Highway photovoltaic array panoramic health assessment operation and maintenance method and system

Abstract

The present application provides a highway photovoltaic array panoramic health assessment operation and maintenance method and system, belonging to the field of panoramic health assessment operation and maintenance, the method comprising: obtaining component images of photovoltaic arrays on highways; for images with availability problems in the images, performing occlusion confidence analysis to generate enhanced images and image confidence scores; based on the enhanced images and image confidence scores, calculating the health score of the current component at the current time; based on the health score of the current component at the current time, training a trend prediction model; based on the health status of the component at a certain time point in the future, combining the geographic coordinates of the corresponding component, constructing a two-dimensional spatial health trend graph, performing spatial clustering analysis on the two-dimensional spatial health trend graph, generating a priority score for each candidate task area, generating a fault early warning scheduling list according to the candidate task area and the corresponding priority score, for generating daily or weekly inspection dispatching tasks.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of panoramic health assessment and operation and maintenance, and particularly relates to a method and system for panoramic health assessment and operation and maintenance of a photovoltaic array of an expressway. BACKGROUND

[0002] In recent years, as an important part of clean energy, photovoltaic power generation has gradually increased in application along expressways. Photovoltaic arrays of expressways are usually deployed in guardrails, soundproof walls, bridge slopes, and service area roofs, and have the advantages of high space utilization efficiency, flexible grid connection, and coordinated development with transportation infrastructure. However, due to the special deployment environment, the photovoltaic arrays are subject to interference from high-speed traffic, large geographical span, dispersed distribution of components, and significant differences in climate conditions. The health assessment and operation and maintenance mechanisms used in conventional centralized power stations cannot be directly applied to the expressway scenario. The main health assessment methods for photovoltaic components currently mainly rely on fixed cameras, unmanned aerial vehicle thermal imaging, infrared spectrum analysis, electrical characteristic comparison, or manual inspection. However, these methods face several major bottlenecks: first, the sensing coverage is limited, and fixed points cannot obtain the status of linearly arranged components in real time; second, environmental interference is severe, such as traffic obstruction, dust accumulation, and complex lighting conditions, which leads to a decrease in image acquisition quality and affects identification accuracy; third, data acquisition and diagnosis are separated, and health assessment mostly stays at the level of "periodic inspection", lacking the ability to model fault evolution trends and the ability to respond to warnings, resulting in late fault discovery and low processing efficiency; and fourth, the deployment cost is high, especially for unmanned aerial vehicle systems, which have difficulty in operation and maintenance in mountainous areas or closed road sections, limiting their normal use. Under these limitations, the current technology cannot build a photovoltaic component health assessment system that is wide in coverage, fast in response, has trend judgment ability, and is suitable for expressway scenarios.

[0003] Therefore, how to build an "low-cost, high-coverage, strong-robust, and predictable" assessment and operation and maintenance method has become a key challenge for the sustainable operation of photovoltaic systems on expressways. SUMMARY

[0004] The purpose of the present application is to propose a method and system for panoramic health assessment and operation and maintenance of a photovoltaic array of an expressway, which fully considers the characteristics of linear deployment, vehicle dynamic flow, and sparse component distribution on expressways, and innovatively builds a health assessment system based on a dynamic inspection mechanism and multi-source sensing fusion.

[0005] In a first aspect, an embodiment of the present application provides a method for panoramic health assessment and operation and maintenance of a photovoltaic array of an expressway, which comprises:

[0006] obtaining component images of the photovoltaic array on the expressway, wherein the component images include cropped image blocks of corresponding photovoltaic components in the images, image acquisition times corresponding to the components, and geographic coordinates of the corresponding components;

[0007] Performing occlusion confidence analysis on the images in the component images that have availability problems, generating enhanced images and image confidence scores;

[0008] Based on the enhanced images and image confidence scores, inputting a single-layer perception machine to obtain a health score of the current component at the current time;

[0009] Based on the health score of the current component at the current time, training a trend prediction model and adding a regular term based on the range of weight distribution in the current sample sequence, obtaining a trained trend prediction model, which is used to estimate the health status of the component at a future time point;

[0010] Based on the health status of the component at a future time point, combining the geographic coordinates of the corresponding component, constructing a two-dimensional spatial health trend graph for judging the health degree and health change speed of the current component, performing spatial clustering analysis on the two-dimensional spatial health trend graph to generate a priority score of each candidate task area, and generating a fault early warning scheduling list according to the candidate task area and the corresponding priority score, for generating daily or weekly inspection task assignment; wherein the candidate task area is an operation and maintenance task candidate area.

[0011] Preferably, the component images of the photovoltaic array on the expressway are collected by a patrol vehicle equipped with a front-view camera, an IMU and a GNSS.

[0012] Each frame of image synchronously records a time stamp, a GNSS position and camera pose information provided by the IMU.

[0013] Preferably, the component images are constructed as follows:

[0014] A lightweight image detection model is used to detect photovoltaic components in each frame of image, and a rectangular frame set is output, which includes pixel coordinate frames of a plurality of photovoltaic component regions in the image.

[0015] For each component region, the component region is mapped to a real geographic coordinate system through spatial perspective transformation according to the GNSS position and the camera pose information, to generate geographic coordinates of the corresponding component.

[0016] According to the pixel coordinate frame and the geographic coordinates, occlusion confidence calculation is performed on each frame of image to determine whether there is image occlusion caused by factors such as traffic flow; wherein the occlusion confidence closest to 1 indicates the highest image quality, and the occlusion confidence closest to 0 indicates occlusion or low quality.

[0017] The cropped image block of each component image, the image acquisition time corresponding to the component, and the geographic coordinates are combined into a triple, which is stored in a component space index library.

[0018] Preferably, the occlusion confidence analysis is performed on the images with availability problems in the component images to generate an enhanced image and an image confidence score, including:

[0019] An image occlusion mask is constructed.

[0020] Based on the image occlusion mask, the adjacent frame difference of the cropped image block, and the relative speed of the vehicle, an image confidence score is determined, calculated as:

[0021]

[0022] wherein, is the total number of pixels in the image identified as an occluded area; is the total number of pixels in the image, representing the maximum unoccluded area; is the L2 norm of the pixel difference between the two images, used to determine whether the component surface state is stable; is the spatial speed of the vehicle at the two sampling times, used to determine the shooting stability; is the weight of adjusting the influence of image difference and vehicle movement on image confidence; is the global confidence rate adjustment term; is a normalization constant to ensure ; is the final image confidence score, with a value closest to 1 indicating the most reliable and a value closest to 0 indicating that it should be shielded or replaced;

[0023] For images with a confidence score below a preset threshold, a cross-time local area compensation mechanism is used to fill in the occluded area to generate an enhanced image.

[0024] Preferably, based on the enhanced image and the image confidence score, a single-layer perception machine is input to obtain a health score of the current component at the current time, including:

[0025] An enhanced image is obtained.

[0026] The enhanced image is input into a lightweight convolutional encoder to extract an image feature vector.

[0027] The image feature vector is scaled and weighted by the image confidence score to obtain a quality-adjusted feature.

[0028] The quality-adjusted feature is input into a single-layer perception machine to output a health score.

[0029] Preferably, the prediction formula of the trend prediction model is:

[0030]

[0031] wherein, is the health prediction of the th component at future time ; is the set of all time points collected historically for the component; is the health score at the historical sampling time ; is the confidence of the image corresponding to the time point; is the confidence decay kernel function, which comprehensively considers time decay and image confidence.

[0032] Preferably, the two-dimensional space health trend graph is constructed, comprising:

[0033] obtaining the health state at a certain future time point, the geographic coordinates of the corresponding component, and the health value change rate calculated from the historical prediction sequence;

[0034] constructing a two-dimensional space health trend graph , each component occupies a grid of the graph at the geographic coordinates of the corresponding component, and the corresponding grid point is given a weighted trend health value, and the weighted health state of the component in the trend graph is output, which is used to judge the health degree and health change speed of the current component; if the health state at a certain future time point is low and changes fast, the weighted health state will rapidly decrease, and the current region corresponds to a pre-warning hot area.

[0035] Preferably, the two-dimensional space health trend graph is subjected to spatial clustering analysis to generate a priority score of each candidate task region, which is calculated as:

[0036]

[0037] wherein, is the number of components in the region ; is the geographic area of the region; is the geographic distance between the center of the region and the nearest operation and maintenance scheduling center; is the weight coefficient, which is used to balance the trend severity, the power influence range and the scheduling cost; is the comprehensive scheduling priority score of the task region .

[0038] In a second aspect, the embodiments of the present application provide a highway photovoltaic array panoramic health evaluation operation and maintenance system, which comprises:

[0039] An acquisition unit is configured to acquire component images of a photovoltaic array on an expressway, wherein the component images include cropped image blocks of corresponding photovoltaic components in the images, image acquisition times corresponding to the components, and geographic coordinates of the corresponding components;

[0040] A confidence analysis unit is configured to perform occlusion confidence analysis on images having availability problems in the component images, to generate enhanced images and image confidence scores.

[0041] A health analysis unit is configured to input a single-layer perception machine based on the enhanced images and the image confidence scores, to obtain a health score of a current component at a current time.

[0042] A health prediction unit is configured to train a trend prediction model based on the health score of the current component at the current time, and add a regular term based on a range of a weight distribution in a current sample sequence, to obtain a trained trend prediction model, which is used to estimate a health state of the component at a future time point.

[0043] A health task generation unit is configured to construct a two-dimensional spatial health trend graph based on the health state of the component at the future time point and the geographic coordinates of the corresponding component, to judge a health degree and a health change speed of the current component, to perform spatial clustering analysis on the two-dimensional spatial health trend graph, to generate a priority score of each candidate task area, to generate a fault early warning scheduling list based on the candidate task area and the corresponding priority score, and to generate a daily or weekly inspection dispatching task, wherein the candidate task area is an operation and maintenance task candidate area.

[0044] The present application has at least the following beneficial technical effects:

[0045] In view of the problems of poor data acquisition coverage, insufficient environmental interference robustness, lack of trend prediction ability and low operation and maintenance response efficiency in the prior art, the present application proposes a panoramic health assessment and operation and maintenance method and system for highway photovoltaic arrays. The method fully considers the scene characteristics such as linear deployment of highways, dynamic flow of vehicles, and sparse distribution of components, and innovatively constructs a health assessment system based on a dynamic inspection mechanism and multi-source perception fusion. On the one hand, by introducing a data collection mechanism with flexible deployment capability, vehicles or roadside infrastructure on the highway are involved in component state perception, significantly improving data coverage and reducing perception cost. On the other hand, a robust state recognition mechanism that can adapt to typical interference factors such as occlusion and light changes in high-speed scenes is introduced in the health assessment process, and a health prediction method combining time evolution trends is established, so that not only the current fault state can be recognized, but also the degradation trend of the component can be perceived in advance, providing intelligent support for maintenance scheduling. The overall system has the characteristics of strong real-time, flexible deployment and high adaptability, and can effectively solve the problems that existing photovoltaic health assessment methods are difficult to apply in highways and lack of reliability. BRIEF DESCRIPTION OF DRAWINGS

[0046] The application is further illustrated by the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the application. For ordinary skilled in the art, other drawings can be obtained without creative labor on the basis of the following drawings.

[0047] Figure 1 Flow chart of the highway photovoltaic array panoramic health assessment and operation and maintenance method of the present application. DETAILED DESCRIPTION

[0048] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation on the present application.

[0049] In one embodiment, as shown in Figure 1 a highway photovoltaic array panoramic health assessment and operation and maintenance method is provided, comprising the following steps:

[0050] S1, acquiring component images of photovoltaic arrays on the highway, wherein the component images include cropped image blocks of corresponding photovoltaic components in the images, image acquisition times corresponding to the components, and geographic coordinates of the corresponding components.

[0051] In particular, this step aims to solve the problems of wide spatial distribution, strong linear extension, high fixed monitoring cost, and real-time coverage difficulty of highway photovoltaic arrays. It uses existing traffic flow resources (such as inspection vehicles or ETC sensing vehicles) for dynamic image acquisition and positioning. The core of this step is to achieve spatial positioning and image quality assurance of components through vehicle-mounted sensor fusion, providing reliable raw data for subsequent health assessment, trend modeling, and other steps. We first use an inspection vehicle equipped with a forward-looking camera, IMU, and GNSS to collect image frames at a certain frequency (e.g., 2Hz) during driving along the highway .

[0052] where each frame of image is recorded with a timestamp , GNSS position , and camera pose information provided by IMU .

[0053] Use a lightweight image detection model (such as YOLOv5-Nano) to detect photovoltaic components in each frame of image , output a set of rectangular frames:

[0054]

[0055] where each represents the pixel coordinate frame (including center point and width and height ) of the th detected photovoltaic component region in the image.

[0056] where, for each component region , according to the current position of the vehicle and the camera pose , the component region is mapped to the real geographic coordinate system through spatial perspective transformation, and the geographic coordinates of the component are calculated :

[0057]

[0058] : represents the pixel coordinate frame (including center point and width and height ) of the th detected photovoltaic component region in the image;

[0059] : camera pose information (such as pitch angle, yaw angle, roll angle) when image is taken, measured by IMU;

[0060] : GNSS position (WGS84 geographic coordinate system) when the vehicle currently collects the image;

[0061] : Geographic coordinates of the th component on the map;

[0062] : Transform function of geographic position based on camera model and pose information to complete perspective back-projection and combine GPS position.

[0063] In order to improve the effectiveness of image data, the occlusion confidence of each frame of image is calculated to determine whether there is image occlusion caused by traffic flow and other factors, and the occlusion confidence is defined as:

[0064]

[0065] : Current image frame;

[0066] : Image frame at the previous moment;

[0067] : Pixel difference between two frames of images, representing the scene change amplitude;

[0068] : Vehicle speed (which can be approximated as displacement length) calculated based on the two position coordinates;

[0069] : Scaling coefficient for controlling the influence of image change on confidence;

[0070] : Weight factor for controlling the sensitivity of vehicle movement to image occlusion;

[0071] : Occlusion confidence value, the closer to 1, the higher the image quality, the closer to 0, the existence of occlusion or low quality.

[0072] Each component image area is combined with its collection time and geographic coordinates to form a triple, which is stored in the component space index library :

[0073]

[0074] : Cropped image block of the th photovoltaic component in the image;

[0075] : Image collection time;

[0076] : Geographical coordinates of the corresponding component

[0077] : Component spatial index library, and as the input data source of subsequent steps (such as the basis of image repair, health identification, trend prediction).

[0078] In this way, the component image data collected by multiple inspection vehicles can be uploaded to the cloud server for fusion and unified storage, forming a structured data platform supporting spatial query + time backtracking.

[0079] S2, occlusion confidence analysis is performed on the images with availability problems in the component images, and enhanced images and image confidence scores are generated.

[0080] Specifically, this step processes the component images output by step one , focusing on the image availability decline problems caused by traffic occlusion (such as large vehicles driving through), strong light reflection, rain and snow interference, etc. for photovoltaic array deployment in highway slope, soundproof wall, shoulder, etc. Enhancement and reconstruction are performed.

[0081] High-speed photovoltaic components in images often exhibit partial occlusion (such as only covering the lower right corner of the component), local highlight / overexposure (such as a mirror or sunlight), which can mislead subsequent health identification (step three). Therefore, a special mechanism must be introduced to complete the confidence modeling, reliable area identification, and occlusion repair enhancement of the image in this step.

[0082] This step first constructs an image occlusion mask , uses a lightweight spatial attention mechanism model to mark the pixel level of the occluded area in the image, and obtains the occlusion probability of each pixel:

[0083] Use self-attention graph and local edge discontinuous area to jointly judge the occlusion probability, , the numerical value of the occluded area is close to 1.

[0084] In order to solve the problem of "image availability is difficult to quantify", this scheme introduces an innovative occlusion-spatiotemporal coupling confidence scoring function, which considers the image occlusion mask , the difference between adjacent frames and , as well as the relative speed of the vehicle (calculated by and ) to jointly evaluate image quality:

[0085]

[0086] : the total sum of pixels identified as occluded regions in the image;

[0087] : the total number of pixels in the image (e.g. ), representing the maximum unoccluded area;

[0088] : the L2 norm of the pixel difference between the two consecutive images, used to determine whether the component surface state is stable;

[0089] : the spatial velocity of the vehicle at the two sampling times, , used to determine the shooting stability;

[0090] : the weight that adjusts the influence of image difference and vehicle movement on image confidence;

[0091] : the global confidence rate adjustment term;

[0092] : the normalization constant, ensuring ;

[0093] : the final image confidence score, the closer to 1, the more reliable, the closer to 0, the more should be shielded or replaced.

[0094] It can be understood that this formula innovatively combines spatial occlusion information with image dynamic stability, and introduces the modeling of "shooting reliability" through the vehicle movement rate , which is designed specifically for the joint problem of image blur / tilt / occlusion in highway scenes, avoiding traditional methods that rely solely on image texture changes to determine quality.

[0095] For images with confidence scores below a threshold (e.g. ), i.e. , a specially designed cross-time local region compensation mechanism in this invention is used to fill in the occluded regions, rather than directly using an image inpainting network. The final output is the enhanced image and its corresponding confidence score , geographical coordinates and timestamp form a four-tuple .

[0096] S3, based on the enhanced image and image confidence score, input a single-layer perception machine to obtain the health score of the current component at the current time.

[0097] Specifically, the task of this step is to obtain the enhanced image Image confidence score In the process, it is deduced that each photovoltaic module at the current time... Health status score and component geographic coordinates Output them together to build a state dataset for trend prediction.

[0098] To avoid image quality from misleading the recognition results, this step employs a confidence-guided lightweight state modeling model. We take the image and its confidence level as input and output a component health score. Instead of using hard-label classification, we use a continuous scoring method to quantify the component state from 0 (severe failure) to 1 (complete health), which adapts to the reality of blurry images and minor anomalies.

[0099] This step uses the following state modeling structure:

[0100] Will Input a lightweight convolutional encoder to extract image feature vectors ;

[0101] Using occlusion confidence The features are scaled and weighted to obtain quality-adjusted features;

[0102] The feature is input into a single-layer perceptron, which outputs a health score. :

[0103]

[0104] :pass Extracted component image features;

[0105] Image quality score, derived from step two, with a range of values. ;

[0106] : Linear layer parameters for state scoring;

[0107] The sigmoid activation function guarantees the output. ;

[0108] : No. Each component in A health score at any given time.

[0109] Understandably, the creativity of this modeling process lies in its use of... As a feature modulation factor, it essentially realizes the "image credibility perception modeling" mechanism, which automatically reduces the output weight when the image credibility is insufficient, thereby reducing misjudgments.

[0110] We further propose a dynamic confidence stretching mechanism to improve the model's stability on partially occluded images, making the scoring results more neutral as the image confidence decreases (avoiding extreme values):

[0111]

[0112] The final health score after confidence level smoothing;

[0113] Explanation: When When the value approaches 1, meaning the image has high confidence, the output approaches the model's judgment. ;

[0114] when At lower values, the output approaches 0.5, indicating uncertain or blurry areas, thus avoiding outputting extreme values ​​for low-quality images.

[0115] Output structured component health status triples This will be used by the next non-uniform time series modeling module.

[0116] S4. Based on the current component's health score at the current moment, train the trend prediction model and add a regularization term based on the range of the weight distribution in the current sample sequence to obtain the trained trend prediction model. The trained trend prediction model is used to predict the health status of the component at a future point in time.

[0117] Specifically, the goal of this step is to build a trend prediction model based on "non-uniform time series + occlusion confidence" after completing the health status identification, in order to predict the component at a future point in time. health status Photovoltaic inspection in highway scenarios presents two typical problems:

[0118] Among them, time sampling is non-uniform: sampling time The interval varies greatly, which does not satisfy the "fixed step size" assumption of the traditional RNN model;

[0119] Data quality varies: the quality of each image acquisition is inconsistent due to factors such as vehicle speed, weather, and occlusion, making it unreliable to model trends based solely on state values.

[0120] Therefore, this step proposes a "non-uniform state convolution modeling method with confidence decay kernel". The overall idea is: to convert the state values ​​at different times... The future state is weightedly estimated through a kernel function convolution with a time difference decay term and a confidence suppression term The prediction formula is constructed as follows:

[0121]

[0122] The health prediction of the first component at future time

[0123] The set of all time points collected by the component in history

[0124] The health score of the historical sampling time

[0125] The confidence of the image corresponding to the time

[0126] The confidence decay kernel function is defined as follows:

[0127]

[0128] The time difference between the prediction point and the sample point

[0129] The time decay rate (controls the rapid weakening of the influence of long-interval samples)

[0130] The image confidence (used to limit the effect of low-quality samples)

[0131] A small positive number (to avoid the product being zero in the case of

[0132] The normalization factor makes the weight sum of all The formula realizes high-order decay control of the time difference

[0133] At the same time, the confidence term is fused The "the higher the image quality, the greater the prediction contribution" mechanism is directly embedded into the state modeling kernel

[0134] In addition, in order to enhance the robustness of the model, the present application introduces an innovative regularization mechanism on the kernel function to control the model to avoid being "misled" by isolated high-confidence points, and the mechanism is:

[0135] ​​​​​

[0136] : confidence non-uniformity penalty term;

[0137] and : range of weight distribution in current sample sequence;

[0138] : regularization coefficient.

[0139] If the weight of a certain moment is much larger than other samples, it will lead to a serious dependence on the prediction results of this point. Through this regularization term, we inhibit the "extreme single-point control" and enhance the model generalization ability.

[0140] The output is the health score prediction of each component in the future time , and the corresponding time and geographical coordinates, forming a triple , which is used to build a trend space heat map and generate operation and maintenance scheduling suggestions in the next step.

[0141] S5, based on the health status of the component at a certain time in the future, combined with the geographical coordinates of the corresponding component, a two-dimensional space health trend chart is constructed, which is used to judge the health degree and health change speed of the current component, and the two-dimensional space health trend chart is subjected to spatial clustering analysis, to generate a priority score of each candidate task area, and generate a fault early warning scheduling list according to the candidate task area and the corresponding priority score, for generating daily or weekly patrol dispatching tasks; wherein the candidate task area is an operation and maintenance task candidate area.

[0142] Specifically, this step aims to combine the future health prediction results of the components obtained in the previous stage with their geographical coordinates to construct a two-dimensional space health trend chart for the entire PV array , and based on the trend evolution speed, to determine the health deterioration degree of different areas, to generate a fault early warning priority list , and finally to link the operation and maintenance system to trigger maintenance scheduling.

[0143] The input data is the triple output by step four:

[0144] : predicted health state value at future time point;

[0145] : predicted target time;

[0146] : the first​ the position of a component in the map coordinate system, denoted as from step one;

[0147] In addition, we calculate the health value change rate from the historical prediction sequence, to evaluate the health degradation trend.

[0148] To map the health state into a spatial graph, we define a two-dimensional spatial health trend graph , where each component occupies a cell in the graph at its geographical coordinate We assign a weighted trend health value to the cell, considering both the current health state and the degradation speed:

[0149]

[0150] : predicted health value;

[0151] : state change rate between the last two time instants;

[0152] : trend penalty coefficient (e.g., 0.5~1.5);

[0153] : weighted health state of the component in the trend graph, used to judge whether it is "currently bad + changing fast".

[0154] It can be understood that this trend index reflects the "health degree x change speed" composite measurement;

[0155] If the health value is low and the change is fast, then it rapidly decreases, corresponding to the early warning hot area;

[0156] Compared with the static state, this value can more truly reflect the real risk of the component being "rapidly deteriorating".

[0157] After constructing the complete graph , we automatically detect potential "high degradation areas" on the graph through a spatial clustering algorithm (such as DBSCAN), i.e., the values of locally continuous components are below a certain threshold , and the spatial distribution is dense. Each clustering area is defined as a "candidate area for operation and maintenance" . For each candidate task area , we calculate its comprehensive priority score , considering three dimensions:

[0158] the average trend health value of the components in the area;

[0159] Area of the region (influence of possible power loss)

[0160] Spatial distance to the nearest service point (influence of scheduling cost)

[0161] The final priority calculation formula is as follows:

[0162]

[0163] Area Number of components in the region

[0164] Area geographic area (can be calculated by the minimum bounding box or convex hull)

[0165] Geographic distance between the region center and the nearest operation and maintenance scheduling center (meters)

[0166] Weight coefficient, used to balance the trend severity, power influence range and scheduling cost

[0167] The comprehensive scheduling priority score of the task region

[0168] It can be understood that the lower the trend value (the more deteriorated), the larger the region, the closer to the service station, the higher the priority

[0169] This mechanism embodies the key system advantage of the present application - from single point judgment to spatial-scheduling integrated comprehensive operation and maintenance evaluation model.

[0170] Finally, a fault warning scheduling list is formed , which is sorted from high to low to generate the daily or weekly inspection task, and the scheduling system retrieves the nearest maintenance personnel, vehicles and other resources according to The center point automatically generates the scheduling path and execution instructions.

[0171] In one embodiment, a highway photovoltaic array panoramic health assessment operation and maintenance system is provided, which comprises:

[0172] An acquisition unit is configured to acquire component images of photovoltaic arrays on the highway, wherein the component images include cropped image blocks of corresponding photovoltaic components in the images, image acquisition times corresponding to the components, and geographic coordinates of the corresponding components

[0173] A confidence analysis unit is configured to perform occlusion confidence analysis on images with availability problems in the component images to generate enhanced images and image confidence scores.​

[0174] a health analysis unit configured to obtain a health score of the current component at the current time based on the enhanced image and the image confidence score by inputting a single-layer perception machine;

[0175] a health prediction unit configured to train a trend prediction model based on the health score of the current component at the current time, and add a regular term based on a range of a weight distribution in the current sample sequence to obtain a trained trend prediction model, the trained trend prediction model being used to estimate a health state of the component at a future time point;

[0176] a health task generation unit configured to construct a two-dimensional spatial health trend graph based on the health state of the component at the future time point and geographical coordinates of the corresponding component, to determine a health degree and a health change speed of the current component, to perform spatial clustering analysis on the two-dimensional spatial health trend graph to generate a priority score of each candidate task area, and to generate a fault early warning scheduling list based on the candidate task area and the corresponding priority score, the fault early warning scheduling list being used to generate a daily or weekly inspection dispatching task, wherein the candidate task area is an operation and maintenance task candidate area.

[0177] The relative steps, numerical expressions and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present application, unless otherwise specifically stated.

[0178] If the functions are realized in the form of software function units and sold or used as independent products, the functions can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present application or parts of the present application that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the system described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0179] In the description of the present application, it should be noted that the terms "upper", "lower", etc. indicate the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly used when the product is used, and are only used to facilitate the description of the present application and simplify the description, and do not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0180] In the description of the application, it also needs to be explained that, unless otherwise explicitly specified and limited, the terms "set", "install", "connect", "connect" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0181] Finally, it should be pointed out that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A panoramic health assessment and maintenance method for photovoltaic arrays on highways, characterized in that: The method includes: Acquire component images of a photovoltaic array on a highway, wherein the component images include cropped image blocks of the corresponding photovoltaic components in the image, the image acquisition time corresponding to the component, and the geographical coordinates of the corresponding component; For images with usability issues in the component images, occlusion confidence analysis is performed to generate enhanced images and image confidence scores, including: Construct an image occlusion mask; Based on the image occlusion mask, the differences between adjacent frames of the cropped image patch, and the relative speed of the vehicle, the image confidence score is determined and calculated as follows: in, The sum of pixels identified as occluded areas in the image; A is the total number of pixels in the image, representing the maximum unoccluded area; The L2 norm of the pixel difference between two consecutive frames is used to determine whether the surface state of the component is stable. The spatial velocity of the vehicle at the two sampling times. This is used to determine shooting stability; To adjust the weights of image differences and vehicle movement on image confidence; This is an adjustment term for the rate of decrease in global confidence. As a normalization constant, ensure The final image confidence score is given, with the closest score being 1 indicating the most reliable image and the closest score being 0 indicating that the image should be masked or replaced. For images with a confidence level below a preset threshold, a cross-time local region compensation mechanism is used to fill in the occluded areas and generate an enhanced image. Based on the enhanced image and image confidence score, the health score of the current component at the current moment is obtained by inputting it into a single-layer perceptron. Based on the current component’s health score at the current moment, the trend prediction model is trained, and a regularization term based on the range of the weight distribution in the current sample sequence is added to obtain the trained trend prediction model. The trained trend prediction model is used to predict the health status of the component at a future point in time. Based on the health status of the components at a future point in time, and combined with the geographical coordinates of the corresponding components, a two-dimensional spatial health trend map is constructed to determine the current health level and rate of health change of the components. Spatial clustering analysis is performed on the two-dimensional spatial health trend map to generate a priority score for each candidate task area. A fault warning scheduling list is generated based on the candidate task areas and their corresponding priority scores to generate inspection dispatch tasks for the current day or week. The candidate task areas are maintenance task candidate areas.

2. The panoramic health assessment and maintenance method for highway photovoltaic arrays according to claim 1, characterized in that, The acquisition of component images of the photovoltaic array on the highway is carried out by an inspection vehicle equipped with a forward-looking camera, IMU and GNSS. Each frame of the image synchronously records a timestamp, GNSS location, and camera attitude information provided by the IMU.

3. The panoramic health assessment and maintenance method for highway photovoltaic arrays according to claim 2, characterized in that, The component image is constructed as follows: A lightweight image detection model is used to detect photovoltaic modules in each frame of the image, and a set of rectangular boxes is output. The set of rectangular boxes includes the pixel coordinate boxes of several photovoltaic module regions in the image. For each component region, based on GNSS location and camera attitude information, the component region is mapped to the real geographic coordinate system through spatial perspective transformation to generate the geographic coordinates of the corresponding component. Based on the pixel coordinate frame and geographic coordinates, an occlusion confidence score is calculated for each frame of the image to determine whether there is image occlusion due to traffic flow factors; wherein, the occlusion confidence score closest to 1 indicates the highest image quality, and the closest to 0 indicates the presence of occlusion or low quality; Each component image's cropped image block is paired with the corresponding image acquisition time and geographic coordinates to form a triple, which is then stored in the component spatial index library.

4. The panoramic health assessment and maintenance method for highway photovoltaic arrays according to claim 1, characterized in that, The step of obtaining the health score of the current component at the current moment by inputting the enhanced image and image confidence score into a single-layer perceptron includes: Obtain enhanced images; The enhanced image is input into a lightweight convolutional encoder to extract image feature vectors; The image feature vector is scaled and weighted using the image confidence score to obtain the quality-adjusted features. The quality-adjusted features are input into a single-layer perceptron, which outputs a health score.

5. The panoramic health assessment and maintenance method for highway photovoltaic arrays according to claim 1, characterized in that, The prediction formula of the trend prediction model is as follows: in, For the j-th component in the future time Health predictions; This is the set of all historical time points collected by this component. Historical sampling time Health score; This represents the confidence level of the image at that moment. The confidence decay kernel function takes into account both time decay and image confidence.

6. The panoramic health assessment and maintenance method for highway photovoltaic arrays according to claim 5, characterized in that, The construction of the two-dimensional spatial health trend map includes: obtaining the health status at a future point in time, the geographical coordinates of the corresponding components, and calculating the rate of change of health values ​​from historical prediction sequences; and constructing the two-dimensional spatial health trend map. Each component occupies a grid cell in the map at its corresponding geographic coordinates, and a weighted trend health value is assigned to the corresponding grid cell. The weighted health status of the component in the trend map is output to determine the current health level and the rate of health change of the component. If the health status is low and changes rapidly at a future point in time, the weighted health status will drop rapidly, and the current area will be a warning hot zone.

7. The panoramic health assessment and maintenance method for highway photovoltaic arrays according to claim 6, characterized in that, The spatial clustering analysis performed on the two-dimensional spatial health trend map generates a priority score for each candidate task region, which is calculated as follows: in, For the region Number of components; The geographical area of ​​the region; This represents the geographical distance between the regional center and the nearest operations and maintenance dispatch center. These are weighting coefficients used to balance the severity of the trend, the extent of the power impact, and dispatch costs. For the mission area The overall scheduling priority score.

8. A system for implementing the panoramic health assessment and maintenance method for highway photovoltaic arrays as described in claim 1, characterized in that, The system includes: The acquisition unit is used to acquire component images of photovoltaic arrays on highways, wherein the component images include cropped image blocks of the corresponding photovoltaic components in the image, the image acquisition time corresponding to the component, and the geographical coordinates of the corresponding component; The confidence analysis unit is used to perform occlusion confidence analysis on images with usability issues in the component images, and generate enhanced images and image confidence scores; A health analysis unit is used to input the enhanced image and image confidence score into a single-layer perceptron to obtain the health score of the current component at the current moment. The health prediction unit is used to train the trend prediction model based on the current component's health score at the current moment, and to add a regularization term based on the range of the weight distribution in the current sample sequence to obtain the trained trend prediction model. The trained trend prediction model is used to predict the health status of the component at a future point in time. The health task generation unit is used to construct a two-dimensional spatial health trend map based on the health status of the component at a future point in time and the geographical coordinates of the corresponding component. This map is used to determine the current health level and rate of health change of the component. Spatial clustering analysis is performed on the two-dimensional spatial health trend map to generate a priority score for each candidate task area. A fault warning scheduling list is generated based on the candidate task areas and their corresponding priority scores to generate inspection dispatch tasks for the current day or week. The candidate task areas are maintenance task candidate areas.

Images