CN-121982355-A - Substation operation fault detection method and fault detection model training method

Abstract

The invention provides a method for detecting operation faults of a transformer substation, a training method, a training device, training equipment and a training storage medium of a fault detection model, wherein after an operation diagram of a photovoltaic grid-connected transformer substation obtained by camera shooting is determined to be an image to be detected, the image to be detected is input into a pre-trained fault detection model, and feature extraction is carried out on the image to be detected by utilizing a convolutional neural network in the fault detection model to obtain a feature diagram corresponding to the image to be detected; and then, the target scores corresponding to the boundary frames and the boundary frames are input into a classification network, and the fault types of the target objects in the boundary frames are output after the processing of the classification network. Therefore, the method and the device for detecting the operation faults of the transformer substation improve accuracy of detection of the operation faults of the transformer substation.

Inventors

• ZHAO HONGWEI
• LI LIUBEN
• ZHANG QING
• ZHANG HONGJIANG
• ZOU MINGJUN
• ZHAO HUANDONG
• CHANG BINGBING
• GE XIANGYANG
• MIN HONGWEI
• LI PENG
• CHENG ZHENFEI
• WANG PEIRUI
• WANG PENGFEI
• SHI JIAYAO
• DU JIANING
• LI YUANXIN

Assignees

• 中国三峡新能源(集团)股份有限公司陕西分公司
• 铜川市峡光新能源发电有限公司
• 中国三峡新能源(集团)股份有限公司
• 中国长江三峡集团有限公司

Dates

Publication Date

20260505

Application Date

20251205

Claims (10)

1. 1. A method for detecting an operational failure of a substation, the method comprising: receiving an image to be detected, wherein the image to be detected is a real-time operation image of the photovoltaic grid-connected transformer substation shot by a camera; inputting the image to be detected into a fault detection model, and extracting features of the image to be detected by using a convolutional neural network in the fault detection model to obtain a feature map corresponding to the image to be detected, wherein the fault detection model is obtained by training based on sample data acquired in advance, and the sample data comprises a fault sample, a boundary box sample and a fault class sample; Inputting the feature images corresponding to the images to be detected into an area proposal network, and outputting a plurality of boundary boxes and target scores corresponding to the boundary boxes respectively after the area proposal network is processed, wherein the target scores are used for representing the possibility of fault equipment in the boundary boxes; inputting the multiple bounding boxes and feature images corresponding to the images to be detected into a classification network, and outputting classification results corresponding to the bounding boxes in the multiple bounding boxes respectively after the processing of the classification network, wherein the classification results are used for representing the fault types of the target objects in the bounding boxes.
2. 2. The method according to claim 1, wherein inputting the plurality of bounding boxes into a classification network, after processing by the classification network, outputting classification results respectively corresponding to the bounding boxes in the plurality of bounding boxes, comprises: inputting a first boundary box in the plurality of boundary boxes into a classification network, and extracting a feature vector corresponding to the first boundary box from a feature map corresponding to the image to be detected by using the classification network; and determining a classification result corresponding to the first boundary frame based on the feature vector corresponding to the first boundary frame, wherein the classification result is used for representing probability distribution of the first boundary frame belonging to each fault class.
3. 3. The method according to claim 2, wherein the method further comprises: And inputting the feature vector corresponding to the first boundary frame and the first boundary frame into a regression network, and outputting correction parameters of the first boundary frame after the regression network is processed, wherein the correction parameters are used for adjusting the coordinates of the first boundary frame to obtain an adjusted boundary frame.
4. 4. A method according to claim 3, characterized in that the method further comprises: And performing non-maximum suppression processing on the adjusted boundary box based on the target score to obtain a target boundary box.
5. 5. A method of training a fault detection model, the method comprising: Obtaining a plurality of sample data, wherein the sample data comprises a fault sample, a boundary box sample and a fault category sample, and the aspect ratio of the boundary box sample is determined for equipment type based on the fault sample; And training a model by using the fault sample, the boundary box sample and the fault class sample pair to obtain the fault detection model, wherein the fault detection model is used for carrying out target detection on an image to be detected and outputting a boundary box with a fault and a fault type.
6. 6. The method of claim 5, wherein the method further comprises: Optimizing the fault class sample by using a cross entropy loss function to reduce the difference between the fault type predicted by the fault detection model and the real fault type; And optimizing the bounding box sample by utilizing a bounding box regression loss function to reduce the difference between the bounding box coordinates predicted by the fault detection model and the real bounding box coordinates.
7. 7. A device for detecting an operational failure of a substation, the device comprising: the receiving module is used for receiving an image to be detected, wherein the image to be detected is a real-time operation image of the photovoltaic grid-connected transformer substation shot by a camera; The extraction module is used for inputting the image to be detected into a fault detection model, and extracting the characteristics of the image to be detected by utilizing a convolutional neural network in the fault detection model to obtain a characteristic diagram corresponding to the image to be detected, wherein the fault detection model is obtained by training based on sample data obtained in advance, and the sample data comprises a fault sample, a boundary box sample and a fault class sample; The first processing module is used for inputting the feature images corresponding to the images to be detected into a regional proposal network, and outputting a plurality of boundary boxes and target scores corresponding to the boundary boxes respectively after the regional proposal network is processed, wherein the target scores are used for representing the possibility of fault equipment in the boundary boxes; and the second processing module is used for inputting the plurality of bounding boxes and the target score into a classification network, and outputting classification results corresponding to the bounding boxes in the plurality of bounding boxes respectively after the processing of the classification network, wherein the classification results are used for representing the fault types of the target objects in the bounding boxes.
8. 8. An electronic device, the electronic device comprising: A processor; A memory for storing the processor-executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method of any of claims 1-6.
9. 9. A computer readable storage medium, characterized in that the storage medium stores a computer program for executing the method according to any one of claims 1-6.
10. 10. A computer program product, characterized in that it comprises a computer program/instruction which, when executed by a processor, implements the method according to any of claims 1-6.

Description

Substation operation fault detection method and fault detection model training method Technical Field The disclosure relates to the field of detection of operation faults of a transformer substation, and in particular relates to a method for detecting operation faults of a transformer substation, a training method, a training device, training equipment and a storage medium of a fault detection model. Background With the rapid development of renewable energy sources, a photovoltaic grid-connected transformer substation is taken as a key component of a solar power generation system, and stable operation of the photovoltaic grid-connected transformer substation is crucial for the reliability of energy source supply. However, due to environmental factors, equipment aging, improper operation and the like, various faults may occur in the photovoltaic modules, connecting wires, equipment and the like in the transformer substation, and if the faults are not found and handled in time, the power generation efficiency and the system safety of the photovoltaic grid-connected transformer substation are seriously affected. At present, when the operation faults of the photovoltaic grid-connected transformer substation are detected, the remote pollution detection is mainly carried out on the transformer substation through a hyperspectral remote sensing technology, however, the fault detection mode is easily affected by environmental conditions, and therefore the accuracy of detection results is low. Disclosure of Invention In order to solve the technical problems, the embodiment of the disclosure provides a method for detecting operation faults of a transformer substation, a training method, a device, equipment and a storage medium for a fault detection model, which can improve the accuracy of operation fault detection of the transformer substation. In a first aspect, an embodiment of the present disclosure provides a method for detecting an operation fault of a substation, where the method includes: receiving an image to be detected, wherein the image to be detected is a real-time operation image of the photovoltaic grid-connected transformer substation shot by a camera; inputting the image to be detected into a fault detection model, and extracting features of the image to be detected by using a convolutional neural network in the fault detection model to obtain a feature map corresponding to the image to be detected, wherein the fault detection model is obtained by training based on sample data acquired in advance, and the sample data comprises a fault sample, a boundary box sample and a fault class sample; Inputting the feature images corresponding to the images to be detected into an area proposal network, and outputting a plurality of boundary boxes and target scores corresponding to the boundary boxes respectively after the area proposal network is processed, wherein the target scores are used for representing the possibility of fault equipment in the boundary boxes; and inputting the multiple bounding boxes and the target scores into a classification network, and outputting classification results corresponding to the bounding boxes in the multiple bounding boxes after the processing of the classification network, wherein the classification results are used for representing the fault types of the target objects in the bounding boxes. In a second aspect, an embodiment of the present disclosure provides a method for training a fault detection model, the method including: Obtaining a plurality of sample pairs, wherein the sample pairs comprise a fault sample, a boundary box sample and a fault category sample, and the aspect ratio of the boundary box sample is determined for the equipment type based on the fault sample; And training a model by using the fault sample, the boundary box sample and the fault class sample pair to obtain the fault detection model, wherein the fault detection model is used for carrying out target detection on an image to be detected and outputting a boundary box with a fault and a fault type. In a third aspect, the present disclosure provides a device for detecting an operation failure of a substation, the device comprising: the receiving module is used for receiving an image to be detected, wherein the image to be detected is a real-time operation image of the photovoltaic grid-connected transformer substation shot by a camera; The extraction module is used for inputting the image to be detected into a fault detection model, and extracting the characteristics of the image to be detected by utilizing a convolutional neural network in the fault detection model to obtain a characteristic diagram corresponding to the image to be detected, wherein the fault detection model is obtained by training based on a sample pair obtained in advance, and the sample pair comprises a fault sample, a boundary box sample and a fault class sample; The first processing module is used for inputting the feature imag