Search

CN-122023912-A - Inversion identification method for partial discharge defects of distribution line insulator

CN122023912ACN 122023912 ACN122023912 ACN 122023912ACN-122023912-A

Abstract

The invention relates to a distribution line insulator partial discharge defect inversion identification method which comprises the following steps of S1, establishing a small sample data set, S2, adding the same condition model to a generator and a judging device for generating an countermeasure network R3GAN by a recursion residual error to obtain CRGAN networks, S3, carrying out iterative training on CRGAN networks, supplementing the generated sample to the small sample data set after training is completed to obtain an expansion data set, S4, introducing SeaFormer modules into YOLOv8_seg networks, constructing S_ YOLOv8_seg networks, setting an improvement loss function, S5, inputting the expansion data set into the S_ YOLOv8_seg networks, and training the S_ YOLOv _seg networks by combining the improvement loss function to obtain the distribution line insulator partial discharge defect inversion identification model. Compared with the prior art, the invention has the advantages of realizing high-precision identification, inversion evaluation and the like in a small sample environment.

Inventors

  • JIN LIJUN
  • ZHU TIANLE
  • LIANG RUILONG

Assignees

  • 同济大学

Dates

Publication Date
20260512
Application Date
20260130

Claims (10)

  1. 1. The inversion identification method for the partial discharge defect of the distribution line insulator is characterized by comprising the following steps of: S1, an unmanned aerial vehicle is adopted to carry an ultraviolet imager system, ultraviolet image data of insulators in different discharge states are collected, gray level image conversion and denoising processing are carried out, and a small sample data set is established; S2, adding the same condition model to a generator and a judger for generating an countermeasure network R3GAN by the recursion residual error to obtain CRGAN networks; s3, based on an ultraviolet imaging forward law, determining classification of discharge intensity in a spot area threshold interval, inputting the classification label, performing iterative training on CRGAN networks, and supplementing generated samples to a small sample data set after training is completed to obtain an extended data set; S4, introducing a SeaFormer module into the YOLOv8_seg network, constructing the S_ YOLOv8_seg network, and setting an improved loss function; S5, inputting the extended data set into the S_ YOLOv8_seg network, and training the S_ YOLOv8_seg network by combining with the improved loss function to obtain a distribution line insulator partial discharge defect inversion identification model, and performing actual distribution line insulator partial discharge defect inversion identification based on the distribution line insulator partial discharge defect inversion identification model.
  2. 2. The distribution line insulator partial discharge defect inversion identification method according to claim 1, wherein the unmanned aerial vehicle-mounted ultraviolet imager system comprises an unmanned aerial vehicle platform, a ground control system and a data processing system.
  3. 3. The inversion identification method for partial discharge defects of a distribution line insulator according to claim 1, wherein the specific steps of performing gray map conversion and denoising processing are as follows: The insulator uv image data is converted to a gray scale and noise is removed using a3 x 3 median filter.
  4. 4. The inversion identification method of partial discharge defects of a distribution line insulator according to claim 1, wherein a SeaFormer module is introduced into a YOLOv8_seg network, and the specific steps of constructing the s_ YOLOv8_seg network are as follows: and adopting SeaFormer modules to replace backbone modules of YOLOv8-seg to obtain the S_ YOLOv8_seg network.
  5. 5. The distribution line insulator partial discharge defect inversion identification method according to claim 1, wherein the improvement loss function is: Wherein, the In order to improve the loss function, In order to expect the cross-ratio loss, Is the minimum point distance cross ratio loss.
  6. 6. The method for inversion identification of partial discharge defects of a distribution line insulator according to claim 5, wherein the expected overlap loss is: is a prediction frame and a truth frame; Is the cross-over ratio; squaring the Euclidean distance; the center point coordinates of the prediction frame and the truth frame; The frame width and the height are predicted; True value frame width and height; The width and the height of the minimum circumscribed rectangle; To cover the minimum circumscribed rectangular diagonal length of the two frames.
  7. 7. The inversion identification method of partial discharge defects of a distribution line insulator according to claim 6, wherein the minimum point distance overlap loss is: 。
  8. 8. The method of claim 1, wherein the classification labels comprise discharge intensity level labels.
  9. 9. The inversion identification method for the partial discharge defects of the distribution line insulator is characterized in that the discharge Area representing the continuous value is less than 250 when the discharge intensity level label is good, the discharge Area representing the continuous value when the discharge intensity level label is weak discharge is more than 250< Area <30000, and the discharge Area representing the continuous value when the discharge intensity level label is flashover is more than 30000.
  10. 10. The method for inversion identification of partial discharge defects of a distribution line insulator according to claim 1, wherein a small batch training mode is adopted in the iterative training of CRGAN networks.

Description

Inversion identification method for partial discharge defects of distribution line insulator Technical Field The invention relates to the technical field of insulator inversion, in particular to a distribution line insulator partial discharge defect inversion identification method. Background The insulator in the distribution line bears the key tasks of isolating high-voltage current and supporting wires, and the working state of the insulator directly influences the safe and stable operation of the power grid. However, since the insulator is exposed to outdoor environment for a long time, the insulator is easily affected by factors such as wind blowing, rain, ice and snow, high temperature, dust and the like, defects such as damage, pollution, foreign matter adhesion, partial discharge and the like are extremely easy to occur, and flashover faults can be caused even large-scale power interruption is caused when the insulator is seriously used. The existing insulator inspection mode mainly relies on manual inspection or ground handheld equipment for detection, and the method has the defects of low efficiency, missed inspection, high labor intensity, high safety risk and the like. In recent years, unmanned aerial vehicle inspection has been applied to power inspection operation, but is still mainly limited to image acquisition, and efficient and accurate automatic identification and intelligent evaluation are not realized. Especially in the aspect of partial discharge detection, the traditional method such as ultraviolet imaging still relies on manual interpretation, and the requirements of a modern power distribution system on fine and intelligent management are difficult to meet. In practical engineering application, because the probability of occurrence of discharge defect events is relatively low, and the field acquisition is limited by weather, flight conditions, time windows and safety specifications, the number of ultraviolet images with defect characteristics which can be acquired is very limited, and only a small sample data set is often formed. The small sample data set is a normal state in the power inspection task, so that model training data is insufficient, class distribution imbalance is easy to occur, and requirements of high-precision identification and inversion are difficult to meet. Disclosure of Invention The invention aims to provide a distribution line insulator partial discharge defect inversion identification method for realizing high-precision identification and inversion evaluation in a small sample environment. The aim of the invention can be achieved by the following technical scheme: The inversion identification method of the partial discharge defect of the distribution line insulator comprises the following steps: S1, an unmanned aerial vehicle is adopted to carry an ultraviolet imager system, ultraviolet image data of insulators in different discharge states are collected, gray level image conversion and denoising processing are carried out, and a small sample data set is established; S2, adding the same condition model to a generator and a judger for generating an countermeasure network R3GAN by the recursion residual error to obtain CRGAN networks; s3, based on an ultraviolet imaging forward law, determining classification of discharge intensity in a spot area threshold interval, inputting the classification label, performing iterative training on CRGAN networks, and supplementing generated samples to a small sample data set after training is completed to obtain an extended data set; S4, introducing a SeaFormer module into the YOLOv8_seg network, constructing the S_ YOLOv8_seg network, and setting an improved loss function; S5, inputting the extended data set into the S_ YOLOv8_seg network, and training the S_ YOLOv8_seg network by combining with the improved loss function to obtain a distribution line insulator partial discharge defect inversion identification model, and performing actual distribution line insulator partial discharge defect inversion identification based on the distribution line insulator partial discharge defect inversion identification model. Further, the unmanned aerial vehicle-mounted ultraviolet imager system comprises an unmanned aerial vehicle platform, a ground control system and a data processing system. Further, the specific steps of gray map conversion and denoising processing are as follows: The insulator uv image data is converted to a gray scale and noise is removed using a3 x 3 median filter. Further, a SeaFormer module is introduced into the YOLOv8_seg network, and the specific steps of constructing the S_ YOLOv8_seg network are as follows: and adopting SeaFormer modules to replace backbone modules of YOLOv8-seg to obtain the S_ YOLOv8_seg network. Further, the improvement loss function is: Wherein, the In order to improve the loss function,In order to expect the cross-ratio loss,Is the minimum point distance cross ratio loss. Further, the