Power transmission line defect identification method and device, computer equipment and program product

Abstract

The invention relates to a power transmission line defect identification method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining multi-source data of a power transmission line, and performing format unification, data cleaning and data enhancement processing on the multi-source data to obtain processed multi-source data; adjusting a fuzzy neural network model based on environment perception data in the processed real-time operation data, and inputting the processed image data into the adjusted fuzzy neural network model to obtain enhanced image data; extracting visual features and semantic features, and fusing the visual features and the semantic features to obtain a fused feature vector; performing quantitative evaluation on the fusion feature vector, and determining the hazard level of each type of defect; and aligning the visual features with the semantic features, and generating an identification result of the power transmission line defect based on the hazard level and the fusion feature vector. By adopting the method, the defect identification accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of power transmission line operation and maintenance technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium and computer program product for identifying power transmission line defects. Background Technology

[0002] As a core component of the power system, the safe and stable operation of transmission lines directly affects the reliability of power supply. With the continuous expansion of the power network, transmission lines cover a wide area, facing complex and diverse terrain and climate conditions, and numerous and highly concealed types of defects, which places higher demands on the accurate identification of defects.

[0003] Among the relevant technologies, existing power transmission defect identification technologies have many shortcomings. In terms of data processing, data acquisition lacks systematic planning, the formats of multi-source data are not uniform, and the processing of duplicate, abnormal and invalid data during the preprocessing process is not thorough enough, resulting in inconsistent data quality, which makes it difficult to support subsequent accurate analysis. Furthermore, the data processing methods are relatively simple, which leads to inaccurate final defect identification results. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for identifying transmission line defects that can improve the accuracy of transmission line defect identification, in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a method for identifying defects in transmission lines, including:

[0006] Multi-source data of the transmission line is acquired, and the multi-source data is subjected to format unification, data cleaning and data enhancement processing to obtain processed multi-source data; the multi-source data includes image data, real-time operation data and historical defect data of the transmission line.

[0007] Based on the environmental perception data in the processed real-time running data, the fuzzy neural network model is adjusted, and the processed image data is input into the adjusted fuzzy neural network model to obtain the enhanced image data.

[0008] Visual features are extracted from the enhanced image data and semantic features are extracted from the processed historical defect data. The visual features and the semantic features are then fused to obtain a fused feature vector.

[0009] For each type of defect, based on preset rules and a defect detection model, the fused feature vector is quantitatively evaluated to determine the hazard level of each type of defect;

[0010] The visual features are aligned with the semantic features, and based on the hazard level and the fused feature vector, the identification result of the transmission line defect is generated.

[0011] In one embodiment, the process of performing format unification, data cleaning, and data augmentation on the multi-source data to obtain processed multi-source data includes:

[0012] The multi-source data is converted into a preset format according to the data type to obtain the converted multi-source data;

[0013] Duplicate images are removed from the image data in the transformed multi-source data, text data lacking key fields is deleted from the transformed multi-source data, and outliers are removed from the transformed multi-source data to obtain the processed multi-source data.

[0014] In one embodiment, the environmental perception data includes light intensity, terrain vibration frequency, and climate type; the adjustment of the fuzzy neural network model based on the environmental perception data in the processed real-time operational data includes:

[0015] Based on the light intensity, adjust the contrast enhancement parameter and brightness adjustment parameter in the fuzzy neural network model;

[0016] Based on the terrain vibration frequency, adjust the filter kernel size and iteration number of the fuzzy neural network model;

[0017] Based on the climate type, the processing method of the fuzzy neural network model is adjusted.

[0018] In one embodiment, fusing the visual features and the semantic features to obtain a fused feature vector includes:

[0019] Remove redundant dimensions from the visual features and retain the features in the visual features whose mutual information value with the preset defect type is greater than a first preset threshold to form the target visual features;

[0020] The semantic features are subjected to dimensionality reduction processing, and semantic features with a cumulative variance contribution rate higher than a second preset threshold are selected as target semantic features.

[0021] Based on the target visual features and the target semantic features, a weighted sum is performed to obtain a fused feature vector.

[0022] In one embodiment, the preset defect types include bird nest defects, external damage defects, and metal equipment corrosion defects.

[0023] In one embodiment, the step of quantitatively evaluating the fused feature vector based on preset rules and a defect detection model for each type of defect to determine the hazard level of each type of defect includes:

[0024] The location of the bird's nest is determined, and based on the location, the volume of the bird's nest, the spatial distance between the bird's nest and the power transmission line, whether it contains metallic foreign objects, and whether there are protective measures, the first quantitative assessment result of the bird's nest defect is determined.

[0025] The three-dimensional positional relationship between the external damage source and the transmission line is reconstructed, and the second quantitative assessment result of the external damage defect is determined based on the rules of external damage source type, spatial distance and direction of movement.

[0026] Acquire image data, depth data, and microstructure data of rust on metal equipment; and determine a third quantitative assessment result of the rust defects on the metal equipment based on the image data, the depth data, and the microstructure data.

[0027] The first quantitative assessment result, the second quantitative assessment result, and the third quantitative assessment result are mapped to preset hazard levels to obtain the hazard level of each type of defect.

[0028] Secondly, this application also provides a transmission line defect identification device, comprising:

[0029] The acquisition module is used to acquire multi-source data of the transmission line, and to perform format unification, data cleaning and data enhancement processing on the multi-source data to obtain processed multi-source data; the multi-source data includes image data, real-time operation data and historical defect data of the transmission line;

[0030] The enhancement module is used to adjust the fuzzy neural network model based on the environmental perception data in the processed real-time running data. The processed image data is input into the adjusted fuzzy neural network model to obtain the enhanced image data.

[0031] The fusion module is used to extract visual features from the enhanced image data and semantic features from the processed historical defect data, and fuse the visual features and the semantic features to obtain a fused feature vector;

[0032] The determination module is used to quantitatively evaluate the fused feature vector based on preset rules and a defect detection model for each type of defect, and determine the hazard level of each type of defect.

[0033] The generation module is used to align the visual features with the semantic features and generate the identification result of the transmission line defect based on the hazard level and the fused feature vector.

[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0035] Multi-source data of the transmission line is acquired, and the multi-source data is subjected to format unification, data cleaning and data enhancement processing to obtain processed multi-source data; the multi-source data includes image data, real-time operation data and historical defect data of the transmission line.

[0036] Based on the environmental perception data in the processed real-time running data, the fuzzy neural network model is adjusted, and the processed image data is input into the adjusted fuzzy neural network model to obtain the enhanced image data.

[0037] Visual features are extracted from the enhanced image data and semantic features are extracted from the processed historical defect data. The visual features and the semantic features are then fused to obtain a fused feature vector.

[0038] For each type of defect, based on preset rules and a defect detection model, the fused feature vector is quantitatively evaluated to determine the hazard level of each type of defect;

[0039] The visual features are aligned with the semantic features, and based on the hazard level and the fused feature vector, the identification result of the transmission line defect is generated.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0041] Multi-source data of the transmission line is acquired, and the multi-source data is subjected to format unification, data cleaning and data enhancement processing to obtain processed multi-source data; the multi-source data includes image data, real-time operation data and historical defect data of the transmission line.

[0042] Based on the environmental perception data in the processed real-time running data, the fuzzy neural network model is adjusted, and the processed image data is input into the adjusted fuzzy neural network model to obtain the enhanced image data.

[0043] Visual features are extracted from the enhanced image data and semantic features are extracted from the processed historical defect data. The visual features and the semantic features are then fused to obtain a fused feature vector.

[0044] For each type of defect, based on preset rules and a defect detection model, the fused feature vector is quantitatively evaluated to determine the hazard level of each type of defect;

[0045] The visual features are aligned with the semantic features, and based on the hazard level and the fused feature vector, the identification result of the transmission line defect is generated.

[0046] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0047] Multi-source data of the transmission line is acquired, and the multi-source data is subjected to format unification, data cleaning and data enhancement processing to obtain processed multi-source data; the multi-source data includes image data, real-time operation data and historical defect data of the transmission line.

[0048] Based on the environmental perception data in the processed real-time running data, the fuzzy neural network model is adjusted, and the processed image data is input into the adjusted fuzzy neural network model to obtain the enhanced image data.

[0049] Visual features are extracted from the enhanced image data and semantic features are extracted from the processed historical defect data. The visual features and the semantic features are then fused to obtain a fused feature vector.

[0050] For each type of defect, based on preset rules and a defect detection model, the fused feature vector is quantitatively evaluated to determine the hazard level of each type of defect;

[0051] The visual features are aligned with the semantic features, and based on the hazard level and the fused feature vector, the identification result of the transmission line defect is generated.

[0052] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for identifying transmission line defects first acquire multi-source data of the transmission line, and then perform format unification, data cleaning, and data enhancement processing on the multi-source data to obtain processed multi-source data. Based on environmental perception data in the processed real-time operational data, the fuzzy neural network model is adjusted, and the processed image data is input into the adjusted fuzzy neural network model to obtain enhanced image data. Visual features from the enhanced image data and semantic features from the processed historical defect data are extracted, and the visual and semantic features are fused to obtain a fused feature vector. For each type of defect, based on preset rules and a defect detection model, the fused feature vector is quantitatively evaluated to determine the hazard level of each type of defect. The visual and semantic features are aligned, and based on the hazard level and the fused feature vector, the identification result of the transmission line defect is generated. In this way, through multi-source data integration and standardized preprocessing, the unification of different types of data formats is achieved, invalid data is effectively eliminated, and data diversity is improved, providing high-quality data support for subsequent stages, ensuring the reliability of the identification work from the source, and making the final defect identification result more accurate. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is an application environment diagram of the transmission line defect identification method in one embodiment;

[0055] Figure 2 This is a flowchart illustrating a method for identifying defects in transmission lines in one embodiment;

[0056] Figure 3 This is a structural block diagram of a transmission line defect identification device in one embodiment;

[0057] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0060] The transmission line defect identification method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart vehicle devices, projection devices, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0061] In one exemplary embodiment, such as Figure 2 As shown, a method for identifying defects in transmission lines is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 210. Wherein:

[0062] Step 202: Obtain multi-source data of the transmission line, and perform format unification, data cleaning and data enhancement processing on the multi-source data to obtain processed multi-source data.

[0063] The multi-source data includes image data of transmission lines, real-time operation data, and historical defect data.

[0064] For example, image data, real-time operation data, and historical defect data of transmission lines are acquired, and the data are subjected to format unification, data cleaning, and data enhancement processing to obtain processed multi-source data.

[0065] In one embodiment, transmission lines in different geographical regions such as mountains, plains, coasts, with voltage levels of 110 kV, 220 kV, 500 kV, and under climate scenarios such as heavy rain, fog, high temperature, and low temperature can be selected as the acquisition areas. Image data is collected by a drone flying at a speed of 5 meters per second and collecting 1 image every 50 meters, a robot with an inspection frequency of once every 2 hours, and a handheld inspection instrument by an operator with a resolution of 4K.

[0066] Obtain real-time operation data from the transmission line monitoring system, for example, current, voltage, temperature; extract historical defect data including defect types, treatment plans, recurrence situations, etc. from the inspection reports and fault record databases of historical defect data.

[0067] In another embodiment, complex fault data with multiple defects superimposed, such as a bird's nest plus corrosion, and defect data under extreme weather conditions, such as defect data under heavy snow, can also be simulated and generated through MATLAB.

[0068] Step 204: Based on the environmental perception data in the processed real-time operation data, adjust the fuzzy neural network model, and input the processed image data into the adjusted fuzzy neural network model to obtain enhanced image data.

[0069] Among them, the fuzzy neural network includes an input layer, a hidden layer, and an output layer.

[0070] Optionally, construct a fuzzy neural network model, where the input layer receives the environmental perception data and the processed image data in the processed real-time operation data, enhances the image data through the hidden layer, and outputs the enhanced image data through the output layer.

[0071] Step 206: Extract the visual features in the enhanced image data and the semantic features in the processed historical defect data, and fuse the visual features and semantic features to obtain a fused feature vector.

[0072] Exemplarily, extract the visual features in the enhanced image data through the ResNet50 network. Input the enhanced image data into the network, perform max-pooling operations after convolutional processing, and output the visual features. The TF-IDF (Frequency-Inverse Document Frequency) algorithm is used to extract the semantic features in the processed historical defect data.

[0073] In one embodiment, perform Jieba word segmentation on the text in the processed historical defect data, remove stop words, such as "de", "zai", "he", etc., calculate the term frequency-inverse document frequency, and output the semantic features.

[0074] Among them, visual features include image information such as shape, color, and texture of image data; semantic features include information such as device model, historical faults, and environmental parameters.

[0075] Step 208: For each type of defect, based on preset rules and defect detection models, the fused feature vector is quantitatively evaluated to determine the hazard level of each type of defect.

[0076] Optionally, for each type of transmission line defect, the fused feature vector is quantitatively evaluated according to preset rules and defect detection models to determine the hazard level of each type of defect.

[0077] Step 210: Align the visual features with the semantic features, and generate the identification results of transmission line defects based on the hazard level and the fused feature vector.

[0078] For example, a Transformer-based cross-modal mapping model is constructed. On the image modality side, the coordinate information contained in the defect detection box and the pixel-level contour contained in the segmentation mask are converted into 1024-dimensional standardized visual features. The position of the detection box is processed by coordinate normalization, and the key shape features of the segmentation mask are extracted by contour sampling to eliminate the format differences of different representations within the same modality. On the language modality side, the BERT-Base model is used to convert long and short text descriptions, such as "There is a bird's nest on tower #35" or "There is a bird's nest with a volume of 0.8 cubic meters 2 meters away from the conductor on the left side of tower #35", and various types of tabular data, such as Excel tables containing defect type, hazard level, and location, into 1024-dimensional structured semantic features to unify the language information format.

[0079] By comparing and learning loss functions, the loss weights are set to minimize the distance between the visual and semantic features of the same defect in the feature space, and to maximize the distance between the feature vectors of different defects. Through iterative training, the batch size and learning rate are set, and the model learns the cross-modal mapping relationship in the power transmission defect scenario, eliminating the semantic gap and realizing a one-to-one correspondence between visual and semantic information.

[0080] Based on the aligned cross-modal information, a standardized natural language text is generated that includes the defect location line number, tower number, specific location, defect type (bird nest, external damage, corrosion), hazard level (mild, moderate, severe), key features such as bird nest volume, distance from external damage source, corrosion depth, and potential risks such as potential short circuits or equipment breakage. For example, a bird nest with a volume of 0.8 cubic meters is located 2 meters from the conductor on the left side of tower #35 of the 220kV Dianqian Line. The minimum distance from the conductor is 0.3 meters, and there are no protective measures. This is judged as a severe risk and may cause a short circuit fault. It is recommended to remove it within 24 hours.

[0081] In one embodiment, the transmission line defect identification method can also integrate the modules that implement the above method based on a microservice architecture.

[0082] The acquisition, enhancement, fusion, determination, and generation modules communicate via interfaces. The PyInstaller tool is used to encapsulate the model and its dependent libraries, including TensorFlow 2.8, PyTorch 1.12, OpenCV 4.5, configuration files containing model parameters, interface addresses, data format rules, and startup scripts, into independent binary files, hiding the internal implementation details. A unified HTTP input / output interface is provided externally: the input is JSON format data containing "image data, device number, inspection time, and environmental parameters", and the output is JSON format results containing structured text.

[0083] In another embodiment, hardware performance testing tools such as CPU-Z and GPU-Z are used to evaluate parameters of the target operating environment, such as the number of CPU cores, clock speed, memory capacity, GPU model, and video memory, to optimize the allocation of computing resources, for example, allocating 8 cores of computing resources to the CPU device and video memory to the GPU device; conditional compilation and the use of the LoadLibrary function are employed to ensure that the model runs stably on different operating systems (e.g., CentOS 7.9, Windows Server 2019); a virtual environment is created using Anaconda to isolate the model's dependent libraries from the system global environment, avoiding runtime errors caused by dependency library version conflicts; for dependent libraries that require specific versions of compilation, such as OpenCV_contrib, the source code is manually compiled and installed to ensure compatibility with the target hardware environment.

[0084] A data format conversion tool is used to achieve bidirectional data adaptation between the model and the existing inspection system. A data format mapping table is established; for example, the system output defect code 001 is mapped to the model input "Bird's Nest" defect, and the model output "severe risk" is mapped to the system-identified risk level 3. Based on the transmission protocol used by the existing system, the corresponding network programming library is selected: for example, for HTTP / HTTPS protocols, Python's requests library is used to send and receive data, handling HTTP request GET / POST methods and status codes such as 200 / 400 / 500. SSL / TLS encryption technology is used to encrypt data transmission; when using the HTTPS protocol, for example, a Symantec SSL certificate is configured to establish a secure encrypted channel. For sensitive data such as equipment parameters and defect locations, AES-256 encryption is used for encryption before transmission, and the receiving end uses the corresponding decryption algorithm to restore the data. An OAuth 2.0 authentication mechanism is implemented to ensure that only legally authorized inspection systems can interact with the model, preventing unauthorized access and data leakage.

[0085] In one embodiment, Docker is used to package the model into a container image. A Dockerfile is written to define the container image building process, such as selecting Ubuntu 20.04 as the base image, installing Python 3.8 and its dependent libraries, copying the model code and configuration files, and setting the startup command. The image is then built using the command and pushed to a private image repository of the power grid, such as Harbor.

[0086] Manage container clusters using Kubernetes: Deploy 3 container instances, set CPU resource limits to 8 cores, memory limits to 32GB, and GPU limits to 1; configure HPA auto-scaling strategy to automatically increase the number of container instances to achieve load balancing when CPU utilization exceeds 70% or concurrent requests exceed 100.

[0087] Establish a comprehensive operation and maintenance plan and emergency response plan: monitor the model's running status, such as CPU utilization, memory usage, and inference speed, and set alarm thresholds, for example, triggering an alarm when CPU utilization exceeds 90%; record model running logs through the ELK log collection system to facilitate troubleshooting; and formulate emergency response plans, such as automatic restart when a container crashes and retrying when data transmission is interrupted, to ensure the long-term stable operation of the model.

[0088] The aforementioned method for identifying transmission line defects involves acquiring multi-source data of the transmission line, and then performing format unification, data cleaning, and data augmentation on this data to obtain processed multi-source data. Based on environmental perception data from the processed real-time operational data, a fuzzy neural network model is adjusted, and the processed image data is input into the adjusted fuzzy neural network model to obtain enhanced image data. Visual features from the enhanced image data and semantic features from the processed historical defect data are extracted, and the visual and semantic features are fused to obtain a fused feature vector. For each type of defect, the fused feature vector is quantitatively evaluated based on preset rules and a defect detection model to determine the hazard level of each type of defect. The visual and semantic features are aligned, and based on the hazard level and the fused feature vector, the identification result of the transmission line defect is generated. Thus, through multi-source data integration and standardized preprocessing, the unification of different data formats is achieved, invalid data is effectively eliminated, and data diversity is improved, providing high-quality data support for subsequent stages and ensuring the reliability of the identification work from the source, resulting in more accurate final defect identification results.

[0089] In an exemplary embodiment, the multi-source data is subjected to format unification, data cleaning, and data augmentation to obtain processed multi-source data, including: converting the multi-source data into a preset format according to the data type to obtain the converted multi-source data; removing duplicate images from the image data in the converted multi-source data, deleting text data lacking key fields in the converted multi-source data, and removing outliers in the converted multi-source data to obtain the processed multi-source data.

[0090] In practical implementation, image data from multi-source data is converted into images at a preset resolution, text data from multi-source data is converted into a preset text format, such as TXT format, real-time running data is converted into CSV format, duplicate images are removed from the converted image data, such as similar scenes captured in succession, text data lacking key fields in the converted multi-source data is deleted, such as data without pole numbers or optional times, and erroneously labeled data is removed, such as data that labels foreign objects as bird nests, and outliers in the converted multi-source data are removed, such as instantaneous overvoltage and extreme temperature data.

[0091] In one embodiment, the processed multi-source data is labeled, and OpenCV is used to perform image enhancement, such as setting the rotation angle to -15° to 15°, the scaling ratio to 0.8 to 1.2, adding Gaussian noise and setting the standard deviation to 0.01. The NLTK toolkit is used to perform text enhancement, such as replacing rust with rust damage, sentence restructuring such as adjusting the word order of defect descriptions, and supplementing information such as the inspection environment temperature of 35°C and humidity of 60% on March 15, 2024, to expand the data volume and obtain the processed multi-source data.

[0092] In the above embodiments, by processing multi-source data and standardizing the data to be processed, subsequent processing becomes more convenient.

[0093] In an exemplary embodiment, the environmental perception data includes light intensity, terrain vibration frequency, and climate type. Based on the environmental perception data in the processed real-time operational data, the fuzzy neural network model is adjusted, including: adjusting the contrast enhancement parameter and brightness adjustment parameter in the fuzzy neural network model based on light intensity; adjusting the filter kernel size and iteration number of the fuzzy neural network model based on terrain vibration frequency; and adjusting the processing method of the fuzzy neural network model based on climate type.

[0094] In practical implementation, based on light intensity, the contrast enhancement and brightness adjustment parameters in the fuzzy neural network model are adjusted. For example, in low-light scenes with light intensity below 500 lux, the contrast enhancement coefficient is optimized from 1.2 to 1.8 and the brightness adjustment parameter is adjusted from 0.2 to 0.3. Based on terrain vibration frequency, the filter kernel size and iteration number of the fuzzy neural network model are adjusted. For example, when the vibration frequency is in the range of 8 to 12 Hz, the degree of motion blur in the image is automatically detected, and the adaptive filter kernel size is adjusted from 3×3 to 7×7, and the iteration number is increased from 10 to 15. Based on climate type, the processing method of the fuzzy neural network model is adjusted. For example, in foggy scenes with visibility below 800 meters, the multi-scale transformation scale factor is set to 1.5 and fused with the deep neural network to remove the fog effect and restore image details.

[0095] In the above embodiments, the blurring is corrected by adjusting the filter kernel size and iteration number of the fuzzy neural network model, and the image detail recognition is improved by adjusting the contrast enhancement parameters and brightness adjustment parameters in the fuzzy neural network model.

[0096] In an exemplary embodiment, visual features and semantic features are fused to obtain a fused feature vector, including: removing redundant dimensions from the visual features, retaining features in the visual features whose mutual information value with a preset defect type is greater than a first preset threshold, forming target visual features; performing dimensionality reduction processing on the semantic features, and selecting semantic features whose cumulative variance contribution rate is higher than a second preset threshold as target semantic features; and performing weighted summation based on the target visual features and target semantic features to obtain the fused feature vector.

[0097] In practice, the mutual information value between the dimensions of the visual features and the preset defect type is calculated. Valid features with mutual information values ​​greater than the first preset threshold are retained, and redundant dimensions are removed to form the target visual features. Principal component analysis is used to reduce the dimensionality of the semantic features, and principal components with cumulative variance contribution rates higher than the second preset threshold are selected to form the target visual features.

[0098] An attention mechanism fusion model is constructed. The target visual features and target semantic features are calculated by the Softmax function, and preset weights are assigned. The two are then weighted and summed to generate a fused feature vector.

[0099] In one embodiment, a classification and recognition algorithm is developed based on support vector machine. The regularization parameter is set to 0.01, the kernel function is RBF and the gamma value is set to 0.001. The training set is 40,000 images and the test set is 10,000 images. The mapping relationship between the fused feature vector and the defect type is optimized through 5-fold cross-validation to achieve fine-grained feature representation and provide accurate feature support for defect classification.

[0100] In the above embodiments, by removing redundant dimensions in visual features and reducing the dimensionality of semantic features, the subsequent computational cost is reduced while retaining important features.

[0101] In one exemplary embodiment, the preset defect types include bird nest defects, external damage defects, and metal equipment corrosion defects.

[0102] In practice, the preset defect types include bird nest defects, external damage defects, and metal equipment corrosion defects.

[0103] The preset defect type can also be other defect types, which can be set by the user according to actual needs. This application embodiment does not limit this.

[0104] In the above embodiments, by clearly defining the specific defect type, the subsequent identification results are more intuitive and the processing efficiency is more efficient.

[0105] In an exemplary embodiment, for each type of defect, the fused feature vector is quantitatively evaluated based on preset rules and a defect detection model to determine the hazard level of each type of defect. This includes: locating the bird's nest and determining a first quantitative evaluation result for the bird's nest defect based on rules regarding location, nest volume, spatial distance between the nest and the transmission line, whether it contains metallic foreign objects, and whether protective measures exist; reconstructing the three-dimensional positional relationship between the external damage source and the transmission line and determining a second quantitative evaluation result for the external damage defect based on rules regarding the type of external damage source, spatial distance, and direction of movement; acquiring image data, depth data, and microstructure data of metal equipment corrosion and determining a third quantitative evaluation result for the metal equipment corrosion defect based on the image data, depth data, and microstructure data; and mapping the first, second, and third quantitative evaluation results to preset hazard levels to obtain the hazard level of each type of defect.

[0106] In practice, bird nest defect cases are identified from historical defect data. The distribution location, volume, and spatial distance of the bird nests from the transmission lines are extracted. These are combined with preset rules, such as the minimum distance from the conductor, the proportion of short circuits caused by foreign objects, the proportion of metal foreign objects in the bird nest, and surrounding protective measures such as whether bird spikes are provided. These rules are encoded into constraints of a neural network. For example, when the minimum distance between the bird nest and the conductor is less than 0.5 meters, the risk coefficient increases by 0.4; when bird spikes are provided, the risk coefficient decreases by 0.2.

[0107] The YOLOv8 algorithm is used to detect bird nests in the image regions corresponding to the fused feature vectors, and a confidence threshold is set for the detection. Preset rules are used to quantify and score the potential threat of bird nest defects, with a score range of 0 to 10. Among them, 0 to 3 points is mild risk, 3 to 6 points is moderate risk, and 6 to 10 points is severe risk. The hazard level of bird nest defects is output.

[0108] In one embodiment, monocular image data of external damage sources are determined from historical defect data to determine the type of external damage source, spatial distance, and direction of movement, such as excavator, transport vehicle, crane, actual distance from the power transmission line, and direction of movement, such as moving towards or away from the line.

[0109] A monocular depth prediction model is constructed based on a convolutional neural network. The input is the image features corresponding to the fused feature vector. A customized loss function is used to fuse distance error loss (MAE) and category loss (cross-entropy), and the weights of the two are set to optimize the model's adaptability to the power transmission channel environment. By learning from a large amount of labeled data, the model reconstructs the three-dimensional spatial structure of the power transmission line and the relative positional relationship with external damage sources, achieving accurate positioning of external damage sources and controlling the positioning error within a preset range. This is combined with hazard level determination; for example, when a crane is 3 meters away from the line and moving towards the line, it is determined to be a severe risk.

[0110] In another embodiment, multimodal image data, laser scanning corrosion depth data, and acoustic detection microstructure data are determined from historical defect data, and features such as corrosion color (e.g., yellowish-brown, reddish-brown), corrosion area percentage, corrosion depth, and equipment type (e.g., wires, poles, fittings) are extracted.

[0111] A fusion model of deep convolutional neural networks and graph convolutional networks is adopted to process the multimodal information corresponding to the corrosion feature vectors. The deep convolutional neural network extracts the corrosion color and area features, while the graph convolutional network mines the correlation between corrosion depth and equipment type. For example, when the corrosion depth of the wire joint is greater than 0.2 mm, the risk level is increased by one level. The contribution of each dimension of information is optimized by an adaptive weight allocation mechanism, which sets the weights of color features, area features, depth features, and equipment type. It adapts to different humidity and temperature ranges to achieve the classification of the severity of corrosion defects in metal equipment into mild, moderate, and severe, and risk assessment.

[0112] In the above embodiments, photos taken by different shifts and under different lighting conditions on the same conductor would previously yield two levels of conclusions; the algorithm maps color, depth, and load current to specific levels, eliminating subjective bias of the human eye and achieving comparability across towers, lines, and provinces.

[0113] To illustrate the transmission line defect identification method in this application in detail, an embodiment is described below. For example, this application describes a transmission line defect identification method in a specific scenario.

[0114] First, image data, real-time operation data, and historical defect data of the transmission line are acquired, and the data is formatted, cleaned, and augmented to obtain processed multi-source data.

[0115] A fuzzy neural network model is constructed. The input layer receives environmental perception data and processed image data from the processed real-time running data. The image data is enhanced through the hidden layer, and the enhanced image data is output through the output layer.

[0116] Visual features are extracted from the enhanced image data using the ResNet50 network. The enhanced image data is input into the network, processed by convolution, and then subjected to max pooling to output visual features. The semantic features in the processed historical defect data are then extracted using the TF-IDF (Frequency-Inverse Document Frequency) algorithm.

[0117] For each type of transmission line defect, the fused feature vector is quantitatively evaluated according to preset rules and defect detection models to determine the hazard level of each type of defect.

[0118] A Transformer-based cross-modal mapping model is constructed. On the image modality side, the coordinate information of the defect detection box and the pixel-level contour of the segmentation mask are converted into 1024-dimensional standardized visual features. The position of the detection box is processed by coordinate normalization, and the key shape features of the segmentation mask are extracted by contour sampling to eliminate the format differences of different representations within the same modality. On the language modality side, the BERT-Base model is used to convert long and short text descriptions, such as "There is a bird's nest on tower #35" or "There is a bird's nest with a volume of 0.8 cubic meters 2 meters away from the left conductor of tower #35", and various types of tabular data, such as Excel tables containing defect type, hazard level, and location, into 1024-dimensional structured semantic features to unify the language information format.

[0119] By comparing and learning loss functions, the loss weights are set to minimize the distance between the visual and semantic features of the same defect in the feature space, and to maximize the distance between the feature vectors of different defects. Through iterative training, the batch size and learning rate are set, and the model learns the cross-modal mapping relationship in the power transmission defect scenario, eliminating the semantic gap and realizing a one-to-one correspondence between visual and semantic information.

[0120] Based on the aligned cross-modal information, a standardized natural language text is generated that includes the defect location line number, tower number, specific location, defect type (bird nest, external damage, corrosion), hazard level (mild, moderate, severe), key features such as bird nest volume, distance from external damage source, corrosion depth, and potential risks such as potential short circuits or equipment breakage. For example, a bird nest with a volume of 0.8 cubic meters is located 2 meters from the conductor on the left side of tower #35 of the 220kV Dianqian Line. The minimum distance from the conductor is 0.3 meters, and there are no protective measures. This is judged as a severe risk and may cause a short circuit fault. It is recommended to remove it within 24 hours.

[0121] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0122] Based on the same inventive concept, this application also provides a transmission line defect identification device for implementing the above-described transmission line defect identification method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the transmission line defect identification device provided below can be found in the limitations of the transmission line defect identification method described above, and will not be repeated here.

[0123] In one exemplary embodiment, such as Figure 3 As shown, a transmission line defect identification device is provided, comprising: an acquisition module 301, an enhancement module 302, a fusion module 303, a determination module 304, and a generation module 305, wherein:

[0124] The acquisition module is used to acquire multi-source data of the transmission line and perform format unification, data cleaning and data enhancement processing on the multi-source data to obtain processed multi-source data; the multi-source data includes image data, real-time operation data and historical defect data of the transmission line.

[0125] The enhancement module is used to adjust the fuzzy neural network model based on the environmental perception data in the processed real-time running data. The processed image data is input into the adjusted fuzzy neural network model to obtain the enhanced image data.

[0126] The fusion module is used to extract visual features from the enhanced image data and semantic features from the processed historical defect data, and fuse the visual features and the semantic features to obtain a fused feature vector.

[0127] The determination module is used to quantitatively evaluate the fused feature vector based on preset rules and a defect detection model for each type of defect, and determine the hazard level of each type of defect.

[0128] The generation module is used to align the visual features with the semantic features and generate the identification result of the transmission line defect based on the hazard level and the fused feature vector.

[0129] In one exemplary embodiment, the above-described enhancement module is further configured to:

[0130] The multi-source data is converted into a preset format according to the data type to obtain the converted multi-source data;

[0131] Duplicate images are removed from the image data in the transformed multi-source data, text data missing key fields is deleted, and outliers are removed to obtain the processed multi-source data.

[0132] In one exemplary embodiment, the above-described apparatus further includes an adjustment module for:

[0133] Based on the light intensity, adjust the contrast enhancement parameters and brightness adjustment parameters in the fuzzy neural network model;

[0134] Based on the terrain vibration frequency, adjust the filter kernel size and iteration number of the fuzzy neural network model;

[0135] Adjust the processing method of the fuzzy neural network model based on climate type.

[0136] In one exemplary embodiment, the fusion module is further configured to:

[0137] Remove redundant dimensions from the visual features and retain the features whose mutual information value with the preset defect type is greater than the first preset threshold to form the target visual features;

[0138] The semantic features are dimensionality reduced, and the semantic features with a cumulative variance contribution rate higher than a second preset threshold are selected as the target semantic features.

[0139] Based on the target's visual features and semantic features, a weighted sum is performed to obtain a fused feature vector.

[0140] In one exemplary embodiment, the preset defect types include bird nest defects, external damage defects, and metal equipment corrosion defects.

[0141] In one exemplary embodiment, the determining module is further configured to:

[0142] The location of the bird's nest is determined based on its location, size, spatial distance from the power transmission line, presence of metallic foreign objects, and the existence of protective measures. This forms the first quantitative assessment result of the bird's nest's defects.

[0143] The three-dimensional positional relationship between the external damage source and the transmission line is reconstructed. Based on the rules of external damage source type, spatial distance and direction of movement, the second quantitative assessment result of the external damage defect is determined.

[0144] Acquire image data, depth data, and microstructure data of metal equipment corrosion, and determine the third quantitative assessment result of metal equipment corrosion defects based on the image data, depth data, and microstructure data;

[0145] The first, second, and third quantitative assessment results are mapped to preset hazard levels to obtain the hazard level of each type of defect.

[0146] Each module in the aforementioned transmission line defect identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0147] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. 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 in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a method for identifying defects in power transmission lines.

[0148] The display unit of this computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this computer device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0149] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform all the steps described above.

[0151] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, performs all of the above steps.

[0152] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs all of the above steps.

[0153] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0154] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0155] 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.

[0156] 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 patent 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 transmission line defect identification, characterized in that, The method comprises: acquiring multi-source data of a power transmission line, and performing format unification, data cleaning and data enhancement processing on the multi-source data to obtain processed multi-source data; the multi-source data comprises image data, real-time operation data and historical defect data of the power transmission line; based on environmental perception data in the processed real-time operation data, adjusting a fuzzy neural network model, inputting the processed image data into the adjusted fuzzy neural network model, and obtaining enhanced image data; extracting visual features in the enhanced image data and semantic features in the processed historical defect data, fusing the visual features and the semantic features to obtain a fusion feature vector; for each type of defect, based on a preset rule and a defect detection model, quantitatively evaluating the fusion feature vector to determine the hazard level of each type of defect; aligning the visual features and the semantic features, and based on the hazard level and the fusion feature vector, generating an identification result of the defects of the power transmission line.

2. The method of claim 1, wherein, The method comprises: transforming the multi-source data into a preset format according to the data type to obtain transformed multi-source data; removing duplicate images in the image data of the transformed multi-source data, deleting text data lacking key fields in the transformed multi-source data, and removing outliers in the transformed multi-source data to obtain processed multi-source data.

3. The method of claim 1, wherein, The environmental perception data comprises light intensity, terrain vibration frequency and climate type; the fuzzy neural network model is adjusted based on the environmental perception data in the processed real-time operation data, comprising: based on the light intensity, adjusting the contrast enhancement parameter and the brightness adjustment parameter in the fuzzy neural network model; based on the terrain vibration frequency, adjusting the filter kernel size and the iteration number of the fuzzy neural network model; based on the climate type, adjusting the processing mode of the fuzzy neural network model.

4. The method of claim 1, wherein, The method comprises: removing redundant dimensions in the visual features, retaining features in the visual features with mutual information values greater than a first preset threshold with respect to preset defect types to form target visual features; dimensionality reduction processing is performed on the semantic features, and semantic features with cumulative variance contribution rates higher than a second preset threshold are selected as target semantic features; based on the target visual features and the target semantic features, weighted summation is performed to obtain a fusion feature vector.

5. The method of claim 4, wherein, The preset defect types comprise bird nest defects, external damage defects and metal equipment corrosion defects.

6. The method of claim 5, wherein, The method comprises: locating the position of the bird nest, determining a first quantitative evaluation result of the bird nest defect based on the position, the volume of the bird nest, the spatial distance between the bird nest and the power transmission line, whether the bird nest contains metal foreign matter, and whether there are protective measures; Reconstruct a three-dimensional position relationship between the external damage source and the power transmission line, and determine a second quantitative evaluation result of the external damage defect based on rules of the type of external damage source, spatial distance, and motion direction; Obtain image data, depth data, and microstructure data of metal equipment corrosion, and determine a third quantitative evaluation result of the metal equipment corrosion defect based on the image data, the depth data, and the microstructure data; Map the first quantitative evaluation result, the second quantitative evaluation result, and the third quantitative evaluation result to a preset hazard level to obtain a hazard level of each type of defect.

7. A power line defect identification apparatus characterized by comprising: The device comprises: An acquisition module configured to acquire multi-source data of a power transmission line, and perform format unification, data cleaning, and data enhancement processing on the multi-source data to obtain processed multi-source data; the multi-source data comprises image data, real-time operation data, and historical defect data of the power transmission line; An enhancement module configured to adjust a fuzzy neural network model based on environment perception data in the processed real-time operation data, input the processed image data into the adjusted fuzzy neural network model, and obtain enhanced image data; A fusion module configured to extract visual features in the enhanced image data and semantic features in the processed historical defect data, fuse the visual features and the semantic features, and obtain a fusion feature vector; A determination module configured to determine a hazard level of each type of defect by quantitatively evaluating the fusion feature vector based on a preset rule and a defect detection model; A generation module configured to align the visual features and the semantic features, and generate an identification result of a power transmission line defect based on the hazard level and the fusion feature vector.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 6.

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

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 6.

Images