CN-121998616-A - Power transformation equipment operation and maintenance fault positioning and automatic repairing system based on deep learning

CN121998616ACN 121998616 ACN121998616 ACN 121998616ACN-121998616-A

Abstract

The invention provides an operation and maintenance fault positioning and automatic repairing system of power transformation equipment based on deep learning, which comprises a multi-mode data acquisition module, a data preprocessing and fusion module, a deep learning fault identification module, a three-dimensional space positioning module, a fault decision and repairing scheduling module, a robot automatic repairing module and a visual monitoring and feedback module, wherein fault signals are captured in real time by means of the multi-mode data acquisition module, a fault locking position of the three-dimensional space positioning module is quickly identified by combining with an improved YOLOv model, manual investigation is not needed, the positioning time of the traditional 2-3 hours is shortened to be within 10 minutes, the fault response efficiency is greatly improved, repairing operation is carried out by the robot automatic repairing module, the fault elimination is ensured by matching with a visual testing module, the dependence on manual experience is remarkably reduced, the on-site work amount of operation and maintenance personnel is greatly reduced, and the safety risks such as electric shock and equipment accidental injury under a high-pressure environment are reduced from the source.

Inventors

WANG FENGXUE
MAO XINYU
MA XINGMING
DU SHOUCHENG
WANG FENGSHUANG
QU XUEWEI
SUN CUIYU
LI JIANGE
ZHANG CHANG
JIAO YUXIN
WANG JINLONG

Assignees

国网黑龙江省电力有限公司大庆供电公司

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. The power transformation equipment operation and maintenance fault positioning and automatic repairing system based on deep learning is characterized by comprising a multi-mode data acquisition module, a data preprocessing and fusion module, a deep learning fault identification module, a three-dimensional space positioning module, a fault decision and repairing scheduling module, a robot automatic repairing module and a visual monitoring and feedback module which are respectively in data connection with the fault decision and repairing scheduling module and the robot automatic repairing module, wherein: the multi-mode data acquisition module acquires appearance images, temperature distribution data and operation sound signals of the power transformation equipment through a high-definition camera, a thermal infrared imager and a sonar sensor, and performs air synchronization; The data preprocessing and fusion module is used for cleaning and standardizing the data acquired by the multi-mode data acquisition module, and fusing visual, infrared and acoustic characteristics through an attention mechanism; The deep learning fault identification module is used for reasoning the fusion characteristics based on an improved YOLO target detection algorithm and identifying fault types and two-dimensional coordinates of faults in the image; The three-dimensional space positioning module is used for converting the two-dimensional coordinates of the image into world coordinates by combining with the three-dimensional space model of the transformer substation; The fault decision and repair scheduling module is used for evaluating the fault grade according to the fault type and the positioning result and scheduling the operation and maintenance robot; The robot automatic repair module is used for controlling the operation and maintenance robot to carry a special tool to reach a fault position, and finishing automatic repair through visual guidance; The visual monitoring and feedback module is used for monitoring the repairing process in real time and verifying the repairing effect through the secondary detection of the multi-mode data.
2. The deep learning-based substation equipment operation and maintenance fault positioning and automatic repairing system is characterized in that a multi-mode data acquisition module is provided with a high-definition camera array, a thermal infrared imager network, a distributed sonar sensor and a data synchronization unit, wherein the data synchronization unit realizes space-time alignment of multi-mode data based on a unified timestamp, in the multi-mode data acquisition module, the high-definition camera array has 3840 multiplied by 2160 resolution and supports day-night dual-mode imaging, and the deployment density is 1 group for each 500m < 2 > substation area; the temperature measuring range of the thermal infrared imager network is-20 ℃ to 300 ℃, the temperature measuring precision is +/-0.5 ℃, the frame frequency is more than or equal to 25fps, and the early warning of the abnormal temperature threshold is supported; the distributed sonar sensor performs sampling work according to set sampling parameters and has an anti-electromagnetic interference design, and the linear distance between a deployment position and a transformer and a circuit breaker is less than or equal to 5 meters.
3. The deep learning-based power transformation equipment operation and maintenance fault positioning and automatic repairing system is characterized in that a data synchronization unit in the multi-mode data acquisition module adopts a GPS timing and local clock calibration mode to realize time stamp synchronization, and the time synchronization precision is less than or equal to 1ms; the data synchronization unit is also configured with a data caching mechanism, an industrial eMMC cache chip is configured for each type of equipment of the multi-mode data acquisition module, multi-mode data can be cached locally for at least 2 hours when the network is interrupted, and automatic transmission is supplemented after the network is recovered.
4. The deep learning-based power transformation equipment operation and maintenance fault positioning and automatic repairing system is characterized in that the data preprocessing and fusing module comprises a data cleaning unit and a multi-mode feature fusing unit, wherein the data cleaning unit is used for carrying out distortion correction, denoising and contrast enhancement on image data, carrying out extreme value rejection and temperature numerical value standardization on infrared data, carrying out Fourier transformation on acoustic data to obtain frequency domain features, and carrying out dynamic weighted fusion on preprocessed visual features, infrared features and acoustic features by the multi-mode feature fusing unit based on an attention mechanism to generate fusion feature vectors strongly related to faults; The data cleaning unit denoises image data by Gaussian filtering, corrects distortion by perspective transformation and enhances contrast by self-adaptive histogram equalization; And (3) eliminating extreme values of the infrared data by adopting a3 sigma criterion, converting a temperature value into a standardized value of an [ -1,1] interval by adopting Z-score standardization, converting the acoustic data into frequency domain characteristics by adopting short-time Fourier transform, and extracting Mel frequency cepstrum coefficients as acoustic characteristic core parameters.
5. The deep learning-based power transformation equipment operation and maintenance fault positioning and automatic repairing system according to claim 4, wherein the multi-mode feature fusion unit distributes weights by calculating mutual information values of each mode feature and fault labels, the higher the mutual information values, the larger the mode weight ratio, and residual connection is adopted in the fusion process to preserve details of original features of each mode.
6. The deep learning-based power transformation equipment operation and maintenance fault locating and automatic repairing system according to claim 1, wherein the deep learning fault identifying module comprises an improved YOLO target detection model which is used for capturing fine fault characteristics by adding a1 x1 convolution layer to the neck of YOLOv basic network to construct a small target detection branch, wherein the improved YOLO target detection model adopts a FocalLoss loss function to solve the problem of unbalanced fault sample categories; The deep learning fault recognition module is further configured with a fault feature library matching unit, the fault feature library stores typical feature vectors of historical faults, and when the confidence of the output faults of the improved YOLO target detection model is 85% -98%, the fault types are verified through feature vector cosine similarity matching complementation.
7. The deep learning-based substation equipment operation and maintenance fault positioning and automatic repairing system is characterized in that the three-dimensional space positioning module comprises a substation three-dimensional modeling unit and a coordinate matching and optimizing unit, the substation three-dimensional modeling unit performs initial modeling by adopting a laser point cloud scanner, performs local point cloud updating on a substation by moving the laser scanning equipment in each quarter, and realizes registration of new point cloud and old point cloud by an ICP algorithm; The three-level mapping relation of the coordinate matching and optimizing unit is established through calibration of a calibration plate, internal parameters and external parameters of the high-definition camera are obtained through the calibration plate, the internal parameters are focal length and principal point coordinates, the external parameters are position and posture relative to a sensor mounting point, pixel coordinate-to-sensor physical coordinate mapping is established, sensor physical coordinate-to-world coordinate mapping is established through the GPS coordinates of the sensor mounting point and the origin of a transformer substation world coordinate system, the number of particles of a particle filtering algorithm is set to be 500-1000, and the iteration times are less than or equal to 10.
8. The deep learning-based substation equipment operation and maintenance fault positioning and automatic repairing system according to claim 1, wherein the fault decision and repairing scheduling module is internally provided with a fault grade assessment model, faults are divided into 1-5 levels based on fault types, the influence range of fault positions on equipment operation and fault diffusion risks, 1-3 levels are simple faults capable of being repaired automatically, 4-5 levels are complex faults requiring manual intervention, a robot scheduling unit is called for the 1-3 levels of faults, an optimal moving path is planned based on the fault positions and the real-time states of robots, and a target operation and maintenance robot is scheduled.
9. The deep learning-based power transformation equipment operation and maintenance fault positioning and automatic repairing system is characterized in that the robot automatic repairing module comprises a wheeled mobile platform and a magnetic attraction type special tool library, wherein the wheeled mobile platform adopts laser radar and a visual camera for fusion navigation, the laser radar is used for remote obstacle detection, the visual camera is used for short-distance texture recognition, a dynamic weight factor is introduced into a navigation algorithm, a path without high-voltage equipment is preferentially selected, obstacle avoidance response time is less than or equal to 100ms, and the maximum moving speed is less than or equal to 0.5m/s; The magnetic type special tool library is provided with a tool identification and positioning unit, the tool identification and positioning unit realizes tool replacement through a mechanical arm and a magnetic type joint at the tail end of the mechanical arm, the single tool replacement time is less than or equal to 5 seconds, a tool loss monitoring sensor is arranged in the tool library, and when the tool loss exceeds a threshold value, the tool replacement or warning is automatically triggered.
10. The power transformation equipment operation and maintenance fault positioning and automatic repairing system based on deep learning according to claim 1 is characterized in that the visual monitoring and feedback module comprises an AR interaction unit and a repairing effect verification unit, wherein the AR interaction unit supports two marking modes, namely fault information marking, fault type, positioning coordinates, tools required for repairing and the like, is superimposed in a repairing picture, operation guide marking, guiding a motion track of a mechanical arm through an AR line, and supporting operation maintenance personnel to manually draw the AR marking through a remote control end, and marking instruction transmission delay is less than or equal to 200ms; The repair effect verification unit adopts a multi-dimensional threshold comparison method, the visual dimension compares the pixel difference of the fault area before and after repair, the infrared dimension compares the temperature value, the acoustic dimension compares the characteristic frequency energy, and the repair is judged to be successful when all three indexes meet the threshold requirement.

Description

Power transformation equipment operation and maintenance fault positioning and automatic repairing system based on deep learning Technical Field The invention relates to a deep learning-based power transformation equipment operation and maintenance fault positioning and automatic repairing system, and belongs to the technical field of power transformation equipment fault positioning and intelligent repairing. Background The power transformation equipment is a core infrastructure of a power transmission and distribution link of a power system, the running state of the power transformation equipment directly determines the stability and safety of power supply, the quantity of large-scale hub substation equipment is increased rapidly along with the expansion of the power grid scale, the structure is complex, the unattended substations in remote areas are difficult to inspect manually frequently due to special geographic positions, and the traditional power transformation equipment operation and maintenance mode is gradually exposed to a remarkable short board; The current transformer equipment fault positioning mainly relies on manual inspection and single sensor monitoring, and operation and maintenance personnel required for manual inspection enter a high-voltage area to be inspected one by one, so that the transformer equipment fault positioning is limited by wide equipment distribution and high-voltage environment risks, and the positioning time is usually 2-3 hours, so that fault handling is easy to delay; In terms of fault positioning accuracy, the traditional method relies on equipment identification plates and experience estimation, the accurate association with the physical space of a transformer substation is lacking, the positioning error is often more than 5 meters, the subsequent repair needs to search for fault points for the second time, the efficiency is low, the fault repair link depends on manual operation, even if small parts are loose, poor in contact and other simple faults, personnel are required to carry tools to enter a high-voltage area, the workload is large, safety risks such as electric shock and equipment accidental injury are faced, and the operation and maintenance requirements of a remote unattended transformer substation are especially not adapted; In summary, the prior art is difficult to meet the requirements of efficient positioning, accurate identification and safe repair of operation and maintenance of the power transformation equipment, and a technical scheme for fusing multidimensional sensing, intelligent analysis and autonomous execution is needed, so that the problems of low efficiency, poor precision and high safety risk of the traditional operation and maintenance are solved. Disclosure of Invention In view of the above, the present invention provides a power transformation equipment operation and maintenance fault positioning and automatic repairing system based on deep learning, so as to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial choice. The technical scheme of the invention is realized by that the operation and maintenance fault positioning and automatic repairing system of the power transformation equipment based on deep learning comprises a multi-mode data acquisition module, a data preprocessing and fusion module, a deep learning fault identification module, a three-dimensional space positioning module, a fault decision and repairing scheduling module, a robot automatic repairing module and a visual monitoring and feedback module which are respectively connected with the fault decision and repairing scheduling module and the robot automatic repairing module in a data way; the multi-mode data acquisition module acquires appearance images, temperature distribution data and operation sound signals of the power transformation equipment through a high-definition camera, a thermal infrared imager and a sonar sensor, and performs air synchronization; The data preprocessing and fusion module is used for cleaning and standardizing the data acquired by the multi-mode data acquisition module, and fusing visual, infrared and acoustic characteristics through an attention mechanism; The deep learning fault identification module is used for reasoning the fusion characteristics based on an improved YOLO target detection algorithm and identifying fault types and two-dimensional coordinates of faults in the image; The three-dimensional space positioning module is used for converting the two-dimensional coordinates of the image into world coordinates by combining with the three-dimensional space model of the transformer substation; The fault decision and repair scheduling module is used for evaluating the fault grade according to the fault type and the positioning result and scheduling the operation and maintenance robot; The robot automatic repair module is used for controlling the operation and maintenance robot to ca