Defect identification method for key components of transmission lines based on cloud-edge collaboration

Abstract

The present invention relates to the field of power transmission line detection technology, and discloses a method for identifying defects in key components of power transmission lines based on cloud-edge collaboration. The method obtains a multi-source inspection data set consisting of transmission line component images collected by drones, edge-end sensor data, and a cloud-end historical defect database. At the edge, component regional features are extracted through an image preprocessing algorithm, and environmental correlation parameters are generated using a multimodal feature fusion algorithm; at the cloud, a spatiotemporal correlation analysis is performed on the historical defect database to generate a component defect evolution map. The above information is input into a cloud-edge collaborative recognition model to obtain a component defect feature vector, and then a multi-level defect recognition network is constructed through a dynamic optimization algorithm to output component defect classification results and confidence levels. The method and system improve the accuracy, real-time nature, and reliability of defect identification of key components of transmission lines, and have good application prospects.

Description

Technical Field

[0001] The present invention relates to the field of transmission line detection technology, and specifically to a method for identifying defects in key components of transmission lines based on cloud-edge collaboration. Background Art

[0002] In modern power systems, transmission lines, as critical infrastructure for power transmission, are crucial for ensuring the normal supply of electricity for production and daily life. Transmission lines are often widely distributed, traversing diverse and complex geographical environments and climatic conditions. Over long-term operation, key components are prone to defects such as broken insulators, broken conductors, and corroded hardware. If these defects are not discovered and addressed promptly, they can lead to line failures, resulting in power outages and significant social and economic losses.

[0003] Traditional methods for detecting defects in transmission line components rely primarily on manual inspections. Manual inspections have numerous limitations. First, they are inefficient and require significant manpower, material resources, and time. Especially for long-distance, large-scale transmission lines, manual inspections are difficult to conduct comprehensively and promptly. Second, manual inspections are significantly affected by environmental factors and subjective human factors. In harsh natural environments, such as those characterized by high temperatures, extreme cold, high altitudes, and strong winds, manual inspections become significantly more difficult and dangerous. Furthermore, manual inspection accuracy is difficult to guarantee, and differences in the technical skills and work experience of different inspectors can lead to discrepancies in test results.

[0004] With technological advancements, drone inspections are increasingly being used in power transmission line inspections. Drones can quickly capture image information of transmission line components, improving inspection efficiency. However, relying solely on drone-collected images for defect identification presents challenges. Firstly, the sheer volume of image data collected by drones can lead to delays and lags during data transmission and processing, compromising the real-time nature of defect identification. Secondly, defect assessment based solely on image information lacks consideration for environmental factors, making it difficult to accurately assess defect severity and development trends.

[0005] Furthermore, existing transmission line defect identification systems often centralize all data processing and analysis in the cloud, which not only places extremely high demands on network bandwidth but also severely impacts system reliability and stability when network signals are unstable. Furthermore, while edge devices can collect large amounts of sensor data, such as temperature, humidity, and vibration, this data is not fully and effectively utilized and cannot be deeply integrated with image data for more accurate defect identification.

[0006] Against the backdrop of the continuous development of big data and artificial intelligence technologies, how to fully utilize multi-source data and combine it with cloud-edge collaboration to improve the accuracy, real-time nature, and reliability of defect identification in key transmission line components has become a critical issue that needs to be addressed in the field of transmission line inspection. The shortcomings of existing technologies in data fusion, computing resource allocation, and defect identification accuracy provide a broad space for research on defect identification technology for key transmission line components based on cloud-edge collaboration. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for identifying defects in key components of transmission lines based on cloud-edge collaboration to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying defects in key components of a power transmission line based on cloud-edge collaboration, the method comprising:

[0009] Acquire a multi-source inspection data set; the multi-source inspection data includes transmission line component images collected by drones, edge sensor data, and a cloud-based historical defect database; the edge sensor data includes temperature, humidity, and vibration information;

[0010] Based on the transmission line component image, extracting component regional features through an edge-end image preprocessing algorithm, wherein the regional features include geometric contour information, texture distribution features, and local defect candidate frames;

[0011] Generate environmental parameters based on the edge sensor data using a multimodal feature fusion algorithm, where the environmental parameters include a temperature anomaly coefficient, a humidity influencing factor, and a vibration fluctuation threshold;

[0012] Performing spatiotemporal correlation analysis on the cloud-based historical defect database to generate a component defect evolution map;

[0013] Inputting the component regional characteristics, environmental correlation parameters and defect evolution map into the cloud-edge collaborative recognition model to generate a component defect feature vector;

[0014] Based on the component defect feature vector, a multi-level defect recognition network is constructed through a dynamic optimization algorithm to output component defect classification results and confidence levels; the levels of the multi-level defect recognition network represent defect type branches, and the nodes represent feature fusion weights and recognition priorities.

[0015] Preferably, the extracting component region features by using an edge image preprocessing algorithm includes:

[0016] performing distortion correction processing on the image of the power transmission line component to generate a standardized image;

[0017] Based on the predefined component template library, the key component area is located through the region matching algorithm, and the component image blocks are segmented;

[0018] A multi-scale filtering algorithm is used to extract the texture gradient features of the component image block and generate candidate defect areas in combination with the geometric contour information;

[0019] The candidate defect area is compared with the reference parameters in the component template library to generate a local defect candidate frame and a confidence score.

[0020] Preferably, the generating of environment-related parameters by a multimodal feature fusion algorithm includes:

[0021] Performing time domain alignment processing on the edge sensor data to eliminate acquisition time deviation;

[0022] Calculate the mean and variance of temperature, humidity, and vibration data based on the sliding window statistical method to generate a dynamic environmental baseline;

[0023] Calculate the temperature anomaly coefficient and humidity impact factor based on the difference between the dynamic environmental baseline and the environmental threshold;

[0024] The spectral characteristics of the vibration data are compared with the preset vibration fluctuation threshold to generate a vibration anomaly index and encode it as an environment-related parameter.

[0025] Preferably, the cloud-edge collaborative recognition model includes a feature enhancement module and a collaborative reasoning module, and the feature enhancement module includes:

[0026] Normalizing the geometric contour information in the component region feature to obtain a first enhancement vector;

[0027] Discretize and encode the temperature anomaly coefficient in the environment-related parameters to generate a second enhancement vector;

[0028] Performing a spatiotemporal convolution operation on the defect evolution map to extract defect propagation characteristics and obtain a third enhancement vector;

[0029] The first enhancement vector, the second enhancement vector and the third enhancement vector are merged into a high-dimensional fusion feature through the feature splicing layer.

[0030] Preferably, the collaborative reasoning module includes:

[0031] Perform channel attention weighting on high-dimensional fusion features to generate a feature importance matrix;

[0032] Extract defect correlation features through cross-level feature interaction algorithm and generate defect correlation matrix;

[0033] Perform element-by-element product operation on the feature importance matrix and the defect correlation matrix to generate collaborative reasoning features;

[0034] The collaborative inference features are superimposed with the original high-dimensional fusion features through residual connection to output the component defect feature vector.

[0035] Preferably, the multi-level defect recognition network is constructed by a dynamic optimization algorithm, including:

[0036] Initialize hierarchical attributes according to defect type branches and generate a weight assignment matrix based on identification priority;

[0037] The component defect feature vector is used as an input feature, and a weight allocation matrix is ​​composed of feature fusion weights and recognition priorities;

[0038] The weight parameters of each layer are iteratively updated through the gradient back-propagation algorithm to adjust the weight distribution matrix;

[0039] The optimal classification sequence covering all defect types is generated based on the adjusted weight distribution matrix.

[0040] Preferably, the method for constructing the component template library includes:

[0041] Collect standard image samples of various typical transmission line components and extract reference geometric contours and texture features;

[0042] Perform affine transformation on the reference geometric outline to generate a multi-view component template;

[0043] Categorize templates according to component type and associate with a benchmark parameter database;

[0044] The classified templates are stored as a component template library.

[0045] Preferably, the parameter optimization method of the dynamic environment baseline includes:

[0046] Calculate the initial sliding window length and statistical interval based on the historical sensor data distribution;

[0047] The grid search algorithm is used to traverse the parameter combinations and select the parameters with the best baseline stability and anomaly detection sensitivity;

[0048] The sliding window length and statistical interval are dynamically adjusted according to the detection results to optimize the calculation accuracy of environmental correlation parameters.

[0049] Preferably, the parameter updating method of the gradient back propagation algorithm includes:

[0050] The loss function between layers is defined as the sum of the classification error and the regularization term of the feature fusion weight;

[0051] Initialize the weight parameters of each layer to random values ​​and the input features to normalized vectors;

[0052] Calculate the gradient update of each layer based on the previous layer using the chain rule and record the optimal weight path;

[0053] The weight distribution matrix is ​​updated based on the back propagation of the optimal weight path to generate the final classification result.

[0054] Preferably, the present invention further includes a system for identifying defects in key components of power transmission lines based on cloud-edge collaboration, the system comprising:

[0055] Multi-source data acquisition module: used to obtain a multi-source inspection data set, including transmission line component images collected by drones, edge sensor data, and a cloud-based historical defect database; wherein the edge sensor data includes temperature, humidity, and vibration information;

[0056] Edge feature extraction module: configured to extract component regional features based on the transmission line component image through an edge end image preprocessing algorithm, wherein the regional features include geometric contour information, texture distribution features and local defect candidate frames;

[0057] Environmental parameter fusion module: used to generate environmental parameters based on the edge sensor data through a multimodal feature fusion algorithm, wherein the environmental parameters include temperature anomaly coefficient, humidity influence factor and vibration fluctuation threshold;

[0058] Cloud-based graph generation module: performs spatiotemporal correlation analysis on the cloud-based historical defect database to generate a component defect evolution graph;

[0059] Collaborative identification module: Inputs the component regional features, environmental correlation parameters and defect evolution map into the cloud-edge collaborative identification model to generate a component defect feature vector; the cloud-edge collaborative identification model includes a feature enhancement module and a collaborative reasoning module;

[0060] Defect classification module: used to construct a multi-level defect recognition network based on the component defect feature vector through a dynamic optimization algorithm, and output the component defect classification results and confidence levels; the levels of the multi-level defect recognition network represent defect type branches, and the nodes represent feature fusion weights and recognition priorities.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] From a data utilization perspective, multi-source inspection data is fully integrated. Images of transmission line components captured by drones provide a visual representation of the components' appearance. Temperature, humidity, and vibration information collected by edge sensors reflect the real-time operating environment of the components. The cloud-based historical defect database contains a wealth of historical data. By organically combining this data, rich and comprehensive information is provided for defect identification. For example, when identifying insulator defects, not only can the insulator's appearance characteristics in the image be determined, but also the temperature and humidity conditions at the time, as well as historical patterns of insulator defect development under similar conditions, significantly improving the accuracy and reliability of defect identification.

[0063] In terms of processing efficiency, the cloud-edge collaborative architecture has obvious advantages. The edge is responsible for image preprocessing and partial feature extraction. For example, it uses edge-side image preprocessing algorithms to quickly extract component area features, reducing data transmission volume and reducing dependence on network bandwidth. At the same time, a multimodal feature fusion algorithm is used at the edge to generate environmental correlation parameters, enabling real-time processing of local data and improving the system's response speed. The cloud, on the other hand, focuses on in-depth analysis of historical data, such as generating component defect evolution maps through spatiotemporal correlation analysis, providing macro-trend judgments for defect identification. This division of labor and collaboration effectively avoids the delays and freezes caused by the concentration of all data processing in the cloud, ensuring the real-time nature of defect identification.

[0064] In terms of defect recognition accuracy, the series of algorithms of the present invention play an important role. The texture gradient features of the component image blocks are extracted through a multi-scale filtering algorithm, and the candidate defect areas are generated in combination with the geometric contour information. Then, the candidate defect areas are compared with the benchmark parameters in the component template library, which can more accurately locate the local defect candidate boxes. The feature enhancement module in the cloud-edge collaborative recognition model processes and fuses the geometric contour information, temperature anomaly coefficient, etc. to obtain high-dimensional fusion features; the collaborative reasoning module further generates accurate component defect feature vectors through operations such as channel attention weighting and cross-level feature interaction. Based on this, the multi-level defect recognition network constructed by the dynamic optimization algorithm can output more accurate component defect classification results and confidence levels according to defect type branching and recognition priority.

[0065] The present invention also has many improvements in terms of the adaptability and stability of the system. The component template library construction method enables the system to adapt to the detection needs of various typical power transmission line components. By collecting standard image samples, generating multi-view component templates and classifying and storing them, the versatility of the system is improved. The parameter optimization method of the dynamic environmental baseline adjusts the sliding window length and statistical interval according to the historical sensor data distribution, optimizes the calculation accuracy of environmental correlation parameters, and enhances the system's adaptability to different environmental conditions. The parameter update method of the gradient backpropagation algorithm ensures the stability and accuracy of the multi-level defect recognition network by defining a reasonable loss function and recording the optimal weight path.

[0066] From a cost-effectiveness perspective, this method and system reduces the frequency and workload of manual inspections, lowering both labor costs and safety risks during the inspection process. Furthermore, by enabling timely and accurate identification of component defects, it avoids line failures and power outages caused by these defects, reducing repair costs and the economic losses associated with power outages, resulting in significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a working principle diagram of the method for identifying defects in key components of power transmission lines based on cloud-edge collaboration according to the present invention;

[0068] Figure 2 This is the workflow diagram of the multimodal feature fusion algorithm;

[0069] Figure 3 This is the workflow diagram of the collaborative reasoning module of the cloud-edge collaborative recognition model;

[0070] Figure 4 Workflow diagram of the method for building a component template library. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0072] See also Figure 1-Figure 4 The present invention provides a technical solution: a method for identifying defects in key components of a power transmission line based on cloud-edge collaboration, the method comprising:

[0073] Beidou differential positioning modules and antenna parameter sensors are used to simultaneously collect positioning data and operating parameters of the target antenna system. Positioning data includes real-time latitude, longitude, and elevation information, which accurately determine the antenna system's geographic location. Operating parameters include signal strength, standing wave ratio, and feeder loss rate, which reflect the antenna system's operating status. For example, in a certain communication base station, Beidou differential positioning modules are installed at key locations in the antenna system, and antenna parameter sensors are deployed to ensure accurate data acquisition.

[0074] The collected positioning data and operating parameters are fed into a pre-set dynamic error compensation model. This model dynamically adjusts the compensation coefficient based on historical error distribution characteristics, thereby correcting for multipath errors and sensor noise in the positioning data, ultimately generating calibrated data. In real-world applications, multipath can cause deviations in positioning data, and sensors can also generate noise due to various factors. The dynamic error compensation model effectively addresses these issues.

[0075] The calibration data is fed into a pre-set signal coverage optimization model. This model determines adjustment weights based on the target area's signal coverage requirements and environmental interference characteristics, generating real-time beam pointing adjustments for the antenna system. Different target areas have different signal coverage requirements and varying ambient interference conditions. The signal coverage optimization model considers these factors to determine the appropriate adjustment plan.

[0076] The generated adjustment instructions drive the antenna system's servo mechanism to perform dynamic adjustments, adjusting the antenna system's beam pointing according to the instructions. After the adjustment is completed, the antenna system's parameter status database is promptly updated, and the adjusted parameter information is accurately recorded for subsequent analysis and management.

[0077] The present invention will be further described below in conjunction with Examples 1 to 5:

[0078] Example 1: This example describes in detail the process of building a dynamic error compensation model. When building a dynamic error compensation model, first obtain a historical positioning data set and corresponding error annotation data, where the error annotation data includes the multipath effect type and error amplitude. Assume that the historical positioning data set is D, and each positioning data sample is d i , i=1,2,...,n (n is the number of data samples), the corresponding error annotation data is e i , e i Contains multipath effect type t i and error amplitude a i。The training subsets are divided according to the multipath effect type and error amplitude. Each training subset corresponds to a typical error scenario. For example, the multipath effect type is divided into direct wave and reflected wave interference type, multiple reflected wave interference type, etc. The training subsets are divided for data of different types with similar error amplitude ranges. Let the divided training subset be S j , j = 1, 2, ..., m (m is the number of training subsets). These training subsets are used to train the initial compensation model in parallel until the initial compensation model's compensation accuracy for each error scenario is greater than or equal to the preset first threshold T1. Training is terminated to obtain the intermediate compensation model. During training, model parameters are continuously adjusted to improve compensation accuracy. The historical positioning dataset D is input into the intermediate compensation model, and the compensation results output by the intermediate compensation model are verified to meet the preset error tolerance. If so, the intermediate compensation model is designated as the dynamic error compensation model. This dynamic error compensation model can effectively adapt to different error scenarios and improve its ability to compensate for positioning data errors.

[0079] In practical applications, a large amount of historical positioning data, such as the antenna feed system of a city's communication network base station, is collected. This data includes positioning data under different weather conditions, time periods, and surrounding environments, along with corresponding error annotations. Through analysis and processing of this data, multiple training subsets are generated, corresponding to different multipath effect types and error amplitude ranges. The initial compensation model is trained in parallel using these training subsets. After multiple iterations, training is terminated when the model's compensation accuracy for the error scenarios corresponding to each training subset reaches a preset first threshold, T1 (e.g., 90%). The intermediate compensation model is then trained by re-entering the historical positioning data set into the intermediate compensation model for verification. If the compensation results meet the preset error tolerance (e.g., within ±5 meters), the intermediate compensation model is determined as the final dynamic error compensation model. This model plays a key role in subsequent error compensation of the antenna feed system's positioning data, effectively reducing the impact of multipath error and sensor noise on positioning data.

[0080] Example 2, this example focuses on the specific implementation method of collecting positioning data through the Beidou differential positioning module. First, establish a communication connection with the target Beidou receiver, and the target Beidou receiver is deployed at the preset monitoring node of the antenna feed system. During actual installation, according to the structure and monitoring requirements of the antenna feed system, the location of the monitoring node is reasonably selected to ensure that the Beidou receiver can accurately obtain the location information of the antenna feed system. The original positioning data stream of the target Beidou receiver is periodically read according to the preset sampling frequency f, and the positioning timestamp is marked based on the timing characteristics of the data stream. Assume that at t k The original positioning data stream read at the moment is P k , then the timestamp is t kAccording to the spatial topology of the antenna feed system, the original positioning data streams of different monitoring nodes at the same timestamp are spatially aligned to form a spatially associated positioning data set. For example, there are N monitoring nodes in the antenna feed system. At time t, the original positioning data streams obtained from each monitoring node are P 1,t 、P 2,t ,...,P n,t , through the spatial alignment algorithm, these data streams are integrated into a spatially associated positioning dataset P t The positioning data set processed in this way can more comprehensively and accurately reflect the position status of the antenna feeder system, providing a reliable data basis for subsequent data analysis and processing.

[0081] In the construction of a communication base station in a mountainous area, due to the complex terrain, the position monitoring of the antenna feed system is particularly important. The antenna feed system of the base station is set up with 5 preset monitoring nodes, each of which is equipped with a target BeiDou receiver. The preset sampling frequency f is set to 10Hz, that is, the raw positioning data stream is read every 0.1 seconds. At a certain time t, the raw positioning data streams obtained from the 5 monitoring nodes are P 1,t 、P 2,t 、P 3,t 、P 4,t 、P 5,t According to the spatial topology of the antenna system, a specially designed spatial alignment algorithm is used to integrate these data streams. The algorithm first determines the relative position relationship of each monitoring node, and then matches and merges the data streams with the same timestamp according to the coordinate information in the positioning data, and finally forms a spatially associated positioning data set P. t Collecting and processing positioning data in this way effectively improves the accuracy of antenna and feeder system position monitoring in complex mountainous environments, providing strong support for subsequent real-time monitoring and adjustment.

[0082] Example 3. This example focuses on the specific operation of inputting positioning data into a preset dynamic error compensation model. First, the abnormal offset segment in the positioning data is extracted. The abnormal offset segment is a data segment whose position offset exceeds the preset offset threshold △ within a continuous time window W. Assume that the positioning data sequence is L(t). In the time window [t1, t 1+ W], if |L(t 1+If W)-L(t1)|>△, the data segment is an abnormal offset segment. An error evaluation metric, E, is generated based on the duration T and offset amplitude A of the abnormal offset segment, where E=f(T,A) (f is a function determined based on the actual situation). A corresponding compensation algorithm is dynamically selected based on the error evaluation metric. For short-duration, high-amplitude offsets (i.e., T≥T0 and A≤A0), a least-squares fitting algorithm is used. In practical applications, using appropriate compensation algorithms for different types of errors can more effectively correct errors in positioning data and improve data accuracy.

[0083] During monitoring of a communication base station antenna feeder system in a coastal area, positioning data was significantly affected by multipath effects in the marine environment. Over a period of time, abnormal positioning data drift was detected. By setting the continuous time window W to 10 seconds and the offset threshold △ to 10 meters, it was found that some data segments exhibited position drift exceeding 10 meters within 10 seconds. These segments were identified as abnormal drift segments. These abnormal drift segments were analyzed to calculate their duration T and amplitude A. For example, one abnormal drift segment had a duration T of 5 seconds and an amplitude A of 15 meters. Since T < T0 (assuming T0 is 8 seconds) and A ≥ A0 (assuming A0 is 12 meters), this was a short-duration, high-amplitude drift, so a Kalman filter algorithm was used for compensation. Another abnormal drift segment had a duration T of 12 seconds and an amplitude A of 8 meters. Since T ≥ T0 and A ≤ A0, this was a long-duration, low-amplitude drift, so a least-squares fitting algorithm was used for compensation. By dynamically selecting the compensation algorithm based on the error characteristics, the correction effect of positioning data errors is significantly improved, providing more accurate data for subsequent signal coverage optimization.

[0084] Example 4: This example introduces in detail the adjustment steps of the signal coverage optimization model. The signal coverage optimization model first constructs a spatial signal strength distribution map based on the beam coverage of the antenna feed system. Assume that there are M preset geographical coordinate points within the beam coverage of the antenna feed system, and each coordinate point is G i, i=1,2,...,m Each distribution node corresponds to a preset geographical coordinate, and the signal strength value S is measured or calculated at each coordinate point. i , thus constructing a spatial signal strength distribution map. The coverage optimization weight is calculated based on the signal strength gradient difference of adjacent distribution nodes. Let the adjacent nodes G i, G i+1 The signal strengths are S i and S i+1 , the signal intensity gradient difference is |S i+1- S i |, based on this difference, a certain algorithm is used to calculate the coverage optimization weight W i (W i =g(|S i+1- Si |), where g is a function relationship determined based on actual conditions. Dynamic adjustment of the target beam pointing is performed based on time series changes in operating parameters. Operating parameters such as signal strength and standing wave ratio (SWR) change over time. By analyzing the time series of these parameters and comprehensively considering coverage optimization weights, the adjustment direction and magnitude of the target beam pointing are determined, thereby dynamically optimizing the antenna system's beam pointing to meet signal coverage requirements in the target area.

[0085] In a communication network optimization project in a certain city's commercial district, 100 preset geographic coordinate points were set within the antenna feed system beam coverage area. Signal strength monitoring equipment was installed at these coordinate points to obtain the signal strength value S at each point. i , and then construct a spatial signal strength distribution map. For example, the signal strengths of two adjacent coordinate points G5 and G6 are S5 and S6 respectively, and the signal strength gradient difference |S 6- S5|. Based on the pre-set algorithm g, the coverage optimization weight W5 is calculated. Meanwhile, long-term monitoring of antenna and feeder system operating parameters, such as signal strength and standing wave ratio (SWR), is performed to generate time series data. Analysis of this time series data reveals that during peak hours on weekdays, signal strength fluctuates significantly and the SWR increases. In conjunction with the coverage optimization weights, the target beam direction is dynamically adjusted. During peak hours, the beam intensity directed toward densely populated commercial areas is appropriately increased, while the beam angle is adjusted to avoid interference sources. This effectively improves signal coverage quality in this area and meets user demand for communication signals during peak hours.

[0086] Example 5. This example mainly describes the update of the parameter status library and related data management operations. When updating the parameter status library, the first-level storage tags are divided according to the parameter type. The first-level storage tags include positioning class, signal strength class, and standing wave ratio class, etc. Under each level of storage tags, the second-level storage sub-tags are further divided based on the parameter value range. For example, under the positioning class tag, the second-level sub-tags are divided according to the latitude and longitude range; under the signal strength class tag, the second-level sub-tags are divided according to the strong and weak intervals of the signal strength. The classified parameter data is stored in different partitions of the distributed database according to the tag level. This storage method facilitates data management and query. At the same time, the access key of the storage tag is configured according to the preset data permissions. Assuming that different user roles have different access rights, a corresponding access key K is set for each storage tag. j ,j=1,2,...q,(q is the number of stored tags). When receiving a data query request, verify the key K provided by the requester rep= K j , then the data access interface corresponding to the storage tag is opened, otherwise the access is denied, thus ensuring the security and privacy of the data.

[0087] In the network management system of a large telecommunications operator, the parameters of antenna feeder systems distributed in various locations are managed. The parameters of the antenna feeder system are divided into first-level storage tags such as positioning, signal strength, and standing wave ratio. Under the positioning tag, the world is divided into multiple regions according to the longitude and latitude range, and each region corresponds to a second-level storage sub-tag; under the signal strength tag, it is divided into multiple intervals such as strong signal, medium signal, and weak signal according to the signal strength value, and each interval corresponds to a second-level storage sub-tag. The collected antenna feeder system parameter data is stored in different partitions of the distributed database according to these tag levels. For example, the positioning data of a certain antenna feeder system is stored in the corresponding second-level sub-tag partition under the positioning tag according to its longitude and latitude information; its signal strength data is stored in the corresponding second-level sub-tag partition under the signal strength tag according to the signal strength value. At the same time, in order to ensure data security, different access keys are configured according to different user roles, such as network administrators, technical maintenance personnel, etc. When the network administrator sends a data query request, the system will verify the key K provided by it. rep Whether to store the access key K of the tag with the target j If there is a match, the corresponding data access interface is opened, and the administrator can obtain the required data for network analysis and management; if there is a mismatch, access is denied, effectively preventing data leakage and illegal access, and ensuring the security and stability of communication network data.

[0088] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0089] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying defects in key components of power transmission lines based on cloud-edge collaboration, characterized in that: include: Acquire a multi-source inspection data set; the multi-source inspection data includes transmission line component images collected by drones, edge sensor data, and a cloud-based historical defect database; the edge sensor data includes temperature, humidity, and vibration information; Based on the transmission line component image, extracting component regional features through an edge-end image preprocessing algorithm, wherein the regional features include geometric contour information, texture distribution features, and local defect candidate frames; Generate environmental parameters based on the edge sensor data using a multimodal feature fusion algorithm, where the environmental parameters include a temperature anomaly coefficient, a humidity influencing factor, and a vibration fluctuation threshold; Performing spatiotemporal correlation analysis on the cloud-based historical defect database to generate a component defect evolution map; Inputting the component regional characteristics, environmental correlation parameters and defect evolution map into the cloud-edge collaborative recognition model to generate a component defect feature vector; Based on the component defect feature vector, a multi-level defect recognition network is constructed through a dynamic optimization algorithm to output component defect classification results and confidence levels; the levels of the multi-level defect recognition network represent defect type branches, and the nodes represent feature fusion weights and recognition priorities; The cloud-edge collaborative recognition model includes a feature enhancement module and a collaborative reasoning module. The feature enhancement module includes: Normalizing the geometric contour information in the component region feature to obtain a first enhancement vector; Discretize and encode the temperature anomaly coefficient in the environment-related parameters to generate a second enhancement vector; Performing a spatiotemporal convolution operation on the defect evolution map to extract defect propagation characteristics and obtain a third enhancement vector; The first enhancement vector, the second enhancement vector and the third enhancement vector are merged into a high-dimensional fusion feature through a feature concatenation layer; The collaborative reasoning module includes: Perform channel attention weighting on high-dimensional fusion features to generate a feature importance matrix; Extract defect correlation features through cross-level feature interaction algorithm and generate defect correlation matrix; Perform element-by-element product operation on the feature importance matrix and the defect correlation matrix to generate collaborative reasoning features; The collaborative inference features are superimposed with the original high-dimensional fusion features through residual connection to output the component defect feature vector.

2. The method for identifying defects in key components of a power transmission line based on cloud-edge collaboration according to claim 1 is characterized in that: The extracting of component region features by edge-end image preprocessing algorithm includes: performing distortion correction processing on the image of the power transmission line component to generate a standardized image; Based on the predefined component template library, the key component area is located through the region matching algorithm, and the component image blocks are segmented; A multi-scale filtering algorithm is used to extract the texture gradient features of the component image block and generate candidate defect areas in combination with the geometric contour information; The candidate defect area is compared with the reference parameters in the component template library to generate a local defect candidate frame and a confidence score.

3. The method for identifying defects in key components of power transmission lines based on cloud-edge collaboration according to claim 1 is characterized in that: The generating of environment-related parameters by a multimodal feature fusion algorithm includes: Performing time domain alignment processing on the edge sensor data to eliminate acquisition time deviation; Calculate the mean and variance of temperature, humidity, and vibration data based on the sliding window statistical method to generate a dynamic environmental baseline; Calculate the temperature anomaly coefficient and humidity impact factor based on the difference between the dynamic environmental baseline and the environmental threshold; The spectral characteristics of the vibration data are compared with the preset vibration fluctuation threshold to generate a vibration anomaly index and encode it as an environment-related parameter.

4. The method for identifying defects in key components of power transmission lines based on cloud-edge collaboration according to claim 1 is characterized in that: The multi-level defect recognition network is constructed by a dynamic optimization algorithm, including: Initialize hierarchical attributes according to defect type branches and generate a weight assignment matrix based on identification priority; The component defect feature vector is used as an input feature, and a weight allocation matrix is ​​composed of feature fusion weights and recognition priorities; The weight parameters of each layer are iteratively updated through the gradient back-propagation algorithm to adjust the weight distribution matrix; The optimal classification sequence covering all defect types is generated based on the adjusted weight distribution matrix.

5. The method for identifying defects in key components of a power transmission line based on cloud-edge collaboration according to claim 2 is characterized in that: The method for constructing the component template library includes: Collect standard image samples of various typical transmission line components and extract reference geometric contours and texture features; Perform affine transformation on the reference geometric outline to generate a multi-view component template; Categorize templates according to component type and associate with a benchmark parameter database; The classified templates are stored as a component template library.

6. The method for identifying defects in key components of power transmission lines based on cloud-edge collaboration according to claim 3 is characterized in that: The parameter optimization method of the dynamic environment baseline includes: Calculate the initial sliding window length and statistical interval based on the historical sensor data distribution; The grid search algorithm is used to traverse the parameter combinations and select the parameters with the best baseline stability and anomaly detection sensitivity; The sliding window length and statistical interval are dynamically adjusted according to the detection results to optimize the calculation accuracy of environmental correlation parameters.

7. The method for identifying defects in key components of a power transmission line based on cloud-edge collaboration according to claim 4 is characterized in that: The parameter updating method of the gradient back propagation algorithm includes: The loss function between layers is defined as the sum of the classification error and the regularization term of the feature fusion weight; Initialize the weight parameters of each layer to random values ​​and the input features to normalized vectors; Calculate the gradient update of each layer based on the previous layer using the chain rule and record the optimal weight path; The weight distribution matrix is ​​updated based on the back propagation of the optimal weight path to generate the final classification result.

8. A system for identifying defects in key components of power transmission lines based on cloud-edge collaboration, characterized in that: include: Multi-source data acquisition module: used to obtain a multi-source inspection data set, including transmission line component images collected by drones, edge sensor data, and a cloud-based historical defect database; wherein the edge sensor data includes temperature, humidity, and vibration information; Edge feature extraction module: configured to extract component regional features based on the transmission line component image through an edge end image preprocessing algorithm, wherein the regional features include geometric contour information, texture distribution features and local defect candidate frames; Environmental parameter fusion module: used to generate environmental parameters based on the edge sensor data through a multimodal feature fusion algorithm, wherein the environmental parameters include temperature anomaly coefficient, humidity influence factor and vibration fluctuation threshold; Cloud-based graph generation module: performs spatiotemporal correlation analysis on the cloud-based historical defect database to generate a component defect evolution graph; Collaborative identification module: Inputs the component regional features, environmental correlation parameters and defect evolution map into the cloud-edge collaborative identification model to generate a component defect feature vector; the cloud-edge collaborative identification model includes a feature enhancement module and a collaborative reasoning module; Defect classification module: used to construct a multi-level defect recognition network based on the component defect feature vector through a dynamic optimization algorithm, and output the component defect classification results and confidence levels; the levels of the multi-level defect recognition network represent defect type branches, and the nodes represent feature fusion weights and recognition priorities; The cloud-edge collaborative recognition model includes a feature enhancement module and a collaborative reasoning module. The feature enhancement module includes: Normalizing the geometric contour information in the component region feature to obtain a first enhancement vector; Discretize and encode the temperature anomaly coefficient in the environment-related parameters to generate a second enhancement vector; Performing a spatiotemporal convolution operation on the defect evolution map to extract defect propagation characteristics and obtain a third enhancement vector; The first enhancement vector, the second enhancement vector and the third enhancement vector are merged into a high-dimensional fusion feature through a feature concatenation layer; The collaborative reasoning module includes: Perform channel attention weighting on high-dimensional fusion features to generate a feature importance matrix; Extract defect correlation features through cross-level feature interaction algorithm and generate defect correlation matrix; Perform element-by-element product operation on the feature importance matrix and the defect correlation matrix to generate collaborative reasoning features; The collaborative inference features are superimposed with the original high-dimensional fusion features through residual connection to output the component defect feature vector.

Images