Distributed photovoltaic power station unmanned aerial vehicle inspection method and system

Abstract

The invention relates to the technical field of photovoltaic power station inspection, and provides a distributed photovoltaic power station unmanned aerial vehicle inspection method and system, and the method comprises the following steps: generating an inspection task list based on the multi-station distribution characteristics of a distributed photovoltaic power station; according to the inspection task list, performing scanning modeling on each photovoltaic sub-power station in sequence through a laser radar, and generating a photovoltaic power station three-dimensional map; dynamically planning an inspection path according to the characteristics of each photovoltaic sub-power station; according to the inspection path, performing multi-angle scanning on each photovoltaic module in different installation environments through a multi-source sensor, and obtaining photovoltaic module image data with position information in combination with the real-time position coordinates of the unmanned aerial vehicle; performing preprocessing, feature extraction and fault identification on the photovoltaic module image data by using a fault identification model, and establishing a fault feature library; and carrying out fusion analysis on the multi-source data, generating a fault detection report of each photovoltaic sub-power station based on the inspection task list, and formulating a corresponding maintenance scheme according to the fault detection report. According to the invention, multi-source data fusion analysis of the photovoltaic module image data, the fault feature library, the position coordinates and the three-dimensional map is realized, and the inspection efficiency and the fault detection accuracy of the distributed photovoltaic power station are improved.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power station inspection technology, and in particular to a distributed photovoltaic power station drone inspection method and system. Background Art

[0002] Photovoltaic power generation is an important clean energy source, and its safe and stable operation is directly related to power generation efficiency and economic benefits. Photovoltaic modules, as the core components of photovoltaic power generation systems, are prone to various defects during long-term operation, such as hot spots, cracks, and hidden cracks. If these defects are not discovered and handled in time, they may lead to reduced power generation efficiency or even safety accidents.

[0003] The existing technology generally uses multiple sensor detection methods to detect distributed photovoltaic power stations. Although multiple sensor detection methods can obtain multiple different types of data information, they cannot fully reflect the operating status of photovoltaic components. At the same time, they lack systematic inspection planning and intelligent path optimization. They often use simple fixed-route inspection modes, resulting in low detection efficiency. In the data collection process, they lack the ability to integrate and analyze multi-source data, and cannot fully utilize the complementary advantages of different types of sensors. Summary of the invention

[0004] In view of this, the present invention proposes a distributed photovoltaic power station drone inspection method and system, which solves the problem that the prior art lacks the ability to integrate and analyze multi-source data in the data collection process and cannot fully utilize the complementary advantages of different types of sensors.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a distributed photovoltaic power station drone inspection method, comprising the following steps:

[0006] S1, based on the multi-site distribution characteristics of distributed photovoltaic power stations, each photovoltaic sub-station is partitioned and numbered, and the inspection sequence between each photovoltaic sub-station is formulated to generate an inspection task list;

[0007] S2, according to the inspection task list, scanning and modeling each photovoltaic sub-station in turn by laser radar, marking the location information of each photovoltaic component, and generating a three-dimensional map of the photovoltaic power station;

[0008] S3, based on the inspection task list and the three-dimensional map of the photovoltaic power station, use the multi-source positioning system to perform rough positioning, obtain the real-time position coordinates of the drone, compare the real-time position coordinates of the drone with the preset three-dimensional map, determine the inspection area of ​​the drone, collect real-time images based on the onboard visual sensor and perform feature matching with the three-dimensional map of the photovoltaic power station, and dynamically plan the inspection path according to the characteristics of each photovoltaic sub-station;

[0009] S4, according to the inspection path, scanning each photovoltaic module in different installation environments at multiple angles by using a multi-source sensor, and acquiring photovoltaic module image data with position information in combination with the real-time position coordinates of the drone;

[0010] S5, based on the environmental information in the three-dimensional map of the photovoltaic power station, using a fault recognition model to preprocess, extract features and identify faults on the photovoltaic module image data, and establish a fault feature library adapted to different installation environments;

[0011] S6, performing a fusion analysis on the photovoltaic component image data, the fault feature library, the real-time position coordinates of the drone and the three-dimensional map of the photovoltaic power station, generating a fault detection report for each photovoltaic sub-station based on the order of the inspection task list, and formulating a corresponding maintenance plan based on the fault detection report of each photovoltaic sub-station.

[0012] On the basis of the above technical solution, preferably, step S1 includes:

[0013] S11, obtaining GPS coordinate information of each photovoltaic sub-station through a pre-flight of the drone, and using the drone's onboard camera to collect installation scene images of each photovoltaic sub-station, and extracting installation environment parameters of each photovoltaic sub-station in the installation scene images through an image recognition algorithm, wherein the installation environment parameters include installation description information, tilt angle, and orientation angle;

[0014] S12, using a clustering algorithm to group photovoltaic sub-stations with similar installation environment parameters, and calculating the optimal inspection route between each group of photovoltaic sub-stations based on the shortest path algorithm, and generating an inspection task list, wherein the inspection task list includes a partition number and an inspection order.

[0015] Based on the above technical solution, preferably, step S2 includes:

[0016] S21, based on the inspection task list, according to the area and complexity parameters of the photovoltaic sub-station, multiple scanning waypoints are set, at each scanning waypoint, the drone is controlled to hover and each photovoltaic sub-station is scanned from multiple angles by an airborne laser radar to obtain point cloud data;

[0017] S22, using an iterative closest point algorithm to align the point cloud data, using a voxel grid filtering algorithm to reduce noise and simplify the point cloud data, generating a three-dimensional map of the photovoltaic power station through a three-dimensional reconstruction algorithm, and identifying and marking the location information of each photovoltaic component in the three-dimensional map of the photovoltaic power station based on a deep learning method.

[0018] On the basis of the above technical solution, preferably, step S3 includes:

[0019] S31, acquiring multi-source position signals of the UAV through a multi-source positioning system, fusing the multi-source position signals based on a Kalman filter algorithm, and obtaining a rough positioning coordinate of the UAV;

[0020] S32, based on the airborne visual sensor, real-time images are collected, and the real-time images are matched with the three-dimensional map of the photovoltaic power station through an image feature matching algorithm to obtain the relative posture information of the UAV; the precise spatial position of the UAV is obtained according to the relative posture information, and the optimal inspection path is generated by a dynamic programming algorithm in combination with the inspection task list and the environmental information in the three-dimensional map of the photovoltaic power station, and the flight parameters of the UAV are adjusted in real time.

[0021] Based on the above technical solution, preferably, step S4 includes:

[0022] S41, based on the inspection path, according to the installation inclination angle of the photovoltaic module, adjusting the flight altitude of the drone, performing multi-angle scanning on each photovoltaic module, and obtaining multi-angle data of the photovoltaic module;

[0023] S42, aligning the real-time position coordinates of the drone with the multi-angle data of the photovoltaic module by time stamp, preprocessing the multi-source image data based on an image enhancement algorithm, and generating photovoltaic module image data.

[0024] Based on the above technical solution, preferably, step S5 includes:

[0025] S51, preprocessing and feature extraction of the photovoltaic module image data, including performing image denoising on the photovoltaic module image data based on an adaptive Gaussian filtering algorithm to obtain denoised photovoltaic module image data, performing image enhancement on the denoised photovoltaic module image data by an adaptive histogram equalization algorithm based on local contrast to obtain enhanced photovoltaic module image data, and extracting fault features of the enhanced photovoltaic module image data by using a deep learning model;

[0026] The calculation formula of the standard deviation of the adaptive Gaussian filtering algorithm is:

[0027]

[0028] Among them, σ zsy is the standard deviation of the adaptive Gaussian filter algorithm, d is the spacing between photovoltaic modules, N zsy is the number of photovoltaic modules in the photovoltaic module image data, α zsy is the first local area adjustment parameter, β zsy is the second local area adjustment parameter, I var is the variance of the local area of ​​the image;

[0029] S52, using an improved convolutional neural network model with multi-scale feature fusion to perform fault classification and identification on the fault features, using an environment-adaptive learning rate adjustment mechanism to dynamically adjust model training parameters according to feature complexity of different installation environments, and constructing an adaptive fault feature library;

[0030] The loss function calculation formula of the improved convolutional neural network model of multi-scale feature fusion is:

[0031]

[0032] Among them, L is the loss function value, β j is the weight coefficient of the jth fault category, J is the number of fault categories, y j is the actual fault type probability distribution of the jth fault category, is the predicted fault type probability distribution of the jth fault category, γ is the difficult-to-classify sample amplitude adjustment parameter, and δ is the error degree sensitivity adjustment parameter.

[0033] On the basis of the above technical solution, preferably, step S6 includes:

[0034] S61, using a multimodal data fusion algorithm, aligns and fuses the PV module image data, fault feature library data, drone real-time location data, and PV power station 3D map data in time and space to build a comprehensive fault assessment model;

[0035] The feature weight calculation formula of the multimodal data fusion algorithm is:

[0036]

[0037] in, is the fusion feature weight of the mth data source, f m is the feature vector of the data corresponding to the mth data source, f n is the feature vector of the data corresponding to the nth data source, N m is the number of data sources, q is the query vector, λ is the time series correlation coefficient, θ m,t-1 is the angle between the eigenvector at the current time t and the eigenvector at the time t-1;

[0038] S62, based on the comprehensive fault assessment model, performing a hierarchical assessment on the fault type, fault location and fault severity of each photovoltaic sub-station, generating a fault detection report including a fault distribution heat map, and formulating a corresponding maintenance plan according to the fault detection report;

[0039] The calculation formula of the comprehensive fault assessment model is:

[0040]

[0041] Among them, S gz is the comprehensive fault assessment score, P is the number of fault types, is the weight coefficient of the pth fault type, F p is the fault feature score of the pth fault type, A p is the fault impact area of ​​the pth fault type, A total is the total area of ​​PV panels, T p is the fault area temperature of the pth fault type, T ref is the reference temperature, β gz and γ gz They are area impact adjustment coefficient and temperature anomaly adjustment coefficient respectively.

[0042] In a second aspect, the present invention further provides a distributed photovoltaic power station drone inspection system, the system comprising:

[0043] The inspection task module is used to partition and number each photovoltaic sub-station based on the multi-site distribution characteristics of the distributed photovoltaic power station, formulate the inspection sequence between each photovoltaic sub-station, and generate an inspection task list;

[0044] A three-dimensional map module is used to scan and model each photovoltaic sub-station in turn through a laser radar according to the inspection task list, mark the location information of each photovoltaic component, and generate a three-dimensional map of the photovoltaic power station;

[0045] A positioning and navigation module is used to perform rough positioning using a multi-source positioning system based on the inspection task list and the three-dimensional map of the photovoltaic power station, obtain the real-time position coordinates of the drone, compare the real-time position coordinates of the drone with the preset three-dimensional map, determine the inspection area of ​​the drone, collect real-time images based on the onboard visual sensor and perform feature matching with the three-dimensional map of the photovoltaic power station, and dynamically plan the inspection path according to the characteristics of each photovoltaic sub-station;

[0046] A data acquisition module is used to perform multi-angle scanning of each photovoltaic module in different installation environments through a multi-source sensor according to the inspection path, and obtain photovoltaic module image data with position information in combination with the real-time position coordinates of the drone;

[0047] A fault identification module is used to pre-process, extract features and identify faults of the photovoltaic module image data using a fault identification model based on the environmental information in the three-dimensional map of the photovoltaic power station, and establish a fault feature library adapted to different installation environments;

[0048] The fault reporting module is used to perform a fusion analysis on the photovoltaic component image data, the fault feature library, the real-time position coordinates of the drone and the three-dimensional map of the photovoltaic power station, generate a fault detection report for each photovoltaic sub-station based on the order of the inspection task list, and formulate a corresponding maintenance plan based on the fault detection report of each photovoltaic sub-station.

[0049] In a third aspect, the present invention further provides an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0050] Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a distributed photovoltaic power station drone inspection method.

[0051] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement steps such as a distributed photovoltaic power station drone inspection method.

[0052] The distributed photovoltaic power station drone inspection method and system of the present invention has the following beneficial effects compared with the prior art:

[0053] (1) By using a multi-source positioning system combined with a three-dimensional map for precise navigation, using multi-source sensors for multi-angle scanning to obtain comprehensive image data, and establishing a fault feature library based on environmental information and performing identification, multi-source data fusion analysis of photovoltaic module image data, fault feature library, location coordinates and three-dimensional maps is achieved, thereby generating a systematic fault detection report and maintenance plan, improving the inspection efficiency and fault detection accuracy of distributed photovoltaic power stations;

[0054] (2) Adaptive Gaussian filtering algorithm and adaptive histogram equalization algorithm are used to perform image denoising and enhancement, and deep learning model is used to extract fault features and classify faults, thereby improving the accuracy of fault identification.

[0055] (3) Through the improved convolutional neural network model of multi-scale feature fusion and the environment-adaptive learning rate adjustment mechanism, the adaptive fault feature library is constructed to adapt to the feature complexity of different installation environments, improve the accuracy and efficiency of fault detection, enhance the adaptability and generalization ability in different environments, and thus improve the operation and maintenance level of photovoltaic power stations;

[0056] (4) By adopting a multimodal data fusion algorithm, the PV module image data, fault feature library data, drone real-time location data and PV power station three-dimensional map data are temporally and spatially aligned and feature fused to construct a comprehensive fault assessment model. The fault type, fault location and fault severity of each PV sub-station are graded and evaluated. A fault detection report containing a fault distribution heat map is generated, and corresponding maintenance plans are formulated based on the report. This not only improves the accuracy of fault detection, the precision and efficiency of fault assessment, but also enhances the targetedness of maintenance plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 This is a flow chart of a distributed photovoltaic power station drone inspection method of the present invention;

[0059] Figure 2 This is a structural diagram of a distributed photovoltaic power station drone inspection system of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] See also Figure 1 The present invention provides a distributed photovoltaic power station drone inspection method, comprising the following steps:

[0062] S1, based on the multi-site distribution characteristics of distributed photovoltaic power stations, each photovoltaic sub-station is partitioned and numbered, and the inspection sequence between each photovoltaic sub-station is formulated to generate an inspection task list;

[0063] S2, according to the inspection task list, scanning and modeling each photovoltaic sub-station in turn by laser radar, marking the location information of each photovoltaic component, and generating a three-dimensional map of the photovoltaic power station;

[0064] S3, based on the inspection task list and the three-dimensional map of the photovoltaic power station, use the multi-source positioning system to perform rough positioning, obtain the real-time position coordinates of the drone, compare the real-time position coordinates of the drone with the preset three-dimensional map, determine the inspection area of ​​the drone, collect real-time images based on the onboard visual sensor and perform feature matching with the three-dimensional map of the photovoltaic power station, and dynamically plan the inspection path according to the characteristics of each photovoltaic sub-station;

[0065] S4, according to the inspection path, scanning each photovoltaic module in different installation environments at multiple angles by using a multi-source sensor, and acquiring photovoltaic module image data with position information in combination with the real-time position coordinates of the drone;

[0066] S5, based on the environmental information in the three-dimensional map of the photovoltaic power station, using a fault recognition model to preprocess, extract features and identify faults on the photovoltaic module image data, and establish a fault feature library adapted to different installation environments;

[0067] S6, performing a fusion analysis on the photovoltaic component image data, the fault feature library, the real-time position coordinates of the drone and the three-dimensional map of the photovoltaic power station, generating a fault detection report for each photovoltaic sub-station based on the order of the inspection task list, and formulating a corresponding maintenance plan based on the fault detection report of each photovoltaic sub-station.

[0068] In a specific embodiment, the photovoltaic assembly image data includes visible light image data, infrared light image data and multi-spectral image data.

[0069] Specifically, this embodiment adopts a multi-source positioning system combined with a three-dimensional map for precise navigation, uses multi-source sensors for multi-angle scanning to obtain comprehensive image data, establishes a fault feature library based on environmental information and performs identification, thereby realizing multi-source data fusion analysis of photovoltaic component image data, fault feature library, location coordinates and three-dimensional maps, thereby generating a systematic fault detection report and maintenance plan, and improving the inspection efficiency and fault detection accuracy of distributed photovoltaic power stations.

[0070] Step S1 includes:

[0071] S11, obtaining GPS coordinate information of each photovoltaic sub-station through a pre-flight of the drone, and using the drone's onboard camera to collect installation scene images of each photovoltaic sub-station, and extracting installation environment parameters of each photovoltaic sub-station in the installation scene images through an image recognition algorithm, wherein the installation environment parameters include installation description information, tilt angle, and orientation angle;

[0072] S12, using a clustering algorithm to group photovoltaic sub-stations with similar installation environment parameters, and calculating the optimal inspection route between each group of photovoltaic sub-stations based on the shortest path algorithm, and generating an inspection task list, wherein the inspection task list includes a partition number and an inspection order.

[0073] Specifically, this embodiment obtains the GPS coordinate information of each photovoltaic sub-station through a drone pre-flight, uses an onboard camera to capture images of the installation scene, and combines an image recognition algorithm to extract installation environment parameters (such as installation description information, tilt angle, and orientation angle) to accurately understand the actual installation environment of each photovoltaic sub-station.

[0074] A clustering algorithm is used to group photovoltaic sub-stations with similar installation environment parameters, and the optimal inspection route between each group of photovoltaic sub-stations is calculated based on the shortest path algorithm to generate an inspection task list, which improves the efficiency of the inspection route and reduces unnecessary time consumption and resource waste during the inspection process.

[0075] By establishing partition numbers and inspection sequences, the comprehensiveness of inspection tasks is ensured, omissions and repeated inspections are avoided, and the overall efficiency of the inspection work is improved.

[0076] Step S2 includes:

[0077] S21, based on the inspection task list, according to the area and complexity parameters of the photovoltaic sub-station, multiple scanning waypoints are set, at each scanning waypoint, the drone is controlled to hover and each photovoltaic sub-station is scanned from multiple angles by an airborne laser radar to obtain point cloud data;

[0078] S22, using an iterative closest point algorithm to align the point cloud data, using a voxel grid filtering algorithm to reduce noise and simplify the point cloud data, generating a three-dimensional map of the photovoltaic power station through a three-dimensional reconstruction algorithm, and identifying and marking the location information of each photovoltaic component in the three-dimensional map of the photovoltaic power station based on a deep learning method.

[0079] In a specific embodiment, the position information includes the spatial position coordinates and inclination information of each photovoltaic module.

[0080] Specifically, this embodiment uses laser radar to scan and model each photovoltaic sub-station in turn, combines the iterative closest point algorithm to accurately align the point cloud data, uses the voxel grid filtering algorithm to reduce noise and simplify, and generates a high-precision three-dimensional map of the photovoltaic power station, ensuring the accuracy of the three-dimensional map.

[0081] In the process of generating a 3D map, the location information of each photovoltaic module, including spatial position coordinates and inclination information, is identified and marked based on deep learning methods, which improves the accuracy of photovoltaic module positioning and ensures accurate identification and detection of each module during the inspection process. By reducing the noise and simplifying the processing of point cloud data, the amount of data is reduced, the efficiency of data processing is improved, and the overall inspection time is shortened.

[0082] Step S3 includes:

[0083] S31, acquiring multi-source position signals of the UAV through a multi-source positioning system, fusing the multi-source position signals based on a Kalman filter algorithm, and obtaining a rough positioning coordinate of the UAV;

[0084] The calculation formula of the Kalman filter algorithm is:

[0085] X k+1 =AX k +BU k +K(Z k -HX k );

[0086] Among them, X k and X k+1 They are the UAV state estimates at time k and time k+1, A is the state transfer matrix, B is the control input matrix, U k is the control input vector at time k, K is the Kalman gain matrix, Z k is the measurement value at time k, and H is the observation matrix;

[0087] S32, based on the airborne visual sensor, real-time images are collected, and the real-time images are matched with the three-dimensional map of the photovoltaic power station through an image feature matching algorithm to obtain relative posture information of the UAV; the precise spatial position of the UAV is obtained according to the relative posture information, and the optimal inspection path is generated by using a dynamic programming algorithm in combination with the inspection task list and the environmental information in the three-dimensional map of the photovoltaic power station, and the flight parameters of the UAV are adjusted in real time;

[0088] The calculation formula of the image feature matching algorithm is:

[0089]

[0090] Among them, S is the similarity score, N is the number of feature points, and w i is the weight coefficient of the i-th feature point, F real (i) and F map (i) are the feature vectors of the i-th feature point of the real-time image and the 3D map of the photovoltaic power station, respectively, and σ is the standard deviation of the Gaussian kernel;

[0091] The calculation formula of the dynamic programming algorithm is:

[0092] C=g(n)+h(n)+α·d(n)+β·e(n);

[0093] Among them, C is the path cost, g(n) is the actual cost from the starting point to node n, h(n) is the heuristic estimated cost from the node to the target node, d(n) is the terrain complexity of node n, e(n) is the estimated energy consumption of node n, and α and β are the influence weight coefficients of terrain complexity and energy consumption, respectively.

[0094] In a specific embodiment, the multi-source position signal includes a GNSS signal and an IMU sensor data, the UAV state estimation includes a UAV position and a UAV speed, the control input vector includes an acceleration control signal and a steering control signal, and the measurement value includes sensor data such as GPS and GLONASS.

[0095] Specifically, this embodiment obtains multi-source position signals of the UAV through a multi-source positioning system, including GNSS signals and IMU sensor data, and uses the Kalman filter algorithm to fuse these signals to obtain the rough positioning coordinates of the UAV. By combining the inspection task list and the environmental information in the three-dimensional map of the photovoltaic power station, a dynamic programming algorithm is used to generate the optimal inspection path, and the flight parameters of the UAV are adjusted in real time to ensure the rationality of the UAV inspection path, reduce the inspection time and inspection energy consumption, collect real-time images through airborne visual sensors, and perform feature matching with the three-dimensional map of the photovoltaic power station. The image feature matching algorithm is used to improve the accuracy of image matching, so that the UAV can accurately identify and locate photovoltaic components during the inspection process.

[0096] Step S4 includes:

[0097] S41, based on the inspection path, controlling the drone to perform multi-angle scanning at a preset height and a preset angle, and synchronously collecting data on the photovoltaic modules through a visible light camera, an infrared thermal imager, and a multispectral camera;

[0098] According to the installation inclination angle of the photovoltaic module, the flight altitude of the drone is adjusted to perform multi-angle scanning on each photovoltaic module to obtain multi-angle data of the photovoltaic module;

[0099] S42, aligning the real-time position coordinates of the drone with the multi-angle data of the photovoltaic module by time stamp, preprocessing the multi-source image data based on an image enhancement algorithm, and generating photovoltaic module image data.

[0100] In a specific embodiment, the flight altitude of the drone is adjusted to be controlled within the range of 3-5 meters, and each photovoltaic module is scanned at three angles of 45°, 60°, and 90°. After obtaining the multi-angle data of the photovoltaic module, the attitude stabilization system is used to ensure that the camera and the photovoltaic module maintain the optimal imaging angle during the acquisition process, and the hardware synchronization trigger mechanism is used to ensure that the data acquisition time error of the three sensors is controlled within 10 milliseconds. When the real-time position coordinates of the drone are aligned with the multi-angle data of the photovoltaic module, the time synchronization accuracy is better than 20 milliseconds. When the multi-source image data is preprocessed based on the image enhancement algorithm, the preprocessing includes: illumination balancing and denoising of visible light images, temperature calibration and atmospheric transmittance compensation of infrared images, and band registration and radiation correction of multi-spectral images.

[0101] Specifically, this embodiment controls the drone to perform multi-angle scanning at a preset height and angle, and uses a visible light camera, an infrared thermal imager, and a multi-spectral camera to synchronously collect data on photovoltaic modules, thereby ensuring the diversity of data collection. According to the installation inclination of the photovoltaic module, the flight altitude of the drone is adjusted within the range of 3-5 meters, and scanning is performed at three angles of 45°, 60°, and 90°. The attitude stabilization system ensures that the camera and the photovoltaic module maintain the optimal imaging angle. This embodiment uses a hardware synchronization trigger mechanism to ensure that the data acquisition time error of the three sensors is controlled within 10 milliseconds, and synchronizes the real-time position coordinates of the drone with the multi-angle data of the photovoltaic module through timestamp alignment, with a time synchronization accuracy of better than 20 milliseconds. Based on the image enhancement algorithm, the multi-source image data is pre-processed, including illumination balancing and denoising of visible light images, temperature calibration and atmospheric transmittance compensation of infrared images, and band registration and radiation correction of multi-spectral images, to generate high-quality photovoltaic module image data, thereby improving the quality of image data.

[0102] Step S5 includes:

[0103] S51, preprocessing and feature extraction of the photovoltaic module image data, including performing image denoising on the photovoltaic module image data based on an adaptive Gaussian filtering algorithm to obtain denoised photovoltaic module image data, performing image enhancement on the denoised photovoltaic module image data by an adaptive histogram equalization algorithm based on local contrast to obtain enhanced photovoltaic module image data, and extracting fault features of the enhanced photovoltaic module image data by using a deep learning model;

[0104] The calculation formula of the standard deviation of the adaptive Gaussian filtering algorithm is:

[0105]

[0106] Among them, σ zsyis the standard deviation of the adaptive Gaussian filter algorithm, d is the spacing between photovoltaic modules, N zsy is the number of photovoltaic modules in the photovoltaic module image data, α zsy is the first local area adjustment parameter, β zsy is the second local area adjustment parameter, I var is the variance of the local area of ​​the image;

[0107] S52, using an improved convolutional neural network model with multi-scale feature fusion to perform fault classification and identification on the fault features, using an environment-adaptive learning rate adjustment mechanism to dynamically adjust model training parameters according to feature complexity of different installation environments, and constructing an adaptive fault feature library;

[0108] The loss function calculation formula of the improved convolutional neural network model of multi-scale feature fusion is:

[0109]

[0110] Among them, L is the loss function value, β j is the weight coefficient of the jth fault category, J is the number of fault categories, y j is the actual fault type probability distribution of the jth fault category, is the predicted fault type probability distribution of the jth fault category, γ is the difficult-to-classify sample amplitude adjustment parameter, and δ is the error degree sensitivity adjustment parameter.

[0111] Specifically, this embodiment performs denoising on the photovoltaic module image data through an adaptive Gaussian filtering algorithm, and performs image enhancement in combination with an adaptive histogram equalization algorithm of local contrast, thereby improving the clarity of the image data and removing the noise and uneven lighting effects in the image.

[0112] The deep learning model is used to extract fault features from the enhanced photovoltaic module image data, and the improved convolutional neural network model with multi-scale feature fusion is adopted for fault classification and identification, which improves the accuracy of fault identification and can effectively identify different types of photovoltaic module faults.

[0113] Through the environment-adaptive learning rate adjustment mechanism, the model training parameters are dynamically adjusted according to the feature complexity of different installation environments, and an adaptive fault feature library is constructed, so that the fault identification model can adapt to different installation environments.

[0114] Step S6 includes:

[0115] S61, using a multimodal data fusion algorithm, aligns and fuses the PV module image data, fault feature library data, drone real-time location data, and PV power station 3D map data in time and space to build a comprehensive fault assessment model;

[0116] The feature weight calculation formula of the multimodal data fusion algorithm is:

[0117]

[0118] in, is the fusion feature weight of the mth data source, f m is the feature vector of the data corresponding to the mth data source, f n is the feature vector of the data corresponding to the nth data source, N m is the number of data sources, q is the query vector, λ is the time series correlation coefficient, θ m,t-1 is the angle between the eigenvector at the current time t and the eigenvector at the time t-1;

[0119] S62, based on the comprehensive fault assessment model, performing a hierarchical assessment on the fault type, fault location and fault severity of each photovoltaic sub-station, generating a fault detection report including a fault distribution heat map, and formulating a corresponding maintenance plan according to the fault detection report;

[0120] The calculation formula of the comprehensive fault assessment model is:

[0121]

[0122] Among them, S gz is the comprehensive fault assessment score, P is the number of fault types, α p gz is the weight coefficient of the pth fault type, F p is the fault feature score of the pth fault type, A p is the fault impact area of ​​the pth fault type, A total is the total area of ​​PV panels, T p is the fault area temperature of the pth fault type, T ref is the reference temperature, β gz and γ gz They are area impact adjustment coefficient and temperature anomaly adjustment coefficient respectively.

[0123] In a specific embodiment, various faults (such as component cracks, loose wiring, hot spot faults, etc.) identified in the photovoltaic sub-station are classified according to the fault detection report. Priorities are set for various faults based on the severity of the fault, the possible system impact, and the difficulty of repair. For example, high-priority faults (such as hot spot faults) need to be handled immediately to prevent further equipment damage and reduced energy efficiency; medium and low priority faults (such as minor loose wiring) can be arranged to be handled during the regular maintenance cycle.

[0124] Evaluate the required maintenance resources, including manpower, tools, spare parts, etc., to ensure that resources are sufficient and available when formulating maintenance plans. Rationally allocate maintenance personnel with corresponding skills according to the priority and complexity of maintenance tasks to ensure that each task is performed by the most suitable person.

[0125] Develop a detailed maintenance schedule that covers the processing time and maintenance cycle of various faults to ensure that all maintenance tasks are completed on time. Assign specific maintenance tasks to the corresponding maintenance teams or individuals, and track and supervise them through a systematic management platform.

[0126] Develop detailed operation guides and repair steps for different types of faults to ensure that maintenance personnel follow standard procedures and reduce human errors. Specify safe operation specifications and emergency plans in the maintenance plan to ensure the personal safety of maintenance personnel and the safety of equipment.

[0127] Record detailed information of each maintenance operation, including fault type, handling method, time and resources used, etc., to establish a complete maintenance file. By analyzing maintenance records and feedback, we can continuously optimize maintenance plans, improve maintenance efficiency and the accuracy of fault handling.

[0128] Specifically, this embodiment achieves orderly management of maintenance tasks through classification, priority evaluation and task scheduling, avoiding waste of resources and task accumulation. Timely processing of high-priority faults reduces the negative impact of faults on the overall performance of the photovoltaic power station.

[0129] Operating guidelines and standardized maintenance processes ensure the consistency and accuracy of each maintenance operation, reduce human errors and omissions, and based on detailed fault detection reports, maintenance plans can accurately handle specific faults, thereby improving the success rate of fault repair.

[0130] By pre-assessing resource requirements, we ensure that the resources required during the maintenance process are sufficient and used reasonably, avoiding maintenance delays due to insufficient resources. Optimized resource allocation and efficient maintenance processes reduce maintenance costs.

[0131] By promptly identifying and handling potential faults, the probability of major failures is reduced, the service life of photovoltaic modules and systems is extended, and maintenance plans are continuously optimized through data recording and feedback to ensure dynamic matching of maintenance strategies with the operating conditions of photovoltaic power stations.

[0132] Maintenance records and data analysis provide strong support for operational management, helping managers make more scientific decisions. Systematic maintenance management processes and detailed record mechanisms improve the transparency and controllability of the maintenance process and enhance the standardization of management.

[0133] See also Figure 2The present invention also provides a distributed photovoltaic power station drone inspection system, the system comprising:

[0134] The inspection task module is used to partition and number each photovoltaic sub-station based on the multi-site distribution characteristics of the distributed photovoltaic power station, formulate the inspection sequence between each photovoltaic sub-station, and generate an inspection task list;

[0135] A three-dimensional map module is used to scan and model each photovoltaic sub-station in turn through a laser radar according to the inspection task list, mark the location information of each photovoltaic component, and generate a three-dimensional map of the photovoltaic power station;

[0136] A positioning and navigation module is used to perform rough positioning using a multi-source positioning system based on the inspection task list and the three-dimensional map of the photovoltaic power station, obtain the real-time position coordinates of the drone, compare the real-time position coordinates of the drone with the preset three-dimensional map, determine the inspection area of ​​the drone, collect real-time images based on the onboard visual sensor and perform feature matching with the three-dimensional map of the photovoltaic power station, and dynamically plan the inspection path according to the characteristics of each photovoltaic sub-station;

[0137] A data acquisition module is used to perform multi-angle scanning of each photovoltaic module in different installation environments through a multi-source sensor according to the inspection path, and obtain photovoltaic module image data with position information in combination with the real-time position coordinates of the drone;

[0138] A fault identification module is used to pre-process, extract features and identify faults of the photovoltaic module image data using a fault identification model based on the environmental information in the three-dimensional map of the photovoltaic power station, and establish a fault feature library adapted to different installation environments;

[0139] The fault reporting module is used to perform a fusion analysis on the photovoltaic component image data, the fault feature library, the real-time position coordinates of the drone and the three-dimensional map of the photovoltaic power station, generate a fault detection report for each photovoltaic sub-station based on the order of the inspection task list, and formulate a corresponding maintenance plan based on the fault detection report of each photovoltaic sub-station.

[0140] Specifically, the inspection task module of a distributed photovoltaic power station drone inspection system in this embodiment generates a detailed inspection task list by partitioning and numbering the multi-site distribution characteristics of the distributed photovoltaic power station and formulating an inspection sequence, thereby ensuring the orderly progress of the inspection work and avoiding repeated or missed inspections.

[0141] The 3D map module uses lidar to scan and model each PV sub-station, mark the location information of each PV module, and generate a high-precision 3D map of the PV station. It provides detailed spatial information and component location data, and improves the level of refinement of inspections.

[0142] The positioning and navigation module obtains the real-time position coordinates of the UAV through a multi-source positioning system, and compares them with the preset three-dimensional map to determine the inspection area of ​​the UAV. Combined with the real-time image acquisition and feature matching of the onboard visual sensor, the module can dynamically plan the optimal inspection path. Through precise positioning and navigation capabilities, it ensures that the UAV can efficiently and accurately cover the predetermined inspection area, reduces unnecessary movement in the flight path, and improves the inspection coverage rate.

[0143] This embodiment achieves seamless connection from task generation, data collection to path planning, which not only improves the overall work efficiency of the system, but also enhances the information sharing and data consistency between modules, ensuring the smooth progress of the inspection process.

[0144] By integrating multiple functional modules such as inspection task management, 3D modeling, positioning and navigation, drones can autonomously execute task lists, automatically perform path planning and data collection, and improve the frequency and reliability of inspections.

[0145] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a distributed photovoltaic power station drone inspection method.

[0146] The present invention also discloses a computer-readable storage medium, which stores computer instructions, and the computer instructions enable the computer to implement all or part of the steps of a distributed photovoltaic power station drone inspection method described in an embodiment of the present invention. The storage medium includes: a U disk, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and other media that can store program codes.

[0147] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed photovoltaic power station drone inspection method, characterized in that: The following steps are involved: S1, based on the multi-site distribution characteristics of distributed photovoltaic power stations, each photovoltaic sub-station is partitioned and numbered, and the inspection sequence between each photovoltaic sub-station is formulated to generate an inspection task list; S2, according to the inspection task list, scanning and modeling each photovoltaic sub-station in turn by laser radar, marking the location information of each photovoltaic component, and generating a three-dimensional map of the photovoltaic power station; S3, based on the inspection task list and the three-dimensional map of the photovoltaic power station, use the multi-source positioning system to perform rough positioning, obtain the real-time position coordinates of the drone, compare the real-time position coordinates of the drone with the preset three-dimensional map, determine the inspection area of ​​the drone, collect real-time images based on the onboard visual sensor and perform feature matching with the three-dimensional map of the photovoltaic power station, and dynamically plan the inspection path according to the characteristics of each photovoltaic sub-station; S4, according to the inspection path, scanning each photovoltaic module in different installation environments at multiple angles by using a multi-source sensor, and acquiring photovoltaic module image data with position information in combination with the real-time position coordinates of the drone; S5, based on the environmental information in the three-dimensional map of the photovoltaic power station, using a fault recognition model to preprocess, extract features and identify faults on the photovoltaic module image data, and establish a fault feature library adapted to different installation environments; S6, performing a fusion analysis on the photovoltaic component image data, the fault feature library, the real-time position coordinates of the drone and the three-dimensional map of the photovoltaic power station, generating a fault detection report for each photovoltaic sub-station based on the order of the inspection task list, and formulating a corresponding maintenance plan based on the fault detection report of each photovoltaic sub-station.

2. A distributed photovoltaic power station drone inspection method as claimed in claim 1, characterized in that: Step S1 includes: S11, obtaining GPS coordinate information of each photovoltaic sub-station through a pre-flight of the drone, and using the drone's onboard camera to collect installation scene images of each photovoltaic sub-station, and extracting installation environment parameters of each photovoltaic sub-station in the installation scene images through an image recognition algorithm, wherein the installation environment parameters include installation description information, tilt angle, and orientation angle; S12, using a clustering algorithm to group photovoltaic sub-stations with similar installation environment parameters, and calculating the optimal inspection route between each group of photovoltaic sub-stations based on the shortest path algorithm, and generating an inspection task list, wherein the inspection task list includes a partition number and an inspection order.

3. A distributed photovoltaic power station drone inspection method as claimed in claim 1, characterized in that: Step S2 includes: S21, based on the inspection task list, according to the area and complexity parameters of the photovoltaic sub-station, multiple scanning waypoints are set, at each scanning waypoint, the drone is controlled to hover and each photovoltaic sub-station is scanned from multiple angles by an airborne laser radar to obtain point cloud data; S22, using an iterative closest point algorithm to align the point cloud data, using a voxel grid filtering algorithm to reduce noise and simplify the point cloud data, generating a three-dimensional map of the photovoltaic power station through a three-dimensional reconstruction algorithm, and identifying and marking the location information of each photovoltaic component in the three-dimensional map of the photovoltaic power station based on a deep learning method.

4. A distributed photovoltaic power station drone inspection method as claimed in claim 1, characterized in that: Step S3 includes: S31, acquiring multi-source position signals of the UAV through a multi-source positioning system, fusing the multi-source position signals based on a Kalman filter algorithm, and obtaining a rough positioning coordinate of the UAV; S32, based on the airborne visual sensor, real-time images are collected, and the real-time images are matched with the three-dimensional map of the photovoltaic power station through an image feature matching algorithm to obtain the relative posture information of the UAV; the precise spatial position of the UAV is obtained according to the relative posture information, and the optimal inspection path is generated by a dynamic programming algorithm in combination with the inspection task list and the environmental information in the three-dimensional map of the photovoltaic power station, and the flight parameters of the UAV are adjusted in real time.

5. A distributed photovoltaic power station drone inspection method as claimed in claim 1, characterized in that: Step S4 includes: S41, based on the inspection path, according to the installation inclination angle of the photovoltaic module, adjusting the flight altitude of the drone, performing multi-angle scanning on each photovoltaic module, and obtaining multi-angle data of the photovoltaic module; S42, aligning the real-time position coordinates of the drone with the multi-angle data of the photovoltaic module by time stamp, preprocessing the multi-source image data based on an image enhancement algorithm, and generating photovoltaic module image data.

6. A distributed photovoltaic power station drone inspection method as claimed in claim 1, characterized in that: Step S5 includes: S51, preprocessing and feature extraction of the photovoltaic module image data, including performing image denoising on the photovoltaic module image data based on an adaptive Gaussian filtering algorithm to obtain denoised photovoltaic module image data, performing image enhancement on the denoised photovoltaic module image data by an adaptive histogram equalization algorithm based on local contrast to obtain enhanced photovoltaic module image data, and extracting fault features of the enhanced photovoltaic module image data by using a deep learning model; The calculation formula of the standard deviation of the adaptive Gaussian filtering algorithm is: Among them, σ zsy is the standard deviation of the adaptive Gaussian filter algorithm, d is the spacing between photovoltaic modules, N zsy is the number of photovoltaic modules in the photovoltaic module image data, α zsy is the first local area adjustment parameter, β zsy is the second local area adjustment parameter, I var is the variance of the local area of ​​the image; S52, using an improved convolutional neural network model with multi-scale feature fusion to perform fault classification and identification on the fault features, using an environment-adaptive learning rate adjustment mechanism to dynamically adjust model training parameters according to feature complexity of different installation environments, and constructing an adaptive fault feature library; The loss function calculation formula of the improved convolutional neural network model of multi-scale feature fusion is: Among them, L is the loss function value, β j is the weight coefficient of the jth fault category, J is the number of fault categories, y j is the actual fault type probability distribution of the jth fault category, is the predicted fault type probability distribution of the jth fault category, γ is the difficult-to-classify sample amplitude adjustment parameter, and δ is the error degree sensitivity adjustment parameter.

7. A distributed photovoltaic power station drone inspection method as claimed in claim 6, characterized in that: Step S6 includes: S61, using a multimodal data fusion algorithm, aligns and fuses the PV module image data, fault feature library data, drone real-time location data, and PV power station 3D map data in time and space to build a comprehensive fault assessment model; The feature weight calculation formula of the multimodal data fusion algorithm is: in, is the fusion feature weight of the mth data source, f m is the feature vector of the data corresponding to the mth data source, f n is the feature vector of the data corresponding to the nth data source, N m is the number of data sources, q is the query vector, λ is the time series correlation coefficient, θ m,t-1 is the angle between the eigenvector at the current time t and the eigenvector at the time t-1; S62, based on the comprehensive fault assessment model, performing a hierarchical assessment on the fault type, fault location and fault severity of each photovoltaic sub-station, generating a fault detection report including a fault distribution heat map, and formulating a corresponding maintenance plan according to the fault detection report; The calculation formula of the comprehensive fault assessment model is: Among them, S gz is the comprehensive fault assessment score, P is the number of fault types, is the weight coefficient of the pth fault type, F p is the fault feature score of the pth fault type, A p is the fault impact area of ​​the pth fault type, A total is the total area of ​​PV panels, T p is the fault area temperature of the pth fault type, T ref is the reference temperature, β gz and γ gz They are area impact adjustment coefficient and temperature anomaly adjustment coefficient respectively.

8. A distributed photovoltaic power station drone inspection system, characterized in that: The system comprises: The inspection task module is used to partition and number each photovoltaic sub-station based on the multi-site distribution characteristics of distributed photovoltaic power stations, formulate the inspection sequence between each photovoltaic sub-station, and generate an inspection task list; A three-dimensional map module is used to scan and model each photovoltaic sub-station in turn through a laser radar according to the inspection task list, mark the location information of each photovoltaic component, and generate a three-dimensional map of the photovoltaic power station; A positioning and navigation module is used to perform rough positioning using a multi-source positioning system based on the inspection task list and the three-dimensional map of the photovoltaic power station, obtain the real-time position coordinates of the drone, compare the real-time position coordinates of the drone with the preset three-dimensional map, determine the inspection area of ​​the drone, collect real-time images based on the onboard visual sensor and perform feature matching with the three-dimensional map of the photovoltaic power station, and dynamically plan the inspection path according to the characteristics of each photovoltaic sub-station; A data acquisition module is used to perform multi-angle scanning of each photovoltaic module in different installation environments through a multi-source sensor according to the inspection path, and obtain photovoltaic module image data with position information in combination with the real-time position coordinates of the drone; A fault identification module is used to pre-process, extract features and identify faults of the photovoltaic module image data using a fault identification model based on the environmental information in the three-dimensional map of the photovoltaic power station, and establish a fault feature library adapted to different installation environments; The fault reporting module is used to perform a fusion analysis on the photovoltaic component image data, the fault feature library, the real-time position coordinates of the drone and the three-dimensional map of the photovoltaic power station, generate a fault detection report for each photovoltaic sub-station based on the order of the inspection task list, and formulate a corresponding maintenance plan based on the fault detection report of each photovoltaic sub-station.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images