Ultraviolet drone 5G cloud high-voltage line fault accurate positioning system and method

Abstract

The present invention discloses a system and method for accurately locating faults in high-voltage lines based on ultraviolet drones and 5G cloud, which solves the objective problems of the mismatch between the ability to quickly restore power supply after tripping and the construction of energy Internet, the weak ability to accurately judge important communication lines, and the insufficient support for improving the quality of service in the implementation of strategies. The positioning system disclosed in the present invention includes: - Inspection terminal: a drone equipped with an ultraviolet imager for collecting line information; - Local workstation: receiving information collected by the inspection terminal through data transmission and performing information processing; - Dispatching display terminal: receiving and displaying processed information transmitted by the local workstation through data transmission; - Station-side security: transmitting the tripping protection information of the substation protection device to the dispatching display terminal through data transmission; and also includes a 5G base station for realizing the above-mentioned data transmission.

Description

Technical Field

[0001] The invention relates to a high-voltage line fault locating technology, which is specifically used for locating and troubleshooting faults of high-voltage power transmission lines. Background Art

[0002] When a 110 kV or higher transmission line trips due to an accident, the fault recorder uses the connected current, voltage, frequency, analog, switch, measured parameters such as positive sequence impedance and zero sequence impedance to determine the fault location according to the power system fault analysis algorithm, and uploads the fault information from the station fault recorder to the recording master station in Zone III through the optical fiber channel. Once the recorder equipment fails or the channel is interrupted, the trip information cannot be uploaded in real time, and the fault cannot be handled quickly. For permanent fault tripping of 110 kV or higher high-voltage transmission lines, if forced delivery is performed before the fault point is inspected, it will affect the reliability of power supply. After a permanent accident trip occurs, line patrol personnel mainly patrol the line by manual, drones, and portable handheld daytime corona and arc imagers rented by the company. 110 kV and higher transmission lines are high above the ground, generally at a height of 25 to 40 meters, making high-altitude patrols difficult and inefficient. The application of drone technology in transmission lines can effectively improve patrol efficiency and reduce the workload of transmission patrol personnel. However, the current drone inspection has low recognition of fine corona discharge points and low data transmission efficiency. At present, 5G cloud big data technology has been applied in distribution networks, that is, the 5G distribution network intelligent distributed protection innovation demonstration project was implemented in 2020, but 5G cloud big data technology has not been applied in production such as control and transmission. Therefore, it is urgent to improve the accuracy of drone detection of transmission line faults and use 5G transmission technology to quickly find and locate fault points.

[0003] Once a 110 kV or higher transmission line trips, if there is a channel interruption, the fault distance information cannot be uploaded in time. The traditional line patrol method is time-consuming and labor-intensive, and cannot fully meet the needs of modern power grid construction and development. If the fault point is on pole No. 100, according to the normal patrol method, a lot of manpower, material and financial resources are consumed, and the line patrol personnel cannot quickly find the fault point. Although the company's current drones can improve the efficiency of line patrols, they cannot quickly process the returned image information in real time. The portable handheld daytime corona and arc imager rented by the company has extremely high sensitivity to surface discharges of objects, but the imaging channel is single-channel with a resolution of 640*480. It has the disadvantages of low imaging resolution and cannot be equipped with drones for image collection and analysis. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a system and method for accurately locating high-voltage line faults based on ultraviolet drone 5G cloud in response to the deficiencies in the above-mentioned prior art, and to use the "ultraviolet drone 5G cloud artificial intelligence" technical solution to solve the mismatch between the ability to quickly restore power supply after tripping and the construction of the energy Internet, the lack of ability to accurately judge important communication lines and the insufficient support for improving the quality of service in strategic implementation, the weak ability of the power grid to deal with failures of important transmission lines and the incompatibility with the level of modern dispatching and control, the lack of innovative mining of power grid data and the mismatch between the level of improvement of modern dispatching service capabilities and other objective problems.

[0005] In order to achieve the above purpose, the UV drone 5G cloud high-voltage line fault accurate positioning system disclosed in the present invention adopts the following technical solutions to achieve it:

[0006] Based on ultraviolet drone 5G cloud high-voltage line fault accurate positioning system, including:

[0007] -Inspection terminal: a drone equipped with a UV imager used to collect line information;

[0008] -Local workstation: receives information collected by the inspection terminal through data transmission and processes the information;

[0009] -Dispatching display terminal: receives and displays the processed information transmitted by the local workstation through data transmission;

[0010] - Station-side information protection: The tripping protection information of the substation protection device is transmitted to the dispatching display terminal through data transmission;

[0011] It also includes a 5G base station for realizing the above data transmission;

[0012] The system comprises the following architecture:

[0013] A perception layer including the above-mentioned station-side security, a network layer including the above-mentioned 5G base station, an application layer including the above-mentioned inspection terminal and local workstation, and a power business management support platform including the above-mentioned dispatching display terminal;

[0014] The power business management support platform includes a data acquisition and control layer, a platform layer, a management application layer and a horizontal interface for external communication connection, and the horizontal interface is controlled by the operation management system.

[0015] As a preferred embodiment of the disclosed positioning system of the present invention:

[0016] The perception layer, in the smart grid 5G application scenario, comprehensively considers the factors of terminal form, mobility, and load, and adopts an embedded module approach for the inspection of power transmission lines equipped with ultraviolet multi-channel spectrum drones, which is specifically manifested in image quality clarity;

[0017] The network layer combines the three major network characteristics of 5G, network slicing technology and the communication requirements of various smart grid services to build customized slices that can meet the business needs of high-voltage line fault inspection. Each type of slice builds multiple network slices on demand, and provides differentiated network slicing services for high-voltage line fault inspection based on the characteristics of slice resources and the business needs of high-voltage line fault inspection.

[0018] The four layers of data acquisition and control layer, platform layer, management application layer and horizontal interface realize the connection management, terminal management and slice management functions in the 5G communication network, and realize the transmission of inspection images to local workstations on the ground through 5G;

[0019] In the application layer, the 5G smart grid application layer has a collection function to collect high-voltage line fault images.

[0020] As a preferred implementation mode of the disclosed positioning system of the present invention: the inspection terminal is equipped with an ultraviolet drone to collect information through 5G, including 9 ultraviolet spectral channel information with a wavelength range of 250-420nm, multiple pictures with a resolution of 2048*2048, video information and inspection high-definition video, providing 100Mbps, 99.9% reliability, a delay of no more than 200ms, and a connection number of one hundred thousand. It realizes real-time return of high-definition video images above 100M, batch processing of tens of thousands of images, and simultaneous processing of multiple users, multiple permissions, multiple models, and multiple images.

[0021] The present invention also discloses a method for accurately locating a fault in a 5G cloud high-voltage line based on an ultraviolet drone, which uses any of the above-mentioned positioning systems to perform the following steps:

[0022] (1) Obtaining station-side security data: The tripping protection information of the protection device in the substation is transmitted to the dispatching end using 5G wireless communication, and the fault information is collected to quickly predict the range of the faulty tower;

[0023] (2) Locating the fault point: Use a drone equipped with an ultraviolet multispectral imager to patrol the faulty line and transmit the inspection images back to the local workstation on the ground via 5G. Finally, use image processing, machine learning, and signal analysis to quickly find the fault point and provide specific fault information to the dispatcher to handle the accident.

[0024] As a preferred implementation of the positioning method disclosed in the present invention: the step (2) specifically includes the following process:

[0025] S100: Memory model: Through high-precision memory encoding and feature extraction in the intelligent detection system, the ultraviolet multi-spectral image is encoded and memorized based on traditional and hierarchical recurrent neural networks, and a hierarchical attention mechanism is introduced on the memory unit to extract high-precision feature quantities of the target;

[0026] S200: Analysis of interference mechanism: Analyze the mechanism of ultraviolet multispectral imaging under the interference factors of jitter and dust, build a visual imaging model for ultraviolet multispectral images, and use dark channel image enhancement technology to clarify blurred images;

[0027] S300: Build a neural network: Use the feature map extraction network to obtain the corresponding feature map, use the regional RPN recommendation network to generate the candidate region RIO on the feature map, obtain a fixed feature map size through the RIO Align layer, input it into the prediction network, and predict the result based on the type, bounding box, and Mask area dimensions.

[0028] As a preferred implementation of the positioning method disclosed in the present invention:

[0029] The extraction of target high-precision features in step S100 is feature extraction based on HOG, and the feature extraction based on HOG includes the following process:

[0030] The histogram of oriented gradient features operate on the original image, mainly describing the contour information of an image, that is, describing the local area information of the input image, thereby statistically analyzing the feature vector of the local area to describe its appearance and shape; the process of generating the HOG feature vector is: through the optimization of the image processing algorithm for the high-voltage fault line image and the quantitative calculation of the algorithm parameters, a high-voltage line surface discharge point image processing algorithm with strong practicality and high precision is obtained for high-voltage line images.

[0031] As a preferred implementation of the positioning method disclosed in the present invention:

[0032] The algorithm flow of the HOG-based feature extraction is as follows:

[0033] S101: input image information;

[0034] S102: grayscale image;

[0035] S103: filtering by Gauss filter convolution of Gabor filter;

[0036] S104: image smoothing;

[0037] S105: judging whether the damage edge is clear by edge density function, if it is clear, executing step S106, if it is not clear, adjusting Gabor filter parameters and re-executing step S103;

[0038] S106: performing image segmentation by using OTSU threshold;

[0039] S107: feature extraction;

[0040] S108: parameter calculation;

[0041] S109: End.

[0042] As a preferred implementation of the positioning method disclosed in the present invention:

[0043] The method also includes determining whether the scratch edge of the image segmented in step S106 is smooth. If it is not smooth, the Gabor filter parameters are adjusted and step S103 is re-executed; if it is smooth, step S107 is executed after morphological reconstruction.

[0044] As a preferred implementation of the positioning method disclosed in the present invention:

[0045] The extraction of target high-precision features in step S100 is feature extraction based on a hierarchical deep attention mechanism, and the feature extraction based on a hierarchical deep attention mechanism includes the following process:

[0046] A hierarchical recurrent neural network is used to encode and memorize the ultraviolet multispectral images, and a hierarchical attention mechanism is introduced on the memory unit for feature extraction: the ultraviolet image of the system at the current moment is input, encoded through the recurrent neural network model to obtain a vectorized representation, and then the attention mechanism is used on the memory unit for semantic activation to obtain a semantic vector, thereby completing the feature extraction of the fault point target.

[0047] As a preferred implementation of the positioning method disclosed in the present invention, it also includes a process of implementing the step (2) by building an artificial intelligence classifier network:

[0048] Based on the improvement of Faster R-CNN network structure, Mask R-CNN network is established to complete classification and bondingBox regression. At the same time, parallel branches are added to each region of interest (RIO) for instance segmentation and joint training.

[0049] The network framework of the artificial intelligence classifier is divided into four parts: feature map extraction network, region recommendation network, RIO Align and prediction network;

[0050] First, the image is passed through the feature map extraction network to obtain the corresponding feature map, and then the region proposal network (RPN) is used to generate the candidate region RIO on the feature map. Then, the fixed feature map size is obtained through the RIO Align layer and input into the prediction network to predict the type, bounding box, and Mask area.

[0051] The feature map network is the backbone network of the high-voltage line fault point detection network. It is divided into two stages. First, the features of the image are extracted, and then the extracted features are fused, and the output result is a feature map. Mask R-CNN uses the ResNet-101 network for feature extraction, and then uses the FPN network for feature fusion.

[0052] The beneficial effects of the present invention are:

[0053] The present invention discloses a high-voltage line fault precise positioning system based on ultraviolet drone 5G cloud. This project uses 5G wireless communication to transmit the trip protection information of the protection device in the substation to the dispatching end, collects the fault information, and quickly predicts the range of the faulty tower. Then, a drone equipped with an ultraviolet multi-spectral imager is used to patrol the faulty line, and the inspection image is transmitted back to the ground workstation via 5G. Finally, image processing, machine learning, signal analysis and other technologies are used to quickly find the fault point, provide theoretical support for dispatchers to handle accidents, improve the efficiency of line patrol personnel in finding fault points, shorten fault finding time, and improve the accident tripping processing rate, so as to achieve the purpose of rapid power supply restoration. The present invention has the following advantages:

[0054] (1) UV-carrying drones can replace manual and traditional drones, complete complex aerial tasks under unmanned conditions, improve operational efficiency, and reduce manpower and material costs.

[0055] 2) 5G's higher bandwidth makes the images transmitted by drones clearer, and the ultra-low latency means more precise control of drones. The wider low-altitude coverage can help drones perform tasks more smoothly and find high-voltage line faults more quickly.

[0056] (3) Ultraviolet multispectral imaging, which obtains information of different ultraviolet bands of the same target at the same time, can achieve a more refined feature description of the target compared to single-channel data, providing a guarantee for the detection and differentiation of targets of interest;

[0057] (4) The application of artificial intelligence technology in high-voltage line fault inspection can improve the detection and positioning accuracy of fault points under imaging fuzzy conditions through technical means such as image enhancement, feature extraction, small sample deep learning, intelligent detection, and positioning measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic structural diagram of a specific implementation mode of the system of the present invention;

[0059] Figure 2 A schematic diagram of a specific application embodiment of the present invention;

[0060] Figure 3HOG feature extraction flow chart used in the preferred embodiment of the present invention;

[0061] Figure 4 A flow chart of an image processing algorithm used in a preferred embodiment of the present invention;

[0062] Figure 5 A flowchart of feature extraction based on a hierarchical attention mechanism adopted by the method of the present invention;

[0063] Figure 6 A high-voltage line fault point accurate positioning network framework in a specific implementation manner of the present invention;

[0064] Figure 7 This is an algorithm flow chart of the feature graph network of the present invention.

[0065] Description of reference numerals:

[0066] 1-5G base station; 2-local workstation, 3-dispatching display terminal, 4-station security, 5-inspection terminal. DETAILED DESCRIPTION

[0067] The specific implementation of the present invention is described below in conjunction with the accompanying drawings and embodiments:

[0068] like Figures 1 to 7 As shown, it shows a specific implementation of the present invention. The ultraviolet drone 5G cloud high-voltage line fault accurate positioning system disclosed in the present invention includes:

[0069] - Inspection terminal 5: a drone equipped with a UV imager for collecting line information;

[0070] - Local workstation 2: receives information collected by the inspection terminal through data transmission and processes the information;

[0071] -Dispatching display terminal 3: receives and displays the processed information transmitted by the local workstation through data transmission;

[0072] - Station-side security information 4: Transmit the tripping protection information of the substation protection device to the dispatching display terminal through data transmission;

[0073] It also includes a 5G base station 1 for realizing the above-mentioned data transmission;

[0074] The system comprises the following architecture:

[0075] A perception layer including the above-mentioned station-side security, a network layer including the above-mentioned 5G base station, an application layer including the above-mentioned inspection terminal and local workstation, and a power business management support platform including the above-mentioned dispatching display terminal;

[0076] The power business management support platform includes a data acquisition and control layer, a platform layer, a management application layer and a horizontal interface for external communication connection, and the horizontal interface is controlled by the operation management system.

[0077] As a preferred embodiment of the positioning system disclosed in the present invention, Figure 2 As shown:

[0078] The perception layer, in the smart grid 5G application scenario, comprehensively considers the factors of terminal form, mobility, and load, and adopts an embedded module approach for the inspection of power transmission lines equipped with ultraviolet multi-channel spectrum drones, which is specifically manifested in image quality clarity;

[0079] The network layer combines the three major network characteristics of 5G, network slicing technology and the communication requirements of various smart grid services to build customized slices that can meet the business needs of high-voltage line fault inspection. Each type of slice builds multiple network slices on demand, and provides differentiated network slicing services for high-voltage line fault inspection based on the characteristics of slice resources and the business needs of high-voltage line fault inspection.

[0080] The four layers of data acquisition and control layer, platform layer, management application layer and horizontal interface realize the connection management, terminal management and slice management functions in the 5G communication network, and realize the transmission of inspection images to local workstations on the ground through 5G;

[0081] In the application layer, the 5G smart grid application layer has a collection function to collect high-voltage line fault images.

[0082] After adopting this preferred embodiment, it is possible to accurately extract the target characteristic quantities of high-voltage line faults, thereby realizing accurate control and refined scheduling of rapid power restoration of transmission lines, providing "stable, accurate, and fast" high-quality services for important corporate users, and realizing internal and external value enhancement. Establish a new model of artificial intelligence multi-spectral inspection, and create a new scenario for the construction and dispatch of the "digital new infrastructure" energy Internet. Intuitively display the status of important transmission lines and important connecting lines, accurately locate faults and discover defects in real time, handle faults in a timely manner, quickly restore power supply, and formulate plans to eliminate defects. Carry out intelligent inspections, realize the return and batch processing of hundreds of megabytes of multi-spectral image information, realize refined inspections, and promote the digitalization and intelligent transformation of enterprises.

[0083] As a preferred embodiment of the positioning system disclosed in the present invention, Figure 1 and 2As shown: The information collected by the ultraviolet drone flight uses 5G to transmit 9 spectral channel information with a wavelength range of 250-420nm, massive pictures and video information with a resolution of 2048*2048, and inspection HD video. It uses the characteristics of 5G high speed, low latency, massive connections, and fast movement to provide 100Mbps, reliability of 99.9%, latency of no more than 200ms, and communication needs of 100,000 connections, realize real-time return of HD video images above 100M, batch processing of tens of thousands of images, support multi-user, multi-authority, multi-model, and multi-image processing at the same time, expand the inspection scope, improve inspection efficiency, and realize intelligent inspection.

[0084] The present invention also discloses a method for accurately locating a fault in a 5G cloud high-voltage line based on an ultraviolet drone, which uses any of the above-mentioned positioning systems to perform the following steps:

[0085] (1) Obtaining station-side security data: The tripping protection information of the protection device in the substation is transmitted to the dispatching end using 5G wireless communication, and the fault information is collected to quickly predict the range of the faulty tower;

[0086] (2) Locating the fault point: Use a drone equipped with an ultraviolet multispectral imager to patrol the faulty line and transmit the inspection images back to the local workstation on the ground via 5G. Finally, use image processing, machine learning, and signal analysis to quickly find the fault point and provide specific fault information to the dispatcher to handle the accident.

[0087] It provides theoretical support for dispatchers to handle accidents, improves the efficiency of line patrol personnel in finding fault points, shortens fault finding time, increases the accident tripping handling rate, and achieves the goal of quickly restoring power supply.

[0088] As a preferred embodiment of the positioning method disclosed in the present invention, Figure 3-6 As shown: the step (2) specifically includes the following process:

[0089] S100: Memory model: Through high-precision memory encoding and feature extraction in the intelligent detection system, the ultraviolet multi-spectral image is encoded and memorized based on traditional and hierarchical recurrent neural networks, and a hierarchical attention mechanism is introduced on the memory unit to extract high-precision feature quantities of the target;

[0090] S200: Analysis of interference mechanism: Analyze the mechanism of ultraviolet multispectral imaging under the interference factors of jitter and dust, build a visual imaging model for ultraviolet multispectral images, and use dark channel image enhancement technology to clarify blurred images;

[0091] S300: Build a neural network: Use the feature map extraction network to obtain the corresponding feature map, use the regional RPN recommendation network to generate the candidate region RIO on the feature map, obtain a fixed feature map size through the RIO Align layer, input it into the prediction network, and predict the result based on the type, bounding box, and Mask area dimensions.

[0092] As a preferred embodiment of the positioning method disclosed in the present invention, Figure 3 As shown:

[0093] The extraction of target high-precision features in step S100 is feature extraction based on HOG, and the feature extraction based on HOG includes the following process:

[0094] The histogram of oriented gradient features operate on the original image, mainly describing the contour information of an image, that is, describing the local area information of the input image, thereby statistically analyzing the feature vector of the local area to describe its appearance and shape; the process of generating the HOG feature vector is: through the optimization of the image processing algorithm for the high-voltage fault line image and the quantitative calculation of the algorithm parameters, a high-voltage line surface discharge point image processing algorithm with strong practicality and high precision is obtained for high-voltage line images.

[0095] As a preferred embodiment of the positioning method disclosed in the present invention, Figure 3-4 As shown:

[0096] The algorithm flow of the HOG-based feature extraction is as follows:

[0097] S101: input image information;

[0098] S102: grayscale image;

[0099] S103: filtering by Gauss filter convolution of Gabor filter;

[0100] S104: image smoothing;

[0101] S105: judging whether the damage edge is clear by edge density function, if it is clear, executing step S106, if it is not clear, adjusting Gabor filter parameters and re-executing step S103;

[0102] S106: performing image segmentation by using OTSU threshold;

[0103] S107: feature extraction;

[0104] S108: parameter calculation;

[0105] S109: End.

[0106] By optimizing the image processing algorithm for high-voltage fault line images and quantitatively calculating the algorithm parameters, a high-voltage line surface discharge point image processing algorithm with strong practicality and high accuracy can be finally obtained for high-voltage line images.

[0107] As a preferred embodiment of the positioning method disclosed in the present invention, Figure 4 As shown:

[0108] The method also includes determining whether the scratch edge of the image segmented in step S106 is smooth. If it is not smooth, the Gabor filter parameters are adjusted and step S103 is re-executed; if it is smooth, step S107 is executed after morphological reconstruction.

[0109] As a preferred embodiment of the positioning method disclosed in the present invention, Figure 5-7 As shown:

[0110] The extraction of target high-precision features in step S100 is feature extraction based on a hierarchical deep attention mechanism, and the feature extraction based on a hierarchical deep attention mechanism includes the following process:

[0111] A hierarchical recurrent neural network is used to encode and memorize the ultraviolet multispectral images, and a hierarchical attention mechanism is introduced on the memory unit for feature extraction: the ultraviolet image of the system at the current moment is input, encoded through the recurrent neural network model to obtain a vectorized representation, and then the attention mechanism is used on the memory unit for semantic activation to obtain a semantic vector, thereby completing the feature extraction of the fault point target.

[0112] As a preferred embodiment of the positioning method disclosed in the present invention, Figure 5-7 As shown:

[0113] It also includes a process for implementing step (2) by building an artificial intelligence classifier network:

[0114] Based on the improvement of Faster R-CNN network structure, Mask R-CNN network is established to complete classification and bondingBox regression. At the same time, parallel branches are added to each region of interest (RIO) for instance segmentation and joint training.

[0115] The network framework of the artificial intelligence classifier is divided into four parts: feature map extraction network, region recommendation network, RIO Align and prediction network;

[0116] First, the image is passed through the feature map extraction network to obtain the corresponding feature map, and then the region proposal network (RPN) is used to generate the candidate region RIO on the feature map. Then, the fixed feature map size is obtained through the RIO Align layer and input into the prediction network to predict the type, bounding box, and Mask area.

[0117] The feature map network is the backbone network of the high-voltage line fault point detection network. It is divided into two stages. First, the features of the image are extracted, and then the extracted features are fused, and the output result is a feature map. Mask R-CNN uses the ResNet-101 network for feature extraction, and then uses the FPN network for feature fusion.

[0118] The solution of the present invention also has the following advantages:

[0119] 1. Visually display the status of important transmission lines and important interconnection lines, accurately locate faults and discover defects in real time, promptly handle faults and quickly restore power supply, and formulate plans to eliminate faults.

[0120] 2. Accurately display information such as transmission line diameter, path, tower size and location, reasonably predict development needs, and guide grid planning and layout.

[0121] 3. It is convenient to carry out the calculation of conductor current carrying capacity, effectively support the evaluation of economic transmission capacity of important transmission lines, reasonably optimize the investment direction, and achieve the purpose of precise investment.

[0122] 4. The progress of fault handling can be accurately predicted, and the controller can timely monitor and accurately control important transmission lines, important contact lines, important sections, currents, voltages, etc. It effectively improves the efficiency of statistical analysis of fault feature quantities, and the statistical work efficiency is increased from 5 images / minute for manual identification to 20 images / minute.

[0123] 5. Accurately determine the grid risk level based on the fault characteristics of the transmission lines and formulate reasonable and effective risk prevention and control measures.

[0124] 6. Formulate a reasonable user access plan based on the transmission line cross-section and equipment load ratio to avoid circuit breaker failure and expand the scope of power outage.

[0125] 7. Scientifically predict the rationality of user's installed capacity based on line cross-section and line load conditions, provide technical support for extracting accessible capacity and conducting grid margin calculations, and ensure smooth implementation of business expansion plans.

[0126] 8. It is convenient for the government to make corresponding business decisions based on the line conditions and the investment situations of different enterprises.

[0127] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the field without departing from the purpose of the present invention. These changes involve related technologies well known to those skilled in the art, which all fall within the scope of protection of the patent of the present invention.

[0128] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for accurately locating high-voltage line faults based on ultraviolet drones and 5G cloud, characterized in that: The following steps are included: (1) Obtain station-side security data: The tripping protection information of the protection device in the substation is transmitted to the dispatching end using 5G wireless communication, and the fault information is collected to quickly predict the range of the faulty tower; (2) Locating the fault point: Use a drone equipped with an ultraviolet multispectral imager to patrol the faulty line and transmit the inspection images back to the local workstation on the ground via 5G. Finally, use image processing, machine learning, and signal analysis to quickly find the fault point and provide specific fault information for the dispatcher to handle the accident. The step (2) specifically includes the following process: S100: Memory model: Through high-precision memory encoding and feature extraction in the intelligent detection system, the ultraviolet multi-spectral image is encoded and memorized based on traditional and hierarchical recurrent neural networks, and a hierarchical attention mechanism is introduced on the memory unit to extract high-precision feature quantities of the target; S200: Analysis of interference mechanism: Analyze the mechanism of ultraviolet multispectral imaging under the interference factors of jitter and dust, build a visual imaging model for ultraviolet multispectral images, and use dark channel image enhancement technology to clarify blurred images; S300: Building a neural network: Using the feature map extraction network to obtain the corresponding feature map, using the regional RPN recommendation network to generate the candidate region RIO on the feature map, obtaining a fixed feature map size through the RIO Align layer, inputting it into the prediction network, and predicting the result based on the type, bounding box, and Mask region dimensions; The extraction of target high-precision features in step S100 is feature extraction based on HOG, and the feature extraction based on HOG includes the following process: The directional gradient histogram feature operates on the original image, mainly describing the contour information of an image, that is, describing the local area information of the input image, thereby counting the feature vector of the local area to describe its appearance and shape; The process of generating HOG feature vector is as follows: by optimizing the image processing algorithm for high-voltage fault line images and quantitatively calculating the algorithm parameters, a high-voltage line surface discharge point image processing algorithm with strong practicality and high accuracy is obtained for high-voltage line images; It also includes a process for implementing step (2) by building an artificial intelligence classifier network: Based on the improvement of Faster R-CNN network structure, Mask R-CNN network is established to complete classification and bonding box regression. At the same time, parallel branches are added to each region of interest RIO for instance segmentation and joint training. The network framework of the artificial intelligence classifier is divided into four parts: feature map extraction network, region recommendation network, RIO Align and prediction network; First, the image is passed through the feature map extraction network to obtain the corresponding feature map, and then the region recommendation network RPN is used to generate the candidate region RIO on the feature map. Then, the fixed feature map size is obtained through the RIO Align layer and input into the prediction network to predict the type, bounding box, and Mask area. The feature map network is the backbone network of the high-voltage line fault point detection network. It is divided into two stages. First, the features of the image are extracted, and then the extracted features are fused, and the output result is a feature map. Mask R-CNN uses the ResNet-101 network for feature extraction, and then uses the FPN network for feature fusion.

2. The positioning method according to claim 1, characterized in that: The algorithm flow of the HOG-based feature extraction is as follows: S101: input image information; S102: grayscale image; S103: filtering by Gauss filter convolution of Gabor filter; S104: image smoothing; S105: judging whether the damage edge is clear by edge density function, if it is clear, executing step S106, if it is not clear, adjusting Gabor filter parameters and re-executing step S103; S106: performing image segmentation by using OTSU threshold; S107: feature extraction; S108: parameter calculation; S109: End.

3. The positioning method according to claim 2, characterized in that: The method also includes determining whether the scratch edge of the image segmented in step S106 is smooth. If it is not smooth, the Gabor filter parameters are adjusted and step S103 is re-executed; if it is smooth, step S107 is executed after morphological reconstruction.

4. The positioning method according to claim 1, characterized in that: The extraction of target high-precision features in step S100 is feature extraction based on a hierarchical deep attention mechanism, and the feature extraction based on a hierarchical deep attention mechanism includes the following process: A hierarchical recurrent neural network is used to encode and memorize the ultraviolet multispectral images, and a hierarchical attention mechanism is introduced on the memory unit for feature extraction: the ultraviolet image of the system at the current moment is input, encoded through the recurrent neural network model to obtain a vectorized representation, and then the attention mechanism is used on the memory unit for semantic activation to obtain a semantic vector, thereby completing the feature extraction of the fault point target.

5. Based on the ultraviolet drone 5G cloud high-voltage line fault precise positioning system, the positioning method as described in any one of claims 1 to 4 is executed, characterized in that: Include: -Inspection terminal: a drone equipped with a UV imager used to collect line information; -Local workstation: receives information collected by the inspection terminal through data transmission and processes the information; -Dispatching display terminal: receives and displays the processed information transmitted by the local workstation through data transmission; - Station-side information protection: The tripping protection information of the substation protection device is transmitted to the dispatching display terminal through data transmission; It also includes a 5G base station for realizing the above data transmission; The system comprises the following architecture: A perception layer including the above-mentioned station-side security, a network layer including the above-mentioned 5G base station, an application layer including the above-mentioned inspection terminal and local workstation, and a power business management support platform including the above-mentioned dispatching display terminal; The power business management support platform includes a data acquisition and control layer, a platform layer, a management application layer and a horizontal interface for external communication connection, and the horizontal interface is controlled by the operation management system.

6. The ultraviolet drone 5G cloud high-voltage line fault accurate positioning system according to claim 5 is characterized in that: The perception layer, in the smart grid 5G application scenario, comprehensively considers the factors of terminal form, mobility, and load, and adopts an embedded module approach for the inspection of power transmission lines equipped with ultraviolet multi-channel spectrum drones, which is specifically manifested in image quality clarity; The network layer combines the three major network characteristics of 5G, network slicing technology and the communication requirements of various smart grid services to build customized slices that can meet the business needs of high-voltage line fault inspection. Each type of slice builds multiple network slices on demand, and provides differentiated network slicing services for high-voltage line fault inspection based on the characteristics of slice resources and the business needs of high-voltage line fault inspection. The four layers of data acquisition and control layer, platform layer, management application layer and horizontal interface realize the connection management, terminal management and slice management functions in the 5G communication network, and realize the transmission of inspection images to local workstations on the ground through 5G; In the application layer, the 5G smart grid application layer has a collection function to collect high-voltage line fault images.

7. The ultraviolet drone 5G cloud high-voltage line fault accurate positioning system according to claim 5 is characterized in that: The inspection terminal is equipped with an ultraviolet drone to collect information through 5G, including 9 ultraviolet spectral channel information with a wavelength range of 250-420nm, multiple pictures with a resolution of 2048*2048, video information, and inspection high-definition video, providing communication requirements of 100Mbps, 99.9% reliability, a delay of no more than 200ms, and a connection number of 100,000. It can realize real-time return of high-definition video images above 100 megabytes, batch processing of tens of thousands of images, and simultaneous processing of multiple users, multiple permissions, multiple models, and multiple images.

Images