Fire risk distribution spatio-temporal evaluation method based on data whole-process processing

Abstract

The application relates to the technical field of forest fire prevention and control, in particular to a mountain fire risk distribution space-time evaluation method based on data whole-process processing, which comprises the following steps: dividing a mountain fire sample data set into a resample data set and an out-of-bag data set; determining all subsets of each mountain fire factor for one resample data set, combining a fire point sequence to obtain the category weight of each mountain fire factor in each resample data set; testing the established decision tree based on the out-of-bag data set, obtaining the factor importance of each mountain fire factor based on the distribution difference characteristics of the test result; comparing the difference characteristics of any two mountain fire factors at different sample points, combining the conditional probability obtained by a Bayesian network model to obtain the correlation confidence of the any two mountain fire factors; and training a mountain fire risk evaluation model to obtain the mountain fire risk probability of a sample point. The application aims to improve the prediction accuracy of the mountain fire risk evaluation model.

Description

Technical Field

[0001] This application relates to the field of forest fire prevention and control technology, specifically to a spatiotemporal assessment method for wildfire risk distribution based on full-process data processing. Background Technology

[0002] Transmission lines are a crucial component of the power system. Transmission networks are deployed across regions, with lines and other equipment exposed to the elements for extended periods. Due to continuous exposure to the natural environment and material aging, transmission line faults pose a high risk of triggering wildfires. Fires caused by power lines often result in widespread safety risks and ecological damage. Furthermore, wildfires reduce the air gap insulation strength of transmission lines, making them highly susceptible to short-circuit breakdowns, inducing line tripping, and resulting in low reclosing success rates, severely jeopardizing the safe and stable operation of the power grid. The distribution of wildfire risk factors, such as geographical factors, vegetation types, and climatic conditions, varies across different transmission line locations, leading to varying levels of wildfire risk from transmission line faults. Establishing wildfire risk assessment models and guiding maintenance personnel in implementing differentiated wildfire prevention measures is a vital means of mitigating wildfire hazards.

[0003] The original wildfire sample dataset contains numerous wildfire risk factors, each with varying degrees of risk and exhibiting differences across regions. Excessive redundant factors introduce unnecessary noise, increasing model complexity and negatively impacting its wildfire risk assessment performance. Current methods for optimizing wildfire sample data ignore the imbalance of factor values ​​within the training set and the correlations between different factors, leading to the erroneous removal of important wildfire risk factors and consequently degrading the model's wildfire risk assessment performance.

[0004] Patent CN109447331A discloses a wildfire risk prediction method based on the stacking algorithm. This method uses stacking to generate secondary features, improving the overall effectiveness of wildfire risk prediction. However, it ignores the noise interference caused by redundant features in the feature set during model training, leading to a decrease in the prediction accuracy. Patent CN109829583A discloses a wildfire risk prediction method based on probabilistic programming technology, utilizing deep learning and probabilistic programming techniques to predict wildfire risk. Due to insufficient analysis of the correlation between different types of data causing wildfire risks, the wildfire risk prediction model established by this method performs poorly in predicting wildfire risks caused by multiple factors. Summary of the Invention

[0005] In view of the above, it is necessary to provide a spatiotemporal assessment method for wildfire risk distribution based on full-process data processing to solve the above problems.

[0006] One embodiment of this application provides a spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing. The method includes:

[0007] Obtain all wildfire factors and fire point sequences for each sample point to form a wildfire sample dataset;

[0008] The wildfire sample dataset is divided into a resampled dataset and an out-of-bag dataset using resampling techniques. For a resampled dataset, all subsets of each wildfire factor are determined. The length characteristics of the fire point sequence of each wildfire factor are analyzed, and combined with the proportion of the number of values ​​of the corresponding subset in the resampled dataset, the class weight of each wildfire factor in each resampled dataset is obtained.

[0009] Decision trees are built based on the set number of wildfire factors in each resampled dataset. The decision trees are tested on the out-of-bag dataset corresponding to the resampled dataset to obtain the out-of-bag error rate of each decision tree. The distribution differences of the out-of-bag error rates of each wildfire factor and the other wildfire factors on all decision trees are analyzed. Combined with the corresponding class weights, the factor importance of each wildfire factor is obtained.

[0010] A pre-set training set is used; based on the numerical difference and distance features between pairwise combined sample points of any two wildfire factors in the training set, combined with the factor importance of each wildfire factor and the conditional probability obtained by the Bayesian network model, the association confidence of any two wildfire factors is obtained.

[0011] Based on the numerical characteristics of the correlation confidence between each wildfire factor and the other wildfire factors, a wildfire risk assessment model is trained to obtain the wildfire risk probability of the sample points.

[0012] Among them, all subsets of each wildfire factor are determined based on the 3σ principle.

[0013] Specifically, the formula for the class weight of each wildfire factor in each resampled dataset is as follows: Let w be the class weight of the k-th wildfire factor in the n-th resampled dataset. k,n Its formula is as follows: Where T represents the number of subsets of the wildfire factor in the resampled dataset; t represents the subset index; r represents the mean length of the fire point sequence of all sample points in the t-th subset of the n-th resampled dataset for the k-th wildfire factor; t This represents the proportion of the number of values ​​for the k-th wildfire factor in the t-th subset of the n-th resampled dataset; ε is a preset value.

[0014] When building the decision tree, a number of wildfire factors will be set as features of the decision tree.

[0015] The specific formula for obtaining the factor importance of each wildfire factor is as follows: Among them, s k w represents the factor importance of the k-th wildfire factor; k,n N represents the class weight of the k-th wildfire factor in the n-th resampled dataset; k represents the number of resampled datasets using the k-th wildfire factor as a feature of the decision tree; exp() represents an exponential function with the natural constant as the base; μ k,n This represents the mean of the average out-of-bag error rates for all wildfire factors except the k-th wildfire factor in the n-th decision tree; er k,n Let represent the average out-of-bag error rate of the k-th wildfire factor in the n-th decision tree.

[0016] Specifically, the average out-of-bag error rate is the average of the out-of-bag error rates of all decision trees with each wildfire factor as a feature of the decision tree.

[0017] The process for determining the association confidence level between any two wildfire factors is as follows:

[0018] By comparing the differences between vectors composed of any two wildfire factors at different sample points and combining them with the distance features between the sample points, the correlation feature value between the two wildfire factors at the sample points is obtained.

[0019] Let CI be the association confidence score between the i-th and j-th wildfire factors. ij Its formula is as follows: in, s represents the average value of the correlation feature between the i-th and j-th wildfire factors across all sample points in the training set; i and s j P represents the factor importance of the i-th and j-th wildfire factors, respectively; i and P j Let represent the conditional probabilities of the i-th and j-th wildfire factors, respectively; norm[] represents the normalization function.

[0020] The process of obtaining the correlation feature values ​​between any two wildfire factors at sample points is as follows:

[0021] The vector formed by any two wildfire factors at a single sample point is denoted as the factor vector.

[0022] The positive correlation mapping result of the distance metric between two sample points is obtained, and combined with the distance metric between the factor vectors of any two wildfire factors in the two sample points, to obtain the correlation feature value of any two wildfire factors between the sample points.

[0023] The training wildfire risk assessment model includes:

[0024] Calculate the sum of the association confidence scores between each wildfire factor and the other wildfire factors. Select the wildfire factor with the smallest sum as the elimination target. Use the wildfire sample dataset of all remaining wildfire factors as the input of the Naive Bayes network model. Eliminate the wildfire factors with the smallest sum of association confidence scores one by one until the model accuracy no longer improves.

[0025] The accuracy of the model is determined by the confusion matrix.

[0026] This application has at least the following beneficial effects:

[0027] This application first reduces the impact of data imbalance on the analysis of wildfire risk caused by different factors by assigning different category weights to multiple wildfire factors, thereby improving the accuracy of wildfire factors used in the training of the wildfire risk assessment model. Simultaneously, it tests the decision trees built on each resampled dataset using out-of-bag datasets to obtain factor importance, enhancing the selection effect for different wildfire factors and reducing noise introduced by redundant wildfire factors in subsequent wildfire risk assessment model training. Furthermore, it analyzes the correlation characteristics between different wildfire factors to obtain correlation confidence, enhancing the model's accuracy in assessing the wildfire risk caused by wildfire factors and avoiding the erroneous removal of wildfire factors. Finally, it evaluates the training results of the wildfire risk assessment model and optimizes factor selection to improve the performance of the trained wildfire risk assessment model. Attached Figure Description

[0028] Figure 1 A flowchart of the spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing provided in this application;

[0029] Figure 2 The process for obtaining the probability of wildfire risk provided in this application. Detailed Implementation

[0030] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0032] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of protection of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0034] This application proposes a spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing, applicable to the field of forest fire prevention and control technology. (See attached document.) Figure 1 The method includes the following steps:

[0035] S1: Obtain all wildfire factors and fire point sequences for each sample point to form a wildfire sample dataset.

[0036] Considering the complex influence of the environment surrounding power transmission lines on the level of wildfire risk, this application establishes a wildfire sample dataset containing three dimensions: geographical feature data, vegetation type factors, and meteorological condition factors, by acquiring multi-source wildfire risk data, in order to enhance the reliability of training the wildfire risk assessment model.

[0037] First, wildfire data for the areas where transmission lines are located for one year were collected from meteorological and power departments. This wildfire data includes geographical features, vegetation types, and meteorological conditions. In this embodiment, box plots were used to detect and remove outliers from the dataset, while missing values ​​were filled. It should be noted that the geographical feature data in this application includes three wildfire factors: altitude, slope, and aspect; the vegetation type data includes two wildfire factors: combustible material risk level and vegetation index; and the meteorological conditions data includes two wildfire factors: annual precipitation and average annual temperature. To facilitate subsequent analysis, all values ​​for each factor were normalized; this embodiment uses the maximum-minimum normalization method.

[0038] Secondly, the area where the transmission line is located is divided into grids of a preset size, with each grid representing a sample point. In this embodiment, the preset size is specifically 1km × 1km, but the implementer can adjust it according to the actual situation. Considering that environmental factors change relatively little between areas in a short period of time, to conduct a time-dimensional wildfire risk assessment, the occurrence times of fire points in each grid throughout the year are arranged chronologically to obtain a fire point sequence for each grid, reflecting the distribution of fire points at different times. The length of the fire point sequence varies between different sample points; the more historical fire points, the longer the sequence. The fire point sequence length for non-fire points is 0.

[0039] Finally, the grid where the fire point is located is used as the fire point sample point, and other grids are used as non-fire point sample points to obtain the fire point distribution in different spaces. Geographic information software is used to extract wildfire factors to the corresponding sample points. The geographical feature data, vegetation type data, meteorological condition data, and fire point sequences of the sample points are combined in a fixed order to form the wildfire sample data for each sample point. Furthermore, the sample data from all sample points in the area where the transmission line is located are used to form the original wildfire sample dataset. It should be noted that in this embodiment, the fixed order of the wildfire factors is specifically altitude, slope, aspect, combustible material risk level, vegetation index, annual precipitation, and average annual temperature. Implementers can set the order of the wildfire factors according to actual conditions.

[0040] S2: The wildfire sample dataset is divided into a resampled dataset and an out-of-bag dataset using resampling techniques. For a resampled dataset, all subsets of each wildfire factor are determined. The length characteristics of the fire point sequence of each wildfire factor are analyzed, and the class weight of each wildfire factor in each resampled dataset is obtained by combining the proportion of the number of values ​​of the corresponding subset in the resampled dataset.

[0041] The original wildfire sample dataset contains numerous wildfire factors. Redundant wildfire factors can introduce noise during the training of the wildfire risk assessment model, thereby affecting the model's performance.

[0042] This application employs a random forest algorithm for optimizing wildfire factors. The wildfire sample dataset is divided into a resampled dataset and an out-of-bag dataset using a Bootstrap resampling technique. In this embodiment, the number of resampling iterations is set to 100, which can be adjusted by the implementer. This yields 100 resampled datasets and their corresponding out-of-bag datasets.

[0043] Considering that extreme environmental conditions can trigger wildfires at a high risk but occur at a low frequency, the dataset exhibits strong imbalance, which in turn affects the reliability of wildfire factor selection.

[0044] This application assigns category weights to different wildfire factors to reduce the impact of data imbalance on the optimal selection of wildfire factors. The specific process is as follows:

[0045] First, because the values ​​of different wildfire factors vary, it is difficult to distinguish the imbalance of different wildfire factor values ​​among the sample data based solely on the magnitude of the wildfire factor values. Therefore, the values ​​of the wildfire factor in a resampled dataset are divided into four subsets.

[0046] For the k-th wildfire factor, calculate the absolute value of the difference between each value of the wildfire factor and the mean of all values ​​of the wildfire factor in each resampled dataset; then, set the absolute value of the difference in the interval [0, σ]. kAll values ​​of the k-th wildfire factor in the dataset are divided into the first subset of the k-th wildfire factor in each resampled dataset; the absolute value of the difference is set in the interval [σ]. k ,2σ k All values ​​corresponding to the k-th wildfire factor are divided into the second subset of each resampled dataset; the absolute value of the difference is set in the interval [2σ]. k ,3σ k The corresponding values ​​in the dataset are divided into the third subset of each resampled dataset for the k-th wildfire factor; the absolute value of the difference is set in the interval [3σ]. k The partitioning of all values ​​corresponding to the k-th wildfire factor in the dataset (+∞) is the fourth subset of each resampled dataset; where σ k It represents the standard deviation of all values ​​of the k-th factor in each resampled dataset.

[0047] Next, the distribution of each wildfire factor value across different subsets in each resampled dataset is analyzed, and the class weight w of the k-th wildfire factor in the n-th resampled dataset is calculated. k,n Its formula is as follows: Where T represents the number of subsets of the wildfire factor in the resampled dataset, which is 4 in this embodiment; t represents the subset number. The larger the subset number, the greater the difference between the factor value in the subset and the mean of the factor in the resampled dataset, reflecting the greater the dispersion of the value of the kth wildfire factor in the subset in the nth resampled dataset. r represents the mean length of the fire point sequence of all sample points in the t-th subset of the n-th resampled dataset for the k-th wildfire factor; t This represents the proportion of the number of values ​​for the k-th wildfire factor in the t-th subset of the n-th resampled dataset, specifically the ratio of the number of values ​​in the t-th subset to the number of values ​​in the n-th resampled dataset; ε is a preset value of 0.1, the purpose of which is to avoid the denominator being zero.

[0048] It should be understood that in the subset corresponding to the wildfire factor, the greater the difference between the value of the wildfire factor and the mean value of the wildfire factor in the resampled dataset, and the smaller the proportion of the wildfire factor values ​​in the subset in the resampled dataset, the greater the dispersion of the wildfire factor values, indicating that the wildfire factor is more unbalanced in the sample point data, and a larger class weight should be set.

[0049] Secondly, considering the temporal distribution of fire points, the greater the mean length of the fire point sequence corresponding to the wildfire factor in the subset, the greater its role in judging wildfire risk, and thus the greater the category weight obtained, so as to improve the assessment performance of wildfire risk caused by extreme environmental factors.

[0050] S3: Build decision trees for each wildfire factor based on the set number of wildfire factors in each resampled dataset. Test the decision trees on the out-of-bag dataset corresponding to the resampled dataset to obtain the out-of-bag error rate of each decision tree. Analyze the distribution differences of the out-of-bag error rates of each wildfire factor and the other wildfire factors on all decision trees. Combine the corresponding class weights to obtain the factor importance of each wildfire factor.

[0051] The wildfire sample data from all sample points in the resampled dataset, along with the class weights of different wildfire factors, are used as input. A random forest algorithm is employed to simultaneously build decision trees on a predetermined number of resampled datasets. A predetermined number of wildfire factors are randomly selected as decision tree features to predict the sample point data. In this embodiment, the predetermined number is 100, and the predetermined value is 3; implementers can adjust this value according to actual conditions. The prediction results of the decision trees in this application are fire points and non-fire points. Based on this, one resampled dataset corresponds to one decision tree, and each decision tree corresponds to three decision tree features. It is important to note that in this embodiment, it is necessary to ensure that each wildfire factor is used as a decision tree feature at least once.

[0052] The decision trees are tested using the out-of-bag dataset corresponding to each resampled dataset as the test set, and the out-of-bag error rate of each decision tree is obtained.

[0053] Calculate the factor importance based on the distribution of out-of-bag error rates: Among them, w k,n N represents the class weight of the k-th wildfire factor in the n-th resampled dataset; k represents the number of resampled datasets using the k-th wildfire factor as a feature for building the decision tree; exp() represents an exponential function with the natural constant as the base; μ k,n This represents the mean of the average out-of-bag error rates for all wildfire factors except the k-th wildfire factor in the n-th decision tree; er k,n Let represent the average out-of-bag error rate of the k-th wildfire factor in the n-th decision tree across all decision trees, and denote it as the average out-of-bag error rate of the k-th wildfire factor; s k This represents the factor importance of the k-th wildfire factor, reflecting its role in assessing wildfire risk.

[0054] It should be understood that in a resampled dataset where the k-th wildfire factor is the feature of the decision tree, the smaller the mean out-of-bag error rate of the k-th wildfire factor relative to the mean of the mean out-of-bag error rates of all other wildfire factors, the greater its role in the subsequent training of the wildfire risk assessment model, and the greater the calculated factor importance. On the other hand, to prevent erroneous removal and further enhance the role of wildfire factors with extreme values ​​that result in higher wildfire risk in subsequent model training, the greater the class weight of a wildfire factor, the greater its calculated factor importance.

[0055] S4: Preset training set; Based on the numerical difference features and distance features between pairwise combined sample points of any two wildfire factors in the training set, combined with the factor importance of each wildfire factor and the conditional probability obtained by the Bayesian network model, the association confidence of any two wildfire factors is obtained.

[0056] To evaluate the impact of different factors on the training effect of the wildfire risk assessment model, 70% of the sample data points were used to form the training set, and the remaining 30% of the sample data points were used as the test set to evaluate the model training effect.

[0057] First, a wildfire sample dataset containing all wildfire factors is used as input to a Naive Bayes network model. The initial and transition networks are learned using the maximum likelihood estimation method, and the conditional probabilities of different wildfire factors are obtained to characterize the risk of a single wildfire factor causing a wildfire.

[0058] The closer the values ​​of two sample points are to different wildfire factors, the greater the probability that the combination of factors will lead to wildfire risk, and the larger the calculated correlation feature value. This application constructs a factor vector from the values ​​of any two wildfire factors in a sample point and calculates the Euclidean distance between the factor vectors of different sample points, which is used to reflect the numerical difference between the two wildfire factors.

[0059] Based on the distribution of wildfire factors in the training set, calculate the correlation feature value between any two factors at sample points: λ ij =ln(δ+D)×E ij Among them, E ij denoted by , represents the Euclidean distance between the factor vectors formed by the i-th and j-th factors of two sample points; D represents the actual distance between the two sample points; ln() represents the logarithmic function with the natural constant as the base, which aims to control the influence of the actual distance between sample points on the calculation of the weights of the associated feature values; δ is a preset parameter, with a value of 1, to avoid the calculation result of the associated feature value being negative; λ ij This represents the correlation feature value between the i-th and j-th factors and two sample points, reflecting the correlation between the two factors in triggering wildfire risk between the sample points.

[0060] It should be understood that wildfires exhibit strong geographical continuity from their development and spread to their eventual extinguishing. Among closely spaced fire points, some points are more likely to experience wildfires due to the spread of the fire itself, rather than being caused by inherent wildfire factors within the points themselves. Therefore, the closer two points are, the weaker the correlation between different factors contributing to the risk of wildfires.

[0061] Furthermore, based on the factor importance and the conditional probability corresponding to each wildfire factor, the association confidence between two wildfire factors is calculated: in, s represents the average value of the correlation feature between the i-th and j-th wildfire factors across all sample points in the training set, reflecting the correlation between the two factors and the risk of wildfires within the sample points of the training set; i and s j P represents the factor importance of the i-th and j-th wildfire factors, respectively; i and P j Let represent the conditional probabilities of the i-th and j-th wildfire factors, respectively; norm[] represents the normalization function, which in this embodiment uses the arctangent normalization function; CI ij This represents the association confidence level between the i-th and j-th wildfire factors, reflecting the probability that the two wildfire factors together will cause wildfire risk.

[0062] S5: Based on the numerical characteristics of the correlation confidence between each wildfire factor and the other wildfire factors, train the wildfire risk assessment model to obtain the wildfire risk probability of the sample point data.

[0063] Using a test set as input and employing a confusion matrix to calculate the model accuracy, the training results of the wildfire risk assessment model are evaluated. The confusion matrix is ​​a well-known existing technique, and will not be elaborated upon here. This application uses a Naive Bayes network model as the wildfire risk assessment model.

[0064] The higher the risk of a wildfire due to the combined effect of two wildfire factors, the greater the association confidence, and the more significant the role of that wildfire factor in assessing wildfire risk. The sum of the association confidence scores between each wildfire factor and the remaining wildfire factors is calculated. The wildfire factor with the smallest sum of association confidence scores is then removed. A wildfire sample dataset containing all remaining wildfire factors is used as input to a Naive Bayes network model. The model is trained using the above method, and the factors with the smallest sum of association confidence scores are gradually eliminated until the model accuracy no longer improves. This completes the training of the wildfire risk assessment model. Based on the wildfire risk probabilities of the sample points output by the model, differentiated wildfire prevention and control work can be guided by operations and maintenance personnel.

[0065] The flowchart for obtaining the probability of wildfire risk is as follows: Figure 2 As shown.

[0066] This application provides a spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing. The method includes: firstly, by assigning different category weights to multiple wildfire factors, reducing the impact of data imbalance on the analysis of wildfire risk caused by different factors, and improving the accuracy of wildfire factors used in the training of the wildfire risk assessment model; secondly, by using out-of-bag datasets to test the decision trees built on each resampled dataset to obtain factor importance, enhancing the selection effect of different wildfire factors and reducing the noise introduced by redundant wildfire factors in the subsequent training of the wildfire risk assessment model; thirdly, by analyzing the correlation characteristics between different wildfire factors to obtain correlation confidence, enhancing the model's accuracy in assessing the wildfire risk caused by wildfire factors and avoiding the erroneous removal of wildfire factors; and finally, evaluating the training results of the wildfire risk assessment model and optimizing factor selection to improve the performance of the trained wildfire risk assessment model.

[0067] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0068] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing, characterized in that: The method includes the following steps: Obtain all wildfire factors and fire point sequences for each sample point to form a wildfire sample dataset; The wildfire sample dataset was divided into a resampled dataset and an out-of-bag dataset using a resampling technique; for a resampled dataset, based on 3 The principle is to determine all subsets of each wildfire factor; analyze the length characteristics of the fire point sequence of each wildfire factor, and combine the proportion of the number of values ​​of the corresponding subset in the resampled dataset to obtain the class weight of each wildfire factor in each resampled dataset. Decision trees are built based on the set number of wildfire factors in each resampled dataset. The decision trees are tested on the out-of-bag dataset corresponding to the resampled dataset to obtain the out-of-bag error rate of each decision tree. The distribution differences of the out-of-bag error rates of each wildfire factor and the other wildfire factors on all decision trees are analyzed. Combined with the corresponding class weights, the factor importance of each wildfire factor is obtained. A pre-set training set is used; based on the numerical difference and distance features between pairwise combined sample points of any two wildfire factors in the training set, combined with the factor importance of each wildfire factor and the conditional probability obtained by the Bayesian network model, the association confidence of any two wildfire factors is obtained. Based on the numerical characteristics of the correlation confidence between each wildfire factor and the other wildfire factors, a wildfire risk assessment model is trained to obtain the wildfire risk probability of the sample points. The formula for the class weight of each wildfire factor in each resampled dataset is as follows: Let the class weight of the k-th wildfire factor in the n-th resampled dataset be denoted as... Its formula is as follows: Where T represents the number of subsets of the wildfire factor in the resampled dataset; t represents the subset index, the larger the subset index, the greater the difference between the factor value in the subset and the mean of the factor in the resampled dataset; This represents the mean length of the fire point sequence of all sample points in the t-th subset of the n-th resampled dataset for the k-th wildfire factor. This represents the proportion of the number of values ​​for the k-th wildfire factor in the t-th subset of the n-th resampled dataset; This is the default value.

2. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 1, characterized in that, When building a decision tree, a number of wildfire factors will be set as features of the decision tree.

3. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 1, characterized in that, The specific formula for obtaining the factor importance of each wildfire factor is as follows: ;in, This indicates the factor importance of the k-th wildfire factor; This represents the class weight of the k-th wildfire factor in the n-th resampled dataset; This represents the number of resampled datasets with the k-th wildfire factor as the feature of the decision tree; exp() represents an exponential function with the natural constant as the base. This represents the mean of the average out-of-bag error rates for all wildfire factors except the k-th wildfire factor in the n-th decision tree. Let represent the average out-of-bag error rate of the k-th wildfire factor in the n-th decision tree.

4. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 3, characterized in that, The average out-of-bag error rate is specifically the average out-of-bag error rate of all decision trees with each wildfire factor as a feature of the decision tree.

5. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 1, characterized in that, The process for determining the confidence level of the association between any two wildfire factors is as follows: By comparing the differences between vectors composed of any two wildfire factors at different sample points and combining them with the distance features between the sample points, the correlation feature value between the two wildfire factors at the sample points is obtained. Let the association confidence between the i-th and j-th wildfire factors be denoted as . Its formula is as follows: ;in, This represents the average value of the correlation feature between the i-th and j-th wildfire factors across all sample points in the training set; and These represent the factor importance of the i-th and j-th wildfire factors, respectively; and Let represent the conditional probabilities of the i-th and j-th wildfire factors, respectively; norm[] represents the normalization function.

6. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 5, characterized in that, The process of obtaining the correlation feature value between any two wildfire factors at the sample points is as follows: The vector formed by any two wildfire factors at a single sample point is denoted as the factor vector. The positive correlation mapping result of the distance metric between two sample points is obtained, and combined with the distance metric between the factor vectors of any two wildfire factors in the two sample points, to obtain the correlation feature value of any two wildfire factors between the sample points.

7. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 1, characterized in that, The training wildfire risk assessment model includes: Calculate the sum of the association confidence scores between each wildfire factor and the other wildfire factors. Select the wildfire factor with the smallest sum as the elimination target. Use the wildfire sample dataset of all remaining wildfire factors as the input of the Naive Bayes network model. Eliminate the wildfire factors with the smallest sum of association confidence scores one by one until the model accuracy no longer improves.

8. The spatiotemporal assessment method for wildfire risk distribution based on end-to-end data processing as described in claim 7, characterized in that, The accuracy of the model is determined by the confusion matrix.

Images