Method for identifying visual alarm high-incidence area of power transmission line channel

Abstract

The invention relates to the field of power transmission line risk hidden danger monitoring, and provides a power transmission line channel visual alarm high-incidence area identification method, and the method comprises the steps: collecting visible light video streams, infrared thermal imaging, environment data, GPS positioning information and other multi-source data, carrying out the preprocessing and multi-modal feature extraction, and constructing a multi-modal feature matrix; based on correlation analysis between features, generating a high-dimensional alarm event feature vector, mapping the high-dimensional alarm event feature vector into a graph node, and constructing a space-time diagram containing spatial distance and time interval information; spatial and temporal distribution characteristics are extracted, clustering analysis is carried out, alarm event clusters are identified, risk scoring is carried out in combination with density and persistence indexes, and finally an alarm high-incidence area with risk rating is calibrated, so that risk assessment based on data driving is realized instead of fixed threshold value alarm, the influence of environmental change on alarm frequency can be adapted, and the risk assessment efficiency is improved. Complete process optimization from data acquisition to risk rating is realized, and the accuracy and comprehensiveness of alarm high-incidence area identification are improved.

Description

Technical Field

[0001] This invention relates to the field of monitoring risks and hazards in power transmission lines, and in particular to a method for identifying high-incidence areas of visual alarms in power transmission line corridors. Background Technology

[0002] In recent years, with the continuous expansion of the power grid and the improvement of its intelligence level, the operation and maintenance management of transmission line corridors has gradually become a key link in ensuring the safe and stable operation of the energy network. Traditional transmission line inspections mainly rely on regular manual inspections and helicopter aerial photography, which suffers from low efficiency, poor real-time performance, and limited coverage. To address these challenges, corridor visualization monitoring technology based on image recognition and sensor fusion has gradually emerged. By deploying high-definition cameras, infrared thermal imagers, and environmental sensors, real-time data acquisition and abnormal condition monitoring of the transmission line corridors can be achieved.

[0003] The main shortcomings of existing technologies are as follows: First, traditional machine learning algorithms (such as decision trees and support vector machines) exhibit significant limitations when processing high-dimensional, nonlinear spatiotemporal data, failing to effectively capture the periodic characteristics of alarm events in the time dimension and the clustering effect in the spatial dimension, resulting in insufficient accuracy and robustness in identifying high-incidence areas. Second, existing research focuses on the independent analysis of single-type alarms (such as smoke and fire alarms or external damage alarms), neglecting the potential correlation between multi-source alarm events (such as the coupling relationship between tree obstacles and smoke and fire alarms), resulting in fragmented analysis of the causes of high-incidence alarm areas and making it difficult to form a global risk prevention and control strategy.

[0004] Therefore, this invention proposes a method for identifying high-incidence areas of visual alarms in power transmission line channels to solve the problems existing in the prior art. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for identifying high-incidence areas of visual alarms in power transmission line channels, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Firstly, this application provides a method for identifying high-incidence areas of visual alarms in power transmission line corridors, comprising the following steps:

[0008] S1. Collect multi-source data and preprocess it. Extract multi-modal features based on the preprocessed multi-source data and construct a multi-modal feature matrix.

[0009] S2. Based on the multimodal feature matrix, obtain the linear correlation between multimodal features and generate a high-dimensional alarm event feature vector;

[0010] S3. Convert the high-dimensional alarm event feature vector into graph nodes, calculate the spatial distance and time interval of each graph node, construct weighted adjacency edges between graph nodes, and generate a spatiotemporal graph.

[0011] S4. Extract the spatiotemporal distribution features of the spatiotemporal map and obtain the alarm event clusters of the spatiotemporal map;

[0012] S5. Calculate the risk score of each cluster based on alarm event clustering, and identify high-incidence areas with risk ratings by combining spatiotemporal distribution characteristics.

[0013] Preferably, the multi-source data in S1 includes visible light video streams, infrared thermal imaging data, environmental data, and GPS positioning information; the preprocessing in S1 includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

[0014] Preferably, the specific steps for extracting multimodal features and constructing a multimodal feature matrix based on preprocessed multi-source data are as follows:

[0015] Based on visible light video stream, alarm targets are identified, and bounding box coordinates and confidence scores are extracted as visual features;

[0016] Based on infrared thermal imaging data, thermal characteristics of temperature anomaly regions are extracted;

[0017] Based on environmental data, extract environmental features;

[0018] By combining GPS positioning coordinate information, multimodal features are spatiotemporally aligned and fused to construct a multimodal feature matrix.

[0019] Preferably, S2 specifically includes:

[0020] S2.1 Based on the multimodal feature matrix, calculate the maximum correlation coefficient between multimodal features and obtain the linear correlation between multimodal features;

[0021] S2.2 Calculate the fusion contribution between multimodal features and combine linear correlation to generate a high-dimensional alarm event feature vector.

[0022] Preferably, S3 specifically includes:

[0023] S3.1 Map the high-dimensional alarm event feature vector to graph nodes, calculate the spatial distance between graph nodes, and obtain the time interval between graph nodes based on the timestamp information of the graph nodes;

[0024] S3.2 Calculate the comprehensive spatiotemporal similarity score between graph nodes based on the spatial distance and time interval of each graph node;

[0025] S3.3. Based on the comprehensive spatiotemporal similarity score, obtain the adjacent nodes of each graph node, and use the comprehensive spatiotemporal similarity score as the edge weight to construct a spatiotemporal graph.

[0026] Preferably, in step S3.1, the high-dimensional alarm event feature vector is mapped to graph nodes using a Transformer encoder. At the same time, the spatial distance between graph nodes is obtained using the Euclidean distance formula, and the time interval between graph nodes is obtained using the time difference method based on the timestamp information of the graph nodes.

[0027] Preferably, in step S3.2, based on the spatial distance and time interval between each graph node, a comprehensive spatiotemporal similarity score between graph nodes is calculated using a Gaussian similarity function, expressed as:

[0028] s(i,j)=e(-αd(i,j) 2 -βΔt(i,j) 2 );

[0029] Where s(i,j) represents the comprehensive spatiotemporal similarity score between graph node i and graph node j, with a value range of (0,1]; e represents the exponential function, α represents the spatial distance decay coefficient, with a value range of α≥0; d(i,j) 2 Δt(i,j) represents the spatial distance between graph node i and graph node j, β represents the time interval decay coefficient, and its value range is β≥0; 2 This represents the time interval between graph node i and graph node j.

[0030] Preferably, S4 specifically includes:

[0031] S4.1 Based on the spatiotemporal graph, aggregate the spatial features of adjacent nodes and capture the dynamic pattern of high-dimensional alarm event feature vector changes over time to extract the spatiotemporal distribution features of the spatiotemporal graph.

[0032] S4.2 Based on the comprehensive spatiotemporal similarity score, the spatiotemporal distribution features of the spatiotemporal map are sorted to generate a spatiotemporal distribution feature matrix;

[0033] S4.3 Calculate the spatiotemporal distribution feature matrix density distribution and divide similar graph nodes into the same cluster to obtain the alarm event clusters of the spatiotemporal graph.

[0034] Preferably, S5 specifically includes:

[0035] S5.1. By statistically analyzing the ratio of the number of alarm event nodes within an alarm event cluster to the corresponding spatial coverage area, a density index is obtained. At the same time, based on the difference in timestamps between the earliest and latest alarm events within the alarm event cluster, a persistence index is obtained.

[0036] S5.2 Based on density and persistence indicators, calculate the risk score of each cluster, and bind the risk score of each alarm event cluster to the corresponding geographic coordinates to generate a spatial point set with risk values.

[0037] S5.3. Identify high-risk areas with risk ratings based on spatial point sets with risk values ​​and spatiotemporal distribution characteristics.

[0038] Secondly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it causes the processor to perform the steps of the method for identifying high-incidence areas of visual alarms in power transmission line channels according to this application.

[0039] Thirdly, this application also provides a computer-readable storage medium storing instructions thereon, which, when executed by a processor, enable the implementation of a method for identifying high-incidence areas of visual alarms in power transmission line channels.

[0040] The present invention discloses a method for identifying high-incidence areas of visual alarms in power transmission line channels, which has the following beneficial effects.

[0041] This invention collects multi-source data, including visible light video streams, infrared thermal imaging, environmental data, and GPS positioning information, and performs preprocessing and multimodal feature extraction to construct a multimodal feature matrix. Based on correlation analysis between features, a high-dimensional alarm event feature vector is generated and mapped to graph nodes, constructing a spatiotemporal graph containing spatial distance and time interval information. This enables the mining of temporal and spatial correlations of alarm events, transforming previously isolated alarm data into graph data with a topological structure, capable of capturing the propagation patterns of alarm events. By extracting spatiotemporal distribution features and performing cluster analysis, alarm event clusters are identified. Risk scoring is then performed using density and persistence indicators, ultimately identifying high-incidence alarm areas with risk ratings. This achieves data-driven risk assessment, rather than fixed threshold alarms, and can dynamically adapt to the impact of environmental changes on alarm frequency. This invention optimizes the entire process from data acquisition to risk rating, improving the accuracy and comprehensiveness of high-incidence alarm area identification. Attached Figure Description

[0042] Figure 1 This is an overall flowchart of a method for identifying high-incidence areas of visual alarms in power transmission line channels, which is involved in this application.

[0043] Figure 2 This is a flowchart of the process for constructing the multimodal feature matrix in this application.

[0044] Figure 3 This is a flowchart of the process for constructing a spatiotemporal map and identifying high-incidence areas in this application.

[0045] Figure 4 This is a schematic diagram of the computer equipment structure in this application. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0047] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0048] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0049] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0050] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0051] As disclosed in the background section, the main shortcomings of the existing technology are as follows: First, traditional machine learning algorithms (such as decision trees and support vector machines) exhibit significant limitations when processing high-dimensional, nonlinear spatiotemporal data, failing to effectively capture the periodic characteristics of alarm events in the time dimension and the clustering effect in the spatial dimension, resulting in insufficient accuracy and robustness in identifying high-incidence areas. Second, existing research focuses on the independent analysis of single-type alarms (such as smoke and fire alarms or external damage alarms), neglecting the potential correlation between multi-source alarm events (such as the coupling relationship between tree obstacles and smoke and fire alarms), resulting in fragmented analysis of the causes of high-incidence alarm areas and making it difficult to form a global risk prevention and control strategy.

[0052] To address the aforementioned problems, this invention proposes a method for identifying high-incidence areas of visual alarms in power transmission line corridors. This method includes the following steps:

[0053] S1. Collect multi-source data and preprocess it. Extract multi-modal features based on the preprocessed multi-source data and construct a multi-modal feature matrix.

[0054] S2. Based on the multimodal feature matrix, obtain the linear correlation between multimodal features and generate a high-dimensional alarm event feature vector;

[0055] S3. Convert the high-dimensional alarm event feature vector into graph nodes, calculate the spatial distance and time interval of each graph node, construct weighted adjacency edges between graph nodes, and generate a spatiotemporal graph.

[0056] S4. Extract the spatiotemporal distribution features of the spatiotemporal map and obtain the alarm event clusters of the spatiotemporal map;

[0057] S5. Calculate the risk score of each cluster based on alarm event clustering, and identify high-incidence areas with risk ratings by combining spatiotemporal distribution characteristics.

[0058] The technical solution provided by this invention, by generating a spatiotemporal graph based on high-dimensional feature vectors, realizes the mining of temporal and spatial correlations of alarm events, transforming originally isolated alarm data into graph data with a topological structure, which can capture the propagation patterns of alarm events; through cluster risk scoring and high-incidence area identification, it realizes data-driven risk assessment, rather than fixed threshold alarms, and can dynamically adapt to the impact of environmental changes on alarm frequency. This invention realizes the complete process optimization from data collection to risk rating, and improves the accuracy and comprehensiveness of high-incidence area identification.

[0059] The above plan will be explained in detail below.

[0060] Please see the appendix Figure 1 , Figure 1This is a schematic flowchart illustrating the main steps of a method for identifying high-incidence areas of visual alarms in power transmission line corridors according to the present invention. Figure 1 As shown, a method for identifying high-incidence areas of visual alarms in power transmission line channels according to an embodiment of the present invention mainly includes the following steps:

[0061] like Figure 2 As shown, S1, collect multi-source data and preprocess it, extract multi-modal features based on the preprocessed multi-source data and construct a multi-modal feature matrix;

[0062] Preferably, in this embodiment, the multi-source data includes visible light video streams, infrared thermal imaging data, environmental data, and GPS positioning information;

[0063] It should be noted that the visible light video stream is acquired in real time by high-definition cameras deployed along the power transmission line; the infrared thermal imaging data uses an infrared thermal imager to actively scan the power transmission equipment, capture the infrared radiation distribution on the equipment surface, and generate thermal map data reflecting temperature differences; the environmental data is obtained by deploying meteorological sensor arrays (including wind speed, temperature, and humidity sensors) at the monitoring points, and directly measuring environmental data through sensing elements; the GPS positioning information uses satellite navigation receiving equipment to directly receive satellite signals and calculate the latitude and longitude coordinates and timestamp information of the current location of the monitoring equipment.

[0064] Preferably, in this embodiment, the preprocessing includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

[0065] Specifically, in this embodiment, the collected multi-source data, including visible light video streams, infrared thermal imaging data, environmental data, and GPS positioning information, are first cleaned to remove incomplete, erroneous, or irrelevant data records. Next, format conversion is performed to ensure that all data sources follow a unified format standard for subsequent processing. Then, deduplication is performed to eliminate duplicate data entries and ensure the uniqueness of the dataset. The normalization step is used to map data values ​​from different sources to the same scale to avoid certain features from having an undue impact on the results due to excessive differences in magnitude. Finally, outlier handling is used to identify and correct data points that deviate from the normal range to ensure the accuracy and reliability of the multi-source data.

[0066] Preferably, in this embodiment, the alarm target is identified by the YOLOv7 target detection algorithm based on the visible light video stream, and the bounding box coordinates and confidence scores are extracted as visual features.

[0067] Specifically, in this embodiment, based on the visible light video stream, the video is first decoded in real time by an edge computing device, and the video is decomposed into continuous image frames. Background subtraction technology is used to compare the changes in the picture frame by frame to detect the area of ​​moving objects. The YOLOv7 target detection algorithm is used to identify potential alarm targets (such as large machinery, fire points, etc.) that may pose a threat around the transmission line. The bounding box coordinates, category and confidence score of the target are output. Through multi-frame correlation verification, short-term false detections are eliminated and persistent high-confidence targets are retained. Finally, stable bounding box coordinates and confidence scores are output as visual features.

[0068] It should be noted that YOLOv7 is a highly efficient object detection algorithm that can simultaneously predict the category and location of multiple objects in an image through a single forward propagation, improving processing speed while maintaining high detection accuracy.

[0069] Preferably, in this embodiment, the thermal features of the temperature anomaly region are extracted based on infrared thermal imaging data using the U-Net semantic segmentation network;

[0070] Specifically, in this embodiment, based on infrared thermal imaging data, the infrared thermal imaging data is first preprocessed (including radiation calibration, noise filtering (such as Gaussian filtering or median filtering), and normalization (linearly mapping pixel values ​​to the [0,1] interval)) to generate a standardized heatmap. The standardized heatmap is then input into the U-Net semantic segmentation network. The U-Net semantic segmentation network extracts multi-scale spatial features step-by-step through the encoder, and then reconstructs a high-resolution feature map through the decoder. The U-Net semantic segmentation network identifies each pixel in the reconstructed high-resolution feature map, obtains the probability value of each pixel belonging to a high-temperature anomaly region, and generates a high-dimensional probability map with the same size as the input heatmap. Subsequently... The U-Net semantic segmentation network performs global thresholding on the high-dimensional probability map (e.g., setting a fixed threshold of 0.5), converting the probability values ​​into binary results (pixels with a probability ≥ 0.5 are marked as foreground pixels 1, representing high-temperature anomaly regions; pixels with a probability < 0.5 are marked as background pixels 0, representing normal regions). The final output is a binary segmentation mask that is completely consistent with the spatial resolution, geometric structure, and size of the input heatmap, where high-temperature anomaly regions are marked as foreground pixels. Based on the segmentation mask, geometric features (such as bounding box coordinates and area) and thermodynamic features (such as maximum temperature and average temperature) of the temperature anomaly regions are extracted. Finally, these features are integrated to generate a thermodynamic feature vector containing the location, range, and intensity of the temperature anomaly regions.

[0071] It should be noted that the U-Net semantic segmentation network is a convolutional neural network architecture specifically designed for image segmentation. It adopts a symmetrical encoder-decoder structure, captures global contextual features of the image through downsampling, restores spatial resolution through upsampling, and fuses multi-scale features with skip connections. It can accurately locate and segment target regions in images and is widely used in fields such as medical imaging, remote sensing, and heat map analysis.

[0072] Preferably, in this embodiment, environmental features are extracted based on environmental data using a Kalman filter algorithm;

[0073] Specifically, in this embodiment, the environmental data is first synchronized with time and outlier removed. The processed environmental data is then input into the Kalman filter algorithm. The Kalman filter algorithm predicts the current state based on the state estimate from the previous moment, and then combines the current sensor measurement values ​​to obtain the optimal correction amount, dynamically updating the environmental parameter estimates and their uncertainties. Finally, it outputs stable values ​​of environmental characteristics such as wind speed, temperature, and humidity after noise smoothing, as well as their changing trends.

[0074] It should be noted that the Kalman filter algorithm is a recursive data processing method. It dynamically adjusts the optimal estimate of the state variables by using the estimated value of the previous moment (the optimal result obtained by the filter in the recursive process by fusing the observed data at that time with historical state information, and through prediction and correction) and the physical quantity observation data directly collected by the sensor at the current moment (such as temperature sensor readings, anemometer measurements, etc.). This enables accurate tracking and smoothing of physical quantities (such as wind speed, temperature, etc.) even in the presence of measurement noise and interference.

[0075] Preferably, in this embodiment, the multimodal features are spatiotemporally aligned and fused by combining GPS positioning coordinate information to construct a multimodal feature matrix;

[0076] Specifically, in this embodiment, by combining GPS positioning coordinate information, the multimodal features are first uniformly converted to the same geographic coordinate system (such as WGS84). Through spatial mapping, data from different sources are associated with the same geographic location. Secondly, the time dimension of each modality data is aligned based on GPS timestamps. The sliding time window method is used to match multi-source features within the same time period. Subsequently, the multimodal features of spatially overlapping areas are weighted and fused (such as the joint weighting of visual confidence and thermal anomaly intensity) to eliminate data redundancy and enhance key features. Finally, a multimodal feature matrix containing spatial location, timestamps, and fused feature values ​​of each modality is constructed to achieve spatiotemporal consistency integration of cross-modal information and provide structured input for subsequent high-incidence area identification.

[0077] It should be noted that multimodal features refer to visual features, thermal features, and environmental features.

[0078] S2. Based on the multimodal feature matrix, obtain the linear correlation between multimodal features and generate a high-dimensional alarm event feature vector;

[0079] Preferably, in this embodiment, S2 specifically includes the following steps:

[0080] S2.1 Based on the multimodal feature matrix, the maximum correlation coefficient between multimodal features is calculated through canonical correlation analysis, and the linear correlation between multimodal features is obtained;

[0081] Specifically, in this embodiment, the standardized multimodal feature matrix is ​​first divided into multiple sub-feature sets according to different modal types to ensure that each modal feature remains independent. Then, the covariance matrix between the two sets of sub-feature sets is constructed. By solving the generalized eigenvalue decomposition problem (specifically, the optimization problem that simultaneously considers the inter-group covariance and the intra-group covariance), the optimal projection direction that maximizes the correlation between inter-group projected variables and minimizes the correlation within the group projected variables is found. The Pearson correlation coefficient between the optimal projection directions (typical variables) is obtained, and the maximum correlation coefficient value is determined as the quantitative indicator of the linear correlation of multimodal features. At the same time, the correlation coefficients of all typical variable pairs are extracted to form a correlation coefficient matrix, which comprehensively reflects the correlation distribution between different modal combinations.

[0082] It should be noted that canonical correlation analysis is a multivariate statistical method used to study the overall linear correlation between two groups of variables. Canonical correlation analysis finds the linear combination of two groups of variables (called canonical variables) that maximizes the correlation coefficient between the combinations, thus transforming the complex relationship between high-dimensional variables into a correlation analysis between a few canonical variables. Canonical correlation analysis can not only obtain the maximum correlation coefficient, but also generate a series of orthogonal canonical variable pairs, comprehensively revealing the multidimensional correlation structure between two groups of variables. It is widely used in fields such as multimodal data fusion and cross-domain feature association analysis.

[0083] S2.2. A gating mechanism is used to dynamically calculate the fusion contribution between multimodal features, and a high-dimensional alarm event feature vector is generated by nonlinear transformation in combination with linear correlation.

[0084] Specifically, in this embodiment, alarm events are first defined based on the multimodal feature matrix through spatiotemporal correlation rules. The fusion weights of each modality feature corresponding to the alarm event are dynamically calculated using the Sigmoid function in the gating mechanism. Then, the weighted multimodal features are input into a multilayer perceptron (MLP) for nonlinear transformation. The hidden layer of the MLP uses the ReLU activation function to capture the complex interactions between features. The linear correlation strength between multimodal features is used as a conditional parameter. The feature fusion ratio is further adjusted through an attention mechanism. Finally, a high-dimensional alarm event feature vector containing complementary multimodal information is generated.

[0085] It should be noted that the alarm event defined by the spatiotemporal correlation rule means that if the same power transmission equipment triggers multimodal feature anomalies within a fixed time window (such as visual features detecting large mechanical intrusion and thermal feature identification equipment temperature ≥100℃), or there is a causal relationship between cross-device features (such as abnormal displacement of device A's GPS positioning coordinates triggering a sudden rise in the thermal feature temperature of device B), then they are aggregated into an alarm event and the timestamp of the alarm event is counted.

[0086] The gating mechanism uses a specific function (such as Sigmoid) to generate weight values ​​between 0 and 1. The weight values ​​serve as attention allocation coefficients to precisely control the contribution of each feature channel or data stream, enabling the processing to automatically identify key information and reduce the impact of interference factors according to actual needs, thereby achieving intelligent filtering and fusion of multi-source information.

[0087] Nonlinear transformation is a mathematical processing method that remaps data using specific nonlinear functions (such as Sigmoid, ReLU, etc.), breaking simple linear proportional relationships. It can capture complex curve correlations and interaction effects between data, transforming the original feature space into a more complex and expressive new feature space, thereby revealing the hidden nonlinear laws and deep structural features in the data.

[0088] like Figure 3 As shown, S3 converts the high-dimensional alarm event feature vector into graph nodes, calculates the spatial distance and time interval of each graph node, constructs weighted adjacency edges between graph nodes, and generates a spatiotemporal graph.

[0089] Preferably, in this embodiment, S3 specifically includes the following steps:

[0090] S3.1 Map the high-dimensional alarm event feature vector to graph nodes, calculate the spatial distance between graph nodes, and obtain the time interval between graph nodes based on the timestamp information of the graph nodes;

[0091] In this embodiment, the high-dimensional alarm event feature vector is mapped to graph nodes by the Transformer encoder. At the same time, the spatial distance between graph nodes is obtained by the Euclidean distance formula, and the time interval between graph nodes is obtained by the time difference method based on the timestamp information of the graph nodes.

[0092] Specifically, in this embodiment, firstly, the feature vector of each high-dimensional alarm event is combined with its corresponding spatial coordinates (longitude and latitude) and timestamp information to form a complete input containing spatiotemporal attributes. Then, the complex nonlinear correlation features between high-dimensional alarm events are extracted through the multi-head self-attention mechanism of the Transformer encoder, while retaining the location and time features of each event. Based on the spatial coordinates of each event, the geographic spatial distance between any two high-dimensional alarm events is obtained using the Euclidean distance formula, and a spatial proximity matrix is ​​constructed. At the same time, based on the timestamp information of each high-dimensional alarm event, the time interval between any two high-dimensional alarm events is calculated using the time difference method, and a temporal proximity matrix is ​​constructed. Finally, by combining the geographic spatial distance and the time interval, a spatial threshold (e.g., 500 meters) and a time threshold (e.g., 5 minutes) are set, and event pairs that simultaneously satisfy spatial proximity and temporal continuity are determined to be connected. Thus, each alarm event feature vector is mapped to a graph node, and edge connections are established between graph nodes to form a complete spatiotemporal graph structure.

[0093] It should be noted that the Transformer encoder is a computational framework that processes sequential data in parallel through a self-attention mechanism. The Transformer encoder does not rely on recursive or convolutional structures, but directly obtains the pairwise association weights (i.e., attention scores) between all elements in the sequence, dynamically capturing global contextual dependencies. The core of the Transformer encoder consists of stacked multi-head attention layers and feedforward neural network layers. It embeds sequence order information through positional encoding, which can flexibly model long-distance feature interactions and is widely used in tasks that require processing sequences or spatiotemporal correlations.

[0094] The timestamp information of a graph node refers to the data identifier that records the specific moment an event occurs. It is usually stored in a standard time format (such as year-month-day hour:minute:second or Unix timestamp) and is used to mark the precise time point when an alarm event occurs. In the construction of a spatiotemporal graph, the timestamp information not only participates in the feature representation of graph nodes as the time attribute of alarm events, but also compares the time interval with other graph nodes through time difference calculation, thereby determining the temporal correlation relationship (such as chronological order or simultaneity) between alarm events. It is the key basis for realizing the alignment and correlation of multimodal features in the time dimension and provides a time reference for analyzing the dynamic evolution process of events.

[0095] S3.2 Calculate the comprehensive spatiotemporal similarity score between graph nodes based on the spatial distance and time interval of each graph node;

[0096] In this embodiment, based on the spatial distance and time interval between graph nodes, a comprehensive spatiotemporal similarity score between graph nodes is calculated using a Gaussian similarity function, expressed as:

[0097] s(i,j)=e(-αd(i,j)2 -βΔt(i,j) 2 );

[0098] Where s(i,j) represents the comprehensive spatiotemporal similarity score between graph node i and graph node j, with a value range of (0,1]; e represents the exponential function, α represents the spatial distance decay coefficient, with a value range of α≥0; d(i,j) 2 Δt(i,j) represents the spatial distance between graph node i and graph node j, β represents the time interval decay coefficient, and its value range is β≥0; 2 This represents the time interval between graph node i and graph node j.

[0099] S3.3. Based on the comprehensive spatiotemporal similarity score, obtain the adjacent nodes of each graph node, and use the comprehensive spatiotemporal similarity score as the edge weight to construct a spatiotemporal graph.

[0100] In this embodiment, based on the comprehensive spatiotemporal similarity score, the neighboring nodes of each graph node are obtained through the K-nearest neighbor algorithm, and the comprehensive spatiotemporal similarity score is used as the edge weight to construct a spatiotemporal graph.

[0101] Specifically, in this embodiment, firstly, a parameter K is set to represent the number of adjacent nodes. For each graph node, it is sorted in descending order according to its comprehensive spatiotemporal similarity score. Then, the top K graph nodes with the highest comprehensive spatiotemporal similarity scores are selected as adjacent nodes from the sorting results, and graph nodes with low comprehensive spatiotemporal similarity scores are excluded to control the sparsity of the graph structure. Next, the comprehensive spatiotemporal similarity score between each graph node and the selected K adjacent nodes is directly used as the weight value of the corresponding edge, preserving the characteristic that high similarity scores correspond to strong associations. Finally, all graph nodes, adjacency relationships (connections between each graph node and the selected K adjacent nodes), and edge weights are integrated to construct a complete weighted spatiotemporal graph structure, where graph nodes represent alarm events, edges represent the spatiotemporal association strength between alarm events, and weights reflect the degree of association between the current graph node and its selected K adjacent nodes.

[0102] It should be noted that the K-nearest neighbors algorithm is a data association method based on distance metric. It calculates the spatial or feature distance (such as Euclidean distance) between the target object and all samples, sorts them from high to low in terms of similarity, and selects the K closest samples as the neighbor reference set of the target object.

[0103] A spacetime diagram is a two-dimensional graph in physics used to visually describe the distribution of events in time and space. It is usually constructed with time as the vertical axis and space as the other orthogonal axis. Each point in the spacetime diagram corresponds to an event with definite time and space coordinates. It is mainly used to visualize and analyze the characteristics of spacetime transformation (such as simultaneity and relativity, time dilation) and the causal relationship of events (such as time interval-like, space interval-like).

[0104] Adjacent nodes are other nodes that are directly connected to a specific node in a graph structure. They are connected by edges. In a spatiotemporal graph, adjacent nodes not only represent spatial proximity (e.g., geographical proximity) or temporal continuity (e.g., events occurring in close succession), but also quantify the strength of this connection through edge weights.

[0105] S4. Extract the spatiotemporal distribution features of the spatiotemporal map and obtain the alarm event clusters of the spatiotemporal map;

[0106] Preferably, in this embodiment, S4 specifically includes the following steps:

[0107] S4.1 Based on the spatiotemporal graph, aggregate the spatial features of adjacent nodes and capture the dynamic pattern of high-dimensional alarm event feature vector changes over time to extract the spatiotemporal distribution features of the spatiotemporal graph.

[0108] In this embodiment, based on the spatiotemporal graph, the spatial features of adjacent nodes are aggregated through graph convolution of the spatiotemporal graph neural network, and temporal convolution is used to capture the dynamic pattern of high-dimensional alarm event feature vector changes over time, thereby extracting the spatiotemporal distribution features of the spatiotemporal graph.

[0109] Specifically, in this embodiment, firstly, based on the constructed spatiotemporal graph, the spatial convolutional layer of the spatiotemporal graph neural network uses a graph convolution algorithm to perform weighted summation of the features of adjacent nodes of each graph node, integrating local spatial topology information to update the spatial features of the graph nodes; then, a temporal convolutional network is used to perform sliding window processing on the graph node feature sequence along the time axis to extract the dynamic pattern of alarm event features changing over time; finally, the spatial features of the graph nodes and the dynamic pattern of alarm event features changing over time are fused, and through feature compression and abstraction operations, spatiotemporal distribution features that simultaneously contain spatial correlation patterns and temporal evolution laws are extracted, completing the joint representation of multimodal alarm events in the spatiotemporal dimension.

[0110] It should be noted that spatiotemporal graph neural networks are a graph data processing technology that integrates spatial and temporal dimension analysis capabilities. It aggregates the spatial correlation features of node neighborhoods through graph convolution operations, and at the same time uses temporal convolution or recurrent structures to capture the dynamic patterns of node features changing over time.

[0111] The training process of a spatiotemporal graph neural network is as follows: First, labeled spatiotemporal graph data is input into the network, and forward propagation is used to obtain the global feature representation of nodes or the graph. Then, the loss function (such as cross-entropy or mean squared error) is calculated based on the prediction results (such as event classification or anomaly scoring) and the true labels to measure the prediction bias. Next, the loss value is backpropagated to each layer of the network using the chain rule, and the parameter gradient is calculated layer by layer. Finally, gradient descent or a variant algorithm (such as Adam) is used to update the parameter weights of graph convolution kernels, temporal convolution kernels, etc., to gradually reduce the prediction error. The above iterative process is repeated until the loss converges or the preset number of training rounds is reached, and finally the parameter configuration of the spatiotemporal graph neural network that can effectively extract spatiotemporal features is obtained.

[0112] The dynamic pattern of the feature vector of a high-dimensional alarm event changes over time refers to the evolution of the multidimensional features of the alarm event in the time dimension. Specifically, it manifests as: the fluctuation trend of different dimension values ​​in the feature vector of a high-dimensional alarm event over time (such as sudden increase, slow decrease or periodic oscillation), the temporal migration of the correlation between features (such as some feature combinations changing from weak correlation to strong correlation), and the dynamic transition of the feature space distribution state (such as the transition from low-dimensional subspace to high-dimensional region).

[0113] The spatiotemporal distribution characteristics of a spatiotemporal map refer to the joint distribution patterns of node features in the spatial location association and temporal evolution process. Specifically, they are manifested in the following ways: in the spatial dimension, the geographical clustering of node feature values ​​(such as the similarity of alarm event features in neighboring areas) and spatial heterogeneity (such as different geographical locations exhibiting differentiated feature patterns), and in the temporal dimension, the dynamic fluctuation trend of feature values ​​(such as short-term sudden changes or long-term periodic changes) and temporal dependence (such as the current feature being continuously affected by historical features). The spatiotemporal distribution characteristics of a spatiotemporal map simultaneously capture the association patterns of alarm events in geographic space (such as local outbreaks or cross-regional propagation) and the evolution patterns on the time axis (such as the sudden-continuous-decay process), providing a multi-dimensional joint characterization basis for analyzing the diffusion path of alarm events and predicting future risk situations.

[0114] S4.2 Based on the comprehensive spatiotemporal similarity score, the spatiotemporal distribution features of the spatiotemporal map are sorted to generate a spatiotemporal distribution feature matrix;

[0115] In this embodiment, the spatiotemporal distribution features of the spatiotemporal graph are sorted based on the comprehensive spatiotemporal similarity score between graph nodes to generate a spatiotemporal distribution feature matrix;

[0116] Specifically, in this embodiment, firstly, based on the calculated comprehensive spatiotemporal similarity scores between graph nodes, a node adjacency matrix is ​​constructed, where the matrix element values ​​directly reflect the spatiotemporal association strength between graph nodes; then, the comprehensive spatiotemporal similarity scores of each graph node are sorted in descending order, and the top K adjacent nodes with the highest similarity are selected as the local spatiotemporal association domain; subsequently, for each graph node, the comprehensive spatiotemporal similarity scores between the graph node and the K adjacent nodes, as well as the graph node's own features (including multi-dimensional feature vectors describing the independent attributes of the node, such as the timestamp of the alarm event, spatial location coordinates, event type encoding, and severity index), are extracted, and the neighborhood information is aggregated by weighted summation to generate a local spatiotemporal distribution feature representation of the adjacent nodes; at the same time, the local spatiotemporal distribution features of each graph node are arranged in a fixed order, finally forming a matrix structure containing the spatiotemporal association features of all nodes.

[0117] S4.3 Calculate the spatiotemporal distribution feature matrix density distribution and divide similar graph nodes into the same cluster to obtain the alarm event clusters of the spatiotemporal graph;

[0118] In this embodiment, the spatiotemporal distribution feature matrix density distribution is calculated by a clustering algorithm, and similar graph nodes are divided into the same cluster to obtain the alarm event clusters of the spatiotemporal graph;

[0119] Specifically, in this embodiment, the spatiotemporal distribution feature matrix is ​​first standardized (e.g., Min-Max normalization or Z-score normalization) to eliminate dimensional differences between different feature dimensions. Then, a density clustering algorithm (e.g., DBSCAN) is selected, and key parameters are set, including the neighborhood radius ε and the minimum sample size MinPts. Local density is obtained by calculating the number of graph nodes contained in the neighborhood radius ε of each graph node's feature vector in the feature space. Subsequently, density-connected nodes (i.e., the set of nodes that are mutually located within each other's ε-neighborhood and meet the minimum sample size requirement) are aggregated into the same cluster, forming a high-density connected region. Graph nodes with extremely low density that cannot be covered by any high-density region are marked as noise points and excluded. Finally, based on the clustering results, the alarm event nodes in the spatiotemporal graph are divided into multiple cluster structures. Each cluster structure represents a set of alarm events with similar spatiotemporal distribution characteristics. Based on the existing multiple cluster structures, inter-cluster similarity is obtained through DTW (Dynamic Time Warping), and similar clusters are merged using a graph clustering algorithm (e.g., Louvain) to obtain the alarm event clusters of the spatiotemporal graph.

[0120] S5. Calculate the risk score of each cluster based on alarm event clustering, and identify high-incidence areas with risk ratings by combining spatiotemporal distribution characteristics.

[0121] S5.1. By statistically analyzing the ratio of the number of alarm event nodes within an alarm event cluster to the corresponding spatial coverage area, a density index is obtained. At the same time, based on the difference in timestamps between the earliest and latest alarm events within the alarm event cluster, a persistence index is obtained.

[0122] Specifically, in this embodiment, each alarm event cluster is first traversed, and the total number of alarm event nodes contained in the cluster is counted as the number of nodes. At the same time, the area of ​​the minimum bounding rectangle of the spatial coordinates of all nodes in the cluster is calculated to obtain the spatial coverage. Then, the density index is obtained by the ratio of the number of nodes to the spatial coverage to quantify the degree of event aggregation in a unit space. Next, the timestamp information of all alarm event nodes in the cluster is extracted to determine the earliest and latest occurrence times, and the difference between the two is calculated to obtain the time span. Finally, the time span is used as a persistence index to characterize the duration of alarm events on the time axis. By jointly analyzing the density index and the persistence index, different types of alarm event aggregation patterns, such as high-density short-term outbreaks and low-density long-term persistence, can be identified, providing a quantitative basis for risk assessment and emergency response.

[0123] S5.2 Based on density and persistence indicators, calculate the risk score of each cluster, and bind the risk score of each alarm event cluster to the corresponding geographic coordinates to generate a spatial point set with risk values.

[0124] In this embodiment, based on density and persistence indicators, the risk score of each cluster is calculated using the analytic hierarchy process (AHP), and the risk score of each alarm event cluster is bound to the corresponding geographic coordinates to generate a spatial point set with risk values.

[0125] Specifically, in this embodiment, the information entropy of density and persistence indicators is first objectively calculated using the entropy method. The weight allocation is automatically determined based on the dispersion of the indicators (the smaller the information entropy, the higher the importance of the indicator), avoiding interference from subjective factors. Then, based on the weight coefficients determined by the entropy method, the risk score of each cluster is obtained by weighted summation. Finally, the risk score of each cluster is bound to the geographic coordinates of the geometric center point of the smallest bounding rectangle of its spatial coverage area to construct a spatial point set data structure containing longitude, latitude, and risk value (risk score of each cluster). This enables accurate mapping of alarm event risk values ​​in geographic space and supports risk visualization and spatial analysis.

[0126] It should be noted that the Analytic Hierarchy Process (AHP) is an analytical method that decomposes decision problems with multi-level structural characteristics into hierarchical structures and calculates relative weights using quantitative relationships between pairs of indicators. The core steps include: decomposing the decision objective layer by layer into a criterion layer and an alternative layer; generating a judgment matrix through pairwise indicator comparisons driven by objective data (e.g., determining comparison values ​​based on historical statistical correlation or entropy methods); obtaining the objective weights of each indicator using the eigenvector method; ensuring the logical rationality of the weights through a consistency ratio test; and finally, deriving a comprehensive evaluation result through weight integration. The AHP avoids reliance on subjective experience and instead relies on data-driven objective quantitative relationships to determine the importance of indicators.

[0127] S5.3 Identify high-risk areas with risk ratings based on spatial point sets with risk values ​​and spatiotemporal distribution characteristics;

[0128] In this embodiment, based on a spatial point set with risk values, high-incidence areas with risk ratings are identified by combining kernel density estimation with spatiotemporal distribution characteristics.

[0129] Specifically, in this embodiment, firstly, a geospatial grid is divided based on the spatial coverage of a set of spatial points with risk values. Then, a kernel density estimation method is used, with each risk point as the center, and weights are set according to the risk value. The density contribution value of each grid unit is calculated using a Gaussian kernel function, where the kernel function bandwidth parameter is automatically selected through cross-validation to balance smoothness and detail preservation. Subsequently, the density contributions of all risk points to the same grid unit are accumulated to obtain the comprehensive risk density value of the grid unit. Combined with spatiotemporal distribution characteristics, the comprehensive risk density value is spatiotemporally weighted and corrected to strengthen the weight influence of recent events and neighboring areas. Finally, based on the corrected comprehensive risk density value, a hierarchical threshold division is performed (such as the natural breakpoint method or equal-interval segmentation) to identify high-incidence areas of different risk levels and generate a risk heat map to intuitively display the spatial distribution pattern of risks, providing a precise location basis for risk prevention and control.

[0130] It should be noted that kernel density estimation is a nonparametric statistical method used to estimate the probability density distribution of random variables in space or feature space. The basic principle is to place a smooth kernel function (such as a Gaussian kernel) centered on each sample point. All kernel functions are superimposed to form a continuous probability density surface. The shape of the kernel function controls the smoothness, and the bandwidth parameter determines the diffusion range. It can capture multimodal features and local clustering patterns without assuming a specific data distribution pattern, and is widely used in geospatial hotspot detection, signal processing, and financial risk analysis. This embodiment also provides a computer device suitable for identifying high-incidence areas of visual alarms in transmission line corridors, including: a memory and a processor; the memory stores computer-executable instructions, and the processor executes these instructions to implement the identification method for high-incidence areas of visual alarms in transmission line corridors as proposed in the above embodiment.

[0131] This embodiment also provides a computer device applicable to a method for identifying high-incidence areas of visual alarms in power transmission line channels, such as... Figure 4 As shown, computer device 200 includes at least one memory 210 (e.g., non-volatile memory such as flash memory, ROM, hard disk drive, magnetic disk, optical disk), at least one processor 220, and computer program 230. Memory 210 stores the computer program 230, which can be executed by processor 220. Processor 230 is configured to run the computer program 230 stored on memory 210. The above-described method can be achieved by running the computer program stored on one or more memories on one or more processors (e.g., in a manner where multiple processors cooperate to run the computer program or a single processor runs the computer program independently). Figure 1 The method includes one or more steps or operations.

[0132] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0133] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements a method for identifying high-incidence areas of visual alarms in power transmission line channels as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0134] In summary, this invention achieves the mining of temporal and spatial correlations of alarm events by generating spatiotemporal graphs based on high-dimensional feature vectors, transforming originally isolated alarm data into graph data with topological structure, and capturing the propagation patterns of alarm events; through cluster risk scoring and high-incidence area identification, it realizes data-driven risk assessment rather than fixed threshold alarms, and can dynamically adapt to the impact of environmental changes on alarm frequency. This invention optimizes the entire process from data collection to risk rating, improving the accuracy and comprehensiveness of high-incidence area identification.

[0135] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0136] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for identifying high-incidence areas of visual alarms in power transmission line corridors, characterized in that, Includes the following steps: S1. Collect multi-source data and preprocess it. Extract multi-modal features based on the preprocessed multi-source data and construct a multi-modal feature matrix. S2. Based on the multimodal feature matrix, obtain the linear correlation between multimodal features and generate a high-dimensional alarm event feature vector; S3. Convert the high-dimensional alarm event feature vector into graph nodes, calculate the spatial distance and time interval of each graph node, construct weighted adjacency edges between graph nodes, and generate a spatiotemporal graph. S4. Extract the spatiotemporal distribution features of the spatiotemporal map and obtain the alarm event clusters of the spatiotemporal map; S5. Calculate the risk score of each cluster based on alarm event clustering, and identify high-incidence areas with risk ratings by combining spatiotemporal distribution characteristics.

2. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 1, characterized in that, The multi-source data in S1 includes visible light video streams, infrared thermal imaging data, environmental data, and GPS positioning information; the preprocessing in S1 includes data cleaning, format conversion, deduplication, normalization, and outlier handling.

3. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 2, characterized in that, The specific steps for extracting multimodal features and constructing a multimodal feature matrix based on preprocessed multi-source data are as follows: Identify alarm targets based on visible light video streams and extract bounding box coordinates and confidence scores as visual features; Based on infrared thermal imaging data, thermal characteristics of temperature anomaly regions are extracted; Based on environmental data, extract environmental features; By combining GPS positioning coordinate information, multimodal features are spatiotemporally aligned and fused to construct a multimodal feature matrix.

4. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 1, characterized in that, S2 specifically includes: S2.1 Based on the multimodal feature matrix, calculate the maximum correlation coefficient between multimodal features and obtain the linear correlation between multimodal features; S2.2 Calculate the fusion contribution between multimodal features and combine linear correlation to generate a high-dimensional alarm event feature vector.

5. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 1, characterized in that, S3 specifically includes: S3.1 Map the high-dimensional alarm event feature vector to graph nodes, calculate the spatial distance between graph nodes, and obtain the time interval between graph nodes based on the timestamp information of the graph nodes; S3.2 Calculate the comprehensive spatiotemporal similarity score between graph nodes based on the spatial distance and time interval of each graph node; S3.

3. Based on the comprehensive spatiotemporal similarity score, obtain the adjacent nodes of each graph node, and use the comprehensive spatiotemporal similarity score as the edge weight to construct a spatiotemporal graph.

6. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 5, characterized in that, In S3.1, the high-dimensional alarm event feature vector is mapped to graph nodes through the Transformer encoder. At the same time, the spatial distance between graph nodes is obtained by using the Euclidean distance formula, and the time interval between graph nodes is obtained by using the time difference method based on the timestamp information of the graph nodes. In step S3.2, based on the spatial distance and time interval between each graph node, a comprehensive spatiotemporal similarity score between graph nodes is calculated using a Gaussian similarity function. The expression is as follows: s(i,j)=e(-αd(i,j) 2 -βΔt(i,j) 2 ); Where s(i,j) represents the comprehensive spatiotemporal similarity score between graph node i and graph node j, with a value range of (0,1]; e represents the exponential function, α represents the spatial distance decay coefficient, with a value range of α≥0; d(i,j) 2 Δt(i,j) represents the spatial distance between graph node i and graph node j, β represents the time interval decay coefficient, and its value range is β≥0; 2 This represents the time interval between graph node i and graph node j.

7. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 5, characterized in that, S4 specifically includes: S4.1 Based on the spatiotemporal graph, aggregate the spatial features of adjacent nodes and capture the dynamic pattern of high-dimensional alarm event feature vector changes over time to extract the spatiotemporal distribution features of the spatiotemporal graph. S4.2 Based on the comprehensive spatiotemporal similarity score, the spatiotemporal distribution features of the spatiotemporal map are sorted to generate a spatiotemporal distribution feature matrix; S4.3 Calculate the spatiotemporal distribution feature matrix density distribution and divide similar graph nodes into the same cluster to obtain the alarm event clusters of the spatiotemporal graph.

8. The method for identifying high-incidence areas of visual alarms in transmission line corridors as described in claim 1, characterized in that, S5 specifically includes: S5.

1. By statistically analyzing the ratio of the number of alarm event nodes within an alarm event cluster to the corresponding spatial coverage area, a density index is obtained. At the same time, based on the difference in timestamps between the earliest and latest alarm events within the alarm event cluster, a persistence index is obtained. S5.2 Based on density and persistence indicators, calculate the risk score of each cluster, and bind the risk score of each alarm event cluster to the corresponding geographic coordinates to generate a spatial point set with risk values. S5.

3. Identify high-risk areas with risk ratings based on spatial point sets with risk values ​​and spatiotemporal distribution characteristics.

9. A computer device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor, when executing the program, implements the steps of a method for identifying high-incidence areas of visual alarms in power transmission line channels as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, the program, when executed by a processor, implementing the steps of a method for identifying high-incidence areas of visual alarms in power transmission line channels as described in any one of claims 1 to 8.

Images