Ice coating grade analysis method and device based on ice region division, computer equipment, storage medium and program product

Abstract

The invention relates to an ice coating grade analysis method and device based on ice region division, computer equipment, a storage medium and a program product, and relates to the technical field of power grids. The method comprises the steps of obtaining at least one piece of region attribute data in response to an icing risk assessment instruction of a target region; performing feature importance analysis on the respective region attribute feature of each region attribute data to obtain the respective feature importance degree of each region attribute feature; carrying out dimension reduction processing on the regional attribute features of which the feature importance degree reaches an importance degree threshold value to obtain target attribute features; based on the target attribute features, carrying out region clustering on the target regions to obtain each target region; performing icing risk analysis on each target area through an icing risk assessment model to obtain an icing risk level of each target area; the icing risk assessment model is a hybrid model obtained by training based on a graph convolutional network and a long short-term memory network. By adopting the method, the accuracy of icing risk assessment can be improved.

Description

Technical Field

[0001] The present application relates to the field of power grid technology, and in particular to an ice coverage level analysis method, device, computer equipment, storage medium and program product based on ice zone division. Background Art

[0002] Icing disasters are one of the most threatening natural disasters to the safe operation of power grids. In recent years, with global warming, atmospheric circulation has changed, which has intensified the frequency and intensity of icing disasters, making the challenges faced by power grid systems increasingly severe. Based on this, it is particularly important to assess the icing risk in areas covered by power grids.

[0003] At present, ice zone maps are usually used as reference to assess icing risks in power grid areas. However, ice zone maps are generally drawn based on manual experience and cannot accurately and comprehensively reflect the icing conditions in the area, resulting in low accuracy in icing risk assessment. Summary of the invention

[0004] Based on this, it is necessary to provide an icing level analysis method, device, computer equipment, computer readable storage medium and computer program product based on ice zone division, which can improve the accuracy of icing risk assessment in response to the above technical problems.

[0005] In a first aspect, the present application provides an icing level analysis method based on ice area division, comprising: in response to an icing risk assessment instruction for a target area, obtaining at least one regional attribute data of the target area; performing feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; performing dimensionality reduction processing on the regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features; based on the target attribute features, performing regional clustering on the target area to obtain at least one target area; performing icing risk analysis on each target area through an icing risk assessment model to obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0006] In one of the embodiments, an icing risk analysis is performed on each target area through an icing risk assessment model to obtain an icing risk level for each target area, including: performing spatial correlation analysis on the regional attribute characteristics of each target area through a graph convolutional network in the icing risk assessment model to obtain a spatial correlation analysis result for each target area; performing temporal correlation analysis on the regional attribute characteristics of each target area through a long short-term memory network in the icing risk assessment model to obtain a temporal correlation analysis result for each target area; and predicting the icing risk level for each target area based on the spatial correlation analysis results and the temporal correlation analysis results to obtain an icing risk level for each target area.

[0007] In one of the embodiments, the target area includes at least one sample clustering point for training the initial regional clustering model; the method also includes: obtaining at least one sample attribute data of the target area, performing data processing on each sample attribute data to obtain sample attribute characteristics of the target area; inputting the sample attribute characteristics into the initial regional clustering model, and using the initial regional clustering model, based on the density peak method, screening out the initial clustering center that matches the optimal number of clusters from each sample clustering point; for other clustering points in each sample clustering point except the initial clustering center, determining the Mahalanobis distance between other clustering points and the initial clustering center; when the Mahalanobis distance meets the clustering condition, clustering other clustering points with the initial clustering center to obtain a first clustering result; based on the first clustering result, training the initial regional clustering model until the first clustering result meets the convergence condition to obtain a regional clustering model; based on the target attribute characteristics, performing regional clustering on the target area to obtain at least one target area, including: inputting the target attribute characteristics into the regional clustering model, performing regional clustering on the target area to obtain at least one target area.

[0008] In one of the embodiments, the process of determining the optimal number of clusters includes: for each candidate number of clusters, inputting the sample attribute data into the initial regional clustering model, performing clustering, and obtaining a second clustering result that matches each candidate number of clusters; obtaining the difference between the second clustering result and the expected value of the clustering result, and taking the candidate number of clusters with the largest difference as the optimal number of clusters.

[0009] In one embodiment, the method further includes: when the first clustering result does not meet the convergence condition, based on the first clustering result, updating each initial cluster center in parallel to obtain an updated cluster center; using the updated cluster center as the initial cluster center, and returning to the step: for other cluster points in each sample cluster point except the initial cluster center, determining the Mahalanobis distance between other cluster points and the initial cluster center.

[0010] In one of the embodiments, dimension reduction processing is performed on regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features, including: dimension reduction processing is performed on regional attribute features whose feature importance reaches an importance threshold to obtain temporary attribute features; the temporary attribute features include at least two features of the terrain features, meteorological features, hydrological features and tower features of the target area; and feature fusion is performed on each temporary attribute feature to obtain the target attribute feature.

[0011] In the second aspect, the present application also provides an icing level analysis device based on ice area division, including: a data acquisition module, used to respond to an icing risk assessment instruction for the target area and acquire at least one regional attribute data of the target area; a feature importance analysis module, used to perform feature importance analysis on the regional attribute features of each regional attribute data respectively, and obtain the feature importance of each regional attribute feature; a feature dimension reduction processing module, used to perform dimensionality reduction processing on the regional attribute features whose feature importance reaches the importance threshold, and obtain the target attribute features; a regional clustering module, used to perform regional clustering on the target area based on the target attribute features, and obtain at least one target area; an icing level analysis module, used to perform icing risk analysis on each target area through an icing risk assessment model, and obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0012] In a third aspect, the present application also provides a computer device, including a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program: in response to an icing risk assessment instruction for a target area, obtaining at least one regional attribute data of the target area; performing feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; performing dimensionality reduction processing on the regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features; based on the target attribute features, performing regional clustering on the target area to obtain at least one target area; performing icing risk analysis on each target area through an icing risk assessment model to obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0013] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which implements the following steps when executed by a processor: in response to an icing risk assessment instruction for a target area, obtaining at least one regional attribute data of the target area; performing feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; performing dimensionality reduction processing on the regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features; based on the target attribute features, performing regional clustering on the target area to obtain at least one target area; performing icing risk analysis on each target area through an icing risk assessment model to obtain an icing risk level for each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0014] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, which implements the following steps when executed by a processor: in response to an icing risk assessment instruction for a target area, obtain at least one regional attribute data of the target area; perform feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; perform dimensionality reduction processing on the regional attribute features whose feature importance reaches the importance threshold to obtain target attribute features; based on the target attribute features, perform regional clustering on the target area to obtain at least one target area; perform icing risk analysis on each target area through an icing risk assessment model to obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0015] The above-mentioned ice level analysis method, device, computer equipment, computer-readable storage medium and computer program product based on ice area division, in response to the ice risk assessment instruction for the target area, first obtains at least one regional attribute data of the target area, thereby extracting the corresponding regional attribute features. Then, feature importance analysis is performed on each regional attribute feature respectively, so as to obtain the feature importance of each regional attribute feature, and the regional attribute features whose feature importance reaches the importance threshold are used as the attribute features to be processed for subsequent dimensionality reduction processing. By screening out important regional attribute features, the accuracy of subsequent regional clustering can be effectively improved. Then, the attribute features to be processed are subjected to dimensionality reduction processing to obtain the target attribute features. The dimensionality reduction processing can effectively reduce the dimension of the feature data, which can not only improve the efficiency of subsequent regional clustering, but also reduce the influence of noise and improve the accuracy of regional clustering. Then, accurate regional clustering is performed on the target area. Finally, the icing risk assessment model is used to analyze the icing risk of each target area obtained by clustering, and the icing risk level of each target area is obtained. Since the icing risk assessment model is a hybrid model trained based on graph convolutional network and long short-term memory network, it can better capture the interdependence and change laws of regional attribute data in time and space dimensions, thereby improving the accuracy of icing risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0017] FIG1 is an application environment diagram of an ice coverage level analysis method based on ice zone division in one embodiment;

[0018] FIG2 is a schematic flow chart of an ice coverage level analysis method based on ice zone division in one embodiment;

[0019] FIG3 is a schematic diagram of a process for predicting ice risk level in one embodiment;

[0020] FIG4 is a schematic diagram of a training process of a regional clustering model in one embodiment;

[0021] FIG5 is a schematic diagram of a process for determining an optimal number of clusters in one embodiment;

[0022] FIG6 is a schematic flow chart of an ice coverage level analysis method based on ice zone division in a specific embodiment;

[0023] FIG7 is a schematic diagram of a training process of a regional clustering model in a specific embodiment;

[0024] FIG8 is a structural block diagram of an ice coverage level analysis device based on ice zone division in one embodiment;

[0025] FIG. 9 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0027] Icing disasters are one of the natural disasters that pose the greatest threat to the safe operation of power grids, especially in complex mountainous areas within power grid areas, where transmission lines face serious icing threats all year round. In recent years, with global warming, changes in atmospheric circulation have occurred, exacerbating the frequency and intensity of icing disasters, which has made the challenges faced by power grid systems increasingly severe. Based on this, it is particularly important to assess the icing risk in areas covered by power grids. At present, ice zone maps are usually used to assess the icing risk in power grid areas. Accurate ice zone maps can not only effectively guide the design and operation and maintenance of transmission lines in power grid areas, but also significantly improve the ability of power grids to resist icing disasters, which is of great significance to ensuring the safe and stable operation of power grids. However, current ice zone maps are generally drawn based on artificial experience, resulting in a high degree of human dependence in the formulation process of ice zone maps, making it difficult to accurately and comprehensively reflect the icing situation in the region, resulting in low accuracy in icing risk assessment. Based on this, the present application proposes a more accurate icing grade analysis method based on ice zone division.

[0028] The ice level analysis method based on ice zone division provided in the embodiment of the present application can be applied in the application environment shown in Figure 1. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. In response to the ice risk assessment instruction for the target area initiated by the terminal 102, the server 104 first obtains at least one regional attribute data of the target area, and performs feature importance analysis on the respective regional attribute features of each regional attribute data, and obtains the respective feature importance of each regional attribute feature. Then, the regional attribute features whose feature importance reaches the importance threshold are subjected to dimensionality reduction processing to obtain the target attribute features, so as to perform regional clustering on the target area based on the target attribute features to obtain at least one target area. Finally, through the ice risk assessment model, the ice risk analysis is performed on each target area to obtain the respective ice risk level of each target area. The ice risk assessment model is a hybrid model trained based on the graph convolution network and the long short-term memory network. Among them, the terminal 102 can be but is not limited to various personal computers, laptops, smart phones, tablets and Internet of Things devices, and the server 104 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0029] In an exemplary embodiment, as shown in FIG. 2 , an ice coverage level analysis method based on ice zone division is provided, and the method is applied to the server 104 in FIG. 1 as an example for description, including the following steps:

[0030] Step S202: in response to an instruction for assessing icing risk in a target area, obtaining at least one area attribute data of the target area.

[0031] The target area may refer to an area with complex terrain, such as mountains, hills, basins, etc. The icing risk assessment instruction may refer to an instruction to conduct an icing risk assessment on the target area. The regional attribute data may refer to basic attribute data of the target area, including but not limited to terrain data, meteorological data, hydrological data and tower data of the target area.

[0032] In one example, terrain data may refer to high-precision digital elevation model (DEM) data of the target area. The high-precision digital elevation model is a digital simulation of the ground terrain (i.e., a digital expression of the terrain surface morphology) through limited terrain elevation data. It is a physical ground model that represents the ground elevation in the form of a set of ordered numerical arrays. The terrain data includes but is not limited to the slope, slope direction, slope change rate and other data of the target area. Meteorological data may refer to the meteorological environment data of the target area, including but not limited to the temperature, humidity, wind speed, wind direction and other data of the target area. Hydrological data may refer to the distribution data of water bodies in the target area, including but not limited to the data of rivers, lakes, reservoirs and other data in the target area. Tower data may refer to the relevant data of transmission towers in the target area. Transmission towers are one of the basic equipment in the power transmission line of the power grid. Their main function is to support the transmission line and ensure that a certain safety distance is maintained between wires and wires, and between wires and the ground. Tower data includes but is not limited to the tower height, tower location coordinates, the location of ice-covered towers, the frequency of ice-covered towers, and the ice thickness of ice-covered towers.

[0033] In one example, regional attribute data can be collected through a multi-source data collection system, which can include a terrain data collection system, a meteorological data collection system, a hydrological data collection system, and a tower data collection system. Among them, the terrain data collection system can be connected to remote sensing satellites such as SRTM (Shuttle Radar Topography Mission) or ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) to collect terrain data of the target area. The meteorological data collection system can be connected to automatic weather stations distributed in the target area. The distribution density and distribution location of the automatic weather stations can be optimized according to the area and terrain complexity of the target area to ensure the representativeness and reliability of the meteorological data. The hydrological data collection system can obtain the hydrological data of the target area by interpreting remote sensing images and field survey reports, and store the hydrological data in GIS (Geographic Information System) vector format. The tower data acquisition system can be connected to the power system database to obtain tower data from the power system database and verify and update it through GPS (Global Positioning System) field measurements. Furthermore, the tower data acquisition system is also connected to the transmission line online monitoring system to collect relevant data on ice-covered towers.

[0034] In one example, the collection frequency of regional attribute data can be set to once per hour to ensure that the dynamic changes in ice cover conditions in the target area are captured. Of course, other collection frequencies can also be set according to actual conditions, and there is no limitation here.

[0035] Specifically, when the server receives an icing risk assessment instruction for an area with complex terrain initiated by the terminal, it can collect regional attribute data of the area, namely terrain data, meteorological data, hydrological data and tower data, through a multi-source data acquisition system.

[0036] Step S204 , performing feature importance analysis on each regional attribute feature of each regional attribute data to obtain the feature importance of each regional attribute feature.

[0037] Among them, the regional attribute features may refer to the features extracted from the regional attribute data, including but not limited to the terrain features corresponding to the terrain data, the meteorological features corresponding to the meteorological data, the hydrological features corresponding to the hydrological data, and the tower features corresponding to the tower data. The feature importance may refer to the importance of the icing risk assessment for the target area. It can be understood that the higher the feature importance, the more conducive it is to the subsequent regional clustering and icing risk assessment.

[0038] Specifically, after the server obtains the attribute data of each region of the target region collected by the multi-source data acquisition system, it can further perform feature extraction on the attribute data of each region to obtain the regional attribute features of each region. In order to improve the efficiency and accuracy of subsequent regional clustering, this embodiment chooses to perform importance analysis on the attribute features of each region, so as to obtain the feature importance of each regional attribute feature. In a preferred example, the Shapley value can be used to evaluate the importance of features. The Shapley value is used to measure the contribution of each participant to the overall value in a cooperative game. In this example, it can be used to describe the contribution rate of each regional attribute feature to characterize its importance, that is, the higher the contribution rate, the higher the feature importance. The Shapley value calculation formula for each regional attribute feature is as follows:

[0039]

[0040] in, represents the Shapley value that should be obtained for the attribute features of the j-th category of regions, F represents the feature set of the attribute features of each region, S is a subset of F, i.e., the feature subset, and v() is the prediction performance metric of the feature subset containing the attribute features of the j-th category of regions. The random forest model can be used as the basic model to predict the performance of the feature subset, and the mean absolute error of the model output is used as the performance metric. Since it is difficult to calculate all possible feature subsets, the Monte Carlo sampling method can be further used in this example to approximate the Shapley values ​​of other feature subsets.

[0041] In one example, after extracting the attribute features of each region, the server can further process the attribute features of each region to improve the quality and reliability of the feature data. Specifically, the server can detect and remove outliers for the attribute features of each region based on the local outlier factor (LOF) algorithm. The expression for identifying outlier features can be:

[0042]

[0043] Among them, i represents the i-th regional attribute data collection point in the target area, j represents the j-th regional attribute feature, represents the mean of the LOF values ​​of the attribute characteristics of the j-th region, The standard deviation of the LOF value of the attribute feature of the j-th region. The local anomaly factor algorithm can effectively identify the abnormal features in the attribute features of each region, thereby improving the reliability of the feature data. Furthermore, the kd tree algorithm can also be used to accelerate the calculation process of the local anomaly factor algorithm to improve the efficiency of identifying abnormal features.

[0044] In one example, after the server completes the abnormal feature recognition, it can normalize the regional attribute features after removing the abnormal features. Specifically, the Min-Max normalization method can be used to introduce nonlinear transformation to enhance the feature discrimination. The normalization conversion function can be:

[0045]

[0046] in, is an adjustable nonlinear factor, and its value range can be between 0.1 and 10. and are the mean and standard deviation of the attribute characteristics of the jth region, respectively. The nonlinear normalization method can better handle features of different scales and distributions, especially for features with long-tail distributions such as ice thickness.

[0047] In one example, after completing feature normalization, the server can further process the missing values ​​of the normalized features. Specifically, a spatiotemporal data interpolation method based on tensor decomposition can be used to fill the missing values. It can be understood that tensor decomposition is an algorithm that decomposes high-dimensional data (tensors) into low-dimensional data combinations to reduce computational complexity and improve computational efficiency. For a missing value X, the calculation formula for its filling value can be:

[0048]

[0049] in, is the core tensor, (n=1, 2, 3) represents the modal product, n represents the dimension of the modal product, a i 、b j and c are spatial, temporal and feature pattern factors respectively. This method can simultaneously consider the spatial and temporal correlation of feature data and provide more accurate missing value estimation. Furthermore, the alternating least squares method can be used to solve the tensor decomposition problem, and the optimal tensor rank can be selected through cross-validation.

[0050] It should be noted that the execution order of abnormal feature detection and elimination, feature normalization processing, and feature missing value filling can be adjusted according to actual conditions and is not limited to the description order in the above embodiments.

[0051] Step S206, performing dimensionality reduction processing on the regional attribute features whose feature importance reaches the importance threshold to obtain the target attribute features.

[0052] The importance threshold may be a pre-set feature importance threshold, and the regional attributes exceeding the threshold are used as features for dimensionality reduction processing. The target attribute feature may refer to the feature obtained after dimensionality reduction processing, which can be used as input data for subsequent regional clustering.

[0053] Specifically, after calculating the feature importance of each regional attribute feature, the server compares the feature importance with the preset feature importance threshold, and screens out the regional attribute features whose feature importance exceeds the threshold. These regional attribute features can be further processed by dimensionality reduction to reduce the feature dimension, which can not only improve the efficiency of subsequent regional clustering, but also reduce the impact of noise, thereby improving the accuracy of subsequent regional clustering results.

[0054] In one example, the server may normalize the Shapley value of each regional attribute feature, and the expression may be as follows:

[0055]

[0056] in, represents the sum of the original Shapley values ​​of all regional attribute features, p represents the total number of categories of regional attribute features, Represents the normalized Shapley value. This step ensures that the sum of all feature weights is 1, which is convenient for subsequent weighted principal component analysis. After completing the normalization of the Shapley value of each regional attribute feature, the server can use weighted principal component analysis to achieve dimensionality reduction of regional attribute features. The expression of weighted principal component analysis is as follows:

[0057]

[0058] Cv = λv

[0059] Where X represents the feature matrix, represents the transpose of X, W is the diagonal weight matrix, and the diagonal elements are , C represents the weighted covariance matrix, v is the eigenvector, and λ is the corresponding eigenvalue.

[0060] In one example, the feature importance threshold can be set to 85%, and the calculated feature value Arrange in descending order and select the first k principal components so that:

[0061]

[0062] Thus, the reduced-dimensional feature matrix X' is obtained. It can be understood that the feature matrix includes the target attribute features. It should be noted that the calculation of feature importance and feature dimensionality reduction processing are not limited to the order described in the above embodiment, and can be performed simultaneously.

[0063] Step S208: clustering the target areas based on the target attribute characteristics to obtain at least one target area.

[0064] The target area may refer to each cluster area in the target region.

[0065] Specifically, after obtaining the target attribute characteristics, the server can further perform regional clustering on the target area, and divide the target area through regional clustering, thereby effectively improving the efficiency of icing risk assessment.

[0066] In one example, the server can implement regional clustering based on a pre-trained regional clustering model, that is, input the target attribute features into the regional clustering model and output each target region. The regional clustering model can be a model trained based on an improved K-means algorithm. The K-means algorithm, also known as the K-means clustering algorithm, is an iterative clustering analysis algorithm that pre-divides the data into K groups, then randomly selects K objects as the initial cluster centers, and then calculates the distance between each object and each cluster center, assigns each object to the cluster center closest to it, and the cluster centers and the objects assigned to them represent a cluster. The cluster centers of the clusters will be recalculated based on the existing objects in the clusters, and the process is repeated until the convergence conditions are met.

[0067] Step S210: Perform icing risk analysis on each target area through an icing risk assessment model to obtain an icing risk level for each target area.

[0068] Among them, the icing risk assessment model is a hybrid model pre-trained based on graph convolutional networks and long short-term memory networks. Graph Convolutional Networks (GCNs) is a deep learning model for processing graph structured data, which can be used to capture the spatial correlation between features. Long Short-Term Memory (LSTM) is a time recurrent neural network that can be used to capture the temporal correlation between features. In this way, the icing risk assessment model can better capture the interdependence and change patterns of regional attributes in time and space dimensions, thereby improving the accuracy of icing risk assessment. The icing risk level can refer to the level of icing disasters in the target area. In one example, a gradient color spectrum can be used to represent different icing risk levels.

[0069] Specifically, after obtaining each target area obtained by regional clustering, the server can input the regional attribute features of each target area into a hybrid model pre-trained based on a graph convolutional network and a long short-term memory network to analyze the spatial correlation and temporal correlation between the regional attribute features, thereby obtaining the ice risk level of each target area. It should be noted that the regional attribute features refer to the attribute features corresponding to the attribute data of each target area, and the regional attribute features have the same meaning as the regional attribute features.

[0070] In this embodiment, in response to the ice risk assessment instruction for the target area, the server first obtains at least one regional attribute data of the target area, thereby extracting the corresponding regional attribute features. Then, feature importance analysis is performed on each regional attribute feature respectively, thereby obtaining the feature importance of each regional attribute feature, and the regional attribute features whose feature importance reaches the importance threshold are used as the attribute features to be processed for subsequent dimensionality reduction processing. By screening out important regional attribute features, the accuracy of subsequent regional clustering can be effectively improved. Then, the attribute features to be processed are subjected to dimensionality reduction processing to obtain target attribute features. Dimensionality reduction processing can effectively reduce the dimension of feature data, which can not only improve the efficiency of subsequent regional clustering, but also reduce the influence of noise and improve the accuracy of regional clustering. Then, accurate regional clustering is performed on the target area. Finally, through the ice risk assessment model, ice risk analysis is performed on each target area obtained by clustering, and the ice risk level of each target area is obtained. Since the ice risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network, it can better capture the interdependence and change law of regional attribute data in time and space dimensions, thereby improving the accuracy of ice risk assessment.

[0071] In an exemplary embodiment, as shown in FIG3 , an icing risk assessment model is used to perform an icing risk analysis on each target area to obtain an icing risk level for each target area, including:

[0072] Step S302: Perform spatial correlation analysis on the regional attribute features of each target area through the graph convolutional network in the ice risk assessment model to obtain the spatial correlation analysis results of each target area.

[0073] Among them, the results of spatial correlation analysis are used to characterize the correlation between the attribute characteristics of each area in the target area in the spatial dimension.

[0074] Specifically, for each target area, the server inputs the regional attribute features of each target area into the graph convolutional network in the icing risk assessment model. The graph convolutional network captures the correlation of the regional attribute features in the spatial dimension by aggregating the neighbor node information of the node (regional attribute features).

[0075] In one example, the expression of the icing risk assessment model is:

[0076]

[0077] Among them, X is the input regional attribute feature, A is the adjacency matrix, , , Represent the GCN, LSTM and multi-layer perceptron layers respectively.

[0078] The calculation formula for the neighbor node information of the graph convolutional network aggregation node (regional attribute feature) is as follows:

[0079]

[0080] in, is an adjacency matrix with self-connection, yes The degree matrix of represents the feature matrix of the lth layer, is a learnable weight matrix, is an activation function such as the ReLU function. In this way, the node features are updated at each layer according to the features of its neighboring nodes.

[0081] Step S304: Perform time correlation analysis on the regional attribute characteristics of each target area through the long short-term memory network in the ice risk assessment model to obtain the time correlation analysis result of each target area.

[0082] Among them, the results of spatial correlation analysis are used to characterize the correlation between the attribute characteristics of each area in the target area in the time dimension.

[0083] Specifically, for each target area, the server inputs the attribute characteristics of each target area into the long short-term memory network in the icing risk assessment model. The long short-term memory network captures the correlation of the attribute characteristics of each area in the time dimension through a gating mechanism.

[0084] In one example, the long short-term memory network introduces three key gating units, namely the forget gate, input gate, and output gate. These gating units control the flow of information so that the network can better capture the long-term dependencies between features. The expressions are as follows:

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] in, represents the output of the forget gate at time step t, represents the output of the input gate at time step t, represents the output of the output gate at time step t, represents the input features at time step t, Both represent weight parameters, Both represent bias parameters, is the hidden state at time step t, is the hidden state at the previous time step. represents the output of the candidate memory cell at time step t, is the hyperbolic tangent activation function, is the output of the memory element at time step t. It should be noted that the forget gate is responsible for deciding which information in the memory state of the previous time step needs to be forgotten, through an activation function () outputs a value between 0 and 1, indicating whether the information at the corresponding position should be retained or forgotten. The ‌Input Gate‌ determines which part of the input information at the current time step should be stored in the memory state. Similar to the forget gate, the input gate also passes through the activation function () outputs a value between 0 and 1, which is used to control the input of information. The output gate determines what the output of the current time step is, which is activated by the function () outputs a value between 0 and 1, controlling which information in the memory state is mapped to the output. Through the gating mechanism, the long short-term memory network is able to selectively retain or forget information when processing sequence data, thereby effectively capturing and processing long-term dependencies in feature sequences.

[0092] Step S306: Based on the spatial correlation analysis results and the temporal correlation analysis results, an icing risk level prediction is performed for each target area to obtain an icing risk level for each target area.

[0093] Specifically, the icing risk assessment model also includes a prediction layer. The spatial correlation analysis results output by the graph convolutional network and the temporal correlation analysis results output by the long short-term memory network are input into the prediction layer. Through the prediction layer, the icing risk level of each target area is predicted, so as to obtain the icing risk level of each target area.

[0094] The prediction expression of the prediction layer is as follows:

[0095]

[0096] in, Indicates the risk level, is the probability distribution predicted by the model, It represents the spatial correlation analysis results of the graph convolutional network and the temporal correlation analysis results of the long short-term memory network.

[0097] In one example, during the training process of the ice risk assessment model, a weighted cross entropy loss function can be used to deal with the category imbalance problem of regional attribute features. The expression of the weighted cross entropy loss function is as follows:

[0098]

[0099] in, represents the weight of the regional attribute feature category c, is the true label, is the prediction probability. When calculating the loss function, the loss value of each feature will be adjusted with the weight of the category, so as to achieve weighted processing of different categories, thereby adjusting the importance between different categories and improving the accuracy of the model.

[0100] In one example, during the training process of the ice risk assessment model, the Adam (Adaptive Moment Estimation) optimizer can also be used for model training. The learning rate can be calculated using the cosine annealing strategy. The calculation expression of the learning rate is as follows:

[0101]

[0102] in, is the learning rate at time t, and are the minimum and maximum learning rates respectively, and T is the total number of training steps.

[0103] In this embodiment, a hybrid model trained by a graph convolutional network and a long short-term memory network is used to implement icing risk assessment for each target area, thereby improving the accuracy of icing risk assessment.

[0104] In an exemplary embodiment, as shown in FIG4 , the training of the regional clustering model further includes:

[0105] Step S402: acquiring at least one sample attribute data of the target area, performing data processing on each sample attribute data, and obtaining the sample attribute characteristics of the target area.

[0106] The target area may include at least one sample sampling point, the sample attribute data may be the regional sample attribute data collected by each sample sampling point, and the sample attribute feature may refer to the attribute feature corresponding to the regional sample attribute data.

[0107] Specifically, before training the regional clustering model, the server needs to collect training samples first, that is, for each sample sampling point in the target area, the server collects the regional sample attribute data at each sample sampling point through a multi-source data acquisition system, and extracts the corresponding attribute features, and performs a series of data processing on these attribute features to obtain the sample attribute features of the target area. It can be understood that the data processing in this embodiment is the same as the abnormal feature detection and elimination, feature normalization processing, and feature missing value filling in step S204, and will not be repeated here.

[0108] Step S404: input the sample attribute features into the initial regional clustering model, and through the initial regional clustering model, based on the density peak method, select the initial cluster center that matches the optimal cluster number from each sample cluster point.

[0109] Among them, the initial regional clustering model is a model established based on the K-means algorithm. The target area includes at least one sample clustering point for training the initial regional clustering model. The Density Peak Clustering Algorithm (DPC) is an unsupervised algorithm that can find the density peak points in the data set. In this embodiment, the density peak method is used to determine the initial clustering center. The optimal number of clusters refers to the number of clusters. The initial clustering center corresponds to the optimal number of clusters. For example, if the optimal number of clusters is 5, then 5 initial clustering centers need to be selected.

[0110] Specifically, the server inputs the sample attribute features into the initial regional clustering model, and the initial regional clustering model can first calculate the local density of each feature point i Distance to points with high local density , the calculation expression is as follows:

[0111] r i = ∑ j x d ij - d c ] [ d i = minutes j: r j > r i d ij

[0112] Among them, if ,but ,otherwise . For the cutoff distance, the 1-2% quantile of all point pair distances in the data set is usually selected as the cutoff distance. is the feature point i and the feature point The Euclidean distance between .

[0113] Then calculate the comprehensive score of each feature point: ,choose The K points with the largest values ​​are used as the initial cluster centers. In this way, points with high density and far distance from each other can be selected as the initial cluster centers to improve the clustering effect.

[0114] Step S406: for the other cluster points in each sample cluster point except the initial cluster center, determine the Mahalanobis distance between the other cluster points and the initial cluster center.

[0115] It should be noted that in order to better capture the data distribution in the feature space, the weighted Mahalanobis distance is introduced as a similarity metric in this embodiment. Compared with the Euclidean distance, the Mahalanobis distance corrects the problem of inconsistent and related scales of various dimensions in the Euclidean distance, and can improve clustering accuracy. The expression of the Mahalanobis distance is as follows:

[0116]

[0117] Where W is the diagonal weight matrix, S k is the covariance matrix of the attribute features of the k-th class of samples.

[0118] Specifically, after the server selects the initial cluster center from each sample cluster point, it further calculates the Mahalanobis distance between other cluster points and the initial cluster center to facilitate subsequent clustering.

[0119] Step S408: When the Mahalanobis distance satisfies the clustering condition, clustering other cluster points with the initial cluster center to obtain a first clustering result.

[0120] The first clustering result may refer to a cluster group formed by other cluster points and the initial cluster center. The clustering condition may refer to the shortest Mahalanobis distance or the Mahalanobis distance resisting a preset distance threshold. In practical applications, appropriate clustering conditions may be set according to actual conditions.

[0121] Specifically, after calculating the Mahalanobis distances between other cluster points and each initial cluster center, the server allocates other cluster centers with the smallest Mahalanobis distances to the corresponding initial cluster centers, thereby forming various cluster families.

[0122] Step S410: Based on the first clustering result, the initial regional clustering model is trained until the first clustering result meets the convergence condition, thereby obtaining the regional clustering model.

[0123] The convergence condition may refer to the convergence of the objective function or the model reaching the number of iterations. The objective function may be the sum of the squares of the Mahalanobis distances in each clustering family. The expression of the objective function is as follows:

[0124]

[0125] in, represents the kth initial cluster center.

[0126] Specifically, after obtaining each cluster group, the server calculates the sum of the squares of the Mahalanobis distances between the cluster points and the cluster center in each cluster group, and counts the sum of the squares of the Mahalanobis distances of all cluster groups. When convergence is achieved, the training is stopped to obtain a regional clustering model.

[0127] In one embodiment, when the first clustering result does not meet the convergence condition, based on the first clustering result, each initial cluster center is updated in parallel to obtain an updated cluster center; the updated cluster center is used as the initial cluster center, and the step is returned to: for other cluster points in each sample cluster point except the initial cluster center, the Mahalanobis distance between the other cluster points and the initial cluster center is determined.

[0128] Specifically, when the first clustering result does not meet the convergence condition, that is, the objective function has not converged or the model has not reached the number of iterations, the cluster center needs to be updated. The cluster center update can be performed in parallel, that is, a distributed parallel computing architecture is used to update each cluster center in parallel to improve the efficiency of model training. That is, the server reselects the cluster center, obtains a new cluster center, and recalculates the Mahalanobis distance between other cluster points and the new cluster center. In one example, the expression for cluster center update is as follows:

[0129]

[0130] in, represents the new cluster center, Represents the initial cluster center.

[0131] In this embodiment, the regional clustering model is obtained by training with an improved K-means algorithm, which improves the accuracy of the regional clustering results, thereby improving the accuracy of the icing risk assessment.

[0132] In an exemplary embodiment, as shown in FIG5 , the process of determining the optimal number of clusters includes:

[0133] Step S502: for each candidate cluster number, the sample attribute data is input into the initial regional clustering model to perform clustering to obtain a second clustering result matching each candidate cluster number.

[0134] It is understandable that in the process of regional clustering, the number of clusters is also one of the important factors affecting the clustering results. Choosing an appropriate number of clusters can greatly improve the accuracy of clustering. The second clustering result refers to the sum of squared Mahalanobis distances in the cluster family. It should be noted that the difference between the second clustering result and the first clustering result is that the second clustering result is used to update the number of clusters, and the first clustering result is used to update the cluster center, and the cluster center needs to be determined after the number of clusters is determined. The candidate number of clusters can be a pre-set random number of clusters.

[0135] Specifically, the server inputs the sample attribute data into the initial regional clustering model based on the candidate cluster data for each set candidate cluster number, and obtains each cluster group output by the model. For example, assuming that the number of candidate clusters is 5, the model can output 5 cluster groups, and the second clustering result can be the sum of squared Mahalanobis distances within the 5 cluster groups.

[0136] Step S504: obtaining the difference between the second clustering result and the expected value of the clustering result, and taking the candidate cluster number with the largest difference as the optimal cluster number.

[0137] The expected value of the clustering result may be the expected value of the sum of squares of Mahalanobis distances in the clustering family set according to the distribution of reference data or standard data.

[0138] Specifically, after calculating the sum of squares of Mahalanobis distances within each cluster, the server calculates the difference between the sum of squares and a preset expected value, and selects the candidate cluster number with the largest difference as the optimal cluster number. In one example, the calculation expression for the difference between the second clustering result and the expected value of the clustering result is:

[0139]

[0140] in, The difference between the second clustering result of K clusters (number of candidate clusters K) and the expected value of the clustering result, represents the expected value of clustering results, Represents the second clustering result.

[0141] In this embodiment, by selecting the optimal number of clusters, the efficiency and accuracy of clustering can be effectively improved, thereby improving the efficiency and accuracy of icing risk assessment.

[0142] In a specific embodiment, as shown in FIG6 , the ice coverage level analysis method based on ice area division further includes the following steps:

[0143] S1: In response to the instruction of ice risk assessment for complex mountainous areas, high-precision digital elevation model data, meteorological data, hydrological data and tower data of complex mountainous areas are collected through the multi-source data collection system. Features are extracted from these data to obtain corresponding regional attribute features.

[0144] S2: Calculate the Shapley value of each regional attribute feature to represent the importance of each regional attribute feature.

[0145] S3: Based on the weighted principal component analysis method, the regional attribute features whose feature importance reaches the importance threshold are subjected to dimensionality reduction processing to obtain temporary attribute features, and the temporary attribute features are subjected to feature fusion to obtain the target attribute features.

[0146] S4: Input the target attribute features into the regional clustering model trained based on the improved K-means algorithm, perform regional clustering on the target area, and obtain at least one target area.

[0147] As shown in FIG7 , the training process of the regional clustering model includes: obtaining at least one sample attribute data of the target area, performing data processing on each sample attribute data, and obtaining the sample attribute characteristics of the target area. The sample attribute characteristics are input into the initial regional clustering model, and the initial cluster center matching the optimal number of clusters is screened out from each sample cluster point based on the density peak method through the initial regional clustering model. For the other cluster points in each sample cluster point except the initial cluster center, the Mahalanobis distance between the other cluster points and the initial cluster center is determined. When the Mahalanobis distance meets the clustering condition, the other cluster points are clustered with the initial cluster center to obtain the first clustering result. Based on the first clustering result, the initial regional clustering model is trained until the first clustering result meets the convergence condition to obtain the regional clustering model. When the first clustering result does not meet the convergence condition, based on the first clustering result, each initial cluster center is updated in parallel to obtain an updated cluster center; the updated cluster center is used as the initial cluster center, and the step is returned to: for other cluster points in each sample cluster point except the initial cluster center, the Mahalanobis distance between other cluster points and the initial cluster center is determined.

[0148] The updating process of the optimal number of clusters includes: for each candidate number of clusters, inputting the sample attribute data into the initial regional clustering model, performing clustering, and obtaining a second clustering result that matches each candidate number of clusters. Obtaining the difference between the second clustering result and the expected value of the clustering result, and taking the candidate number of clusters with the largest difference as the optimal number of clusters.

[0149] S5: Through the graph convolutional network in the icing risk assessment model, the spatial correlation analysis of the regional attribute characteristics of each target area is performed respectively, and the spatial correlation analysis results of each target area are obtained. Through the long short-term memory network in the icing risk assessment model, the temporal correlation analysis of the regional attribute characteristics of each target area is performed respectively, and the temporal correlation analysis results of each target area are obtained. Based on the spatial correlation analysis results and the temporal correlation analysis results, the icing risk level of each target area is predicted to obtain the icing risk level of each target area.

[0150] In this embodiment, in response to the ice risk assessment instruction for the target area, at least one regional attribute data of the target area is first obtained to extract the corresponding regional attribute features. Then, feature importance analysis is performed on each regional attribute feature to obtain the feature importance of each regional attribute feature, and the regional attribute features whose feature importance reaches the importance threshold are used as the attribute features to be processed for subsequent dimensionality reduction processing. By screening out important regional attribute features, the accuracy of subsequent regional clustering can be effectively improved. Then, the attribute features to be processed are subjected to dimensionality reduction processing to obtain target attribute features. Dimensionality reduction processing can effectively reduce the dimension of feature data, which can not only improve the efficiency of subsequent regional clustering, but also reduce the influence of noise and improve the accuracy of regional clustering. Then, accurate regional clustering is performed on the target area. Finally, through the ice risk assessment model, ice risk analysis is performed on each target area obtained by clustering to obtain the ice risk level of each target area. Since the ice risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network, it can better capture the interdependence and change law of regional attribute data in time and space dimensions, thereby improving the accuracy of ice risk assessment.

[0151] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0152] Based on the same inventive concept, the embodiment of the present application also provides an ice coverage level analysis device based on ice zone division for implementing the ice coverage level analysis method based on ice zone division involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations of one or more embodiments of the ice coverage level analysis device based on ice zone division provided below can refer to the limitations of the ice coverage level analysis method based on ice zone division above, and will not be repeated here.

[0153] In an exemplary embodiment, as shown in FIG8 , an ice level analysis device based on ice area division is provided, including: a data acquisition module 802, for acquiring at least one regional attribute data of a target area in response to an ice risk assessment instruction for a target area; a feature importance analysis module 804, for performing feature importance analysis on the respective regional attribute features of each regional attribute data, and obtaining the respective feature importance of each regional attribute feature; a feature dimension reduction processing module 806, for performing dimension reduction processing on the regional attribute features whose feature importance reaches the importance threshold, and obtaining the target attribute features; a regional clustering module 808, for performing regional clustering on the target area based on the target attribute features, and obtaining at least one target area; an ice level analysis module 810, for performing ice risk analysis on each target area through an ice risk assessment model, and obtaining the respective ice risk level of each target area; the ice risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0154] In one of the embodiments, the icing level analysis module 810 is also used to perform spatial correlation analysis on the regional attribute characteristics of each target area through the graph convolutional network in the icing risk assessment model, so as to obtain the spatial correlation analysis results of each target area; to perform temporal correlation analysis on the regional attribute characteristics of each target area through the long short-term memory network in the icing risk assessment model, so as to obtain the temporal correlation analysis results of each target area; and to predict the icing risk level of each target area based on the spatial correlation analysis results and the temporal correlation analysis results, so as to obtain the icing risk level of each target area.

[0155] In one of the embodiments, the ice coverage level analysis device based on ice zone division is also used to obtain at least one sample attribute data of the target area, perform data processing on each sample attribute data, and obtain sample attribute characteristics of the target area; input the sample attribute characteristics into the initial regional clustering model, and through the initial regional clustering model, based on the density peak method, screen out the initial cluster center that matches the optimal number of clusters from each sample cluster point; for other cluster points in each sample cluster point except the initial cluster center, determine the Mahalanobis distance between other cluster points and the initial cluster center; when the Mahalanobis distance meets the clustering condition, cluster the other cluster points with the initial cluster center to obtain a first clustering result; based on the first clustering result, train the initial regional clustering model until the first clustering result meets the convergence condition to obtain a regional clustering model; the regional clustering module 808 is also used to input the target attribute characteristics into the regional clustering model, perform regional clustering on the target area, and obtain at least one target area.

[0156] In one of the embodiments, the ice coverage level analysis device based on ice zone division is also used to input sample attribute data into the initial regional clustering model for clustering for each candidate cluster number, to obtain a second clustering result matching each candidate cluster number; obtain the difference between the second clustering result and the expected value of the clustering result, and take the candidate cluster number with the largest difference as the optimal cluster number.

[0157] In one of the embodiments, the ice coverage level analysis device based on ice zone division is also used to update each initial cluster center in parallel based on the first cluster result to obtain an updated cluster center when the first cluster result does not meet the convergence condition; use the updated cluster center as the initial cluster center, and return to the step: for other cluster points in each sample cluster point except the initial cluster center, determine the Mahalanobis distance between other cluster points and the initial cluster center.

[0158] In one of the embodiments, the feature dimension reduction processing module 806 is also used to perform dimension reduction processing on the regional attribute features whose feature importance reaches the importance threshold to obtain temporary attribute features; the temporary attribute features include at least two features of the terrain features, meteorological features, hydrological features and tower features of the target area; and the temporary attribute features are feature fused to obtain the target attribute features.

[0159] Each module in the ice coverage level analysis device based on ice zone division can be implemented in whole or in part by software, hardware and a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0160] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG9. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store ice risk assessment data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an ice level analysis method based on ice area division is implemented.

[0161] Those skilled in the art will understand that the structure shown in FIG. 9 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0162] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the following steps when executing the computer program: in response to an instruction for assessing icing risk for a target area, obtaining at least one regional attribute data of the target area; performing feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; performing dimensionality reduction processing on the regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features; based on the target attribute features, performing regional clustering on the target area to obtain at least one target area; performing icing risk analysis on each target area through an icing risk assessment model to obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0163] In one embodiment, when the processor executes the computer program, the following steps are also implemented: through the graph convolutional network in the icing risk assessment model, spatial correlation analysis is performed on the regional attribute characteristics of each target area, and the spatial correlation analysis results of each target area are obtained; through the long short-term memory network in the icing risk assessment model, time correlation analysis is performed on the regional attribute characteristics of each target area, and the time correlation analysis results of each target area are obtained; based on the spatial correlation analysis results and the time correlation analysis results, the icing risk level of each target area is predicted to obtain the icing risk level of each target area.

[0164] In one embodiment, when the processor executes the computer program, the following steps are also implemented: obtaining at least one sample attribute data of the target area, performing data processing on each sample attribute data, and obtaining the sample attribute characteristics of the target area; inputting the sample attribute characteristics into the initial regional clustering model, and through the initial regional clustering model, based on the density peak method, screening out the initial cluster center that matches the optimal number of clusters from each sample cluster point; for other cluster points in each sample cluster point except the initial cluster center, determining the Mahalanobis distance between other cluster points and the initial cluster center; when the Mahalanobis distance meets the clustering condition, clustering the other cluster points with the initial cluster center to obtain a first clustering result; based on the first clustering result, training the initial regional clustering model until the first clustering result meets the convergence condition to obtain a regional clustering model; based on the target attribute characteristics, performing regional clustering on the target area to obtain at least one target area, including: inputting the target attribute characteristics into the regional clustering model, performing regional clustering on the target area to obtain at least one target area.

[0165] In one embodiment, when the processor executes the computer program, the following steps are also implemented: for each candidate number of clusters, the sample attribute data is input into the initial regional clustering model to perform clustering to obtain a second clustering result that matches each candidate number of clusters; the difference between the second clustering result and the expected value of the clustering result is obtained, and the candidate number of clusters with the largest difference is used as the optimal number of clusters.

[0166] In one embodiment, the processor further implements the following steps when executing the computer program: when the first clustering result does not meet the convergence condition, based on the first clustering result, each initial cluster center is updated in parallel to obtain an updated cluster center; the updated cluster center is used as the initial cluster center, and the process returns to the step of: for each sample cluster point except the initial cluster center, determining the Mahalanobis distance between the other cluster points and the initial cluster center.

[0167] In one embodiment, when the processor executes the computer program, the following steps are also implemented: dimensionality reduction processing is performed on the regional attribute features whose feature importance reaches the importance threshold to obtain temporary attribute features; the temporary attribute features include at least two features of the terrain features, meteorological features, hydrological features and tower features of the target area; and feature fusion is performed on each temporary attribute feature to obtain the target attribute feature.

[0168] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: in response to an instruction for icing risk assessment for a target area, obtaining at least one regional attribute data of the target area; performing feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; performing dimensionality reduction processing on the regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features; based on the target attribute features, performing regional clustering on the target area to obtain at least one target area; performing icing risk analysis on each target area through an icing risk assessment model to obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0169] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: through the graph convolutional network in the icing risk assessment model, spatial correlation analysis is performed on the regional attribute characteristics of each target area, and a spatial correlation analysis result of each target area is obtained; through the long short-term memory network in the icing risk assessment model, temporal correlation analysis is performed on the regional attribute characteristics of each target area, and a temporal correlation analysis result of each target area is obtained; based on the spatial correlation analysis results and the temporal correlation analysis results, the icing risk level of each target area is predicted to obtain an icing risk level of each target area.

[0170] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining at least one sample attribute data of the target area, performing data processing on each sample attribute data, and obtaining the sample attribute characteristics of the target area; inputting the sample attribute characteristics into the initial regional clustering model, and through the initial regional clustering model, based on the density peak method, screening out the initial cluster center that matches the optimal number of clusters from each sample cluster point; for other cluster points in each sample cluster point except the initial cluster center, determining the Mahalanobis distance between other cluster points and the initial cluster center; when the Mahalanobis distance meets the clustering condition, clustering the other cluster points with the initial cluster center to obtain a first clustering result; based on the first clustering result, training the initial regional clustering model until the first clustering result meets the convergence condition to obtain a regional clustering model; based on the target attribute characteristics, performing regional clustering on the target area to obtain at least one target area, including: inputting the target attribute characteristics into the regional clustering model, performing regional clustering on the target area to obtain at least one target area.

[0171] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: for each candidate number of clusters, the sample attribute data is input into the initial regional clustering model to perform clustering to obtain a second clustering result that matches each candidate number of clusters; the difference between the second clustering result and the expected value of the clustering result is obtained, and the candidate number of clusters with the largest difference is used as the optimal number of clusters.

[0172] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: when the first clustering result does not meet the convergence condition, based on the first clustering result, each initial cluster center is updated in parallel to obtain an updated cluster center; the updated cluster center is used as the initial cluster center, and the step of returning to: for other cluster points in each sample cluster point except the initial cluster center, the Mahalanobis distance between other cluster points and the initial cluster center is determined.

[0173] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: dimensionality reduction processing is performed on the regional attribute features whose feature importance reaches the importance threshold to obtain temporary attribute features; the temporary attribute features include at least two features of the terrain features, meteorological features, hydrological features and tower features of the target area; and feature fusion is performed on each temporary attribute feature to obtain the target attribute feature.

[0174] In one embodiment, a computer program product is provided, including a computer program, which implements the following steps when executed by a processor: in response to an instruction for icing risk assessment for a target area, obtaining at least one regional attribute data of the target area; performing feature importance analysis on the regional attribute features of each regional attribute data respectively to obtain the feature importance of each regional attribute feature; performing dimensionality reduction processing on the regional attribute features whose feature importance reaches an importance threshold to obtain target attribute features; based on the target attribute features, performing regional clustering on the target area to obtain at least one target area; performing icing risk analysis on each target area through an icing risk assessment model to obtain the icing risk level of each target area; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

[0175] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: through the graph convolutional network in the icing risk assessment model, spatial correlation analysis is performed on the regional attribute characteristics of each target area, and a spatial correlation analysis result of each target area is obtained; through the long short-term memory network in the icing risk assessment model, temporal correlation analysis is performed on the regional attribute characteristics of each target area, and a temporal correlation analysis result of each target area is obtained; based on the spatial correlation analysis results and the temporal correlation analysis results, the icing risk level of each target area is predicted to obtain an icing risk level of each target area.

[0176] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining at least one sample attribute data of the target area, performing data processing on each sample attribute data, and obtaining the sample attribute characteristics of the target area; inputting the sample attribute characteristics into the initial regional clustering model, and through the initial regional clustering model, based on the density peak method, screening out the initial cluster center that matches the optimal number of clusters from each sample cluster point; for other cluster points in each sample cluster point except the initial cluster center, determining the Mahalanobis distance between other cluster points and the initial cluster center; when the Mahalanobis distance meets the clustering condition, clustering the other cluster points with the initial cluster center to obtain a first clustering result; based on the first clustering result, training the initial regional clustering model until the first clustering result meets the convergence condition to obtain a regional clustering model; based on the target attribute characteristics, performing regional clustering on the target area to obtain at least one target area, including: inputting the target attribute characteristics into the regional clustering model, performing regional clustering on the target area to obtain at least one target area.

[0177] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: for each candidate number of clusters, the sample attribute data is input into the initial regional clustering model to perform clustering to obtain a second clustering result that matches each candidate number of clusters; the difference between the second clustering result and the expected value of the clustering result is obtained, and the candidate number of clusters with the largest difference is used as the optimal number of clusters.

[0178] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: when the first clustering result does not meet the convergence condition, based on the first clustering result, each initial cluster center is updated in parallel to obtain an updated cluster center; the updated cluster center is used as the initial cluster center, and the step of returning to: for other cluster points in each sample cluster point except the initial cluster center, the Mahalanobis distance between other cluster points and the initial cluster center is determined.

[0179] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: dimensionality reduction processing is performed on the regional attribute features whose feature importance reaches the importance threshold to obtain temporary attribute features; the temporary attribute features include at least two features of the terrain features, meteorological features, hydrological features and tower features of the target area; and feature fusion is performed on each temporary attribute feature to obtain the target attribute feature.

[0180] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0181] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0182] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0183] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for analyzing ice coverage levels based on ice zone division. It is characterized in that The method comprises: In response to an instruction for assessing ice risk in a target area, obtaining at least one area attribute data of the target area; Performing feature importance analysis on each of the regional attribute features of the regional attribute data to obtain the feature importance of each of the regional attribute features; Performing dimensionality reduction processing on the regional attribute features whose feature importance reaches the importance threshold to obtain the target attribute features; Based on the target attribute characteristics, the target area is clustered to obtain at least one target area; Through the icing risk assessment model, an icing risk analysis is performed on each of the target areas to obtain the icing risk level of each of the target areas; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

2. The method according to claim 1, It is characterized in that The icing risk assessment model is used to perform icing risk analysis on each target area to obtain an icing risk level for each target area, including: Through the graph convolutional network in the ice risk assessment model, spatial correlation analysis is performed on the regional attribute characteristics of each target area to obtain the spatial correlation analysis results of each target area; Through the long short-term memory network in the ice risk assessment model, the time correlation analysis is performed on the attribute characteristics of each area of ​​each target area, and the time correlation analysis results of each target area are obtained; Based on the spatial correlation analysis result and the temporal correlation analysis result, an icing risk level prediction is performed on each of the target areas to obtain an icing risk level of each of the target areas.

3. The method according to claim 1, It is characterized in that The target area includes at least one sample cluster point for training an initial regional clustering model; the method further includes: Acquire at least one sample attribute data of a target area, perform data processing on each sample attribute data, and obtain the sample attribute characteristics of the target area; Inputting the sample attribute characteristics into the initial regional clustering model, and using the initial regional clustering model, based on the density peak method, selecting the initial clustering center matching the optimal clustering number from each of the sample clustering points; For other clustering points in each of the sample clustering points except the initial clustering center, determining the Mahalanobis distance between the other clustering points and the initial clustering center; When the Mahalanobis distance satisfies the clustering condition, clustering the other clustering points with the initial clustering center to obtain a first clustering result; Based on the first clustering result, the initial regional clustering model is trained until the first clustering result meets a convergence condition to obtain a regional clustering model; The step of clustering the target area based on the target attribute characteristics to obtain at least one target area includes: The target attribute characteristics are input into the regional clustering model, and the target area is regionally clustered to obtain at least one target area.

4. The method according to claim 3, It is characterized in that The process of determining the optimal number of clusters includes: For each candidate cluster number, the sample attribute data is input into the initial regional clustering model to perform clustering to obtain a second clustering result matching each candidate cluster number; The difference between the second clustering result and the expected value of the clustering result is obtained, and the number of candidate clusters with the largest difference is used as the optimal number of clusters.

5. The method according to claim 3, It is characterized in that The method further comprises: When the first clustering result does not meet the convergence condition, based on the first clustering result, each of the initial cluster centers is updated in parallel to obtain an updated cluster center; The updated cluster center is used as the initial cluster center, and the process returns to the step of: for each of the sample cluster points except the initial cluster center, determining the Mahalanobis distance between the other cluster points and the initial cluster center.

6. The method according to claim 1, It is characterized in that The dimension reduction process is performed on the regional attribute features whose feature importance reaches the importance threshold to obtain the target attribute features, including: Performing dimensionality reduction processing on the regional attribute features whose feature importance reaches the importance threshold to obtain temporary attribute features; the temporary attribute features include at least two features of the terrain features, meteorological features, hydrological features and tower features of the target area; The temporary attribute features are fused to obtain the target attribute features.

7. An ice coverage level analysis device based on ice zone division, It is characterized in that The device comprises: A data acquisition module, configured to acquire at least one area attribute data of a target area in response to an ice risk assessment instruction for the target area; A feature importance analysis module is used to perform feature importance analysis on each of the regional attribute features of the regional attribute data to obtain the feature importance of each of the regional attribute features; A feature dimension reduction processing module is used to perform dimension reduction processing on the regional attribute features whose feature importance reaches the importance threshold to obtain the target attribute features; A regional clustering module, used for performing regional clustering on the target area based on the target attribute characteristics to obtain at least one target area; The icing level analysis module is used to perform icing risk analysis on each of the target areas through an icing risk assessment model to obtain the icing risk level of each of the target areas; the icing risk assessment model is a hybrid model trained based on a graph convolutional network and a long short-term memory network.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images