A method for power asset lifecycle management based on a cloud platform

Abstract

The present invention provides a cloud platform-based power asset lifecycle management method, which relates to the field of image recognition and is particularly suitable for the monitoring and management of early risks such as dynamic arcs and corona of power equipment. The cloud platform-based power asset lifecycle management method of the present invention specifically includes: collecting multi-source image information of power asset equipment; identifying multi-target positions in the power asset equipment based on the multi-source image information; identifying corona and arcs occurring in the power asset equipment based on the multi-source image information and the multi-target positions; performing lifecycle assessment on the power asset equipment based on the corona and arc identification results; through the above steps, the present invention realizes the efficient detection of hidden risks of power assets and realizes the safety management and maintenance of power assets.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and particularly to a method for managing the life cycle of power assets based on a cloud platform, which is particularly suitable for monitoring and managing early risks such as dynamic arcs and corona of power equipment. Background Art

[0002] In the operation of modern society, electricity, as one of the most fundamental forms of energy, directly determines the normal operation of social production and life. The large-scale operation of the power system relies on a huge number of power equipment, and the operating states of these equipment are directly related to the stability and security of the power grid. Therefore, the management of power equipment is not only a necessary means to ensure power supply but also an important part of modern energy management. However, with the complexity of the power system and the expansion of equipment scale, traditional power equipment management methods have gradually revealed problems such as low efficiency and lagging response, making it difficult to meet the high-efficiency operation requirements of modern power systems.

[0003] In the prior art, the life cycle management method of power equipment mainly relies on manual inspection. Manual inspection usually requires maintenance personnel to regularly check power equipment and record the operating state of the equipment and changes in the external environment. However, this method has obvious limitations. First, the efficiency of manual inspection is low, and it is impossible to achieve real-time monitoring of a large number of power equipment. Second, due to the limited inspection frequency, many early abnormal signals during equipment operation are difficult to detect in a timely manner. Although these signals may have little impact on the equipment in the initial stage, they are often precursors of potential equipment failures. If not processed in a timely manner, it may lead to a gradual deterioration of equipment performance and ultimately cause major failures. In addition, for the life cycle management of power assets, existing methods often lack systematic prediction and analysis means. Usually, only when obvious failures occur in the equipment will repair or even replacement measures be taken. This lag not only increases the maintenance cost but also may cause power supply interruption, seriously threatening the safety and stability of the power grid.

[0004] Especially for early power equipment risks such as dynamic arcs and corona, existing management methods are almost powerless. Arcs and corona are common abnormal phenomena in power equipment under high-voltage operating conditions, usually manifested as discharge phenomena on the surface or joints of power equipment. An arc is accompanied by intense light, heat, and noise, while corona is characterized by weak ultraviolet light and ozone generation. Although these phenomena may not cause direct damage to the equipment in the initial stage, they are early signals of problems such as equipment aging and insulation failure. If they cannot be effectively monitored and managed, it may lead to further deterioration of the equipment and increase the instability of the power system. Therefore, the dynamic monitoring of arcs and corona is a major technical difficulty in power equipment management. The current manual inspection method is difficult to capture these short-lived and weak abnormal signals, lacking technical means to identify and evaluate the early risks of equipment, resulting in the management and maintenance of power assets lagging behind the actual needs.

[0005] To address the above problems, the present invention proposes a cloud-platform-based power asset lifecycle management method. This method uses machine vision to detect arcs and corona in power equipment in real time, combines the big data processing capabilities of the cloud platform, comprehensively analyzes the operating status of the equipment, predicts the lifecycle of the equipment, and provides optimized maintenance suggestions. Summary of the Invention

[0006] The present invention provides a cloud-platform-based power asset lifecycle management method, which specifically includes the following steps:

[0007] S1: Collect multi-source image information of power asset equipment;

[0008] S2: Identify multi-target positions in power asset equipment based on the multi-source image information;

[0009] S3: Identify corona and arcs occurring in power asset equipment based on the multi-source image information and the multi-target positions;

[0010] S4: Conduct lifecycle assessment of the power asset equipment based on the identification results of corona and arcs.

[0011] The present invention also provides a cloud-platform-based power asset lifecycle management system, which includes:

[0012] Image acquisition device: The image acquisition device collects multi-source image information of power asset equipment;

[0013] Target recognition module: The target recognition module identifies multi-target positions in power asset equipment based on the multi-source image information;

[0014] Corona and arc recognition module: The corona and arc recognition module recognizes corona and arc occurring in power asset equipment based on multi-source image information and the multi-target positions;

[0015] Life cycle assessment module: The life cycle assessment module performs a life cycle assessment on the power asset equipment based on the corona and arc identification results.

[0016] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned cloud platform-based power asset lifecycle management method when executing the computer program.

[0017] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements the above-mentioned cloud platform-based power asset lifecycle management method.

[0018] Compared with the prior art, the present invention proposes a power asset lifecycle management method based on a cloud platform. The present invention obtains the location information of corona-prone locations in high-voltage switchgear based on visible light images, and effectively obtains the corona occurrence location by combining location priors and noise filtering. At the same time, visible light video information and ultraviolet image information are combined in arc detection. According to the arc development characteristics and the physical characteristics when it occurs, a multi-scale feature extraction network and an ultraviolet high-frequency information extraction module are designed to improve the accuracy of arc detection, and the life cycle of the equipment is evaluated based on the occurrence of corona and arc, thereby realizing efficient detection of hidden risks of power assets and realizing safe management and maintenance of power assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 It is a flow chart of the present invention for performing cloud platform-based power asset lifecycle management. DETAILED DESCRIPTION

[0021] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0022] The following describes the implementation manners of the present application through specific specific examples. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope protected by the present application.

[0023] It should be noted that the following describes various aspects of the embodiments within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on the present application, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects described herein can be used to implement the device and / or practice the method. Additionally, this device and / or practice of this method can be implemented using other structures and / or functionality in addition to one or more of the aspects described herein.

[0024] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the examples can be practiced without these specific details.

[0025] The embodiments of this specification propose a method for power asset lifecycle management based on a cloud platform. The method specifically includes the following steps:

[0026] S1: Collect multi-source image information of power asset equipment;

[0027] S2: Identify multiple target positions in the power asset equipment based on the multi-source image information;

[0028] S3: Identify corona and arc occurring in the power asset equipment based on the multi-source image information and the multiple target positions;

[0029] S4: Perform lifecycle assessment on the power asset equipment based on the identification results of corona and arc.

[0030] Among them, the power asset equipment is specifically a high-voltage switchgear, and the multi-source image information includes visible light video and ultraviolet light image.

[0031] In the process of collecting specific multi-source image information, the visible light industrial camera preferably uses a high-speed industrial camera with a resolution of 1920×1080 pixels (1080P) or above and a frame rate of not less than 60 fps. The lens field of view is set between 90° and 120° to achieve a larger coverage range, and is equipped with an automatic aperture adjustment function to adapt to different lighting conditions in the switch cabinet. The ultraviolet imaging module preferably uses an ultraviolet sensor with a wavelength response range of 240nm to 280nm, and the sensitivity requirement is a minimum detectable ultraviolet signal power of 0.01 μW / cm², which can capture weak ultraviolet signals released by corona under strong ambient light conditions. The ultraviolet imaging module is equipped with a narrow-band filter to improve the response accuracy to the target ultraviolet signal.

[0032] In the process of selecting specific acquisition equipment, arc discharge is usually accompanied by strong light and heat, and its duration is short (milliseconds), but the light intensity is high. Therefore, the high frame rate and high dynamic range of the above-mentioned high-speed industrial camera can effectively record the rapid occurrence and disappearance of the arc, and capture its light intensity changes, arc length and expansion trend. In order to achieve accurate acquisition of arc transient discharge in high-voltage switchgear, for example, for shooting arc transient discharge, the high-speed industrial camera used can be PhotronFASTCAM Mini UX100 or Vision Research Phantom VEO 710. The former supports up to 4000 fps (resolution 1280×1024), while the latter supports 7000 fps (resolution 1280×800), which can capture the rapid optical dynamic changes of arc transient discharge and meet the needs of capturing arc strong light, arc light movement trajectory and duration.

[0033] The corona phenomenon is characterized by weak light intensity and uneven distribution, which is usually continuously released in the form of weak ultraviolet light on the surface of the insulator or the busbar joint area. Due to the weak characteristics of the corona signal, the equipment needs to have high sensitivity and narrow-band filtering function (ultraviolet bandwidth 2 nm) to accurately capture weak ultraviolet light and eliminate visible light interference. For the collection of corona phenomena, OFIL DayCor Superb or Corocam 8 UAV are used as examples. These two devices have high sensitivity and can detect ultraviolet light signals with a wavelength range of 240 nm to 280 nm. They also have good background noise suppression capabilities and are particularly suitable for corona monitoring under strong ambient light conditions.

[0034] The deployment location of the acquisition device is optimized based on the internal structural characteristics of the high-voltage switchgear and the occurrence law of discharge phenomena. The high-voltage switchgear consists of a busbar, a circuit breaker, and terminal connectors, etc. Among them, arc discharge and corona discharge are mostly concentrated in the busbar connection points, circuit breaker joints, the surface of insulators, and the terminal connector areas. To achieve full coverage, the device is preferably installed at the top and sidewall positions inside the switchgear. The visible light industrial camera is installed at the center of the switchgear top, tilted downward by 30° to 45°, and the lens is aimed at the busbar connection points and circuit breaker joints to ensure that the key areas inside the switchgear are within the field of view. The ultraviolet imaging module is preferably installed on the sidewall of the switchgear, and the lens angle is adjusted to be perpendicular or slightly tilted to the target area to capture the clearest ultraviolet light signal. To avoid device occlusion or signal interference, the installation of the acquisition device needs to fully consider the spatial layout of the internal devices of the switchgear, and at the same time, the housing of the acquisition device is insulated and protected to ensure its stable operation in a high electromagnetic interference and high-voltage environment.

[0035] To solve the possible problem of dark environment inside the high-voltage switchgear, this solution is equipped with a low-interference LED lighting device. The lighting device uses a wide-spectrum LED light source, preferably with a wavelength range of 400nm to 700nm (visible light range) to avoid ultraviolet spectrum interference with the detection of the ultraviolet imaging module. The LED lamp preferably uses a light source with a high color rendering index (CRI≥90) to ensure the uniformity and naturalness of the lighting and avoid the generation of strong light spots or shadows. The lighting device is installed on the top or sidewall of the switchgear and can be integrated with the visible light industrial camera so that the lighting direction is consistent with the camera lens viewing angle to achieve the best lighting effect. To further reduce the interference with ultraviolet detection, the LED lighting device can operate in a controllable pulse mode, only lighting up briefly when the visible light industrial camera captures video and turning off when the ultraviolet imaging module captures images. In addition, the light intensity of the lighting device is adjustable to meet the requirements of different internal environment brightness of the switchgear.

[0036] The acquisition device realizes the synchronous acquisition of visible light video and ultraviolet light image through a preset acquisition program. During the acquisition process, the visible light industrial camera continuously records video at a preset frame rate to capture the optical characteristics of the dynamic arc inside the high-voltage switch cabinet in real time; the ultraviolet imaging module acquires static images in an interval shooting manner, which is mainly used to detect the distribution range and intensity of ultraviolet light of corona discharge. To ensure the synchronization of data, a unified timestamp can be set in the acquisition system to mark the time information of each frame of visible light video and each ultraviolet light image. The acquisition device transmits the collected data to the edge computing device through industrial Ethernet or wireless communication module for preliminary analysis. The edge computing device preferably uses a high-performance embedded computing platform (such as NVIDIA Jetson Xavier or Intel Movidius) to process the collected image data in real time, including denoising, enhancement and other operations, and the processed data is packaged and uploaded to the cloud platform for storage and analysis.

[0037] The data storage module of the cloud platform adopts a distributed database architecture. The data uploaded to the cloud platform is first authenticated and encrypted to ensure the security of data transmission. The cloud platform assigns a unique device number to each high-voltage switchgear and stores the collected data in the corresponding folder structure in chronological order. The storage module of the cloud platform automatically classifies and indexes the visible light video and ultraviolet image data to facilitate subsequent rapid retrieval and analysis. At the same time, the cloud platform will perform quality inspection on the uploaded data to eliminate invalid data caused by abnormal acquisition equipment or transmission errors. In order to save storage space, the visible light video can be compressed and the ultraviolet image can be stored in a high-quality lossless format to ensure the accuracy of subsequent analysis. The cloud platform is also equipped with data backup and disaster recovery functions, which regularly copies the collected data to a remote server to prevent data loss due to system failures.

[0038] Identify multiple target locations in power asset equipment based on multi-source image information, including identifying terminal blocks, busbar connection points, and insulators based on visible light video key frame images; key frame image recognition is implemented using the Yolo series network;

[0039] Since the captured high-voltage switchgear is a static target, the main content of the video will not change. Therefore, the selection of key frames only needs to consider the quality of the image frame, which has no direct connection with the acquisition time. Therefore, the visible light video key frame can select the first frame of the video as the key frame or the video frame captured at a preset time point as the key frame. This application does not specifically limit the selection of key frames, and any frame in the visible light video can be used as a key frame.

[0040] Before the input is recognized by the Yolo series network, first, the key frames are preprocessed. The key frames are normalized by the preprocessing module, and the pixel values are normalized to the range of [0, 1].

[0041] The feature extraction module of the Yolo series network adopted in the present invention uses the improved CSPDarknet53 as the backbone network. The improved CSPDarknet53 extracts multi-scale features through cascaded convolutional layers and residual blocks, and can capture the image characteristics of terminal blocks, bus connection points, and insulators at different scales. For terminal blocks, their image characteristics are mainly manifested as regular metal structures, clear edges, and small sizes. Therefore, 3×3 and 5×5 convolutional kernels are added to the shallow features to enhance the ability to extract edge features; for bus connection points, their image characteristics are obvious geometric shapes and usually accompanied by insulators around. Therefore, local geometric information is extracted by combining shallow and middle-level features; for insulators, their image characteristics are strong symmetry, complex texture, and single color. Therefore, the repeated texture patterns and symmetry are captured through the deep feature extraction module.

[0042] The specific feature extraction network structure is defined as follows:

[0043] The characteristics of the terminal block are regular metal structure, clear edges, and small sizes. The feature extraction mainly relies on shallow features, and 3×3 and 5×5 convolutional kernels are used to extract edge information. The convolutional operation is described as:

[0044] ;

[0045] and respectively represent the convolutional calculations of 3×3 and 5×5, which are used to extract high-frequency edge features and local features in a larger range. The output enhances the expression ability of the edge features of the terminal block in the image, is the key frame after normalization;

[0046] The characteristics of the bus connection point are obvious geometric shapes and usually accompanied by insulators around. It is necessary to combine shallow and middle-level features to extract geometric information. The middle-level feature extraction is realized through residual blocks:

[0047] ;

[0048] ;

[0049] represents being processed by two residual modules. The output combines the detailed information of the shallow features and the geometric information of the middle-level features through the residual blocks, and can better express the position and shape of the bus;

[0050] The image features of insulators are strong symmetry, complex texture, and single color. It mainly relies on the deep feature extraction module to capture its repetitive texture patterns and symmetry. Deep feature extraction is achieved through further convolution and pooling operations:

[0051] ;

[0052] Among them, represents a 7×7 convolution kernel, which is used to capture large-scale texture patterns and symmetry, The max pooling operation is used to reduce the size of the feature map and enhance the extraction of global information, and the output feature extracts the expression suitable for the global symmetry and repetitive texture patterns of insulators, and can capture the overall structural characteristics of insulators; the extracted 、 and are input into the neck network of the Yolo series network for feature fusion, and after fusion, they are input into the prediction head for target classification and localization;

[0053] In the target classification and localization module, the prediction head of the YOLO series detection framework is used to output the category and position coordinates of the target. The output of the prediction head includes the category probability and the bounding box parameters. The category probability is calculated by the Softmax function, and the bounding box parameters include the center point coordinates (x, y) and the width and height (w, h). A joint loss function including category loss, bounding box loss, and confidence loss is used to optimize the model. The bounding box loss is calculated using the CIoU loss (Complete IoU).

[0054] Based on the multi-source image information and the multi-target positions, corona and arc occurrences in power asset equipment are identified. Among them, the corona identification process specifically includes:

[0055] S3-1-1: Register the visible light key frame with the ultraviolet light image;

[0056] S3-1-2: Generate a target weight matrix according to the multi-target positions identified in step S2;

[0057] S3-1-3: Combine the ultraviolet light image and the target weight matrix for corona detection.

[0058] Among them, the process of generating the target weight matrix includes:

[0059] For the j-th target region in the ultraviolet light image, with its center point generate the corresponding Gaussian weight matrix :

[0060] ;

[0061] ;

[0062] ;

[0063] Among them, and are proportionality coefficients, which are the reference values for controlling the expansion range; and are the width and height of the j-th target area; is the shape correction factor, which controls the influence of the aspect ratio on the expansion range, , where exp is the exponential function, and represent the variances in the x and y directions respectively;

[0064] When the target area is very wide, the horizontal expansion range significantly increases, while the vertical expansion range is restricted by the aspect ratio and remains small.

[0065] When the target area is very high, the vertical expansion range will significantly increase, while the horizontal expansion range, remains small.

[0066] The expansion range formulas in the horizontal and vertical directions are symmetric with each other, which can balance the width and height characteristics of the target area and adapt to target areas of various shapes. This balance ensures that the Gaussian distribution can cover the key characteristics of the target area without unnecessary over-expansion.

[0067] The Gaussian distribution weight functions of each target area are superimposed to generate the global weight matrix ;

[0068] ;

[0069] α is the global scaling coefficient, which is used to adjust the overall strength of the weight matrix, is the minimum weight value of the non-target area, which ensures that the non-target area will not be completely ignored. M is the total number of target areas.

[0070] The corona detection by combining the ultraviolet image and the target weight matrix specifically includes:

[0071] Calculating the average brightness , the brightness range and the brightness standard deviation of the same pixel of multiple collected ultraviolet images :

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] and The brightest and darkest pixel values of all frames for each pixel are extracted respectively, which is the total number of ultraviolet light images;

[0078] Using the brightness characteristic map, low-brightness noise and fixed noise are filtered out to extract the candidate corona region ,

[0079] ;

[0080] ；

[0081] ;

[0082] ;

[0083] wherein, 、 、 、 are all threshold parameters, , are the filtering masks respectively, is the initial candidate corona region;

[0084] Corona and noise show significant differences in ultraviolet images, mainly reflected in aspects such as the dynamic range of brightness, spatial distribution characteristics, and temporal stability. Corona is usually a discharge phenomenon caused by overly strong local electric fields on the surface of high-voltage equipment. Its bright spots appear as high-brightness regions in the image and exhibit certain dynamic fluctuations. This kind of fluctuation includes changes in brightness and intermittency in time, that is, the corona bright spots may disappear or weaken in some frames, but in multiple frames of images, their positions are generally relatively stable. In addition, corona is usually concentrated in specific high-risk areas (such as terminal blocks, busbar connection points, insulators, etc.), and has certain brightness attenuation characteristics: high brightness in the center and gradually decreasing brightness at the edges, showing a diffuse spatial distribution characteristic. In contrast, noise can be divided into two types: fixed noise and random noise. The brightness of fixed noise is highly stable in multiple frames of images, with almost no change. A typical example is sensor defects or background light interference. Random noise is usually caused by the external environment (such as thermal noise or accidental light interference). Its brightness and position change irregularly between frames, usually appearing in isolation, without the spatial concentration and persistence of corona. Random noise may suddenly increase in brightness in a certain frame, but its overall fluctuation is small and it will not persist. Based on these characteristics, the present invention constructs an effective filtering logic to distinguish corona bright spots and noise. First, we define the candidate regions of corona bright spots as those pixels with high brightness and significant fluctuations in multiple frames of images. By calculating the range and standard deviation of the brightness of each pixel, the dynamic change characteristics of the bright spots can be captured. The range is used to measure the maximum amplitude of brightness change, while the standard deviation is used to quantify the overall fluctuation of brightness. To avoid false detection, both conditions of a large brightness range and a high standard deviation must be met to determine a corona bright spot; if only one condition is met, it may be noise. For example, the brightness of fixed noise hardly changes in multiple frames, and its range and standard deviation are both small, so it can be eliminated; while the brightness of random noise occasionally increases in a certain frame, which may lead to a large range, but because it changes little in other frames, the standard deviation is usually low, and thus it can also be filtered. Finally, we combine the global weight matrix, assign higher weights to key regions (such as terminal blocks), making them more likely to be retained in the judgment, while non-key regions require more stringent filtering conditions. This logic based on brightness dynamics, temporal stability, and spatial distribution characteristics can effectively distinguish corona and noise, minimizing the risk of false detection and missed detection.

[0085] Based on multi-source image information and the multi-target positions, identify the corona and arcs that appear in the power asset equipment, wherein the arc identification is realized by using a dual-channel identification network;

[0086] The dual-channel identification network is divided into a visible light video channel and an ultraviolet light image channel;

[0087] The visible light video channel extracts the collected video images, and performs differential calculation on the video images to extract the arc image sequence :

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] Among them, is the inter-frame differential intensity, , are the grayscale images of the t-th and (t - 1)-th frames of the visible light video, H and W are the height and width scales of the image, is the arc index set;

[0093] According to the arc image sequence, the ultraviolet light image collected during the arc occurrence is extracted as the key frame. When the corresponding ultraviolet light image is not collected during the arc occurrence period, the last ultraviolet light image collected before the arc occurrence is extracted as the key frame;

[0094] The two-channel recognition network is divided into three stages of feature extraction. Among them, the feature extraction modules in the first stage and the third stage are the same:

[0095] ;

[0096] ;

[0097] ;

[0098] ;

[0099] Among them, and respectively represent the feature maps of the first stage or the third stage of the visible light channel and the ultraviolet light channel, and correspond to the feature maps of the second stage of the visible light channel and the ultraviolet light channel, is the concatenation by channel, is the input feature map of the first stage or the third stage of the visible light channel. Among them, the input feature map of the first stage is the differential image of the arc image sequence, and the input feature map of the third stage is , is the input feature map for the first or third stage of the ultraviolet light channel. Among them, the input feature map of the first stage is the weighted fusion of the ultraviolet light key frame and the candidate corona region, and the input feature map of the third stage is ;

[0100] The dual-channel recognition network also includes three feature fusion stages for fusing the visible light feature map and the ultraviolet light feature map obtained by three-stage feature extraction:

[0101] Enhance the attention of the visible light feature map before fusion:

[0102] In arc detection, the visible light video channel extracts multi-scale information in the image because the visible light manifestations of arcs may have different spatial scales and morphological characteristics, and these characteristics will show significant differences in actual scenarios due to distance, ambient light, and the intensity and development stage of the arc. An arc in a visible light video may appear as a small and concentrated bright spot to a larger diffuse halo, and may even be accompanied by complex dynamic changes, such as bright and dark flickering or diffusion. Therefore, single-scale information extraction may not be able to completely capture these features. For example, small-scale features help detect the initial concentration point of the arc, while large-scale features can capture the overall halo distribution or occlusion phenomenon when the arc spreads. In addition, the visible light video is also affected by ambient light interference. For example, other bright spots in the background (such as reflections or lights) may be visually similar to the arc at a certain scale, but multi-scale analysis can more comprehensively capture the significant changes of the arc at different spatial scales, thereby enhancing the robustness to background interference.

[0103] ;

[0104] At each scale s, capture the context relationship in the feature sequence through the multi-head self-attention mechanism:

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] will The outputs of the attention heads are concatenated, and the feature dimension is restored through a linear mapping:

[0113] ;

[0114] ;

[0115] For each scale 𝑠, the features of other scales 𝑠′ are used as Key and Value for cross-scale attention:

[0116] ;

[0117] ;

[0118] By concatenating and fusing the features of all scales, the visible light features to be fused at the k-th stage are obtained ;

[0119] ;

[0120] PatchEmbed(⋅) is a patch embedding function, which obtains the feature map for feature extraction at the k-th stage, where k = 1, 2, 3; represents the feature map at the s-th resolution, is the multi-scale feature sequence, with a total of ones, , and are the projection matrices used to calculate the values of Query, Key, and Value respectively; is the dimension of each attention head, is the output of the h-th attention head, where h = 1, 2, ……, ; and represent the normalization layer and the fully connected layer respectively;

[0121] High-frequency information enhancement is performed on the ultraviolet light feature map before fusion:

[0122] ;

[0123] ;

[0124] The enhanced visible light feature map and the ultraviolet light feature map are fused:

[0125] ;

[0126] In arc detection, the ultraviolet (UV) image channel extracts high-frequency information because the occurrence of an arc is usually accompanied by obvious spatial texture changes and local intensity mutations, and these features are mainly concentrated in the high-frequency part of the image. High-voltage arc discharge is a localized and high-intensity phenomenon, which appears as a spot or high-brightness area with significant edges in the UV band, showing an obvious contrast difference from the background UV radiation. This difference usually exists in the image in the form of high-frequency components. By extracting this high-frequency information, the edges and detailed features of the arc area can be more effectively highlighted, while suppressing the low-frequency components in the background (such as uniformly distributed ambient UV noise or slowly changing radiation on the surface of the device). Extracting high-frequency information can effectively filter out these irrelevant or weakly related low-frequency components, focusing attention on the significant features of arc discharge, thereby improving the accuracy and robustness of arc detection. In this way, the UV image channel can not only capture the physical characteristics of the arc but also distinguish it from the background environment.

[0127] The three fused scales , and are input into the Head detection head of Yolov8 to obtain the position area information of the arc.

[0128] In the arc detection of high-voltage switchgear, this application combines the dual-modal information of visible light video sequence frames and UV images. The visible light video sequence frames can intuitively capture the bright spot or spark phenomenon when the arc occurs, which is the direct visual manifestation of arc discharge. Visible light imaging is easy to implement, with clear imaging, and its video sequence can provide continuity in the time dimension, helping to analyze the dynamic behavior of the arc. On the other hand, the UV image provides information related to the arc in the UV band, especially the UV radiation characteristics when the arc occurs, which are very important physical features in arc detection. High-voltage arcs are usually accompanied by strong UV radiation, and UV imaging can effectively capture this feature while avoiding environmental light interference in the visible light band. Therefore, the UV image has a certain degree of robustness under complex lighting conditions. In addition, in the early stage before the arc occurs, UV imaging can also capture precursor signals such as corona discharge, which are the precursor phenomena of arc discharge. Fusing the dual-channel information of visible light and UV images can give full play to the complementary advantages of the two modalities. The UV image can provide the physical characteristics of arc discharge, while the visible light video can intuitively present the dynamic behavior and visual features of the arc. By fusing the information of the two modalities, a more comprehensive arc detection model can be established in the spatial and time dimensions and verified with each other. Furthermore, it helps to reduce the false detection rate.

[0129] The life cycle assessment of the power asset equipment based on the corona and arc recognition results specifically includes:

[0130] S4-1: Statistically analyze the area information of the corona occurrence location and the area information of the arc occurrence location respectively;

[0131] S4-2: Calculate the discreteness of the corona occurrence distribution and the discreteness of the arc occurrence distribution;

[0132] S4-3: Determine whether there is an overlap between the corona occurrence location area and the arc occurrence location area;

[0133] S4-4: Conduct a life cycle assessment based on the number of corona and arc occurrences, the discreteness of the distribution, and the number of overlaps;

[0134] When the discreteness of the corona occurrence distribution is greater than the preset threshold, it indicates that the corona phenomenon is widely distributed and there may be insulation deterioration in multiple places in the equipment. Arcs are usually more concentrated than coronas. When the discreteness of the arc occurrence distribution is greater than the preset threshold, it indicates that there is a serious spread of local faults in the equipment. When the number of overlaps between the corona occurrence location area and the arc occurrence location area is greater than the threshold, it indicates that the insulation deterioration of the equipment has developed to a serious level.

[0135] The present invention also provides a cloud platform-based power asset life cycle management system, which includes:

[0136] Image acquisition device: The image acquisition device acquires multi-source image information of power asset equipment;

[0137] Target recognition module: The target recognition module recognizes multiple target positions in the power asset equipment based on the multi-source image information;

[0138] Corona and arc recognition module: The corona and arc recognition module recognizes coronas and arcs that appear in the power asset equipment based on the multi-source image information and the multiple target positions;

[0139] Life cycle assessment module: The life cycle assessment module conducts a life cycle assessment on the power asset equipment based on the corona and arc recognition results.

[0140] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned cloud platform-based power asset life cycle management method.

[0141] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned cloud platform-based power asset life cycle management method.

[0142] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0143] In this specification, the same or similar parts among the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the embodiments described later, the description is relatively simple, and the relevant parts can be referred to the partial description of the foregoing embodiments.

[0144] As described above, the above are only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A power asset lifecycle management method based on a cloud platform, characterized in that: The method comprises the following steps: S1: Collect multi-source image information of power asset equipment; S2: Identify multiple target locations in power asset equipment based on multi-source image information; the identification of multiple target locations is based on visible light video key frame image recognition; S3: Identify corona and arc occurring in power asset equipment based on multi-source image information and the multi-target positions; S4: Performing a life cycle assessment on the power asset equipment based on the corona and arc identification results; The corona identification process specifically includes: S3-1-1: registering the visible light key frame with the ultraviolet light image; S3-1-2: Generate a target weight matrix according to the multi-target positions identified in step S2; The Gaussian distribution weight functions of each target area are superimposed to generate a global weight matrix W(x, y) and calculate the collected multiple UV images Calculate the average brightness of the same pixel Brightness difference ΔI ult (x, y) and brightness standard deviation Extract candidate corona region M final (x, y), M candidate (x,y)=(M1(x,y)∩M2(x,y))·W(x,y); Among them, T avg 、T diff1 、T diff2 、T final are threshold parameters, M1(x, y), M2(x, y) are filter masks, M candidate (x, y) is the initial candidate corona region; Arc recognition is realized by using a dual-channel recognition network. The dual-channel recognition network is divided into three stages of feature extraction. The first stage input feature maps are the difference image of the arc image sequence and the ultraviolet key frame and the candidate corona region M. final Weighted fusion of (x, y).

2. According to the cloud platform-based power asset lifecycle management method of claim 1, it is characterized in that: The identification of multiple target positions specifically includes identifying wiring terminals, busbar connection points, and insulators based on visible light video key frame images.

3. The power asset lifecycle management method based on a cloud platform according to claim 1 is characterized in that: For the jth target area in the UV image, its center point (x j ,y j ) Generate the corresponding Gaussian weight matrix W j (x, y): Among them, k x , k y is the proportionality coefficient, w j 、h j is the width and height of the jth target area; β is the shape correction factor, exp is the exponential function, and Represent the variance in the x and y directions respectively; The Gaussian distribution weight functions of each target area are superimposed to generate a global weight matrix W(x, y); α is the global scaling factor, W min is the minimum weight value of non-target areas, and M is the total number of target areas.

4. The cloud platform-based power asset lifecycle management method according to claim 3 is characterized in that: The feature extraction modules of the first and third stages are the same: in, and Represent the characteristic graphs of the first or third stage of the visible light channel and the ultraviolet light channel, and Corresponding to the second stage feature map of the visible light channel and ultraviolet light channel, is the input feature map of the first or third stage of the visible light channel, where the first stage input feature map is the differential image of the arc image sequence, and the third stage input feature map is is the input feature map of the first or third stage of the UV channel, where the first stage input feature map is the weighted fusion of the UV keyframe and the candidate corona region, and the third stage input feature map is f concat For channel-by-channel splicing, f avgpool and f maxpool are average pooling and maximum pooling respectively, ReLU is the activation function, f conv3 Represents a 3×3 convolution calculation.

5. The method for managing the lifecycle of power assets based on a cloud platform according to claim 4, characterized in that: The dual-channel recognition network also includes three feature fusion stages, which are used to fuse the visible light feature map and the ultraviolet light feature map obtained by the three-stage feature extraction.

6. A cloud-based power asset lifecycle management system, characterized in that: The system is used to execute a cloud platform-based power asset lifecycle management method as described in any one of claims 1 to 5, and the system includes: Image acquisition device: the image acquisition device acquires multi-source image information of power asset equipment; Target recognition module: The target recognition module recognizes multiple target locations in power asset equipment based on multi-source image information; the recognition of multiple target locations is based on visible light video key frame image recognition; Corona and arc recognition module: The corona and arc recognition module recognizes corona and arc occurring in power asset equipment based on multi-source image information and the multi-target positions; Life cycle assessment module: the life cycle assessment module performs life cycle assessment on the power asset equipment based on the corona and arc identification results; The corona identification process specifically includes: S3-1-1: registering the visible light key frame with the ultraviolet light image; S3-1-2: Generate a target weight matrix based on the identified multi-target positions; The Gaussian distribution weight functions of each target area are superimposed to generate a global weight matrix W(x, y) and calculate the collected multiple UV images Calculate the average brightness of the same pixel Brightness difference ΔI ult (x, y) and brightness standard deviation Extract candidate corona region M final (x, y), M candidate (x,y)=(M1(x,y)∩M2(x,y))·W(x,y); Among them, T avg , T diff1 , T diff2 , T final are threshold parameters, M1(x, y), M2(x, y) are filter masks, M candidate (x, y) is the initial candidate corona region; Arc recognition is realized by using a dual-channel recognition network. The dual-channel recognition network is divided into three stages of feature extraction. The first stage input feature maps are the difference image of the arc image sequence and the ultraviolet key frame and the candidate corona region M. final Weighted fusion of (x, y).

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for managing the lifecycle of power assets based on a cloud platform is implemented.

8. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a cloud platform-based power asset lifecycle management method as described in any one of claims 1 to 5.

Images