AR intelligent photovoltaic inspection method and system

Abstract

The invention discloses an AR intelligent photovoltaic inspection method and system, and the method comprises the following steps: S1, obtaining a panoramic image and heat distribution data of a photovoltaic module, and generating a data set; s2, multi-modal data preprocessing is carried out, key fault features are extracted, and fault areas are marked; s3, focusing and analyzing the marked area, and identifying a thermal abnormal point; s4, constructing an electrical topological graph, and generating a fault point priority; s5, performing ant colony optimization to adjust the inspection path, and marking the position of a fault point; s6, displaying the path, the fault point and the risk heat map in the AR equipment; and S7, generating an inspection report and performing feedback optimization. According to the invention, through multi-modal data fusion, dynamic path optimization and augmented reality technologies, efficient detection, accurate positioning and visual display of photovoltaic module faults are realized, and the inspection efficiency and the intelligent level are improved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic engineering technology, and in particular to an AR intelligent photovoltaic inspection method and system. Background Art

[0002] Photovoltaic power generation technology is an important part of the renewable energy field. It has been widely used around the world because of its clean, efficient and resource-rich characteristics. With the continuous expansion of the scale of photovoltaic power stations and the acceleration of deployment, the operation, maintenance and performance monitoring of photovoltaic modules have become key links to ensure system stability and economic benefits. Photovoltaic modules, the core equipment of photovoltaic power stations, are easily affected by factors such as dust, shading, module aging, hot spot effects, etc. due to their long-term exposure to the natural environment, which can lead to performance degradation or even failure. Therefore, inspection of photovoltaic power stations is an important means to discover problems, improve system efficiency and extend equipment life.

[0003] Traditional inspections of photovoltaic power stations mainly rely on manual inspections or drone equipment for filming. These methods have improved efficiency to a certain extent, but there are still many shortcomings. For example, manual inspections are limited by high labor costs, low detection efficiency, and limited coverage; and although drone inspections have improved coverage, they usually only rely on a single thermal imaging or ordinary visual inspection technology, making it difficult to fully obtain multimodal characteristic data of photovoltaic modules. In addition, these inspection methods are mostly aimed at fault detection, lack in-depth analysis of fault type, severity, and propagation path, and are difficult to prioritize and accurately locate faults.

[0004] In terms of processing and displaying fault information, existing technologies usually use traditional reporting formats to present inspection results in the form of static data or two-dimensional charts. Although this model can meet basic information transmission needs, it lacks intuitiveness, especially when facing large-scale photovoltaic power plants. It cannot quickly reflect the overall picture of inspection paths, fault point locations, and risk distribution. In addition, traditional inspection result displays are often divorced from actual geographical scenarios, making it difficult for users to intuitively perceive the relationship between fault distribution and power plant spatial layout, thus affecting decision-making efficiency.

[0005] In recent years, due to its ability to overlay virtual information in real scenarios, augmented reality (AR) technology has provided new possibilities for the visualization of the inspection results of photovoltaic power stations. However, the application of existing inspection schemes based on AR technology in the photovoltaic field is still in its initial stage, with problems such as single function, poor interactivity, and insufficient real-time performance. For example, some schemes only achieve simple display of the path or a single fault point, fail to integrate multi-source data such as thermal imaging, geographic information, and fault analysis, and even more fail to dynamically generate a comprehensive thermal map reflecting the inspection coverage and risk distribution. In addition, the interactive operations of users and system feedback often do not form a closed loop, and it is difficult for the system to optimize the display content or adjust the inspection plan according to real-time feedback.

[0006] Therefore, how to provide an AR intelligent photovoltaic inspection method and system is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose an AR intelligent photovoltaic inspection method and system. The present invention comprehensively processes information such as thermal anomaly points, fault point priorities, inspection path coverage, and risk distribution through multi-modal data fusion technology, and combines a dynamic path optimization algorithm and a risk distribution thermal map generation technology to generate an intuitive display of inspection results. Through augmented reality technology, the present invention realizes the dynamic visualization display of the inspection path, fault points, and repair suggestions superimposed in the actual photovoltaic power station scenario, and at the same time provides a thermal map to display the inspection coverage and risk distribution, supporting users to perform interactive operations on the display content through gestures or voices. The present invention makes full use of the intuitiveness of augmented reality technology, the comprehensiveness of multi-modal data, and the high efficiency of dynamic interaction, solves the technical problems of non-intuitive data presentation, insufficient dynamic performance, lack of risk distribution display, and decision-making support in traditional inspection methods, and has the remarkable advantages of clear display content, comprehensive information integration, accurate inspection path optimization, and excellent user experience, providing a brand-new solution for the intelligent operation and maintenance of photovoltaic power stations.

[0008] An AR intelligent photovoltaic inspection method according to an embodiment of the present invention includes the following steps:

[0009] S1. Obtain a panoramic visible light image of a photovoltaic module through a bionic compound eye camera, and combine a thermal imaging sensor to obtain thermal distribution data of the photovoltaic module to generate a data set;

[0010] S2. Use a multi-modal data fusion model to preprocess the data set, extract key fault feature data, and mark the existing fault areas;

[0011] S3. Based on bionic thermal perception technology, perform focused analysis on the marked fault areas, simulate the sensitive perception of cold-blooded animals to heat sources, and identify the positions of thermal anomaly points through a dynamic sensitivity adjustment mechanism;

[0012] S4. Use the key fault feature data and the positions of thermal anomaly points to construct the electrical topology diagram of the photovoltaic module, analyze the electrical connection relationship and the fault propagation path by combining the graph neural network algorithm, and generate the priority of the fault points;

[0013] S5. Based on the key fault feature data, the positions of thermal anomaly points and the priority of the fault points, use the ant colony optimization algorithm to dynamically adjust the inspection path and mark the positions of the fault points;

[0014] S6. Display the inspection path, the positions of the fault points and the repair suggestions in the AR device, and overlay them on the actual scene in the form of augmented reality to generate a heat map representing the risk level;

[0015] S7. Generate an inspection report containing fault information, inspection path, fault propagation analysis and optimization suggestions, and feedback the inspection results to the system model.

[0016] Optionally, the S2 specifically includes:

[0017] S21. Receive the generated dataset, including panoramic images, thermal distribution data and environmental parameters, and perform integrity checks on the dataset to remove missing and abnormal data points;

[0018] S22. Perform standardization processing on the valid data, including unifying the resolution and denoising the panoramic images, normalizing the thermal distribution data, converting the units of the environmental parameters and removing outliers, to generate a standardized dataset;

[0019] S23. Use the feature extraction algorithm to analyze the standardized dataset, extract the texture features of the panoramic images, the temperature gradient features in the thermal distribution data, and the change trend features in the environmental parameters respectively, and construct a preliminary feature vector:

[0020]

[0021] where, T(i,j) is the texture feature of the panoramic image, M and N are the width and height of the image, I(x,y) is the pixel gray value of the image, f(I(x,y),i,j) is the statistical relationship of the pixel gray value, x and y are the pixel positions, i and j are the indexes of the pixel gray levels in the image, G(x,y) is the temperature gradient value, is the change rate of the temperature in the x direction, is the change rate of the temperature in the y direction, E k is the change rate of the environmental parameter, ΔP k is the change value of the environmental parameter, and Δt is the measurement time interval;

[0022] S24. Input the preliminary feature vectors into the multi-modal data fusion model, strengthen the weights of fault-related features through the adaptive attention mechanism, fuse the multi-modal feature data, and generate key fault feature data:

[0023]

[0024] Among them, F fuse is the key fault feature data, F k is the feature vector of the k-th modality, w k is the weight vector corresponding to the k-th modality feature, F j is the feature vector of the j-th modality, w j is the weight vector corresponding to the j-th modality feature, n is the total number of multi-modal features, and k is the index of the modality;

[0025] S25. Based on the key fault feature data, dynamically calculate the cluster centers through the clustering algorithm, and assign the data points to the corresponding clusters according to the distance from the cluster centers, identify the fault points and mark the fault areas:

[0026]

[0027] Among them, Cluster(x) is the cluster number to which the data point x belongs, is to select the cluster number j that minimizes the objective function, |C j | is the number of data points included in the j-th cluster, and x' is the set of data points belonging to the j-th cluster;

[0028] S26. Combine the fault area marking results with the original data for verification, and store the final fault marking results as structured data.

[0029] Optionally, the S3 specifically includes:

[0030] S31. Receive the marked fault area data, including the preliminary fault area location and thermal distribution characteristics of the photovoltaic module, extract the thermal distribution map from the thermal imaging sensor, and form a basic data set;

[0031] S32. Based on the thermal distribution map, simulate the sensitive perception mechanism of cold-blooded animals to heat sources, establish a bionic thermal perception model, and initially screen out potential thermal anomaly points:

[0032]

[0033] Among them, S(x,y) is the significance score, T(x,y) is the temperature value of the point (x,y) in the thermal distribution map, T bg is the background temperature, σ T is the standard deviation of the background temperature, is the temperature gradient, ε is the weight factor, P(x, y) indicates whether the point (x, y) is a potential abnormal point, and S thresh is the threshold of the significance score;

[0034] S33. Using the selected potential thermal abnormal points, combined with the temperature distribution characteristics in the target area, dynamically adjust the sensitivity range, optimize the detection threshold, eliminate non-significant abnormal points, and classify them into point-like anomalies and line-like anomalies:

[0035] S34. Conduct in-depth analysis on point-like anomalies and line-like anomalies. Combining the spatial continuity characteristics of the thermal distribution in the area, process point-like anomalies and line-like anomalies separately, and screen out significant thermal abnormal points:

[0036]

[0037] Among them, S point (x, y) is the significance score of the point-like anomaly, T(x, y) is the temperature value of the point (x, y) in the thermal distribution map, max{T neighbors (x, y)} is the maximum value of all temperature values in the neighborhood around the point (x, y), and σ T is the standard deviation of the global temperature in the thermal distribution map, and S line (L) is the significance score of the line-like anomaly area L, Continuity(x, y) is the spatial continuity of the point (x, y), dA is the area of the integration region L, Length(L) is the total length of the line-like anomaly area, P(x, y) is the formula for screening significant thermal abnormal points, and θ point is the significance score threshold of the point-like anomaly, and θ line is the significance score threshold of the line-like anomaly;

[0038] S35. Compare and verify the selected significant thermal abnormal points with the thermal distribution map, eliminate the false abnormal points caused by environmental interference or sensor noise, and generate a list of thermal abnormal points;

[0039] S36. According to the generated list of thermal abnormal points, obtain the position coordinates of each thermal abnormal point and output them in a structured form.

[0040] Optionally, the specific steps of S4 include:

[0041] S41. Extract the electrical characteristics and the location information of the thermal abnormal points from the key fault feature data, identify the photovoltaic modules, and obtain the electrical connection relationship between the photovoltaic modules as the initial connection data;

[0042] S42. Regard the photovoltaic modules as the nodes of the graph and the electrical connection relationship as the edges to construct the electrical topology graph of the photovoltaic modules. Attach the electrical characteristics and the feature data of the thermal abnormal points to each node to form the topology graph structure;

[0043] S43. Normalize the electrical topology diagram of the photovoltaic module, including normalizing node features, initializing the weights of electrical connection edges, and removing isolated nodes and invalid edges;

[0044] S44. Input the normalized electrical topology diagram of the photovoltaic module into the graph neural network model, design a message passing mechanism between nodes, simulate the propagation of electrical characteristics among photovoltaic modules, extract the global representation features of nodes, and generate the fault propagation information of each node:

[0045]

[0046] Among them, is the feature vector of node i at the t-th layer, is the fault propagation information of node i, N(i) is the set of neighbor nodes of node i, is the message source data passed from neighbor node j to node i, w ij is the strength of the electrical connection relationship, W is a parameter in the graph neural network, and τ is an activation function;

[0047] S45. Based on the fault propagation information of nodes, analyze the electrical connection relationship and fault propagation path of the photovoltaic modules in the topology diagram, and generate the fault propagation probability distribution of nodes:

[0048]

[0049] Among them, is the fault propagation probability of the node, W out is a learnable linear transformation matrix, and softmax is a normalization function;

[0050] S46. Use the propagation probability distribution of nodes, combined with the thermal anomaly significance score and electrical characteristic anomaly score of nodes, to calculate the fault point priority:

[0051]

[0052] Among them, P i is the fault point priority of node i, S heat (i) is the thermal anomaly significance score of node i, S elec (i) is the electrical characteristic anomaly score of node i, and α, β, γ are weight parameters.

[0053] Optionally, the specific content of S5 includes:

[0054] S51. Extract key fault feature data, the location of thermal anomaly points, and the fault point priority, generate a path planning data set, and generate an initial path according to the candidate node set, combined with the dynamic weight adjustment mechanism and heuristic rules:

[0055]

[0056] Among them, Path initial is the dynamically generated initial path, d ij is the distance between node i and node j, W j (t) is the dynamic priority weight of node j, R j is the regional risk factor of node j, and E is the set of edges in the path;

[0057] S52. Initialize the ant colony optimization algorithm based on the path planning dataset and the initial path, set the pheromone distribution of the path, allocate the pheromone concentration, define the heuristic factor according to the priority of the candidate nodes and the weight of the high-risk area, and generate a set of candidate nodes through the path planning dataset:

[0058] N i = {j | (i, j) ∈ E, d ij ≤ d max , W j (t) ≥ W threshold , j ∈ (Path initial ∪ Adj(i))};

[0059] Among them, N i is the set of candidate nodes, d max is the distance threshold, W threshold is the priority threshold, and Adj(i) is the set of adjacent nodes of node i;

[0060] S53. Based on the pheromone concentration, the heuristic factor, and the set of candidate nodes, according to the path selection mechanism of the ant colony optimization algorithm, combined with the position of the thermal anomaly point and the priority of the fault point, dynamically generate the inspection path:

[0061]

[0062] Among them, Path ant (t) is the inspection path, Path is the set of edges in the path, η ij and η ik are the heuristic factors, τ ij (t) is the pheromone concentration from node i to node j, τ ik (t) is the pheromone concentration from node i to node k, α is the influence weight of the pheromone concentration in path selection, and β is the influence weight of the heuristic factor in path selection;

[0063] S54. Compare the inspection path with the initial path, mark the position of the fault point, and dynamically adjust the path planning parameters according to the improvement of the initial path by the inspection path:

[0064] S55. Update the pheromone concentration in the inspection path according to the fault point information and inspection feedback, adjust the heuristic factor and node priority, and through multiple iterations, gradually improve the coverage efficiency of the inspection path for key nodes and dynamically adjust the inspection path;

[0065] S56. After multiple iterations, determine the optimal inspection path according to the pheromone concentration and path performance indicators, mark the final fault point location, and output the optimized path result in a structured form.

[0066] Optionally, the specific steps of S6 are as follows:

[0067] S61. Obtain the inspection path, including the fault point location, fault type, and repair suggestion data, and store the data in a structured manner;

[0068] S62. Load the 3D model of the photovoltaic module and the geospatial information, map the inspection path and fault point data to the actual geographical scene, and align the virtual content with the actual scene through coordinate matching technology;

[0069] S63. Based on the actual geographical scene and 3D model data, generate an augmented reality display layer for the inspection path, dynamically identify the inspection path using arrows or line segments, and use high - light marking and additional text information display for the fault points;

[0070] S64. Combine the displayed fault point information and overlay the repair suggestions, including the recommended repair solutions, necessary tools and materials, and estimated repair time;

[0071] S65. Use the fault point data and inspection path coverage data to generate a heat map representing the risk level, and map the risk level with the color gradient of the heat map.

[0072] An AR intelligent photovoltaic inspection system according to an embodiment of the present invention includes the following modules:

[0073] A data acquisition and processing module, which is used to collect and process the key data generated during the inspection process;

[0074] A three - dimensional scene modeling module, which is used to generate a 3D model of the photovoltaic module and align the inspection data with the actual scene;

[0075] A visualization generation module, which is used to dynamically generate the inspection path, fault points, and a heat map representing the risk level;

[0076] A repair suggestion module, which is used to provide repair suggestions for each fault point;

[0077] A dynamic update and feedback module, which is used to dynamically adjust the displayed content and record the feedback data generated during the inspection process;

[0078] An augmented reality display module for presenting inspection routes, fault points, heat maps indicating risk levels, and repair suggestions on AR devices.

[0079] The beneficial effects of the present invention are as follows:

[0080] By integrating multi-modal data fusion, dynamic path optimization, and augmented reality technologies, the present invention significantly improves the efficiency and intelligence level of photovoltaic power plant inspections, and solves the technical problems of low inspection efficiency, inaccurate fault location, unintuitive risk display, and insufficient decision-making support existing in the prior art. By superimposing the inspection route, fault point location, and repair suggestions onto the actual scene, the present invention realizes the dynamic visualization display of inspection results, enabling users to intuitively understand the inspection coverage, fault point priorities, and risk distributions. Compared with the display forms of traditional static reports or two-dimensional charts, the application of augmented reality technology not only enhances the intuitiveness of data display but also closely combines the inspection results with the actual layout of the photovoltaic power plant, facilitating users to quickly locate problem components and high-risk areas.

[0081] In addition, the dynamic path optimization algorithm of the present invention effectively improves the flexibility of inspection route planning. By preferentially covering high-risk areas and key nodes, it realizes the optimal allocation of inspection resources. At the same time, combined with the generated risk distribution heat map, users can clearly understand the uninspected areas and potential risks, providing important references for subsequent inspections and maintenance. The superimposed display of repair suggestions further improves the decision-making support function of the system, providing specific repair plans, required tools, and estimated repair times to help users quickly formulate scientific repair plans.

[0082] The present invention also supports users to switch inspection views and repair suggestions using gestures or voices through a dynamic interaction function, enhancing the operation flexibility and user experience of the system. The real-time updated display content can be dynamically adjusted according to inspection results and user feedback, forming a closed-loop optimization mechanism, making the system perform more precisely, efficiently, and intelligently during the inspection process. The present invention ultimately realizes the full-process optimization of photovoltaic power plant inspection data from collection, processing, visualization to decision-making support, significantly improving the operation and maintenance efficiency and economic benefits of photovoltaic power plants. Description of the Drawings

[0083] The drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0084] Figure 1 is a flowchart of an AR intelligent photovoltaic inspection method proposed by the present invention;

[0085] Figure 2 is a structural schematic diagram of an AR intelligent photovoltaic inspection system proposed by the present invention. Detailed implementation mode

[0086] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0087] Reference Figure 1 , an AR intelligent photovoltaic inspection method, comprising the following steps:

[0088] S1. Obtain a panoramic visible light image of the photovoltaic module through a bionic compound eye camera, and combine with a thermal imaging sensor to obtain the thermal distribution data of the photovoltaic module, and generate a data set;

[0089] S2. Use a multi-modal data fusion model to preprocess the data set, extract key fault feature data, and mark the existing fault areas;

[0090] S3. Based on the bionic thermal perception technology, perform focused analysis on the marked fault areas, simulate the sensitive perception of cold-blooded animals to heat sources, and identify the positions of thermal anomaly points through a dynamic sensitivity adjustment mechanism;

[0091] S4. Use the key fault feature data and the positions of thermal anomaly points to construct an electrical topology diagram of the photovoltaic module, combine with the graph neural network algorithm to analyze the electrical connection relationship and the fault propagation path, and generate the priority of the fault points;

[0092] S5. Based on the key fault feature data, the positions of thermal anomaly points and the priority of the fault points, use the ant colony optimization algorithm to dynamically adjust the inspection path and mark the positions of the fault points;

[0093] S6. Display the inspection path, the positions of the fault points and the repair suggestions in the AR device, and overlay them on the actual scene in the form of augmented reality to generate a heat map representing the risk level;

[0094] S7. Generate an inspection report including fault information, inspection path, fault propagation analysis and optimization suggestions, and feedback the inspection results to the system model.

[0095] In this implementation mode, the specific content of S2 includes:

[0096] S21. Receive the generated data set, including panoramic images, thermal distribution data and environmental parameters, and perform integrity checks on the data set to remove missing and abnormal data points;

[0097] S22. Perform standardization processing on the valid data, including unifying the resolution and denoising the panoramic images, normalizing the thermal distribution data, converting the units of the environmental parameters and removing outliers, and generating a standardized data set;

[0098] S23. Analyze the standardized data set using feature extraction algorithms, extract the texture features of the panoramic image, the temperature gradient features in the thermal distribution data, and the change trend features in the environmental parameters respectively, and construct a preliminary feature vector:

[0099]

[0100] Among them, T(i,j) is the texture feature of the panoramic image, M and N are the width and height of the image, I(x,y) is the pixel gray value of the image, f(I(x,y),i,j) is the statistical relationship of the pixel gray value, x and y are the pixel positions, i and j are the indexes of the pixel gray levels in the image, G(x,y) is the temperature gradient value, is the change rate of temperature in the x direction, is the change rate of temperature in the y direction, E k is the change rate of the environmental parameter, ΔP k is the change value of the environmental parameter, Δt is the measurement time interval;

[0101] S24. Input the preliminary feature vector into the multi-modal data fusion model, strengthen the weights of the fault-related features through the adaptive attention mechanism, fuse the multi-modal feature data, and generate the key fault feature data:

[0102]

[0103] Among them, F fuse is the key fault feature data, F k is the feature vector of the k-th modality, w k is the weight vector corresponding to the k-th modality feature, F j is the feature vector of the j-th modality, w j is the weight vector corresponding to the j-th modality feature, n is the total number of multi-modal features, and k is the index of the modality;

[0104] S25. Based on the key fault feature data, dynamically calculate the cluster centers through the clustering algorithm, and assign the data points to the corresponding clusters according to the distance from the cluster centers, identify the fault points and mark the fault areas:

[0105]

[0106] Among them, Cluster(x) is the cluster number to which the data point x belongs, is to select the cluster number j that minimizes the objective function, |C j | is the number of data points included in the j-th cluster, and x' is the set of data points belonging to the j-th cluster;

[0107] S26. Verify by combining the fault area marking result with the original data, and store the final fault marking result as structured data.

[0108] In this embodiment, the S3 specifically includes:

[0109] S31. Receive the marked fault area data, including the preliminary fault area position and thermal distribution characteristics of the photovoltaic module, extract the thermal distribution map from the thermal imaging sensor, and form a basic data set;

[0110] S32. Based on the thermal distribution map, simulate the sensitive perception mechanism of cold-blooded animals to heat sources, establish a bionic thermal perception model, and preliminarily screen out potential thermal abnormal points:

[0111]

[0112] Among them, S(x,y) is the significance score, T(x,y) is the temperature value of the point (x,y) in the thermal distribution map, T bg is the background temperature, σ T is the standard deviation of the background temperature, is the temperature gradient, ε is the weight factor, P(x,y) is whether the point (x,y) is a potential abnormal point, S thresh is the threshold of the significance score;

[0113] S33. Use the screened potential thermal abnormal points, combine the temperature distribution characteristics in the target area, dynamically adjust the sensitivity range, optimize the detection threshold, eliminate non-significant abnormal points, and classify them into dot-shaped anomalies and linear anomalies:

[0114] S34. Conduct in-depth analysis on dot-shaped anomalies and linear anomalies, combine the spatial continuity characteristics of the thermal distribution in the area, process dot-shaped anomalies and linear anomalies separately, and screen out significant thermal abnormal points:

[0115]

[0116] Among them, S point (x,y) is the significance score of the dot-shaped anomaly, T(x,y) is the temperature value of the point (x,y) in the thermal distribution map, max{T neighbors (x,y)} is the maximum value of all temperature values in the neighborhood around the point (x,y), σ T is the standard deviation of the global temperature in the thermal distribution map, S line (L) is the significance score of the linear anomaly area L, Continuity(x,y) is the spatial continuity of the point (x,y), dA is the area of the integration area L, Length(L) is the total length of the linear anomaly area, P(x,y) is the significant thermal abnormal point screening formula, θ point is the significance score threshold of the dot-shaped anomaly, θline is the significance score threshold for linear anomalies;

[0117] S35. Compare and verify the selected significant thermal anomaly points with the thermal distribution map, eliminate the false anomaly points caused by environmental interference or sensor noise, and generate a list of thermal anomaly points;

[0118] S36. According to the generated list of thermal anomaly points, obtain the position coordinates of each thermal anomaly point and output them in a structured form.

[0119] In this embodiment, the specific steps of S4 include:

[0120] S41. Extract the electrical characteristics and the position information of the thermal anomaly points from the key fault feature data, identify the photovoltaic modules, and obtain the electrical connection relationship between the photovoltaic modules as the initial connection data;

[0121] S42. Regard the photovoltaic modules as the nodes of the graph and the electrical connection relationship as the edges, construct the electrical topology graph of the photovoltaic modules, and attach the electrical characteristics and the feature data of the thermal anomaly points to each node to form the topology graph structure;

[0122] S43. Perform normalization processing on the electrical topology graph of the photovoltaic modules, including normalizing the node features, initializing the weights of the electrical connection edges, and removing the isolated nodes and invalid edges;

[0123] S44. Input the normalized electrical topology graph of the photovoltaic modules into the graph neural network model, design the message passing mechanism between the nodes, simulate the propagation of the electrical characteristics between the photovoltaic modules, extract the global representation features of the nodes, and generate the fault propagation information of each node:

[0124]

[0125] Among them, is the feature vector of node i at the t-th layer, is the fault propagation information of node i, N(i) is the set of neighbor nodes of node i, is the message source data passed from neighbor node j to node i, w ij is the strength of the electrical connection relationship, W is the parameter in the graph neural network, and τ is the activation function;

[0126] S45. Based on the fault propagation information of the nodes, analyze the electrical connection relationship and the fault propagation path of the photovoltaic modules in the topology graph, and generate the fault propagation probability distribution of the nodes:

[0127]

[0128] Among them, is the fault propagation probability of the node, W outis a learnable linear transformation matrix, and softmax is a normalization function;

[0129] S46. Using the propagation probability distribution of nodes, combined with the thermal anomaly significance score and electrical characteristic anomaly score of nodes, calculate the fault point priority:

[0130]

[0131] Among them, P i is the fault point priority of node i, S heat (i) is the thermal anomaly significance score of node i, S elec (i) is the electrical characteristic anomaly score of node i, and α, β, γ are weight parameters.

[0132] In this embodiment, the S5 specifically includes:

[0133] S51. Extract key fault feature data, thermal anomaly point locations, and fault point priorities to generate a path planning dataset. According to the candidate node set, combined with the dynamic weight adjustment mechanism and heuristic rules, generate an initial path:

[0134]

[0135] Among them, Path initial is the dynamically generated initial path, d ij is the distance between node i and node j, W j (t) is the dynamic priority weight of node j, R j is the regional risk factor of node j, and E is the set of edges in the path;

[0136] S52. Based on the path planning dataset and the initial path, initialize the ant colony optimization algorithm, set the pheromone distribution of the path, allocate pheromone concentration, define the heuristic factor according to the priority of the candidate nodes and the high-risk area weight, and generate a candidate node set through the path planning dataset:

[0137] N i ={j∣(i,j)∈E,d ij ≤d max ,W j (t)≥W threshold ,j∈(Path initial ∪Adj(i))};

[0138] Among them, N i is the candidate node set, d max is the distance threshold, W threshold is the priority threshold, and Adj(i) is the set of adjacent nodes of node i;

[0139] S53. Dynamically generate an inspection path based on the pheromone concentration, heuristic factor, and candidate node set, in accordance with the path selection mechanism of the ant colony optimization algorithm, and combining the position of the thermal anomaly point and the priority of the fault point:

[0140]

[0141] Among them, Path ant (t) is the inspection path, Path is the set of edges in the path, η ij and η ik are heuristic factors, τ ij (t) is the pheromone concentration from node i to node j, τ ik (t) is the pheromone concentration from node i to node k, α is the influence weight of the pheromone concentration in path selection, and β is the influence weight of the heuristic factor in path selection;

[0142] S54. Compare the inspection path with the initial path, mark the position of the fault point, and dynamically adjust the path planning parameters according to the improvement of the initial path by the inspection path:

[0143] S55. Update the pheromone concentration in the inspection path according to the fault point information and inspection feedback, adjust the heuristic factor and node priority, and gradually improve the coverage efficiency of the inspection path for key nodes through multiple iterations, and dynamically adjust the inspection path;

[0144] S56. After multiple iterations, determine the optimal inspection path according to the pheromone concentration and path performance indicators, mark the final position of the fault point, and output the optimized path result in a structured form.

[0145] In this embodiment, the specific steps of S6 are as follows:

[0146] S61. Obtain the inspection path, including the position of the fault point, the type of the fault, and the repair suggestion data, and store the data in a structured manner;

[0147] S62. Load the three-dimensional model and geospatial information of the photovoltaic module, map the inspection path and fault point data to the actual geographical scene, and align the virtual content with the actual scene through coordinate matching technology;

[0148] S63. Generate an augmented reality display layer of the inspection path based on the actual geographical scene and three-dimensional model data, dynamically identify the inspection path using arrows or line segments, and use highlighter markings and additional text information to display the fault points;

[0149] S64. Combine the displayed fault point information and overlay the repair suggestions, including the recommended repair plan, necessary tools and materials, and estimated repair time;

[0150] S65. Generate a heat map representing the risk level using the fault point data and the inspection path coverage data, and map the risk levels with the heat map color gradient.

[0151] Reference Figure 2 , an AR intelligent photovoltaic inspection system, including the following modules:

[0152] Data acquisition and processing module, used to collect and process the key data generated during the inspection process;

[0153] Three-dimensional scene modeling module, used to generate a three-dimensional model of the photovoltaic modules and align the inspection data with the actual scene;

[0154] Visualization generation module, used to dynamically generate the inspection path, fault points, and a heat map representing the risk level;

[0155] Repair suggestion module, used to provide repair suggestions for each fault point;

[0156] Dynamic update and feedback module, used to dynamically adjust the display content and record the feedback data generated during the inspection process;

[0157] Augmented reality display module, used to display the inspection path, fault points, a heat map representing the risk level, and repair suggestions on an AR device.

[0158] Example 1:

[0159] To verify the feasibility of the present invention in implementation, the present invention is applied to a large-scale photovoltaic power station. The power station covers an area of 500 mu and contains 15,000 photovoltaic modules. The terrain is complex, the module layout is dense and unevenly distributed, and it is easily affected by environmental factors, resulting in performance degradation and frequent failures. Traditional manual inspections require about 15 person-times and take more than 5 days. Not only is the efficiency low, but there are also problems such as insufficient coverage and inaccurate fault location. Although drone inspections can cover a large area, they mainly rely on single thermal imaging analysis, cannot provide multi-modal data support, and lack dynamic path optimization capabilities and risk display functions.

[0160] In this experiment, a photovoltaic inspection result visualization system based on the present invention was deployed. The system includes a data acquisition and processing module, a three-dimensional scene modeling module, a visualization generation module, a repair suggestion module, a dynamic update and feedback module, and an augmented reality display module. During the experiment, drones equipped with thermal imaging and electrical sensors were used to collect inspection data, and AR devices were used to assist in display and interaction.

[0161] At the beginning of the experiment, the drone conducted a preliminary inspection according to the preset path, collecting thermal anomaly points, electrical anomaly points, and environmental data. Combining with the 3D scene modeling module, the system mapped the inspection path and the location of the fault points to the power station's geographical space and generated a heat map of the priority and risk distribution of key nodes. The operation and maintenance personnel could intuitively view the inspection path, fault points, and high-risk areas through the AR device and obtain repair suggestions. The system supported users to adjust the inspection plan or switch the display of fault point information through the interaction function.

[0162] During the experiment, the system discovered a total of 112 fault points, including 18 high-priority fault points, 40 medium-priority fault points, and 54 low-priority fault points. The risk distribution heat map showed that the high-risk areas were concentrated in the southeast and central regions of the power station. The system recommended that the operation and maintenance personnel give priority to dealing with high-priority fault points and dynamically optimize the path according to the inspection coverage to reduce the time of missed areas and repeated inspections. The repair suggestion module provided detailed repair guidance for each fault point, including the required tools and materials and the estimated repair time. Through the AR device, the operation and maintenance personnel could view the inspection results and repair guidelines in real time, greatly improving the inspection efficiency and repair effect.

[0163] Table 1 Comparison between Traditional Inspection and the Method of the Present Invention

[0164]

[0165] Table 2 Analysis of Fault Point Priority and Risk Distribution Heat Map

[0166]

[0167] As can be seen from Table 1, the photovoltaic inspection method of the present invention performs excellently in terms of efficiency and accuracy compared with the traditional method. Manual inspection took 120 hours and the coverage rate was only 85%, and the fault marking accuracy was 70%; although the coverage rate of drone inspection was increased to 95%, the marking accuracy was only 80%. In contrast, the method of the present invention reduced the time to 12 hours, the coverage rate reached 100%, the fault marking accuracy was increased to 98%, and the priority classification efficiency was even up to 20 fault points per hour, significantly higher than the traditional method.

[0168] Table 2 further demonstrates the advantages of the method of the present invention in fault point priority classification and risk distribution heat map generation. The average repair time for high-priority fault points was only 25 minutes and the coverage rate reached 100%; medium- and low-priority fault points were repaired in 40 minutes and 50 minutes respectively, and the coverage rates were 95% and 90% respectively. These data indicate that the present invention can accurately identify the priority of fault points and optimize resource allocation, and at the same time clarify high-risk areas through the dynamic risk heat map, improving the inspection and repair efficiency.

[0169] In summary, the method of the present invention not only greatly improves the efficiency of photovoltaic inspection, but also significantly improves the accuracy of fault point marking and the classification effect, especially showing obvious advantages in risk distribution analysis and inspection path optimization. Data shows that the method of the present invention provides a new intelligent solution for the efficient operation and maintenance of photovoltaic power stations, with important application value and promotion prospects.

[0170] As described above, only the preferred specific embodiments of the present invention are given, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. An AR intelligent photovoltaic inspection method, characterized in that: The steps include: S1. Obtain panoramic visible light images of photovoltaic modules through a bionic compound eye camera, and obtain thermal distribution data of photovoltaic modules in combination with a thermal imaging sensor to generate a data set; S2. Use the multimodal data fusion model to preprocess the data set, extract key fault feature data, and mark the existing fault areas; S3. Based on the bionic thermal sensing technology, the marked fault area is analyzed in a focused manner, simulating the sensitive perception of cold-blooded animals to heat sources, and the location of thermal anomalies is identified through a dynamic sensitivity adjustment mechanism; S4. Use key fault feature data and thermal anomaly point locations to build an electrical topology diagram of the photovoltaic module, combine the graph neural network algorithm to analyze the electrical connection relationship and fault propagation path, and generate fault point priority; S5. Based on key fault feature data, thermal anomaly point locations, and fault point priorities, the inspection path is dynamically adjusted using the ant colony optimization algorithm to mark the fault point locations; S6. Display the inspection path, fault location and repair suggestions in the AR device, and superimpose them on the actual scene in the form of augmented reality to generate a heat map indicating the degree of risk; S7. Generate an inspection report including fault information, inspection path, fault propagation analysis and optimization suggestions, and feed back the inspection results to the system model.

2. The AR intelligent photovoltaic inspection method according to claim 1, characterized in that: The S2 specifically includes: S21, receiving the generated data set, including the panoramic image, thermal distribution data and environmental parameters, and performing integrity check on the data set to remove missing and abnormal data points; S22, standardizing the valid data, including unifying the resolution and removing noise of the panoramic image, normalizing the thermal distribution data, converting the units of the environmental parameters and removing outliers, and generating a standardized data set; S23. Analyze the standardized data set using a feature extraction algorithm to extract the texture features of the panoramic image, the temperature gradient features in the thermal distribution data, and the change trend features in the environmental parameters, and construct a preliminary feature vector: Where T(i, j) is the texture feature of the panoramic image, M and N are the width and height of the image, I(x, y) is the grayscale value of the image pixel, f(I(x, y), i, j) is the statistical relationship of the pixel grayscale, x and y are the pixel positions, i and j are the indexes of the pixel grayscale levels in the image, G(x, y) is the temperature gradient value, is the rate of change of temperature in the x direction, is the rate of change of temperature in the y direction, E k is the rate of change of environmental parameters, ΔP k is the change value of the environmental parameter, Δt is the measurement time interval; S24. Input the preliminary feature vector into the multimodal data fusion model, strengthen the fault-related feature weights through the adaptive attention mechanism, fuse the multimodal feature data, and generate key fault feature data: Among them, F fuse is the key fault characteristic data, F k is the eigenvector of the kth mode, w k is the weight vector corresponding to the kth modal feature, F j is the eigenvector of the jth mode, w j is the weight vector corresponding to the jth modal feature, n is the total number of multimodal features, and k is the index of the modality; S25. Based on the key fault feature data, the cluster center is dynamically calculated through the clustering algorithm, and the data points are assigned to the corresponding clusters according to the distance from the cluster center, the fault points are identified and the fault areas are marked: Among them, Cluster(x) is the cluster number to which the data point x belongs. To select the cluster number j that minimizes the objective function, |C j | is the number of data points contained in the jth cluster, and x′ is the set of data points belonging to the jth cluster; S26. Verify the fault area marking result in combination with the original data, and store the final fault marking result as structured data.

3. The AR intelligent photovoltaic inspection method according to claim 1, characterized in that: The S3 specifically includes: S31, receiving the marked fault area data, including the preliminary fault area location and thermal distribution characteristics of the photovoltaic module, extracting the thermal distribution map from the thermal imaging sensor, and forming a basic data set; S32. Based on the heat distribution map, simulate the sensitive perception mechanism of cold-blooded animals to heat sources, establish a bionic thermal perception model, and preliminarily screen out potential thermal anomalies: Among them, S(x, y) is the significance score, T(x, y) is the temperature value of the point (x, y) in the heat distribution map, and T bg is the background temperature, σ T is the standard deviation of the background temperature, is the temperature gradient, ε is the weight factor, P(x, y) is whether the point (x, y) is a potential abnormal point, S thresh The threshold for the significance score; S33. Using the potential thermal anomaly points screened out, combined with the temperature distribution characteristics in the target area, dynamically adjust the sensitivity range, optimize the detection threshold, remove non-significant anomalies, and classify them into point anomalies and line anomalies: S34. Conduct in-depth analysis on point anomalies and line anomalies, and process them separately based on the spatial continuity characteristics of thermal distribution in the region to screen out significant thermal anomaly points: Among them, S point (x, y) is the significance score of the point anomaly, T(x, y) is the temperature value of the point (x, y) in the thermal distribution map, max{T neighbors (x, y)} is the maximum value of all temperature values ​​in the neighborhood around point (x, y), σ T is the standard deviation of the global temperature in the heat distribution map, S line (L) is the significance score of the linear anomaly region L, Continuity(x, y) is the spatial continuity of the point (x, y), dA is the area of ​​the integral region L, Length(L) is the total length of the linear anomaly region, P(x, y) is the screening formula for significant thermal anomaly points, θ point is the significance score threshold of point anomalies, θ line is the significance score threshold of linear anomalies; S35, comparing and verifying the selected significant thermal anomaly points with the thermal distribution map, eliminating false anomaly points caused by environmental interference or sensor noise, and generating a list of thermal anomaly points; S36. According to the generated thermal anomaly point list, the position coordinates of each thermal anomaly point are obtained and output in a structured form.

4. The AR intelligent photovoltaic inspection method according to claim 1, characterized in that: The S4 specifically includes: S41, extracting electrical characteristics and thermal anomaly point location information from key fault feature data, identifying photovoltaic modules, and obtaining electrical connection relationships between photovoltaic modules as initial connection data; S42, considering the photovoltaic modules as nodes of the graph and the electrical connection relationships as edges, constructing an electrical topology graph of the photovoltaic modules, and attaching electrical characteristics and thermal anomaly point feature data to each node to form a topology graph structure; S43, normalizing the electrical topology of the photovoltaic module, including normalizing node features, initializing electrical connection edge weights, and removing isolated nodes and invalid edges; S44. Input the normalized PV module electrical topology diagram into the graph neural network model, design the message transmission mechanism between nodes, simulate the propagation of electrical characteristics between PV modules, extract the global characterization features of the nodes, and generate the fault propagation information of each node: in, is the feature vector of node i in layer t, is the fault propagation information of node i, N(i) is the set of neighbor nodes of node i, is the source data of the message transmitted by neighbor node j to node i, W ij is the strength of the electrical connection relationship, W is the parameter in the graph neural network, and τ is the activation function; S45. Based on the fault propagation information of the node, the electrical connection relationship and fault propagation path of the photovoltaic components in the topology diagram are analyzed to generate the fault propagation probability distribution of the node: in, is the fault propagation probability of the node, W out is a learnable linear transformation matrix, and softmax is a normalization function; S46. Calculate the fault point priority by using the propagation probability distribution of the node, combined with the thermal anomaly significance score and electrical characteristic anomaly score of the node: Among them, P i is the fault point priority of node i, S heat (i) is the thermal anomaly significance score of node i, S elec (i) is the electrical characteristic anomaly score of node i, and α, β, and γ are weight parameters.

5. The AR intelligent photovoltaic inspection method according to claim 1, characterized in that: The S5 specifically includes: S51, extract key fault feature data, thermal anomaly point location and fault point priority, generate a path planning data set, and generate an initial path based on the candidate node set, combined with a dynamic weight adjustment mechanism and heuristic rules: Among them, Path initial is the dynamically generated initial path, d ij is the distance between node i and node j, W j (t) is the dynamic priority weight of node j, R j is the regional risk factor of node j, and E is the set of edges in the path; S52. Based on the path planning data set and the initial path, the ant colony optimization algorithm is initialized, the pheromone distribution of the path is set, the pheromone concentration is allocated, the heuristic factor is defined according to the priority of the candidate node and the weight of the high-risk area, and the candidate node set is generated through the path planning data set: N i ={j|(i,j)∈E,d ij ≤d max ,W j (t)≥W threshold ,j∈(Path initial ∪Adj(i))}; Among them, N i is the candidate node set, d max is the distance threshold, W threshold is the priority threshold, Adj(i) is the set of adjacent nodes of node i; S53, based on pheromone concentration, heuristic factors and candidate node set, according to the path selection mechanism of the ant colony optimization algorithm, combined with the location of thermal anomaly points and the priority of fault points, dynamically generate inspection paths: Among them, Path ant (t) is the inspection path, Path is the edge set in the path, η ij and η ik is the heuristic factor, τ ij (t) is the pheromone concentration from node i to node j, τ ik (t) is the pheromone concentration from node i to node k, α is the influence weight of pheromone concentration in path selection, and β is the influence weight of heuristic factor in path selection; S54, compare the inspection path with the initial path, mark the location of the fault point, and dynamically adjust the path planning parameters according to the improvement of the inspection path to the initial path: S55. Based on the fault point information and inspection feedback, update the pheromone concentration in the inspection path, adjust the heuristic factor and node priority, gradually improve the coverage efficiency of the inspection path for key nodes through multiple iterations, and dynamically adjust the inspection path; S56. After multiple iterations, the optimal inspection path is determined based on the pheromone concentration and the path performance index, the final fault point location is marked, and the optimized path result is output in a structured form.

6. The AR intelligent photovoltaic inspection method according to claim 1, characterized in that: The S6 specifically includes: S61, obtaining the inspection path, including the location of the fault point, the fault type, and the repair suggestion data, and storing the data in a structured manner; S62, loading the three-dimensional model and geographic space information of the photovoltaic module, mapping the inspection path and fault point data to the actual geographic scene, and aligning the virtual content with the actual scene through coordinate matching technology; S63, based on the actual geographic scene and the three-dimensional model data, generate an augmented reality display level of the inspection path, use arrows or line segments to dynamically mark the inspection path, and highlight the fault point and display it with additional text information; S64, combining the displayed fault point information, superimposing repair suggestions, including recommended repair solutions, necessary tools and materials, and estimated repair time; S65. Generate a heat map representing the degree of risk using the fault point data and the inspection path coverage data, where the color gradient of the heat map maps the risk level.

7. An AR intelligent photovoltaic inspection system, an AR intelligent photovoltaic inspection method according to any one of claims 1 to 6, characterized in that: Includes the following modules: Data collection and processing module, used to collect and process key data generated during the inspection process; 3D scene modeling module, used to generate 3D models of PV panels and align inspection data with actual scenes; Visualization generation module, used to dynamically generate inspection paths, fault points and heat maps indicating risk levels; Repair suggestion module, used to provide repair suggestions for each fault point; Dynamic update and feedback module, used to dynamically adjust the display content and record the feedback data generated during the inspection process; The augmented reality display module is used to display the inspection path, fault points, heat maps indicating the degree of risk, and repair suggestions on the AR device.

Images