CN-121981949-A - DFINE model-based substation site leakage oil identification method and system

CN121981949ACN 121981949 ACN121981949 ACN 121981949ACN-121981949-A

Abstract

The invention discloses a method and a system for identifying leakage oil on site of a transformer substation based on DFINE model, which belong to the technical field of transformer substation leakage oil detection, and ensure that the sample covers the full leakage condition by taking images recorded in the historical operation of the transformer substation as data samples, meanwhile, the difference between the leakage oil and a normal scene is enhanced by carrying out characteristic extraction on the sample, and a leakage oil identification model is built by taking DFINE model as a core, so that the automation, quasi-real-time identification and early warning of the leakage oil are realized, the traditional manual regular inspection is replaced, the problems of low manual inspection efficiency, limited coverage range and difficult detection of pain points of hidden/tiny leakage are solved, and the operation and maintenance labor cost and the environmental risk are reduced.

Inventors

JIANG SHIJIN
WANG CHENGUANG
GONG YUNING
DONG SHENG
ZHU CHENHAO
WEI WEI
XU KE
TANG JUN
Zhang saifan
PENG CHANG
YU YIFENG
Weng Xiaqing
LU SHUAISHUAI
TIAN ZHIPING
YU JIAN

Assignees

国网浙江省电力有限公司金华供电公司

Dates

Publication Date: 20260505
Application Date: 20251222

Claims (10)

1. The method for identifying the transformer substation on-site leakage oil based on DFINE model is characterized by comprising the following steps: S1, collecting historical operation images of a transformer substation, grouping the historical operation images to obtain a normal sample and an oil leakage sample by taking whether oil leakage exists as a grouping condition, and labeling the oil leakage sample; s2, extracting features of a normal sample and an oil leakage sample to obtain target features, and fusing the target features based on a dynamic feature fusion mechanism to obtain a sample data vector; s3, carrying out parameter training on the sample data vector by taking DFINE model as a core and taking the oil seepage existing in the identified transformer substation as a target to obtain an oil seepage identification model; S4, acquiring a site operation picture of the transformer substation in real time, taking the site operation picture as input of an oil seepage identification model, and outputting the identification result.
2. The DFINE model-based substation site leakage oil identification method as claimed in claim 1, wherein: s1, collecting historical operation images of a transformer substation, and grouping the historical operation images to obtain a normal sample and an oil leakage sample by taking whether oil leakage exists as a grouping condition, and marking the oil leakage sample, wherein the method comprises the following steps of; Extracting visible light images of all equipment of a transformer substation historical time node from monitoring equipment of a transformer substation to serve as historical operation images; taking the image with the oil leakage condition as an oil leakage sample, and taking the rest other images as normal samples; and framing the region where the oil leakage occurs in a boundary box form for each image in the oil leakage sample, and marking the positions of oil stains, oil drops and the liquid level of the oil level gauge.
3. The DFINE model-based substation site leakage oil identification method as claimed in claim 1, wherein: In S2, the target features include device type, environmental conditions, extent of leakage, and morphology of leakage; the equipment type at least comprises a transformer, a reactor and a sleeve; the environmental conditions include sunny days, overcast days, nights, strong light and weak light; The leakage degree includes slight leakage, moderate leakage and severe leakage; the leak patterns include dot, line, and sheet.
4. A method for identifying oil leakage in a transformer substation site based on DFINE model as set forth in claim 3, wherein: S2, fusing target features based on a dynamic feature fusion mechanism to obtain a sample data vector, wherein the method comprises the following steps of: Based on the leakage degree and leakage form in the target feature, introducing an attention mechanism to allocate feature weights to the leakage degree and leakage form; the greater the extent of leakage or the greater the area of oil stain formation in the leakage pattern, the higher the weight assigned; And mapping all types of features in the target features to the same semantic space to obtain sample data vectors.
5. A method for identifying oil leakage in a transformer substation site based on DFINE model as set forth in claim 3, wherein: The seepage oil identification model comprises a data layer, a multi-level feature layer and a framing layer; the data layer adopts a hierarchical grid network as a backbone network of the leakage oil identification model and is used for carrying out image optimization processing on the field operation picture; The multi-level feature layer distributes high weight to the oil stain-containing area in the field operation picture based on leakage degrees and leakage forms of different degrees, and distributes low weight to the background area for performing feature enhancement treatment on the field operation picture; The framing layer frames and displays the leakage oil stain on the field operation picture based on the output of the multi-level characteristic layer.
6. The DFINE model-based substation site leakage oil identification method as claimed in claim 5, wherein: The multi-level feature layer comprises a first feature layer, a second feature layer and a third feature layer, wherein the initial weights of the first feature layer, the second feature layer and the third feature layer are the same, and the sum of the weights is 1; If oil stains exist in the field operation picture and the area of the oil stains is smaller than or equal to a first area threshold value, the weight of the first characteristic layer is increased; If the oil stain area is between the first area threshold value and the second area threshold value, the weight of the second characteristic layer is increased; If the area of the oil stain is larger than or equal to the second area threshold value, the weight of the third characteristic layer is increased; If no oil stain exists, the initial weight is kept unchanged.
7. The DFINE model-based substation site leakage oil identification method as claimed in claim 1, wherein: And S4, the identification result comprises a field operation picture after oil stain framing, an area coordinate where oil leakage occurs and the position of the liquid level of the oil level gauge.
8. The substation site leakage oil identification system based on DFINE model is suitable for the substation site leakage oil identification method based on DFINE model according to any one of claims 1-7, and is characterized by comprising a data reading module, a feature extraction module, a model construction module, a detection module and an image acquisition module; The data reading module reads historical operation images of the transformer substation, groups the historical operation images by taking whether oil leakage exists as a grouping condition to obtain a normal sample and an oil leakage sample, and marks the oil leakage sample; the feature extraction module performs feature extraction on the normal sample and the oil leakage sample to obtain target features, and fuses the target features based on a dynamic feature fusion mechanism to obtain sample data vectors; the model construction module takes DFINE models as cores, performs parameter training on sample data vectors with the aim of identifying leakage oil existing in the transformer substation to obtain leakage oil identification models, and carries the leakage oil identification models into the detection module; The detection module is carried with a leakage oil identification model, the image acquisition module acquires the on-site operation picture of the transformer substation in real time, and the on-site operation picture is sent to the detection module to be output to obtain an identification result.
9. The computer equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are in communication with each other through the communication bus, the memory is used for storing a computer program, and the processor is used for realizing the steps of the DFINE-model-based substation field seepage oil identification method according to any one of claims 1 to 7 when the program stored on the memory is executed.
10. A computer readable storage medium, characterized in that, a computer program is stored in the computer readable storage medium, and the computer program realizes the steps of the method for identifying the leakage oil of the transformer substation site based on DFINE model according to any one of claims 1-7 when being executed by a processor.

Description

DFINE model-based substation site leakage oil identification method and system Technical Field The invention relates to the technical field of transformer substation leakage oil detection, in particular to a method and a system for identifying transformer substation field leakage oil based on DFINE models. Background The transformer substation is used as a core hub of the power system, and the stable and reliable operation of the equipment directly relates to the safety and the power supply quality of the power grid. Among the key equipment of many transformer substations, oil-filled electrical equipment such as transformers, circuit breakers, reactors, transformers and the like play a vital role. These devices are internally filled with insulating oil (e.g. transformer oil) whose main functions include insulation, heat dissipation and arc extinction. However, in the long-term operation of the apparatus, leakage of insulating oil is extremely liable to occur due to various factors such as material aging, deterioration of sealing members, mechanical vibration, thermal expansion and contraction, damage by external force, or manufacturing and installation defects. If the on-site oil leakage problem cannot be found and treated in time, a series of serious consequences such as equipment performance reduction, fault risk increase, environmental pollution, operation and maintenance cost increase and the like are brought. Traditionally, oil leakage inspection has relied on manual periodic inspection. However, the transformer substation equipment is numerous and widely distributed, part of the equipment is higher in position or complex in environment (such as night, rainy and snowy weather and strong electromagnetic interference areas), and the problems of low efficiency, limited coverage, susceptibility to subjective factors, difficulty in finding tiny or hidden leakage points and the like exist in manual inspection. Especially in unattended or unattended stations, timeliness is difficult to guarantee. Therefore, how to efficiently, accurately and real-timely automatically identify the leakage oil of the field equipment of the transformer substation, and realize early warning becomes an urgent need for guaranteeing the safety of a power grid, improving the intelligent level of operation and maintenance and reducing the cost of operation and maintenance and environmental risk. Chinese patent, publication No. CN113033322A, publication No. 2021, 6 and 25, discloses a method for identifying leakage oil hidden danger of transformer substation oil filling equipment based on deep learning, which comprises the steps of firstly collecting leakage oil hidden danger image samples of transformer substation equipment, then expanding the leakage oil hidden danger image samples of the transformer substation equipment, marking the severity of the leakage oil hidden danger samples of the transformer substation equipment, training a leakage oil hidden danger detection model of the transformer substation equipment, deploying the leakage oil hidden danger detection model of the transformer substation equipment after training, and finally detecting the severity of leakage oil hidden danger of a substation inspection image, wherein the problems of tiny leakage points or hidden leakage points are difficult to find under the condition of higher positions of part of equipment or complex environments, and the capability of the method for coping with complex environments is low. Disclosure of Invention Aiming at the problem of poor recognition efficiency and precision caused by low automation degree of the existing leakage oil recognition method, the invention provides a DFINE model-based substation field leakage oil recognition method and system, which ensure that the sample covers the full leakage situation by taking images recorded in the historical operation of the substation as data samples, simultaneously perform characteristic extraction on the sample to enhance the difference between leakage oil and normal scenes, and take a DFINE model as a core to construct a leakage oil recognition model, thereby realizing automatic, quasi-real-time recognition and early warning of the leakage oil, replacing the traditional manual regular inspection, solving the problems of low manual inspection efficiency, limited coverage and difficult detection of hidden/tiny leakage pain points, and further reducing the operation and maintenance labor cost and environmental risk. In a first aspect, the technical scheme provided by the embodiment of the invention is that the substation on-site leakage oil identification method based on DFINE model comprises the following steps: S1, collecting historical operation images of a transformer substation, grouping the historical operation images to obtain a normal sample and an oil leakage sample by taking whether oil leakage exists as a grouping condition, and labeling the oil leakage sample; s2, extracting features