mechanism model into the 3D model of the forest area to generate an initial digital twin framework include: By pre-setting vegetation water transport calculation rules, the tree species distribution information in geographic information data and historical ecological data is calculated to obtain the initial field of vegetation water status. Based on the vegetation moisture state in the initial field of vegetation moisture state, the change of combustible load is calculated to obtain combustible load distribution information. Climate field information of the forest area is obtained by combining information on the distribution of combustible load with meteorological parameters in the target forest area. Based on the initial field of vegetation moisture status, combustible load distribution information and forest climate field information, data coupling and parameter binding are performed on the three-dimensional model of the forest area to generate an initial digital twin framework.

[0009] Furthermore, the target digital twin is composed of multiple spatial units. The steps for calculating fire risk warning information by substituting the current forest area status represented by the target digital twin into a preset fire point prediction model include: Extract forest area status from the target digital twin, which represents parameters such as vegetation moisture, combustible load, and forest climate. The vegetation moisture parameter, combustible load parameter, and forest climate parameter are weighted and combined with the weight coefficients obtained by training based on historical fire sample data to calculate the comprehensive fire risk index of each spatial unit. Each comprehensive fire risk index is substituted into a preset nonlinear probability mapping function for transformation to obtain the fire probability of each spatial unit in the future time period. After filtering out the target fire probabilities that exceed the preset fire probabilities from the various fire probabilities, the geographical location of the target spatial unit corresponding to the target fire probability is obtained, and fire risk warning information is generated. The fire risk warning information includes the geographical location of each target spatial unit and the target fire probability corresponding to the target spatial unit.

[0010] Furthermore, based on fire risk warning information and / or real-time fire information, the steps for generating fire spread prediction information by performing fire simulation in the target digital twin include: Based on fire risk warning information and / or real-time fire information, set the initial fire point location and initial fire intensity on the target spatial unit corresponding to the target digital twin; Based on the current forest area status and initial fire intensity on each target spatial unit, calculate the fire spread direction on the target digital twin starting from the initial fire point, as well as the fire spread speed and fire intensity changes in each fire spread direction; Spatiotemporal integration and iterative updates are performed on the changes in fire spread rate and fire intensity to obtain the fire boundary location and fire point intensity distribution of each fire field in the future time period. The terrain features of the target forest area are extracted from the target digital twin. Combined with the terrain features, the location of the fire boundary of each fire site and the intensity distribution of fire points in each fire site, the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark are calculated. The fire spread prediction information includes the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark.

[0011] Furthermore, the steps for generating fire management information by making decisions based on fire spread prediction information include: Determine the predicted fire points and the protection sequence between each predicted fire point based on the fire spread path, fire spread channel and / or time point; Based on the protection sequence and pre-positioned fire-fighting resources, the scheduling and matching scheme between each predicted fire point and the pre-positioned fire-fighting resources is calculated; Based on the scheduling and matching scheme, path planning and time coordination are performed to obtain fire management information for pre-set fire-fighting resources.

[0012] This application also provides an AI-based intelligent forest fire management system, including: The twin construction module is used to construct a target digital twin of the target forest area based on multi-source sensing information of the target forest area; The data calculation module is used to input the current forest area status represented by the target digital twin into the preset fire point prediction model to calculate fire risk warning information; The data prediction module is used to perform fire simulation in the target digital twin based on fire risk warning information and / or real-time fire information, and generate fire spread prediction information. The data generation module is used to make decisions based on fire spread prediction information and generate fire management information.

[0013] Beneficial effects achieved: This application provides an AI-based intelligent forest fire management method, comprising: constructing a target digital twin of the target forest area based on multi-source sensing information of the target forest area; substituting the current forest area status represented by the target digital twin into a preset fire point prediction model to calculate fire risk warning information; performing fire simulation in the target digital twin based on the fire risk warning information and / or real-time fire information to generate fire spread prediction information; and making decisions based on the fire spread prediction information to generate fire management information.

[0014] In this application, a dynamic and computable target digital twin is constructed as the core, enabling real-time simulation and quantitative representation of the current forest area state driven by physical mechanisms. This transforms the management basis from static observation data to dynamic physical state variables, allowing the pre-set fire point prediction model to directly calculate based on the pre-forest area state represented by the target digital twin. This allows for the identification of high-risk fire areas and the generation of fire risk warnings before open flames or dense smoke appear, shifting from perceiving already burning to predicting flammability and solving the problem of delayed warnings. Furthermore, upon obtaining fire risk warnings and / or real-time fire information, a fire spread prediction based on physical laws is performed on the target digital twin, replacing static predictions relying on human experience. This generates fire spread prediction information, transforming the assessment of fire development from static analysis to dynamic prediction. Finally, fire management information is generated based on fire spread prediction information to guide the pre-positioning of fire resources before the fire arrives and the precise deployment of defenses along critical paths. This transforms fire management from a passive response to an already occurring fire to an active intervention in key aspects of fire development, and realizes a shift in management model from relying on post-event perception and static analysis to dynamic calculation and active intervention. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of an AI-based smart forest fire management method proposed in this application. Figure 2 A flowchart illustrating the steps involved in constructing the target digital twin for this application; Figure 3 This is a flowchart illustrating the steps involved in calculating fire risk warning information in this application. Figure 4 A flowchart illustrating the steps involved in generating fire spread prediction information for this application; Figure 5 A flowchart illustrating the steps involved in generating fire management information for this application; Figure 6 This is a schematic diagram of the modules of the AI-based smart forest fire management system proposed in this application.

[0016] Explanation of reference numerals in the attached figures: 10. Twin construction module; 20. Data calculation module; 30. Data prediction module; 40. Data generation module; 50. Data acquisition module. Detailed Implementation

[0017] The following combination Figures 1-6 This application will be described in further detail.

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0020] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0021] This application discloses an AI-based intelligent forest fire management method.

[0022] Please refer to Figure 1The intelligent forest fire management method based on AI (Artificial Intelligence) analysis proposed in this embodiment includes steps S10 to S30: Step S10: Based on the multi-source sensing information of the target forest area, construct a target digital twin of the target forest area.

[0023] In this step, by constructing a target digital twin, the aim is to fuse and elevate multi-source sensing information into a computable virtual image capable of real-time synchronous and dynamic simulation of the target forest area's internal physical state. This provides a high-fidelity, computable core foundational model for subsequent proactive fire risk assessment and accurate spread prediction. Integrating previously discrete and heterogeneous multi-source sensing data into a unified, spatiotemporally consistent target digital twin allows for the quantification and continuous calculation of the real-time status and interrelationships of key elements such as vegetation moisture dynamics, combustible material load changes, and microclimate field evolution within the target forest area. This enables a leap from surface observation to a deep understanding of the underlying mechanisms of the target forest area. Fire management no longer relies solely on the identification of post-event phenomena such as smoke and fire, but can instead, based on the dynamic state revealed by the target digital twin, proactively anticipate the gestation process of fire hazards. This provides reliable and dynamic input for subsequent quantitative early warning and predictive simulations, supporting the transformation of the entire fire management process from passive response to proactive intervention.

[0024] Step S20: Substitute the current forest area status represented by the target digital twin into the preset fire point prediction model to calculate the fire risk warning information.

[0025] By utilizing the real-time quantification of the current forest area state, which reflects the inherent physical mechanisms of the target forest area, as derived from the target digital twin, and inputting it into a pre-set fire point prediction model, the current passive identification of smoke and fire is transformed into an active quantitative assessment and probability prediction of potential ignition conditions. By correlating the current forest area state, which characterizes the forest environment, i.e., key dynamic variables such as vegetation moisture content, combustible load, temperature, humidity, and wind speed, with disaster-causing patterns learned from historical fires, the pre-set fire point prediction model calculates and assesses the potential probability of fire occurring in each spatial unit of the target digital twin under the current forest area state. This enables the output of fire risk warning information, including the geographical location of specific ignition points and their corresponding target ignition probabilities, before open flames or visible smoke are generated. This shifts the fire prevention and control focus forward, allowing management to move from post-disaster emergency response to pre-disaster risk control.

[0026] Step S30: Based on fire risk warning information and / or real-time fire information, perform fire simulation in the target digital twin to generate fire spread prediction information.

[0027] Under the conditions of fire risk warning information and / or actual fire information, a computable virtual image constructed using the target digital twin is used to simulate the fire development trend in the target forest area. This combines the data in the target digital twin that reflects the current state of the forest area with the physical laws of fire behavior, thereby deduce the possible spread path, speed and intensity changes of the fire in the real forest environment in the computable virtual image. By transforming probability-based static risk warning or isolated observation of existing fire points into prediction of the dynamic evolution of the fire field in the future, fire spread prediction information including fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark is generated. This transforms fire fighting command decision-making from rough judgment based on past experience to scientific judgment based on high-confidence dynamic prediction, providing key and direct decision-making basis for the accurate allocation of fire-fighting resources, setting the best interception position and timing of action in subsequent steps.

[0028] Step S40: Make decisions based on fire spread prediction information and generate fire management information.

[0029] Decision-making is based on fire spread prediction information. The simulated dynamic development of the fire scene is transformed into executable operational instructions, thereby guiding the optimized scheduling of pre-positioned fire-fighting resources and the deployment of fire-fighting operations. Based on the descriptions of the future direction and speed of fire spread in the fire spread prediction information, combined with the real-time status and spatial distribution of currently available fire-fighting resources, i.e., pre-positioned fire-fighting resources, the optimal plan is formulated through analysis and optimization calculations, determining where to deploy pre-positioned fire-fighting resources, in what order, through what path, and when to perform what tasks. Finally, systematic fire management information is generated, realizing the connection between dynamic prediction and actual action. This allows the scheduling and deployment of fire-fighting resources to be matched with the development of the fire in advance, and to deploy or effectively block the fire in advance on key paths and channels of fire spread. This improves the predictability, accuracy, and overall efficiency of fire-fighting operations, effectively overcoming the resource waste and passive response problems caused by information lag and reliance on experience in traditional fire-fighting.

[0030] Furthermore, steps S10 to S03 are included before step S10: Step S01: Collect polarized light information, multispectral information, and thermal radiation information through a sensor network and remote sensing platform deployed in the target forest area.

[0031] A sensor network consisting of ground nodes is deployed within the target forest area. These ground nodes integrate polarization sensors to continuously measure the polarization state of skylight from different directions, thereby acquiring polarized light information reflecting the characteristics of particles in the atmosphere. Simultaneously, remote sensing platforms such as UAVs, airships, or satellites equipped with multispectral cameras and thermal infrared sensors periodically scan and image the target forest area along preset routes or orbits. The multispectral cameras collect the reflectance spectrum information of vegetation in different narrow bands, while the thermal infrared sensors capture the thermal radiation intensity information of the ground surface and canopy. This achieves the coordinated acquisition of polarized light information, multispectral information, and thermal radiation information. The synchronous acquisition of polarized light information, multispectral information, and thermal radiation information provides raw observation data for subsequent interference suppression and feature fusion processing. In particular, the introduction of polarized light information provides a data dimension to solve the problem of distinguishing between clouds and fog and early smoke in traditional observations.

[0032] Step S02: Perform interference suppression processing on the polarized light information and multispectral information to obtain spectral feature information.

[0033] By utilizing the characteristics of light wave vibration direction contained in polarized light information, this data is analyzed and compared with reflection intensity data of different bands in multispectral information. By analyzing the differences in polarization degree and morphological characteristics of spectral curves at specific angles, and considering the differences in polarization characteristics between Mie scattering from spherical water droplets in atmospheric clouds and the scattering from non-spherical particles in smoke, and the specific polarization angle of specular reflection from vegetation leaves, a discrimination feature can be constructed using these polarization parameters, such as polarization degree and polarization angle. This allows for the preliminary separation of specular reflection components with high polarization degree and polarization angle conforming to specular reflection geometry, as well as cloud and fog scattering components with specific polarization degree ranges, from multispectral information.

[0034] By combining multispectral information, the spectral curves of each observation unit are analyzed. An inversion method based on a spectral mixing model is established, incorporating the separated specular reflection and cloud scattering components as known endmembers or constraints into the spectral mixing model. This spectral mixing model assumes that the multi-band reflection intensity signal acquired by each observation unit is the result of a linear mixture of multiple components, including surface vegetation, cloud scattering, and specular reflection, in a certain proportion (abundance). The observed polarization parameters are related to the polarization characteristics and abundance of each component.

[0035] By solving the spectral mixing model, particularly utilizing the physical constraints on the scattering and reflection components provided by polarized light information, the spectral contributions of the specular reflection and cloud / fog scattering components—two interfering components—can be accurately estimated. Finally, by subtracting the estimated cloud / fog scattering and specular reflection contributions band by band from the polarized light and multispectral information, the interfering components are removed, yielding spectral characteristic information that primarily reflects the vegetation's own reflectivity and has suppressed cloud / fog scattering and specular reflection interference.

[0036] An example of this processing method is to utilize a nonnegative matrix factorization or linear spectral unmixing algorithm framework, and use the discriminant features derived from polarization parameters as constraints on the proportion of certain components or the spectral shape during the unmixing process. Through iterative optimization algorithms, the quantitative separation of interfering components is finally achieved. This greatly improves the accuracy and reliability of identifying early smoke in target forest areas, especially white smoke that is similar to the visual characteristics of clouds and fog, and real surface thermal anomalies from the data source. It effectively overcomes the problem of false alarms and missed alarms of fire caused by environmental interference in traditional single-spectrum or thermal imaging technologies.

[0037] It should be noted that the observation unit refers to the smallest spatial unit in which multispectral cameras and thermal infrared sensors that make up remote sensing images observe and record information about the ground.

[0038] Step S03: Spatiotemporal registration and feature fusion of spectral feature information and thermal radiation information are performed to obtain multi-source sensing information.

[0039] By reading and analyzing the spectral and thermal radiation information, along with their associated geographic location information and imaging time, the spectral and thermal radiation information is uniformly converted to the same geographic coordinate system, such as WGS84 UTM. It should be noted that in remote sensing data processing, both spectral and thermal radiation information are essentially organized as a two-dimensional grid, with each observation unit storing a data array of corresponding observation values; their data structure and storage format are exactly the same as those of digital images.

[0040] Next, since the spatial resolution of spectral feature information and thermal radiation information, i.e., the ground size represented by a single observation unit, is usually inconsistent, it is necessary to resample the higher-resolution spectral feature information or the lower-resolution thermal radiation information. Algorithms such as bilinear interpolation or cubic convolution interpolation are used to adjust the observation units of both to the same ground sampling interval, ensuring that the same geographic coordinates point to the same ground area in both spectral feature information and thermal radiation information. Then, through registration methods based on scale-invariant feature transform (SIFT) or phase correlation, fine geometric correction at the sub-pixel level is performed on the coordinate-transformed and resampled spectral feature information and thermal radiation information to eliminate residual positional deviations caused by sensor perspective, terrain undulations, etc., ultimately achieving precise spatial overlap between spectral feature information and thermal radiation information, meaning that each observation unit's location strictly corresponds to the same ground observation unit. After completing spatiotemporal registration and establishing a unified spatial grid, for each observation unit on this spatial grid, its corresponding spectral feature information and the thermal radiation information at that location are directly spliced ​​or stacked at the data level to form a new feature vector. This new feature vector is the multi-source sensing information that integrates reflectance and thermal radiation characteristics.

[0041] In one feasible implementation, refer to Figure 2 As shown, step S10 may specifically include steps S11 to S13: Step S11: Obtain geographic information data and historical ecological data of the target forest area, combine multi-source sensing information to perform 3D modeling and data fusion, and construct a 3D model of the forest area.

[0042] Geographic information data and historical ecological data of the target forest area are obtained by calling or accessing existing geographic information system databases and long-term ecological monitoring archives. These data include, but are not limited to, topographic elevation, slope and aspect, soil type, historical vegetation distribution maps and long-term phenological observation records.

[0043] Next, using topographic data from geographic information data as a base, a three-dimensional geometric grid representing the surface undulations is constructed. Historical ecological data, such as tree species distribution and soil properties, are assigned as initial attribute information to the corresponding areas of the three-dimensional geometric grid. Then, the multi-source sensing information acquired at the current moment, namely the spectral feature vector corresponding to each observation unit and the radiation temperature value acquired synchronously by the corresponding observation unit, are mapped and fused to the corresponding spatial location and surface attributes of the constructed three-dimensional geometric grid through their attached geographic coordinates and timestamps.

[0044] This fusion process is specifically manifested in the dynamic attachment or binding of spectral feature vectors reflecting material composition and thermal radiation values ​​reflecting energy state to the corresponding grid cells of the three-dimensional geometric grid as attribute layers. This forms a three-dimensional forest area model that has both accurate three-dimensional terrain features and carries detailed information on the surface spectrum and temperature at the current moment. It creates a unified spatial carrier that combines static geographical and historical background with dynamic real-time observation, enabling subsequent physical process simulations to run in a three-dimensional scene that reflects historical patterns and fits the current reality.

[0045] Step S12: Embed a preset physical mechanism model into the three-dimensional model of the forest area to generate an initial digital twin framework.

[0046] The purpose of embedding a pre-defined physical mechanism model that couples the dynamics of vegetation moisture, changes in combustible load, and the evolution of the microclimate field into the mathematical and computational framework of the 3D forest area model is to transform the original 3D forest area model, which only reflects spatial morphology and instantaneous attributes, into an initial digital twin framework that can simulate the spatiotemporal evolution of key fire risk elements in forest areas, namely moisture, combustibles, and local climate, based on the basic laws of physics, ecology, and meteorology. This establishes a dynamic coupling relationship of mutual driving and feedback among key fire risk elements in various forest areas, laying an indispensable simulation foundation with inherent causal logic for the subsequent realization of forward-looking early warning of fire risks and dynamic prediction of fire situation based on physical mechanisms rather than simple statistical laws.

[0047] It should be noted that the pre-defined physical mechanism model, which couples the dynamics of vegetation moisture, changes in combustible load, and the evolution of the microclimate field, is a computational framework that correlates and calculates the dynamic evolution of moisture state, combustible load, and climate by simultaneously solving a set of physical control equations describing surface energy exchange, momentum transfer, and water vapor diffusion on a unified three-dimensional grid space.

[0048] Furthermore, step S12 may include steps S121 to S124: Step S121: By using preset vegetation water transport calculation rules, the tree species distribution information in geographic information data and historical ecological data is calculated to obtain the initial field of vegetation water status.

[0049] It should be noted that the preset vegetation water transport calculation rule is a set of mathematical equations based on the physical process of water transport in the soil-plant-atmosphere continuum. It comprehensively considers the physiological characteristics of tree species, such as stomatal conductance and root depth, soil hydrological parameters such as soil texture and water holding capacity, as well as basic meteorological driving factors.

[0050] In the specific calculation, the topographic elevation, slope aspect and soil type of each observation unit are first extracted from the geographic information data. The tree species composition, forest age and root distribution parameters of the corresponding observation unit are obtained from the historical ecological data. Then, the above parameters are input into the preset vegetation water transport calculation rules. By solving the equations describing water infiltration, root water absorption, vegetation transpiration and soil evaporation, the soil moisture content, vegetation moisture content and water stress index of each observation unit at the initial time are calculated, thus forming an initial field of vegetation water state covering the entire target forest area, with each observation unit assigned a quantitative water state value.

[0051] The equation describing water infiltration is shown below:

[0052] in, It is the soil volumetric water content; t is time; z is vertical depth; It is the unsaturated hydraulic conductivity, which varies with water content; is the soil water potential; S is the root water absorption rate per unit time and unit volume of soil.

[0053] The equation describing root water absorption is shown below:

[0054] in, It is the water absorption rate at depth z; It is a weighting function that describes the distribution of root density with depth; It is the potential transpiration rate, which is determined by meteorological conditions and vegetation characteristics; It is related to soil water potential The relevant stress response function represents the inhibitory effect of soil drought on water absorption; It is the maximum depth of the root system.

[0055] The equation describing vegetation transpiration is shown below:

[0056] in, It is the slope of the saturated water vapor pressure-temperature curve; It is net canopy radiation; It is air density; It is the specific heat of air at constant pressure; It is the saturated water vapor pressure difference; It is the latent heat of vaporization of water; It is the constant of the wet and dry meter; It is aerodynamic drag; This refers to stomatal resistance in the canopy. Actual transpiration... ,in This refers to the stress coefficient related to soil water potential mentioned above.

[0057] The equation describing soil evaporation is shown below:

[0058] in, It is the soil evaporation rate; It is the net radiation at the Earth's surface; It is soil heat flux; This is the aerodynamic drag between the ground surface and a reference altitude. When the soil surface is dry, evaporation is limited by the surface soil water potential.

[0059] Step S122: Based on the vegetation moisture state in the initial field of vegetation moisture state, calculate the change in combustible load and obtain combustible load distribution information.

[0060] Based on the vegetation moisture status of each grid cell in the initial vegetation moisture status field, such as soil moisture content or vegetation moisture stress index, the organic matter decomposition rate of the grid cell under the current moisture conditions is calculated through a function describing the dependence of microbial decomposition activities on water. At the same time, combined with the tree species type and phenological information of the grid cell, the natural addition rate of litter per unit time is determined through a function reflecting the laws of vegetation growth and litter production. Next, these two rates are substituted into an equation describing the dynamic equilibrium of matter. The core form of this equation is "Current combustible load = Previous load + Natural addition - Decomposition consumption". By solving this equation or performing time-step iterative calculations, the total combustible load of each grid cell at a specific time point is quantified, forming a spatially continuous distribution of combustible load information. This establishes the correlation between vegetation moisture and combustible load, enabling the target digital twin to simulate the physical process that combustibles are not static but dynamically accumulate or are consumed depending on moisture conditions. This provides input data for accurately assessing the potential fire intensity in different areas and simulating the spread of fire under real combustible distribution environments.

[0061] The function describing the water dependence of microbial decomposition activities is shown below: ,when ,when in, It is the optimal soil moisture content for microbial activity; This is an empirical parameter determined based on litter type and soil properties. This function reflects a nonlinear relationship where the decomposition rate is inhibited under both drought and waterlogging conditions, peaking under optimal moisture conditions. The baseline decomposition rate (typically determined by temperature via a modified Arrhenius equation) is multiplied by a moisture correction factor. Thus, the rate of organic matter decomposition under the current moisture conditions is obtained.

[0062] The functions reflecting the patterns of vegetation growth and litter production are shown below:

[0063] in, It is the rate of litter addition at time t; It is the maximum possible rate of addition for this vegetation type under ideal conditions; It is a phenological function related to time t, and is often described by sine, S-curve or accumulated temperature-based models to simulate the variation pattern of high addition rate during the vigorous growth period, concentrated leaf fall during the deciduous season, and very little addition during the dormant period. It is an adjustment factor that takes into account site productivity, such as soil fertility and slope.

[0064] Decomposition consumption = organic matter decomposition rate under current moisture conditions × combustible load on the corresponding grid cell × time step ( t).

[0065] Step S123: Combine the information on the distribution of combustible load with the meteorological parameters of the target forest area to simulate the climate field and obtain the climate field information of the forest area.

[0066] Meteorological driving parameters of the target forest area, such as air temperature, air humidity, wind speed, wind direction, solar radiation and atmospheric pressure obtained from meteorological stations, are used as the background field input for the simulation. At the same time, information on the distribution of combustible load is used as input data to describe the properties of the forest ground surface.

[0067] On the 3D computational grid of the forest area 3D model, the two inputs mentioned above are coupled with topographic data. The simulation is completed by simultaneously solving the physical control equations describing surface energy exchange, momentum transfer, and water vapor diffusion. Specifically, the combustible material load distribution information is used to calculate key parameters such as surface albedo, surface heat capacity, and surface roughness for each grid cell. For example, high-load dry combustible materials have lower albedo and smaller heat capacity, thus directly affecting the distribution of net radiation and the sensible and latent heat fluxes between the ground and the atmosphere in the surface energy balance equation. At the same time, the combustible material load distribution information also affects the momentum transfer equation by changing the surface roughness, thereby affecting the vertical distribution and horizontal variation of near-surface wind speed. By numerically solving the physical control equations on the 3D computational grid, the detailed states of temperature, humidity, wind speed, and wind direction at different altitudes for each 3D grid cell can be calculated, thus obtaining forest microclimate field information that reflects the combined influence of forest topography and combustible material distribution, thereby generating a dynamic background field that is highly consistent with the physical environment of the real forest area.

[0068] The equations describing energy exchange at the Earth's surface are shown below:

[0069] in, It is the net radiation at the Earth's surface, calculated as , It is the surface albedo; For incident shortwave radiation; Earth's surface emissivity; Long-wave radiation incident downwards from the atmosphere; Stefan-Boltzmann constant; It represents the long-wave radiation emitted outward from the Earth's surface. It is the sensible heat flux. , It is the thermal turbulent exchange coefficient; It refers to the wind speed at the reference altitude; It refers to the surface temperature; The air temperature at the reference altitude. It is latent heat flux. , It is the latent heat of vaporization of water; Water vapor turbulent exchange coefficient; Indicates the surface temperature The saturated air below is more humid; This indicates the actual specific humidity of the air at the reference altitude. It is the heat flux entering the soil or combustible layer. , Thermal conductivity of soil / combustible medium; This represents the vertical temperature gradient at the Earth's surface.

[0070] Step S124: Based on the initial field of vegetation moisture status, combustible load distribution information and forest climate field information, data coupling and parameter binding are performed on the three-dimensional model of the forest area to generate an initial digital twin framework.

[0071] Using the spatial grid of a 3D forest model as a unified spatiotemporal reference framework, the initial field of vegetation moisture status (characterizing vegetation moisture condition), the distribution information of combustible load reflecting spatial differences in fuel load, and the forest climate field information including elements such as temperature, humidity, wind speed, and wind direction are spatially aligned and matched through shared geographic coordinates. This achieves the integration of multi-source data on the same spatial grid, completing data coupling. Next, dynamic relationships between attributes are established on the 3D grid cells based on physical mechanisms. For example, the vegetation moisture status parameters of each 3D grid cell, such as soil moisture content, are correlated with the potential evaporation capacity of the corresponding 3D grid cell. Combustible load and type information are converted into underlying surface physical attribute parameters for the corresponding 3D grid cell. By substituting these underlying surface physical attribute parameters into the physical control equations, the three originally independently calculated fields—moisture, combustibles, and microclimate—are linked into a computable organic whole within each 3D grid cell through physical laws, thus obtaining the initial digital twin framework.

[0072] Step S13: Based on the multi-source sensing information at the current moment, drive and calibrate the state variables of the initial digital twin framework to obtain the target digital twin.

[0073] The multi-source sensing information at the current moment is input into the initial digital twin framework. Using the spectral feature vector and thermal radiation value contained therein, the observation estimate value that corresponds to the state variable in the initial digital twin framework and can reflect the current forest area condition is calculated through physical inversion or data assimilation algorithm. For example, real-time vegetation moisture content and leaf area index are obtained from the spectral feature vector, and the surface temperature field is obtained from the thermal radiation value.

[0074] Next, the calculated observation estimates are compared with the predicted state variables at the corresponding time points obtained by the simulation and deduction of the initial digital twin framework through its internal preset physical mechanism model, such as the predicted vegetation moisture state and surface temperature. The difference between the two is calculated and minimized, thereby adjusting and correcting the state variables of the corresponding three-dimensional grid units within the initial digital twin framework, such as soil moisture content, vegetation canopy water potential, and sensible / latent heat flux distribution coefficients in the surface energy balance. This ensures that the simulated state of the initial digital twin framework approximates the observed real state as closely as possible. This process is repeated periodically as new multi-source sensing information continues to flow in, thereby achieving continuous and dynamic updates to the initial digital twin framework. This ensures that the final target digital twin not only possesses the physical evolution laws given by the initial digital twin framework, but also that the vegetation moisture, combustible dryness, and microclimate field it exhibits at each moment are synchronized with the real spatiotemporal state of the target forest area in the real world. This provides a highly reliable simulation environment for high-confidence fire risk warning and fire situation prediction based on the current real situation.

[0075] In one feasible implementation, refer to Figure 3 As shown, step S20 may specifically include steps S21 to S24: Step S21: Extract forest status parameters representing vegetation moisture, combustible load, and forest climate parameters from the target digital twin.

[0076] Based on the three-dimensional spatial grid structure and data organization method established by the target digital twin, through its internal data query interface, according to the unique geographic identifier of each predefined three-dimensional grid unit, the instantaneous values ​​of three types of key state variables stored or calculated by each three-dimensional grid unit at the latest time step are read synchronously: vegetation moisture parameters, such as soil volumetric water content or canopy water potential; combustible load parameters, such as effective surface combustible load; and forest climate parameters, such as near-surface air temperature, relative humidity, and wind speed.

[0077] Step S22: The vegetation moisture parameter, combustible load parameter, and forest climate parameter are weighted and combined with the weight coefficients obtained from training based on historical fire sample data to calculate the comprehensive fire risk index of each spatial unit.

[0078] For each three-dimensional mesh cell extracted from the target digital twin, its corresponding vegetation moisture parameter, combustible load parameter, and forest climate parameter constitute a multi-dimensional feature vector. Then, each parameter value in this multi-dimensional feature vector is multiplied by its corresponding specific weight coefficient, which is obtained by training and analyzing a large amount of historical environmental data during fires and background data before fires through machine learning methods. This is achieved through a linear weighted combination mathematical operation.

[0079] Specifically, for a three-dimensional spatial unit, the extracted vegetation moisture parameter W, combustible material load parameter F, and forest climate parameter T are first arranged in a predetermined order to form a multi-dimensional feature vector X=[W, F, T]. Simultaneously, a large number of historical samples, including environmental state data during fires as positive samples and normal data before fires as negative samples, are trained using machine learning algorithms such as logistic regression. This training process aims to find an optimal weight vector w=[w1, w2, w3] and a bias term b, enabling the target digital twin to distinguish between historical fires and safe conditions. Each parameter in the feature vector X of the current spatial unit is multiplied element-wise with the corresponding weight coefficient in the trained weight vector w, i.e., W×w1, F×w2, T×w3. All these products are then summed, and the trained bias term z is added to obtain the comprehensive fire risk index Z of the three-dimensional spatial unit, i.e., Z=W×w1+F×w2+T×w3+z.

[0080] By transforming the forest area status parameters provided by the target digital twin into a comprehensive fire risk index consistent with historical statistical patterns, the inaccuracy and poor generalization ability caused by the reliance on fixed formulas and human experience to set parameters in traditional fire risk indices are fundamentally overcome.

[0081] Step S23: Substitute each comprehensive fire risk index into a preset nonlinear probability mapping function for conversion to obtain the fire probability of each spatial unit in the future time period.

[0082] The comprehensive fire risk index of each three-dimensional spatial unit is used as an input variable and substituted into a preset nonlinear probability mapping function, namely the Sigmoid function, for calculation.

[0083] The form of the preset nonlinear probability mapping function is:

[0084] Where Z represents the comprehensive fire risk index; P represents the probability of fire, with a value between 0 and 1; e is a constant; m and n are key shape parameters determined by logistic regression fitting of historical datasets, including the comprehensive fire risk index at the time of historical fires and the background value when no fires occurred, using machine learning methods. m controls the steepness of the function curve, i.e., the sensitivity of the probability to exponential growth, while n determines the fire risk index threshold position corresponding to a probability of 0.5. This preset nonlinear probability mapping function describes the nonlinear process of a three-dimensional spatial unit transitioning from a low-risk state to a high-risk state.

[0085] When the comprehensive fire risk index is low, the probability of fire approaches 0, indicating that it is almost impossible for a fire to start. As the comprehensive fire risk index increases, the probability of fire initially increases slowly, but after a rapid transition range determined by m and n, the probability of fire will rise sharply and eventually approach 1, indicating that the probability of fire is extremely high.

[0086] By pre-setting a nonlinear probability mapping function, each three-dimensional spatial unit can obtain a fire probability value. This fire probability value not only directly indicates the level of fire risk, but its nonlinear characteristics can also more realistically reflect the critical change phenomenon in the fire occurrence mechanism. For example, after the combustible material dries to a certain critical point, the probability of ignition will increase dramatically, thus making the final generated fire risk warning information more accurate.

[0087] Step S24: After selecting the target fire probability that exceeds the preset fire probability from the various fire probabilities, obtain the geographical location of the target spatial unit corresponding to the target fire probability and generate fire risk warning information. The fire risk warning information includes the geographical location of each target spatial unit and the target fire probability corresponding to the target spatial unit.

[0088] The fire probability values ​​of all three-dimensional spatial units are compared one by one with a preset fire probability set by domain experts based on historical fire statistics, risk tolerance, and resource allocation capabilities. All fire probabilities greater than the preset fire probability are identified and screened out, and the screened fire probabilities are marked as target fire probabilities.

[0089] Next, based on the unique identifier of the three-dimensional spatial unit to which each target fire probability belongs, the geographic coordinate data established when constructing the target digital twin and strictly bound to the corresponding unique identifier is traced back. After obtaining the geographic location of the target spatial unit corresponding to each target fire probability, the geographic location of each target spatial unit is paired and encapsulated with its corresponding target fire probability to generate fire risk warning information. By grounding the abstract mathematical probability output into specific warning targets marked on geographic space, the fire prevention command department can clearly know "where" and "how likely" a fire risk exists based on this information, thereby enabling immediate and accurate guidance for early intervention actions such as focusing patrol forces and pre-positioning fire-fighting resources.

[0090] In one feasible implementation, refer to Figure 4 As shown, step S30 may specifically include steps S31 to S34: Step S31: Based on the fire hazard warning information and / or real-time fire information, set the initial fire point location and initial fire intensity on the target spatial unit corresponding to the target digital twin.

[0091] Based on the fire hazard warning information, the geographical location of each target spatial unit contained in the fire hazard warning information is read. A unique corresponding 3D mesh unit is found in the 3D spatial grid of the target digital twin through coordinate matching. The center point or designated location of this 3D mesh unit is marked as the initial fire point location. Simultaneously, the target ignition probability number attached to the target spatial unit is mapped to an initial fire intensity through a transformation function.

[0092] Furthermore, if real-time fire information exists, such as open flames or heat sources detected by infrared monitoring, the latitude and longitude coordinates corresponding to the real-time fire information are matched and located with the three-dimensional spatial grid of the target digital twin, and the monitored thermal radiation intensity or estimated burned area is converted into an initial fire intensity value through a physical inversion model.

[0093] If fire risk warning information and real-time fire information coexist, the real-time fire information is used first for setting, and the overlapping points in the fire risk warning information are covered and superimposed. This provides accurate and physically consistent initial spatiotemporal and energy conditions for subsequent physical simulations, ensuring that the calculation starting point of the fire spread simulation is highly consistent with the risk warning point or actual ignition point of the target forest area, thus making the entire simulation process based on a real and reliable state.

[0094] It should be noted that the conversion function is a conversion relationship established based on historical data statistics or physical experience, which maps the probability value of fire occurrence to the typical initial combustion energy release rate; while the physical inversion model is a calculation model constructed based on physical laws such as thermal radiation transfer, used to quantitatively infer the actual energy release rate of the fire scene from the observed infrared radiation intensity or burned area.

[0095] Step S32: Based on the current forest area status and initial fire intensity on each target spatial unit, calculate the fire spread direction on the target digital twin starting from the initial fire point, as well as the fire spread speed and fire intensity changes in each fire spread direction.

[0096] For each initial fire point location, the current forest status of the target spatial unit and its neighboring target spatial units is obtained in real time from the target digital twin, including the combustible load, combustible moisture content, near-surface wind speed and direction, and terrain slope and aspect of the target spatial unit. Then, using the set initial fire intensity as energy input, a set of governing equations describing the physical processes such as energy balance at the fire front, combustible preheating and ignition, and flame convection heat transfer are solved. According to the laws of energy conservation and heat transfer, the main direction of fire spread is determined by the strongest driving force with the strongest instantaneous combined effect. This is usually calculated by synthesizing the wind speed and direction vector and the terrain slope vector to obtain a dominant spread direction. The fire spread rate in each direction is calculated by a physical formula that uses the above-mentioned dominant driving force, combustible load, and combustible moisture content as key variables.

[0097] For example, the rate of fire spread in downwind and uphill directions is significantly accelerated due to increased driving force; meanwhile, changes in fire intensity are characterized by calculating the energy released per unit time and per unit length of fireline, which is directly related to the rate of fire spread and the rate of combustible material consumption. Through this series of calculations based on physical laws, a set of fire behavior parameters (i.e., the direction of fire spread, and the changes in fire spread rate and fire intensity in each direction) can be output for each initial fire location and its neighboring units that may be affected in the next moment. This transforms the static forest state and initial fire intensity represented by the target digital twin into dynamic fire behavior parameters with direction and velocity.

[0098] Step S33: Perform spatiotemporal integration and iterative updates on the changes in fire spread rate and fire intensity to obtain the fire boundary locations of each fire site and the distribution of fire point intensity within each fire site in the future time period.

[0099] Taking the changes in fire spread speed and fire intensity from the initial fire point as input, a discretized time step is set. Within each time step, the fire spread speed of each point on the fire field is integrated over time. That is, by multiplying the fire spread speed of the point by the time step, the displacement vector of the point in that time period is calculated, thereby determining the new position of the point after the movement. The set of the new positions of all fire field boundary points constitutes the updated fire field boundary position.

[0100] Simultaneously, the fire intensity change at each point on the fire field is integrated over time, that is, the fire intensity change is multiplied by the time step to accumulate or decrease the fire intensity at that point, thereby updating the fire intensity change within the fire field. After completing the integration for one time step, based on the updated fire field boundary position and fire point intensity distribution, combined with the current forest area state provided by the target digital twin corresponding to the current simulation time, the calculation method in step S32 is called again to calculate the new fire spread rate and fire intensity change at each point on the fire field at the start of the next time step. In this way, the process of "calculation-integration-update-recalculation" is iterated in a loop until the simulation clock reaches the end of the future time period, outputting the fire field boundary position from the current to multiple consecutive future times and the fire point intensity information distributed within the fire field at each time, realizing the dynamic deduction of the entire spatiotemporal evolution of the fire field in a complex real environment, thereby generating a high spatiotemporal resolution fire development prediction.

[0101] Step S34: Extract the terrain features of the target forest area from the target digital twin. Combine the terrain features, the location of the fire boundary of each fire site and the intensity distribution of fire points in each fire site to calculate the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark. The fire spread prediction information includes the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark.

[0102] From the 3D terrain model upon which the target digital twin is based, continuously distributed terrain features such as elevation, slope, aspect, and automatically generated ridgelines and valley lines reflecting terrain morphology are extracted. Next, the fire boundary locations and fire intensity distribution within each fire zone in future time periods are overlaid and spatiotemporally correlated with these terrain features in a unified geographic space. By tracking the leading edge trajectory of the fire boundary expansion over time, the main spatial propagation axis of fire spread, i.e., the fire spread path, is determined. Simultaneously, combining the fire intensity distribution within the fire zone at each time point, especially the movement of high-intensity fire areas, within the framework of terrain features, terrain features that simultaneously satisfy narrow terrain characteristics, such as ridgelines, canyons, and terrain with continuous combustible material and whose main spread direction is highly consistent with the prevailing wind direction or terrain uplift direction, such as spatial corridors, are further used to determine the fire spread channels from which the fire is most likely to accelerate or break through.

[0103] For the pre-set landmark, by calculating the shortest spatial distance from the current fire boundary to the pre-set landmark, and based on the fire spread speed, the expected time when the fire will reach the pre-set landmark is calculated. This specifically indicates "which path" the fire will take, "which key path" it will pass through, and "when and where it will threaten". This provides highly targeted and actionable predictive information support for the fire department to intercept key passages in advance, organize the protection of key targets before the fire arrives, and scientifically plan firefighting resources and evacuation routes.

[0104] It should be noted that the fire spread path is determined based on a comprehensive assessment of topography, combustible continuity, and wind direction. It refers to the narrow geographical path through which fire is most likely to spread rapidly or break through, such as ridgelines, canyons, or specific vegetation corridors. Pre-defined landmarks refer to the pre-defined geographical locations on the fire spread path that have important protective value, such as settlements and infrastructure, or that have a key impact on the development of the fire, such as gaps in firebreaks or rivers.

[0105] In one feasible implementation, refer to Figure 5 As shown, step S40 may specifically include steps S41 to S43: Step S41: Determine the predicted fire points and the protection sequence between each predicted fire point based on the fire spread path, fire spread channel and / or time point.

[0106] Under a unified geographic coordinate system, the fire spread path and fire spread channel are spatially overlaid and calculated. Through line and surface overlay analysis or raster calculation in the geographic information system, the geometric intersection area or overlapping segment with high spatial proximity of the fire spread path and fire spread channel is automatically identified. These areas are determined as key points due to their geographic constraints and fire behavior intensification effect. They also include preset landmarks with clearly provided expected arrival times. The above key points and preset landmarks are determined as predicted fire points.

[0107] Next, two methods are used to determine its time attribute: for the preset landmark, the time point provided in step S34 is used directly; for the key point, the specific method for estimating its expected spread time point is as follows: first, in the fire boundary location, the nearest front point to the key point on the fire boundary at each future moment is extracted and its corresponding time is recorded. Then, through spatial interpolation, such as interpolation and time extrapolation based on the spatiotemporal trend surface constructed based on multiple nearest front points, the remaining time for the fire to spread from these known locations and times to the key point is estimated. Specifically, a dynamic time warping algorithm is used to align the movement sequence of the fire boundary front point with the spatial relationship of the key point, thereby calculating the time point when the fire head arrives at the key point at the currently predicted spread speed sequence.

[0108] Finally, based on the time points of each of the predicted fire points, all points are uniformly sorted in chronological order to generate a protection timeline that integrates all predicted fire points and their time points, thereby transforming the spatial threat of fire development into a clear timeline of action.

[0109] Step S42: Based on the protection sequence and pre-set fire-fighting resources, calculate the scheduling and matching scheme between each predicted fire point and the pre-set fire-fighting resources.

[0110] It should be noted that pre-set fire-fighting resources refer to the collection of information on various fire-fighting forces and assets that are pre-entered into the system and can be dispatched. These typically include the real-time status, type, quantity, effectiveness, and current geographical location of fire brigades, fire trucks, aircraft, large equipment, and fixed facilities such as fire pools and material warehouses.

[0111] By using the protection sequence as the task requirement input and the real-time status and location of pre-set fire-fighting resources as the supply capacity input, a many-to-many task-resource matching optimization model is constructed. The core optimization objective of this task-resource matching optimization model is to minimize the global response time and scheduling cost or maximize the overall protection success rate, while satisfying the hard time constraint that "fire-fighting resources arrive before the latest time required by the task".

[0112] For example, a genetic algorithm or particle swarm optimization algorithm can be used to solve the problem: the algorithm treats each predicted fire point as a task with a latest arrival time window, and each schedulable fire resource as a service provider with a specific movement speed, fire extinguishing efficiency, and initial position. Through iterative calculations in a simulated virtual spacetime, it allocates appropriate fire resources capable of arriving within their time window to each task, ensuring that tasks performed by the same fire resource on the timeline do not conflict. Finally, it outputs a scheduling and matching scheme that clearly specifies "which fire resource or team goes to which predicted fire point to perform which protection task." This enables the global coordination and matching of limited, static pre-set fire resources with dynamic, time- and space-distributed predicted fire points to generate a scheduling scheme that is executable in terms of time, space, and resource capabilities. This maximizes the utilization rate of fire resources and the overall efficiency of emergency response in complex multi-task concurrent scenarios.

[0113] Step S43: Based on the scheduling matching scheme, perform path planning and time coordination processing to obtain fire management information for the pre-set fire resources.

[0114] For each pair of pre-positioned fire resources and predicted fire points determined in the scheduling and matching scheme, the path planning process calculates an optimal or feasible safe path from the current location of the pre-positioned fire resource to the predicted fire point, based on the terrain and road data in the target digital twin and combined with real-time traffic and accessibility constraints.

[0115] Next, through time-coordinated processing, based on the length of each planned path, the preset travel speed, and the road condition complexity in the scheduling and matching scheme, the travel time required for the corresponding pre-positioned fire-fighting resources to travel along the planned path is calculated, and then their expected arrival time is estimated.

[0116] Subsequently, the estimated arrival times of all tasks are compared and coordinated with the protection sequence. By adjusting the departure order of the pre-positioned fire-fighting resources corresponding to the predicted fire points with lower priority in the protection sequence or by fine-tuning the travel speed within a safe range, it is ensured that the tasks corresponding to the predicted fire points with higher priority in the protection sequence can be completed within the required time window, and that multiple fire-fighting resources avoid time and space conflicts in the path or task area as much as possible.

[0117] Finally, the above processing results are structured and integrated to generate a complete set of action elements for each dispatch task, including the starting point (i.e., the current location of pre-positioned fire-fighting resources), the destination (i.e., the predicted fire point), the route, the estimated arrival time, and the resource capacity generated, such as creating firebreaks or implementing fire suppression by spraying water. The collection of all these elements constitutes fire management information that can be directly issued and executed, realizing a closed loop from optimized dispatch to precise execution. This ensures that the entire emergency response action is coordinated in time, orderly in space, and clear in tasks, greatly improving the overall efficiency and execution accuracy of multi-force dispatch in complex fire environments.

[0118] This application also provides an AI-based intelligent forest fire management system, referring to... Figure 6 As shown, the AI-based smart forest fire management system includes: The twin construction module 10 is used to construct a target digital twin of the target forest area based on multi-source sensing information of the target forest area; The data calculation module 20 is used to input the current forest area status represented by the target digital twin into the preset fire point prediction model to calculate fire risk warning information; Data prediction module 30 is used to perform fire simulation in the target digital twin based on fire risk warning information and / or real-time fire information, and generate fire spread prediction information. The data generation module 40 is used to make decisions based on fire spread prediction information and generate fire management information.

[0119] Optionally, this application also includes a data acquisition module 50, used for: By deploying sensor networks and remote sensing platforms in the target forest area, polarized light information, multispectral information, and thermal radiation information are collected. By suppressing interference in polarized light information and multispectral information, spectral feature information is obtained. By performing spatiotemporal registration and feature fusion of spectral feature information and thermal radiation information, multi-source sensing information is obtained.

[0120] Optionally, the twin building block 10 is also used for: Acquire geographic information data and historical ecological data of the target forest area, combine multi-source sensing information to perform 3D modeling and data fusion, and construct a 3D model of the forest area. A pre-defined physical mechanism model is embedded into a 3D model of the forest area to generate an initial digital twin framework; Based on the multi-source sensing information at the current moment, the state variables of the initial digital twin framework are driven and calibrated to obtain the target digital twin.

[0121] Optionally, the twin building block 10 is also used for: By pre-setting vegetation water transport calculation rules, the tree species distribution information in geographic information data and historical ecological data is calculated to obtain the initial field of vegetation water status. Based on the vegetation moisture state in the initial field of vegetation moisture state, the change of combustible load is calculated to obtain combustible load distribution information. Climate field information of the forest area is obtained by combining information on the distribution of combustible load with meteorological parameters in the target forest area. Based on the initial field of vegetation moisture status, combustible load distribution information and forest climate field information, data coupling and parameter binding are performed on the three-dimensional model of the forest area to generate an initial digital twin framework.

[0122] Optionally, the data calculation module 20 is also used for: Extract forest area status from the target digital twin, which represents parameters such as vegetation moisture, combustible load, and forest climate. The vegetation moisture parameter, combustible load parameter, and forest climate parameter are weighted and combined with the weight coefficients obtained by training based on historical fire sample data to calculate the comprehensive fire risk index of each spatial unit. Each comprehensive fire risk index is substituted into a preset nonlinear probability mapping function for transformation to obtain the fire probability of each spatial unit in the future time period. After filtering out the target fire probabilities that exceed the preset fire probabilities from the various fire probabilities, the geographical location of the target spatial unit corresponding to the target fire probability is obtained, and fire risk warning information is generated. The fire risk warning information includes the geographical location of each target spatial unit and the target fire probability corresponding to the target spatial unit.

[0123] Optionally, the data prediction module 30 is also used for: Based on fire risk warning information and / or real-time fire information, set the initial fire point location and initial fire intensity on the target spatial unit corresponding to the target digital twin; Based on the current forest area status and initial fire intensity on each target spatial unit, calculate the fire spread direction on the target digital twin starting from the initial fire point, as well as the fire spread speed and fire intensity changes in each fire spread direction; Spatiotemporal integration and iterative updates are performed on the changes in fire spread rate and fire intensity to obtain the fire boundary location and fire point intensity distribution of each fire field in the future time period. The terrain features of the target forest area are extracted from the target digital twin. Combined with the terrain features, the location of the fire boundary of each fire site and the intensity distribution of fire points in each fire site, the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark are calculated. The fire spread prediction information includes the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark.

[0124] Optionally, the data generation module 40 is also used for: Determine the predicted fire points and the protection sequence between each predicted fire point based on the fire spread path, fire spread channel and / or time point; Based on the protection sequence and pre-positioned fire-fighting resources, the scheduling and matching scheme between each predicted fire point and the pre-positioned fire-fighting resources is calculated; Based on the scheduling and matching scheme, path planning and time coordination are performed to obtain fire management information for pre-set fire-fighting resources.

[0125] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A smart forest fire management method based on AI analysis, characterized in that, include: Based on multi-source sensing information of the target forest area, a target digital twin of the target forest area is constructed; Substitute the current forest area status represented by the target digital twin into the preset fire point prediction model to calculate fire risk warning information; Based on the fire risk warning information and / or real-time fire information, fire simulation is performed in the target digital twin to generate fire spread prediction information; Decision-making is carried out based on the fire spread prediction information to generate fire management information.

2. The AI-based intelligent forest fire management method according to claim 1, characterized in that, The steps prior to the multi-source sensing information based on the target forest area include: Polarized light information, multispectral information, and thermal radiation information are collected through sensor networks and remote sensing platforms deployed in the target forest area. The polarization information and the multispectral information are subjected to interference suppression processing to obtain spectral feature information; The spectral feature information and the thermal radiation information are spatiotemporally registered and feature fused to obtain the multi-source sensing information.

3. The AI-based intelligent forest fire management method according to claim 1, characterized in that, The step of constructing a target digital twin of the target forest area based on multi-source sensing information of the target forest area includes: The geographic information data and historical ecological data of the target forest area are acquired, and the multi-source sensing information is combined to perform three-dimensional modeling and data fusion to construct a three-dimensional model of the forest area. A preset physical mechanism model is embedded in the three-dimensional model of the forest area to generate an initial digital twin framework; Based on the multi-source sensing information at the current moment, the state variables of the initial digital twin framework are driven and calibrated to obtain the target digital twin.

4. The AI-based intelligent forest fire management method according to claim 3, characterized in that, The step of embedding a preset physical mechanism model into the three-dimensional model of the forest area to generate an initial digital twin framework includes: By using preset vegetation moisture transport calculation rules, the tree species distribution information in the geographic information data and the historical ecological data is calculated to obtain the initial field of vegetation moisture state. Based on the vegetation moisture state in the initial field of the vegetation moisture state, the change in combustible load is calculated to obtain combustible load distribution information. By combining the information on the distribution of combustible load with the meteorological parameters of the target forest area, a climate field simulation is performed to obtain the climate field information of the forest area. Based on the initial field of vegetation moisture status, the information on combustible material load distribution, and the information on forest climate field, data coupling and parameter binding are performed on the three-dimensional model of the forest area to generate the initial digital twin framework.

5. The AI-based intelligent forest fire management method according to claim 1, characterized in that, The target digital twin is composed of multiple spatial units. The step of substituting the current forest area status represented by the target digital twin into a preset fire point prediction model to calculate fire risk warning information includes: Extract the forest area status, which represents vegetation moisture parameters, combustible load parameters, and forest area climate parameters from the target digital twin; The vegetation moisture parameter, the combustible material load parameter, and the forest climate parameter are respectively weighted and combined with the weight coefficients obtained by training based on historical fire sample data to calculate the comprehensive fire risk index of each spatial unit. Each of the comprehensive fire risk indices is substituted into a preset nonlinear probability mapping function for conversion to obtain the fire probability of each of the spatial units in the future time period. After filtering out the target fire probabilities that exceed the preset fire probabilities from the various fire probabilities, the geographical location of the target spatial unit corresponding to the target fire probability is obtained, and the fire risk warning information is generated. The fire risk warning information includes the geographical location of each target spatial unit and the target fire probability corresponding to the target spatial unit.

6. The AI-based intelligent forest fire management method according to claim 1, characterized in that, The step of performing fire simulation and generating fire spread prediction information in the target digital twin based on the fire risk warning information and / or real-time fire information includes: Based on the fire risk warning information and / or the real-time fire information, the initial fire point position and initial fire intensity are set on the target space unit corresponding to the target digital twin; Based on the current forest area status and the initial fire intensity of each target spatial unit, calculate the fire spread direction on the target digital twin starting from the initial fire point location, as well as the fire spread speed and fire intensity changes in each fire spread direction; Spatiotemporal integration and iterative updates are performed on the fire spread rate and the fire intensity changes to obtain the fire boundary positions of each fire field and the fire point intensity distribution within each fire field in future time periods. The terrain features of the target forest area are extracted from the target digital twin. Combined with the terrain features, the fire boundary positions of each fire site and the intensity distribution of fire points within each fire site, the fire spread path, fire spread channel and / or the time point when the fire spreads to the preset landmark are calculated. The fire spread prediction information includes the fire spread path, the fire spread channel and / or the time point when the fire spreads to the preset landmark.

7. The AI-based intelligent forest fire management method according to claim 6, characterized in that, The step of making decisions based on the fire spread prediction information and generating fire management information includes: Based on the fire spread path, the fire spread channel, and / or the time point, the predicted fire points and the protection sequence between each predicted fire point are determined. Based on the protection sequence and the pre-set fire-fighting resources, a scheduling and matching scheme between each of the predicted fire sites and the pre-set fire-fighting resources is calculated. Based on the aforementioned scheduling and matching scheme, path planning and time coordination processes are performed to obtain fire management information for scheduling the pre-set fire-fighting resources.

8. A smart forest fire management system based on AI analysis, characterized in that, include: The twin construction module is used to construct a target digital twin of the target forest area based on multi-source sensing information of the target forest area; The data calculation module is used to substitute the current forest area status represented by the target digital twin into the preset fire point prediction model to calculate fire risk warning information; The data prediction module is used to perform fire simulation in the target digital twin based on the fire risk warning information and / or real-time fire information, and generate fire spread prediction information. The data generation module is used to make decisions based on the fire spread prediction information and generate fire management information.

Images