A method and system for lightning strike fire spread simulation

Abstract

The application discloses a lightning-caused fire spreading simulation method and system, and the method comprises the steps: acquiring multi-source spatio-temporal characteristic data corresponding to lightning-caused fire; performing feature extraction and fusion on the corresponding multi-source spatio-temporal characteristic data through a pre-constructed spatio-temporal feature fusion network, and outputting prediction parameters for representing fire state conversion; constructing a multi-state cellular automaton including a smoldering state and an open fire state, locating a lightning ignition point, initializing the state of the cell corresponding to the lightning ignition point to the smoldering state, and giving the cell corresponding to the lightning ignition point its initial thermal energy; performing phase change discrimination on the cell in the smoldering state based on the prediction parameters and dynamic time sequence characteristics; performing a neighborhood propagation rule based on vector correction based on the phase change discrimination result to simulate lightning-caused fire spreading, and outputting a simulation result. Through deep coupling of deep learning feature extraction and physical mechanism modeling, the accuracy and scientificity of lightning-caused fire simulation are significantly improved by the scheme.

Description

Technical Field

[0001] This application generally relates to the field of forest fire monitoring and risk early warning technology. More specifically, this application relates to a method and system for simulating the spread of lightning-caused fires. Background Technology

[0002] Lightning-induced fires are a type of natural fire caused by the direct ignition of combustibles by atmospheric discharge. They are widespread in arid, high-altitude, or densely vegetated forest ecosystems. Unlike man-made fires, lightning-induced fires have a significant "latent-sudden" characteristic: after a lightning strike, combustibles often enter a smoldering stage that lasts for hours or even days. During this stage, the ignition source produces little smoke and is extremely difficult to detect by satellites or lookout towers; once the ambient wind speed reaches a critical value or the humidity decreases, the smoldering ignition source will instantly undergo a phase transition and transform into an open flame, leading to an explosive spread of the fire.

[0003] Current technologies for predicting and simulating lightning-induced fires primarily focus on using standard machine learning algorithms to predict fire occurrence or employing physical models such as FARSITE to simulate the spread of open flames. However, these existing technologies still have the following shortcomings when dealing with the unique mechanisms of lightning-induced fires: First, traditional models often simply combine topography and meteorological factors, which fails to effectively capture the trend of cumulative changes in meteorological conditions over time, and also makes it difficult to characterize the complex nonlinear interactions between meteorological factors and topography.

[0004] Secondly, existing cellular automata models typically assume "ignition equals spread," neglecting the unique energy accumulation and sudden turning point processes inherent in lightning-caused fires. This simplification of the physical process makes it impossible for the models to accurately predict the outbreak time and path of delayed-type fires.

[0005] Third, the spread rules are too idealistic. The traditional 8-propagation model is prone to "chessboard effect" in complex wind field environments, making it difficult to truly reproduce the heterogeneous spread pattern of fire driven by wind direction.

[0006] In view of this, there is an urgent need to provide a lightning strike fire spread simulation scheme to solve the above problems and meet the needs of precise prevention and control of lightning strike fires. Summary of the Invention

[0007] In order to at least solve one or more of the technical problems mentioned above, this application proposes a lightning strike fire spread simulation scheme in several aspects.

[0008] In a first aspect, this application provides a method for simulating the spread of lightning-induced fire, comprising: acquiring multi-source spatiotemporal feature data corresponding to a lightning-induced fire, wherein the multi-source spatiotemporal feature data includes static spatial features, dynamic temporal features, and lightning physical features; extracting and fusing features from the static spatial features and the dynamic temporal features through a pre-constructed spatiotemporal feature fusion network, and outputting prediction parameters for characterizing the fire state transition; constructing a multi-state cellular automaton including smoldering and open flame states, locating the lightning ignition point, initializing the state of the cell corresponding to the lightning ignition point to the smoldering state, and assigning initial thermal energy to the cell corresponding to the lightning ignition point; performing phase transition discrimination on the cells in the smoldering state based on the prediction parameters and the dynamic temporal features; executing a neighborhood propagation rule based on vector correction based on the phase transition discrimination result to simulate the spread of lightning-induced fire, and outputting the simulation results.

[0009] In some embodiments, the spatiotemporal feature fusion network includes a spatial feature extraction branch, a temporal feature extraction branch, an attention fusion module, and an output layer. The spatial feature extraction branch is used to extract a spatial hindering feature vector representing the effect of terrain on fire based on the input static spatial features. The temporal feature extraction branch is used to extract a dryness cumulative feature vector representing the change of meteorological factors over time based on the input dynamic temporal features using a bidirectional long short-term memory network. The attention fusion module is used to adaptively weight the feature channels of the spatial feature extraction branch and the feature channels of the temporal feature extraction branch, and concatenate the spatial hindering feature vector and the dryness cumulative feature vector based on the adaptive weighting result to obtain a comprehensive spatiotemporal feature map. The output layer is used to input the comprehensive spatiotemporal feature map into a fully connected mapping layer and output the corresponding prediction parameters.

[0010] In some embodiments, the prediction parameters include a smoldering-to-open flame phase transition threshold and a baseline probability of open flame spread.

[0011] In some embodiments, the multi-state cellular automaton is a four-state cellular automaton, and the states of the four-state cellular automaton are defined as follows: .

[0012] In some embodiments, during the process of performing phase transition discrimination on cells in a smoldering state based on the predicted parameters and the dynamic temporal characteristics, the following steps are performed: calculating the internal thermal energy of the cells in a smoldering state through energy accumulation and dissipation mechanisms, and obtaining the corresponding smoldering duration based on the internal thermal energy; constructing a phase transition discrimination function; calculating the phase transition discrimination value of the cells in a smoldering state through the phase transition discrimination function; determining whether the phase transition discrimination value of the cells in a smoldering state is greater than the smoldering-open flame phase transition threshold and whether the smoldering duration is greater than the threshold threshold. If the phase transition discrimination value of a cell in a smoldering state is greater than the smoldering-open flame phase transition threshold and the smoldering time is less than the maximum tolerance time threshold, the cell in a smoldering state is determined to undergo a phase transition, and its state is converted to an open flame state at the next moment; if the phase transition discrimination value of a cell in a smoldering state is not greater than the smoldering-open flame phase transition threshold or the smoldering time is not less than the maximum tolerance time threshold, the cell in a smoldering state is determined not to undergo a phase transition, and its state is converted to an extinguished state at the next moment.

[0013] In some embodiments, the expression for the energy accumulation and dissipation mechanism is: , Let be the internal thermal energy of the cell at time t. The internal thermal energy of the cell at time t-1 For thermal energy gain, This is the term for heat energy dissipation.

[0014] In some embodiments, the expression for the phase transition discrimination function is: ,in, This is the phase transition discrimination value. For adjustment coefficients, Let be the wind speed at time t. Let t be the humidity at time t.

[0015] In some embodiments, during the execution of a neighborhood propagation rule based on vector correction based on the phase transition discrimination result, the following steps are performed: For a cell undergoing a phase transition, identify its eight neighboring cells within its molar neighborhood; calculate the angle between the current wind speed vector and the propagation direction vector from the cell undergoing the phase transition to its neighboring cells; and construct a heterogeneous propagation weight based on the angle, wherein the formula for calculating the heterogeneous propagation weight is: ,in, The non-homogeneous propagation weight corresponding to the k-th neighbor cell. As the baseline spread probability, This is the wind sensitivity coefficient. For wind speed vectors, The magnitude of the wind speed vector. Let be the angle between the current wind speed vector and the propagation direction vector from the cell undergoing phase change to the k-th neighbor cell. The state of neighbor cells is updated using heterogeneous propagation weights. During this process, the following steps are performed: neighbor cells in the unburned state are designated as candidate neighbor cells; a random judgment value uniformly distributed within the interval [0,1] is generated for each candidate neighbor cell; the heterogeneous propagation weight corresponding to each candidate neighbor cell is determined to be less than its corresponding random judgment value; if the heterogeneous propagation weight of a candidate neighbor cell is less than its corresponding random judgment value, the state of that candidate neighbor cell remains unchanged; if the heterogeneous propagation weight of a candidate neighbor cell is not less than its corresponding random judgment value, the state of that candidate neighbor cell is updated from unburned to smoldering.

[0016] In some embodiments, the simulation results include a latent risk map, a fire evolution sequence, and a variable sensitivity map; the latent risk map marks cells currently in a smoldering state and their estimated time to transition to an open flame state; the fire evolution sequence shows the fire evolution sequence from a smoldering state to an open flame state and from an open flame state to fire line expansion; the variable sensitivity map shows the contribution of the coupling of wind speed and humidity to sudden changes in the fire field.

[0017] In a second aspect, this application provides a lightning-fire spread simulation system, which uses the lightning-fire spread simulation method as described in any embodiment of the first aspect to simulate lightning-fire spread. The system includes: a multi-source data acquisition module for acquiring multi-source spatiotemporal feature data corresponding to lightning-fire, wherein the multi-source spatiotemporal feature data includes static spatial features, dynamic temporal features, and lightning physical features; and a prediction parameter acquisition module for extracting and fusing features from the static spatial features and the dynamic temporal features through a pre-constructed spatiotemporal feature fusion network, and outputting a feature characterizing the fire. The system includes: a prediction parameter for state transitions; a cellular automaton construction module for constructing a multi-state cellular automaton including smoldering and open flame states, locating the lightning ignition point, initializing the state of the cell corresponding to the lightning ignition point to the smoldering state, and assigning initial thermal energy to the cell corresponding to the lightning ignition point; a phase transition discrimination module for performing phase transition discrimination on cells in the smoldering state based on the prediction parameter and the dynamic temporal characteristics; and a propagation simulation module for performing neighborhood propagation rule based on vector correction based on the phase transition discrimination result to simulate the propagation of lightning-induced fire and outputting the simulation results.

[0018] Through the lightning-induced fire spread simulation scheme provided above, this application significantly improves the accuracy and scientific rigor of lightning-induced fire simulation by deeply coupling deep learning feature extraction with physical mechanism modeling. First, by constructing a multi-state cellular automaton and introducing a smoldering energy accumulation mechanism, the unique concealed, latent, and abruptly eruptive physical characteristics of lightning-induced fire are accurately captured, solving the technical challenge of traditional models simulating the transition from latent to open flame. Second, a spatiotemporal feature fusion network is used to transform multi-source dynamic meteorological and static geographical data into key prediction parameters. Compared to traditional physical formulas, this approach can more sensitively perceive the nonlinear influence of environmental factors on the phase transition of the fire field, improving prediction accuracy in complex environments. Finally, a neighborhood propagation rule based on vector correction is adopted, dynamically adjusting the spread weights through the wind direction vector angle. This fundamentally eliminates the chessboard effect commonly found in traditional cellular automaton simulations, making the generated fire evolution pattern more consistent with the diffusion patterns under real strong wind conditions, providing more valuable auxiliary decision support for disaster prevention and mitigation.

[0019] Furthermore, in some embodiments, by defining a four-state cell and a phase transition discrimination mechanism, a refined dynamic simulation of the entire life cycle of lightning-induced fire—from latency to sudden change to extinction—is achieved. By constructing a state space that conforms to the physical nature of lightning-induced fire and clearly distinguishing between smoldering and open flame states, the limitation of traditional binary cells (burning only / unburning) in describing the long-term concealed smoldering characteristics of lightning-induced fire is overcome. By establishing a dual-constraint phase transition discrimination logic, the environmental sudden change threshold predicted by deep learning and the time tolerance limit based on physical mechanisms are comprehensively considered. This mechanism can capture both explosive ignition caused by a sudden increase in wind speed and a sudden drop in humidity, and simulate natural extinction caused by fuel depletion or environmental obstruction, significantly improving the model's discrimination accuracy under critical conditions, thereby effectively reducing the false alarm rate and false negative rate of lightning-induced fire spread trends.

[0020] Furthermore, in some embodiments, the realism and physical accuracy of fire spatial morphology simulation are significantly improved by using heterogeneous weighting of the physical vector field. First, by calculating the cosine of the angle between the wind speed vector and the propagation direction, an exponentially corrected heterogeneous propagation weight is constructed, which greatly increases the probability of propagation in the downwind direction and suppresses it in the upwind direction. This accurately reproduces the natural diffusion morphology of the fire field under strong wind conditions, which is "teardrop-shaped" or "ellipsoidal," on the grid. Second, the sensitivity of environmental response is enhanced by introducing a wind sensitivity coefficient and an exponential function to simulate the nonlinear accelerating effect of wind speed on fire spread. Finally, a macroscopic expression of microscopic randomness is achieved. By comparing the calculated physical weights with random judgment values, the decisive role of the dominant wind direction is ensured, and the random ignition phenomenon caused by uneven fuel distribution in nature is simulated. It is also clarified that the ignited cells enter the "smoldering state" rather than direct open flame, strictly following the physical temporal law of energy transfer in lightning strike fires.

[0021] Furthermore, in some embodiments, simulation results are used to transform single fire prediction data into comprehensive decision support information with high interpretability and practical guidance. Latent risk maps visually locate hidden smoldering points and quantify the estimated time for them to convert into open flames, providing emergency management departments with a valuable window of opportunity for firefighting, thus supporting a tactical shift from passive firefighting to proactive prevention. Fire evolution sequences clearly demonstrate the complete spatiotemporal trajectory of lightning-caused fires from microscopic smoldering to macroscopic spread, helping commanders to fully grasp the development of the fire. Variable sensitivity maps deeply analyze the contribution of wind speed and humidity coupling to sudden changes in the fire, helping decision-makers quickly identify key meteorological drivers that lead to fire deterioration, thereby formulating more targeted and scientific key prevention and control strategies. Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein: Figure 1 An exemplary flowchart of the lightning strike fire spread simulation method according to an embodiment of this application is shown; Figure 2 An exemplary flowchart illustrating phase transition determination performed by a cell in a smoldering state according to an embodiment of this application is shown. Figure 3 An exemplary flowchart illustrating the execution of a vector-corrected neighborhood propagation rule based on a phase transition discrimination result according to an embodiment of this application is shown. Figure 4 An exemplary flowchart illustrating the state update determination of neighboring cells by non-homogeneous propagation weights according to an embodiment of this application is shown. Figure 5 An exemplary structural block diagram of a lightning strike fire spread simulation system according to an embodiment of this application is shown. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be understood that the terms "comprising" and "including" used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0025] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0026] Figure 1 An exemplary flowchart of a lightning strike fire spread simulation method 100 according to an embodiment of this application is shown.

[0027] like Figure 1 As shown, in step S110, multi-source spatiotemporal feature data corresponding to lightning strikes are obtained. The multi-source spatiotemporal feature data includes static spatial features, dynamic temporal features, and lightning physical features.

[0028] In the embodiments of this application, static spatial features include topographic factors (such as slope, aspect, altitude, etc.), land cover (such as vegetation type, canopy density, etc.), and flame retardant factors (such as water bodies, roads, firebreaks, etc.).

[0029] In the embodiments of this application, the dynamic time series features include time series meteorological data of the past 24 hours (e.g., temperature, humidity, wind speed, wind direction, precipitation, etc.), and the dynamic time series features are used to capture the cumulative effect of environmental dryness.

[0030] In the embodiments of this application, the physical characteristics of lightning include lightning polarity, lightning current amplitude (kA), and lightning time.

[0031] In the embodiments of this application, during the acquisition of multi-source spatiotemporal feature data corresponding to lightning strikes, the study area is first divided into regular grid cells with a spatial resolution of 30m×30m. Each grid cell serves as a simulation unit, and a three-dimensional feature dataset including static spatial features, dynamic temporal features, and lightning physical features is collected. Then, the collected data is resampled, projected, and normalized on a GIS platform to form a unified spatiotemporal feature tensor.

[0032] Based on step S110, by integrating static spatial features such as terrain and vegetation with dynamic temporal features such as meteorological changes, the model accurately depicts the spatial distribution of fuel and hindering conditions at the fire site and keenly captures the dynamic trend of environmental dryness accumulating over time, significantly improving the model's ability to perceive complex natural environments. By incorporating lightning physical features such as polarity and current amplitude, the model fills the gap in traditional fire hazard models that neglect the physical properties of ignition sources, providing a core basis for subsequent accurate calculation of ignition energy and location of the ignition point. By unifying heterogeneous data into a spatiotemporal feature tensor of a 30m×30m high-resolution raster, the dimensional differences between multi-source data are effectively eliminated, ensuring the consistency and comparability of input data, and the model's computational efficiency and spatial resolution at a fine scale are greatly improved.

[0033] After completing step S110, in step S120, static spatial features and dynamic temporal features are extracted and fused through a pre-constructed spatiotemporal feature fusion network, and prediction parameters are output to characterize the fire state transformation.

[0034] In the embodiments of this application, during the construction of the spatiotemporal feature fusion network, based on historical lightning-induced fire data, a sliding window method is used to construct positive samples (sequences of fire transition) and negative samples (sequences of long-term smoldering or extinguishment). Focal Loss is used instead of conventional cross-entropy as the loss function to address the extreme imbalance problem of long smoldering periods and few fire transitions in the lightning-induced fire samples. Furthermore, the AdamW algorithm is selected for optimization, with an initial learning rate set to 0.001, and the learning rate is adjusted using a cosine annealing strategy.

[0035] In the embodiments of this application, the spatiotemporal feature fusion network includes a spatial feature extraction branch, a temporal feature extraction branch, an attention fusion module, and an output layer.

[0036] Specifically, the spatial feature extraction branch is used to extract spatially hindering feature vectors representing the effect of terrain on fire based on the static spatial features of the input.

[0037] In the embodiments of this application, the spatial feature extraction branch uses a 3-layer one-dimensional convolutional neural network (CNN) to extract the spatial obstruction feature vector of terrain on fire.

[0038] Specifically, the temporal feature extraction branch is used to extract the aridity cumulative feature vector of meteorological factors changing over time based on the dynamic temporal feature representation of the input using a bidirectional long short-term memory (BiLSTM) network.

[0039] Specifically, the attention fusion module is used to adaptively weight the feature channels of the spatial feature extraction branch and the feature channels of the temporal feature extraction branch, and cascade the spatial retardation feature vector and the dryness cumulative feature vector based on the adaptive weighting result to obtain a comprehensive spatiotemporal feature map.

[0040] In the embodiments of this application, the SE-Block (Squeeze-and-Excitation) mechanism is introduced into the aforementioned attention fusion module to adaptively weight the feature channels of the spatial feature extraction branch and the temporal feature extraction branch. For example, in the time step where the wind speed increases sharply, the network automatically assigns a higher weight to the wind speed channel.

[0041] Specifically, the output layer is used to input the comprehensive spatiotemporal feature map into the fully connected mapping layer and output the corresponding prediction parameters.

[0042] In the embodiments of this application, the prediction parameters include the smoldering-to-open flame phase transition threshold and the open flame spread baseline probability.

[0043] Based on step S120, a dual-stream architecture combining CNN and Bi-LSTM is used to accurately extract the static spatial obstruction features of terrain on fire, and to keenly capture the cumulative effect of meteorological factors (such as aridity) over time, overcoming the lack of historical meteorological memory in traditional methods. By adaptively weighting feature channels through the SE-Block attention mechanism, high-risk factors can be automatically focused at critical time steps such as rapid wind speed increases, effectively solving the problem of nonlinear coupling of multi-source heterogeneous data. For the data characteristics of long smoldering times and few open-flame conversions in lightning-induced fires, Focal Loss is used instead of the conventional loss function, significantly improving the model's sensitivity to sparse but deadly smoldering-to-open-flame events, ensuring that the output phase transition threshold and baseline probability have extremely high practical confidence.

[0044] After completing step S120, in step S130, a multi-state cellular automaton including smoldering state and open flame state is constructed, the lightning ignition point is located, the state of the cell corresponding to the lightning ignition point is initialized to smoldering state, and the cell corresponding to the lightning ignition point is given initial thermal energy.

[0045] In embodiments of this application, a global lightning location network or weather radar acquires the coordinates of the lightning strike point, thereby locating the lightning ignition point.

[0046] In the embodiments of this application, the cell corresponding to the lightning ignition point is given initial thermal energy based on the amplitude of the lightning current.

[0047] In the embodiments of this application, the multi-state cellular automaton adopts a four-state cellular automaton, and the state definition of the four-state cellular automaton is as follows: .

[0048] Based on step S130, a four-state cellular automaton comprising "unignited - smoldering - open flame - extinguished" was constructed, accurately capturing the unique "latent smoldering" stage of lightning-induced fires. This not only recreates the real physical process of a fire from concealment to outbreak but also provides independent state identifiers for early warning. Unlike traditional models that rely solely on probabilistic random ignition, this model uses global lightning location data to precisely locate the ignition point and directly assigns initial thermal energy based on the lightning current amplitude, establishing a quantitative mapping relationship between "lightning energy and initial fire intensity." This significantly improves the physical realism and scientific rigor of the simulation's initial stage.

[0049] After step S130 is completed, in step S140, phase transition discrimination is performed on the cells in the smoldering state based on the predicted parameters and dynamic time series characteristics.

[0050] The specific process involved in step S140 in the embodiments of this application can be found in [reference needed]. Figure 2 .

[0051] Figure 2 An exemplary flowchart illustrating phase transition determination performed by a cell in a smoldering state according to an embodiment of this application is shown.

[0052] like Figure 2 As shown, in step S210, the internal thermal energy of the cell in the smoldering state is calculated through an energy accumulation and dissipation mechanism, and the corresponding smoldering duration is obtained based on the internal thermal energy. In step S210, a phase transition discrimination function is constructed. In step S230, the phase transition discrimination value of the cell in the smoldering state is calculated using the phase transition discrimination function. In step S240, it is determined whether the phase transition discrimination value of the cell in the smoldering state is greater than the smoldering-open flame phase transition threshold and whether the smoldering duration is less than the maximum tolerable duration threshold. In response to the cell in the smoldering state having a phase transition value greater than the smoldering-open flame phase transition threshold and a smoldering duration less than the maximum tolerable duration threshold, in step S250, it is determined that the cell in the smoldering state has undergone a phase transition, and its state is converted to the open flame state at the next moment. In response to the fact that the phase transition discrimination value of a cell in the smoldering state is not greater than the smoldering-open flame phase transition threshold or the smoldering duration is not less than the maximum tolerance duration threshold, in step S260, it is determined that the cell in the smoldering state does not undergo a phase transition, and its state is converted to the extinguished state at the next moment.

[0053] In the embodiments of this application, the expression for the energy accumulation and dissipation mechanism is as follows: , Let be the internal thermal energy of the cell at time t. The internal thermal energy of the cell at time t-1 This is a thermal energy gain term, which is positively correlated with wind speed. This is a heat energy dissipation term, which is positively correlated with humidity.

[0054] In the embodiments of this application, the expression for the phase transition discrimination function is: ,in, This is the phase transition discrimination value. For adjustment coefficients, Let be the wind speed at time t. Let t be the humidity at time t.

[0055] Based on step S140, the changes in internal thermal energy of the cells are dynamically tracked through an energy accumulation and dissipation model, realistically reproducing the physical accumulation process of the fire source during the smoldering stage, thus avoiding the random bias of purely probabilistic models. A dynamic criterion coupled with meteorological conditions is constructed, using the ratio of wind speed to humidity to build a discriminant function, quantifying the nonlinear driving effect of external environmental dryness and wind-assisted combustion on the qualitative change of fire intensity. The logic of fire generation and destruction is improved. By introducing dual constraints of the maximum tolerance time threshold and the phase transition threshold, it can not only accurately predict the moment when smoldering erupts into open flame, but also identify spontaneous extinction phenomena caused by energy depletion or environmental humidity, effectively eliminating false alarms of "infinite smoldering," and significantly improving the system's practical fault tolerance and prediction accuracy.

[0056] After step S140 is completed, in step S150, a neighborhood propagation rule based on vector correction is executed based on the phase transition discrimination result to simulate the spread of lightning-induced fire and output the simulation results.

[0057] In the embodiments of this application, the specific process of executing the neighborhood propagation rule based on vector correction based on the phase transition discrimination result can be found in [reference needed]. Figure 3 .

[0058] Figure 3 An exemplary flowchart illustrating the execution of a vector-corrected neighborhood propagation rule based on the phase transition discrimination result according to an embodiment of this application is shown.

[0059] like Figure 3 As shown, in step S310, the eight neighboring cells within the molar neighborhood of the cell undergoing phase transition are identified. In step S320, the angle between the current wind speed vector and the propagation direction vector from the cell undergoing phase transition to the neighboring cells is calculated. In step S330, a heterogeneous propagation weight is constructed based on the angle. In step S340, the state update of the neighboring cells is determined using the heterogeneous propagation weight.

[0060] In the embodiments of this application, the formula for calculating the non-homogeneous propagation weight is as follows: ,in, The non-homogeneous propagation weight corresponding to the k-th neighbor cell. As the baseline spread probability, This is the wind sensitivity coefficient. For wind speed vectors, The magnitude of the wind speed vector. It is the angle between the wind speed vector at the current moment and the propagation direction vector from the cell where the phase transition occurred to the k-th neighbor cell.

[0061] In the embodiments of this application, the specific process involved in step S340 can be found in [reference needed]. Figure 4 .

[0062] Figure 4 An exemplary flowchart illustrating how the state update determination of neighboring cells is performed by non-homogeneous propagation weights according to an embodiment of this application is shown.

[0063] like Figure 4 As shown, in step S410, neighboring cells in the unburned state are selected as candidate neighboring cells. In step S420, a random judgment value uniformly distributed in the interval [0,1] is generated for each candidate neighboring cell. In step S430, it is determined whether the heterogeneous propagation weight corresponding to each candidate neighboring cell is less than its corresponding random judgment value. In response to the heterogeneous propagation weight corresponding to a candidate neighboring cell being less than its corresponding random judgment value, in step S440, it is determined that the state of the candidate neighboring cell remains unchanged. In response to the heterogeneous propagation weight corresponding to a candidate neighboring cell being not less than its corresponding random judgment value, in step S450, it is determined that the state of the candidate neighboring cell is updated from the unburned state to the smoldering state.

[0064] By implementing a neighborhood propagation rule based on vector correction using phase transition discrimination results, a highly realistic anisotropic fire spread model was constructed, effectively solving the distortion problem of traditional cellular automata simulations tending towards regular geometric shapes (such as squares or circles). Non-homogeneous propagation weights were constructed by calculating the angle between the wind speed vector and the propagation direction vector, accurately simulating the directional driving effect of wind on fire, achieving non-linear anisotropic spread with acceleration in the downwind direction and obstruction in the upwind direction, making the simulated fire shape more consistent with real elliptical or irregular morphologies. Using a weighting formula incorporating wind sensitivity coefficients and cosine functions, macroscopic meteorological data was mapped in real time to differentiated propagation probabilities between microscopic cells, enhancing the model's immediate response to environmental disturbances such as sudden wind direction changes. By comparing the propagation weights with random [0,1] judgment values ​​to determine state updates (from unburned to smoldering), the dominance of physical laws was preserved while simulating the randomness and uncertainty caused by fuel discontinuity or stray sparks in nature, significantly improving the natural realism of the spread simulation.

[0065] In the embodiments of this application, the simulation results include latent risk maps, fire evolution sequences, and variable sensitivity maps.

[0066] Specifically, the latent risk map marks cells currently in a smoldering state and the estimated time for them to transition to an open flame state.

[0067] Specifically, the fire evolution sequence shows the entire process of fire evolution from smoldering to open flame and from open flame to fire line expansion.

[0068] Specifically, the variable sensitivity plot shows the contribution of the coupling of wind speed and humidity to sudden changes in the fire situation.

[0069] In the embodiments of this application, the latent risk map, fire evolution sequence, and variable sensitivity map are stored in GeoTIFF and Shapefile formats and can be imported into a GIS system for visualization and emergency dispatch analysis.

[0070] In the process of outputting simulation results, a latent risk map is generated to visualize and mark the difficult-to-detect underground smoldering state and the expected outbreak time, moving the fire prevention and extinguishing focus forward and buying valuable time for firefighting. A complete recap of the entire process from smoldering to open flame to fire line expansion is provided through the fire evolution sequence, helping commanders to intuitively grasp the development trajectory and future spread trend of the fire. The coupling contribution of wind speed and humidity is quantified through variable sensitivity maps, revealing the intrinsic driving mechanism of sudden fire changes and assisting in the formulation of more targeted containment strategies. Data is stored using common geographic information formats such as GeoTIFF and Shapefile, enabling seamless integration with existing GIS command and dispatch systems and achieving rapid transformation from algorithmic models to emergency response.

[0071] In summary, the embodiments of this application significantly improve the accuracy and scientific rigor of lightning-induced fire simulation through the deep coupling of deep learning feature extraction and physical mechanism modeling. First, by constructing a multi-state cellular automaton and introducing a smoldering energy accumulation mechanism, the unique concealment, latency, and sudden eruption physical characteristics of lightning-induced fire are accurately captured, solving the technical challenge of traditional models simulating the transition from latent to open flame. Second, a spatiotemporal feature fusion network is used to transform multi-source dynamic meteorological and static geographical data into key prediction parameters. Compared to traditional physical formulas, this approach can more sensitively perceive the nonlinear influence of environmental factors on the phase transition of the fire field, improving prediction accuracy in complex environments. Finally, a neighborhood propagation rule based on vector correction is adopted, dynamically adjusting the spread weights through the wind direction vector angle. This fundamentally eliminates the chessboard effect commonly found in traditional cellular automaton simulations, making the generated fire evolution pattern more consistent with the diffusion patterns under real strong wind conditions, providing more valuable auxiliary decision support for disaster prevention and mitigation.

[0072] Furthermore, in some embodiments, by defining a four-state cell and a phase transition discrimination mechanism, a refined dynamic simulation of the entire life cycle of lightning-induced fire—from latency to sudden change to extinction—is achieved. By constructing a state space that conforms to the physical nature of lightning-induced fire and clearly distinguishing between smoldering and open flame states, the limitation of traditional binary cells (burning only / unburning) in describing the long-term concealed smoldering characteristics of lightning-induced fire is overcome. By establishing a dual-constraint phase transition discrimination logic, the environmental sudden change threshold predicted by deep learning and the time tolerance limit based on physical mechanisms are comprehensively considered. This mechanism can capture both explosive ignition caused by a sudden increase in wind speed and a sudden drop in humidity, and simulate natural extinction caused by fuel depletion or environmental obstruction, significantly improving the model's discrimination accuracy under critical conditions, thereby effectively reducing the false alarm rate and false negative rate of lightning-induced fire spread trends.

[0073] Furthermore, in some embodiments, the realism and physical accuracy of fire spatial morphology simulation are significantly improved by using heterogeneous weighting of the physical vector field. First, by calculating the cosine of the angle between the wind speed vector and the propagation direction, an exponentially corrected heterogeneous propagation weight is constructed, which greatly increases the probability of propagation in the downwind direction and suppresses it in the upwind direction. This accurately reproduces the natural diffusion morphology of the fire field under strong wind conditions, which is "teardrop-shaped" or "ellipsoidal," on the grid. Second, the sensitivity of environmental response is enhanced by introducing a wind sensitivity coefficient and an exponential function to simulate the nonlinear accelerating effect of wind speed on fire spread. Finally, a macroscopic expression of microscopic randomness is achieved. By comparing the calculated physical weights with random judgment values, the decisive role of the dominant wind direction is ensured, and the random ignition phenomenon caused by uneven fuel distribution in nature is simulated. It is also clarified that the ignited cells enter the "smoldering state" rather than direct open flame, strictly following the physical temporal law of energy transfer in lightning strike fires.

[0074] Furthermore, in some embodiments, simulation results are used to transform single fire prediction data into comprehensive decision support information with high interpretability and practical guidance. Latent risk maps visually locate hidden smoldering points and quantify the estimated time for them to convert into open flames, providing emergency management departments with a valuable window of opportunity for firefighting, thus supporting a tactical shift from passive firefighting to proactive prevention. Fire evolution sequences clearly demonstrate the complete spatiotemporal trajectory of lightning-caused fires from microscopic smoldering to macroscopic spread, helping commanders to fully grasp the development of the fire. Variable sensitivity maps deeply analyze the contribution of wind speed and humidity coupling to sudden changes in the fire, helping decision-makers quickly identify key meteorological drivers that lead to fire deterioration, thereby formulating more targeted and scientific key prevention and control strategies.

[0075] This application also provides a lightning strike fire spread simulation system, which can use the aforementioned lightning strike fire spread simulation method 100 to simulate lightning strike fire spread, or other methods can be used to simulate lightning strike fire spread, and this application does not limit it here.

[0076] Figure 5 An exemplary structural block diagram of a lightning strike fire spread simulation system according to an embodiment of this application is shown.

[0077] like Figure 5 As shown, the system 500 includes a multi-source data acquisition module 510, a prediction parameter acquisition module 520, a cellular automata construction module 530, a phase transition discrimination module 540, and a propagation simulation module 550.

[0078] Specifically, the multi-source data acquisition module 510 is used to acquire multi-source spatiotemporal feature data corresponding to lightning strikes. The multi-source spatiotemporal feature data includes static spatial features, dynamic temporal features, and lightning physical features.

[0079] Specifically, the prediction parameter acquisition module 520 is used to extract and fuse static spatial features and dynamic temporal features through a pre-built spatiotemporal feature fusion network, and output prediction parameters to characterize the fire state transformation.

[0080] Specifically, the cellular automaton construction module 530 is used to construct a multi-state cellular automaton including smoldering and open flame states, locate the lightning ignition point, initialize the state of the cell corresponding to the lightning ignition point to the smoldering state, and assign initial thermal energy to the cell corresponding to the lightning ignition point.

[0081] Specifically, the phase transition discrimination module 540 is used to perform phase transition discrimination on cells in the smoldering state based on predicted parameters and dynamic temporal characteristics.

[0082] Specifically, the propagation simulation module 550 is used to execute a neighborhood propagation rule based on vector correction based on the phase transition discrimination result to simulate the propagation of lightning-induced fire and output the simulation results.

[0083] When system 500 uses the aforementioned lightning-fire propagation simulation method 100 to simulate lightning-fire propagation, the multi-source data acquisition module 510 executes the aforementioned step S110, the prediction parameter acquisition module 520 executes the aforementioned step S120, the cellular automata construction module 530 executes the aforementioned step S130, the phase transition discrimination module 540 executes the aforementioned step S140, and the propagation simulation module 550 executes the aforementioned step S150. The specific execution process can be found above and will not be repeated here.

[0084] While numerous embodiments of this application have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will arise for those skilled in the art without departing from the spirit and intent of this application. It should be understood that various alternatives to the embodiments of this application described herein may be employed in the practice of this application. The appended claims are intended to define the scope of protection of this application and therefore cover equivalents or alternatives within the scope of these claims.

Claims

1. A method for simulating the spread of fire caused by lightning strikes, characterized in that, include: Acquire multi-source spatiotemporal feature data corresponding to lightning strikes, wherein the multi-source spatiotemporal feature data includes static spatial features, dynamic temporal features, and lightning physical features; The static spatial features and the dynamic temporal features are extracted and fused by a pre-constructed spatiotemporal feature fusion network, and prediction parameters are output to characterize the fire state transformation. Construct a multi-state cellular automaton that includes smoldering and open flame states, locate the lightning ignition point, initialize the state of the cell corresponding to the lightning ignition point to the smoldering state, and assign initial thermal energy to the cell corresponding to the lightning ignition point. Phase transition discrimination is performed on cells in the smoldering state based on the predicted parameters and the dynamic temporal characteristics. Based on the phase transition discrimination results, a neighborhood propagation rule based on vector correction is executed to simulate the spread of lightning-induced fire, and the simulation results are output.

2. The lightning-fire spread simulation method according to claim 1, characterized in that, The spatiotemporal feature fusion network includes a spatial feature extraction branch, a temporal feature extraction branch, an attention fusion module, and an output layer; The spatial feature extraction branch is used to extract spatial hindering feature vectors representing the effect of terrain on fire based on the input static spatial features. The time feature extraction branch is used to extract the aridity cumulative feature vector of meteorological factors changing over time based on the dynamic temporal feature representation of the input using a bidirectional long short-term memory network; The attention fusion module is used to adaptively weight the feature channels of the spatial feature extraction branch and the feature channels of the temporal feature extraction branch, and to concatenate the spatial retardation feature vector and the dryness accumulation feature vector based on the adaptive weighting result to obtain a comprehensive spatiotemporal feature map. The output layer is used to input the comprehensive spatiotemporal feature map into the fully connected mapping layer and output the corresponding prediction parameters.

3. The lightning-fire spread simulation method according to claim 1 or 2, characterized in that, The prediction parameters include the smoldering-to-open flame phase transition threshold and the baseline probability of open flame spread.

4. The lightning strike fire spread simulation method according to claim 3, characterized in that, The multi-state cellular automaton employs a four-state cellular automaton, and the state definition of the four-state cellular automaton is as follows: .

5. The lightning-fire spread simulation method according to claim 4, characterized in that, In the process of performing phase transition discrimination on cells in the smoldering state based on the predicted parameters and the dynamic temporal characteristics, the following steps are performed: The internal thermal energy of a cell in a smoldering state is calculated using energy accumulation and dissipation mechanisms, and the corresponding smoldering duration is obtained based on the internal thermal energy. Construct a phase transition discrimination function; The phase transition discrimination value of a cell in the smoldering state is calculated using the phase transition discrimination function; Determine whether the phase transition discrimination value of a cell in a smoldering state is greater than the smoldering-open flame phase transition threshold and whether the smoldering duration is less than the maximum tolerance duration threshold. In response to a cell in a smoldering state having a phase transition discrimination value greater than the smoldering-open flame phase transition threshold and a smoldering duration less than the maximum tolerance duration threshold, the cell in a smoldering state is determined to undergo a phase transition, and its state is converted to an open flame state at the next moment. If the phase transition discrimination value of a cell in a smoldering state is not greater than the smoldering-open flame phase transition threshold or the smoldering duration is not less than the maximum tolerance duration threshold, the cell in a smoldering state is determined not to undergo a phase transition, and its state is converted to an extinguished state at the next moment.

6. The lightning strike fire spread simulation method according to claim 5, characterized in that, The expression for the energy accumulation and dissipation mechanism is as follows: , Let be the internal thermal energy of the cell at time t. The internal thermal energy of the cell at time t-1 For thermal energy gain, This is the term for heat energy dissipation.

7. The lightning-fire spread simulation method according to claim 5, characterized in that, The expression for the phase transition discriminant function is: ,in, This is the phase transition discrimination value. For adjustment coefficients, Let be the wind speed at time t. Let t be the humidity at time t.

8. The lightning-fire spread simulation method according to claim 5, characterized in that, In the process of executing the neighborhood propagation rule based on vector correction based on the phase transition discrimination result, the following steps are performed: For a cell undergoing a phase transition, identify its eight neighboring cells within its molar neighborhood; Calculate the angle between the current wind speed vector and the propagation direction vector from the cell where the phase transition occurred to the neighboring cell; The heterogeneous propagation weights are constructed based on the included angle, and the calculation formula for the heterogeneous propagation weights is as follows: ,in, The non-homogeneous propagation weight corresponding to the k-th neighbor cell. As the baseline spread probability, This is the wind sensitivity coefficient. For wind speed vectors, The magnitude of the wind speed vector. The angle between the wind speed vector at the current moment and the propagation direction vector from the cell where the phase transition occurred to the k-th neighbor cell; The state update of neighboring cells is determined by non-homogeneous propagation weights. In the process of determining the state update of neighboring cells through heterogeneous propagation weights, the following steps are performed: Select neighboring cells that are not in a burning state as candidate neighboring cells; For each candidate neighbor cell, generate a random decision value that is uniformly distributed in the interval [0,1]. Determine whether the heterogeneous propagation weight corresponding to each candidate neighbor cell is less than its corresponding random determination value; In response to the non-homogeneous propagation weight of a candidate neighbor cell being less than its corresponding random determination value, the state of the candidate neighbor cell is determined to remain unchanged. In response to the heterogeneous propagation weight corresponding to the candidate neighbor cell being no less than its corresponding random determination value, the state of the candidate neighbor cell is determined to be updated from the unburned state to the smoldering state.

9. The lightning strike fire spread simulation method according to claim 5, characterized in that, The simulation results include a latent risk map, a fire evolution sequence, and a variable sensitivity map. The latent risk map marks cells currently in a smoldering state and the estimated time for them to transition to an open flame state; The fire evolution sequence shows the entire process of fire evolution from smoldering to open flame and from open flame to fire line expansion. The variable sensitivity plot shows the contribution of the coupling of wind speed and humidity to sudden changes in the fire situation.

10. A lightning strike fire spread simulation system, characterized in that, The lightning strike fire spread simulation is performed using the lightning strike fire spread simulation method as described in any one of claims 1-9, wherein the system comprises: A multi-source data acquisition module is used to acquire multi-source spatiotemporal feature data corresponding to lightning strikes, wherein the multi-source spatiotemporal feature data includes static spatial features, dynamic temporal features, and lightning physical features. The prediction parameter acquisition module is used to extract and fuse the static spatial features and the dynamic temporal features through a pre-constructed spatiotemporal feature fusion network, and output prediction parameters to characterize the fire state transformation. The cellular automaton construction module is used to construct a multi-state cellular automaton including smoldering and open flame states, locate the lightning ignition point, initialize the state of the cell corresponding to the lightning ignition point to the smoldering state, and assign initial thermal energy to the cell corresponding to the lightning ignition point. A phase transition discrimination module is used to perform phase transition discrimination on cells in a smoldering state based on the predicted parameters and the dynamic temporal characteristics; The propagation simulation module is used to execute a neighborhood propagation rule based on vector correction based on the phase transition discrimination result to simulate the propagation of lightning-induced fire and output the simulation results.

Images