Disaster damage prediction method and system based on multi-dimensional data fusion of machine learning

Abstract

The invention provides a multi-dimensional data fusion disaster damage prediction method and system based on machine learning, and relates to the technical field of power distribution network disaster damage prediction. Data of different sources of a power distribution network is acquired, and multi-dimensional data of the power distribution network is obtained; performing related screening on disaster-forming variables in the multi-dimensional data of the power distribution network to obtain a key disaster-forming variable set, and determining a key sample data set based on the key disaster-forming variable set; dividing the key sample data set to obtain a training data set and a test data set, performing random sampling on the training data set to obtain a plurality of training subsets, and further determining sample dispersion corresponding to each training subset; and constructing a disaster damage decision forest according to the test data set and the sample dispersion corresponding to each training subset, and performing disaster damage prediction on the power distribution network by using the disaster damage decision forest, so that key variables in multi-dimensional data can be effectively extracted and the model calculation efficiency can be improved, and the reliability of the model during disaster damage prediction can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of power distribution network disaster loss prediction, and more specifically, to a disaster loss prediction method and system based on multi-dimensional data fusion based on machine learning. Background Art

[0002] Disaster loss prediction based on multi-dimensional data fusion of machine learning has important application value in distribution networks, especially when facing natural disasters (such as typhoons, heavy rains, earthquakes, etc.). It can help predict the degree of damage and recovery of power grid facilities (such as substations, distribution lines, etc.), improve post-disaster recovery efficiency, and optimize emergency repair and scheduling plans.

[0003] However, in practical applications, the heterogeneity and high dimensionality of data pose challenges to multidimensional data fusion disaster loss prediction based on machine learning. Meteorological data, power grid equipment data, historical disaster data, etc. usually come from different data sources, and the data format and quality vary greatly. In addition, as the data dimension increases, the computational complexity of the model will also increase dramatically, and the training process requires a lot of computing resources and time. Therefore, how to effectively extract key variables from multidimensional data and improve the model's computational efficiency to improve the reliability of the model in disaster loss prediction is a difficult problem faced by the industry. Summary of the invention

[0004] The present application provides a disaster loss prediction method and system based on multidimensional data fusion of machine learning, which can effectively extract key variables in multidimensional data and improve the calculation efficiency of the model, so as to improve the reliability of the model when performing disaster loss prediction.

[0005] In a first aspect, the present application provides a disaster loss prediction method based on multi-dimensional data fusion of machine learning, the prediction method comprising the following steps:

[0006] Obtain data from different sources of the distribution network to obtain multi-dimensional data of the distribution network;

[0007] Performing relevant screening on each disaster-causing variable in the multidimensional data of the distribution network to obtain a key disaster-causing variable set of the distribution network, and determining a key sample data set of the distribution network based on the key disaster-causing variable set;

[0008] Dividing the key sample data set to obtain a training data set and a test data set of the distribution network, randomly sampling the training data set to obtain a plurality of training subsets, and then determining the sample discreteness corresponding to each training subset;

[0009] A disaster loss decision forest of the distribution network is constructed according to the sample discreteness corresponding to the test data set and each training subset, and the disaster loss decision forest is used to predict the disaster loss of the distribution network.

[0010] In this embodiment, the various disaster-causing variables in the multidimensional data of the distribution network are screened to obtain a set of key disaster-causing variables of the distribution network, which specifically includes:

[0011] Determining the variable information entropy of each disaster-causing variable in the multidimensional data of the distribution network;

[0012] Determining the redundancy between various disaster-causing variables in the multidimensional data of the distribution network;

[0013] A disaster variable is selected from all the disaster variables in the multidimensional data of the distribution network, and a variable correlation sequence corresponding to the selected disaster variable is determined according to the variable information entropy of the selected disaster variable and the redundancy between the selected disaster variable and other disaster variables, thereby obtaining a variable correlation sequence corresponding to each disaster variable in the multidimensional data of the distribution network;

[0014] A set of key disaster-causing variables of the distribution network is screened out based on all variable-related sequences.

[0015] In this embodiment, the key disaster-causing variable set of the distribution network is screened out based on all variable-related sequences and specifically includes:

[0016] Using all the variable correlation sequences to construct a variable correlation matrix of the distribution network;

[0017] Screening out all key variable pairs according to the variable correlation matrix;

[0018] Eliminate the correlation of each key variable pair, and then obtain multiple key disaster variables;

[0019] The set of all key disaster-causing variables is taken as the key disaster-causing variable set of the distribution network.

[0020] In this embodiment, determining the key sample data set of the distribution network based on the key disaster variable set is to eliminate the disaster variables in the multidimensional data of the distribution network that are not in the key disaster variable set, and then use the multidimensional data of the distribution network after elimination as the key sample data set of the distribution network.

[0021] In this embodiment, the random sampling of the training data set is a random sampling with replacement of the training data set.

[0022] In this embodiment, determining the sample discreteness corresponding to each training subset specifically includes:

[0023] For each training subset, determining the data similarity between each data sample in the training subset;

[0024] Extracting discrete data clusters of the training subset based on all data similarities;

[0025] The sample discreteness corresponding to the training subset is determined by the discrete data cluster, and then the sample discreteness corresponding to each training subset is obtained.

[0026] In this embodiment, constructing the disaster loss decision forest of the distribution network according to the sample discreteness corresponding to the test data set and each training subset specifically includes:

[0027] Construct a disaster loss decision tree corresponding to each training subset according to the sample discreteness corresponding to each training subset;

[0028] Using each disaster loss decision tree to predict the disaster loss of the test data set, and then obtaining the decision coefficient of each disaster loss decision tree;

[0029] A disaster loss decision forest of the distribution network is constructed based on each disaster loss decision tree and the corresponding decision coefficient.

[0030] In this embodiment, constructing a disaster loss decision tree corresponding to each training subset according to the sample discreteness corresponding to each training subset specifically includes:

[0031] For each training subset, the decision order of magnitude is determined according to the sample discreteness corresponding to the training subset;

[0032] Filtering data samples in the training subset by the decision magnitude;

[0033] A disaster loss decision tree corresponding to the training subset is constructed based on all the screened data samples, and then a disaster loss decision tree corresponding to each training subset is obtained.

[0034] In this embodiment, using the disaster loss decision forest to predict the damage to the distribution network is to input the key sample data set of the distribution network into each disaster loss decision tree of the disaster loss decision forest to perform disaster loss prediction.

[0035] In a second aspect, the present application provides a disaster loss prediction system based on multidimensional data fusion of machine learning, which is used to perform a disaster loss prediction method based on multidimensional data fusion of machine learning, and the prediction system includes:

[0036] A data acquisition module is used to acquire data from different sources of the distribution network and obtain multi-dimensional data of the distribution network;

[0037] A key variable determination module is used to perform relevant screening on each disaster-causing variable in the multidimensional data of the distribution network to obtain a key disaster-causing variable set of the distribution network, and determine a key sample data set of the distribution network based on the key disaster-causing variable set;

[0038] A discreteness determination module is used to divide the key sample data set to obtain a training data set and a test data set of the distribution network, randomly sample the training data set to obtain multiple training subsets, and then determine the sample discreteness corresponding to each training subset;

[0039] The disaster loss prediction module is used to construct a disaster loss decision forest for the distribution network based on the sample discreteness corresponding to the test data set and each training subset, and use the disaster loss decision forest to predict the disaster loss of the distribution network.

[0040] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0041] The multidimensional data of the distribution network is obtained by acquiring data from different sources of the distribution network; the disaster-causing variables in the multidimensional data of the distribution network are screened to obtain the key disaster-causing variable set of the distribution network, and the key sample data set of the distribution network is determined based on the key disaster-causing variable set; the key sample data set is divided to obtain a training data set and a test data set of the distribution network, the training data set is randomly sampled to obtain a plurality of training subsets, and the sample discreteness corresponding to each training subset is determined; a disaster loss decision forest of the distribution network is constructed according to the sample discreteness corresponding to the test data set and each training subset, and the disaster loss decision forest is used to predict the disaster loss of the distribution network.

[0042] It can be seen that in the present application, firstly, by performing relevant screening on disaster-causing variables in the multidimensional data of the distribution network, redundant disaster-causing variables can be removed, the dimension of the data can be significantly reduced, thereby reducing the computational complexity and helping to reduce the risk of overfitting of the model; then, through random sampling, the robustness and adaptability of the model can be enhanced, so that it can make reliable disaster loss predictions in different scenarios and disaster loss modes, and by calculating the sample discreteness of the training subset, the degree of data discreteness of the training subset can be measured, which helps to assign different weights to different training subsets and improve the accuracy of the model in predicting disaster losses; finally, by integrating multiple disaster loss decision trees through the disaster loss decision forest, the deviation and overfitting problems that may occur in a single disaster loss decision tree can be reduced, the generalization ability of the model can be enhanced, and the reliability of disaster loss prediction can be improved.

[0043] In summary, the technical solution adopted in this application can effectively extract key variables in multidimensional data and improve the calculation efficiency of the model, so as to improve the reliability of the model when predicting disaster losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0045] Figure 1 is a flow chart of a disaster loss prediction method based on multidimensional data fusion based on machine learning provided in this application;

[0046] Figure 2 is an exemplary flow chart for determining the sample discreteness corresponding to each training subset provided in the present application;

[0047] Figure 3 is an exemplary flow chart of constructing a disaster loss decision forest for a distribution network according to the present application;

[0048] Figure 4 This is a module structure diagram of a disaster loss prediction system based on multi-dimensional data fusion based on machine learning provided in this application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and drawings. The schematic implementation modes and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention. It should be noted that the present invention is already in the actual development and use stage.

[0050] Embodiment 1

[0051] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 As shown in FIG. 1 , this figure is an exemplary flow chart of a method for disaster loss prediction based on multidimensional data fusion based on machine learning according to this embodiment of the present application. The prediction method includes the following steps:

[0052] In step S1, data from different sources of the distribution network are acquired to obtain multi-dimensional data of the distribution network.

[0053] It should be noted that in the present application, data from different sources of the distribution network include distribution line data, meteorological data and geographical environment data. The data from different sources of the distribution network can be integrated to obtain multi-dimensional data of the distribution network. The multi-dimensional data of the distribution network is a multi-dimensional data sample set related to the distribution network, wherein the data samples are composed of multi-dimensional disaster-causing variables. For example, the disaster-causing variables contained in a data sample may include the service life of the distribution line, historical fault information of the distribution line, the slope of the location of the distribution network equipment, the surface type of the location of the distribution network equipment, the maximum wind speed, the minimum wind speed, the average wind speed, the rainfall, etc., which are not limited here.

[0054] In step S2, relevant screening is performed on each disaster variable in the multidimensional data of the distribution network to obtain a key disaster variable set of the distribution network, and a key sample data set of the distribution network is determined based on the key disaster variable set.

[0055] In this embodiment, the following method is used to perform relevant screening on each disaster-causing variable in the multidimensional data of the distribution network to obtain a key disaster-causing variable set of the distribution network, namely:

[0056] Determining the variable information entropy of each disaster-causing variable in the multidimensional data of the distribution network;

[0057] Determining the redundancy between various disaster-causing variables in the multidimensional data of the distribution network;

[0058] A disaster variable is selected from all the disaster variables in the multidimensional data of the distribution network, and a variable correlation sequence corresponding to the selected disaster variable is determined according to the variable information entropy of the selected disaster variable and the redundancy between the selected disaster variable and other disaster variables, thereby obtaining a variable correlation sequence corresponding to each disaster variable in the multidimensional data of the distribution network;

[0059] A set of key disaster-causing variables of the distribution network is screened out based on all variable-related sequences.

[0060] In the specific implementation, first, the variable information entropy of each disaster variable in the multidimensional data of the distribution network can be determined by the existing information entropy calculation method. The variable information entropy is used to measure the amount of information contained in the disaster variable. The larger the variable information entropy, the greater the contribution of the disaster variable to the disaster loss prediction; then, the redundancy between each disaster variable in the multidimensional data of the distribution network can be determined, where the redundancy represents the information duplication between the disaster variables. Disaster variables with too high redundancy will lead to overfitting of the model, and the Pearson correlation coefficient between the disaster variables can be used as the redundancy between the disaster variables.

[0061] In addition, in specific implementation, a disaster variable can be selected from all the disaster variables in the multidimensional data of the distribution network as the selected disaster variable, so that the variable correlation sequence corresponding to the selected disaster variable can be determined according to the variable information entropy of the selected disaster variable and the redundancy between the selected disaster variable and other disaster variables, wherein the variable correlation sequence is a sequence composed of multiple variable correlations, and the variable correlation is an indicator used to measure the size of the correlation between disaster variables. In actual implementation, the variable correlation can be determined in the following way:

[0062]

[0063] Among them, α(a, b) represents the variable correlation between disaster variables a and disaster variables b, β(a, b) represents the redundancy between disaster variables a and disaster variables b, γ(a) represents the variable information entropy of disaster variable a, and γ(b) represents the variable information entropy of disaster variable b. The variable correlation sequence corresponding to the selected disaster variable can be obtained by the above method, thereby obtaining the variable correlation sequence corresponding to each disaster variable in the multidimensional data of the distribution network.

[0064] In this embodiment, the following method can be used to select the key disaster-causing variable set of the distribution network based on all variable-related sequences, namely:

[0065] Using all the variable correlation sequences to construct a variable correlation matrix of the distribution network;

[0066] Screening out all key variable pairs according to the variable correlation matrix;

[0067] Eliminate the correlation of each key variable pair, and then obtain multiple key disaster variables;

[0068] The set of all key disaster-causing variables is taken as the key disaster-causing variable set of the distribution network.

[0069] In specific implementation, first, all variable correlation sequences can be used to construct a variable correlation matrix of the distribution network, wherein the variable correlation matrix is ​​a correlation matrix composed of variable correlations between all disaster variables, and the variable correlations in all variable correlation sequences can be arranged according to the corresponding relationship between the disaster variables, so as to obtain the variable correlation matrix of the distribution network; then, all key variable pairs can be screened out according to the variable correlation matrix, wherein the key variable pair is composed of a pair of disaster variables, and a variable correlation threshold can be set according to historical experiments and data analysis. The variable correlation threshold set in the present application is 0.78, and other values ​​can also be set in actual implementation, which is not limited here, so that the disaster variables corresponding to the variable correlations greater than the variable correlation threshold in the variable correlation matrix can be used as key variable pairs, and all key variable pairs can be obtained in the above manner.

[0070] In addition, in the specific implementation, each key variable pair can be correlated and eliminated to obtain multiple key disaster variables, where the key disaster variable represents the disaster variable that has a greater contribution to the model prediction. For each key variable pair, the variable correlation mean between the two disaster variables in the key variable pair and other disaster variables can be calculated based on the variable correlation matrix, so that the disaster variables with larger variable correlation mean can be eliminated, and the remaining disaster variables can be used as key disaster variables. Multiple key disaster variables can be obtained in the above manner; finally, the set of all key disaster variables can be used as the key disaster variable set of the distribution network.

[0071] In this embodiment, determining the key sample data set of the distribution network based on the key disaster variable set is to eliminate the disaster variables in the multidimensional data of the distribution network that are not in the key disaster variable set, and then use the multidimensional data of the distribution network after elimination as the key sample data set of the distribution network.

[0072] It should be noted that those disaster variables that do not belong to the key disaster variable set are eliminated from the multidimensional data of the distribution network. The purpose of elimination is to remove redundant data and improve the accuracy of the model. Non-key disaster variables do not contribute significantly to disaster loss prediction, or their information has been covered by other key disaster variables. Eliminating non-key disaster variables can reduce the complexity of the model and avoid overfitting of the machine learning model.

[0073] In addition, it should be noted that by performing relevant screening on the disaster-causing variables in the multidimensional data of the distribution network, redundant disaster-causing variables can be removed, which can significantly reduce the dimension of the data, thereby reducing the computational complexity and helping to reduce the risk of model overfitting.

[0074] In step S3, the key sample data set is divided to obtain a training data set and a test data set of the distribution network, the training data set is randomly sampled to obtain a plurality of training subsets, and then the sample discreteness corresponding to each training subset is determined.

[0075] In specific implementation, the key sample data set can be divided according to a preset division ratio to obtain a training data set and a test data set for the distribution network. For example, 70% of the data samples in the key sample data set can be divided as a training data set, and 30% of the data samples in the key sample data set can be divided as a test data set. This will not be repeated here.

[0076] In this embodiment, the random sampling of the training data set is a random sampling with replacement of the training data set to obtain multiple training subsets; it should be noted that the number of data samples contained in each training subset is the same, and the random sampling with replacement can increase the diversity of the data, so that the model has better adaptability to noise and changes in the data, thereby enhancing its stability.

[0077] Preferably, in this embodiment, reference Figure 2 As shown in FIG. 1 , this figure is an exemplary flow chart of determining the sample discreteness corresponding to each training subset in an embodiment of the present application. In this embodiment, determining the sample discreteness corresponding to each training subset can be implemented by the following steps:

[0078] First, in step S31, for each training subset, the data similarity between each data sample in the training subset is determined;

[0079] Then, in step S32, discrete data clusters of the training subset are extracted based on all data similarities;

[0080] Finally, in step S33, the sample discreteness corresponding to the training subset is determined by the discrete data cluster, and then the sample discreteness corresponding to each training subset is obtained.

[0081] In the specific implementation, first, for each training subset, the data similarity between each data sample in the training subset can be calculated by the cosine similarity method; then, the discrete data cluster of the training subset can be extracted based on all the data similarities, wherein the discrete data cluster is a data cluster containing multiple discrete data, and the similarity threshold can be set according to historical experience, and the mean data similarity corresponding to each data sample in the training subset can be calculated, so that the data samples whose data similarity mean is less than the similarity threshold are regarded as discrete data. Through the above method, all the discrete data of the training subset can be obtained, and the data cluster composed of all the discrete data is regarded as the discrete data cluster of the training subset; finally, the sample discreteness corresponding to the training subset can be determined through the discrete data cluster, wherein the sample discreteness represents the proportion of discrete data in the training subset, and the ratio of the number of discrete data in the discrete data cluster to the number of data samples in the training subset can be regarded as the sample discreteness corresponding to the training subset. Through the above method, the sample discreteness corresponding to each training subset can be obtained.

[0082] It should be noted that random sampling can enhance the robustness and adaptability of the model, so that it can make reliable disaster loss predictions in different scenarios and disaster loss patterns. By calculating the sample discreteness of the training subset, the degree of data discreteness of the training subset can be measured, which helps to assign different weights to different training subsets and improve the accuracy of the model's disaster loss prediction.

[0083] In step S4, a disaster loss decision forest of the distribution network is constructed according to the sample discreteness corresponding to the test data set and each training subset, and the disaster loss decision forest is used to predict the disaster loss of the distribution network.

[0084] Preferably, in this embodiment, reference Figure 3 As shown, this figure is an exemplary flow chart of constructing a disaster loss decision forest for a distribution network in an embodiment of the present application. In this embodiment, the disaster loss decision forest for the distribution network is constructed based on the sample discreteness corresponding to the test data set and each training subset, which can be implemented by the following steps:

[0085] First, in step S41, a disaster loss decision tree corresponding to each training subset is constructed according to the sample discreteness corresponding to each training subset;

[0086] Then, in step S42, each disaster loss decision tree is used to perform disaster loss prediction on the test data set, thereby obtaining a decision coefficient of each disaster loss decision tree;

[0087] Finally, in step S43, a disaster loss decision forest of the distribution network is constructed based on each disaster loss decision tree and the corresponding decision coefficient.

[0088] In specific implementation, first, a disaster loss decision tree corresponding to each training subset can be constructed based on the sample discreteness corresponding to each training subset; then, each disaster loss decision tree can be used to predict the disaster loss of the test data set, and then the decision coefficient of each disaster loss decision tree can be obtained, wherein the decision coefficient is used to represent the allocation weight of the corresponding disaster loss decision tree in the model, and the test data set can be input into each disaster loss decision tree for disaster loss prediction. During the disaster loss prediction process, each disaster loss decision tree will classify each data sample in the test data set to determine its disaster risk level or specific disaster loss value. The disaster loss decision tree is split according to the characteristics of the data sample until the leaf node obtains the final prediction result, thereby determining the accuracy of the prediction result of the disaster loss decision tree, and the accuracy can be used as the decision coefficient of the disaster loss decision tree. The decision coefficient of each disaster loss decision tree can be obtained in the above manner; finally, a disaster loss decision forest of the distribution network can be constructed based on each disaster loss decision tree and the corresponding decision coefficient, that is, each decision coefficient can be used as the construction weight of the corresponding disaster loss decision tree to construct the disaster loss decision forest of the distribution network. It should be noted that the disaster loss decision forest used in this application is a random forest model based on machine learning.

[0089] In this embodiment, the disaster loss decision tree corresponding to each training subset is constructed according to the sample discreteness corresponding to each training subset, which can be specifically constructed in the following manner, namely:

[0090] For each training subset, the decision order of magnitude is determined according to the sample discreteness corresponding to the training subset;

[0091] Filtering data samples in the training subset by the decision magnitude;

[0092] A disaster loss decision tree corresponding to the training subset is constructed based on all the screened data samples, and then a disaster loss decision tree corresponding to each training subset is obtained.

[0093] In the specific implementation, first, for each training subset, the decision order of magnitude can be determined according to the sample discreteness corresponding to the training subset, wherein the decision order of magnitude represents the number of data samples that need to be screened, and the total number of data samples in the training subset can be obtained, thereby rounding down the product of the sample discreteness and the total number of data samples in the training subset, and taking the result as the decision order of magnitude of the training subset; then, the corresponding number of data samples can be screened out in the training subset according to the decision order of magnitude; finally, a disaster loss decision tree corresponding to the training subset can be constructed based on all the screened data samples, that is, a decision tree is constructed for all the screened data samples according to the existing decision tree algorithm, thereby obtaining a disaster loss decision tree corresponding to the training subset. The specific process will not be repeated here, and the disaster loss decision tree corresponding to each training subset can be obtained in the above manner.

[0094] In this embodiment, using the disaster loss decision forest to predict the distribution network's disaster losses is to input the key sample data set of the distribution network into each disaster loss decision tree of the disaster loss decision forest for disaster loss prediction; in specific implementation, each data sample in the key sample data set is input into each disaster loss decision tree in the disaster loss decision forest, and in each disaster loss decision tree, the input data sample is classified by the path from the root node to the leaf node, and the disaster loss decision tree calculates the degree of disaster loss corresponding to the data sample based on the path (such as the loss of electricity, the degree of equipment damage, etc.), and after obtaining the prediction result of each decision tree, the weighted average method is used to perform weighted summation on the prediction results of each disaster loss decision tree, wherein the decision coefficient corresponding to each disaster loss decision tree is used as the corresponding weight in the weighted average method, thereby obtaining a final numerical prediction (such as the loss of electricity or the degree of equipment damage).

[0095] It should be noted that by integrating multiple disaster loss decision trees through the disaster loss decision forest, the deviation and overfitting problems that may occur in a single disaster loss decision tree can be reduced, the generalization ability of the model can be enhanced, and thus the reliability of disaster loss prediction can be improved.

[0096] It can be seen that in the present application, firstly, by performing relevant screening on disaster-causing variables in the multidimensional data of the distribution network, redundant disaster-causing variables can be removed, the dimension of the data can be significantly reduced, thereby reducing the computational complexity and helping to reduce the risk of overfitting of the model; then, through random sampling, the robustness and adaptability of the model can be enhanced, so that it can make reliable disaster loss predictions in different scenarios and disaster loss modes, and by calculating the sample discreteness of the training subset, the degree of data discreteness of the training subset can be measured, which helps to assign different weights to different training subsets and improve the accuracy of the model in predicting disaster losses; finally, by integrating multiple disaster loss decision trees through the disaster loss decision forest, the deviation and overfitting problems that may occur in a single disaster loss decision tree can be reduced, the generalization ability of the model can be enhanced, and the reliability of disaster loss prediction can be improved.

[0097] In summary, the technical solution adopted in this application can effectively extract key variables in multidimensional data and improve the calculation efficiency of the model, so as to improve the reliability of the model when predicting disaster losses.

[0098] Embodiment 2

[0099] This application provides a disaster loss prediction system based on multidimensional data fusion based on machine learning. Figure 4 As shown, this figure is a schematic diagram of a prediction system according to this embodiment of the present application, and the prediction system includes:

[0100] The data acquisition module 100 is used to acquire data from different sources of the distribution network to obtain multi-dimensional data of the distribution network;

[0101] A key variable determination module 200 is used to perform relevant screening on each disaster variable in the multidimensional data of the distribution network to obtain a key disaster variable set of the distribution network, and determine a key sample data set of the distribution network based on the key disaster variable set;

[0102] The discreteness determination module 300 is used to divide the key sample data set to obtain a training data set and a test data set of the distribution network, randomly sample the training data set to obtain multiple training subsets, and then determine the sample discreteness corresponding to each training subset;

[0103] The damage prediction module 400 is used to construct a damage decision forest for the distribution network according to the sample discreteness corresponding to the test data set and each training subset, and use the damage decision forest to predict damage to the distribution network.

[0104] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A disaster loss prediction method based on multidimensional data fusion based on machine learning, characterized in that: The prediction method comprises the following steps: Obtain data from different sources of the distribution network to obtain multi-dimensional data of the distribution network; Performing relevant screening on each disaster-causing variable in the multidimensional data of the distribution network to obtain a key disaster-causing variable set of the distribution network, and determining a key sample data set of the distribution network based on the key disaster-causing variable set; Dividing the key sample data set to obtain a training data set and a test data set of the distribution network, randomly sampling the training data set to obtain a plurality of training subsets, and then determining the sample discreteness corresponding to each training subset; A disaster loss decision forest of the distribution network is constructed according to the sample discreteness corresponding to the test data set and each training subset, and the disaster loss decision forest is used to predict the disaster loss of the distribution network.

2. The method for disaster loss prediction based on multidimensional data fusion based on machine learning as claimed in claim 1, characterized in that: The various disaster-causing variables in the multidimensional data of the distribution network are screened to obtain a set of key disaster-causing variables of the distribution network, which specifically includes: Determining the variable information entropy of each disaster-causing variable in the multidimensional data of the distribution network; Determining the redundancy between various disaster-causing variables in the multidimensional data of the distribution network; A disaster variable is selected from all the disaster variables in the multidimensional data of the distribution network, and a variable correlation sequence corresponding to the selected disaster variable is determined according to the variable information entropy of the selected disaster variable and the redundancy between the selected disaster variable and other disaster variables, thereby obtaining a variable correlation sequence corresponding to each disaster variable in the multidimensional data of the distribution network; A set of key disaster-causing variables of the distribution network is screened out based on all variable-related sequences.

3. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 2, characterized in that: The key disaster-causing variable set of the distribution network is screened out based on all variable related sequences, including: Using all the variable correlation sequences to construct a variable correlation matrix of the distribution network; Screening out all key variable pairs according to the variable correlation matrix; Eliminate the correlation of each key variable pair, and then obtain multiple key disaster variables; The set of all key disaster-causing variables is taken as the key disaster-causing variable set of the distribution network.

4. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 1, characterized in that: Determining the key sample data set of the distribution network based on the key disaster variable set is to eliminate the disaster variables that are not in the key disaster variable set in the multidimensional data of the distribution network, and then use the eliminated multidimensional data of the distribution network as the key sample data set of the distribution network.

5. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 1, characterized in that: The random sampling of the training data set is a random sampling with replacement of the training data set.

6. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 1, characterized in that: Determining the sample discreteness corresponding to each training subset specifically includes: For each training subset, determining the data similarity between each data sample in the training subset; Extracting discrete data clusters of the training subset based on all data similarities; The sample discreteness corresponding to the training subset is determined by the discrete data cluster, and then the sample discreteness corresponding to each training subset is obtained.

7. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 1, characterized in that: Constructing the disaster loss decision forest of the distribution network according to the sample discreteness corresponding to the test data set and each training subset specifically includes: Construct a disaster loss decision tree corresponding to each training subset according to the sample discreteness corresponding to each training subset; Using each disaster loss decision tree to predict the disaster loss of the test data set, and then obtaining the decision coefficient of each disaster loss decision tree; A disaster loss decision forest of the distribution network is constructed based on each disaster loss decision tree and the corresponding decision coefficient.

8. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 7, characterized in that: Constructing the disaster loss decision tree corresponding to each training subset according to the sample discreteness corresponding to each training subset specifically includes: For each training subset, the decision order of magnitude is determined according to the sample discreteness corresponding to the training subset; Filtering data samples in the training subset by the decision magnitude; A disaster loss decision tree corresponding to the training subset is constructed based on all the screened data samples, and then a disaster loss decision tree corresponding to each training subset is obtained.

9. The method for disaster loss prediction based on multi-dimensional data fusion based on machine learning as claimed in claim 1, characterized in that: Using the disaster loss decision forest to predict the damage to the distribution network is to input the key sample data set of the distribution network into each disaster loss decision tree of the disaster loss decision forest to perform disaster loss prediction.

10. A disaster loss prediction system based on multidimensional data fusion of machine learning, used to execute the disaster loss prediction method based on multidimensional data fusion of machine learning as claimed in any one of claims 1 to 9, characterized in that: The prediction system comprises: A data acquisition module is used to acquire data from different sources of the distribution network and obtain multi-dimensional data of the distribution network; A key variable determination module is used to perform relevant screening on each disaster-causing variable in the multidimensional data of the distribution network to obtain a key disaster-causing variable set of the distribution network, and determine a key sample data set of the distribution network based on the key disaster-causing variable set; A discreteness determination module is used to divide the key sample data set to obtain a training data set and a test data set of the distribution network, randomly sample the training data set to obtain multiple training subsets, and then determine the sample discreteness corresponding to each training subset; The disaster loss prediction module is used to construct a disaster loss decision forest for the distribution network based on the sample discreteness corresponding to the test data set and each training subset, and use the disaster loss decision forest to predict the disaster loss of the distribution network.

Images