Transformer substation insulation fault monitoring device and monitoring method thereof

Abstract

The invention relates to the technical field of power equipment monitoring, and discloses a transformer substation insulation fault monitoring device and a monitoring method thereof.The transformer substation insulation fault monitoring device comprises a background terminal, an installation shell, an imaging unit, a processing module and a communication unit, the imaging unit is installed on the installation shell, and the imaging unit comprises a visible blind imaging module, a visible light imaging module and a yellow light imaging module; light of different wave bands can penetrate through the different modules, insulation terminals of transformer substation equipment are imaged respectively, and the processing module judges whether an early-stage discharge accident or a medium-stage discharge accident occurs according to visible blind image signals, monitoring equipment image signals and yellow light image signals collected by the imaging unit. The communication unit sends the accident information judged by the processing module to a background terminal; therefore, the monitoring device can identify the equipment characteristics of the equipment of the transformer substation when the equipment of the transformer substation has faults at all stages, thereby solving the problem that the single-band imaging equipment in the prior art is difficult to monitor the discharge fault accident of the equipment of the transformer substation in time.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment monitoring, and particularly to a substation insulation fault monitoring device and a monitoring method thereof. Background Art

[0002] With the development of the power system, the equipment maintenance and management of substations have received increasing attention. Among them, insulation faults have become one of the main problems affecting the safe operation of power equipment. Traditional insulation fault monitoring methods mainly rely on manual regular inspections. This method is difficult to detect potential faults in a timely manner, and there are risks of misdetection and missed detection, with a relatively low defect detection rate. At the same time, manual inspections also consume a large amount of human and material resources, and are prone to cause equipment damage and power grid blackout accidents during the defect discrimination process.

[0003] Existing substation insulation fault monitoring and processing means often rely only on imaging devices of a single wavelength band; such devices can only monitor after the insulation problem has deteriorated to a certain extent and obvious discharge faults have occurred, and cannot accurately identify the conditions of insulation terminals installed inside substation equipment, resulting in a relatively slow response of maintenance personnel to insulation fault accidents.

[0004] Therefore, existing monitoring methods cannot provide sufficient early warning and judgment information in the early and middle stages of insulation faults in substation equipment, and it is difficult to take effective measures in a timely manner according to the severity of the faults, resulting in further expansion of the accident scope. Summary of the Invention

[0005] The object of the present invention is to provide a substation insulation fault monitoring device and a monitoring method thereof, which solve the problem that imaging devices of a single wavelength band in the prior art are difficult to monitor discharge fault accidents in substation equipment in a timely manner.

[0006] To achieve the above object, the technical solution of the present invention is as follows: A substation insulation fault monitoring device includes: a monitoring device body and a background terminal, and the monitoring device body includes:

[0007] An installation housing;

[0008] An imaging unit, the imaging unit is installed in the installation housing, and the imaging unit includes a visible-blind imaging module, a visible light imaging module, and a yellow light imaging module. The visible-blind imaging module, the visible light imaging module, and the yellow light imaging module can transmit light of different wavelength bands, and respectively image the insulation terminals of substation equipment, and respectively collect visible-blind image signals, monitoring equipment image signals, and yellow light image signals. The visible-blind image signals are collected by the visible-blind imaging module, the monitoring equipment image signals are collected by the visible light imaging module, and the yellow light image signals are collected by the yellow light imaging module;

[0009] A processing module, which is installed in the installation housing. The processing module is communicatively connected to a visible-blind imaging module, a visible-light imaging module, and a yellow-light imaging module. The processing module determines whether it belongs to an early-stage discharge accident or a mid-stage discharge accident based on the visible-blind image signal, the monitoring device image signal, and the yellow-light image signal collected by the imaging unit. When the image area collected by the visible-blind image signal > 50% of the monitoring area collected by the monitoring device image signal, it is determined as an early-stage discharge accident. When the image area collected by the yellow-light image signal > 50% of the monitoring area collected by the monitoring device image signal, it is determined as a mid-stage discharge accident;

[0010] A communication unit, which sends the accident information determined by the processing module to a background terminal.

[0011] As an optional solution, the background terminal is further configured to:

[0012] When it is determined as an early-stage discharge accident, send a first warning message;

[0013] When it is determined as a mid-stage discharge accident, send a second warning message.

[0014] As an optional solution, the processing module determines whether the substation equipment belongs to an early-stage discharge accident or a mid-stage discharge accident, specifically including:

[0015] Insulated terminal contour recognition, using an optical frequency feature decoding network to fuse and decode the image information collected by the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module to ensure that the shape of the insulated terminal is accurately recognized under different spectra;

[0016] Insulated terminal image semantic segmentation, after the insulated terminal contour recognition step is completed, using the DeepLabV3 model to segment the image information of the insulated terminal into precise functional areas;

[0017] Statistics on the situation of a single insulated terminal. After segmenting the functional areas of the insulated terminal, further adjust the boundaries of the functional areas through an adaptive edge dilation algorithm, and then calculate the proportion of the imaging of the visible-blind module or the yellow-light module in each functional area and compare it with the set boundary value of 50%.

[0018] As an optional solution, the monitoring device body further includes a carrying platform and universal wheels;

[0019] The carrying platform is used to install the installation housing;

[0020] The universal wheels are arranged at the bottom end face of the carrying platform, and the universal wheels are used for the carrying platform to move.

[0021] As an alternative, a solar panel is further provided on the bearing platform, and the power output end of the solar panel is electrically connected to the processing module.

[0022] As an alternative, a push rod is further provided on the bearing platform. The fixed end of the push rod is fixedly connected to the bearing platform, and a rubber pad is provided at the holding end of the push rod.

[0023] As an alternative, the optical axes of the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module are parallel.

[0024] As an alternative, a first filter is provided in the visible-blind imaging module, and the light transmission band of the first filter is 300 - 400 nm;

[0025] A second filter is provided in the visible-light imaging module, and the light transmission band of the second filter is 400 - 760 nm;

[0026] A third filter is provided in the yellow-light imaging module, and the light transmission band of the third filter is 580 - 597 nm.

[0027] A substation insulation fault monitoring method includes the substation insulation fault monitoring device according to claims 1 - 8, and the monitoring method includes:

[0028] The insulation terminal is imaged by the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module respectively, and a visible-blind image signal, a monitoring device image signal, and a yellow-light image signal are collected respectively.

[0029] When the image area collected by the visible-blind image signal > 50% of the monitoring area collected by the monitoring device image signal, it is determined as an early discharge accident; when the image area collected by the yellow-light image signal > 50% of the monitoring area collected by the monitoring device image signal, it is determined as a mid-term discharge accident.

[0030] As an alternative, the monitoring method further includes the registration steps of the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module;

[0031] Optimization in the feature extraction stage, using multi-scale convolutional feature extraction to optimize the different imaging features extracted by different modules.

[0032] Optical flow estimation and dense registration, using PWC-Net for multi-modal optical flow estimation, processing the images collected by different modules through feature pyramid hierarchical processing, and introducing a self-supervised learning mechanism in the optical flow estimation process.

[0033] Based on the detail optimization of geometric transformation, after optical flow estimation, a model combining affine transformation and perspective transformation is introduced. By calculating the dense correspondence of different pixels, the image details are further optimized, and the edge features in the corresponding module image are added as constraint conditions in the registration process for the processing of the edge of the insulating material.

[0034] Network loss function design. A loss function including optical flow loss, smoothness loss, loss function based on structural similarity, and feature alignment loss function is constructed. The feature alignment loss function measures the alignment of image features under different modules through the cosine similarity in the convolutional feature space.

[0035] A substation insulation fault monitoring device and its monitoring method according to the present invention. Compared with the prior art, the beneficial effects are as follows: In the installation shell of the embodiment of the present invention, an imaging unit is installed. The imaging unit monitors the insulating terminal through a visible blind imaging module, a visible light imaging module, and a yellow light imaging module that can identify light of multiple bands. The processing module determines the fault condition of the monitored insulating terminal according to the image data collected by the visible blind imaging module, the visible light imaging module, and the yellow light imaging module, and sends the determined fault condition to the background terminal through the communication unit for the staff to view, so that the monitoring device can identify the equipment characteristics of the substation equipment in each stage of the fault, thereby solving the problem that the imaging equipment of a single band in the prior art is difficult to monitor the discharge fault accident of the substation equipment in time. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is the overall schematic diagram of the substation insulation fault monitoring device according to the embodiment of the present invention;

[0037] Figure 2 is the side view of the substation insulation fault monitoring device according to the embodiment of the present invention;

[0038] Figure 3 is the specific steps for the processing module of the substation insulation fault monitoring device according to the embodiment of the present invention to determine whether the substation equipment belongs to an early discharge accident or a mid-term discharge accident;

[0039] In the figure, 1, installation shell; 2, communication unit; 3, bearing platform; 4, universal wheel; 5, solar panel; 6, push rod. DETAILED DESCRIPTION OF THE INVENTION

[0040] [[ID=??]]The following combines the drawings and embodiments to further describe in detail the specific embodiments of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0041] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention.

[0042] In the description of the present invention, it should be understood that the terms "connected", "connected to", "fixed", etc. used in the present invention should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection, or a welded connection; it can be directly connected, or indirectly connected through an intermediate medium. It can be the communication inside two elements or the interaction relationship between two elements, unless otherwise clearly defined. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0043] As Figures 1 to 3 shown, a preferred embodiment of the present invention provides a substation insulation fault monitoring device, which is characterized by including: a monitoring device body and a background terminal, and the monitoring device body includes:

[0044] An installation housing 1, the installation housing 1 is a camera housing without a lens, and the observation surface of the camera housing for installing the lens can cover the area where the substation equipment is located;

[0045] An imaging unit, the imaging unit is installed in the installation housing 1, and the imaging unit includes a visible blind imaging module, a visible light imaging module, and a yellow light imaging module. The visible blind imaging module, the visible light imaging module, and the yellow light imaging module can transmit light of different bands and respectively image the insulation terminals of the substation equipment. The identified light band of the visible light imaging module is set relatively wide, and the overall situation of the substation equipment can be observed. However, due to the special nature of light, when the equipment emits light, the light emitted by it spreads out in a surface. At this time, for the module with a relatively wide identified light band, the specific location of the fault cannot be located in the imaging of this module. Therefore, after using the visible light imaging module to identify the faulty equipment, we use the yellow light imaging module to accurately locate the fault point of the corresponding equipment; the visible blind imaging module, the visible light imaging module, and the yellow light imaging module respectively collect visible blind image signals, monitoring equipment image signals, and yellow light image signals. The visible blind image signals are collected by the visible blind imaging module, the monitoring equipment image signals are collected by the visible light imaging module, and the yellow light image signals are collected by the yellow light imaging module;

[0046] The processing module is installed in the installation housing 1. The processing module is communicatively connected to the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module. The processing module determines whether it belongs to an early-stage discharge accident or a mid-stage discharge accident according to the visible-blind image signal, the monitoring device image signal, and the yellow-light image signal collected by the imaging unit. When the image area collected by the visible-blind image signal > 50% of the monitored area collected by the monitoring device image signal, it is determined as an early-stage discharge accident. When the image area collected by the yellow-light image signal > 50% of the monitored area collected by the monitoring device image signal, it is determined as a mid-stage discharge accident;

[0047] The communication unit 2 sends the accident information judged by the processing module to the background terminal. Relevant personnel can manually confirm the imaging of the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module through the background terminal, and can also confirm the fault information;

[0048] Based on this, the monitoring device completes the monitoring of the substation equipment through the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module arranged on the installation housing 1, and the processing module judges and analyzes the image information identified by the visible-blind imaging module, the visible-light imaging module, and the yellow-light imaging module. The communication unit 2 uploads the judgment result of the processing module to the background terminal, so that the monitoring device can identify the equipment characteristics of the substation equipment in various stages of failure, thereby solving the problem that it is difficult for a single-band imaging device in the prior art to timely monitor the discharge fault accident of the substation equipment.

[0049] Further, as Figures 1 to 3 shown, the background terminal is further configured to:

[0050] When it is determined as an early-stage discharge accident, send a first warning message;

[0051] When it is determined as a mid-stage discharge accident, send a second warning message;

[0052] The specific ways of the first warning message and the second warning message are not further limited here. There are obvious differences in the warning methods for personnel between the first warning message and the second warning message. For example, the first warning message is that the background terminal sends a text message warning to the personnel, and the second warning message is that a buzzer and a warning light sound on-site, so that relevant staff can quickly judge the insulation accident stage of the substation through different warning messages.

[0053] Further, as Figure 3 shown, the processing module determines whether the substation equipment belongs to an early-stage discharge accident or a mid-stage discharge accident, specifically including:

[0054] Step 1, Insulated Terminal Profile Recognition. Use the optical frequency feature decoding network to fuse and decode the image information collected by the visible blind imaging module, visible light imaging module, and yellow light imaging module to ensure that the outer shape of the insulated terminal is accurately recognized under different spectra.

[0055] The Spectral Feature Decoding Network (SFDN) is used to recognize the outer contour of the insulated terminal. The images captured by the multispectral imaging module (visible light, yellow light, visible blind imaging) are fused and decoded to ensure that the outer shape of the terminal is accurately recognized under different spectra. SFDN is optimized for the shape and edge characteristics of the terminal to ensure accurate recognition of the terminal's contour under complex background or lighting change conditions.

[0056] 1. Multispectral Feature Extraction. The outer contour of the terminal presents different edges and details under different spectra. SFDN first extracts edge features from visible light, yellow light, and visible blind imaging through a convolutional neural network (such as ResNet). For example, yellow light may highlight the fine contours on the surface of the terminal, while visible light can clearly present the overall shape of the terminal. These features are integrated into a multi-scale feature pyramid to comprehensively capture the details of the outer contour of the terminal.

[0057] 2. Frequency Domain Feature Analysis. To more accurately identify the edges of the terminal, SFDN analyzes the edge features of the image in the frequency domain through the Fast Fourier Transform (FFT). This module can distinguish the frequency differences between the terminal and the background, filter out noise, and enhance the saliency of the terminal contour to ensure the consistency of the outer contour in the frequency domain under different spectra.

[0058] 3. Self-Attention Mechanism. For the recognition of the outer contour of the terminal, SFDN introduces a multi-head self-attention mechanism to highlight the key edge regions of the terminal. Through the spatial attention mechanism, the system can automatically focus on the edges of the terminal and reduce background interference. The channel attention mechanism automatically adjusts the weights of different spectra according to the edge detection ability of the spectra. For example, when visible light performs well in a large edge area, the attention mechanism will increase its weight to ensure the coherence and integrity of the outer contour recognition.

[0059] 4. Feature Fusion and Decoding. After multispectral feature extraction and attention processing, SFDN fuses the edge features under each spectrum to form a comprehensive contour image of the terminal. The decoding layer generates the complete outer contour of the terminal through these fused features, ensuring clear edges, no noise, and adaptability to spectral differences under different imaging conditions.

[0060] This step focuses on the recognition of the outer contour of the insulating terminal. Through multispectral fusion, it ensures that the edges of the terminal can be accurately captured even in a complex industrial environment. During the recognition process, the frequency-domain feature analysis module can effectively distinguish the terminal from the background and enhance the edge clarity; the self-attention mechanism can intelligently adjust the spectral weights to ensure that the outer contour can be accurately presented whether in visible light, blue light, or visible-blind imaging.

[0061] Step 2, semantic segmentation of the insulating terminal image. After the completion of the insulating terminal contour recognition step, the DeepLabV3 model is used to segment the image information of the insulating terminal into precise functional regions.

[0062] After the recognition of the outer contour of the insulating terminal is completed, the DeepLabV3 model can be used for semantic segmentation of the terminal. DeepLabV3 is a powerful deep convolutional neural network, especially suitable for dealing with complex semantic segmentation tasks.

[0063] First, the outer contour image is input into the DeepLabV3 model. The model will use its unique dilated convolution to expand the receptive field, so as to obtain more context information while maintaining the resolution. This convolution method can capture the overall structure and local details of the terminal. Whether it is a large insulating layer or a small wiring area, DeepLabV3 can accurately identify through multi-scale feature extraction methods.

[0064] Subsequently, the feature fusion mechanism of DeepLabV3 combines feature maps of different resolutions to achieve precise segmentation at various scales. Through its Atrous Spatial Pyramid Pooling (ASPP) module, DeepLabV3 can maintain precise discrimination of different parts of the insulating terminal under various complex backgrounds.

[0065] Finally, the model assigns each pixel point to different semantic categories, such as the insulating layer, wiring area, etc., through a classification layer to generate an accurate segmentation result. This method is not only efficient but also easy to implement, providing precise geometric information for industrial automation operations.

[0066] Step 3, statistics of the situation of a single insulating terminal. After the functional regions of the insulating terminal are segmented, after further adjusting the boundaries of the functional regions through the adaptive edge dilation algorithm, the proportion of the imaging of the visible-blind module or the yellow light module in each functional region is statistically calculated and compared with the set boundary value of 50%.

[0067] And in the process of statistically calculating the situation of a single insulating terminal:

[0068] Boundary confirmation: After segmenting the internal area of the insulating terminal, the Adaptive Edge Expansion Algorithm can be used. This algorithm adaptively adjusts the boundary of the segmented area by analyzing the spectral differences between the inside and the edge of the terminal, ensuring a more natural transition between the edge and the internal area. This method can avoid the inaccuracies caused by simple binary processing. Especially in visible-blind imaging, the optical properties of the internal area and the edge area are often similar, and traditional segmentation is prone to misjudgment.

[0069] Then, within the boundary, count the proportion of the visible-blind image in the internal area using the Dynamic Feature Clustering method. This method divides the internal area into different sub-areas (such as high-reflection areas, low-transmission areas, etc.) by clustering and analyzing the spectral responses of each pixel, and then adjusts the weights according to the physical properties of different sub-areas to generate a multi-level regional proportion statistic. This dynamic clustering statistic can distinguish the differences in the internal area of the image terminal according to characteristics such as the terminal material properties, shooting angle, and the formation of light spots by sunlight, and comprehensively obtain the proportion of the visible-blind image.

[0070] Dynamic Feature Clustering is used in the detection of transformer insulating terminals for detailed division and statistics of the internal area. First, based on the pixel features in the visible-blind image, use K-Means Clustering to divide the terminal image into K different regions. The spectral response feature of each pixel is represented as F(x, y), and preliminary clustering is performed through the Euclidean distance. To adapt to the changes in the terminal material and spectral properties, an adaptive weight adjustment mechanism is introduced to dynamically update the features of each pixel. The weight coefficient α(x, y) is calculated according to the variance of the local area: F′(x, y) = α(x, y)

[0071] Among them, the larger the variance, the more it represents that the pixel is in the transition area or the edge, and the lower the weight. The updated features are used for re-clustering to generate an adaptive clustering result. After completing the clustering, calculate the area proportion of each region Ck:

[0072]

[0073] Furthermore, different regions are given weights wk according to the spectral properties of the regions, and the final dynamic proportion is:

[0074] Pk′ = wk·Pk

[0075] Use 50% as the fault experience threshold to further predict faults.

[0076] After the above three steps, the processing module can accurately identify the fault area and fault location of the insulating terminal, providing conditions for timely detecting discharge fault accidents of substation equipment and issuing correct warnings.

[0077] Further, as Figures 1 to 2 shown, the monitoring device body further includes a carrying platform 3 and universal wheels 4;

[0078] The carrying platform 3 is used to install the installation shell 1. The installation shell 1 is fixedly connected to the carrying platform 3 through a fixed rod. The fixed rod with a certain length can ensure the viewing range of the visible blind imaging module, visible light imaging module, and yellow light imaging module arranged in the installation shell 1.

[0079] The universal wheels 4 are arranged on the bottom end surface of the carrying platform 3, and the universal wheels 4 are used for moving the carrying platform 3.

[0080] A push rod 6 is further arranged on the carrying platform 3. The fixed end of the push rod 6 is fixedly connected to the carrying platform 3, and a rubber pad is arranged at the holding end of the push rod 6. The arrangement of the universal wheels 4 and the push rod 6 enables the monitoring device to be pushed away from or close to the fault point by the staff when necessary, ensuring the equipment safety of the monitoring device or making the imaging clarity of the visible blind imaging module, visible light imaging module, and yellow light imaging module higher.

[0081] Further, as Figures 1 to 2 shown, a solar panel 5 is further arranged on the carrying platform 3. The power output end of the solar panel 5 is electrically connected to the processing module. The solar panel 5 conveys the converted electric energy to the processing module. The solar panel 5 can provide part of the power for the processing module, enabling the monitoring device to still maintain operation for a period of time in the event of a power outage emergency, thereby indirectly ensuring the equipment safety of the substation.

[0082] Further, not shown in the drawings in this embodiment, the optical axes of the visible blind imaging module, the visible light imaging module, and the yellow light imaging module are parallel to ensure that the outer contours of the images of the three modules can correspond, facilitating comparison by the processing module or the staff.

[0083] Further, this embodiment is not shown in the drawings. A first filter is provided in the visible-blind imaging module. The light waveband that can pass through the first filter is 300 - 400 nm, and the first filter can be adjusted according to actual use conditions to transmit invisible light of other wavebands. A first area CCD is also provided in the visible-blind imaging module. The first area CCD converts the invisible light transmitted through the first filter into image information. The material of the first area CCD is not further limited herein. When the invisible light transmitted through the first filter is ultraviolet light, the first area CCD uses a special silicon-based material to improve the sensitivity to ultraviolet light. If necessary, an ultraviolet light filter can be added at the corresponding position of the first area CCD. When the invisible light transmitted through the first filter is infrared light, the first area CCD uses materials sensitive to infrared light such as germanium or indium gallium arsenide.

[0084] A second filter is provided in the visible light imaging module. The light waveband that can pass through the second filter is 400 - 760 nm. Since this waveband covers the spectral range that can be perceived by the human naked eye, the visible light imaging module can provide the monitoring area picture information to the processing module or relevant staff as comprehensively as possible. A second area CCD is also provided in the visible light imaging module. The second area CCD is arranged behind the second filter, and the imaging optical axis of the second area CCD is parallel to the transmissible optical axis of the second filter. Since silicon material has good transmissibility to visible light, the material used for the second area CCD does not require additional design. The second area CCD converts the visible light transmitted through the second filter into image information.

[0085] A third filter is provided in the yellow light imaging module. The light waveband that can pass through the third filter is 580 - 597 nm, and this waveband is yellow light. If necessary, the third filter can be replaced according to the light waveband actually emitted when the equipment in the substation fails. The third filter narrows the waveband of the transmissible light to match the light waveband emitted when the equipment in the substation fails, greatly reducing the imaging area of the yellow light imaging module. After being identified by the processing module, precise positioning of the equipment fault point in the substation is achieved.

[0086] This embodiment is not shown in the drawings. The present invention also provides a method for monitoring substation insulation faults, including the substation insulation fault monitoring device described in claims 1 - 8. The monitoring method includes:

[0087] Imaging the insulating terminal through the visible-blind imaging module, visible light imaging module, and yellow light imaging module respectively, and collecting visible-blind image signals, monitoring device image signals, and yellow light image signals respectively.

[0088] When the image area captured by the visible blind image signal > 50% of the monitored area captured by the monitoring device's image signal, it is determined as an early discharge accident. When the image area captured by the yellow light image signal > 50% of the monitored area captured by the monitoring device's image signal, it is determined as a mid-term discharge accident.

[0089] During the actual operation process, the visible blind imaging module, the visible light imaging module, and the yellow light imaging module may have the following problems during imaging, resulting in the inability of the modules to correspond in imaging; to avoid mutual occlusion of parts between the modules, there will be some deviations in the installation positions of the three modules; for the equipment in different substations, the surface of its insulating terminals may have complex textures, minute structures, or local reflections; different modules have different responses to different materials.

[0090] Furthermore, not shown in the attached drawings in this embodiment, the monitoring method further includes the registration steps of the visible blind imaging module, the visible light imaging module, and the yellow light imaging module.

[0091] Step 1, optimization of the feature extraction stage, using multi-scale convolutional feature extraction to optimize the different imaging features extracted by different modules.

[0092] Since the performance of the insulating terminal under visible light, yellow light, and visible blind imaging may vary greatly, it is necessary to use multi-scale convolutional feature extraction in the feature extraction stage, especially focusing on extracting local features in areas such as surface texture and edges. And fine-tune according to the material of the insulating terminal through a pre-trained deep network (such as ResNet or VGG).

[0093] For the yellow light imaging module, since it may have a better response in terms of high-frequency surface texture, it can enhance the capture of high-frequency features; while under the visible blind imaging module, it may pay more attention to the internal defects and penetrability features of the material. According to these characteristics, an adaptive convolutional kernel can be used during feature extraction to adapt to the feature differences of different modules processed at different scales.

[0094] Step 2, optical flow estimation and dense registration, using PWC-Net for multi-modal optical flow estimation, and processing the images captured by different modules through feature pyramid hierarchical processing, and introducing a self-supervised learning mechanism during the optical flow estimation process.

[0095] The optical flow field estimation uses PWC-Net for multi-modal optical flow estimation. This network processes the image pair through feature pyramid hierarchical processing, first estimating the optical flow in a larger range at the rough level, and then gradually refining to the detailed level. This method is especially suitable for scenarios such as the insulating terminals of substation equipment, which have a delicate surface structure but may also have large-scale deformations.

[0096] Due to the different spectral ranges of different modules having different reflectivities, light transmittances, and surface responses to the insulating material, a self-supervised learning mechanism may need to be introduced during the optical flow estimation process. This mechanism allows the network to automatically adjust the optical flow field, enabling the features of the same object to be aligned in images from different modules, even if the features appear visually inconsistent.

[0097] Step 3, Detail optimization based on geometric transformation. After optical flow estimation, introduce a model that combines affine transformation and perspective transformation. Further optimize the image details by calculating the dense correspondence of different pixels, and add the edge features in the corresponding module images as constraint conditions during the registration process for the edges of the insulating material.

[0098] Regarding the geometric deformation problems that may be caused by surface reflections or texture differences of the insulating terminals of substation equipment, a model that combines affine transformation and perspective transformation can be introduced. After preliminary optical flow estimation, calculate the dense correspondence of each pixel and further optimize it through geometric transformation to ensure the registration accuracy, especially in the edge and detail areas of substation equipment.

[0099] Moreover, since the edges of the insulating material used for the insulating terminals may be relatively clear and have important reference significance, edge feature constraints can be introduced during registration. By detecting the edge features in the module images and using these features as constraint conditions during the registration process, ensure that the structural features in the images can be accurately aligned.

[0100] Step 4, Network loss function design. Construct a loss function that includes traditional optical flow loss (based on pixel intensity differences), smoothness loss, loss functions based on structural similarity (such as SSIM loss), and feature alignment loss function. The feature alignment loss function measures the alignment of image features under different modules through the cosine similarity in the convolutional feature space, so as to ensure that the imaging features of the insulating terminals for each module are maximally matched in the high-dimensional feature space.

[0101] After the above steps, the registration of the visible blind imaging module, visible light imaging module, and yellow light imaging module is achieved. At this time, further process the three groups of image information, and align the image features identified by the three modules to facilitate comparison by the processing module or relevant staff.

[0102] Processing step 1, Fuse the image information identified by the visible blind imaging module, visible light imaging module, and yellow light imaging module

[0103] Fuse the registration results using Laplacian Pyramid Blending to preserve the valuable details in the images of different modules. In industrial applications, especially during material surface defect detection, this can ensure that the key features in each module are retained.

[0104] The image presented by the module after final registration can be further improved in resolution and detail through super-resolution technology, allowing subsequent detection algorithms to more accurately identify tiny defects or irregular structures on the insulating terminals.

[0105] Processing step 2, mathematical model of optical flow estimation and dense registration scene

[0106] Assume that for two module images I vis and I uv (visible light and visible blind imaging, respectively), their pixel values do not match exactly (i.e., the surface response is different under different spectra), we need to find the dense registration optical flow field u(x, y), v(x, y) describing the vis to I uv Pixel displacement.

[0107] The goal of optical flow estimation is to minimize the feature difference between the two modules:

[0108]

[0109] Among them F vis and F uv are feature maps extracted from visible and visible-blind images.

[0110] Processing step 3, feature alignment loss

[0111] For multimodal imaging, the feature alignment loss is defined as:

[0112]

[0113] By minimizing this loss, the features between different modules are guaranteed to be optimally aligned in the high-dimensional feature space.

[0114] The working process of the present invention is: first, a visible blind imaging module, a visible light imaging module and a yellow light imaging module are set at the monitoring end and aligned through the alignment steps recorded in the application document. The imaging of the visible blind imaging module, the visible light imaging module and the yellow light imaging module is subjected to fault identification through the identification method recorded in the application document, and the communication unit 2 transmits the image information to the equipment maintenance personnel through the transceiver module.

[0115] In summary, the present invention provides a substation insulation fault monitoring device and its monitoring method. The device uses a visible blind imaging module, a visible light imaging module, and a yellow light imaging module to identify light of different wavelengths emitted by substation equipment due to insulation faults. The processing module judges the information identified by the visible blind imaging module, the visible light imaging module, and the yellow light imaging module and generates an accident information result. The communication unit 2 transmits the accident information judged by the processing module to the background terminal, enabling the monitoring device to identify the equipment characteristics of substation equipment in various stages of faults, thereby solving the problem that it is difficult for imaging devices with a single wavelength band in the prior art to timely monitor discharge fault accidents of substation equipment.

[0116] It should be understood that in the present invention, terms such as "first" and "second" are used to describe various information, but these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, "first" information can also be referred to as "second" information, and similarly, "second" information can also be referred to as "first" information. In addition, the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0117] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art in the technical field, without departing from the technical principle of the present invention, several improvements and substitutions can be made, and these improvements and substitutions should also be regarded as the protection scope of the present invention.

Claims

1. A substation insulation fault monitoring device, characterized in that, include: The monitoring device body and the backend terminal, the monitoring device body includes: Install the housing; An imaging unit, the imaging unit being installed in the installation housing, the imaging unit comprising a visible blind imaging module, a visible light imaging module, and a yellow light imaging module. The visible blind imaging module, the visible light imaging module, and the yellow light imaging module can transmit light of different wavelengths and respectively image the insulated terminals of the substation equipment, respectively collecting visible blind image signals, monitoring equipment image signals, and yellow light image signals. The visible blind image signals are collected by the visible blind imaging module, the monitoring equipment image signals are collected by the visible light imaging module, and the yellow light image signals are collected by the yellow light imaging module. a processing module, the processing module being installed in the installation housing and being communicatively connected to the visible blind imaging module, the visible light imaging module, and the yellow light imaging module; the processing module determining whether it is an early-stage discharge accident or a mid-stage discharge accident based on the visible blind image signal, the monitoring device image signal, and the yellow light image signal collected by the imaging unit; and determining whether it is an early-stage discharge accident when the image area collected by the visible blind image signal is greater than 50% of the monitoring portion area collected by the monitoring device image signal, and determining whether it is a mid-stage discharge accident when the image area collected by the yellow light image signal is greater than 50% of the monitoring portion area collected by the monitoring device image signal; A communication unit is configured to send the accident information determined by the processing module to a background terminal.

2. The substation insulation fault monitoring device according to claim 1, characterized in that, The backend terminal is further configured to: When it is determined to be an early discharge accident, the first warning information is issued; When it is determined to be a mid-term discharge accident, a second warning message is issued.

3. The substation insulation fault monitoring device according to claim 1, characterized in that, The processing module determines whether the substation equipment has an early discharge accident or a mid-term discharge accident, specifically including: Insulated terminal contour recognition uses an optical frequency feature decoding network to fuse and decode image information collected by the visible blind imaging module, visible light imaging module, and yellow light imaging module to ensure that the shape of the insulated terminal can be accurately identified under different spectra; Semantic segmentation of insulated terminal images: After the insulated terminal contour recognition step is completed, the DeepLabV3 model is used to segment the image information of the insulated terminal into precise functional areas; For statistics on a single insulated terminal, after dividing the functional areas of the insulated terminal, the boundaries of each functional area are further adjusted through an adaptive edge expansion algorithm. The proportion of imaging of the visible blind module or imaging of the yellow light module in each functional area is counted and compared with the set boundary value of 50%.

4. The substation insulation fault monitoring device according to claim 1, characterized in that, The monitoring device body also includes a carrying platform and universal wheels; The bearing platform is used to install the installation shell; The universal wheel is arranged on the bottom end surface of the carrying platform, and the universal wheel is used for moving the carrying platform.

5. The substation insulation fault monitoring device according to claim 1, characterized in that A solar panel is further provided on the carrying platform, and a power output end of the solar panel is electrically connected to the processing module.

6. The substation insulation fault monitoring device according to claim 1, characterized in that, A push rod is also provided on the bearing platform, a fixed end of the push rod is fixedly connected to the bearing platform, and a gripping end of the push rod is provided with a rubber pad.

7. The substation insulation fault monitoring device according to claim 1, characterized in that, The optical axes of the visible blind imaging module, the visible light imaging module and the yellow light imaging module are parallel.

8. The substation insulation fault monitoring device according to claim 1, characterized in that, A first filter is provided in the visible-blind imaging module, and the light wave band that can pass through the first filter is 300 - 400 nm; A second filter is provided in the visible light imaging module, and the light wave band that can pass through the second filter is 400 - 760 nm; A third filter is provided in the yellow light imaging module, and the light wave band that can pass through the third filter is 580 - 597 nm.

9. A substation insulation fault monitoring method, characterized in that, Including the substation insulation fault monitoring device according to any one of claims 1 - 8, the monitoring method includes: Respectively image the insulating terminal through the visible-blind imaging module, the visible light imaging module and the yellow light imaging module, and respectively collect the visible-blind image signal, the monitoring device image signal and the yellow light image signal; When the image area collected by the visible-blind image signal > 50% of the monitoring area collected by the monitoring device image signal, it is determined as an early discharge accident. When the image area collected by the yellow light image signal > 50% of the monitoring area collected by the monitoring device image signal, it is determined as a mid-term discharge accident.

10. The substation insulation fault monitoring method according to claim 9, characterized in that, The monitoring method further includes the registration steps of the visible-blind imaging module, the visible light imaging module and the yellow light imaging module; Feature extraction stage optimization, using multi-scale convolutional feature extraction to optimize different imaging features extracted by different modules; Optical flow estimation and dense registration, using PWC-Net for multi-modal optical flow estimation, stratifying the images collected by different modules through a feature pyramid, and introducing a self-supervised learning mechanism in the optical flow estimation process; Detail optimization based on geometric transformation, after the optical flow estimation, introducing a model that combines affine transformation and perspective transformation, further optimizing the image details by calculating the dense correspondence of different pixels, and adding the edge features in the corresponding module images as constraint conditions in the processing of the insulating material edge during the registration process; Network loss function design, constructing a loss function including optical flow loss, smoothness loss, loss function based on structural similarity and feature alignment loss function, and the feature alignment loss function measures the image feature alignment situation under different modules through the cosine similarity in the convolutional feature space.

Images