A photovoltaic power station inspection method and system for multimodal collaboration

Abstract

The present invention relates to the technical field of photovoltaic power station inspection, and discloses a multi-modal collaborative photovoltaic power station inspection method and system. The method includes obtaining power station location information, spectral reflection data, and an initial geographical fence; performing aging analysis based on the spectral reflection data to obtain material aging characteristics; inputting the material aging characteristics into a pre-trained aging evaluation model to output an aging evaluation result; performing spatial clustering based on the power station location information to obtain power station clustering data; performing iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence; performing inspection path planning based on the optimized geographical fence and the aging evaluation result to obtain an optimal inspection path, and inspecting the photovoltaic power station according to the optimal inspection path. The present method has the following effects: The present method can improve the inspection efficiency of photovoltaic power stations.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station inspection, and particularly to a multi-modal collaborative photovoltaic power station inspection method and system. Background Art

[0002] As an important part of sustainable energy solutions, the efficient operation of photovoltaic power stations is crucial for promoting the development of green energy. However, in the actual inspection process, photovoltaic power stations face many challenges. Since photovoltaic modules are installed in a vast area, manual inspection is not only time-consuming but also labor-intensive, especially in areas with complex terrain or harsh climate conditions. In addition, traditional inspection methods are difficult to achieve a comprehensive inspection of all photovoltaic panels, resulting in some problems not being detected in time, such as hidden cracks, hot spot effects, etc. If these problems are not dealt with in time, they will directly affect the power generation efficiency and service life of the power station. Therefore, how to effectively improve the inspection efficiency and ensure the detection quality has become a key issue in the operation and maintenance of photovoltaic power stations.

[0003] In one prior art, in order to improve the efficiency and accuracy of photovoltaic power station inspection, a widely used technical means is to perform automated inspection based on an unmanned aerial vehicle (UAV) equipped with a multi-sensor system. First, the UAV is equipped with a high-resolution RGB camera, an infrared thermal imager, and a light detection and ranging (LiDAR) to capture visible light images, thermal images, and three-dimensional spatial information respectively. Before the inspection task starts, a detailed flight plan is formulated according to the specific layout and environmental characteristics of the photovoltaic power station, and automatic obstacle avoidance parameters are set to ensure that the UAV can complete the task safely and effectively. During the flight, the UAV navigates autonomously along the preset path, collects data in real time, and synchronizes the data to the ground station through wireless transmission technology. Then, image processing algorithms are used to analyze the collected data. For example, by comparing thermal images at different time points to identify abnormally heated areas, or applying computer vision technology to detect physical damage from RGB images. These analysis results can quickly locate the fault location and provide accurate guidance for subsequent maintenance work.

[0004] However, although the multi-sensor inspection system based on UAVs has improved the efficiency and accuracy of photovoltaic power station inspection, its efficiency is still limited when faced with large-scale power stations. Specifically, when the photovoltaic power station covers a very large area, it becomes extremely time-consuming to use UAVs to inspect each area one by one, resulting in an extended overall inspection cycle. In fact, the state of photovoltaic panels in many areas is good, and obvious damage or abnormalities can be easily judged by conventional high-definition pictures without further detailed inspection using thermal imaging or other advanced sensors. In this case, the full-coverage inspection method of UAVs not only wastes time but also increases unnecessary workload, resulting in low efficiency of photovoltaic power station inspection. Summary of the Invention

[0005] The present invention provides a multi-modal collaborative photovoltaic power station inspection method and system to improve the efficiency of photovoltaic power station inspection.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a multi-modal collaborative photovoltaic power station inspection method, including:

[0007] Obtain the power station location information, spectral reflection data, and initial geographical fence;

[0008] Conduct aging analysis based on the spectral reflection data to obtain material aging characteristics;

[0009] Input the material aging characteristics into a pre-trained aging evaluation model to output an aging evaluation result;

[0010] Conduct spatial clustering based on the power station location information to obtain power station clustering data;

[0011] Conduct iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence;

[0012] Conduct inspection path planning based on the optimized geographical fence and the aging evaluation result to obtain an optimal inspection path, and inspect the photovoltaic power station according to the optimal inspection path.

[0013] In an optional implementation manner, the conducting aging analysis based on the spectral reflection data to obtain material aging characteristics includes:

[0014] Conduct linear regression analysis based on the spectral reflection data and a preset first wavelength range to obtain a spectral slope;

[0015] Conduct Gaussian fitting based on the spectral reflection data and a preset second wavelength range to obtain a wavelength reflection function;

[0016] Conduct second derivative calculation based on the wavelength reflection function to obtain a reflection second derivative;

[0017] Use the wavelength corresponding to the maximum value of the reflection second derivative as the reflection peak position;

[0018] Construct a spectral matrix based on the spectral reflection data;

[0019] Conduct principal component analysis on the spectral matrix to obtain a dimensionality-reduced spectral matrix;

[0020] Calculate the pollution index according to the following formula: where represents the pollution index, represents the dimension number of the dimensionality-reduced spectral matrix, represents the total number of dimensions of the dimensionality-reduced spectral matrix, Represents the weight coefficient of the [[ID=]]-th dimension, represents the dimensionality-reduced spectral matrix the value of the [[ID=]]-th dimension;

[0021] Among them, the material aging characteristics include the spectral slope, the reflection peak position, and the pollution index.

[0022] In an alternative embodiment, the training process of the aging evaluation model includes:

[0023] Construct an aging evaluation model based on historical aging characteristics and historical aging results, train the model, and determine that the training is completed when the number of training times reaches the set upper limit or the loss function of the detection model meets the conditions, and obtain the trained model;

[0024] Input the material aging characteristics into the trained model to obtain an aging evaluation result.

[0025] In an alternative embodiment, the spatial clustering according to the power station location information to obtain power station clustering data includes:

[0026] Perform coordinate system conversion on the power station location information to obtain plane position data;

[0027] Obtain the neighborhood radius and the minimum number of core points;

[0028] Traverse the plane position data;

[0029] Mark the points whose number of coordinate points within the neighborhood radius is greater than the minimum number of core points as core points;

[0030] Mark the coordinate points that do not meet the core standard within the neighborhood of the core points as boundary points, and incorporate the boundary points into the core point cluster of the core points;

[0031] Mark the coordinate points that are not incorporated into any core point cluster as noise points;

[0032] Among them, the power station clustering data includes the distribution of the core points, the boundary points, and the noise points.

[0033] In an alternative embodiment, the iterative optimization according to the power station clustering data and the initial geographical fence to obtain an optimized geographical fence includes:

[0034] Generate an initial convex bounding fence according to the power station clustering data;

[0035] Perform spatial topological analysis according to the initial geographical fence and the power station clustering data to obtain an abnormal fence;

[0036] Optimize the abnormal fence according to a preset fence optimization rule to obtain an updated geographical fence;

[0037] Merge the updated geographical fence and the initial convex bounding fence to obtain an optimized geographical fence.

[0038] In an alternative embodiment, the spatial topology analysis based on the initial geographical fence and the power station clustering data to obtain an abnormal fence includes:

[0039] When there are core points outside the initial geographical fence, it is determined as an out-of-bounds component fence;

[0040] When there are no core points inside the initial geographical fence, it is determined as a void area fence;

[0041] Calculate the power station density based on the initial geographical fence and the power station clustering data to obtain the power station density;

[0042] When the peak difference of the power station density inside the initial geographical fence is greater than a preset density threshold, it is determined as a density mutation area fence.

[0043] In an alternative embodiment, the inspection path planning based on the optimized geographical fence and the aging evaluation result to obtain an optimal inspection path and performing inspections on the photovoltaic power station according to the optimal inspection path includes:

[0044] Divide the inspection area according to the optimized geographical fence;

[0045] Mark high-risk components according to the aging evaluation result and the optimized geographical fence;

[0046] Initialize a basic inspection path that can traverse all the inspection areas;

[0047] Generate key detection points according to the high-risk components;

[0048] Insert the key detection points into the basic inspection path to obtain an updated inspection path;

[0049] Smooth the updated inspection path to obtain an optimal inspection path.

[0050] In a second aspect, the present invention provides a photovoltaic power station inspection system for multimodal collaboration, including:

[0051] A data acquisition module for acquiring power station location information, spectral reflection data, and an initial geographical fence;

[0052] An aging analysis module for performing aging analysis based on the spectral reflection data to obtain material aging characteristics;

[0053] An aging evaluation module, configured to input the material aging characteristics into a pre-trained aging evaluation model and output an aging evaluation result;

[0054] A power station clustering module, configured to perform spatial clustering based on the power station location information to obtain power station clustering data;

[0055] A fence optimization module, configured to perform iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence;

[0056] A path planning module, configured to perform inspection path planning based on the optimized geographical fence and the aging evaluation result to obtain an optimal inspection path, and perform inspections on the photovoltaic power station according to the optimal inspection path.

[0057] In a third aspect, the present invention further provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the multi-modal collaborative photovoltaic power station inspection method described in any one of the above is implemented.

[0058] In a fourth aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the multi-modal collaborative photovoltaic power station inspection method described in any one of the above.

[0059] Compared with the prior art, the present invention has the following beneficial effects: The present invention discloses a multi-modal collaborative photovoltaic power station inspection method and system. The method includes obtaining power station location information, spectral reflection data, and an initial geographical fence; performing aging analysis based on the spectral reflection data to obtain material aging characteristics; inputting the material aging characteristics into a pre-trained aging evaluation model and outputting an aging evaluation result; performing spatial clustering based on the power station location information to obtain power station clustering data; performing iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence; performing inspection path planning based on the optimized geographical fence and the aging evaluation result to obtain an optimal inspection path, and performing inspections on the photovoltaic power station according to the optimal inspection path. This method has the following effects: This method can improve the inspection efficiency of photovoltaic power stations.

[0060] Specifically, by obtaining power station location information, spectral reflection data, and an initial geographical fence, a comprehensive data basis is provided for subsequent analysis. In particular, using spectral reflection data for aging analysis can accurately identify the aging characteristics of materials. This method is more scientific and accurate than traditional time- or experience-based evaluations and can effectively locate areas that need attention.

[0061] Further, inputting the material aging characteristics into a pre-trained aging assessment model to output an aging assessment result not only improves the accuracy of the assessment but also accelerates the assessment process. Since the model is pre-trained, it means that it has learned patterns and regularities from a large amount of historical data, enabling it to provide fast and reliable prediction results on new data.

[0062] Further, perform spatial clustering based on the power station location information to obtain power station clustering data, and iteratively optimize it in combination with the initial geographical fence to obtain an optimized geographical fence. This approach helps improve the accuracy and practicality of the geographical fence, enabling the inspection activities to cover the target area more efficiently, while reducing unnecessary repeated inspections and enhancing the inspection efficiency.

[0063] Further, plan the optimal inspection path based on the optimized geographical fence and the aging assessment result. This step fully considers the actual status and distribution of the power station, ensuring the pertinence and effectiveness of the inspection work. The inspection path generated by the intelligent algorithm can minimize the inspection time while also ensuring that all key points are properly inspected.

[0064] In summary, the invention shows significant advantages in improving the inspection efficiency of photovoltaic power stations by integrating multiple data sources and applying advanced data analysis methods. It can not only more accurately identify the components that need maintenance but also optimize the inspection route, thereby enhancing the inspection efficiency of photovoltaic power stations. Brief Description of the Drawings

[0065] Figure 1 is a schematic flowchart of a multi-modal collaborative photovoltaic power station inspection method provided by the first embodiment of the present invention;

[0066] Figure 2 is a schematic structural diagram of a multi-modal collaborative photovoltaic power station inspection system provided by the second embodiment of the present invention. Detailed Embodiments

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0068] Refer to Figure 1 , the first embodiment of the present invention provides a multi-modal collaborative photovoltaic power station inspection method, including the following steps:

[0069] S11, obtain the power station location information, spectral reflection data, and the initial geographical fence;

[0070] S12, performing aging analysis according to the spectral reflectance data to obtain material aging characteristics;

[0071] S13, inputting the material aging characteristics into a pre-trained aging assessment model, and outputting an aging assessment result;

[0072] S14, performing spatial clustering according to the power station location information to obtain power station clustering data;

[0073] S15, performing iterative optimization according to the power station clustering data and the initial geo-fence to obtain an optimized geo-fence;

[0074] S16, performing inspection path planning according to the optimized geographic fence and the aging assessment result to obtain an optimal inspection path, and inspecting the photovoltaic power station according to the optimal inspection path.

[0075] In step S11 , power station location information, spectral reflectance data and initial geographic fence are acquired.

[0076] In one implementation, the power station location information is obtained through differential GPS technology, and a dual-frequency receiver is used for spatial positioning with centimeter-level accuracy. The data is stored in the WGS-84 coordinate system as a three-dimensional coordinate group containing longitude, latitude and elevation. The format adopts the GeoJSON standard structure and comes with metadata fields such as positioning accuracy (HDOP≤1.5) and acquisition timestamp. Each set of data corresponds to the physical location of a specific inverter or combiner box.

[0077] In one embodiment, spectral reflectance data is collected by a hyperspectral imager in the visible light-near infrared band (400-2500nm), carried on a patrol drone or remote sensing satellite platform, and generated into 256-band reflectance cube data after radiation calibration. The stored content includes the original DN value, the surface reflectance after atmospheric correction, the solar altitude angle at the time of shooting, and the sensor attitude parameters. It is encapsulated in a hierarchical data format (HDF5) and associated with the GPS position index. The spatial resolution is set to 0.5-2 meters based on the payload performance.

[0078] In one implementation, the initial geographic fence is based on sub-meter satellite remote sensing images, and the photovoltaic array contour is extracted using an improved Canny edge detection algorithm. The OpenStreetMap vector road network data is combined for spatial alignment to generate an ESRI Shapefile file containing a polygon vertex coordinate sequence. The boundary accuracy is controlled within the range of ±5 meters, and 10-15 meters of redundant space is allowed to be compatible with subsequent optimization algorithms. At the same time, the fence confidence level (level 1-3) and the topological relationship attributes of adjacent areas are marked.

[0079] In step S12, aging analysis is performed based on the spectral reflectance data to obtain material aging characteristics.

[0080] In one implementation, linear regression analysis is performed based on the spectral reflectance data and a preset first wavelength range to obtain a spectral slope;

[0081] Gaussian fitting is performed based on the spectral reflectance data and a preset second wavelength range to obtain a wavelength reflection function;

[0082] Second derivative calculation is performed based on the wavelength reflection function to obtain a reflection second derivative;

[0083] The wavelength corresponding to the maximum value of the reflection second derivative is used as the reflection peak position;

[0084] A spectral matrix is constructed based on the spectral reflectance data;

[0085] Principal component analysis is performed on the spectral matrix to obtain a reduced-dimensional spectral matrix;

[0086] The pollution index is calculated according to the following formula: where, represents the pollution index, represents the dimension number of the reduced-dimensional spectral matrix, represents the total number of dimensions of the reduced-dimensional spectral matrix, represents the weight coefficient of the th dimension, represents the value of the

[0087] th dimension of the reduced-dimensional spectral matrix;

[0088] It should be noted that the first wavelength range is the visible light band (i.e., 450 - 700 nm). In linear regression analysis, first, the collected original spectral data needs to be preprocessed: the Savitzky-Golay filtering algorithm is used to eliminate random noise interference, and at the same time, the cubic spline interpolation method is used to resample the non-uniformly sampled spectral data to a standardized wavelength sequence with a 1 nm interval. Subsequently, 20 characteristic wavelength points (such as key nodes like 450 nm, 475 nm, 500 nm, etc.) within this band are selected to construct a wavelength-reflectance scatter plot, and the weighted least squares method is used for linear regression modeling - where the weight coefficient is dynamically adjusted according to the signal-to-noise ratio of each wavelength point to ensure that high signal-to-noise ratio data points contribute more to the fitting result. The slope parameter obtained by fitting characterizes the average change in reflectance when the wavelength increases by 1 nm: a negative slope indicates a decrease in reflectance in the long-wavelength direction, and a positive slope reflects an increasing trend in reflectance.

[0089] It should be noted that the second wavelength range is the near-infrared band (set to 900 - 1700 nm in this method). The Gaussian fitting process is as follows: First, baseline correction is performed on the original data, and the locally weighted scatterplot smoothing (LOESS) algorithm is used to eliminate the baseline drift caused by the detector thermal noise. Abnormal reflection values caused by surface contamination are removed through the adaptive threshold method, and the proportion of valid data points retained should be ≥ 95%. The preprocessed spectral data is normalized to unify the reflectance value range to the 0 - 1 interval. The wavelength reflection function is fitted using a seven-parameter Gaussian model: where represents the wavelength reflection function at wavelength , represents the exponential function, represents the wavelength, represents the function parameter coefficient;

[0090] The second derivative calculation is performed using the five-point method for discrete second derivative calculation: where represents the second derivative of the wavelength reflection function, represents the wavelength, represents the instrument resolution (taking a value of 5 nm), represents the wavelength reflection function at wavelength , represents the wavelength reflection function at wavelength , the wavelength reflection function at wavelength , represents the wavelength reflection function at wavelength , represents the wavelength reflection function at wavelength .

[0091] In one implementation, first, the spectral reflectance data of multiple samples are arranged in rows to form the original data matrix X. This matrix has m rows (number of samples) and 251 columns (corresponding to the 450 - 700 nm wavelength range, 1 nm interval). Each element represents the reflectance value of the corresponding sample at a specific wavelength point. Subsequently, data standardization is performed: The mean and standard deviation are calculated for the reflectance values at each wavelength point, and then each value is subtracted by the mean of the column and divided by the standard deviation to ensure that each wavelength dimension has zero mean and unit variance. Next, the covariance matrix of the standardized data is calculated by multiplying the transposed standardized matrix by itself and dividing by m - 1. The covariance matrix is eigen-decomposed to obtain the eigenvalues arranged in descending order of magnitude and their corresponding eigenvectors (principal component directions). The number of principal components is determined by the cumulative variance contribution rate n , and the smallest number of principal components that makes the cumulative contribution rate reach more than 95% is selected. Finally, the standardized data is projected onto the selected firstn The principal component direction, and through matrix multiplication, the spectral matrix after dimensionality reduction is obtained. Its number of rows remains the same as the number of samples, and the number of columns is reduced to n the number of principal component dimensions.

[0092] It should be noted that the principal component dimensions of the dimensionality-reduced spectral matrix are 3, corresponding to the pollution characteristics of VOCs, heavy metals, and particulate matter respectively. The larger the value of the pollution index, the higher the degree of pollution of the photovoltaic power station.

[0093] In step S13, the material aging characteristics are input into a pre-trained aging evaluation model, and the aging evaluation result is output.

[0094] In one implementation, an aging evaluation model is constructed based on historical aging characteristics and historical aging results, and the model is trained. When the number of training times reaches the set upper limit or the loss function of the detection model meets the conditions, it is determined that the training is completed, and the trained model is obtained; the material aging characteristics are input into the trained model to obtain the aging evaluation result.

[0095] In one implementation, the aging evaluation model is based on an improved convolutional neural network. The input features include: (1) spectral slope, which characterizes the change rate of the surface chemical structure of the material (the mean value of the first derivative of the reflection spectrum in the wavelength range of 400 - 700 nm); (2) reflection peak position, which refers to the central wavelength offset of the characteristic reflection peak of the material (with an accuracy of ±0.5 nm); (3) pollution index, which calculates the proportion of the concentration of foreign atoms through EDX elemental analysis. The depth residual network (ResNet-18 variant) is used for feature fusion and aging prediction. The network architecture contains 4 residual blocks and embeds an SE channel attention module in the third layer to enhance the weight learning of key aging characteristics. During training, a sample set containing 50,000 groups of historical aging data is used, and 200 rounds of iterative training are carried out through the AdamW optimizer (learning rate 3e-4) in cooperation with the cross-entropy loss function with class weights (weight ratio 1:2:3:4:5). When the accuracy of the validation set fluctuates less than 0.5% for 15 consecutive rounds, the training is terminated. The final output includes: ① aging level (level 1 - 5, level 5 is severe aging), which is divided according to the weighted distance of the features from the reference value; ② estimated service life (unit: month), which is output through the regression of the fully connected layer and takes the lower limit value of the 95% confidence interval. For example, the output "level 3 (confidence level 82%), remaining life 23 ± 5 months" indicates that the material has entered the accelerated aging stage and needs to be detected more intensively.

[0096] In step S14, spatial clustering is performed according to the power station location information to obtain power station clustering data.

[0097] In one implementation, the coordinate system of the power station location information is converted to obtain plane position data; the neighborhood radius and the minimum number of core points are acquired; the plane position data is traversed; the points whose number of coordinate points within the neighborhood radius is greater than the minimum number of core points are marked as core points; the coordinate points that do not meet the core standard within the neighborhood of the core points are marked as boundary points, and the boundary points are incorporated into the core point cluster of the core points; the coordinate points that are not incorporated into any core point cluster are marked as noise points; wherein, the power station clustering data includes the distribution of the core points, the boundary points and the noise points.

[0098] It should be noted that the core standard is that the points whose number of coordinate points within the neighborhood radius is greater than the minimum number of core points are core points.

[0099] It should be noted that spatial clustering is implemented using an improved DBSCAN algorithm. Specifically, during implementation: First, the geographical location (longitude, latitude) of the power station is converted into plane rectangular coordinate system data through UTM projection to eliminate the influence of the earth's curvature; the neighborhood radius is set to 5 kilometers and the minimum number of core points is set to 3. When performing clustering, the algorithm traverses all coordinate points. When there are at least 3 adjacent power stations within the neighborhood of a certain point, it is marked as a core point, and then all adjacent points (including boundary points that do not meet the core standard) within 5 kilometers of the core point are classified into the same cluster; this process is recursively extended until no new members can be added, and finally three types of distributions are formed: core point clusters (representing high-density power station groups), boundary points (located at the edge of the cluster), and noise points (isolated power stations, more than 5 kilometers away from the nearest cluster).

[0100] In step S15, iterative optimization is performed according to the power station clustering data and the initial geographical fence to obtain an optimized geographical fence.

[0101] In one implementation, an initial convex bounding fence is generated according to the power station clustering data; spatial topological analysis is performed according to the initial geographical fence and the power station clustering data to obtain an abnormal fence; the abnormal fence is optimized according to a preset fence optimization rule to obtain an updated geographical fence; the updated geographical fence and the initial convex bounding fence are merged to obtain an optimized geographical fence.

[0102] In one implementation, the spatial topological analysis according to the initial geographical fence and the power station clustering data to obtain an abnormal fence includes: when there are core points outside the initial geographical fence, it is determined as an out-of-bounds component fence; when there are no core points inside the initial geographical fence, it is determined as a hollow area fence; the power station density is calculated according to the initial geographical fence and the power station clustering data to obtain the power station density; when the peak difference of the power station density inside the initial geographical fence is greater than a preset density threshold, it is determined as a density mutation area fence.

[0103] It should be noted that during the iterative optimization of the geofence, an initial convex bounding fence is first constructed through the Graham scan algorithm, which generates the minimum convex hull boundary based on the core point clusters generated by DBSCAN clustering. Subsequently, three types of abnormal fences are identified through three-dimensional spatial topological relationship analysis for targeted processing:

[0104] It should be noted that the optimization of the out-of-bounds component fence occurs when there are high-density core point clusters outside the initial fence. For example, in the case where some points of a discrete power station group exceed the boundary, the discrete points will be incorporated into the new boundary through polygon vertex expansion or bounding box extension operations. For the fence of the void area formed by the original fence accidentally including non-constructible areas (such as lake protection areas), spatial Boolean operations are used to cut off the redundant spatial range. In the face of the special case of the density mutation area fence, when there is a significant difference in the density of power stations in a certain area, with 15 power stations per square kilometer on the east side and a sudden drop to 3 power stations per square kilometer on the west side, a transition sub-fence will be constructed based on the Voronoi diagram segmentation technology, and a density gradient threshold of ±20% will be set to trigger the dynamic reconstruction mechanism.

[0105] It should be noted that in the final stage, the geometry union operation function of the GEOS library is used to perform polygon fusion on the optimized abnormal fence and the initial convex bounding fence to form a seamlessly connected closed optimized geofence. This fence not only completely covers all core point clusters but also can accurately reflect the spatial distribution characteristics of power stations, effectively eliminating geographical redundant areas.

[0106] In step S16, according to the optimized geofence and the aging assessment result, a patrol path is planned to obtain an optimal patrol path, and the photovoltaic power station is patrolled according to the optimal patrol path.

[0107] In one implementation, the patrol area is divided according to the optimized geofence; high-risk components are marked according to the aging assessment result and the optimized geofence; a basic patrol path capable of traversing all the patrol areas is initialized; key detection points are generated according to the high-risk components; the key detection points are inserted into the basic patrol path to obtain an updated patrol path; and the updated patrol path is smoothed to obtain an optimal patrol path.

[0108] In one implementation, based on the polygon boundary of the optimized geofence, the Voronoi diagram segmentation technology is used to divide the photovoltaic power station into logical patrol units. Each unit establishes a mapping relationship with the geographical coordinates of the equipment through a spatial index, and at the same time, a color temperature heat map generated by superimposing the aging assessment result is used to automatically identify high-risk components with a corrosion rate exceeding the threshold (such as an annual corrosion amount > 0.5 mm) or significant electrical performance degradation (such as a conversion efficiency drop > 15%), forming a three-dimensional marked point cloud data set including geographical coordinates, risk levels, and detection parameters.

[0109] In one implementation, an improved genetic algorithm is used to construct an initial inspection path. This algorithm sets double constraint conditions: the spatial constraint requires that the path completely covers the minimum bounding rectangle of all inspection units, and the time constraint limits the single operation duration to no more than 8 hours. A weighted graph model is generated through device positioning data. The node weights include the terrain undulation degree (DEM elevation data) and the obstacle distribution density, and the edge weights are calculated based on the UAV flight energy consumption model. Finally, a Hamiltonian cycle that meets the full coverage requirement is output.

[0110] In one implementation, for the marked high-risk component clusters, a two-layer detection strategy is initiated: in-situ fixed-point detection is carried out for the first-level risk points (such as insulation impedance < 2 MΩ), and a spherical detection area with a diameter of 50 meters is generated; for the second-level risk points (such as the backplane cracking rate > 30%), a scanning path along the component array is set. The shortest broken-line distance between the basic path and the risk points is calculated by the Dijkstra algorithm, and the detection area is seamlessly embedded into the original path by using the cubic spline interpolation method to ensure the curvature continuity between the newly added detection points and the basic path.

[0111] In one implementation, the Pareto front analysis method is introduced to handle multi-dimensional optimization objectives: in the energy consumption dimension, the lithium battery consumption rate of different flight segments is calculated according to the UAV power model; in the time dimension, the influence weight of the change in light conditions on visible light inspection is considered; in the detection accuracy dimension, the hover time is adjusted in combination with the wind speed perturbation model. Non-dominated sorting is carried out by the NSGA-II algorithm, and finally the path scheme with the optimal comprehensive score is selected. This scheme has the following characteristics: the detection time ratio in the high-risk area ≥ 40%, the total flight path volatility < 15%, and the length of the redundant flight segment is controlled within 5% of the total path.

[0112] In one implementation, a path smoothing engine based on the NURBS curve theory is deployed. This engine realizes three levels of smoothing processing through control point optimization: at the global level, the acute turns in the path are eliminated, and the turning radius is increased to the safety threshold (≥ 10 meters); at the local level, a spiral progressive approach strategy is adopted in the equipment-intensive area; at the micro level, the flight point spacing is dynamically adjusted according to the real-time meteorological data. Finally, a three-dimensional flight track file that meets the RTK positioning accuracy is output, and a digital twin scenario is constructed through the Unity3D engine, and the path feasibility is verified by using Monte Carlo simulation to ensure that the cloud error of more than 98% of the detection points is controlled within ±0.3 meters.

[0113] In one implementation, an evaluation index system is established, which includes the device body status (such as the cracking area of the backplane and the proportion of dark spots in the EL image), environmental corrosion parameters (salt spray deposition rate and acid rain pH value), and electrical performance degradation (I-V curve distortion degree and insulation impedance decline rate). The analytic hierarchy process is used to determine the weights of each index. For example, the backplane defect accounts for 35%, the electrical performance accounts for 40%, and the environmental factors account for 25%. The device is classified into four levels of risk through fuzzy comprehensive evaluation: level one (comprehensive score ≥ 85 points) requires emergency treatment, level two (70 - 84 points) requires key monitoring, level three (50 - 69 points) requires routine detection, and level four (< 50 points) requires sampling inspection. Taking a coastal power station as an example, 12.7% of its components reach the second-level risk due to salt spray corrosion, and the detection frequency needs to be increased to twice a week.

[0114] In one implementation, a "trinity" detection is implemented for high-risk devices: deploy an infrared thermal imager for all-weather temperature field monitoring (sampling interval ≤ 10 minutes); configure a drone carrying an EL detection module to perform fixed-point hovering imaging (hovering accuracy ±0.1m); arrange manual handheld IV testers for contact diagnosis. Medium-risk devices use a mobile IV curve scanning vehicle for daytime patrol inspection (coverage rate 100%). Low-risk devices are subject to monthly general surveys through intelligent cleaning robots equipped with simple sensors, and the detection duration is reduced from the conventional 45 minutes / group to 8 minutes / group.

[0115] In summary, the present invention discloses a multi-modal collaborative photovoltaic power station inspection method, aiming to improve the inspection efficiency of photovoltaic power stations by integrating multiple data sources and applying data analysis methods. The method first obtains the power station location information, spectral reflection data, and initial geographical fence to provide a comprehensive data basis for subsequent analysis. In the aging analysis stage, the spectral reflection data is used to accurately identify the aging characteristics of materials, such as spectral slope, reflection peak position, and pollution index, etc. This method is more scientific and accurate than traditional time- or experience-based evaluations and can effectively locate the areas that need attention. Then, these material aging characteristics are input into a pre-trained aging evaluation model to output the aging evaluation results. This process not only improves the accuracy of the evaluation but also accelerates the evaluation process. Since the model is trained based on a large amount of historical data, it can quickly provide reliable prediction results on new data. In addition, the power station clustering data is obtained through spatial clustering based on the power station location information, and the initial geographical fence is iteratively optimized to obtain an optimized geographical fence, which helps to improve the accuracy and practicality of the geographical fence, enabling the inspection activities to cover the target area more efficiently while reducing unnecessary repeated inspections. In terms of inspection path planning, the optimal inspection path is generated based on the optimized geographical fence and aging evaluation results, fully considering the actual status and distribution of the power station to ensure the pertinence and effectiveness of the inspection work.

[0116] For example, when processing spectral reflectance data, first perform linear regression analysis according to a preset first wavelength range (450 - 700 nm) to obtain the spectral slope; then perform Gaussian fitting according to the second wavelength range (900 - 1700 nm) to obtain the wavelength reflection function, and calculate the second derivative to determine the reflection peak position. The combined effect of these steps can effectively capture the signs of material surface aging. Subsequently, by performing principal component analysis (PCA) on the spectral matrix, the data dimension can be reduced, simplifying subsequent analysis while retaining key information. It is worth mentioning that throughout the process, the concept of a pollution index is also introduced to quantify the degree of contamination of photovoltaic modules, which is crucial for judging the operating status of the modules.

[0117] As one of the cores of the entire system, the construction and training process of the aging assessment model also reflects a high level of technical content. The model is constructed based on historical aging characteristics and historical aging results and is repeatedly trained until specific conditions are met. After training, the model can receive new input of material aging characteristics and quickly give aging assessment results. This machine learning-based method greatly improves the speed and accuracy of traditional manual assessment. At the same time, the application of spatial clustering algorithms has also brought significant improvements to the inspection work. Through the DBSCAN algorithm or its variants, power station units with close geographical locations and similar states can be grouped together to form a manageable clustering dataset. Then, combined with the information of the initial geographical fence, the design of the geographical fence is further optimized to ensure that the inspection plan covers all necessary areas without excessive expansion.

[0118] Finally, in the inspection path planning stage, the system divides the inspection area according to the optimized geographical fence and marks high-risk components in combination with the aging assessment results. The initialized basic inspection path will be adjusted on this basis, adding key detection points for high-risk components, and finally generating the optimal inspection path after smoothing. This process not only takes into account physical distance and terrain factors but also incorporates complex variables such as the drone flight energy consumption model, striving to achieve the most efficient inspection route design. Generally speaking, the present invention shows significant advantages in improving the inspection efficiency of photovoltaic power stations. It can not only more accurately identify the components that need maintenance but also optimize the inspection route to ensure that every key point can be properly inspected.

[0119] Refer to Figure 2 , the second embodiment of the present invention provides a photovoltaic power station inspection system with multimodal collaboration, including:

[0120] A data acquisition module for acquiring power station location information, spectral reflectance data, and an initial geographical fence;

[0121] An aging analysis module for performing aging analysis based on the spectral reflectance data to obtain material aging characteristics;

[0122] An aging assessment module, configured to input the material aging characteristics into a pre-trained aging assessment model and output an aging assessment result;

[0123] A power station clustering module, configured to perform spatial clustering based on the power station location information to obtain power station clustering data;

[0124] A fence optimization module, configured to perform iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence;

[0125] A path planning module, configured to perform inspection path planning based on the optimized geographical fence and the aging assessment result to obtain an optimal inspection path, and perform inspections on the photovoltaic power station according to the optimal inspection path.

[0126] It should be noted that a multi-modal collaborative photovoltaic power station inspection system provided by an embodiment of the present invention is used to execute all the process steps of a multi-modal collaborative photovoltaic power station inspection method in the above embodiment. The working principles and beneficial effects of the two correspond one by one, and thus will not be elaborated herein.

[0127] An embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition program. When the processor executes the computer program, the steps in the above embodiments of various multi-modal collaborative photovoltaic power station inspection methods are implemented, such as Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, the functions of each module / unit in the above device embodiments are implemented, such as the data acquisition module.

[0128] Exemplarily, the computer program may be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0129] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device, and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine certain components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0130] The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects various parts of the entire electronic device through various interfaces and circuits.

[0131] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory, the processor realizes various functions of the electronic device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0132] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0133] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0134] The specific embodiments described above have further elaborated on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only specific embodiments of the present invention and is not used to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A photovoltaic power station inspection method for multimodal collaboration, characterized in that, Including: Obtaining power station location information, spectral reflection data, and an initial geographical fence; Performing aging analysis based on the spectral reflection data to obtain material aging characteristics; Inputting the material aging characteristics into a pre-trained aging evaluation model to output an aging evaluation result; Performing spatial clustering based on the power station location information to obtain power station clustering data; Performing iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence, including: Constructing an initial convex bounding fence through the Graham scan algorithm, where the Graham scan algorithm generates a minimum convex hull boundary based on the core point clusters generated by DBSCAN clustering; performing spatial topological analysis based on the initial geographical fence and the power station clustering data to obtain abnormal fences; performing targeted processing on the abnormal fences to obtain optimized abnormal fences; using the geometric body union operation function of the GEOS library to perform polygon fusion on the optimized abnormal fences and the initial convex bounding fence to obtain a seamlessly connected closed optimized geographical fence; Wherein, the targeted processing includes: for the optimization of the over-bound component fence when there is a high-density core point cluster outside the initial fence, incorporating discrete points into the new boundary through polygon vertex expansion or bounding box extension operations; for the hole area fence formed by the original fence mistakenly including non-constructible areas, using spatial Boolean operations to cut off redundant spatial ranges; for the special case of the density mutation area fence, constructing a transition sub-fence based on the Voronoi diagram segmentation technology and setting a density gradient threshold to trigger a dynamic reconstruction mechanism; Performing inspection path planning based on the optimized geographical fence and the aging evaluation result to obtain an optimal inspection path, and performing inspections on the photovoltaic power station according to the optimal inspection path.

2. The multi-modal collaborative photovoltaic power plant inspection method according to claim 1, characterized in that, The performing aging analysis based on the spectral reflection data to obtain material aging characteristics includes: Performing linear regression analysis based on the spectral reflection data and a preset first wavelength range to obtain a spectral slope; Performing Gaussian fitting based on the spectral reflection data and a preset second wavelength range to obtain a wavelength reflection function; Performing second-order derivative calculation based on the wavelength reflection function to obtain a reflection second derivative; Taking the wavelength corresponding to the maximum value of the reflection second derivative as the reflection peak position; Constructing a spectral matrix based on the spectral reflection data; Performing principal component analysis on the spectral matrix to obtain a dimensionality-reduced spectral matrix; Calculate the pollution index according to the following formula: where represents the pollution index, represents the dimension number of the dimensionality-reduced spectral matrix, represents the total number of dimensions of the dimensionality-reduced spectral matrix, represents the weight coefficient of the th dimension, represents the value of the th dimension of the dimensionality-reduced spectral matrix; Wherein, the material aging characteristics include the spectral slope, the reflection peak position, and the pollution index.

3. The multi-modal collaborative photovoltaic power station inspection method according to claim 1, wherein The training process of the aging evaluation model includes: Constructing an aging evaluation model based on historical aging characteristics and historical aging results, training the model, and determining that the training is completed when the number of training times reaches the set upper limit or the loss function of the detection model meets the conditions, to obtain a trained model; Inputting the material aging characteristics into the trained model to obtain an aging evaluation result.

4. The multi-modal collaborative photovoltaic power plant inspection method according to claim 1, characterized in that The performing spatial clustering based on the power station location information to obtain power station clustering data includes: Performing coordinate system conversion on the power station location information to obtain planar position data; Obtaining a neighborhood radius and a minimum number of core points; Traversing the planar position data; Mark the points whose number of coordinate points within the neighborhood radius is greater than the minimum number of core points as core points; Mark the coordinate points that do not meet the core criteria within the neighborhood of the core points as boundary points, and incorporate the boundary points into the core point cluster of the core points; Mark the coordinate points that are not incorporated into any core point cluster as noise points; Among them, the power station clustering data includes the distribution of the core points, the boundary points, and the noise points.

5. The multi-modal collaborative photovoltaic power station inspection method according to claim 1, wherein, Performing spatial topology analysis based on the initial geographical fence and the power station clustering data to obtain an abnormal fence, including: When there are core points outside the initial geographical fence, it is determined as an out-of-bounds component fence; When there are no core points within the initial geographical fence, it is determined as a void area fence; Calculating the power station density based on the initial geographical fence and the power station clustering data to obtain the power station density; When the peak difference of the power station density within the initial geographical fence is greater than a preset density threshold, it is determined as a density mutation area fence.

6. The multi-modal collaborative photovoltaic power station inspection method according to claim 1, characterized in that Performing inspection path planning based on the optimized geographical fence and the aging assessment result to obtain an optimal inspection path, and inspecting the photovoltaic power station according to the optimal inspection path, including: Dividing the inspection area according to the optimized geographical fence; Marking high-risk components according to the aging assessment result and the optimized geographical fence; Initializing a basic inspection path that can traverse all the inspection areas; Generating key detection points according to the high-risk components; Inserting the key detection points into the basic inspection path to obtain an updated inspection path; Smoothing the updated inspection path to obtain an optimal inspection path.

7. A photovoltaic power station inspection system for multimodal collaboration, characterized in that, A photovoltaic power station inspection method for implementing multimodal collaboration as described in any one of claims 1 to 6, including: A data acquisition module for acquiring power station location information, spectral reflection data, and an initial geographical fence; An aging analysis module for performing aging analysis based on the spectral reflection data to obtain material aging characteristics; An aging assessment module for inputting the material aging characteristics into a pre-trained aging assessment model and outputting an aging assessment result; A power station clustering module for performing spatial clustering based on the power station location information to obtain power station clustering data; A fence optimization module for performing iterative optimization based on the power station clustering data and the initial geographical fence to obtain an optimized geographical fence; A path planning module for performing inspection path planning based on the optimized geographical fence and the aging assessment result to obtain an optimal inspection path, and inspecting the photovoltaic power station according to the optimal inspection path.

8. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the photovoltaic power station inspection method for multimodal collaboration as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the photovoltaic power station inspection method for multimodal collaboration as described in any one of claims 1 to 6.

Images