Mountain fire spreading prediction optimization method and device suitable for complex terrain environment
By utilizing clustering algorithms and neural network technology in complex terrain environments, the real-time correction coefficients of wildfire spread models are obtained and corrected, improving the accuracy of wildfire spread prediction, solving the problem of insufficient prediction by existing models in complex terrain, and supporting the safe operation of the power grid.
Patent Information
- Application Number
- CN202510993656.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-11-21
AI Technical Summary
Existing wildfire spread models have low prediction accuracy in complex terrain environments, making them difficult to meet practical needs.
By acquiring historical environmental datasets, clustering algorithms are used to divide the data into multiple environmental data groups. Neural networks are then used to train wildfire spread correction coefficients for each data group. Real-time data is acquired when wildfires are present, and the group to which the data belongs is determined. The real-time correction coefficients are then used to correct the prediction results of the wildfire spread model.
It improves the accuracy of wildfire spread prediction in complex terrain environments, provides more reliable wildfire spread prediction results, and supports the design and emergency control of power grid transmission lines.
Smart Images
Figure CN120994991A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present disclosure relates to the technical field of mountain fire prediction, in particular to a mountain fire spread prediction optimization method and device suitable for complex terrain environment. BACKGROUND
[0002] Mountain fire spread prediction refers to predicting the mountain fire spread speed and direction by using a mountain fire spread model after a mountain fire occurs in a certain place.
[0003] However, the mountain fire spread model in the related art is often only suitable for simple terrain environment, and the prediction result of the mountain fire spread model in complex terrain environment is less accurate. SUMMARY
[0004] The present disclosure provides a mountain fire spread prediction optimization method and device suitable for complex terrain environment, which can realize accurate mountain fire spread prediction in complex terrain environment. The technical solution at least includes the following solutions: In a first aspect, a mountain fire spread prediction optimization method suitable for complex terrain environment is provided, including: obtaining a historical environment data set of a first region, the historical environment data set including historical combustible species, wind speed, slope, slope direction, humidity, air relative humidity, air temperature, calculated mountain fire spread speed and measured mountain fire spread speed; using a clustering algorithm, dividing the historical data into a plurality of environment data groups according to the combustible species; training a neural network using each of the environment data groups, the output of the neural network being a mountain fire spread correction coefficient; in the case of a mountain fire in the first region, obtaining real-time environment data of the first region; determining the first environment data group to which the real-time environment data belongs, the first environment data group being one of the plurality of environment data groups; inputting the real-time environment data into the first neural network corresponding to the first environment data group to obtain a real-time correction coefficient corresponding to the real-time environment data; using the real-time correction coefficient to correct the mountain fire spread prediction result of the real-time environment data by a mountain fire spread model to obtain a corrected mountain fire spread prediction result.
[0005] Optionally, the clustering algorithm is used to divide the historical data into a plurality of environmental data groups according to combustible species, including: dividing the historical environmental data set into a plurality of environmental feature vectors according to time and location, the environmental feature vector of the i-th location at the t-th time step including the combustible species, wind speed, slope, aspect, humidity, air relative humidity, air temperature, calculated wildfire spread speed and measured wildfire spread speed of the i-th location at the t-th time step; grouping the environmental feature vectors according to combustible species to obtain an environmental feature vector set corresponding to each combustible species; selecting environmental feature vectors with calculated wildfire spread speed or measured wildfire spread speed greater than a wildfire spread speed threshold from the environmental feature vector set corresponding to the h-th combustible species to obtain a subset of environmental feature vectors corresponding to the h-th combustible species; weighting the wind speed, slope, aspect, humidity, air relative humidity and air temperature in the subset of environmental feature vectors corresponding to the h-th combustible species based on the entropy weight method to obtain six weights corresponding to the h-th combustible species; and clustering the environmental feature vector set corresponding to the h-th combustible species based on the six weights corresponding to the h-th combustible species using a weighted clustering algorithm to obtain K environmental data groups corresponding to the h-th combustible species.
[0006] Optionally, the combustible species in the real-time environmental data belongs to a first combustible species, and the determination of the first environmental data group to which the real-time environmental data belongs includes: obtaining K environmental data groups corresponding to the first combustible species; obtaining a clustering center of each environmental data group in the K environmental data groups corresponding to the first combustible species to obtain a plurality of clustering centers; calculating a weighted Euclidean distance between the real-time environmental data and each clustering center, and taking the environmental data group corresponding to the clustering center with the shortest weighted Euclidean distance as the first environmental data group.
[0007] Optionally, the weighted Euclidean distance between the real-time environmental data and the clustering center of the k-th environmental data group is calculated using the following formula:
[0008] wherein, is the weighted Euclidean distance between the real-time environmental data and the clustering center of the k-th environmental data group, is the six weights corresponding to the first combustible species, representing the weights of wind speed, slope, aspect, humidity, air relative humidity and air temperature, respectively, represent the wind speed, slope, aspect, humidity, air relative humidity and air temperature at the clustering center of the k-th environmental data group, respectively, represent the wind speed, slope, aspect, humidity, air relative humidity and air temperature in the real-time environmental data, respectively.
[0009] Optionally, the following formula is used to obtain the clustering center of the kth environment data set corresponding to the first combustible species:
[0010] wherein, is the clustering center of the kth environment data set corresponding to the first combustible species, is the total number of environment vectors in the kth environment data set corresponding to the first combustible species, represents the ith environment vector in the kth environment data set corresponding to the first combustible species, respectively represent the wind speed, slope, slope direction, humidity, air relative humidity and air temperature in the ith environment vector in the kth environment data set.
[0011] The second aspect also provides a wildfire spread prediction optimization device suitable for complex terrain environment, comprising: a first acquisition module configured to acquire a historical environment data set of a first region, wherein the historical environment data set comprises historical combustible species, wind speed, slope, slope direction, humidity, air relative humidity, air temperature, calculated wildfire spread speed and measured wildfire spread speed; a clustering module configured to divide the historical data into a plurality of environment data sets according to combustible species by using a clustering algorithm; a training module configured to train a neural network using each of the environment data sets, wherein the output of the neural network is a wildfire spread correction coefficient; a second acquisition module configured to acquire real-time environment data of the first region when a wildfire exists in the first region; an environment data set determination module configured to determine a first environment data set to which the real-time environment data belongs, wherein the first environment data set is one of the plurality of environment data sets; a correction coefficient acquisition module configured to input the real-time environment data into a first neural network corresponding to the first environment data set to obtain a real-time correction coefficient corresponding to the real-time environment data; and a correction module configured to correct a wildfire spread prediction result of a wildfire spread model for the real-time environment data by using the real-time correction coefficient to obtain a corrected wildfire spread prediction result.
[0012] Optionally, the clustering module is further configured to divide the historical environment data set into a plurality of environment feature vectors according to time and location, an environment feature vector of the i-th location at the t-th time step including a fuel species, a wind speed, a slope, a slope direction, a humidity, an air relative humidity, an air temperature, a calculated fire spread speed, and a measured fire spread speed of the i-th location at the t-th time step; grouping the environment feature vectors according to the fuel species to obtain a set of environment feature vectors corresponding to each fuel species; filtering, from the set of environment feature vectors corresponding to the h-th fuel species, an environment feature vector having a calculated fire spread speed or a measured fire spread speed greater than a fire spread speed threshold to obtain a subset of environment feature vectors corresponding to the h-th fuel species; weighting, based on an entropy weight method, the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature in the subset of environment feature vectors corresponding to the h-th fuel species to obtain six weights corresponding to the h-th fuel species; and clustering, based on the six weights corresponding to the h-th fuel species, the set of environment feature vectors corresponding to the h-th fuel species using a weighted clustering algorithm to obtain K environment data groups corresponding to the h-th fuel species.
[0013] Optionally, the combustible material in the real-time environment data belongs to a first fuel species, and the environment data group determination module is further configured to obtain the K environment data groups corresponding to the first fuel species; obtain a clustering center of each of the K environment data groups corresponding to the first fuel species to obtain a plurality of clustering centers; calculate a weighted Euclidean distance between the real-time environment data and each clustering center, and determine an environment data group corresponding to a clustering center having the shortest weighted Euclidean distance as the first environment data group.
[0014] Optionally, in the environment data group determination module, the weighted Euclidean distance between the real-time environment data and the clustering center of the k-th environment data group is calculated using the following formula:
[0015] wherein, is the weighted Euclidean distance between the real-time environment data and the clustering center of the k-th environment data group, is the six weights corresponding to the first fuel species, representing the weights of the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature, respectively, represent the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature at the clustering center of the k-th environment data group, respectively, represent the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature in the real-time environment data, respectively.
[0016] Optionally, in the environment data set determination module, the following formula is used to obtain the clustering center of the kth environment data set corresponding to the first combustible species:
[0017] wherein, is the clustering center of the kth environment data set corresponding to the first combustible species, is the total number of environment vectors in the kth environment data set corresponding to the first combustible species, represents the ith environment vector in the kth environment data set corresponding to the first combustible species, respectively represent the wind speed, slope, slope direction, humidity, air relative humidity and air temperature in the ith environment vector in the kth environment data set.
[0018] The third aspect further provides a computer device, comprising a memory and a processor, at least one computer program is stored in the memory, the at least one computer program is loaded and executed by the processor, so as to execute the optimization method for predicting the spread of mountain fire in complex terrain environment described in the above embodiments.
[0019] The fourth aspect further provides a computer readable storage medium, at least one computer program is stored in the computer readable storage medium, the at least one computer program is loaded and executed by the processor, so as to execute the optimization method for predicting the spread of mountain fire in complex terrain environment described in the above embodiments.
[0020] The fifth aspect provides a computer program product, comprising computer programs / instructions, which are executed by the processor to realize the method of the first aspect.
[0021] The technical scheme provided by the embodiments of the present disclosure has at least the following beneficial effects: In this embodiment, historical environmental datasets for a first region are acquired; a clustering algorithm is used to divide the historical data into multiple environmental data groups according to the type of combustible material; a neural network is trained for each environmental data group, and the output of the neural network is a wildfire spread correction coefficient; when a wildfire exists in the first region, real-time environmental data for the first region is acquired; the first environmental data group to which the real-time environmental data belongs is determined, and the first environmental data group is one of multiple environmental data groups; the real-time environmental data is input into the first neural network corresponding to the first environmental data group to obtain the real-time correction coefficient corresponding to the real-time environmental data; the real-time correction coefficient is used to correct the wildfire spread prediction result of the wildfire spread model based on the real-time environmental data to obtain the corrected wildfire spread prediction result. This is equivalent to correcting the wildfire spread prediction result output by the wildfire spread model based on real-time environmental data, thereby effectively improving the accuracy of wildfire spread prediction in complex environments. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 A flowchart illustrating an exemplary embodiment of this disclosure provides a wildfire spread prediction optimization method applicable to complex terrain environments; Figure 2 These are schematic diagrams of the fireline profile at time step t and time step (t+1). Figure 3 A flowchart of a wildfire spread prediction optimization method applicable to complex terrain environments, provided by another exemplary embodiment of this disclosure, is shown; Figure 4 This illustration shows a schematic diagram of a wildfire spread prediction and optimization device suitable for complex terrain environments, provided in an exemplary embodiment of this disclosure. Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. Detailed Implementation
[0024] Unless otherwise defined, technical terms or scientific terms used herein shall have the same meaning as is commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terms "first", "second", "third", and similar terms used herein do not denote any order, quantity, or importance, but are used to distinguish one element from another, and the terms "one", "another", "an", and similar terms used herein do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced element. The terms "comprises", "comprising", "includes", "including", and the like can mean the presence of one or more elements or components in the described implementations.
[0025] For the purpose of making the objectives, technical solutions and advantages of the present disclosure clearer, the embodiments of the present disclosure will be further described in detail below with reference to the drawings.
[0026] Figure 1 A flowchart of a method for predicting and optimizing the spread of a forest fire in a complex terrain environment is shown, which can be executed by a computer device. Referring to FIG. 1, the method comprises the following steps. Figure 1 The method comprises the following steps. In step 101, a historical environment data set of a first region is obtained.
[0027] The historical environment data set comprises historical combustible species, wind speed, slope, slope direction, humidity, air relative humidity, air temperature, calculated forest fire spread speed, and measured forest fire spread speed.
[0028] The first region herein is a region for which the spread of a forest fire needs to be predicted. For example, it is a region in which a southern power grid is laid.
[0029] In order to ensure the accuracy of the data collected in the historical environment data set, the following three steps are used to collect the historical environment data in the embodiments of the present disclosure.
[0030] First, meteorological data is collected and preprocessed.
[0031] The meteorological data comprises wind speed, air relative humidity, humidity, and air temperature.
[0032] When collecting the meteorological data, micro-meteorological collection devices are laid near typical dense power transmission channels in the southern power grid, and meteorological data near the power transmission channels is collected in combination with the meteorological stations already established in each region.
[0033] It is necessary to ensure that the collected meteorological data near the power transmission channels has an original spatial resolution of less than 1 km. In addition, the meteorological collection devices need to ensure that the meteorological data at the stations can be obtained in real time, with a time resolution of at least minutes; and the data types that can be obtained by the meteorological stations are not limited to air temperature, humidity, air relative humidity, wind speed, etc.
[0034] Optionally, considering that meteorological data is mainly fluid motion and generally does not change abruptly, a Kriging interpolation method based on strong spatial correlation is adopted to perform spatial interpolation on several types of meteorological data at the original spatial resolution, to obtain meteorological data at a finer spatial resolution, and to realize preprocessing of the meteorological data. The implementation of the Kriging interpolation method is relatively common in the related art, and is thus omitted here.
[0035] Secondly, terrain data is obtained.
[0036] The terrain data includes slope and slope direction.
[0037] In obtaining the terrain data, high-precision terrain DEM elevation data can be obtained based on a power grid GIS platform, a cooperative remote sensing surveying and mapping platform, or purchased or cooperative satellite data. In the DEM elevation data, the rasterized data has a spatial resolution of no higher than 100 meters, preferably 30 meters or below, and the data types obtainable based on the DEM data are not limited to slope and slope direction.
[0038] Thirdly, combustible data is obtained.
[0039] The combustible data is combustible species and combustible humidity.
[0040] In obtaining the preparation data, combustible data near typical dense power transmission channels in the southern power grid region can be obtained based on unmanned aerial vehicle laser point cloud scanning inversion and field investigation. The spatial resolution of the combustible data is no lower than that of the DEM elevation data.
[0041] Finally, the spatial resolutions of the terrain data, the combustible data, and the meteorological data need to be unified, and the unification is performed based on the spatial resolution of the DEM elevation data.
[0042] The calculation of the mountain fire spread speed and the measured mountain fire spread speed is determined based on a mountain fire that has occurred in the first region. The calculation of the mountain fire spread speed can be determined based on a mountain fire spread model, and the measured mountain fire spread speed can be determined according to the mountain fire spread distance between two adjacent time steps.
[0043] The following describes the obtaining method of the calculation of the mountain fire spread speed, taking the Wang Zhengfei spread model as an example. The Wang Zhengfei spread model is a semi-empirical forest fire spread model obtained based on multiple fire test data in the Greater and Lesser Hinggan Mountains and Sichuan Province and combined with analysis of the physical mechanism in the forest fire burning process. The model is strong in pertinence and convenient to use, and is most widely applicable in China and meets the application requirements. In implementation, the model can be replaced with other mountain fire spread models as required, and the present disclosure does not limit this.
[0044] In the Wang Zhengfei spread model, the calculation of the fire spread speed is represented by formula (1).
[0045] (1) In formula (1), to calculate the fire spread speed, is the daily spread index, is the slope correction coefficient, is the combustible correction coefficient, is the humidity correction coefficient, is the wind speed correction coefficient. 、 、 、 The combustible correction coefficient is obtained based on Table 1.
[0046] (2) Here, the fire spread model is to predict the fire spread speed of a location at the next time step (i.e., calculate the fire spread speed) by inputting the environmental data of the location at a certain time step into the fire spread model, so the parameters involved in formula (2) are all environmental data input into the fire spread model, such as the environmental data of the i-th location at the t-th time step.
[0047] In formula (2), is the daily maximum temperature (unit: ), is the noon wind power (unit: grade), is the daily minimum humidity (unit: %), is the slope (unit: degree), is the combustible humidity (unit: %), is the wind speed (unit: m / s).
[0048] Table 1: Combustible correction coefficient table
[0049] In obtaining the calculation of the fire spread speed of the i-th location at the t-th time step, only the relevant environmental data of the i-th location at the t-th time step is substituted into the above formula (1) to formula (2) for calculation.
[0050] Alternatively, the measured fire spread speed can be determined according to the fire spread distance between two adjacent time steps. In implementation, it can be calculated based on the images containing the fire of the two adjacent time steps.
[0051] The following takes the image containing the forest fire at the tth time step and the t+1th time step as an example for illustration. For the image containing the forest fire at the tth time step and the t+1th time step, the fire line contour can be extracted, and the fire line contour at the tth time step and the fire line contour at the t+1th time step are obtained. The image containing the forest fire can be obtained by a drone or collected by a remote sensing satellite.
[0052] Optionally, the fire line contour is extracted from the image containing the forest fire in the following manner: first, the image is binarized, and then an edge detection algorithm is used to extract the fire line contour in the image.
[0053] The essence of fire line extraction is the edge extraction of the fire field region. The edge is a discontinuous pixel position in the local feature of the image, and is a place where the gray scale changes sharply in the image. The three common forms of edges are step type, pulse type and roof type. The edge detection algorithm based on the extraction of the binarized fire line edge has good effect. Optionally, the edge detection algorithm includes Sobel algorithm, Canny algorithm, Laplacian algorithm and the like. In the embodiment of the present disclosure, the first-order Sobel algorithm is used to extract the fire line contour.
[0054] Then, the fire line contour at the tth time step and the fire line contour at the t+1th time step are divided into N segments, and the measured forest fire spread speed of each segment is calculated by formula (3).
[0055] Figure 2 is a schematic diagram of the fire line contour at the tth time step and the fire line contour at the t+1th time step. As shown in Figure 2 For the nth segment of the fire line, the distance between the nth segment of the fire line at the tth time step and the nth segment of the fire line at the t+1th time step can be calculated, and the distance is substituted into formula (3) to calculate the measured forest fire spread speed corresponding to the nth segment of the fire line.
[0056] (3) In formula (3), The measured forest fire spread speed corresponding to the nth segment of the fire line, here, the nth segment of the fire line corresponds to multiple positions, that is The measured forest fire spread speed at the multiple positions of the nth segment of the fire line at the tth time step; The time corresponding to the t+1th time step, The time corresponding to the tth time step, The distance between the nth segment of the fire line at the tth time step and the nth segment of the fire line at the t+1th time step.
[0057] In step 102, a clustering algorithm is used to divide the historical data into multiple environment data groups according to the combustible species.
[0058] In step 103, a neural network is trained respectively for each environment data set.
[0059] The output of the neural network is a fire spread correction coefficient.
[0060] The neural network herein is, for example, a BP (Back Propagation) neural network, but is not limited thereto.
[0061] In step 104, real-time environment data of the first region is obtained in the case that a fire exists in the first region.
[0062] Here, the real-time environment data is obtained in the same way as the historical environment data, and the type of the real-time environment data is the same as that of the historical environment data.
[0063] In step 105, a first environment data set to which the real-time environment data belongs is determined.
[0064] The first environment data set is one of the plurality of environment data sets.
[0065] In step 106, the real-time environment data is input into the first neural network corresponding to the first environment data set to obtain a real-time correction coefficient corresponding to the real-time environment data.
[0066] In step 107, the real-time correction coefficient is used to correct a fire spread prediction result of the fire spread model for the real-time environment data to obtain a corrected fire spread prediction result.
[0067] Here, in implementing step 107, first, the real-time environment data is input into the fire spread model to obtain a prediction result of the fire spread model for the real-time environment data (for example, the real-time environment data is input into formula (1) to formula (2) to obtain a prediction result of the fire spread model for the real-time environment data), and then the real-time correction coefficient is multiplied by the prediction result of the fire spread model for the real-time environment data to obtain the corrected fire spread prediction result.
[0068] In the embodiments of the present disclosure, a historical environment data set of a first region is obtained; a clustering algorithm is used to divide the historical data into a plurality of environment data groups according to combustible species; a neural network is trained using each environment data group, and the output of the neural network is a wildfire spread correction coefficient; in the case that there is a wildfire in the first region, real-time environment data of the first region is obtained; a first environment data group to which the real-time environment data belongs is determined, and the first environment data group is one of the plurality of environment data groups; the real-time environment data is input into the first neural network corresponding to the first environment data group to obtain a real-time correction coefficient corresponding to the real-time environment data; and the wildfire spread prediction result of the wildfire spread model for the real-time environment data is corrected using the real-time correction coefficient to obtain a corrected wildfire spread prediction result. In other words, the wildfire spread prediction result output by the wildfire spread model based on the real-time environment data is corrected, thereby effectively improving the accuracy of wildfire spread prediction in a complex environment.
[0069] Figure 3 A flowchart of a wildfire spread prediction optimization method suitable for a complex terrain environment provided by an example embodiment of the present disclosure is shown, which can be executed by a computer device. Referring to Figure 3 , the method comprises: In step 301, a historical environment data set of a first region is obtained.
[0070] For related content of step 301, refer to the aforementioned step 101, and detailed description is omitted here.
[0071] In step 302, a clustering algorithm is used to divide the historical data into a plurality of environment data groups according to combustible species.
[0072] Optionally, step 302 comprises steps a-e as follows.
[0073] Step a, the historical environment data set is divided into a plurality of environment feature vectors according to time and location.
[0074] The environment feature vector of the i-th location at the t-th time step comprises combustible species, wind speed, slope, slope direction, humidity, air relative humidity, air temperature, calculated wildfire spread speed and measured wildfire spread speed of the i-th location at the t-th time step.
[0075] That is, each environment feature vector comprises 9-dimensional data.
[0076] Step b, the environment feature vectors are grouped according to combustible species to obtain an environment feature vector set corresponding to each combustible species.
[0077] In the implementation of step b, the environment feature vectors of the same combustible type are divided into the same environment feature vector set. Assuming that there are H combustible types, H environment feature vector sets can be finally obtained, and the environment feature vectors in the hth environment feature vector set include all the environment feature vectors of the hth combustible type. The value of h ranges from 1 to H.
[0078] In step c, the environment feature vectors whose fire spread speed or measured fire spread speed is greater than the fire spread speed threshold are screened out from the environment feature vector set corresponding to the hth combustible type, to obtain a subset of environment feature vectors corresponding to the hth combustible type.
[0079] Optionally, the value of the fire spread speed threshold ranges from 3 m / s to 10 m / s, for example, 3 m / s, 5 m / s, 7 m / s, 10 m / s, etc.
[0080] The subset of environment feature vectors screened according to the fire spread threshold can reflect the environment vectors related to the hth combustible type and related to the fire in the historical environment data, and the accuracy of the six weights obtained in step d based on the subset of environment feature vectors corresponding to the hth combustible type is higher.
[0081] In step d, the wind speed, slope, aspect, humidity, relative air humidity, and air temperature in the subset of environment feature vectors corresponding to the hth combustible type are weighted based on the entropy weight method, to obtain six weights corresponding to the hth combustible type.
[0082] The six weights corresponding to the hth combustible type here are the weights of wind speed, slope, aspect, humidity, relative air humidity, and air temperature. Different types of combustibles have different influences of wind speed, slope, aspect, humidity, relative air humidity, and air temperature during combustion (fire spread), and the six weights obtained by objectively weighting through the entropy weight method can reflect the influences of the six factors on the hth combustible type during combustion. If the weight of a factor is large, it means that the factor has a greater influence on the hth combustible type during combustion; if the weight of a factor is small, it means that the factor has a smaller influence on the hth combustible type during combustion.
[0083] The implementation of the entropy weight method is more in related technologies, which is omitted here.
[0084] In step e, based on the six weights corresponding to the hth combustible type, a weighted clustering algorithm is used to cluster the environment feature vector set corresponding to the hth combustible type, to obtain K environment data groups corresponding to the hth combustible type.
[0085] In a complex environment, the parameters such as wind speed, slope, slope direction, humidity, air relative humidity and air temperature have large differences. If all the data are analyzed together, the neural network training effect in the subsequent step 303 will be poor. By clustering similar data types into a group, the neural network is trained for the data group of similar types, which can improve the accuracy of the trained neural network.
[0086] Since the six weights corresponding to the hth combustible species have been obtained in the preceding step d, the six weights can be directly used for weighted clustering. In the weighted clustering, any weighted clustering algorithm in the related art can be used, for example, a weighted Kmeans clustering algorithm, and finally the K environment data groups corresponding to the hth combustible species can be obtained.
[0087] The value of K can be set as required. For example, the value of K ranges from 3 to 5, for example, 3, 4 or 5.
[0088] Steps c to e are for processing the hth combustible species. For other combustible species except the hth combustible species, the same processing method is also required to obtain the K environment data groups corresponding to each combustible species. In the case that there are a total of H combustible species, a total of H*K environment data groups can be obtained.
[0089] In step 303, one neural network is trained for each environment data group.
[0090] The output of the neural network is a forest fire spread correction coefficient.
[0091] Here, the input of the neural network is the wind speed, slope, slope direction, humidity, air relative humidity and air temperature in the environment vector, and the output of the neural network is a forest fire spread correction coefficient, which is used to correct the forest fire spread prediction result of the forest fire spread model.
[0092] The environment vectors belonging to the same environment data group can be considered to have environmental similarity, and different environment data groups may have certain differences. Each environment data group can be considered as a type of environment data, and different types of environment data have different effects on forest fires. Training one neural network for each environment data group can obtain the neural network corresponding to each type of environment data, and then the forest fire spread correction coefficient can be obtained based on the neural network.
[0093] In step 304, in the case that there is a forest fire in the first region, real-time environment data of the first region is obtained.
[0094] Here, the real-time environment data is acquired in the same way as the historical environment data, and the real-time environment data and the historical environment data are of the same type.
[0095] In step 305, a first environment data group to which the real-time environment data belongs is determined.
[0096] The first environment data group is one of a plurality of environment data groups.
[0097] Optionally, the combustible in the real-time environment data belongs to a first combustible category, and step 305 includes steps f to h.
[0098] Step f, K environment data groups corresponding to the first combustible category are acquired.
[0099] The K environment data groups corresponding to each combustible category have been obtained in the foregoing step 302, and in the case where the combustible category in the real-time environment data is known (the first combustible category), the K environment data groups corresponding to the first combustible category can be directly determined.
[0100] Step g, the cluster centers of each environment data group in the K environment data groups corresponding to the first combustible category are acquired, to obtain a plurality of cluster centers.
[0101] Optionally, formula (4) is used to acquire the cluster center of the kth environment data group corresponding to the first combustible category: (4) In formula (4), is the cluster center of the kth environment data group corresponding to the first combustible category, respectively represent the wind speed, slope, aspect, humidity, air relative humidity, and air temperature at the cluster center of the kth environment data group, is the total number of environment vectors in the kth environment data group corresponding to the first combustible category, represents the ith environment vector in the kth environment data group corresponding to the first combustible category, respectively represent the wind speed, slope, aspect, humidity, air relative humidity, and air temperature in the ith environment vector in the kth environment data group.
[0102] k is an integer, and the value range of k is 1 to K. Formula (4) can be used to calculate the cluster center for any one of the K environment data groups corresponding to the first combustible category, thereby obtaining a plurality of cluster centers.
[0103] Step h, the weighted Euclidean distance between the real-time environment data and each cluster center is calculated, and the environment data group corresponding to the cluster center with the shortest weighted Euclidean distance is taken as the first environment data group.
[0104] Optionally, the weighted Euclidean distance between the real-time environment data and the clustering center of the kth environment data group is calculated by using formula (5): (5) In formula (5), is the weighted Euclidean distance between the real-time environment data and the clustering center of the kth environment data group, is the weight corresponding to the first combustible species, which represents the weight of wind speed, slope, aspect, humidity, air relative humidity and air temperature, respectively, which has been obtained in step d. is the wind speed, slope, aspect, humidity, air relative humidity and air temperature at the clustering center of the kth environment data group, respectively, is the wind speed, slope, aspect, humidity, air relative humidity and air temperature in the real-time environment data, respectively.
[0105] Based on formula (5), the weighted Euclidean distance between the real-time environment data and the clustering center of each environment data group corresponding to the first combustible species in the K environment data groups can be calculated. If the weighted Euclidean distance of a certain clustering center is the shortest, it means that the real-time environment data is closest to the environment data group corresponding to the clustering center. At this time, the environment data group corresponding to the clustering center with the shortest weighted Euclidean distance can be taken as the first environment data group to which the real-time environment data belongs.
[0106] In step 306, the real-time environment data is input into the first neural network corresponding to the first environment data group to obtain the real-time correction coefficient corresponding to the real-time environment data.
[0107] In step 303, the neural network corresponding to each environment data group has been obtained. After the first environment data group is determined in step 305, the first neural network is also determined. At this time, the real-time environment data can be input into the first neural network corresponding to the first environment data group to obtain the real-time correction coefficient corresponding to the real-time environment data.
[0108] In step 307, the real-time correction coefficient is used to correct the mountain fire spread prediction result of the mountain fire spread model for the real-time environment data to obtain the corrected mountain fire spread prediction result.
[0109] In implementing step 307, first, the real-time environment data is input into the mountain fire spread model to obtain the prediction result of the mountain fire spread model for the real-time environment data (for example, inputting the real-time environment data into formula (1) to formula (2) to obtain the prediction result of the mountain fire spread model for the real-time environment data), and then multiplying the real-time correction coefficient by the prediction result of the mountain fire spread model for the real-time environment data to obtain the corrected mountain fire spread prediction result.
[0110] The modified mountain fire spread prediction result has high accuracy, can provide reference for design, operation and maintenance of the power transmission line and emergency control under mountain fire conditions, can further serve the power grid mountain fire trip risk assessment, and provide scientific basis for emergency disposal of important power transmission corridors of China Southern Power Grid under large-area mountain fire conditions.
[0111] The following is the device embodiment of the present application. For details not described in detail in the device embodiment, reference can be made to the above method embodiment.
[0112] Figure 4 The structure schematic diagram of the mountain fire spread prediction optimization device suitable for complex terrain environment provided by one exemplary embodiment of the present application is shown. Referring to Figure 4 The mountain fire spread prediction optimization device 400 suitable for complex terrain environment includes a first acquisition module 401, a clustering module 402, a training module 403, a second acquisition module 404, an environment data set determination module 405, a correction coefficient acquisition module 406, and a correction module 407.
[0113] The first acquisition module 401 is configured to acquire a historical environment data set of a first region, and the historical environment data set includes historical combustible species, wind speed, slope, slope direction, humidity, air relative humidity, air temperature, calculated mountain fire spread speed, and measured mountain fire spread speed. The clustering module 402 is configured to divide the historical data into a plurality of environment data sets according to the combustible species by using a clustering algorithm. The training module 403 is configured to train a neural network using each environment data set, and the output of the neural network is a mountain fire spread correction coefficient. The second acquisition module 404 is configured to acquire real-time environment data of the first region in the case that there is a mountain fire in the first region. The environment data set determination module 405 is configured to determine a first environment data set to which the real-time environment data belongs, and the first environment data set is one of the plurality of environment data sets. The correction coefficient acquisition module 406 is configured to input the real-time environment data into a first neural network corresponding to the first environment data set to obtain a real-time correction coefficient corresponding to the real-time environment data. The correction module 407 is configured to correct a mountain fire spread prediction result of a mountain fire spread model for the real-time environment data by using the real-time correction coefficient to obtain a modified mountain fire spread prediction result.
[0114] Optionally, the clustering module 402 is further configured to divide the historical environment data set into a plurality of environment feature vectors according to time and location, an environment feature vector of the i-th location at the t-th time step including a flammable species, a wind speed, a slope, a slope direction, a humidity, an air relative humidity, an air temperature, a calculated wildfire spread speed, and a measured wildfire spread speed of the i-th location at the t-th time step; grouping the environment feature vectors according to the flammable species to obtain an environment feature vector set corresponding to each flammable species; filtering, from the environment feature vector set corresponding to the h-th flammable species, an environment feature vector with a calculated wildfire spread speed or a measured wildfire spread speed greater than a wildfire spread speed threshold to obtain a sub-set of environment feature vectors corresponding to the h-th flammable species; weighting, based on an entropy weight method, the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature in the sub-set of environment feature vectors corresponding to the h-th flammable species to obtain six weights corresponding to the h-th flammable species; and clustering, based on the six weights corresponding to the h-th flammable species, the environment feature vector set corresponding to the h-th flammable species using a weighted clustering algorithm to obtain K environment data groups corresponding to the h-th flammable species.
[0115] Optionally, the flammable material in the real-time environment data belongs to a first flammable species, and the environment data group determination module 405 is further configured to obtain K environment data groups corresponding to the first flammable species; obtain a clustering center of each of the K environment data groups corresponding to the first flammable species to obtain a plurality of clustering centers; calculate a weighted Euclidean distance between the real-time environment data and each of the clustering centers, and determine an environment data group corresponding to a clustering center with the shortest weighted Euclidean distance as the first environment data group.
[0116] Optionally, in the environment data group determination module 405, the weighted Euclidean distance between the real-time environment data and the clustering center of the k-th environment data group is calculated using the following formula:
[0117] wherein, is the weighted Euclidean distance between the real-time environment data and the clustering center of the k-th environment data group, is the six weights corresponding to the first flammable species, representing the weights of the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature, respectively, represent the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature at the clustering center of the k-th environment data group, respectively, represent the wind speed, the slope, the slope direction, the humidity, the air relative humidity, and the air temperature in the real-time environment data, respectively.
[0118] Optionally, in the environment data group determination module 405, the clustering center of the k-th environment data group corresponding to the first flammable species is obtained using the following formula:
[0119] wherein, is the clustering center of the kth environmental data set corresponding to the first combustible species, is the total number of environmental vectors in the kth environmental data set corresponding to the first combustible species, represents the ith environmental vector in the kth environmental data set corresponding to the first combustible species, respectively represent the wind speed, slope, slope direction, humidity, air relative humidity and air temperature in the ith environmental vector in the kth environmental data set.
[0120] It should be noted that the above-mentioned embodiment provides a mountain fire spread prediction optimization device suitable for complex terrain environment for mountain fire spread prediction optimization, and only the above-mentioned division of each functional module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the mountain fire spread prediction optimization device suitable for complex terrain environment and the mountain fire spread prediction optimization method embodiment provided by the above-mentioned embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.
[0121] The division of modules in the embodiments of the present disclosure is illustrative, and is only a logical function division. In actual implementation, there can be another division manner. In addition, each functional module in each embodiment of the present disclosure can be integrated in one processor, or can be a separate physical existence, or two or more modules can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.
[0122] When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present disclosure essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing an end device (which can be a personal computer, a mobile phone, or a communication device, etc.) or a processor to execute all or part of the steps of the method of each embodiment of the present disclosure. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.
[0123] Figure 5 is a structural schematic diagram of a computer device provided by an embodiment of the present disclosure. As shown in the figure, the computer device 500 includes a processor 501 and a memory 502. Figure 5
[0124] The processor 501 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 501 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), a PLA (Programmable Logic Array). The processor 501 can also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 501 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content required to be displayed by the display screen. In some embodiments, the processor 501 can also include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.
[0125] The memory 502 can include one or more computer-readable storage media, which can be non-transitory. The memory 502 can also include a high-speed random access memory, and a non-volatile memory such as one or more disk storage devices, flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 502 is used to store at least one instruction for being executed by the processor 501 to implement the method for optimizing the spread of forest fires in complex terrain environments provided by an embodiment of the present disclosure.
[0126] Those skilled in the art can understand that the structure shown in the figure does not constitute a limitation on the computer device 500, and can include more or fewer components than the figure, or combine certain components, or use different component arrangements. Figure 5
[0127] The embodiment of the present disclosure further provides a non-transitory computer readable storage medium, when instructions in the storage medium are executed by a processor of a computer device, the computer device is enabled to execute the method for optimizing the wildfire spread prediction in a complex terrain environment provided in the embodiment of the present disclosure.
[0128] The embodiment of the present disclosure further provides a computer program product, comprising computer programs / instructions, when the computer programs / instructions are executed by a processor, the method for optimizing the wildfire spread prediction in a complex terrain environment provided in the embodiment of the present disclosure is realized.
[0129] The above merely provides optional embodiments of the present disclosure, and is not intended to limit the present disclosure, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.
Claims
1. A wildfire spread prediction optimization method suitable for complex terrain environments, characterized in that, The method comprises: acquiring a historical environment data set of a first region, the historical environment data set comprising historical combustible species, wind speed, slope, aspect, humidity, air relative humidity, air temperature, calculated wildfire spread speed and measured wildfire spread speed; using a clustering algorithm, dividing the historical data into a plurality of environment data groups according to the combustible species; training a neural network using each of the environment data groups, the output of the neural network being a wildfire spread correction coefficient; in the case of a wildfire in the first region, acquiring real-time environment data of the first region; determining a first environment data group to which the real-time environment data belongs, the first environment data group being one of the plurality of environment data groups; inputting the real-time environment data into a first neural network corresponding to the first environment data group to obtain a real-time correction coefficient corresponding to the real-time environment data; correcting a wildfire spread prediction result of the real-time environment data using the wildfire spread model and the real-time correction coefficient to obtain a corrected wildfire spread prediction result.
2. The method of claim 1, wherein, The method comprises: dividing the historical environment data set into a plurality of environment feature vectors according to time and location, the environment feature vector of the ith location at the tth time step comprising the combustible species, wind speed, slope, aspect, humidity, air relative humidity, air temperature, calculated wildfire spread speed and measured wildfire spread speed of the ith location at the tth time step; grouping the environment feature vectors according to the combustible species to obtain an environment feature vector set corresponding to each combustible species; selecting, from the environment feature vector set corresponding to the hth combustible species, an environment feature vector whose calculated wildfire spread speed or measured wildfire spread speed is greater than a wildfire spread speed threshold to obtain a subset of environment feature vectors corresponding to the hth combustible species; based on the entropy weight method, weighting the wind speed, slope, aspect, humidity, air relative humidity and air temperature in the subset of environment feature vectors corresponding to the hth combustible species to obtain six weights corresponding to the hth combustible species; based on the six weights corresponding to the hth combustible species, using a weighted clustering algorithm to cluster the environment feature vector set corresponding to the hth combustible species to obtain K environment data groups corresponding to the hth combustible species.
3. The method of claim 2, wherein, The combustible species in the real-time environment data belongs to a first combustible species, The method comprises: acquiring K environment data groups corresponding to the first combustible species; acquiring a clustering center of each of the K environment data groups corresponding to the first combustible species to obtain a plurality of clustering centers; calculating a weighted Euclidean distance between the real-time environment data and each clustering center, and taking the environment data group corresponding to the clustering center with the shortest weighted Euclidean distance as the first environment data group.
4. The method of claim 3, wherein, The weighted Euclidean distance between the real-time environment data and the clustering center of the kth environment data group is calculated using the following formula: wherein, is a weighted Euclidean distance between the real-time environment data and the cluster center of the kth environment data group, are six weights corresponding to the first combustible species, respectively representing the weights of wind speed, slope, aspect, humidity, air relative humidity and air temperature, respectively represent the wind speed, slope, aspect, humidity, air relative humidity and air temperature at the cluster center of the kth environment data group, respectively represent the wind speed, slope, aspect, humidity, air relative humidity and air temperature in the real-time environment data.
5. The method of claim 4, wherein, The clustering center of the kth environment data group corresponding to the first combustible species is acquired using the following formula: wherein, is the cluster center of the kth environmental data set corresponding to the first combustible species, is the total number of environmental vectors in the kth environmental data set corresponding to the first combustible species, represents the ith environmental vector in the kth environmental data set corresponding to the first combustible species, respectively represent the wind speed, slope, aspect, humidity, air relative humidity and air temperature in the ith environmental vector in the kth environmental data set.
6. A wildfire spread prediction optimization device suitable for complex terrain environments, comprising: The device comprises: a first acquisition module configured to acquire a historical environment data set of a first region, the historical environment data set comprising historical combustible species, wind speed, slope, aspect, humidity, air relative humidity, air temperature, calculated wildfire spread speed, and measured wildfire spread speed; a clustering module configured to divide the historical data into a plurality of environment data groups according to combustible species by using a clustering algorithm; a training module configured to train a neural network using each of the environment data groups, the output of the neural network being a wildfire spread correction coefficient; a second acquisition module configured to acquire real-time environment data of the first region in the presence of a wildfire in the first region; an environment data group determination module configured to determine a first environment data group to which the real-time environment data belongs, the first environment data group being one of the plurality of environment data groups; a correction coefficient acquisition module configured to input the real-time environment data into a first neural network corresponding to the first environment data group to obtain a real-time correction coefficient corresponding to the real-time environment data; a correction module configured to correct a wildfire spread prediction result of a wildfire spread model for the real-time environment data using the real-time correction coefficient to obtain a corrected wildfire spread prediction result.
7. The apparatus of claim 6, wherein, The clustering module is further configured to divide the historical environment data set into a plurality of environment feature vectors according to time and location, an environment feature vector at the i-th location at the t-th time step comprising combustible species, wind speed, slope, aspect, humidity, air relative humidity, air temperature, calculated wildfire spread speed, and measured wildfire spread speed at the i-th location at the t-th time step; grouping the environment feature vectors according to combustible species to obtain an environment feature vector set corresponding to each combustible species; selecting, from the environment feature vector set corresponding to the h-th combustible species, an environment feature vector having a calculated wildfire spread speed or a measured wildfire spread speed greater than a wildfire spread speed threshold to obtain a sub-set of environment feature vectors corresponding to the h-th combustible species; based on an entropy weight method, weighting wind speed, slope, aspect, humidity, air relative humidity, and air temperature in the sub-set of environment feature vectors corresponding to the h-th combustible species to obtain six weights corresponding to the h-th combustible species; based on the six weights corresponding to the h-th combustible species, clustering the environment feature vector set corresponding to the h-th combustible species by using a weighted clustering algorithm to obtain K environment data groups corresponding to the h-th combustible species.
8. A computer device, comprising: The computer device comprises a memory and a processor, at least one computer program is stored in the memory, the at least one computer program is loaded and executed by the processor to implement the method of any one of claims 1 to 5.
9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one computer program, the at least one computer program is loaded and executed by the processor to implement the method of any one of claims 1 to 5.
10. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to implement the method of any one of claims 1 to 5.