Fire spread prediction method and device for power transmission line

Abstract

The application provides a fire spreading prediction method and device for a power transmission line, and relates to the technical field of computer processing.The method comprises the following steps: collecting combustible branch data and meteorological branch data; inputting the combustible branch data and the meteorological branch data into a fire burning probability model, and extracting combustible features and meteorological features which are associated with fire burning; wherein the fire burning probability model adopts a least squares support vector machine to classify data, analyzes conversion values corresponding to burning state records in a preset range of the power transmission line, and determines data which is associated with the fire burning state according to the conversion values; and integrating the combustible features and the meteorological features into a three-dimensional cellular automaton, simulating a fire spreading process based on a starting position and a starting time of the fire, and determining a fire spreading prediction result.

Description

Technical Field

[0001] This invention relates to the field of computer processing technology, and more particularly to a method and apparatus for predicting the spread of fire in power transmission lines. Background Technology

[0002] This section is intended to provide background or context for embodiments of the present invention. The description herein is not intended to imply that it is prior art simply because it is included in this section.

[0003] Currently, research on forest fire behavior lacks an effective, rational, and systematic approach to wildfire and forest fire management. In particular, theoretical modeling research is largely imitative, lacking a systematic study of the spread mechanisms and modeling methods of forest fires. The most widely used theoretical model is the Wang Zhengfei model, which, through extensive spot burning experiments and regression analysis of fire experimental data, established a model for calculating forest fire spread rate. Mao Xianmin, Wen Guangyu, and Wang Haihui, among others, have made certain modifications to the Wang Zhengfei model. This type of theoretical model is often based on statistical models, using optimization methods derived from numerous experiments to fit the relationship between environmental parameters, combustible material parameters, and forest fire spread rate. While such models possess a certain degree of theoretical persuasiveness and often achieve relatively good predictive results in local environments, they have significant limitations when applied to macroscopic, conventional, or environmentally complex conditions with significant topographical changes, as they cannot describe the changes in the fire field during specific seasons.

[0004] In summary, there is an urgent need for a method that can overcome the shortcomings of existing technologies and effectively and reasonably predict the spread of fires along power transmission lines. Based on spatial propagation process modeling, this method combines deep learning with existing forest fire spread prediction technologies to improve the accuracy of forest fire spread speed and direction, providing technical support for forest fire control and early warning. Summary of the Invention

[0005] To address the problems existing in current technologies, this invention proposes a method and device for predicting the spread of wildfires along power transmission lines. This invention overcomes the shortcomings of existing wildfire spread prediction methods in terms of prediction efficiency, accuracy, applicability, and data quality. Based on spatial propagation process modeling, this invention achieves wildfire spread prediction along power transmission lines. It uses deep learning to extract geographical and meteorological features, forming a wildfire spread prediction model. The extracted features are then used in a method for simulating wildfire spread based on a three-dimensional surface cellular automata model, resulting in more accurate wildfire spread prediction.

[0006] In a first aspect of the present invention, a method for predicting the spread of fire in a power transmission line is proposed, the method comprising:

[0007] Collect data on combustible materials and meteorological data;

[0008] The combustible material branch data and meteorological branch data are input into the fire combustion probability model to extract combustible material features and meteorological features that are associated with fire combustion. The fire combustion probability model uses a least squares support vector machine to classify the data, analyzes the conversion values ​​corresponding to the combustion state records within a preset range of the transmission line, and determines the data associated with the fire combustion state based on the conversion values.

[0009] By integrating combustible material characteristics and meteorological characteristics into a three-dimensional cellular automaton, the fire spread process is simulated based on the fire's starting location and time, and the fire spread prediction results are determined.

[0010] In a second aspect of the present invention, a fire spread prediction device for power transmission lines is provided, the device comprising:

[0011] The data acquisition module is used to collect combustible material branch data and meteorological branch data;

[0012] The data feature extraction module is used to input the combustible branch data and meteorological branch data into the fire combustion probability model to extract combustible features and meteorological features that are related to fire combustion; wherein, the fire combustion probability model uses a least squares support vector machine to classify the data, analyzes the conversion values ​​corresponding to the combustion state records within a preset range of the transmission line, and determines the data that are related to the fire combustion state based on the conversion values;

[0013] The fire spread prediction module integrates combustible material characteristics and meteorological characteristics into a three-dimensional cellular automata, simulates the fire spread process based on the fire's starting location and start time, and determines the fire spread prediction results.

[0014] In a third aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for predicting the spread of fire in a power transmission line.

[0015] In a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for predicting the spread of fire in a power transmission line.

[0016] In a fifth aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements a method for predicting the spread of fire in power transmission lines.

[0017] This invention proposes a method and device for predicting the spread of fires along power transmission lines. Based on spatial propagation process modeling, it deeply integrates spatial propagation modeling theory and deep learning technology, providing a novel technical solution for fire prevention in power facilities. At the technical implementation level, this invention first constructs a three-dimensional curved surface cellular automata model as the basic framework, which can accurately simulate the fire propagation process under complex terrain conditions. Based on this, a deep neural network architecture is innovatively introduced, constructing a core database, a high-precision micro-topographic feature database, a normalized vegetation database, and a multi-dimensional meteorological database. The overall solution employs an advanced CBR feature extraction module, using a multi-layer convolutional neural network to perform deep feature learning on the aforementioned multi-source heterogeneous data. Simultaneously, this invention designs a multi-modal data fusion algorithm based on least squares support vector machines, which can effectively integrate the influence weights of various climatic conditions to achieve accurate quantitative calculation of the fire probability. Regarding model optimization, the system establishes a dynamic tuning mechanism, continuously iteratively optimizing model parameters by real-time access to satellite remote sensing, meteorological monitoring, and field observation data. This invention significantly improves the timeliness and accuracy of wildfire early warning for power transmission lines, providing strong technical support for ensuring the safe and stable operation of energy systems. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart of a fire spread prediction method for power transmission lines according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the steps for constructing an elevation sample database according to an embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the specific process of analyzing fire spread prediction results according to an embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the steps for optimizing and adjusting model parameters according to an embodiment of the present invention.

[0023] Figure 5 This is a schematic diagram of the overall data branch relationship of a specific embodiment of the present invention.

[0024] Figure 6 This is a schematic diagram illustrating the steps of data acquisition, processing, and fire prediction according to a specific embodiment of the present invention.

[0025] Figure 7This is a schematic diagram of the process for establishing a tree species training sample library according to a specific embodiment of the present invention.

[0026] Figure 8 This is a schematic diagram of the fire spread prediction device for power transmission lines according to an embodiment of the present invention.

[0027] Figure 9 This is a schematic diagram of a computer device structure according to an embodiment of the present invention. Detailed Implementation

[0028] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are given merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make this disclosure more thorough and complete, and to fully convey the scope of this disclosure to those skilled in the art.

[0029] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0030] According to an embodiment of the present invention, a method and apparatus for predicting the spread of fire in power transmission lines are proposed, relating to the field of computer processing technology.

[0031] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0032] Figure 1 This is a schematic flowchart of a fire spread prediction method for power transmission lines according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0033] S101, collects combustible material branch data and meteorological branch data;

[0034] S102, the combustible branch data and meteorological branch data are input into the fire combustion probability model to extract combustible features and meteorological features that are related to fire combustion; wherein, the fire combustion probability model uses a least squares support vector machine to classify the data, analyzes the conversion values ​​corresponding to the combustion state records within a preset range of the transmission line, and determines the data that are related to the fire combustion state based on the conversion values.

[0035] S103 integrates combustible material characteristics and meteorological characteristics into a three-dimensional cellular automaton, simulates the fire spread process based on the fire's starting location and starting time, and determines the fire spread prediction results.

[0036] The fire spread prediction method for power transmission lines proposed in this invention is based on spatial propagation process modeling and deeply integrates spatial propagation modeling theory with deep learning technology, providing a novel technical solution for fire prevention in power facilities. At the technical implementation level, this invention first constructs a three-dimensional curved surface cellular automata model as the basic framework, which can accurately simulate the fire propagation process under complex terrain conditions. Based on this, a deep neural network architecture is innovatively introduced to construct three core databases: 1. A high-precision micro-topographic feature database, containing three-dimensional terrain parameters such as slope, aspect, and altitude; 2. A normalized vegetation database, integrating key indicators such as NDVI, vegetation type, and combustible material humidity; 3. A multi-dimensional meteorological database, covering meteorological elements such as wind speed and direction, temperature, humidity, and precipitation.

[0037] The overall solution employs an advanced CBR (convolutional + batch normalization + ReLU) feature extraction module, using a multi-layer convolutional neural network to perform deep feature learning on the aforementioned multi-source heterogeneous data. In particular, this invention designs a multi-modal data fusion algorithm based on least squares support vector machine (LSSVM), which can effectively integrate the influence weights of various climatic conditions to achieve accurate quantitative calculation of fire probability.

[0038] Regarding model optimization, the system establishes a dynamic tuning mechanism, continuously iteratively optimizing model parameters by real-time access to satellite remote sensing, meteorological monitoring, and field observation data. Experimental results show that after 200 iterations of training, the predicted results highly match the actual wildfire spread, with a significant improvement in the accuracy of predicting spread speed and direction. This invention significantly improves the timeliness and accuracy of wildfire early warning for power transmission lines, providing strong technical support for ensuring the safe and stable operation of energy systems.

[0039] To provide a clearer explanation of the above-mentioned method for predicting the spread of fires on power transmission lines, a detailed explanation is provided below with reference to specific embodiments.

[0040] In one embodiment, to ensure the accuracy of the fire combustion probability model (least squares support vector machine) in classifying data, the model is trained by constructing elevation sample data. Specifically, refer to... Figure 2 The method includes:

[0041] S201, Construct an elevation sample database; wherein, the elevation sample database includes: a combustible material sample database and a meteorological sample database; the combustible material sample database is constructed by using the TensorFlow framework to pre-train a model on the ImageNet dataset and identifying tree species images through drones; the meteorological sample database is constructed by obtaining local meteorological data from the meteorological bureau and combining it with real-time wind speed and wind direction data collected by drones.

[0042] Furthermore, the sample data needs to be preprocessed to optimize data quality. Specific methods include:

[0043] S202, preprocess the data in the elevation sample database; including checking whether there are missing values, outliers or duplicate records in the dataset of the elevation sample database, filling or deleting missing values, and deleting outliers and duplicate records.

[0044] The training method for the fire combustion probability model includes:

[0045] A least squares support vector machine is trained using the elevation sample database, and then evaluated and optimized to obtain a trained fire combustion probability model. Specifically, for fire spread, the model analyzes whether the current unit state in the fire combustion probability model transitions to a burning state. If so, the current unit is a fire point in the next moment and is recorded as 1; otherwise, the current unit is not a fire point in the next moment and is recorded as 0. The influence of the current unit is recorded as a sample space, and the fire spread situation is determined through data classification and transformation.

[0046] In one embodiment, reference Figure 3 This is a schematic diagram illustrating the specific process of analyzing fire spread prediction results according to an embodiment of the present invention. Figure 3 As shown, the specific process of integrating combustible material characteristics and meteorological characteristics into a three-dimensional cellular automata (CA model), simulating the fire spread process based on the fire's initiation location and initiation time, and determining the fire spread prediction results is as follows:

[0047] S301, integrate combustible material characteristics and meteorological characteristics into a three-dimensional cellular automaton, calculate the probability that each adjacent grid cell in the three-dimensional cellular automaton is affected by fire, and integrate the probability of fire impact into the three-dimensional cellular automaton.

[0048] S302: Set the starting location and starting time of the fire, simulate the fire spread process using a three-dimensional cellular automata, and determine the fire spread prediction results.

[0049] In one embodiment, in the three-dimensional cellular automaton, the combustion state includes at least: unburned, just ignited, intensely burning, beginning to extinguish, and already burned;

[0050] Among them, the unburned becomes just ignited after the ignition probability exceeds the preset probability value;

[0051] It is ignited and then, after a first preset time, it transitions to intense combustion.

[0052] Intense combustion continues when the characteristics of the combustible material reach the first preset condition (tree species factor is greater than normal);

[0053] The intense combustion transitions to extinguishing when the characteristics of the combustible material reach the second preset condition (tree species factor equals normal);

[0054] After the initial extinguishing time has elapsed for the second preset period, it will be considered as having already burned.

[0055] In one embodiment, for the CA model, the model parameters can be further optimized and adjusted by analyzing the prediction results, referring to... Figure 4 The specific method is as follows:

[0056] S401, determine the starting location and starting time of the fire based on the actual fire data, and input the fire status into the three-dimensional cellular automaton as the starting condition for initializing the model;

[0057] S402, by running the three-dimensional cellular automaton, the process of a fire spreading from its starting position is simulated, and the state of each cell is updated;

[0058] S403, Analyze the fire spread prediction results output by the three-dimensional cellular automata and compare them with actual fire data, and optimize and adjust the three-dimensional cellular automata model based on the comparison results.

[0059] Compared with the prior art, the present invention makes improvements in at least the following aspects:

[0060] 1. It achieves an organic combination of cellular automata and deep learning:

[0061] This invention proposes a method for predicting the spread of wildfires along power transmission lines based on spatial propagation process modeling. It innovatively transforms the discrete state transition rules of cellular automata into activation functions of neural networks. By designing a spatiotemporal feature alignment module, it solves the problem of mismatch in the output dimensions of the two models, thereby achieving online collaborative optimization of the parameters of the two models. This method can effectively integrate the influence weights of various climatic conditions and achieve accurate quantitative calculation of the fire probability.

[0062] 2. Specialized wildfire prediction model for power facility protection:

[0063] This invention constructs a specialized wildfire prediction model for power facility protection. Its core innovation lies in building a prediction system that perceives power grid characteristics. The technical breakthroughs are: first, the development of a dynamic propagation correction algorithm based on the tower coordinate system, which can accurately reflect the fire spread characteristics around transmission lines; second, the design of a tiered early warning mechanism considering equipment importance, directly linking the prediction results to power grid safety assessments. Protection efforts should focus on core technologies such as a power-specific feature extraction network structure, equipment vulnerability assessment models, and industry-customized loss functions. These proprietary technologies constitute the application barriers of this invention in the power industry.

[0064] 3. Scalable multi-source data fusion framework:

[0065] This invention develops a spatiotemporal encoder that supports unified representation of heterogeneous data, achieving feature alignment of multi-source data such as meteorological, remote sensing, and topographic data. It also designs a dynamic weight allocation strategy based on an attention mechanism, capable of adaptively adjusting the contribution of each data source. The protection focus of this framework should include standardized processing methods for multimodal data, incremental learning interface specifications, and online data source access protocols. These technological innovations ensure the system's continuous optimization capability as data dimensions expand.

[0066] It should be noted that although the operation of the method of the present invention has been described in a specific order in the above embodiments and figures, this does not require or imply that the operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0067] The fire spread prediction method for power transmission lines of the present invention will be described in detail below with reference to a specific embodiment. (See reference...) Figure 5 This is a schematic diagram of the overall data branching relationship in a specific embodiment of the present invention. Figure 5 As shown, the geographical branch is used to depict the spatial basis of wildfire spread.

[0068] Elevation data acquisition: DEM (Digital Elevation Model), which reflects the topographic relief (slope, aspect, altitude).

[0069] Tree species data collection: The vegetation type (coniferous forest, broad-leaved forest, shrubs, etc.) around the transmission line is directly related to the characteristics of combustibles (combustibility, load capacity).

[0070] Elevation feature extraction: slope (affects the spread rate, steep slopes accelerate the spread), aspect (affects light and temperature, indirectly affecting combustion).

[0071] Tree species characteristics extraction: combustible material type (e.g., high flammability of coniferous forests) and vegetation density (which affects the spread of fire).

[0072] The meteorological branch is used to capture the dynamic driving forces behind wildfire spread.

[0073] Wind speed and direction information collection: The role of wind in propelling the fire (direction determines the direction of spread, speed determines the rate of spread).

[0074] Temperature and humidity information collection: Increased temperature accelerates combustion, while increased humidity inhibits combustion (the core effect on combustion probability).

[0075] Wind field feature extraction: wind speed vector (magnitude + direction), wind contribution to fire spread (e.g., faster spread in the downwind direction).

[0076] Temperature and humidity feature extraction: temperature threshold (probability of exceeding the ignition point), humidity inhibition coefficient (the higher the humidity, the more difficult it is to burn).

[0077] Feature fusion: Integrating geographical and meteorological features (spatial and dynamic coupling of the current environment) and then incorporating fire status features (supplementing historical patterns), the synergy of multi-source features is achieved through weighting, concatenation, or model fusion (such as LSSVM, neural networks).

[0078] The fire status branch can provide historical experience on wildfire spread.

[0079] Historical fire information collection: the spread path, burning duration, and affected area of ​​past wildfires (such as historical fire data around power transmission lines).

[0080] Fire feature extraction: historical spread patterns (average spread rate in a certain area) and combustion intensity patterns (duration of high temperature in coniferous forest fires).

[0081] Fire spread prediction: The fused feature input prediction model (improved CA cellular automata and deep learning model) outputs the wildfire spread path, speed, and affected area, providing a basis for wildfire early warning for power transmission lines.

[0082] refer to Figure 6 This is a schematic diagram illustrating the steps of data acquisition, processing, and fire prediction according to a specific embodiment of the present invention. Figure 6 As shown, the specific methods include:

[0083] S601, collect raw data and construct an elevation sample database.

[0084] The DEM grid data was extracted from contour lines (1:10000) using the Arc View Spatial Analyst module to construct an elevation sample database. Using the TensorFlow framework, a model pre-trained on the ImageNet dataset based on VGG16 was used to identify tree species in UAV visible light images, ultimately forming a training sample library of tree species (combustible material / normalized vegetation index (NDVI)).

[0085] Local meteorological data was obtained from the meteorological bureau's official website, and real-time wind speed and direction data were collected using drones. The specific process for establishing a tree species training sample library is as follows: Figure 7As shown, the drone collects tree species data and obtains large-scale tree species distribution data through visible light imagery (covering the area around the power transmission line); it also collects other data on-site to supplement detailed data that are difficult for the drone to obtain (such as the thickness of dead branches and fallen leaves, the humidity of hanging fuel, etc., which directly affect the combustion intensity).

[0086] During the data processing stage, the VGG16 pre-trained model under the TensorFlow framework (optimized based on the ImageNet dataset and good at image classification) is used to extract tree species features (such as leaf texture, canopy density, etc.) from UAV images. The feature extraction can classify tree species according to their combustibility (such as coniferous forest / broadleaf forest / shrub, corresponding to different combustion probabilities).

[0087] A combustible sample database was created, including key parameters such as tree species type, combustible humidity, and fallen leaf load, providing standardized data for subsequent Fuel Feature extraction.

[0088] S602, preprocess the data.

[0089] Load the raw data from S601, check for missing values, outliers, or duplicate records in the dataset, and record these issues. For missing data, imputation (e.g., using the mean, median, or mode) or deletion of records containing missing values ​​is performed as needed. Meteorological data is cleaned using Python. The cleaned data is then used to train a least squares support vector machine (LSSVM), which is evaluated and optimized to derive a nonlinear rule for fire combustion probability, ultimately constructing a meteorological branch sample database. DEM grid data is cleaned to construct an elevation sample database. VGG16 is used to identify and classify tree species data collected by UAVs, and combined with surface vegetation / fallen leaf cover, a combustible material sample database is constructed.

[0090] S603 uses preprocessed data to analyze the relationship between meteorological factors and fire spread.

[0091] Meteorological factors include temperature, humidity, wind speed, and wind direction.

[0092] Using the LSSVM model obtained in step two, the corresponding meteorological features are extracted, and the corresponding wind field features are also extracted.

[0093] Then, based on the analysis results, the probability of each adjacent CA grid cell being affected by the fire is calculated, and the calculated impact probability is integrated into the CA model to simulate the spread of forest fire.

[0094] S604 simulates the fire spread process.

[0095] After the model is integrated and calibrated, the starting location and time of the fire are determined, and these conditions are input into the CA model. The CA model is then run to simulate the process of the fire spreading from its starting location, and the final fire spread prediction results are obtained.

[0096] S605 continuously optimizes the model.

[0097] Collect real fire spread data corresponding to the model predictions, including the burning area, spread path, and spread speed. Then, compare the fire spread prediction data with the real fire spread data to evaluate the accuracy of the model's predictions. Finally, continuously optimize the model based on the prediction results.

[0098] The specific processes from S601 to S605 will be explained in detail below.

[0099] S601, collect raw data and construct an elevation sample database.

[0100] In the first sub-step, DEM grid data is extracted from contour lines (1:10000) using the Arc View Spatial Analyst module. Then, auxiliary data such as DEM (Digital Elevation Model), DOM (Digital Orthophoto Map), and SHP are used to enhance the samples and construct an elevation sample database.

[0101] Sub-step two involves using the TensorFlow framework to perform tree species identification in UAV visible light images based on a pre-trained model on the ImageNet dataset using VGG16. The identified tree species information is then combined with Normalized Difference Vegetation Index (NDVI) data to ultimately form a training sample library of tree species (combustible material / NNDVI) data.

[0102] The third step involves obtaining local meteorological data from the meteorological bureau's official website, including temperature, humidity, wind speed, and wind direction, and combining this with real-time wind speed and direction data collected by drones to ultimately form a comprehensive meteorological dataset.

[0103] S602, preprocess the data.

[0104] Sub-step one: Data cleaning.

[0105] The collected meteorological, elevation, combustible material, and fire data are loaded into the data processing environment. The dataset is checked for missing values, outliers, or duplicate records, and these data issues are recorded. For missing data, imputation (e.g., using the mean, median, or mode) or deletion of records containing missing values ​​is performed as appropriate. Next, features related to the probability of fire combustion are selected, and irrelevant or redundant features are removed to improve the model's predictive performance. Finally, the data is transformed as necessary, such as normalized or standardized, to facilitate model training.

[0106] Sub-step two: Establish the model.

[0107] In this invention, the transition of a cell in a cellular automaton from one state to another is considered a classification problem. Therefore, Least Squares Support Vector Machine (LSSVM) is used to classify the data. When studying fire spread, the main focus is on whether the current cell state can transition to a burning state; if so, the cell is a fire point in the next time step and is recorded as 1; otherwise, the cell is not a fire point in the next time step and is recorded as 0. Therefore, the cell transition space is {0,1}, and the conditions for transition and non-transition are mainly due to factors that can currently influence the cell state. Thus, the influence of the cell itself can be recorded as a sample space, making the determination of forest fire spread using cellular automaton transition rules a classification problem. That is, the transition value of a cell can be determined based on conditions.

[0108] Based on the original data, CBR processing is performed to extract features. The specific implementation is as follows:

[0109] Calculate the mean of the input data in a batch-standardized neural network:

[0110]

[0111] Calculate the variance:

[0112]

[0113] Standardized input data:

[0114]

[0115] In the formula, It is the standardized value of the i-th input sample, x. i B is the value of the original input sample, that is, the value of the i-th sample in the original input sample; B is the number of samples contained in a single batch. It is the mean of all samples in a batch, used to calculate the standard deviation and variance. It is the variance of all samples in the batch, used to calculate the standard deviation and variance. It is a small constant that is usually added to the denominator to avoid the denominator being zero and to ensure numerical stability.

[0116] The formula for calculating the BN layer is:

[0117]

[0118] In the formula, y represents the final output value after batch standardization; This represents the scaling factor (a learnable parameter) used to adjust the magnitude of the standardized data; This represents the offset coefficient (a learnable parameter) used to adjust the offset of the standardized data;

[0119] The above Substituting, we get:

[0120]

[0121] The collected data is processed by convolution. For each position (i,j), the convolution operation can be represented as:

[0122]

[0123] In the formula, Output[i,j] represents the output value at position (i,j) in the feature map after convolution, where i and j represent the position coordinates in the feature map after convolution; Kernel[m,n] represents the weight value at position (m,n) in the convolution kernel, where m and n represent the position coordinates in the convolution kernel; Input[i+m,j+n] represents the pixel value at position (i+m,j+n) in the input feature map (the area covered by the sliding window); K is the size of the convolution kernel, and b is the bias term.

[0124] Alternatively, a simpler calculation method can be used:

[0125]

[0126] In the formula, Indicates the weighting coefficient;

[0127] Substitution We can obtain:

[0128]

[0129] ReLU is a non-linear activation function, defined as:

[0130]

[0131] In the formula, negative values ​​of y are reduced to zero and positive values ​​are retained through calculation; x represents the input value of the activation function (the output of the BN layer or convolutional layer).

[0132] After ReLU activation, all negative values ​​become 0, while positive values ​​are retained.

[0133]

[0134] Secondly, based on the processed feature data, the least squares support vector machine (LSSVM) is used to classify the data. The specific implementation process is as follows:

[0135] Given a training set containing N samples, where the i-th input data is... The corresponding i-th output data is The goal is to build a classifier of the following form:

[0136]

[0137] In the formula, f(x) represents the classification decision function of LSSVM, which outputs the sample class (positive / negative); sign is the sign function, which outputs 1 (positive class) or -1 (negative class); T represents transpose; Represents the kernel function mapping function. This means mapping the input sample x to a high-dimensional feature space;

[0138] The least squares support vector classification problem is ultimately an optimization problem that solves the following equation:

[0139]

[0140] In the formula, N represents the total number of training samples; e represents the natural constant; e i This represents the prediction error for the i-th sample;

[0141] And satisfy the following constraints:

[0142]

[0143] In the formula, y i The true label of the i-th sample is represented by the value {0,1}, which represents non-fire point and fire point, respectively.

[0144] The Lagrange polynomial for the problem described in the above equation is:

[0145]

[0146] Where L represents the Lagrange function; This represents the objective function for optimization, balancing model complexity and prediction error; a i These are Lagrange multipliers, which can be positive or negative due to equation constraints. The optimality condition is that all partial derivatives are 0, i, e, i.e., the equation is:

[0147]

[0148] Where I is the identity matrix, Z is an N×d matrix, N is the total number of training samples, d is the dimension of the high-order feature space, and y is an N×1 sample label vector, corresponding to fire points or non-fire points in the wildfire scene. Represents the regularization parameter; Represents the Lagrange multiplier vector;

[0149] To eliminate e and w, we will use the Mercer condition, i.e., the equation:

[0150]

[0151] In the formula, y represents the element in the k-th row and j-th column of the kernel matrix, characterizing the similarity between samples k and j; k y j Represents the true labels of the k-th and j-th samples; x k x j K(x) represents the k-th and j-th training samples; k ,x j ) represents the kernel function, which calculates the kernel value of a sample x in a high-dimensional space. k With x j The inner product; N' represents the total number of rows in the kernel matrix.

[0152] The resulting system of equations relates to b and α. Using the above conditions, the system of equations is transformed into the following equations:

[0153]

[0154] Assumption Here, A is a symmetric positive semi-definite matrix, therefore its inverse matrix exists; Let represent an N×N kernel matrix. Ultimately, the LSSVM classifier can be represented by the equation:

[0155]

[0156] K(x,x i () represents the kernel function, which calculates the kernel function between samples x and x in a high-dimensional space. i The inner product;

[0157] Here, the classification decision function It can be softened into a hard classification through planar logical transformation, and then the probability of each cell being converted into a fire point can be calculated according to the following equation:

[0158]

[0159] Among them, P c This represents the probability of fire impact factors influencing unit transformation. Here, a kernel function with strong localization is chosen, such as the Gaussian radial basis function, i.e.:

[0160]

[0161] Indicates sample x k With x i The Euclidean distance. The width coefficient of the radial basis kernel controls the degree of localization of the kernel function;

[0162] The above equation can then be rewritten as:

[0163]

[0164] Where c is the penalty parameter of the decision function; c and By comparing combinations, for example, c=1 and σ=1, the above equation can then be transformed into:

[0165]

[0166] However, its combustion probability should also consider various interactions within the neighborhood and the influence of adjacent wind effects, which can be summarized into two parts: internal transformation and local interaction. For each unit A(i,j) in this invention (A(i,j) represents the unit (cell) located at (i,j) in the symmetric positive semi-definite matrix A, where (i,j) is the unit coordinate, i.e., the i-th row and j-th column), the unit state transition probability can be expressed as a function of two sub-components:

[0167]

[0168] in, Let p(i,j) represent the state transition probability (whether it becomes a fire point) of the cell at coordinate (i,j) at time t. t θ(i,j) is the ignition probability of the cell at coordinate (i,j) at time t, used to measure the ignition probability of the cell affected by its own factors. t Let represent the adjacent wind effect of the cell at coordinate (i,j) at time t. Therefore, the formula can be modified to the following calculation:

[0169]

[0170] In the formula, the right side of the equal sign contains two parts of parameters: the part on the left side represents the LSSVM basic probability (sigmoid term), and the numerator on the right side represents the weighted sum of the "weights and states" of the neighboring units, K. w Kw represents the weight coefficient of the neighboring unit (the closer to the current cell, the larger Kw is, representing the degree of influence); c(i,j) represents the state coefficient of the neighboring unit at position (i,j) (e.g., 1 represents "combustible / burning", 0 represents "non-combustible / extinguished"); n is the neighborhood range (e.g., n=1 corresponds to a 3×3 neighborhood, covering 8 neighboring units of the current cell).

[0171] S603 uses preprocessed data to analyze the relationship between meteorological factors and fire spread.

[0172] The first sub-step involves using the cleaned data as model input to extract features from meteorological factors (such as temperature, humidity, wind speed, and wind direction); secondly, based on the results of the meteorological factor analysis, the probability of each adjacent CA grid cell being affected by the fire is calculated.

[0173] Sub-step two involves integrating the calculated impact probabilities into the CA model.

[0174] In the two-dimensional cellular automata model, the combustion state is divided into four categories: unable to burn, not yet ignited, burning, and already burning.

[0175] The state of non-combustibility can only be transformed into the state of non-combustibility. The state of not yet ignited will transform into the state of being combusted when the ignition probability P>0.5. The state of being combusted will transform into the state of being combusted after a period of time.

[0176] This invention utilizes a three-dimensional surface cellular automaton model, which is based on this model and incorporates an elevation model to confine the spread of fire within a three-dimensional surface grid. Simultaneously, the influence probabilities obtained in sub-step one are integrated into the current model, and the combustion probability is calculated using the following formula:

[0177]

[0178]

[0179] Among them, P i Let represent the ignition probability of the i-th neighboring cell on the target cell; B represents the base ignition probability constant (benchmark value) in the above formula; K1, K2, K3, and K4 represent weighting coefficients, which adjust the contribution of each influencing factor respectively; Q represents the comprehensive influencing factor (integrating the synergistic effect of tree species, wind field, and topography); S i W represents the tree species factor (such as combustible material type, humidity, etc.) of the i-th neighboring cell; i H represents the wind field factor (wind speed, wind direction influence) of the i-th neighboring cell; H represents the topographic factor (elevation, slope, aspect, etc.); N i This represents the state of the i-th neighboring cell (burning means "burning").

[0180] The total ignition probability P is obtained by summing the ignition probabilities of neighboring cells to the current cell:

[0181]

[0182] P represents the total ignition probability of the target cell (the sum of the ignition probabilities of its 8 neighboring cells).

[0183] At the same time, the number of combustion states has been increased to 5: unburned, just ignited, burning fiercely, starting to extinguish, and already burned.

[0184] Among them, the unburned becomes just ignited after the ignition probability exceeds the preset probability value;

[0185] It is ignited and then, after a first preset time, it transitions to intense combustion.

[0186] Intense combustion continues when the characteristics of the combustible material reach the first preset condition (tree species factor is greater than normal);

[0187] The intense combustion transitions to extinguishing when the characteristics of the combustible material reach the second preset condition (tree species factor equals normal);

[0188] After the initial extinguishing time has elapsed for the second preset period, it will be considered as having already burned.

[0189] S604 simulates the fire spread process.

[0190] Simulations and verifications were conducted using a wildfire scenario in a specific region. The starting location and time of the fire were determined based on real data and input into the CA (Catch-Up) model along with the fire's ignition status as initialization conditions. Next, the CA model was run to simulate the fire's spread from its starting location. The model updated the state of each cell based on the probability of meteorological factors and the CA rules. Finally, the fire spread prediction results output by the model were analyzed, including the predicted burning area, spread path, and spread rate. Based on these predictions, the model was optimized and adjusted.

[0191] S605 continuously optimizes the model.

[0192] This invention simulates the fire spread process every 5 hours from the initial fire point to the end of the fire, and conducts a comparative analysis of the coverage ratio of three-dimensional cellular automata (CA) and two-dimensional cellular automata (CA) models in wildfire spread prediction. For all three models, the simulated fire spread direction is basically consistent with the actual fire spread direction reported in official reports. This indicates that these models can capture the main trend of wildfire spread well. Although the simulated wildfire spread direction is consistent among the three models, they show significant differences in the predicted coverage ratio. The coverage ratio refers to the ratio of the fire-affected area predicted by the model to the actual fire-affected area, which is an important indicator for measuring the accuracy of model prediction. Table 1 shows a comparison of the simulated coverage ratios of the proposed method, three-dimensional cellular automata (CA), and two-dimensional cellular automata (traditional CA) for wildfire spread.

[0193] Table 1 Coverage Ratio of Wildfire Spread Prediction by Method

[0194]

[0195] The parameters in Table 1 are explained below:

[0196] a: The number and ratio of simulated grids;

[0197] β: The number and percentage of unsimulated grids;

[0198] γ: The number and ratio of grids in the simulation;

[0199] S1: Number of grid cells actually affected by wildfire spread;

[0200] S2: Number of grid cells simulating wildfire spread.

[0201] Analysis of the implementation data shows that the present invention generally maintains consistency with actual fire behavior in terms of simulating the direction and speed of wildfire spread. The wildfire simulation coverage is significantly higher than that of three-dimensional cellular automata and two-dimensional cellular automata, while the over-simulation rate is significantly lower than that of three-dimensional cellular automata and two-dimensional cellular automata.

[0202] After introducing the method of exemplary embodiments of the present invention, the following references are made. Figure 8 A fire spread prediction device for power transmission lines according to an exemplary embodiment of the present invention will be described.

[0203] The implementation of the fire spread prediction device for power transmission lines can refer to the implementation of the above-described method, and will not be repeated here. The term "module" or "unit" used below can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0204] Based on the same inventive concept, this invention also proposes a fire spread prediction device for power transmission lines, such as... Figure 8 As shown, the device includes:

[0205] Data acquisition module 810 is used to collect combustible branch data and meteorological branch data;

[0206] The data feature extraction module 820 is used to input the combustible branch data and meteorological branch data into the fire combustion probability model to extract combustible features and meteorological features that are related to fire combustion; wherein, the fire combustion probability model uses a least squares support vector machine to classify the data, analyzes the conversion values ​​corresponding to the combustion state records within a preset range of the transmission line, and determines the data that are related to the fire combustion state based on the conversion values;

[0207] The fire spread prediction module 830 integrates combustible material characteristics and meteorological characteristics into a three-dimensional cellular automaton, simulates the fire spread process based on the fire's starting location and starting time, and determines the fire spread prediction results.

[0208] It should be noted that although several modules of the fire spread prediction device for transmission lines have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the invention, the features and functions of two or more modules described above can be embodied in a single module. Conversely, the features and functions of a single module described above can be further divided and embodied by multiple modules.

[0209] Based on the aforementioned inventive concept, such as Figure 9 As shown, the present invention also proposes a computer device 900, including a memory 910, a processor 920, and a computer program 930 stored in the memory 910 and executable on the processor 920. When the processor 920 executes the computer program 930, it implements the aforementioned method for predicting the spread of fire in power transmission lines.

[0210] Based on the aforementioned inventive concept, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned method for predicting the spread of fire in power transmission lines.

[0211] Based on the aforementioned inventive concept, this invention proposes a computer program product, which includes a computer program that, when executed by a processor, implements a method for predicting the spread of fires in power transmission lines.

[0212] This invention proposes a method and device for predicting the spread of fires along power transmission lines. Based on spatial propagation process modeling, it deeply integrates spatial propagation modeling theory and deep learning technology, providing a novel technical solution for fire prevention in power facilities. At the technical implementation level, this invention first constructs a three-dimensional curved surface cellular automata model as the basic framework, which can accurately simulate the fire propagation process under complex terrain conditions. Based on this, a deep neural network architecture is innovatively introduced, constructing a core database, a high-precision micro-topographic feature database, a normalized vegetation database, and a multi-dimensional meteorological database. The overall solution employs an advanced CBR feature extraction module, using a multi-layer convolutional neural network to perform deep feature learning on the aforementioned multi-source heterogeneous data. Simultaneously, this invention designs a multi-modal data fusion algorithm based on least squares support vector machines, which can effectively integrate the influence weights of various climatic conditions to achieve accurate quantitative calculation of the fire probability. Regarding model optimization, the system establishes a dynamic tuning mechanism, continuously iteratively optimizing model parameters by real-time access to satellite remote sensing, meteorological monitoring, and field observation data. This invention significantly improves the timeliness and accuracy of wildfire early warning for power transmission lines, providing strong technical support for ensuring the safe and stable operation of energy systems.

[0213] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0214] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0215] This invention is described with reference to flowchart illustrations and / or block diagrams of methods and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0216] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0217] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0218] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the spread of fire in a power transmission line, characterized in that, The method includes: Collect data on combustible materials and meteorological data; The combustible material branch data and meteorological branch data are input into the fire combustion probability model to extract combustible material features and meteorological features that are associated with fire combustion. The fire combustion probability model uses a least squares support vector machine to classify the data, analyzes the conversion values ​​corresponding to the combustion state records within a preset range of the transmission line, and determines the data associated with the fire combustion state based on the conversion values. By integrating combustible material characteristics and meteorological characteristics into a three-dimensional cellular automaton, the fire spread process is simulated based on the fire's starting location and time, and the fire spread prediction results are determined. The method also includes: An elevation sample database is constructed, comprising a combustible material sample database and a meteorological sample database. The combustible material sample database is constructed by using the TensorFlow framework, pre-training a model on the ImageNet dataset, and identifying tree species images using drones. The meteorological sample database is constructed by obtaining local meteorological data from the meteorological bureau and combining it with real-time wind speed and direction data collected by drones. The training method for the fire combustion probability model includes: A least squares support vector machine is trained using the elevation sample database, and then evaluated and optimized to obtain a trained fire combustion probability model. Specifically, regarding fire spread, the model analyzes whether the current unit state in the fire combustion probability model transitions to a burning state. If so, the current unit is a fire point in the next moment and is recorded as 1; otherwise, the current unit is not a fire point in the next moment and is recorded as 0. The influence of the current unit is recorded as a sample space, and the fire spread situation is determined through data classification and transformation. This involves integrating combustible material characteristics and meteorological features into a three-dimensional cellular automata, simulating the fire spread process based on the fire's initiation location and time, and determining the fire spread prediction results, including: Combustible material characteristics and meteorological characteristics are integrated into a three-dimensional cellular automaton. The probability of each adjacent grid cell in the three-dimensional cellular automaton being affected by fire is calculated, and the probability of fire impact is integrated into the three-dimensional cellular automaton. The starting location and time of the fire are set, and the fire spread process is simulated using a three-dimensional cellular automata to determine the fire spread prediction results.

2. The fire spread prediction method for transmission lines according to claim 1, characterized in that, The method also includes: The data in the elevation sample database is preprocessed; specifically, the dataset in the elevation sample database is checked for missing values, outliers, or duplicate records. Missing values ​​are filled or deleted, and outliers and duplicate records are deleted.

3. The fire spread prediction method for transmission lines according to claim 1, characterized in that, The method also includes: In the three-dimensional cellular automaton, the combustion state includes at least: unburned, just ignited, intensely burning, beginning to extinguish, and already burned; wherein, unburned changes to just ignited after the ignition probability is greater than a preset probability value; just ignited changes to intensely burning after a first preset time; intensely burning continues when the combustible material characteristics reach a first preset condition; intensely burning changes to beginning to extinguish when the combustible material characteristics reach a second preset condition; beginning to extinguish changes to already burned after a second preset time.

4. The fire spread prediction method for transmission lines according to claim 1, characterized in that, The method also includes: The starting location and time of the fire are determined based on actual fire data, and the fire status is input into the three-dimensional cellular automaton as the initial condition for initializing the model. By running the three-dimensional cellular automaton, the process of a fire spreading from its starting position is simulated, and the state of each cell is updated. The fire spread prediction results output by the three-dimensional cellular automata are analyzed and compared with actual fire data. Based on the comparison results, the three-dimensional cellular automata model is optimized and adjusted.

5. A fire spread prediction device for power transmission lines, characterized in that, The device includes: The data acquisition module is used to collect combustible material branch data and meteorological branch data; The data feature extraction module is used to input the combustible branch data and meteorological branch data into the fire combustion probability model to extract combustible features and meteorological features that are related to fire combustion; wherein, the fire combustion probability model uses a least squares support vector machine to classify the data, analyzes the conversion values ​​corresponding to the combustion state records within a preset range of the transmission line, and determines the data that are related to the fire combustion state based on the conversion values; The fire spread prediction module integrates combustible material characteristics and meteorological characteristics into a three-dimensional cellular automata, simulates the fire spread process based on the fire's starting location and starting time, and determines the fire spread prediction results. This device is also used for: An elevation sample database is constructed, comprising a combustible material sample database and a meteorological sample database. The combustible material sample database is constructed by using the TensorFlow framework, pre-training a model on the ImageNet dataset, and identifying tree species images using drones. The meteorological sample database is constructed by obtaining local meteorological data from the meteorological bureau and combining it with real-time wind speed and direction data collected by drones. A least squares support vector machine is trained using the elevation sample database, and then evaluated and optimized to obtain a trained fire combustion probability model. Specifically, regarding fire spread, the model analyzes whether the current unit state in the fire combustion probability model transitions to a burning state. If so, the current unit is a fire point in the next moment and is recorded as 1; otherwise, the current unit is not a fire point in the next moment and is recorded as 0. The influence of the current unit is recorded as a sample space, and the fire spread situation is determined through data classification and transformation. Specifically, the fire spread prediction module is used for: Combustible material characteristics and meteorological characteristics are integrated into a three-dimensional cellular automaton. The probability of each adjacent grid cell in the three-dimensional cellular automaton being affected by fire is calculated, and the probability of fire impact is integrated into the three-dimensional cellular automaton. The starting location and time of the fire are set, and the fire spread process is simulated using a three-dimensional cellular automata to determine the fire spread prediction results.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 4.

Images