A flexible photovoltaic support damage identification method based on improved LK optical flow method

Abstract

This invention discloses a method for identifying damage to flexible photovoltaic (PV) supports based on an improved LK optical flow method in the field of PV health monitoring. The method includes: acquiring a pure target video; performing edge detection on the image data in the pure target video to extract an edge skeleton; using the edge skeleton as a constraint for a pre-constructed optical flow optimization objective function, minimizing the objective function to obtain the optimal optical flow displacement component, and acquiring the continuous dynamic displacement time history of key nodes of the support; acquiring the modal parameters of the flexible PV support based on the continuous dynamic displacement time history; calculating the curvature mode and strain mode based on the modal parameters of the flexible PV support; calculating the modal difference between the curvature mode and strain mode and a pre-set health benchmark mode; and determining the damage area based on the abrupt change location of the modal difference. This invention solves the technical problem of low displacement identification accuracy and difficulty in accurately locating PV support damage in existing technologies under complex background interference.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic health monitoring technology, and in particular to a method for identifying damage to flexible photovoltaic supports based on an improved LK optical flow method. Background Technology

[0002] Structural health monitoring of photovoltaic (PV) supports is a core means of ensuring their safety, reliability, and durability. Displacement response, as a key indicator reflecting the dynamic characteristics of PV supports, is of great significance for assessing structural safety status and providing early warnings of potential risks. Traditional contact-based monitoring methods are limited by installation conditions and the interference of additional mass on the structural dynamics. Furthermore, the Global Positioning System (GPS) suffers from low measurement accuracy and sampling frequency, making it difficult to meet the refined monitoring needs of flexible PV supports in complex environments.

[0003] In contrast, non-contact measurement technologies, represented by computer vision, have become a research hotspot for photovoltaic (PV) support health monitoring due to their advantages such as real-time performance, remote sensing, multi-point measurement across the entire field, and high precision. Among visual monitoring methods, displacement extraction based on sub-pixel edge detection is a commonly used technique. This method calculates displacement by identifying the edge features of the PV support and tracking changes in edge position. However, in actual field environments, flexible PV supports often face complex background interference such as water surface reflections, crop fluctuations, and changes in lighting, which can easily generate false edges, leading to a decrease in displacement recognition accuracy and consequently affecting the accuracy of damage identification for the PV support.

[0004] Therefore, how to overcome the inherent limitations of existing visual methods under complex background interference of flexible photovoltaic supports and significantly improve the accuracy and environmental robustness of photovoltaic support damage identification is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying damage to flexible photovoltaic supports based on an improved LK optical flow method. This method can solve the technical problem of low displacement identification accuracy and difficulty in accurately locating damage to photovoltaic supports under complex background interference in the prior art.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] In a first aspect, the present invention provides a method for damage identification of flexible photovoltaic supports based on an improved LK optical flow method, comprising:

[0008] Obtain video of the pre-confirmed key monitoring areas;

[0009] The video of the key monitoring area is input into a pre-constructed fusion segmentation network to obtain a pure target video containing only the target flexible photovoltaic bracket;

[0010] Edge detection is performed on image data in pure target video to extract the edge skeleton of flexible photovoltaic support;

[0011] The edge frame of the flexible photovoltaic support is used as a constraint of the pre-constructed optical flow optimization objective function. The optimal optical flow displacement component is obtained by minimizing the objective function. Based on the optimal optical flow displacement component, the continuous dynamic displacement time history of key nodes on the edge frame is obtained. The key nodes are equally spaced discrete sampling nodes of the corresponding high-damage sensitive area members.

[0012] The Hankel matrix is ​​constructed based on the continuous dynamic displacement time history. The eigenvalues ​​and eigenvectors of the Hankel matrix are calculated using a data-driven random subspace identification method. Based on the eigenvalues ​​and eigenvectors, the modal parameters of the flexible photovoltaic support are obtained. The modal parameters include natural frequency, damping ratio and mode shape.

[0013] Based on the modal parameters of the flexible photovoltaic support, the curvature mode and strain mode are calculated. The modal difference between the curvature mode and strain mode and the preset health benchmark mode is calculated. The damage area is determined according to the abrupt change location of the modal difference. The degree of damage is quantitatively assessed according to the absolute value of the modal difference.

[0014] Furthermore, the key monitoring areas were identified as including:

[0015] The damage sensitivity factor of each element in the pre-constructed finite element model of the flexible photovoltaic support is calculated. The expression for calculating the damage sensitivity factor includes:

[0016] ;

[0017] ;

[0018] ;

[0019] in, Let i be the damage sensitivity factor of the i-th unit. Let be the change in modal strain energy before and after damage to the i-th element. The total modal strain energy under healthy structural conditions. Let be the modal strain energy of the i-th element under the damaged state. Let be the modal strain energy of the i-th element under healthy conditions, and n be the total number of elements in the finite element model of the flexible photovoltaic support.

[0020] Element units with damage sensitivity factors greater than a preset sensitivity factor threshold and mode amplitudes greater than a preset mode amplitude threshold are selected as key monitoring areas.

[0021] Furthermore, the video of the key monitoring area is input into a pre-constructed fusion segmentation network to obtain pure target video containing only the target flexible photovoltaic support, including:

[0022] The convolutional branches of the fusion segmentation network are used to extract local features of the video images in the key monitoring areas. Through multi-scale downsampling and feature pyramid fusion, a local feature map containing detailed information such as support edges and textures is generated.

[0023] The global context features of the images in the video of the key monitoring area are extracted by the Transformer branch of the fusion segmentation network to generate a global feature map containing the overall outline information of the support structure.

[0024] The local feature map is fused with the global feature map to obtain the fused output feature map;

[0025] The resolution of the fused output feature map is restored by upsampling, and the output is a pure target video containing only the target flexible photovoltaic support.

[0026] Furthermore, a data-driven random subspace identification method is used to obtain the eigenvalues ​​and eigenvectors of the Hankel matrix. Based on the eigenvalues ​​and eigenvectors, the modal parameters of the flexible photovoltaic support are obtained, including:

[0027] Singular value decomposition is performed on the Hankel matrix to separate the system signal and noise components, and a reduced-order observation matrix is ​​obtained.

[0028] The system matrix is ​​solved from the reduced-order observation matrix using a data-driven random subspace method. The system matrix is ​​then subjected to eigenvalue decomposition to obtain eigenvalues ​​and eigencomponents. A stable graph is then constructed based on the eigenvalues ​​and eigencomponents.

[0029] Based on the stability diagram, points in the identification results of different orders where the rate of change of frequency, the rate of change of damping ratio, and the rate of change of mode shape are all less than a preset change threshold are marked as stable points. Noise modes and false modes are eliminated to obtain the modal parameters of the flexible photovoltaic support.

[0030] Furthermore, the expression for the optical flow optimization objective function is:

[0031] ;

[0032] in, To optimize the objective function for optical flow, For the tracking area of ​​key nodes, The edge frame of the extracted flexible photovoltaic support. These are the edge constraint weighting coefficients, balancing the optical flow fitting accuracy with the structural constraint strength. Let be the optical flow displacement vector. The normal vector of the edge skeleton. These represent the spatial gradients of the image grayscale in the x and y directions, respectively. This represents the gradient of image grayscale over time. These are the optical flow displacement components of the pixel in the x and y directions, respectively.

[0033] Furthermore, the strain modes are obtained by calculating the mode shape and strain transformation matrix, and the expressions include:

[0034] ;

[0035] ;

[0036] in, For the k-th strain mode, For the k-th mode shape, The strain transformation matrix, The distance between adjacent key nodes;

[0037] The curvature modes are calculated using the second-order central difference of the mode shapes. The expressions include:

[0038] ;

[0039] ;

[0040] in, For the k-th mode at node Curvature mode value at that point, For the k-th mode at node The modal displacement value at that point, The distance between adjacent nodes. For the k-th mode at node Curvature mode difference at that point, These are the measured curvature mode values. This represents the baseline curvature mode value under healthy conditions.

[0041] Furthermore, before constructing the Hankel matrix based on the continuous dynamic displacement time history, the following steps are also included:

[0042] Polynomial fitting is used to remove the trend term in the continuous dynamic displacement time history and eliminate low-frequency drift; Butterworth low-pass filter is used to remove high-frequency noise interference in the continuous dynamic displacement time history and retain the main frequency component of structural vibration.

[0043] Furthermore, using the continuous dynamic displacement time history and modal parameters as the target response, the finite element model is periodically updated to keep the model synchronized with the actual structural state.

[0044] The regression coefficients of the pre-constructed second-order polynomial response surface surrogate model are solved by the least squares method to complete the calibration of the response surface surrogate model, which is used to replace the finite element model for response calculation.

[0045] A pre-built BP neural network is used to optimize and correct the response surface surrogate model, which is used to improve the response calculation accuracy of the second-order polynomial response surface surrogate model.

[0046] The optimization objective is to minimize the residual between the measured target response and the target response calculated by the optimized response surface surrogate model, and to obtain the model correction parameters that match the actual structural state.

[0047] Based on the model correction parameters, the parameters of the finite element model are updated to complete this periodic update;

[0048] By periodically repeating the update process, the finite element model is kept in sync with the actual structural state.

[0049] Secondly, the present invention provides a flexible photovoltaic support damage identification device based on an improved LK optical flow method, comprising:

[0050] The video acquisition module is used to acquire video of the pre-confirmed key monitoring areas;

[0051] The segmentation module is used to input the video of the key monitoring area into a pre-constructed fusion segmentation network to obtain a pure target video containing only the target flexible photovoltaic bracket;

[0052] The edge extraction module is used to perform edge detection on image data in pure target video and extract the edge skeleton of flexible photovoltaic support.

[0053] The solution module is used to take the edge frame of the flexible photovoltaic support as a constraint of the pre-constructed optical flow optimization objective function, and obtain the optimal optical flow displacement component by minimizing the objective function. Based on the optimal optical flow displacement component, the continuous dynamic displacement time history of key nodes on the edge frame is obtained. The key nodes are equally spaced discrete sampling nodes of the corresponding high-damage sensitive area members.

[0054] The modal parameter calculation module is used to construct a Hankel matrix based on continuous dynamic displacement time history, calculate the eigenvalues ​​and eigenvectors of the Hankel matrix using a data-driven random subspace identification method, and obtain the modal parameters of the flexible photovoltaic support based on the eigenvalues ​​and eigenvectors. The modal parameters include natural frequency, damping ratio and mode shape.

[0055] The damage determination module is used to calculate curvature mode and strain mode based on the modal parameters of flexible photovoltaic support, calculate the modal difference between the curvature mode and strain mode and the preset health benchmark mode, determine the damage area based on the abrupt change location of the modal difference, and quantitatively assess the degree of damage based on the absolute value of the modal difference.

[0056] Thirdly, the present invention provides an electronic device including at least one processor and a memory communicatively connected to the at least one processor; the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method described in any one of the first aspects.

[0057] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0058] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0059] This invention constructs a fusion segmentation network based on CNN and Transformer architectures, utilizing the local detail extraction capabilities of CNN and the global context modeling capabilities of Transformer to effectively eliminate complex background interference such as water surface reflections and crop fluctuations, providing clean image data for subsequent high-precision displacement extraction. Building upon traditional optical flow methods, it introduces an edge skeleton as a geometric constraint, preserving the ability of edge detection to characterize the support structure while overcoming the deficiency of relying solely on edges in complex backgrounds where false edges are easily generated. Furthermore, it combines visual displacement data with structural dynamics modal recognition, achieving precise damage localization and quantitative assessment through curvature modal difference and strain modal difference, improving displacement recognition accuracy under complex background interference and realizing precise damage localization of photovoltaic supports.

[0060] By periodically updating the finite element model based on measured data, the model can always remain synchronized with the actual structural state. This allows for structural response simulations under extreme wind conditions, enabling a leap from post-event diagnosis to pre-event early warning, and providing technical support for the long-term safe operation and maintenance of flexible photovoltaic supports. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating a method for identifying damage to flexible photovoltaic supports based on an improved LK optical flow method, provided in Embodiment 1 of the present invention.

[0062] Figure 2 This is a structural diagram of the fused segmentation network in a flexible photovoltaic support damage identification method based on the improved LK optical flow method provided in Embodiment 1 of the present invention;

[0063] Figure 3 This is a flowchart of modal parameter identification for a flexible photovoltaic support damage identification method based on the improved LK optical flow method provided in Embodiment 1 of the present invention.

[0064] Figure 4 This is a flowchart of the finite element model update and early warning process for a flexible photovoltaic support damage identification method based on the improved LK optical flow method provided in Embodiment 1 of the present invention.

[0065] Figure 5 This is a schematic diagram of the structure of a flexible photovoltaic support damage identification device based on the improved LK optical flow method provided in Embodiment 2 of the present invention;

[0066] Figure 6 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0067] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0068] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0069] Example 1:

[0070] Figure 1 This is a flowchart of the flexible photovoltaic support damage identification method based on the improved LK optical flow method in Embodiment 1 of the present invention. This flowchart only illustrates the logical sequence of the methods described in this embodiment. Under the premise of no conflict, different methods may be used in other possible embodiments of the present invention. Figure 1 Complete the steps shown or described in the order indicated.

[0071] The LK optical flow method (Lucas-Kanade-Optical-Flow) refers to the most classic and widely used sparse optical flow algorithm in the field of computer vision. Its core is to solve the instantaneous motion vector of sparse feature points in an image sequence through the least squares method of local windows.

[0072] See Figure 1 As shown, the method in this embodiment specifically includes the following steps:

[0073] Step 1: Obtain video of the pre-confirmed key monitoring area. It should be noted that the video here is attached... Figure 2 Specifically, the video images of wind-induced vibration are obtained by deploying industrial area array vision sensors in key monitoring areas. These sensors have a resolution of at least 2 megapixels and a sampling frame rate of at least 30fps. The sensor orientation is adjusted so that its optical axis is perpendicular to the vibration plane of the support structure, allowing for the acquisition of continuous video footage of the flexible photovoltaic support structure under natural wind-induced vibration conditions. The acquired video, after cropping, grayscale conversion, and Gaussian filtering preprocessing, can be used as input data for subsequent analysis.

[0074] The key monitoring areas identified include:

[0075] The modal confidence criterion value and damage sensitivity factor of each element in the pre-constructed finite element model of the flexible photovoltaic support are calculated. The expression for calculating the damage sensitivity factor includes:

[0076] ;

[0077] ;

[0078] ;

[0079] in, Let i be the damage sensitivity factor of the i-th unit. Let be the change in modal strain energy before and after damage to the i-th element. The total modal strain energy under healthy structural conditions. Let be the modal strain energy of the i-th element under the damaged state. Let be the modal strain energy of the i-th element under healthy conditions, and n be the total number of elements in the finite element model of the flexible photovoltaic support.

[0080] The expressions for calculating the modal confidence criterion values ​​include:

[0081] ;

[0082] in, Let be the modal confidence criterion values ​​for the k-th and j-th vibration modes. For the k-th reference mode shape, For the j-th measured mode shape, The closer the value is to 1, the better the mode shape correlation;

[0083] Units with a damage sensitivity factor greater than a preset sensitivity factor threshold and a mode amplitude greater than a preset mode amplitude threshold are selected as key monitoring areas. In this embodiment, the sensitivity factor threshold can be 0.8.

[0084] Step Two: As Figure 2 As shown, the video of the key monitoring area is input into a pre-constructed fusion segmentation network to obtain a pure target video containing only the target flexible photovoltaic support, including:

[0085] The convolutional branches of the fusion segmentation network are used to extract local features of the video images in the key monitoring areas. Through multi-scale downsampling and feature pyramid fusion, a local feature map containing detailed information such as support edges and textures is generated.

[0086] The global context features of images in the video of key monitoring areas are extracted by the Transformer branch of the fusion segmentation network to generate a global feature map containing the overall contour information of the scaffold. The Transformer branch includes a self-attention mechanism.

[0087] The local feature map and the global feature map are fused using an adaptive weighted fusion strategy to achieve effective fusion, resulting in a fused output feature map. The expression includes:

[0088] ;

[0089] ;

[0090] in, The output feature map after fusion. The local feature map output by the CNN branch. This is the global feature map output by the Transformer branch. These are adaptive weighting coefficients, with values ​​ranging from 0 to 1. The Sigmoid activation function is used to normalize the weight coefficients to the 0-1 range. This is a channel attention mechanism used to dynamically adjust the weighting coefficients based on the information entropy of the feature map, while also taking into account local edges;

[0091] By upsampling to restore the resolution of the fused output feature map, pixel-level segmentation results are output, resulting in a pure target video containing only the target flexible photovoltaic support. This network effectively removes background interference such as water surface shadows and crop undulations, providing clean image data for subsequent displacement extraction.

[0092] Step 3: Perform edge detection on the image data in the pure target video to extract the edge skeleton of the flexible photovoltaic support;

[0093] Edge detection here can be achieved using existing subpixel edge detection methods, which are conventional existing technologies and will not be elaborated upon here.

[0094] Step 4: Using the edge frame of the flexible photovoltaic support as a constraint of the pre-constructed optical flow optimization objective function, the optimal optical flow displacement component is obtained by minimizing the objective function. Based on the optimal optical flow displacement component, the continuous dynamic displacement time history of key nodes on the edge frame is obtained. The key nodes are equally spaced discrete sampling nodes corresponding to the high-sensitivity damage area of ​​the rod. The discrete sampling nodes are pixels selected at equal intervals along the axial direction of the rod edge. The continuous dynamic displacement time history is a displacement sequence formed by splicing the optical flow displacement values ​​of the key nodes in each video frame in the order of acquisition time.

[0095] The expression for the optical flow optimization objective function is as follows:

[0096] ;

[0097] in, The optimal optical flow displacement can be obtained by finding the minimum value of the objective function for optical flow optimization. For the tracking area of ​​key nodes, The edge frame of the extracted flexible photovoltaic support. These are the edge constraint weighting coefficients, balancing the optical flow fitting accuracy with the structural constraint strength. Let be the optical flow displacement vector. The normal vector of the edge skeleton. These represent the spatial gradients of the image grayscale in the x and y directions, respectively. This represents the gradient of image grayscale over time. These are the optical flow displacement components of the pixel in the x and y directions, respectively.

[0098] Before constructing the Hankel matrix based on the continuous dynamic displacement time history, the following steps are also included:

[0099] Preprocessing of the continuous dynamic displacement time history: Polynomial fitting is used to remove the trend term in the continuous dynamic displacement time history and eliminate low-frequency drift caused by factors such as temperature changes; Butterworth low-pass filter is used to remove high-frequency noise interference in the continuous dynamic displacement time history, retain the main frequency component of structural vibration, and finally obtain the preprocessed continuous dynamic displacement time history.

[0100] Step 5: As Figure 3 As shown, the Hankel matrix is ​​constructed based on the preprocessed continuous dynamic displacement time history, and its formula is as follows:

[0101]

[0102] in: For Hankel matrix, The number of rows in the block is taken as twice the system prediction order. The number of blocks is 1 / 2 of the number of sampling points.

[0103] Singular value decomposition is performed on the Hankel matrix, and the order is reduced by truncating the singular values ​​to separate the system signal and noise components and obtain the reduced-order observation matrix.

[0104] The system matrix is ​​solved from the reduced-order observation matrix using a data-driven random subspace method. The system matrix is ​​then subjected to eigenvalue decomposition to obtain eigenvalues ​​and eigencomponents. A stable graph is then constructed based on the eigenvalues ​​and eigencomponents.

[0105] Based on the stability diagram, points in the identification results of different orders where the rate of change of frequency, the rate of change of damping ratio, and the rate of change of mode shape are all less than a preset change threshold are marked as stable points. Noise modes and false modes are eliminated to obtain the modal parameters of the flexible photovoltaic support.

[0106] Step Six: Based on the modal parameters of the flexible photovoltaic support, calculate the curvature mode and strain mode. When the structure is damaged, the stiffness of the damaged area decreases, causing the curvature mode and strain mode of the area to change abruptly relative to the healthy state. Calculate the modal difference between the curvature mode and strain mode and the preset healthy baseline mode. Determine the damaged area based on the location of the abrupt change in the modal difference. Quantitatively assess the degree of damage based on the absolute value of the modal difference.

[0107] The modal parameters include natural frequency, damping ratio, and high-resolution mode shape;

[0108] The strain modes are obtained by calculating the mode shape and strain transformation matrix. The expressions include:

[0109] ;

[0110] ;

[0111] in, For the k-th strain mode, For the k-th mode shape, The strain transformation matrix, The distance between adjacent key nodes;

[0112] The curvature modes are calculated using the second-order central difference of the mode shapes. The expressions include:

[0113] ;

[0114] ;

[0115] in, For the k-th mode at node Curvature mode value at that point, For the k-th mode at node The modal displacement value at that point, The distance between adjacent nodes. For the k-th mode at node Curvature mode difference at that point, These are the measured curvature mode values. This represents the baseline curvature mode value under healthy conditions.

[0116] Furthermore, in cases of multiple damages coexisting, singular values ​​at a single measuring point may not accurately reflect the degree of damage. This embodiment further employs the modal superposition energy method: the measured vibration modes are expanded according to the modal modes under healthy conditions, the energy changes of each mode are calculated, and the degree of damage under multiple damage conditions is comprehensively evaluated by the amplitude of energy changes, thereby improving the accuracy of the evaluation.

[0117] like Figure 4 As shown in the figure, the flexible photovoltaic support damage identification method based on the improved LK optical flow method provided in this embodiment also includes:

[0118] Using the continuous dynamic displacement time history and modal parameters as the target response, the finite element model is periodically updated to keep the model synchronized with the actual structural state.

[0119] The regression coefficients of the pre-constructed second-order polynomial response surface surrogate model are solved by the least squares method to complete the calibration of the response surface surrogate model, which is used to replace the finite element model for response calculation.

[0120] A pre-constructed BP neural network is used to optimize and correct the response surface surrogate model, thereby improving the response calculation accuracy of the second-order polynomial response surface surrogate model. The deep learning algorithm used in this embodiment is implemented through the BP neural network.

[0121] The optimization objective is to minimize the residual between the measured target response and the target response calculated by the optimized response surface surrogate model, and to obtain the model correction parameters that match the actual structural state.

[0122] Based on the model correction parameters, the parameters of the finite element model are updated to complete this periodic update;

[0123] By periodically repeating the above update steps, the finite element model is kept synchronized with the actual structural state. The updated finite element model achieves a consistency of over 95% with the measured data.

[0124] The expression for the second-order polynomial response surface model includes:

[0125] ;

[0126] Where y is the target response. The input variables include key design parameters such as the elastic modulus of the members, prestress values, stiffness of the connectors, and damping ratio in the finite element model. m represents the number of input variables. These are the regression coefficients of the second-order polynomial response surface model. This is the random error term.

[0127] Based on the updated finite element model, a dynamic response simulation under a 50-year return period extreme wind condition was conducted. Using the extreme wind speed time history as input load, the displacement response, stress response, and safety margin of the structure were calculated. When the calculated safety margin falls below the preset threshold specified in the code, the system immediately issues a proactive safety warning to maintenance personnel. Simultaneously, based on the response analysis results, targeted reinforcement and maintenance recommendations are provided, realizing a shift from post-event diagnosis to pre-event warning.

[0128] Example 2:

[0129] Embodiment 2 of the present invention provides a damage identification device for flexible photovoltaic supports based on an improved LK optical flow method, such as... Figure 5 As shown, it includes:

[0130] The video acquisition module is used to acquire video of the pre-confirmed key monitoring areas;

[0131] The segmentation module is used to input the video of the key monitoring area into a pre-constructed fusion segmentation network to obtain a pure target video containing only the target flexible photovoltaic bracket;

[0132] The edge extraction module is used to perform edge detection on image data in pure target video and extract the edge skeleton of flexible photovoltaic support.

[0133] The solution module uses the edge frame of the flexible photovoltaic support as a constraint of a pre-constructed optical flow optimization objective function. It obtains the optimal optical flow displacement component by minimizing the objective function. Based on this optimal optical flow displacement component, it acquires the continuous dynamic displacement time history of key nodes on the edge frame. These key nodes are equally spaced discrete sampling nodes corresponding to high-damage-sensitive areas of the support members. Each discrete sampling node is a pixel selected at equal intervals along the axial direction of the support member edge. The continuous dynamic displacement time history is a displacement sequence formed by stitching together the optical flow displacement values ​​of the key nodes in each video frame according to the acquisition time order.

[0134] The modal parameter calculation module is used to construct the Hankel matrix based on the continuous dynamic displacement time history, and to calculate the eigenvalues ​​and eigenvectors of the Hankel matrix using the data-driven random subspace identification method (Data-SSI). Based on the eigenvalues ​​and eigenvectors, modal parameters such as the natural frequency, mode shape and damping ratio of the flexible photovoltaic support are obtained. The modal parameters include the natural frequency, damping ratio and mode shape.

[0135] The damage determination module is used to calculate curvature modes and strain modes based on the modal parameters of the flexible photovoltaic support. When the structure is damaged, the stiffness of the damaged area decreases, causing the curvature modes and strain modes in that area to change abruptly relative to the healthy state. The module calculates the modal difference between the curvature modes and strain modes and the preset healthy baseline modes, determines the damaged area based on the location of the abrupt change in the modal difference, and quantitatively assesses the degree of damage based on the absolute value of the modal difference.

[0136] The flexible photovoltaic support damage identification device based on the improved LK optical flow method provided in Embodiment 2 of the present invention can execute the flexible photovoltaic support damage identification method based on the improved LK optical flow method provided in Embodiment 1 of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0137] Example 3:

[0138] like Figure 6 As shown, this embodiment also provides an electronic device, including at least one processor and a memory communicatively connected to the processor. The memory stores a computer program executable by the processor, which, when executed by the processor, implements the flexible photovoltaic support damage identification method based on the improved LK optical flow method described in any of the above embodiments. Specifically, this electronic device may be an industrial computer, an edge computing device, or a cloud server.

[0139] Example 4:

[0140] Embodiment 4 of the present invention also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method described in Embodiment 1, and has the corresponding functional modules and beneficial effects of the method.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0143] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0144] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0145] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A flexible photovoltaic support damage identification method based on an improved LK optical flow method, characterized in that, include: Obtain video of the pre-confirmed key monitoring areas; The video of the key monitoring area is input into a pre-constructed fusion segmentation network to obtain a pure target video containing only the target flexible photovoltaic bracket; Edge detection is performed on image data in pure target video to extract the edge skeleton of flexible photovoltaic support; The edge frame of the flexible photovoltaic support is used as a constraint of the pre-constructed optical flow optimization objective function. The optimal optical flow displacement component is obtained by minimizing the objective function. Based on the optimal optical flow displacement component, the continuous dynamic displacement time history of key nodes on the edge frame is obtained. The key nodes are equally spaced discrete sampling nodes of the corresponding damage-sensitive area members. The Hankel matrix is ​​constructed based on the continuous dynamic displacement time history. The eigenvalues ​​and eigenvectors of the Hankel matrix are calculated using a data-driven random subspace identification method. Based on the eigenvalues ​​and eigenvectors, the modal parameters of the flexible photovoltaic support are obtained. The modal parameters include natural frequency, damping ratio and mode shape. Based on the modal parameters of the flexible photovoltaic support, the curvature mode and strain mode are calculated. The modal difference between the curvature mode and strain mode and the preset health benchmark mode is calculated. The damage area is determined according to the abrupt change location of the modal difference. The degree of damage is quantitatively assessed according to the absolute value of the modal difference.

2. The flexible photovoltaic support damage identification method based on the improved LK optical flow method according to claim 1, wherein the key monitoring areas are identified as including: The damage sensitivity factor of each element in the pre-constructed finite element model of the flexible photovoltaic support is calculated. The expression for calculating the damage sensitivity factor includes: ； ； ； wherein, is the damage sensitive factor of the i-th element, is the modal strain energy variation of the i-th element before and after damage, is the total modal strain energy under the structure health state, is the modal strain energy of the i-th element under the damage state, is the modal strain energy of the i-th element under the health state, and n is the total number of elements of the flexible photovoltaic support finite element model. Element units with damage sensitivity factors greater than a preset sensitivity factor threshold and mode amplitudes greater than a preset mode amplitude threshold are selected as key monitoring areas.

3. The improved LK optical flow method based flexible photovoltaic support damage identification method according to claim 1, characterized in that, The video of the key monitoring area is input into a pre-constructed fusion segmentation network to obtain pure target video containing only the target flexible photovoltaic support, including: The convolutional branches of the fusion segmentation network are used to extract local features of the video images in the key monitoring areas. Through multi-scale downsampling and feature pyramid fusion, a local feature map containing detailed information such as support edges and textures is generated. The global context features of the images in the video of the key monitoring area are extracted by the Transformer branch of the fusion segmentation network to generate a global feature map containing the overall outline information of the support structure. The local feature map is fused with the global feature map to obtain the fused output feature map; The resolution of the fused output feature map is restored by upsampling, and the output is a pure target video containing only the target flexible photovoltaic support.

4. The improved LK optical flow method based flexible photovoltaic support damage identification method according to claim 1, characterized in that, A data-driven random subspace identification method is used to obtain the eigenvalues ​​and eigenvectors of the Hankel matrix. Based on the eigenvalues ​​and eigenvectors, the modal parameters of the flexible photovoltaic support are obtained, including: Singular value decomposition is performed on the Hankel matrix to separate the system signal and noise components, and a reduced-order observation matrix is ​​obtained. The system matrix is ​​solved from the reduced-order observation matrix using a data-driven random subspace method. The system matrix is ​​then subjected to eigenvalue decomposition to obtain eigenvalues ​​and eigencomponents. A stable graph is then constructed based on the eigenvalues ​​and eigencomponents. Based on the stability diagram, points in the identification results of different orders where the rate of change of frequency, the rate of change of damping ratio, and the rate of change of mode shape are all less than a preset change threshold are marked as stable points. Noise modes and false modes are eliminated to obtain the modal parameters of the flexible photovoltaic support.

5. The improved LK optical flow based flexible photovoltaic support damage identification method according to claim 1, wherein, The expression for the optical flow optimization objective function is: ； wherein, is an optical flow optimization objective function, is a tracking area of key nodes, is an edge skeleton of the extracted flexible PV support, is an edge constraint weight coefficient, balancing the optical flow fitting accuracy and the structural constraint strength, is an optical flow displacement vector, is a normal vector of the edge skeleton, are spatial gradients of image gray in x, y directions, respectively, is a gradient of image gray in time dimension, are optical flow displacement components of a pixel in x, y directions, respectively.

6. The method for damage identification of flexible photovoltaic supports based on the improved LK optical flow method according to claim 1, characterized in that, The strain modes are obtained by calculating the mode shape and strain transformation matrix. The expressions include: ； ； in, For the k-th strain mode, For the k-th mode shape, The strain transformation matrix, The distance between adjacent key nodes; The curvature modes are calculated using the second-order central difference of the mode shapes. The expressions include: ； ； in, For the k-th mode at node Curvature mode value at that point, For the k-th mode at node The modal displacement value at that point, The distance between adjacent nodes. For the k-th mode at node Curvature mode difference at that point, These are the measured curvature mode values. This represents the baseline curvature mode value under healthy conditions.

7. The method for damage identification of flexible photovoltaic supports based on the improved LK optical flow method according to claim 1, characterized in that, Before constructing the Hankel matrix based on the continuous dynamic displacement time history, the following steps are also included: Polynomial fitting is used to remove the trend term in the continuous dynamic displacement time history and eliminate low-frequency drift; Butterworth low-pass filter is used to remove high-frequency noise interference in the continuous dynamic displacement time history and retain the main frequency component of structural vibration.

8. The method for damage identification of flexible photovoltaic supports based on the improved LK optical flow method according to claim 1, characterized in that, Using the continuous dynamic displacement time history and modal parameters as the target response, the finite element model is periodically updated to keep the model synchronized with the actual structural state. The regression coefficients of the pre-constructed second-order polynomial response surface surrogate model are solved by the least squares method to complete the calibration of the response surface surrogate model, which is used to replace the finite element model for response calculation. A pre-built BP neural network is used to optimize and correct the response surface surrogate model, which is used to improve the response calculation accuracy of the second-order polynomial response surface surrogate model. The optimization objective is to minimize the residual between the measured target response and the target response calculated by the optimized response surface surrogate model, and to obtain the model correction parameters that match the actual structural state. Based on the model correction parameters, the parameters of the finite element model are updated to complete this periodic update; By periodically repeating the update process, the finite element model is kept in sync with the actual structural state.

9. A method and device for damage identification of flexible photovoltaic supports based on an improved LK optical flow method, characterized in that, include: The video acquisition module is used to acquire video of the pre-confirmed key monitoring areas; The segmentation module is used to input the video of the key monitoring area into a pre-constructed fusion segmentation network to obtain a pure target video containing only the target flexible photovoltaic bracket; The edge extraction module is used to perform edge detection on image data in pure target video and extract the edge skeleton of flexible photovoltaic support. The solution module is used to take the edge frame of the flexible photovoltaic support as a constraint of the pre-constructed optical flow optimization objective function, and obtain the optimal optical flow displacement component by minimizing the objective function. Based on the optimal optical flow displacement component, the continuous dynamic displacement time history of key nodes on the edge frame is obtained. The key nodes are equally spaced discrete sampling nodes of the corresponding high-damage sensitive area members. The modal parameter calculation module is used to construct a Hankel matrix based on continuous dynamic displacement time history, calculate the eigenvalues ​​and eigenvectors of the Hankel matrix using a data-driven random subspace identification method, and obtain the modal parameters of the flexible photovoltaic support based on the eigenvalues ​​and eigenvectors. The modal parameters include natural frequency, damping ratio and mode shape. The damage determination module is used to calculate curvature mode and strain mode based on the modal parameters of flexible photovoltaic support, calculate the modal difference between the curvature mode and strain mode and the preset health benchmark mode, determine the damage area based on the abrupt change location of the modal difference, and quantitatively assess the degree of damage based on the absolute value of the modal difference.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 8.

Images