A forest fire fire dispatch auxiliary decision-making method and system based on combustible material classification and physical spread rate-driven directed graph model construction.
By constructing a directed graph model driven by combustible material classification and physical spread rate, the accuracy and efficiency problems of existing forest fire dispatch models are solved, enabling scientific and efficient deployment of fire-fighting resources and improving fire emergency response capabilities.
Patent Information
- Application Number
- CN202611531778.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-30
- Publication Date
- 2026-11-17
AI Technical Summary
Existing forest fire dispatch models lack physical realism and are difficult to adapt to complex fire situations, resulting in insufficient accuracy and low efficiency in dispatch results, making it difficult to provide scientific and differentiated optimal resource deployment solutions.
A directed graph model driven by combustible material classification and physical spread rate is constructed. An initial directed graph is generated by satellite remote sensing imagery, digital elevation model and overall wind field. The spread rate is calculated by combining combustible material type and terrain features. Logical Benders decomposition algorithm is used to optimize the deployment of fire suppression resources.
It achieves a fully automated closed loop from satellite imagery to fire dispatch, possesses physical realism and differentiated protection capabilities, provides a scientific and efficient fire dispatch decision-making tool, and improves emergency response efficiency.
Smart Images

Figure CN123119291A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of forest fire prevention and control technology, specifically to a forest fire dispatching auxiliary decision-making method and system that uses combustible material classification and physical spread rate-driven directed graph model construction. Background Technology
[0002] Forest fire dispatch decisions primarily rely on experience and heuristics, making them ill-suited for the rapid response and precise decision-making demands of complex fire situations. In recent years, mathematical programming-based fire resource dispatch models have gained increasing attention. These methods typically discretize the fire scene into a grid and model the dispatch problem as a combinatorial optimization problem. However, as the fire scale and the number of grids increase, the model size grows rapidly, significantly reducing solution efficiency and failing to meet the real-time requirements of forest fire emergency response. Furthermore, existing dispatch models often describe fire spread using simplified propagation rules or empirical parameters, failing to adequately consider the influence of combustible material type, terrain, and wind field on fire spread behavior, resulting in insufficient accuracy in dispatch results. Current methods struggle to provide scientific and differentiated optimal resource deployment schemes. Summary of the Invention
[0003] The purpose of this invention is to address the technical problems in existing forest fire command and rescue decision-making, such as isolated links in each stage, lack of physical realism in the scheduling model, and lack of differentiated protection weights leading to insufficient accuracy and low efficiency in scheduling results. This invention provides a forest fire fire dispatching auxiliary decision-making method and system that uses combustible material classification and physical spread rate-driven directed graph model construction.
[0004] The technical solution adopted by this invention to solve the above problems is: a forest fire fire dispatch auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction, the method comprising the following steps:
[0005] Based on satellite optical remote sensing images, digital elevation model (DEM) and overall wind field of the study area, an initial directed graph model covering the entire study area is constructed.
[0006] Based on the results of the combustible material classification model, potential protection targets are automatically identified and basic protection weights are generated based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, the protection weight of each node is generated, and node weights are assigned to the directed graph model.
[0007] Based on the physical parameters of combustible material type and combined with terrain features, the Rothermel equation and elliptical fire shape model are used to calculate the spread rate in 8 directions and convert it into arc travel time to generate weighted directed arc edges for directed graph model.
[0008] With the goal of minimizing the weighted number of burned nodes, the logical Benders decomposition algorithm is used to solve for the optimal fire extinguishing resource deployment scheme on the directed graph model.
[0009] Furthermore, the specific process of automatically identifying potential protected targets and generating node protection weights based on the combustible material classification model results includes: extracting a 10-dimensional image-spectral feature vector for each grid from satellite optical remote sensing images. The 10-dimensional feature vector includes 6 dimensions of spectral features (mean and standard deviation of RGB three channels), 2 dimensions of vegetation index (mean and standard deviation of NDVI approximation), and 2 dimensions of color features (mean of brightness and mean of saturation). The 10-dimensional feature vector is input into the XGBoost combustible material classification model. After training with a training sample set, the model simultaneously outputs the combustible material type number, classification confidence, and protected target category for each grid. The protected target categories include built-up areas, farmland, ecologically sensitive areas, general vegetation areas, and areas with no protection value.
[0010] Furthermore, the basic protection weights are generated based on the JRC (Joint Research Centre for European Union) wildfire risk framework. The JRC wildfire risk framework is a standard system for assessing potential fire losses. This framework classifies land cover types into categories such as urban / built areas, farmland, shrubland, woodland / grassland, and water / bare rock, and assigns a basic protection value to each category. According to the framework's assignment rules, the basic protection weights are set as follows: urban / built areas: 10; farmland: 6-8; shrubland: 4-6; woodland / grassland: 1-4; water / bare rock: 0.
[0011] Furthermore, the step of receiving the user-annotated mandatory protection grids through the interactive interface and performing spatial weight propagation includes: receiving the mandatory protection grids annotated by the user on the base map, performing spatial weight propagation on the 8 neighboring grids around the mandatory protection grids, assigning a weight bonus of 2 to each, and the upper limit of the weight after the bonus is 9.
[0012] Further, the specific process of calculating the 8-directional spread rate and converting it into arc travel time includes: extracting the terrain features of each grid based on the digital elevation model (DEM), the terrain features including elevation, slope, and aspect; querying 12 physical parameters and 1 dynamic flag based on the combustible material type number, the 12 physical parameters including load (h1, h10, h100, lh, lw), surface area to volume ratio (sav1, savlh, savlw), combustible bed depth, extinguishing moisture content xmext, and calorific value (heatd, heatl); the dynamic flag is used to indicate whether the combustible material participates in the spread process; and calculating the basic spread rate using the Rothermel spread equation.
[0013]
[0014] In the formula, This represents the base spread rate of the current grid without any classification confidence correction. The intensity of the reaction depends on the surface area to volume ratio and the quenching water content. The fire front propagation flux ratio depends on the surface area to volume ratio and the extinguishing moisture content. The density of the combustible bed depends on the load and the depth of the combustible bed. Indicates the effective heating coefficient; It represents the amount of heat required to ignite a unit mass of combustible material, and depends on its calorific value.
[0015] Using the aforementioned terrain features and combined with overall wind field data, the local topographic wind field of each grid is calculated to obtain local wind speed and direction; the comprehensive wind slope factor is then calculated.
[0016]
[0017]
[0018] In the formula, For comprehensive wind slope factors; For wind factors; Wind speed; Wind direction; The direction in which the fire starts moving; The slope angle is denoted by .
[0019] Then calculate the flame rate:
[0020]
[0021] For the angle with the direction of the fire head is The direction and propagation rate are calculated using the elliptic interpolation formula:
[0022]
[0023]
[0024] The 8-directional (N, NE, E, SE, S, SW, W, NW) propagation rates for each grid are calculated using the above formula. When the classification confidence level of a certain grid... When the spread rate falls below a preset spread confidence threshold, a conservative buffer is added to the base spread rate of the grid, and the buffer is calculated using the following formula:
[0025]
[0026] In the formula, For classification confidence; The maximum spread rate; This represents the base spread rate of the current grid without classification confidence correction. This is the scaling factor.
[0027] The corrected 8-directional spread rate is converted into the arc travel time of the directed graph model structure:
[0028]
[0029] In the formula, For the node to adjacent nodes The time of advance of the directed arc; This refers to the grid spacing; The value derived from the modified base spread rate combined with the elliptic fire model is obtained from the node. Point to adjacent nodes The rate of directional spread.
[0030] Furthermore, the process of solving the optimal fire extinguishing resource deployment scheme using the logistic Benders decomposition algorithm with the objective of minimizing the weighted number of burned nodes includes: treating each grid as a node, constructing a logistic Benders decomposition solution framework, wherein the main problem adopts mixed integer programming, and the decision variables include... (Indicates whether resources are deployed at node n and time t) and (Indicates whether node n is burned), the objective function is:
[0031]
[0032] In the formula, The set of all nodes; For nodes Protection weight; For the indicator variable of being burned ( This indicates that the node has been burned. (This indicates that it was not burned).
[0033] The subproblem uses Dijkstra's shortest path algorithm to verify the fire arrival time under a given resource deployment; when the integer solution to the main problem is infeasible, a Benders cut constraint is generated through a callback function; the optimal time for each node to deploy fire suppression resources is output.
[0034] On the other hand, this invention proposes a forest fire fire dispatching auxiliary decision-making system driven by combustible material classification and physical spread rate to construct a directed graph model. The system includes:
[0035] The fire environment modeling module is configured to construct an initial directed graph model covering the entire study area based on satellite optical remote sensing images, digital elevation model (DEM), and overall wind field of the study area.
[0036] The fire risk and protection value assessment module is configured to automatically identify potential protection targets based on the results of the combustible classification model and generate basic protection weights based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, it generates the protection weight of each node and assigns node weights to the directed graph model.
[0037] The fire spread arc weight calculation module is configured to calculate the 8-directional spread rate from each grid to the adjacent grid based on the physical parameters of the combustible material type, combined with the terrain features, using the Rothermel equation and the elliptical fire shape model, and perform robust correction based on the classification confidence. Then, the spread rate is converted into the directed arc travel time to generate weighted directed arc edges for the directed graph model.
[0038] The fire resource optimization and scheduling module is configured to minimize the number of weighted burned nodes and use the logical Benders decomposition algorithm to solve for the optimal fire extinguishing resource deployment scheme on the directed graph model.
[0039] The beneficial technical effects of this invention are:
[0040] This invention constructs an initial directed graph model to lay the foundation for subsequent node weight assignment and arc edge generation. Then, based on the XGBoost combustible material classification model, it automatically identifies potential protection targets and generates basic protection weights based on the JRC framework, combining user-interactive annotation to achieve differentiated node weight assignment. Next, based on the Rothermel physical equations and the elliptic fire shape model, it calculates the 8-directional spread rate and uses classification confidence for robust correction, converting the spread rate into directed arc travel time, thus completing the construction of the weighted directed graph model. Finally, with the goal of minimizing the number of weighted burned nodes, it uses the logistic Benders decomposition algorithm to solve for the optimal fire extinguishing resource deployment scheme on the directed graph model, achieving a fully automated closed loop from satellite imagery to fire extinguishing dispatch scheme. This approach offers advantages such as physical realism and strong differentiated protection capabilities, providing a scientific and efficient auxiliary decision-making tool for fire command. Attached Figure Description
[0041] The present invention can be better understood by referring to the description given below in conjunction with the accompanying drawings, which together with the following detailed description are included in and form part of this specification, and are used to further illustrate preferred embodiments of the invention and explain the principles and advantages of the invention.
[0042] Figure 1 This is a flowchart of a forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate driven by a directed graph model, according to an embodiment of the present invention.
[0043] Figure 2This is another flowchart of a forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction, according to an embodiment of the present invention.
[0044] Figure 3 This is a schematic diagram showing the results of grid division and combustible material type classification in the study area in this embodiment;
[0045] Figure 4 This is a diagram illustrating the user-interactive annotation of mandatory protected cells and the propagation effect of spatial weights in this embodiment;
[0046] Figure 5 This is a schematic diagram of the directed graph model in this embodiment;
[0047] Figure 6 This is a schematic diagram showing the result of not setting up fire extinguishing resource deployment in this embodiment;
[0048] Figure 7 This is a schematic diagram of the optimal fire extinguishing resource deployment result in this embodiment. Detailed Implementation
[0049] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.
[0050] This invention proposes a forest fire fire dispatch auxiliary decision-making method that integrates combustible material classification and physical spread simulation, such as... Figures 1-2 As shown, it includes the following steps:
[0051] S1: Based on satellite optical remote sensing images, digital elevation model (DEM) and overall wind field of the study area, construct an initial directed graph model covering the entire study area;
[0052] S2: Based on the results of the combustible material classification model, potential protection targets are automatically identified and basic protection weights are generated based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, the protection weight of each node is generated, and node weights are assigned to the directed graph model.
[0053] S3: Based on the physical parameters of combustible material type and combined with terrain features, the Rothermel equation and elliptical fire shape model are used to calculate the spread rate in 8 directions and convert it into arc travel time to generate weighted directed arc edges for the directed graph model.
[0054] S4: The optimal fire extinguishing resource deployment scheme is solved by using the logical Benders decomposition algorithm with the goal of minimizing the weighted number of burned nodes.
[0055] The method begins with S1. In S1, satellite optical remote sensing images, digital elevation models (DEMs), and overall wind fields of the study area are acquired to construct an initial directed graph model covering the entire study area.
[0056] According to an embodiment of the present invention, Landsat-8 OLI multispectral imagery (30m resolution) and the ASTER GDEMV3 digital elevation model are first used as inputs, while overall wind field data (including time series of wind speed and direction) is acquired from meteorological stations or WRF models. The study area is projected onto the UTM coordinate system and divided into an N-row × M-column grid array according to a regular grid of 30m × 30m. Each grid corresponds to a node in the directed graph model. Initially, there are no curved edges between nodes and the node weights are not assigned, forming an initial directed graph model.
[0057] Then S2 is executed. In S2, based on the results of the combustible material classification model, potential protection targets are automatically identified and basic protection weights are generated based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, protection weights are generated for each node, thus assigning node weights to the directed graph model.
[0058] According to an embodiment of the present invention, an XGBoost ensemble learning classification model is used for the classification of combustible materials. As an example, the specific steps include:
[0059] S21. Extract the 10-dimensional image spectral feature vector and terrain features for each grid. The 10-dimensional image spectral feature vector includes: 6-dimensional spectral features (mean and standard deviation of RGB three channels), 2-dimensional vegetation index (mean and standard deviation of NDVI approximation), and 2-dimensional color features (mean of brightness and mean of saturation). The terrain features are extracted from the digital elevation model and include: elevation, slope, and aspect. The image spectral feature vector serves as the input feature for combustible material classification and protected target identification.
[0060] S22. Select 300 sample plots (each plot is 30m × 30m), and determine the combustible material types (13 categories in total) for each plot through manual labeling. Extract the aforementioned 10-dimensional feature vector for each sample plot to form a training sample set:
[0061]
[0062]
[0063] in, Indicates the first 10-dimensional image spectral feature vectors extracted from each sample plot; Indicates the first Each sample plot has a corresponding combustible material type label, with the label value being an integer from 1 to 13;
[0064] S23. Construct an XGBoost tree ensemble model for classifying 13 types of combustibles. For the first... Training samples and combustible material categories , Indicates the category number of combustible materials. The model outputs a category score. :
[0065]
[0066] In the formula, Indicates the number of base learners; Indicates the first Trees for categories The output function;
[0067] The softmax function is then used to convert the class scores into class probabilities. :
[0068]
[0069] In the formula, Indicates the category number of the combustible material used in the summation; Indicates the first The training samples belong to the combustible material category. The predicted probability;
[0070] The objective function for model training is:
[0071]
[0072] In the formula, Indicates the first The predicted probability vector is composed of the probabilities of the 13 categories corresponding to each training sample. ; Represents the cross-entropy function for multi-class classification; Indicates the first A decision tree; This represents the model complexity penalty term; This represents the penalty coefficient for the number of leaf nodes; Indicates the first The number of leaf nodes in a decision tree; The L2 regularization coefficient represents the weight of the leaf node; Indicates the first The weight vectors of the leaf nodes of a decision tree.
[0073] S24. Set hyperparameters and use cross-validation for training;
[0074] S25. After training, each grid within the study area is identified and classified. Indicates the grid number, , This represents the total number of grid cells within the study area. The model output is the [number]th [cell]. The grid belongs to the first Predicted probability of combustible materials , No. The combustible material type number for each grid is selected from the category with the highest predicted probability:
[0075]
[0076] No. The classification confidence score for each grid cell is the maximum value among the predicted probabilities of the 13 categories:
[0077]
[0078] In the formula, Indicates the first The combustible material type number for each network; Indicates the first The classification confidence of a network.
[0079] Based on the pre-defined correspondence between combustible material types and protected target categories , will the Combustible material type number for each network Mapped to protection target category:
[0080]
[0081] In the formula, Indicates the first The protection target category of the network, step S26 according to the protection target category Determine the basic protection weights for this grid. Figure 3 This diagram illustrates the grid division and combustible material type classification results for the study area.
[0082] S26. Based on the JRC wildfire risk framework, preset basic protection weights for each protection target category: 10 for buildings, 6-8 for farmland, 4-6 for ecologically sensitive areas, 1-4 for general vegetation areas, and 0 for areas with no protection value.
[0083] S27. Users can mark "must-protect grids" on satellite base maps through an interactive interface;
[0084] S28. The system automatically propagates spatial weights to the 8 neighboring grids around the protected grid, increasing the weight of each by 2, with the maximum weight after the addition being 9. Figure 4 The diagram shows the effect of user-annotated cells and spatial weight propagation.
[0085] S29, Final protection weight for each grid As node weights in a directed graph model;
[0086] S210. For grids with confidence levels below the classification confidence threshold, mark them as "uncertain" and use the neighborhood majority voting method for smoothing in subsequent propagation calculations.
[0087] Then, S3 is executed. In S3, based on the physical parameters of the combustible material type and combined with the terrain features, the Rothermel equation and the elliptic fire shape model are used to calculate the spread rate in 8 directions and convert it into arc travel time, generating weighted directed arc edges for the directed graph model.
[0088] According to an embodiment of the present invention, each type of combustible material corresponds to 12 physical parameters. As an example, the specific calculation steps are as follows:
[0089] S31. Query the 12 physical parameters and 1 dynamic flag corresponding to each type of combustible material. The 12 physical parameters include: load (h1, h10, h100, lh, lw), surface area to volume ratio (sav1, savlh, savlw), combustible bed depth, extinguishing moisture content xmext, and calorific value (heatd, heatl); the dynamic flag is used to indicate whether the combustible material participates in the propagation process.
[0090] S32, Calculate the base spread rate The Rothermel propagation equation is adopted:
[0091]
[0092] In the formula, This represents the base spread rate of the current grid without any classification confidence correction. Indicates the reaction intensity (determined by the surface area-to-volume ratio and the quenching moisture content); This indicates the fire front propagation flux ratio (determined by the surface area-to-volume ratio and the extinguishing moisture content). This indicates the density of the combustible bed (determined by the load and the depth of the combustible bed). Indicates the effective heating coefficient; It represents the amount of heat required to ignite a unit mass of combustible material (determined by calorific value).
[0093] S33. Using the terrain features extracted in S21 and combined with the overall wind field data, calculate the local terrain wind field for each grid. The WindNinja model is used to simulate local wind speed and direction.
[0094] S34. Calculate the comprehensive wind slope factor:
[0095]
[0096]
[0097] In the formula, For comprehensive wind slope factors; For wind factors; Wind speed; Wind direction; The direction in which the fire starts moving; The slope angle is denoted by .
[0098] S35. Use the elliptical fire pattern model to determine the fire head direction and calculate the fire head rate:
[0099]
[0100] S36, For the angle between the flame direction and the fire direction is... The direction and propagation rate are calculated using the elliptic interpolation formula:
[0101]
[0102]
[0103] S37. Repeat S36 to calculate the spread rate in each grid in 8 directions (N, NE, E, SE, S, SW, W, NW).
[0104] S38. For each network, if the classification confidence level... Calculate the buffer size:
[0105]
[0106] In the formula, For the first Classification confidence of each network; The maximum spread rate; This represents the base spread rate of the current grid without classification confidence correction. This is the scaling factor.
[0107] S39, Corrected propagation rate for:
[0108]
[0109] S310. Convert the 8-directional spread rate into the arc travel time of the directed graph model structure: the directed arc travel time between adjacent nodes:
[0110]
[0111] In the formula, This refers to the node spacing; For the modified base spread rate The data obtained from the elliptical fire-shaped model is derived from the nodes. Point to adjacent nodes directional spread rate;
[0112] The set of all directed arcs is denoted as The set of travel times is denoted as To form a directed graph model . Figure 5 A schematic diagram of the principle of a directed graph model is shown.
[0113] Then, S4 is executed. In S4, with the objective of minimizing the weighted number of burned nodes, the logical Benders decomposition algorithm is used to solve for the optimal fire extinguishing resource deployment scheme on the directed graph model.
[0114] According to an embodiment of the present invention, the optimal fire suppression resource deployment scheme is solved using the logical Benders decomposition algorithm, and the specific steps are as follows:
[0115] S41. Define decision variables and parameters. Decision variables include: Indicates at time Is it at the node? Deploy firefighting resources; This represents the upper limit of the target time window starting from the start of scheduling; Indicates the discrete time step; Represents the maximum discrete time sequence required to cover the target time window. ; Represents a node Whether it was burned out within the target time window. Input parameters include: set of ignition points. Firefighting delay This indicates the delay time after resource deployment; time. Number of available firefighting resources Node protection weight ; set of nodes that must be protected ;
[0116] S42. Establish the objective function as minimizing the weighted number of burned nodes:
[0117]
[0118] S43. Establish constraints. Resource quantity constraints:
[0119]
[0120] Resource deployment frequency constraints:
[0121]
[0122] Ignition point resource constraints:
[0123]
[0124] Required protection node constraints:
[0125]
[0126] S44. Solve using the logical Benders decomposition framework. The main problem is mixed-integer programming, solving for resource deployment schemes:
[0127]
[0128] The subproblem is given a deployment scheme The fire spread simulation was performed, and Dijkstra's shortest path algorithm was used to calculate the time it takes for the fire to reach each node.
[0129] Directed edge The weights are:
[0130]
[0131] In the formula, A collection of nodes on which resources have been deployed; For indicator functions;
[0132] Dijkstra's algorithm for initializing ignition point arrival time:
[0133]
[0134] Update adjacent nodes according to the shortest path:
[0135]
[0136] S45. Generate Benders optimal cut. If the principal problem is determined... (considering nodes) (Protected), but sub-problem verification revealed (The actual fire arrival time is earlier than the target time window, i.e., the node) If it is not protected, then calculate:
[0137]
[0138] Add cut:
[0139]
[0140] The meaning of this cut is: to protect the node. It needs to be deployed at least along the fire path. Resource nodes.
[0141] S46. Generate Benders feasibility cuts. If resources are deployed on nodes... The moment Later than the time when fire arrives at that node (i.e. If the resource is invalid, the calculation will be performed.
[0142]
[0143] Add cut:
[0144]
[0145] The meaning of this cut is: if in of Deploying resources at any given time is ineffective. To satisfy the constraints, there are two necessary scenarios: first, cancel the resource deployment on this node; second, add an effective deployment on an upstream node in the fire path.
[0146] S47. Iterate until convergence. Steps S41 to S43 are used to define decision variables and parameters and establish the objective function and constraints, and are executed only once during the modeling phase. During the iteration phase, steps S44 to S46 are repeated.
[0147] As an example, the specific steps to solve S47 include:
[0148] S471. Construct a mixed integer programming model for the main problem and add an initial cut (based on the resource-free Dijkstra result).
[0149] S472. Enter the loop iteration: Solve the main problem and obtain integer solutions. and ;
[0150] S473. Perform the Dijkstra subproblem with deployment options to obtain the fire arrival time. ;
[0151] S474, For each node Execute judgment: If Then add Benders optimal cut; if it exists Then add Benders feasibility cut;
[0152] S475. If no new cut is added in this iteration, the loop converges and exits; otherwise, return to S472 to continue iterating.
[0153] S476, Output the optimal deployment plan And the target value. If there is no solution, return to S41 to redefine the parameters. Figure 6 This diagram illustrates the result of not deploying firefighting resources. Figure 7 A schematic diagram showing the optimal deployment of firefighting resources is presented.
[0154] Another embodiment of the present invention proposes a forest fire fire dispatching auxiliary decision-making system based on combustible material classification and physical spread rate-driven directed graph model construction. The system includes:
[0155] The fire environment modeling module is configured to construct an initial directed graph model covering the entire study area based on satellite optical remote sensing images, digital elevation model (DEM), and overall wind field of the study area.
[0156] The fire risk and protection value assessment module is configured to automatically identify potential protection targets based on the results of the combustible classification model and generate basic protection weights based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, it generates the protection weight of each node and assigns node weights to the directed graph model.
[0157] The fire spread arc weight calculation module is configured to calculate the 8-directional spread rate from each grid to the adjacent grid based on the physical parameters of the combustible material type, combined with the terrain features, using the Rothermel equation and the elliptical fire shape model, and perform robust correction based on the classification confidence. Then, the spread rate is converted into the directed arc travel time to generate weighted directed arc edges for the directed graph model.
[0158] The fire resource optimization and scheduling module is configured to minimize the number of weighted burned nodes and uses the logical Benders decomposition algorithm to solve for the optimal fire extinguishing resource deployment scheme.
[0159] The function of the forest fire dispatching auxiliary decision-making system driven by combustible material classification and physical spread rate to construct a directed graph model can be described by the aforementioned forest fire dispatching auxiliary decision-making method driven by combustible material classification and physical spread rate to construct a directed graph model. Therefore, for the parts not described in detail in this embodiment, please refer to the above method embodiments, and they will not be repeated here.
[0160] In summary, this invention discloses a forest fire fire dispatching auxiliary decision-making method and system driven by combustible material classification and physical spread rate to construct a directed graph model. Key technical points include: constructing an initial directed graph model based on satellite remote sensing imagery, digital elevation model, and overall wind field; automatically identifying combustible material types and protection target categories based on the combustible material classification model, generating basic protection weights using the JRC risk framework, and then spatially propagating these weights to the directed graph model by user-labeled mandatory protection grids; calculating the 8-direction spread rate based on the Rothermel equation and elliptic fire shape model and converting it into arc travel time to generate weighted directed arc edges; and using the logistic Benders decomposition algorithm to solve for the optimal fire-fighting resource deployment scheme on the directed graph model with the objective of minimizing the weighted number of burned nodes. This invention achieves a fully automated closed loop from satellite imagery to fire-fighting dispatching, possessing advantages such as high physical realism and support for differentiated protection. It can assist commanders in making scientific and efficient command decisions, thereby significantly improving the efficiency of forest fire emergency response.
[0161] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. The disclosure of the invention is illustrative and not restrictive, and the scope of the invention is defined by the appended claims.
Claims
1. A forest fire fire dispatching auxiliary decision-making method driven by combustible material classification and physical spread rate-driven directed graph model construction, characterized in that, Includes the following steps: S1: Based on satellite optical remote sensing images, digital elevation models and overall wind fields of the study area, construct an initial directed graph model covering the entire study area; S2: Based on the results of the combustible material classification model, potential protection targets are automatically identified and basic protection weights are generated based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, the protection weight of each node is generated, and node weights are assigned to the directed graph model. S3: Based on the physical parameters of combustible material type and combined with terrain features, the Rothermel equation and elliptical fire shape model are used to calculate the spread rate in 8 directions and convert it into arc travel time to generate weighted directed arc edges for the directed graph model. S4: The optimal fire extinguishing resource deployment scheme is solved by using the logical Benders decomposition algorithm with the goal of minimizing the weighted number of burned nodes.
2. The forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction according to claim 1, characterized in that, Step S2 includes: A 10-dimensional image-spectral feature vector is extracted from each grid cell in satellite optical remote sensing imagery, serving as input features for combustible material classification and target protection identification. The 10-dimensional feature vector includes: 6-dimensional spectral features (mean and standard deviation of the RGB channels); 2-dimensional vegetation index (mean and standard deviation of the approximate NDVI value); and 2-dimensional color features (mean of brightness and mean of saturation). This 10-dimensional feature vector is input into the XGBoost combustible material classification model, which is trained using a training sample set. After training, the model outputs values for each grid cell simultaneously. The system includes fuel type number, classification confidence level, and protection target category. The protection target categories include built-up areas, farmland, ecologically sensitive areas, general vegetation areas, and areas with no protection value. Based on the JRC wildfire risk framework, corresponding basic protection weights are preset for each protection target category, with built-up areas assigned a value of 10, farmland assigned a value of 6-8, ecologically sensitive areas assigned a value of 4-6, general vegetation areas assigned a value of 1-4, and areas with no protection value assigned a value of 0. For grids with classification confidence levels lower than the classification confidence level threshold, a neighborhood majority voting method is used for smoothing.
3. The forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction according to claim 1, characterized in that, The method for calculating the 8-directional propagation rate in step S3 specifically includes: The terrain features of each grid are extracted from the digital elevation model and used as input parameters for terrain wind field correction and spread rate calculation. These terrain features include elevation, slope, and aspect. Twelve physical parameters and one dynamic flag are queried based on the combustible material type number. The 12 physical parameters include: h1, representing the dead combustible material load in 1 hour; h10, representing the dead combustible material load in 10 hours; h100, representing the dead combustible material load in 100 hours; lh, representing the live herbaceous combustible material load; lw, representing the live woody combustible material load; sav1, The following values represent the surface area-to-volume ratio of dead combustibles in 1 hour: savlh, savlw, depth, xmext, heatd, heatl, and dynamic flags. The dynamic flag indicates whether the corresponding combustible model is a dynamic combustible model and is used to determine the conversion method from live herbaceous load to dead herbaceous load. The base propagation rate was calculated using the Rothermel propagation equation: In the formula, This represents the base spread rate of the current grid without any classification confidence correction. Indicates the intensity of the reaction; Indicates the fire front propagation flux ratio; Indicates the density of the combustible material bed; Indicates the effective heating coefficient; This represents the amount of heat required to ignite a unit mass of combustible material. Using the aforementioned terrain features and combined with overall wind field data, the local terrain wind field of each grid is calculated to obtain local wind speed and wind direction; Calculate the comprehensive wind slope factor: In the formula, For comprehensive wind slope factors; For wind factors; Wind speed; Wind direction; The direction in which the fire starts moving; The slope angle; Then calculate the flame rate: For the angle with the direction of the fire head is The direction and propagation rate are calculated using the elliptic interpolation formula: This yields the 8-directional propagation rate for each grid.
4. The forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction according to claim 1, characterized in that, Step S3 also includes robust correction based on classification confidence, specifically including: When the classification confidence of a certain grid Below the propagation confidence threshold At that time, a conservative buffer is added to the base spread rate of the grid, and the conservative buffer is calculated by the following formula: In the formula, For classification confidence; The maximum spread rate; This represents the base spread rate of the current grid without classification confidence correction. This is the scaling factor; The base spread rate, corrected for classification confidence, is used for calculating arc travel time in subsequent scheduling.
5. The forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction according to claim 1, characterized in that, Step S4 includes: Treating each grid cell as a node, the travel time of the directed arc between adjacent grid cells is: In the formula, This refers to the node spacing; For the node To the node Robust spread rate in the direction; Construct a logical Benders decomposition framework, with the main problem being mixed-integer programming, and the decision variables including: and , Indicates at time Is it at the node? Deploy firefighting resources, Represents a node Whether it was burned out within the target time window; The objective function is N represents the set of all nodes. Represents a node Protection weight; Constraints include resource quantity constraints, single deployment constraints, no-deployment constraints at ignition points, and mandatory protection constraints; the subproblem uses Dijkstra's algorithm to simulate fire spread and calculate the fire arrival time under a given resource deployment. When the solution to the main problem does not satisfy the subproblems, a Benders optimal cut or feasible cut is generated through a callback function; the dominance inequality is injected based on the fire path tree structure to accelerate the solution; the optimal time for deploying fire extinguishing resources for each grid is output.
6. The forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction according to claim 5, characterized in that, The Benders optimality cut includes: if And the time when fire reaches node n Then add a cut: In the formula, The deadline for the target time window; Fire suppression delays caused by deploying a single effective resource; This represents the distance from the ignition point to the node obtained from the Dijkstra subproblem. The shortest fire spread path node sequence; This indicates that the path node sequence excluding the target node... All upstream nodes except those mentioned above; express Any upstream node in; The feasibility cut includes: if resources are deployed on nodes The moment Later than the time when fire arrives at that node, that is If the resource is invalid, add a cut: 。 7. The forest fire fire dispatching auxiliary decision-making method based on combustible material classification and physical spread rate-driven directed graph model construction according to claim 5, characterized in that, Preset discrete time step Determine the maximum discrete time sequence required to cover the target time window. : The resource quantity constraint is as follows: in, For a moment The number of available firefighting resources, ; The single deployment constraint is: The restriction prohibiting deployment at the ignition point is as follows: in, Gather at the point of ignition. ; The mandatory protection constraint is: in, To protect the set of nodes, .
8. A forest fire fire dispatching auxiliary decision-making system driven by combustible material classification and physical spread rate-based directed graph model construction, characterized in that, include: The fire environment modeling module is configured to construct an initial directed graph model covering the entire study area based on satellite optical remote sensing images, digital elevation models, and overall wind fields of the study area. The fire risk and protection value assessment module is configured to automatically identify potential protection targets based on the results of the combustible classification model and generate basic protection weights based on the JRC wildfire risk framework. Combined with the user-annotated mandatory protection grid information, it generates the protection weight of each node and assigns node weights to the directed graph model. The fire spread arc weight calculation module is configured to calculate the 8-directional spread rate from each grid to the adjacent grid based on the physical parameters of the combustible material type, combined with the terrain features, using the Rothermel equation and the elliptical fire shape model, and perform robust correction based on the classification confidence. Then, the spread rate is converted into the directed arc travel time to generate weighted directed arc edges for the directed graph model. The fire resource optimization and scheduling module is configured to minimize the number of weighted burned nodes and uses the logical Benders decomposition algorithm to solve for the optimal fire extinguishing resource deployment scheme.
9. A computer device, characterized in that: The system includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, it performs the steps of the forest fire fire dispatch auxiliary decision-making method according to any one of claims 1 to 7, which is based on the classification of combustibles and the physical spread rate-driven construction of a directed graph model.
10. A computer-readable storage medium, characterized in that: The storage medium is used to store a computer program that executes a forest fire fire dispatching auxiliary decision-making method based on the construction of a directed graph model driven by combustible material classification and physical spread rate, as described in any one of claims 1 to 7.
