Forest fire spread simulation method and system based on UAV LiDAR technology

Abstract

This invention provides a forest fire spread simulation method and system based on drone LiDAR technology. The method includes: using drone LiDAR technology, cameras, and sensors to collect forest area data; the forest area data includes LiDAR point cloud data, thermal infrared data, image data, and meteorological data; extracting forest area features based on the forest area data to obtain characteristic parameters; constructing a fire spread dynamic model based on the characteristic parameters using a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm; simulating and predicting fire spread based on the fire spread dynamic model, and providing dynamic early warnings based on the simulation and prediction results to inform firefighting and prevention decisions. The technical solution of this invention is of great significance for protecting forest resources and maintaining ecological balance.

Description

Technical Field

[0001] The present invention belongs to the technical field of dynamic simulation of forest fire spread, and in particular relates to a method and system for simulating forest fire spread based on UAV LiDAR technology. Background Art

[0002] As global climate change intensifies, the frequency and intensity of extreme climate events, such as high temperatures and droughts, have significantly increased the frequency and intensity of forest fires. Traditional forest fire spread models typically rely on ground observations and satellite data, but these methods have significant limitations. Ground observations have limited coverage and long data acquisition cycles, making it difficult to meet the needs of real-time fire simulations. Satellite data typically has a resolution of 10-30 meters and a temporal resolution of 5-16 days, which cannot accurately depict complex forest fuel structures or rapidly changing fire conditions. In addition, existing models have low accuracy for complex terrain and heterogeneous fuel distribution areas, making it difficult to fully capture the dynamic characteristics of forest fire spread.

[0003] The rapid development of drone-based LiDAR technology in recent years has provided new data support for forest fire spread simulations. LiDAR point cloud data, with its high spatial resolution (over 200 points per square meter), high accuracy, and three-dimensional structure, can provide detailed characterization of forest fuel distribution and topographic features. This technology has particular potential in subtropical forest environments, addressing the shortcomings of traditional methods and providing a crucial technical foundation for improving the accuracy and real-time performance of forest fire spread simulations. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention utilizes UAV LiDAR technology and multiple sensors to collect multimodal data to achieve dynamic prediction and early warning of forest fire spread trends.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A forest fire spread simulation method based on UAV LiDAR technology, the method comprising:

[0007] Using drone LiDAR technology, cameras, and sensors, the system collects forest area data, including LiDAR point cloud data, thermal infrared data, image data, and meteorological data.

[0008] Extracting forest area features based on the forest area data to obtain feature parameters;

[0009] Based on the characteristic parameters, a fire spread dynamic model is constructed through a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm;

[0010] Based on the fire spread dynamic model, fire spread simulation and prediction are carried out, and dynamic early warning is carried out based on the simulation and prediction results to obtain fire fighting and fire prevention decisions.

[0011] Preferably, the characteristic parameters include terrain parameters, fuel parameters and meteorological parameters;

[0012] The terrain parameters include the altitude, slope and aspect of the forest area;

[0013] The fuel parameters include surface fuel type, fuel moisture characteristics, surface fuel loading, canopy density, vertical fuel distribution, and canopy gap distribution;

[0014] The meteorological parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure and light intensity.

[0015] Preferably, the method for constructing a fire spread dynamic model includes:

[0016] Based on terrain parameters, fuel parameters, and meteorological parameters, a deep reinforcement learning model is constructed, which treats the spread of surface fire and canopy fire as a Markov decision process to obtain the dynamic fire line position and intensity.

[0017] Based on the flame heat radiation transfer equation and air turbulence model, combined with the canopy gap distribution, the flame jump distance and probability are calculated to obtain a coupled model of the vertical spread and horizontal diffusion of the canopy flame.

[0018] Based on the coupling model, a three-dimensional fire propagation path is simulated;

[0019] The fire spread dynamic model is constructed based on the dynamic fire line position and intensity and the three-dimensional fire propagation path.

[0020] Preferably, the method for obtaining the dynamic firing line position and intensity comprises:

[0021] discretizing the terrain parameter, the fuel parameter, and the meteorological parameter;

[0022] The spatial distribution characteristics of discretized terrain parameters and discretized fuel parameters are obtained based on convolutional neural networks, and the temporal characteristics of discretized meteorological parameters are captured based on long short-term memory networks.

[0023] Based on the spatial distribution characteristics and the temporal characteristics, a state space is constructed; the position and intensity of the line of fire are used as the action space;

[0024] Using the state space as the input of the deep Q network and the action space as the output of the deep Q network, updating the network weights by minimizing the loss function to train the intelligent agent;

[0025] The intelligent agent selects the optimal action according to the current state, simulates the Markov decision process, and obtains the dynamic fire line position and intensity.

[0026] Preferably, the method for obtaining a coupled model of vertical and horizontal spread of a canopy fire includes:

[0027] Using the radiation heat transfer theory, the flame heat radiation transfer equation is established, and based on the flame heat radiation transfer equation, the propagation and attenuation process of flame heat radiation in the forest canopy is obtained;

[0028] Based on the air turbulence intensity, an air turbulence model is used to simulate the effect of air turbulence on flame spread;

[0029] Based on the propagation and attenuation process of flame thermal radiation in the forest canopy and the influence of air turbulence on flame spread, combined with the canopy gap distribution, the flame jump distance and probability are calculated to obtain a coupled model of vertical and horizontal spread of forest canopy flames.

[0030] Preferably, it also includes integrating LiDAR point cloud data, fire spread simulation and prediction results, and real-time meteorological data to establish a three-dimensional dynamic visualization platform, dynamically display the movement trend of the fire line through thermal maps and particle effects, and divide the fire risk level.

[0031] The present invention also provides a forest fire spread simulation system based on UAV LiDAR technology, which is used to implement the method, including:

[0032] A data acquisition module, configured to collect forest area data using drone LiDAR technology, cameras, and sensors; wherein the forest area data includes LiDAR point cloud data, thermal infrared data, image data, and meteorological data;

[0033] A feature parameter extraction module is used to extract forest area features based on the forest area data to obtain feature parameters;

[0034] a fire spread dynamic model construction module, configured to construct a fire spread dynamic model based on the characteristic parameters by using a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm;

[0035] The fire spread simulation module is used to simulate and predict the spread of fire based on the fire spread dynamic model, and to issue dynamic warnings based on the simulation and prediction results to obtain fire fighting and fire prevention decisions.

[0036] Preferably, in the characteristic parameter extraction module, the characteristic parameters include terrain parameters, fuel parameters and meteorological parameters;

[0037] The terrain parameters include the altitude, slope and aspect of the forest area;

[0038] The fuel parameters include surface fuel type, fuel moisture characteristics, surface fuel loading, canopy density, vertical fuel distribution, and canopy gap distribution;

[0039] The meteorological parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure and light intensity.

[0040] Compared with existing technologies, this invention offers the following advantages: by integrating multi-source data, such as drone LiDAR technology, it can more accurately identify fire points and predict fire spread trends, thereby improving the accuracy of early warnings. Simulations and predictions based on dynamic fire spread models enable more scientific and rational firefighting and prevention decisions, improving rescue efficiency and safety. Dynamic early warnings and real-time data updates enable faster responses to fires, shortening rescue response times and reducing fire losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is a flow chart of a forest fire spread simulation method based on UAV LiDAR technology according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Example 1

[0046] like Figure 1 As shown in FIG, a forest fire spread simulation method based on UAV LiDAR technology includes:

[0047] S1: Using drone LiDAR technology, cameras, and sensors, collect forest area data. This data includes LiDAR point cloud data, thermal infrared data, image data, and meteorological data. Specifically, a thermal infrared camera is used to collect thermal infrared data, and an RGB camera is used to collect RGB image data.

[0048] In this embodiment, the UAV's dynamic flight path optimization is performed based on real-time meteorological conditions (such as wind speed, temperature) and terrain complexity (through pre-collected DEM data). The three-dimensional forest model is discretized into a grid map, and each grid node stores the terrain height, fire intensity and risk level. The initial path from the starting point to the target point is searched by the A* algorithm. Based on the initial path generated by the A* algorithm, the local path is optimized by the dynamic programming method to reduce the path length and risk. The UAV automatically adjusts the flight altitude (30-150 meters) and overlap rate (30%-50%) according to the fire risk level to ensure that the data collection density in high-risk areas is greater than or equal to 300 points / m2, and transmits LiDAR point cloud data in real time through edge computing compression to reduce latency.

[0049] S2: extracting forest area features based on the forest area data to obtain characteristic parameters; in a further embodiment, the characteristic parameters include terrain parameters, fuel parameters, and meteorological parameters;

[0050] Terrain parameters include the altitude, slope and aspect of the forest area; in this embodiment, a multi-scale point cloud layered processing method is adopted to separate ground points from non-ground points through a density clustering algorithm. For complex terrain (such as steep slopes and gullies), an adaptive filtering algorithm is introduced to dynamically adjust the filter window size (3×3 to 15×15 pixels) to eliminate vegetation interference. Kriging interpolation method (Kriging) is combined to generate a high-precision DEM (resolution ≤ 0.5 meters), and the vector differential algorithm of slope and aspect is used to extract terrain parameters. In this embodiment, the pressure sensor carried by the drone is synchronously calibrated with the LiDAR elevation data to eliminate the influence of atmospheric pressure fluctuations on altitude measurement. Combined with real-time wind speed data, a relationship model between terrain undulation and local turbulence intensity is established, and the influence weight of slope aspect on fire spread is dynamically corrected.

[0051] Fuel parameters include surface fuel type, fuel moisture characteristics, surface fuel load, canopy density, vertical fuel distribution and canopy gap distribution; in this embodiment, the surface fuel type is obtained through image recognition based on RGB image data. The fuel moisture characteristics are obtained using the surface temperature data of the infrared thermal camera as well as the air temperature and relative humidity. Combining the ground reflection intensity of the LiDAR point cloud with the texture characteristics of the RGB image, a convolutional neural network (CNN) model is trained to classify and identify surface fuel types such as fallen leaves, shrubs, and tree trunks, and the load per unit area is inverted through a regression model. The LiDAR point cloud is segmented using a voxelization method (voxel size 0.1m 3 ) to calculate the point cloud density within each voxel layer and generate a vertical canopy profile. Combined with canopy temperature gradients inverted from thermal infrared data, this method optimizes the vertical flammability assessment of fuels. Using a 3D point cloud skeleton extraction algorithm, the geometric characteristics (area, shape index) of canopy gaps are identified and their spatial connectivity is calculated.

[0052] Meteorological parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure, and light intensity.

[0053] S3: Based on the characteristic parameters, a fire spread dynamic model is constructed using a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm. A further embodiment is that the method for constructing the fire spread dynamic model includes:

[0054] S31: Based on terrain parameters, fuel parameters, and meteorological parameters, a deep reinforcement learning model is constructed, and the surface fire and canopy fire spread process is used as a Markov decision process to obtain the dynamic fire line position and intensity. A further embodiment is that the method for obtaining the dynamic fire line position and intensity includes:

[0055] The terrain parameters, fuel parameters and meteorological parameters are discretized; specifically, in this embodiment, the terrain parameters (altitude, slope, slope direction) are divided into a regular grid (such as 10m×10m), and each grid cell stores the terrain characteristic value. Missing data is filled by bilinear interpolation to ensure spatial continuity. The terrain complexity index (TCI) is introduced to dynamically adjust the grid resolution. High-complexity areas (such as steep slopes and gullies) use a higher resolution (5m×5m). The fuel parameters (surface fuel load, canopy density, vertical fuel distribution, etc.) are discretized in layers, with each layer being 0.5m thick, and the grid cells are aligned with the terrain parameters. The fuel distribution is smoothed by a three-dimensional convolution kernel to eliminate noise. The fuel flammability index (FCI) is proposed to comprehensively consider fuel load, humidity and vertical distribution to quantify the flammability of grid cells. The meteorological parameters (temperature, humidity, wind speed, wind direction, etc.) are discretized in time series with a time step of 5 minutes. The sliding window method is used to extract local meteorological features to reduce data fluctuations.

[0056] The spatial distribution characteristics of discretized terrain parameters and discretized fuel parameters are obtained based on convolutional neural networks, and the temporal characteristics of discretized meteorological parameters are captured based on long-short-term memory networks. Specifically, regarding the extraction of spatial distribution characteristics, a multi-layer CNN network is constructed, with the input being the discretized grid data of terrain and fuel parameters and the output being a high-dimensional feature vector. The network structure includes:

[0057] Input layer: multi-channel grid data of terrain + fuel parameters.

[0058] Convolutional layer: A 3×3 convolution kernel extracts local spatial features. Specifically, an attention mechanism is introduced to dynamically weight the importance of terrain and fuel features, improving the targeted nature of feature extraction.

[0059] Pooling layer: Maximum pooling to reduce feature dimensions.

[0060] Fully connected layer: Outputs a 128-dimensional feature vector. The terrain and fuel feature vectors are concatenated and reduced to 64 dimensions through principal component analysis (PCA) as part of the state space.

[0061] Regarding the extraction of time series features, a bidirectional LSTM (BiLSTM) is introduced to simultaneously capture past and future meteorological change trends and output a 64-dimensional time series feature vector.

[0062] Based on spatial distribution features and temporal features, a state space is constructed; the position and intensity of the line of fire are used as the action space; specifically, the spatial features (64 dimensions) and the temporal features (64 dimensions) are spliced ​​together to construct a 128-dimensional state space vector.

[0063] The state space is used as the input of a Deep Q-Network (DQN), and the action space is used as the output of the DQN. The network weights are updated by minimizing the loss function to train the agent. Specifically, the state space is a 128-dimensional feature vector representing the combined state of terrain, fuel, and weather. The action space is the location (grid coordinates) and intensity (low / medium / high) of the fire line, with a total of N×3 actions (N is the number of grid cells).

[0064] The DQN network structure includes:

[0065] Input layer: 128-dimensional state space vector.

[0066] Hidden layer: 2 fully connected layers (256 neurons), ReLU activation function.

[0067] Output layer: N×3 dimensional action space vector, Softmax activation function.

[0068] Loss function:

[0069] Among them, Q(s,a) is the current Q value, r is the immediate reward, and γ is the discount factor.

[0070] Prioritized Experience Replay (PER) is introduced to prioritize high-error samples for training and accelerate convergence. Specifically, state transition samples (s, a, r, s′) are stored and randomly sampled for training. The target Q network is regularly updated to stabilize the training process.

[0071] The agent selects the optimal action based on its current state, simulating a Markov decision process to determine the dynamic position and intensity of the fire line. Specifically, the agent updates the fire line's position and intensity based on the current state and action, and predicts the next state based on weather changes.

[0072] Reward function: r = w1*fire intensity + w2*spread speed + w3*risk area coverage,

[0073] Among them, w1, w2, and w3 are weight coefficients, which are calibrated through experiments.

[0074] S32: Based on the flame heat radiation transfer equation and air turbulence model, combined with the canopy gap distribution, the flame jump distance and probability are calculated to obtain a coupled model of the vertical spread and horizontal diffusion of the canopy flame;

[0075] A further embodiment is that the method for obtaining a coupled model of vertical and horizontal spread of a canopy fire includes:

[0076] S321: Using the radiation heat transfer theory, the flame heat radiation transfer equation is established, and based on the flame heat radiation transfer equation, the flame heat radiation propagation and attenuation process in the forest canopy is obtained. In this embodiment, the non-gray body radiation model is used, the flame is regarded as a multi-wavelength radiation source, and the canopy fuel density distribution extracted by LiDAR is combined to calculate the radiation intensity attenuation coefficient (κ λ ) and scattering coefficient (σ λ ). The RTE is solved by the discrete coordinate method (DOM), and the formula is:

[0077]

[0078] Among them, I λ is the radiation intensity, s is the path length, and Φ is the scattering phase function. κ is dynamically adjusted by introducing the vertical distribution data of canopy fuel. λ and σ λ , improving the scene adaptability of radiation heat transfer calculations.

[0079] S322: Based on the air turbulence intensity, an air turbulence model is used to simulate the influence of air turbulence on flame spread. In this embodiment, a three-dimensional turbulent wind field model based on large eddy simulation (LES) is established to solve the Navier-Stokes equations and capture the wind speed fluctuations and turbulent vortices in the canopy area. The flame is regarded as a heat source term, and the buoyancy effect is embedded in the momentum equation through the Boussinesq approximation, and the temperature field and velocity field are dynamically updated. Among them, with regard to the velocity field, the fire spread rate formula is improved based on the Rothermel model, and the turbulent kinetic energy (TKE) correction term is introduced:

[0080]

[0081] Among them, R0 is the benchmark spreading rate and α is the turbulence intensity coefficient, which is calibrated through experiments.

[0082] S323: Based on the propagation and attenuation of flame thermal radiation in the forest canopy and the influence of air turbulence on flame spread, combined with the distribution of canopy gaps, the flame jump distance and probability are calculated to obtain a coupled model of the vertical and horizontal spread of forest canopy flames.

[0083] Specifically, the preheating effect of flame thermal radiation on combustibles and the promotion of flame jumping by air turbulence were considered to establish the flame jumping mechanism in the gaps between forest canopies, including flame propagation mode, jumping conditions, and other influencing factors. Combining the distribution of gaps between forest canopies and the flame jumping mechanism, and considering the impact of different fire intensities, wind speeds, and combustible types on flame jumping distance and probability, a random walk algorithm was used to simulate the flight trajectory of Mars to establish a flame jumping distance and probability calculation model.

[0084] S33: Based on the coupled model, the three-dimensional fire propagation path is simulated;

[0085] S34: Construct a dynamic fire spread model based on the dynamic fire line position and intensity and the three-dimensional fire propagation path.

[0086] S4: Based on the dynamic model of fire spread, simulate and predict the spread of fire, and issue dynamic warnings based on the simulation and prediction results to obtain fire fighting and fire prevention decisions.

[0087] A further implementation method includes integrating LiDAR point cloud data, fire spread simulation and prediction results, and real-time meteorological data to establish a three-dimensional dynamic visualization platform, dynamically display the movement trend of the fire line through thermal maps and particle effects, and divide the fire risk level.

[0088] Example 2

[0089] The present invention also provides a forest fire spread simulation system based on UAV LiDAR technology, which is used to implement the method, including:

[0090] A data acquisition module, which is used to collect forest area data based on drone LiDAR technology, cameras, and sensors. The forest area data includes LiDAR point cloud data, thermal infrared data, image data, and meteorological data.

[0091] A feature parameter extraction module is used to extract forest area features based on forest area data and obtain feature parameters;

[0092] A fire spread dynamic model construction module is used to construct a fire spread dynamic model based on characteristic parameters through a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm;

[0093] The fire simulation module is used to simulate and predict the spread of fire based on the dynamic model of fire spread, and to issue dynamic warnings based on the simulation and prediction results to obtain fire fighting and fire prevention decisions.

[0094] A further embodiment is that, in the characteristic parameter extraction module, the characteristic parameters include terrain parameters, fuel parameters, and meteorological parameters;

[0095] Topographic parameters include the elevation, slope, and aspect of the forest area;

[0096] Fuel parameters include surface fuel type, fuel moisture characteristics, surface fuel loading, canopy density, vertical fuel distribution, and canopy gap distribution;

[0097] Meteorological parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure, and light intensity.

[0098] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A forest fire spread simulation method based on UAV LiDAR technology, characterized by: The method comprises: Using drone LiDAR technology, cameras, and sensors, the system collects forest area data, including LiDAR point cloud data, thermal infrared data, image data, and meteorological data. Extracting forest area features based on the forest area data to obtain feature parameters; Based on the characteristic parameters, a fire spread dynamic model is constructed through a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm; Based on the fire spread dynamic model, fire spread simulation and prediction are performed, and dynamic early warning is performed based on the simulation and prediction results to obtain fire fighting and fire prevention decisions; The characteristic parameters include terrain parameters, fuel parameters and meteorological parameters; The terrain parameters include the altitude, slope and aspect of the forest area; The fuel parameters include surface fuel type, fuel moisture characteristics, surface fuel loading, canopy density, vertical fuel distribution, and canopy gap distribution; The meteorological parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure and light intensity; Methods for constructing a dynamic model of fire spread include: Based on terrain parameters, fuel parameters, and meteorological parameters, a deep reinforcement learning model is constructed, which treats the spread of surface fire and canopy fire as a Markov decision process to obtain the dynamic fire line position and intensity. Based on the flame heat radiation transfer equation and air turbulence model, combined with the canopy gap distribution, the flame jump distance and probability are calculated to obtain a coupled model of the vertical spread and horizontal diffusion of the canopy flame. Based on the coupling model, a three-dimensional fire propagation path is simulated; The fire spread dynamic model is constructed based on the dynamic fire line position and intensity and the three-dimensional fire propagation path.

2. The method according to claim 1, characterized in that Methods for obtaining dynamic fire line position and intensity include: discretizing the terrain parameter, the fuel parameter, and the meteorological parameter; The spatial distribution characteristics of discretized terrain parameters and discretized fuel parameters are obtained based on convolutional neural networks, and the temporal characteristics of discretized meteorological parameters are captured based on long short-term memory networks. Based on the spatial distribution characteristics and the temporal characteristics, a state space is constructed; the position and intensity of the line of fire are used as the action space; Using the state space as the input of the deep Q network and the action space as the output of the deep Q network, updating the network weights by minimizing the loss function to train the intelligent agent; The intelligent agent selects the optimal action according to the current state, simulates the Markov decision process, and obtains the dynamic fire line position and intensity.

3. The method according to claim 1, characterized in that Methods for obtaining a coupled model of vertical and horizontal spread of canopy fire include: Using the radiation heat transfer theory, the flame heat radiation transfer equation is established, and based on the flame heat radiation transfer equation, the propagation and attenuation process of flame heat radiation in the forest canopy is obtained; Based on the air turbulence intensity, an air turbulence model is used to simulate the effect of air turbulence on flame spread; Based on the propagation and attenuation process of flame thermal radiation in the forest canopy and the influence of air turbulence on flame spread, combined with the canopy gap distribution, the flame jump distance and probability are calculated to obtain a coupled model of vertical and horizontal spread of forest canopy flames.

4. The method according to claim 1, wherein It also includes integrating LiDAR point cloud data, fire spread simulation and prediction results, and real-time meteorological data to establish a three-dimensional dynamic visualization platform, dynamically display the movement trend of the fire line through thermal maps and particle effects, and divide the fire risk level.

5. A forest fire spread simulation system based on UAV LiDAR technology, used to implement the method according to any one of claims 1 to 4, characterized in that: include: A data acquisition module, configured to collect forest area data using drone LiDAR technology, cameras, and sensors; wherein the forest area data includes LiDAR point cloud data, thermal infrared data, image data, and meteorological data; A feature parameter extraction module is used to extract forest area features based on the forest area data to obtain feature parameters; a fire spread dynamic model construction module, configured to construct a fire spread dynamic model based on the characteristic parameters by using a fire spread dynamics reinforcement learning framework and a vertical-horizontal fire spread coupling algorithm; A fire simulation module is used to simulate and predict the spread of fire based on the fire spread dynamic model, and to issue dynamic warnings based on the simulation and prediction results to obtain fire fighting and fire prevention decisions; The characteristic parameters include terrain parameters, fuel parameters and meteorological parameters; The terrain parameters include the altitude, slope and aspect of the forest area; The fuel parameters include surface fuel type, fuel moisture characteristics, surface fuel loading, canopy density, vertical fuel distribution, and canopy gap distribution; The meteorological parameters include temperature, humidity, wind speed, wind direction, atmospheric pressure and light intensity; Methods for constructing a dynamic model of fire spread include: Based on terrain parameters, fuel parameters, and meteorological parameters, a deep reinforcement learning model is constructed, which treats the spread of surface fire and canopy fire as a Markov decision process to obtain the dynamic fire line position and intensity. Based on the flame heat radiation transfer equation and air turbulence model, combined with the canopy gap distribution, the flame jump distance and probability are calculated to obtain a coupled model of the vertical spread and horizontal diffusion of the canopy flame. Based on the coupling model, a three-dimensional fire propagation path is simulated; The fire spread dynamic model is constructed based on the dynamic fire line position and intensity and the three-dimensional fire propagation path.

Images