Photovoltaic array power generation performance prediction method in combination with unmanned aerial vehicle inspection data

Abstract

The invention relates to a photovoltaic array power generation performance prediction method in combination with unmanned aerial vehicle inspection data. The method comprises the following steps: acquiring an initial data set comprising sensor data, unmanned aerial vehicle inspection data and terrain and meteorological data, wherein the unmanned aerial vehicle inspection data comprises visible light image data, infrared image data and spectral image data; based on the visible light image data, identifying the boundaries of the components through image segmentation, extracting geographic position marks, calculating a light transmittance loss value, and generating a potential defect component set in combination with the infrared image data; based on the set, a defect influence weight matrix is obtained through mismatch power analysis, and a matrix is constructed by fusing topographic feature parameters and the like; the matrix is input into a machine learning model to generate a power generation performance prediction curve, the power generation performance of the photovoltaic array can be accurately predicted, the maintenance priority is scientifically and reasonably formulated through weighted scoring by combining the weight matrix, a maintenance decision list is generated, and scientific and reliable decision support is provided for operation and maintenance management of the photovoltaic array.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power generation technology, and in particular relates to a method for predicting the power generation performance of photovoltaic arrays by combining UAV inspection data. Background Technology

[0002] With the continued growth of global demand for clean energy, photovoltaic (PV) power generation plays a vital role in the energy sector due to its clean and renewable advantages. As the core component of a PV power generation system, the stability and efficiency of PV arrays directly affect the overall system's operational effectiveness. However, in actual operation, PV arrays face numerous complex factors, making accurate prediction of their power generation performance a significant challenge. Firstly, meteorological conditions such as solar irradiance, temperature, humidity, and wind speed are constantly changing, and these factors have a substantial impact on the power generation performance of PV arrays. For example, the intensity of irradiance directly determines the amount of light energy received by the PV modules, thus affecting power generation, while increased temperature may lead to reduced module efficiency. However, traditional PV array power generation performance prediction methods often struggle to accurately capture the nonlinear relationship between these complex and variable meteorological factors and power generation performance, resulting in insufficient prediction accuracy. Secondly, during long-term operation, PV modules inevitably develop various defects such as microcracks, hot spots, dust accumulation, and reduced light transmittance. These defects gradually erode the performance of the PV modules, ultimately negatively impacting the overall power generation capacity of the PV array. However, traditional defect detection methods rely on manual inspection, which is not only inefficient and has blind spots, but also makes it difficult to fully and timely detect potential defects. Summary of the Invention

[0003] Therefore, it is necessary to provide a method for predicting the power generation performance of photovoltaic arrays by combining UAV inspection data, in order to address the above-mentioned technical problems, thereby improving the accuracy and reliability of photovoltaic array power generation performance prediction and enhancing the ability to identify potential defects in photovoltaic modules.

[0004] In a first aspect, this application provides a method for predicting the power generation performance of a photovoltaic array by combining UAV inspection data, including:

[0005] Obtain the initial dataset, which includes sensor data, UAV inspection data, terrain data, and meteorological data. The sensor data includes irradiance data and temperature data, and the UAV inspection data includes visible light image data, infrared image data, and spectral image data.

[0006] Based on the visible light image data in the initial dataset, the boundaries of photovoltaic modules are identified by image segmentation algorithm, and the geographical location markers of each module are extracted. The transmittance loss value of each module is calculated, and a transmittance loss feature matrix including the transmittance loss value and the corresponding geographical location markers of the modules is generated. Based on the infrared image data and transmittance loss feature matrix in the initial dataset, modules with transmittance degradation and temperature anomalies are identified, and a set of potentially defective modules is generated.

[0007] Based on the set of potentially defective components, the measured current and voltage data of the associated strings are obtained. The power loss contribution weight value caused by the defects is calculated by the mismatch power analysis method, and the defect influence weight matrix is ​​generated. The terrain feature parameters are extracted from the terrain data, and the degradation values ​​in the terrain feature parameters, the transmittance loss feature matrix and the weight values ​​in the defect influence weight matrix are fused to construct a dynamic feature matrix.

[0008] The dynamic feature matrix is ​​input into a machine learning model based on time series analysis for processing to generate a power generation performance prediction curve. Based on the power generation performance prediction curve and the defect impact weight matrix, a weighted scoring algorithm is used to calculate the maintenance priority score and generate a maintenance decision list with geographic coordinates.

[0009] In one embodiment, based on infrared image data and a transmittance loss feature matrix in the initial dataset, components exhibiting transmittance degradation and temperature anomalies are identified, generating a set of potentially defective components, including:

[0010] The infrared image data is processed by image segmentation based on adaptive temperature threshold to identify areas where the temperature exceeds the preset offset value of the ambient temperature, and output hot spot feature data including hot spot location coordinates and hot spot area ratio.

[0011] Spatial matching is performed between the locations of components whose transmittance loss values ​​exceed the first threshold in the transmittance loss feature matrix and the location coordinates in the hot spot feature data to generate a set of candidate defect locations.

[0012] Based on spectral image data, for each component in the candidate defect location set, contrast and entropy features are extracted through gray-level co-occurrence matrix, the hidden crack density value is calculated, and components with hidden crack density values ​​exceeding the second threshold and corresponding hot spot area proportions exceeding the third threshold are selected to generate a potential defect component set.

[0013] In one embodiment, based on a set of potentially defective components, measured current and voltage data of associated strings are obtained. The power loss contribution weight value caused by the defect is calculated using a mismatch power analysis method, generating a defect impact weight matrix, including:

[0014] Based on the set of potentially defective components, locate the associated strings, obtain the measured string current and voltage data of the associated strings, and generate the measured power curve.

[0015] The theoretical power value of the associated string is calculated based on the preset standard test conditions, and the power loss value is calculated by the difference between the theoretical power value and the measured power curve.

[0016] Using a multiple linear regression algorithm, the hot spot area ratio and hidden crack density values ​​in the hot spot feature data are used as independent variables, and the power loss value is used as the dependent variable to construct a multiple linear regression model.

[0017] Based on the regression coefficients of the multiple linear regression model, the weighting coefficients of hot spot defects and hidden crack defects on the power loss value are analyzed, and a defect influence weight matrix is ​​generated.

[0018] In one embodiment, the terrain feature parameters, the degradation values ​​in the transmittance loss feature matrix, and the weight values ​​in the defect influence weight matrix are fused to construct a dynamic feature matrix. The terrain feature parameters include terrain slope and roughness parameters, including:

[0019] Based on terrain data, terrain slope and roughness parameters are extracted through a digital elevation model to generate a terrain complexity index.

[0020] From meteorological data in the initial dataset, the average wind speed and precipitation frequency of the target area are obtained to generate a set of environmental corrosion factors.

[0021] The degradation values ​​in the transmittance loss feature matrix and the weight values ​​in the defect impact weight matrix are multiplied by the terrain complexity index to generate primary fusion features.

[0022] A dynamic feature matrix is ​​generated by using a three-layer convolutional neural network to perform feature fusion encoding on primary fusion features and environmental corrosion factors.

[0023] In one embodiment, the dynamic feature matrix is ​​input into a time-series analysis-based machine learning model for processing to generate a power generation performance prediction curve, including:

[0024] Construct a machine learning model, which includes a first channel and a second channel;

[0025] The dynamic feature matrix is ​​input into the first channel, and the temporal features are extracted through the LSTM layer. The key time node features are weighted based on the self-attention mechanism, and the temporal feature vector is output.

[0026] The sensor data from the initial dataset is input into the second channel, and a nonlinear transformation is performed through a fully connected layer to output an environmental feature vector.

[0027] The initial power prediction curve is generated by dynamically weighting and fusing the time feature vector and the environmental feature vector through a gating mechanism.

[0028] By acquiring and optimizing the model parameters of the machine learning model based on the actual power generation data of the photovoltaic array through backpropagation algorithm, a power generation performance prediction curve is generated.

[0029] In one embodiment, based on the power generation performance prediction curve and the defect impact weight matrix, a maintenance priority score is calculated using a weighted scoring algorithm to generate a maintenance decision list with geographic coordinates, including:

[0030] The power decline rate within a preset period is calculated from the power generation performance prediction curve using the time difference method.

[0031] Read the defect contribution weight value of each component from the defect impact weight matrix;

[0032] The degradation values ​​in the transmittance loss feature matrix are processed by a normalization algorithm to generate normalized degradation values.

[0033] The power degradation rate, defect contribution weight value and normalized degradation value are weighted and integrated by a preset weighted summation formula to generate a maintenance priority score.

[0034] Based on the maintenance priority score, the geographical location markers of the corresponding components in the transmittance loss feature matrix are associated, and a maintenance decision list is generated by sorting in descending order.

[0035] Secondly, this application also provides a photovoltaic array power generation performance prediction device that combines UAV inspection data, comprising:

[0036] The multi-source data acquisition module is used to acquire the initial dataset, which includes sensor data, UAV inspection data, terrain data and meteorological data. The sensor data includes irradiance data and temperature data, and the UAV inspection data includes visible light image data, infrared image data and spectral image data.

[0037] The image analysis and defect identification module is used to identify the boundaries of photovoltaic modules based on visible light image data in the initial dataset, extract the geographical location markers of each module, calculate the transmittance loss value of each module, generate a transmittance loss feature matrix including the transmittance loss value and the corresponding geographical location markers of the modules, and identify modules with transmittance degradation and temperature anomalies based on infrared image data and transmittance loss feature matrix in the initial dataset, thereby generating a set of potentially defective modules.

[0038] The data fusion and matrix construction module is used to obtain measured current and voltage data of associated strings based on the set of potentially defective components, calculate the power loss contribution weight value caused by defects through the mismatch power analysis method, generate a defect impact weight matrix, extract terrain feature parameters from terrain data, and fuse the terrain feature parameters, the degradation value in the transmittance loss feature matrix and the weight value in the defect impact weight matrix to construct a dynamic feature matrix.

[0039] The performance prediction and maintenance decision module is used to input the dynamic feature matrix into a machine learning model based on time series analysis for processing, generate a power generation performance prediction curve, and calculate the maintenance priority score through a weighted scoring algorithm based on the power generation performance prediction curve and the defect impact weight matrix, and generate a maintenance decision list with geographic coordinates.

[0040] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the first aspect.

[0041] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the first aspect.

[0042] The aforementioned method for predicting the power generation performance of photovoltaic arrays by combining UAV inspection data comprehensively acquires environmental factors, component status, and terrain features affecting the power generation performance of photovoltaic arrays by obtaining an initial dataset, providing a data foundation for subsequent analysis. Secondly, it calculates the transmittance loss value of each component using visible light image data, generating a transmittance loss feature matrix. Combined with infrared image data, it identifies components with transmittance degradation and temperature anomalies, achieving efficient identification and location of potential defects in photovoltaic modules. Furthermore, based on the set of potentially defective components, it calculates the power loss contribution weight value caused by defects, generating a defect impact weight matrix. This matrix is ​​then integrated with terrain feature parameters, transmittance loss degradation values, and defect impact weight values ​​to construct a dynamic feature matrix, comprehensively considering the combined impact of terrain, defects, and environmental factors on power generation performance, providing more comprehensive input features for subsequent performance prediction. Finally, the dynamic feature matrix is ​​input into a machine learning model, which captures the time-varying characteristics of power generation performance and automatically learns complex nonlinear relationships, improving the accuracy of photovoltaic array power generation performance prediction. Furthermore, based on the curve and the defect impact weight matrix, a maintenance decision list with geographic coordinates is generated, enabling scientific assessment of maintenance priorities and rational allocation of maintenance resources, providing strong support for the efficient operation and maintenance of photovoltaic power plants.

[0043] Compared with traditional photovoltaic array power generation performance prediction methods, this method can not only accurately predict the power generation performance of photovoltaic arrays, but also scientifically and rationally formulate maintenance priorities based on the defect impact weight and power generation performance prediction, and generate a maintenance decision list with geographical coordinates, providing scientific and reliable decision support for the operation and maintenance management of photovoltaic arrays. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart of a method for predicting the power generation performance of a photovoltaic array by combining UAV inspection data is provided as an exemplary embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram of a photovoltaic array power generation performance prediction device that combines UAV inspection data, as an exemplary embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] In one embodiment, such as Figure 1 As shown, a method for predicting the power generation performance of a photovoltaic array by combining UAV inspection data is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0049] S101: Obtain the initial dataset, which includes sensor data, UAV inspection data, terrain data, and meteorological data. The sensor data includes irradiance data and temperature data, and the UAV inspection data includes visible light image data, infrared image data, and spectral image data.

[0050] Specifically, sensor data can be collected by sensors installed at the photovoltaic array site. Irradiance data reflects the intensity of solar radiation and is a key factor affecting the power generation of the photovoltaic array. Temperature data includes the surface temperature of the photovoltaic modules and the ambient temperature, used to assess the impact of temperature on power generation efficiency. Drone inspection data can be collected by drones equipped with various imaging devices, including visible light image data, infrared image data, and spectral image data. For example, visible light image data can visually display the appearance of the photovoltaic modules, such as damage or shading; infrared image data reflects the heat distribution of the modules and is used to identify anomalies such as hot spots; while spectral image data provides richer information, helping to analyze the material properties and aging of the modules. Furthermore, topographic data describes the terrain features of the area where the photovoltaic array is located, such as slope and aspect. This topographic data allows for further analysis of the factors affecting the sunlight received by the photovoltaic modules. Meteorological data, including wind speed, wind direction, and rainfall, provides environmental background information for subsequent analysis.

[0051] S102: Based on the visible light image data in the initial dataset, the boundaries of photovoltaic modules are identified through image segmentation algorithms, and the geographical location markers of each module are extracted. The transmittance loss value of each module is calculated, and a transmittance loss feature matrix including the transmittance loss value and the corresponding geographical location markers of the modules is generated. Based on the infrared image data and the transmittance loss feature matrix in the initial dataset, modules with transmittance degradation and temperature anomalies are identified, and a set of potentially defective modules is generated.

[0052] Specifically, image segmentation algorithms can separate photovoltaic (PV) modules from the background by analyzing features such as grayscale and texture in an image. Once the boundaries of the PV modules are identified, the geographic coordinates recorded during image acquisition or image matching techniques can be used to obtain the geographic location markers of each module. Subsequently, the transmittance loss value of each module can be calculated. This loss value reflects the degree of decline in light transmittance performance caused by surface contamination, aging, etc. For example, the transmittance loss value can be calculated by comparing the grayscale differences between clean and contaminated module images. Combining the calculated transmittance loss value with the corresponding geographic location marker of the module can form a transmittance loss feature matrix. Furthermore, by analyzing infrared image data, abnormalities such as hot spots inside the modules can be detected. If a module exhibits both transmittance degradation and temperature anomalies, it can be identified as a potentially defective module and added to the set of potentially defective modules. This process comprehensively considers the appearance, heat distribution, and light transmittance performance of the PV modules, accurately identifying modules with potential defects and providing a basis for subsequent maintenance decisions.

[0053] S103: Based on the set of potentially defective components, obtain the measured current and voltage data of the associated strings, calculate the power loss contribution weight value caused by defects through the mismatch power analysis method, generate the defect influence weight matrix, extract the terrain feature parameters from the terrain data, and fuse the terrain feature parameters, the degradation value in the transmittance loss feature matrix and the weight value in the defect influence weight matrix to construct a dynamic feature matrix.

[0054] Specifically, based on a set of potentially defective components, measured current and voltage data of the associated strings can be obtained through power quality monitoring equipment installed in the photovoltaic array. This measured current and voltage data reflects the actual operating status of the associated strings, and the actual power generation of the associated strings can be calculated from this data. Subsequently, the power loss contribution weight value caused by defective components can be calculated using the mismatch power analysis method. This weight value reflects the degree of impact of defective components on power generation performance, and the calculated weight values ​​can be combined to form a defect impact weight matrix. Furthermore, terrain feature parameters, such as slope and aspect, can be extracted from terrain data. The aspect can be obtained by calculating the direction of the slope change rate. By integrating and normalizing the terrain feature parameters, the degradation values ​​in the transmittance loss feature matrix, and the weight values ​​in the defect impact weight matrix, they can be analyzed uniformly in the same matrix, resulting in a dynamic feature matrix. This dynamic feature matrix integrates information from multiple aspects such as the geographical location of photovoltaic modules, transmittance loss, power loss contribution weight, and terrain features, providing comprehensive feature inputs for subsequent power generation performance prediction.

[0055] S104: Input the dynamic feature matrix into the machine learning model based on time series analysis for processing to generate a power generation performance prediction curve. Based on the power generation performance prediction curve and the defect impact weight matrix, calculate the maintenance priority score through a weighted scoring algorithm to generate a maintenance decision list with geographic coordinates.

[0056] Specifically, time series analysis is an analytical method suitable for processing time-dependent data, capable of considering the changing trends of photovoltaic array power generation performance over time. Therefore, this machine learning model can employ a long short-term memory network architecture. During model training, historical dynamic feature matrix data and corresponding actual power generation data are used as training samples. After multiple rounds of iterative training, it can learn patterns and regularities in historical data, and then generate a power generation performance prediction curve based on the feature information in the dynamic feature matrix. This prediction curve reflects the expected power generation of the photovoltaic array at different time points. Based on the power generation performance prediction curve and the defect impact weight matrix, a maintenance priority score can be assigned to each potentially defective component, taking into account both the power generation performance prediction results and the degree of impact of defective components on power generation performance. The higher the score, the greater the impact of the component on power generation performance, and the higher the maintenance priority. Finally, based on the maintenance priority score, the geographical coordinates of the corresponding components can be associated to generate a maintenance decision list with geographical coordinates. This list details the location, maintenance priority, and defect information of the components requiring maintenance, providing clear maintenance guidance for operation and maintenance personnel. This helps to rationally allocate maintenance resources, prioritize components with a greater impact on power generation performance, and thus improve the overall power generation efficiency and operational reliability of the photovoltaic array.

[0057] The aforementioned method, by acquiring an initial dataset including data from multiple sources, comprehensively collects multi-dimensional information affecting the power generation performance of photovoltaic arrays, avoiding the limitations of a single data source and providing reliable data support for subsequent analysis. Secondly, based on visible light image data, image segmentation algorithms are used to identify the boundaries of photovoltaic modules, extract geographical location markers, and calculate transmittance loss values. Combined with infrared image data, a set of potentially defective modules is generated, enabling accurate identification of defects such as transmittance degradation and temperature anomalies in photovoltaic modules. Compared to traditional manual inspection, this improves the efficiency and accuracy of defect identification. Furthermore, based on the set of potentially defective modules, the power loss contribution weight value caused by defects is calculated, and a dynamic feature matrix is ​​constructed by integrating terrain feature parameters, transmittance loss degradation values, and defect impact weight values. This fully considers the impact of terrain factors and module defects on power generation performance, quantifying the power loss caused by defects. Furthermore, the dynamic feature matrix is ​​input into the machine learning model to generate a power generation performance prediction curve. Combined with the defect impact weight matrix, a maintenance decision list is generated through weighted fusion. This not only allows the machine learning model to capture the time series variation pattern of power generation performance and improve prediction accuracy, but also scientifically evaluates maintenance priorities based on the prediction results and defect impact weights, achieving a reasonable allocation of maintenance resources. This, in turn, improves the accuracy of photovoltaic array power generation performance prediction and the efficiency of operation and maintenance management.

[0058] In one embodiment, based on infrared image data and a transmittance loss feature matrix in an initial dataset, components exhibiting transmittance degradation and temperature anomalies are identified, generating a set of potentially defective components, including:

[0059] The infrared image data is processed by image segmentation based on adaptive temperature threshold to identify areas where the temperature exceeds the preset offset value of the ambient temperature, and output hot spot feature data including hot spot location coordinates and hot spot area ratio.

[0060] Spatial matching is performed between the locations of components whose transmittance loss values ​​exceed the first threshold in the transmittance loss feature matrix and the location coordinates in the hot spot feature data to generate a set of candidate defect locations.

[0061] Based on spectral image data, for each component in the candidate defect location set, contrast and entropy features are extracted through gray-level co-occurrence matrix, the hidden crack density value is calculated, and components with hidden crack density values ​​exceeding the second threshold and corresponding hot spot area proportions exceeding the third threshold are selected to generate a potential defect component set.

[0062] Specifically, the adaptive temperature threshold is used to distinguish between normal temperature regions and abnormally high temperature regions. The preset ambient temperature offset value can be pre-set based on the temperature range of the photovoltaic module under normal operating conditions. For example, if the temperature range of the photovoltaic module under normal operating conditions is the ambient temperature plus a certain temperature rise, then a specific offset value can be added to the ambient temperature as the adaptive temperature threshold. This offset value can be adjusted based on empirical data from actual applications to ensure accurate identification of abnormally high temperature regions. The image segmentation algorithm can analyze the temperature values ​​of infrared images and segment them into different regions. For each pixel, if its temperature value exceeds the adaptive temperature threshold, it is marked as an abnormally high temperature region. By segmenting all pixels whose temperature exceeds the adaptive temperature threshold, a binary image can be formed, where the pixel value of the abnormally high temperature region is 1, and the pixel value of the normal temperature region is 0. This binary image can identify regions whose temperature exceeds the preset ambient temperature offset value, i.e., hot spot regions. Furthermore, for each hot spot region, its position coordinates and hot spot area percentage are extracted. The position coordinates can be determined by calculating the centroid coordinates of the hot spot region, which is the average of the coordinates of all pixels within the hot spot region. The hot spot area ratio can be obtained by calculating the ratio of the hot spot area to the total area of ​​the photovoltaic module. By combining the location coordinates of each hot spot area with its area ratio, hot spot characteristic data can be generated.

[0063] Furthermore, in the transmittance loss feature matrix, components with transmittance loss values ​​exceeding a first threshold are selected. This first threshold can be preset based on the transmittance loss range of the photovoltaic module under normal operating conditions. For example, if the transmittance loss range of the photovoltaic module under normal operating conditions is 0% to 5%, the first threshold can be set to 6%. This threshold can be adjusted based on empirical data from actual applications to ensure accurate selection of components with abnormal transmittance loss. Subsequently, spatial matching is performed between the locations of the selected components with transmittance loss values ​​exceeding the first threshold and the location coordinates in the hot spot feature data. If the distance between the component location and the hot spot location coordinates is within a certain range, such as less than a preset distance threshold, it can be considered that the component and the hot spot region have a spatial association, generating a candidate defect location set. For each component in the candidate defect location set, spectral image data can be converted into a grayscale image, and a preset window can be slid across the image to calculate the grayscale relationship between pixels within the window. Based on the calculation results, a grayscale co-occurrence matrix (GCM) can be constructed. The GCM is a matrix used to describe the texture features of an image, reflecting the grayscale relationship between pixels in the image. By calculating the gray-level co-occurrence matrix, texture features of an image, such as contrast and entropy, can be extracted. Contrast reflects the sharpness of the image's texture, while entropy reflects the complexity of the image's texture.

[0064] Specifically, hidden cracks are a common defect in photovoltaic (PV) modules, leading to a decrease in power generation performance. Based on extracted contrast and entropy features, the hidden crack density can be calculated using a mathematical model relating it to contrast and entropy. A second threshold based on hidden crack density and a third threshold based on hot spot area ratio are then established. These thresholds can be set according to the range of hidden crack density and hot spot area ratio under normal operating conditions. Finally, modules with hidden crack density values ​​exceeding the second threshold and corresponding hot spot area ratios exceeding the third threshold are selected, representing modules with a higher defect risk and generating a set of potentially defective modules.

[0065] In one embodiment, based on a set of potentially defective components, measured current and voltage data of associated strings are obtained. The power loss contribution weight value caused by defects is calculated using a mismatch power analysis method, generating a defect impact weight matrix, including:

[0066] Based on the set of potentially defective components, locate the associated strings, obtain the measured string current and voltage data of the associated strings, and generate the measured power curve.

[0067] The theoretical power value of the associated string is calculated based on the preset standard test conditions, and the power loss value is calculated by the difference between the theoretical power value and the measured power curve.

[0068] Using a multiple linear regression algorithm, the hot spot area ratio and hidden crack density values ​​in the hot spot feature data are used as independent variables, and the power loss value is used as the dependent variable to construct a multiple linear regression model.

[0069] Based on the regression coefficients of the multiple linear regression model, the weighting coefficients of hot spot defects and hidden crack defects on the power loss value are analyzed, and a defect influence weight matrix is ​​generated.

[0070] Specifically, in a photovoltaic (PV) array, modules are typically connected in strings, with each string containing multiple PV modules. Based on the set of potentially defective modules, the string containing each potentially defective module can be determined by analyzing the electrical connection diagram and module layout diagram of the PV array. After locating the associated strings, power quality monitoring equipment installed in the PV array can collect measured current and voltage data for these strings, including current and voltage values ​​measured at different time points. This data can then be used to calculate the actual power generation of the associated strings at each time point. Subsequently, by plotting the power change curve over time—the measured power curve—the changes in the power generation performance of the strings can be visually observed.

[0071] Furthermore, the preset standard test conditions are the performance test conditions of photovoltaic modules under standard environmental conditions, typically including specific irradiance, temperature, and air quality. Based on these conditions, the theoretical power value of the associated strings can be calculated according to the rated power of the photovoltaic modules and the string connection method. Subsequently, the power loss value can be calculated by comparing the difference between the theoretical power value and the measured power curve. This power loss value reflects the degree of power reduction of the strings due to factors such as module defects and changes in environmental conditions during actual operation. In addition, by collecting the hot spot area ratio, hidden crack density value, and corresponding power loss value at multiple time points, a dataset is constructed. A multiple linear regression model is obtained by using a multiple linear regression algorithm to determine the regression coefficient fitting model by minimizing the sum of squared errors. The regression coefficients of this model can represent the influence of the hot spot area ratio and hidden crack density on the power loss value, respectively. The larger the absolute value of the regression coefficient, the more significant the influence of this factor on the power loss. Based on these regression coefficients, a defect influence weight matrix can be generated. This defect influence weight matrix stores the weight coefficients of each potentially defective module in matrix form, thus providing a quantitative basis for subsequent maintenance decisions.

[0072] In one embodiment, the terrain feature parameters, the degradation values ​​in the transmittance loss feature matrix, and the weight values ​​in the defect impact weight matrix are fused to construct a dynamic feature matrix. The terrain feature parameters include terrain slope and roughness parameters, including:

[0073] Based on terrain data, terrain slope and roughness parameters are extracted through a digital elevation model to generate a terrain complexity index.

[0074] From meteorological data in the initial dataset, the average wind speed and precipitation frequency of the target area are obtained to generate a set of environmental corrosion factors.

[0075] The degradation values ​​in the transmittance loss feature matrix and the weight values ​​in the defect impact weight matrix are multiplied by the terrain complexity index to generate primary fusion features.

[0076] A dynamic feature matrix is ​​generated by using a three-layer convolutional neural network to perform feature fusion encoding on primary fusion features and environmental corrosion factors.

[0077] Specifically, digital elevation modeling (DEM) is a three-dimensional terrain representation method that provides terrain elevation information. Based on terrain data, the terrain slope can be obtained by calculating the square root of the sum of the squares of the rates of change of elevation in the horizontal and vertical directions. The roughness parameter can be represented by terrain undulation, defined as the difference between the maximum and minimum elevations within a preset window. By weighted fusion of the terrain slope and roughness parameters, a terrain complexity index can be obtained. Subsequently, the average wind speed and precipitation frequency of the target area can be obtained from meteorological data in the initial dataset. The average wind speed can be calculated using a 30-day moving average to eliminate the influence of short-term fluctuations, and the precipitation frequency can be defined as the proportion of days with precipitation to the total number of days within a certain period. Furthermore, based on material corrosion theory, power function models and linear models can be established to calculate the wind speed factor and precipitation factor to assess their impact on photovoltaic modules. By comprehensively considering the synergistic effect of both and performing a weighted summation, a set of environmental corrosion factors can be obtained.

[0078] By performing matrix multiplication operations on the degradation values ​​in the transmittance loss feature matrix and the weight values ​​in the defect impact weight matrix with the terrain complexity index, two primary fusion features are generated. These features reflect the combined impact of transmittance loss, defect impact weights, and terrain complexity. Finally, the primary fusion features and environmental corrosion factors are input into a three-layer convolutional neural network for feature fusion encoding, generating a dynamic feature matrix. The convolutional neural network is a deep learning model capable of automatically extracting features from the input data. This dynamic feature matrix integrates information from terrain features, transmittance loss, defect impact weights, and environmental corrosion factors, providing comprehensive feature input for subsequent power generation performance prediction.

[0079] In one embodiment, the dynamic feature matrix is ​​input into a time-series analysis-based machine learning model for processing to generate a power generation performance prediction curve, including:

[0080] Construct a machine learning model, which includes a first channel and a second channel;

[0081] The dynamic feature matrix is ​​input into the first channel, and the temporal features are extracted through the LSTM layer. The key time node features are weighted based on the self-attention mechanism, and the temporal feature vector is output.

[0082] The sensor data from the initial dataset is input into the second channel, and a nonlinear transformation is performed through a fully connected layer to output an environmental feature vector.

[0083] The initial power prediction curve is generated by dynamically weighting and fusing the time feature vector and the environmental feature vector through a gating mechanism.

[0084] By acquiring and optimizing the model parameters of the machine learning model based on the actual power generation data of the photovoltaic array through backpropagation algorithm, a power generation performance prediction curve is generated.

[0085] Specifically, the first channel processes the dynamic feature matrix to uncover temporal dependencies in the data, while the second channel processes sensor data from the initial dataset to extract features related to environmental factors. Illustratively, this machine learning model can be built using the deep learning framework TensorFlow. After the dynamic feature matrix is ​​input into the first channel, it first enters the Long Short-Term Memory (LSTM) network layer. This layer, through the collaborative work of the forget gate, input gate, and output gate, effectively handles long-term dependencies in time-series data. Multiple memory units can be set in this layer; during training, each memory unit learns weight parameters to capture the changing patterns of the dynamic feature matrix at different time steps. Furthermore, a self-attention mechanism is introduced based on the output of this layer. Through this self-attention mechanism, the correlation between features at different time points can be calculated, assigning corresponding weights to each time point to highlight features at key time points, such as features after equipment maintenance or before and after extreme weather, ultimately outputting a temporal feature vector.

[0086] Specifically, sensor data from the initial dataset, namely irradiance and temperature data, are input into the second channel. First, the data is normalized using the Min-Max normalization method to map it to the [0,1] interval, thus eliminating dimensional differences between different features. Then, the normalized data is input into a fully connected layer for nonlinear transformation, enabling the extraction of high-order features related to environmental factors from the sensor data, such as irradiance variation trends and the nonlinear impact of temperature on power generation performance, ultimately outputting an environmental feature vector. Furthermore, a gating mechanism dynamically weights and fuses the time and environmental feature vectors. Illustratively, the gating mechanism consists of two gating units: a time gate and an environment gate. The time gate weights the time feature vector using calculated weights, and the environment gate similarly weights the environmental feature vector. By adding the two weighted vectors, the fused feature vector is obtained. This fused feature vector is input into the output layer, and a linear activation function is used. Through linear transformation, an initial power prediction curve is obtained. This initial power prediction curve reflects the preliminary prediction result of the photovoltaic array's power generation based on the current model parameters and input data.

[0087] Furthermore, actual power generation data of the photovoltaic array can be collected through power quality monitoring equipment, reflecting the actual power generation of the photovoltaic array at different time points. The actual power generation data is compared with the predicted values ​​corresponding to the initial power prediction curve, and the mean squared error is calculated as the loss function. Subsequently, based on the backpropagation algorithm, the model parameters are updated along the gradient descent direction by calculating the gradient of the loss function with respect to the model parameters, thereby reducing the loss function value. The Adam optimizer is selected as the optimizer. During training, the Adam optimizer can adaptively adjust the learning rate according to the gradients of different parameters, accelerating the convergence speed of the model. After multiple rounds of iterative training, training stops when the loss function value converges to a preset loss threshold. The model obtained at this point is the final machine learning model, and its output prediction curve is the power generation performance prediction curve.

[0088] In one embodiment, based on the power generation performance prediction curve and the defect impact weight matrix, a maintenance priority score is calculated using a weighted scoring algorithm to generate a maintenance decision list with geographic coordinates, including:

[0089] The power decline rate within a preset period is calculated from the power generation performance prediction curve using the time difference method.

[0090] Read the defect contribution weight value of each component from the defect impact weight matrix;

[0091] The degradation values ​​in the transmittance loss feature matrix are processed by a normalization algorithm to generate normalized degradation values.

[0092] The power degradation rate, defect contribution weight value and normalized degradation value are weighted and integrated by a preset weighted summation formula to generate a maintenance priority score.

[0093] Based on the maintenance priority score, the geographical location markers of the corresponding components in the transmittance loss feature matrix are associated, and a maintenance decision list is generated by sorting in descending order.

[0094] Specifically, the power generation performance prediction curve is a time series data set, reflecting the expected power generation of the photovoltaic array at different time points. The time difference method is a method for calculating the rate of change of time series data; it obtains the power variation by calculating the power difference between adjacent time points. By calculating the power decline rate at each time point, a power decline rate curve can be obtained, reflecting the power change trend of the photovoltaic array within a preset period. The defect impact weight matrix stores the weight coefficients of each potentially defective component; these weight coefficients reflect the degree of impact of defects on power generation performance. Illustratively, an index mapping relationship can be established using component IDs to ensure the rapid and accurate acquisition of the defect weight vector of any component. Furthermore, the degradation values ​​in the transmittance loss feature matrix can be mapped to a standard normal distribution using the Z-score normalization method to eliminate the influence of dimensions. Subsequently, the analytic hierarchy process (AHP) can be used to construct a judgment matrix, calculate the weights of the power decline rate, defect contribution weight value, and normalized degradation value, and then perform a weighted summation to obtain the maintenance priority score.

[0095] Based on maintenance priority scores, the geographic location markers of corresponding components in the transmittance loss feature matrix can be associated. The geographic location marker for each component is stored in the metadata of the feature matrix. Then, the association between the score and geographic location is established using the component ID, generating a list of tuples containing the score and geographic coordinates. This list is then sorted in descending order according to the maintenance priority score. The sorted list constitutes the maintenance decision checklist, allowing maintenance personnel to plan specific maintenance routes and resource allocations based on the priority and geographic location information in the checklist.

[0096] like Figure 2 As shown, based on the same inventive concept, this application also provides a photovoltaic array power generation performance prediction device 200 that combines UAV inspection data to implement the above-mentioned method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of a photovoltaic array power generation performance prediction device combining UAV inspection data provided below can be found in the above-described limitations of a photovoltaic array power generation performance prediction method combining UAV inspection data, and will not be repeated here. The device includes:

[0097] The multi-source data acquisition module 201 is used to acquire an initial dataset, which includes sensor data, UAV inspection data, terrain data and meteorological data. The sensor data includes irradiance data and temperature data, and the UAV inspection data includes visible light image data, infrared image data and spectral image data.

[0098] The image analysis and defect identification module 202 is used to identify the boundaries of photovoltaic modules based on visible light image data in the initial dataset through image segmentation algorithm, extract the geographical location markers of each module, calculate the transmittance loss value of each module, generate a transmittance loss feature matrix including the transmittance loss value and the corresponding geographical location markers of the modules, and identify modules with transmittance degradation and temperature anomalies based on infrared image data and transmittance loss feature matrix in the initial dataset, thereby generating a set of potentially defective modules.

[0099] The data fusion and matrix construction module 203 is used to obtain the measured current and voltage data of the associated strings based on the set of potential defective components, calculate the power loss contribution weight value caused by defects through the mismatch power analysis method, generate the defect impact weight matrix, extract the terrain feature parameters from the terrain data, and fuse the terrain feature parameters, the degradation value in the transmittance loss feature matrix and the weight value in the defect impact weight matrix to construct a dynamic feature matrix.

[0100] The performance prediction and maintenance decision module 204 is used to input the dynamic feature matrix into a machine learning model based on time series analysis for processing, generate a power generation performance prediction curve, and calculate the maintenance priority score through a weighted scoring algorithm based on the power generation performance prediction curve and the defect impact weight matrix, and generate a maintenance decision list with geographic coordinates.

[0101] In the aforementioned device, the multi-source data acquisition module 201 acquires an initial dataset, enabling a comprehensive understanding of environmental variables, component status, and terrain features affecting the photovoltaic array's power generation performance. This lays a multi-dimensional data foundation for subsequent defect identification and performance prediction. The image analysis and defect identification module 202 performs component boundary identification and transmittance loss calculation based on visible light image data. It also combines infrared image data to generate a set of potentially defective components, achieving visualized detection of photovoltaic component transmittance degradation and temperature anomalies. This improves the efficiency and accuracy of defect identification and reduces detection blind spots. The data fusion and matrix construction module 203 quantifies the contribution weight of defects to power loss using mismatch power analysis. It integrates terrain feature parameters and transmittance degradation values ​​to construct a dynamic feature matrix, enabling in-depth analysis of the nonlinear correlation between terrain shading, component defects, and power generation performance. The performance prediction and maintenance decision module 204 inputs the dynamic feature matrix into the time series machine learning model to generate prediction curves, and combines the defect impact weight matrix to generate a maintenance decision list with geographical coordinates. This not only captures the time-varying pattern of power generation performance, but also achieves geographically accurate allocation of maintenance resources through weighted scoring, thereby improving the level of intelligent operation and maintenance of photovoltaic power plants.

[0102] In one exemplary embodiment, the present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the photovoltaic array power generation performance prediction method combining UAV inspection data according to this application. A multi-core processor is preferred to improve the system's parallel processing capability. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate a large amount of supply information and computational tasks.

[0103] In one exemplary embodiment, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a photovoltaic array power generation performance prediction method combining UAV inspection data according to the present application.

[0104] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for predicting the power generation performance of a photovoltaic array by combining UAV inspection data, characterized in that, The method includes: Obtain an initial dataset, which includes sensor data, UAV inspection data, terrain data, and meteorological data. The sensor data includes irradiance data and temperature data, and the UAV inspection data includes visible light image data, infrared image data, and spectral image data. Based on the visible light image data in the initial dataset, the boundaries of photovoltaic modules are identified by an image segmentation algorithm, and the geographical location markers of each module are extracted. The transmittance loss value of each module is calculated, and a transmittance loss feature matrix including the transmittance loss value and the corresponding geographical location markers of the modules is generated. Based on the infrared image data in the initial dataset and the transmittance loss feature matrix, modules with transmittance degradation and temperature anomalies are identified, and a set of potentially defective modules is generated. Based on the set of potentially defective components, the measured current and voltage data of the associated strings are obtained. The power loss contribution weight value caused by the defect is calculated by the mismatch power analysis method, and a defect influence weight matrix is ​​generated. The terrain feature parameters are extracted from the terrain data, and the terrain feature parameters, the degradation value in the transmittance loss feature matrix and the weight value in the defect influence weight matrix are fused to construct a dynamic feature matrix. The dynamic feature matrix is ​​input into a machine learning model based on time series analysis for processing to generate a power generation performance prediction curve. Based on the power generation performance prediction curve and the defect impact weight matrix, a maintenance priority score is calculated using a weighted scoring algorithm to generate a maintenance decision list with geographic coordinates.

2. The method according to claim 1, characterized in that, The step involves identifying components with transmittance degradation and temperature anomalies based on the infrared image data in the initial dataset and the transmittance loss feature matrix, generating a set of potentially defective components, including: The infrared image data is processed by image segmentation based on an adaptive temperature threshold to identify areas where the temperature exceeds a preset offset value of the ambient temperature, and output hot spot feature data including hot spot location coordinates and hot spot area ratio. Spatial matching is performed between the locations of components whose transmittance loss values ​​exceed the first threshold in the transmittance loss feature matrix and the location coordinates in the hot spot feature data to generate a set of candidate defect locations. Based on the spectral image data, for each component in the candidate defect location set, contrast and entropy features are extracted using the gray-level co-occurrence matrix, the hidden crack density value is calculated, and components whose hidden crack density value exceeds a second threshold and whose corresponding hot spot area ratio exceeds a third threshold are selected to generate the potential defect component set.

3. The method according to claim 1, characterized in that, Based on the set of potentially defective components, the measured current and voltage data of the associated strings are obtained. The power loss contribution weight value caused by the defects is calculated using the mismatch power analysis method, generating a defect impact weight matrix, including: Based on the set of potentially defective components, locate the associated string, obtain the measured string current and voltage data corresponding to the associated string, and generate a measured power curve. The theoretical power value of the associated string is calculated based on the preset standard test conditions, and the power loss value is calculated by the difference between the theoretical power value and the measured power curve. Using a multiple linear regression algorithm, the hot spot area ratio and the hidden crack density value in the hot spot feature data are used as independent variables, and the power loss value is used as the dependent variable to construct a multiple linear regression model. Based on the regression coefficients of the multiple linear regression model, the weighting coefficients of hot spot defects and hidden crack defects on the power loss value are analyzed, and the defect influence weight matrix is ​​generated.

4. The method according to claim 1, characterized in that, The method involves fusing the terrain feature parameters, the degradation values ​​in the transmittance loss feature matrix, and the weight values ​​in the defect influence weight matrix to construct a dynamic feature matrix. The terrain feature parameters include terrain slope and roughness parameters, including: Based on the terrain data, the terrain slope and roughness parameters are extracted using a digital elevation model to generate a terrain complexity index. From the meteorological data in the initial dataset, the average wind speed and precipitation frequency of the target area are obtained to generate a set of environmental corrosion factors. The degradation values ​​in the transmittance loss feature matrix and the weight values ​​in the defect influence weight matrix are respectively multiplied by the terrain complexity index to generate a primary fusion feature. The dynamic feature matrix is ​​generated by performing feature fusion encoding on the primary fusion features and the environmental corrosion factors through a three-layer convolutional neural network.

5. The method according to claim 1, characterized in that, The step of inputting the dynamic feature matrix into a machine learning model based on time series analysis for processing to generate a power generation performance prediction curve includes: Construct the machine learning model, which includes a first channel and a second channel; The dynamic feature matrix is ​​input into the first channel, and the temporal features are extracted through the LSTM layer. The key time node features are weighted based on the self-attention mechanism, and the temporal feature vector is output. The sensor data in the initial dataset is input into the second channel, and a nonlinear transformation is performed through a fully connected layer to output an environmental feature vector. The time feature vector and the environment feature vector are dynamically weighted and fused using a gating mechanism to generate an initial power prediction curve. The actual power generation data of the photovoltaic array is acquired and the model parameters of the machine learning model are iteratively optimized through the backpropagation algorithm to generate the power generation performance prediction curve.

6. The method according to claim 1, characterized in that, The step involves calculating a maintenance priority score based on the power generation performance prediction curve and the defect impact weight matrix using a weighted scoring algorithm, and generating a maintenance decision list with geographic coordinates, including: The power decrease rate within a preset period is calculated from the power generation performance prediction curve using the time difference method. Read the defect contribution weight value corresponding to each component from the defect impact weight matrix; The degradation values ​​in the transmittance loss feature matrix are processed by a normalization algorithm to generate normalized degradation values. The power decline rate, the defect contribution weight value, and the normalized degradation value are weighted and fused using a preset weighted summation formula to generate the maintenance priority score. Based on the maintenance priority score, the geographical location markers of the corresponding components in the transmittance loss feature matrix are associated, and the maintenance decision list is generated by sorting in descending order.

7. A photovoltaic array power generation performance prediction device combining UAV inspection data, characterized in that, The device includes: A multi-source data acquisition module is used to acquire an initial dataset, which includes sensor data, UAV inspection data, terrain data, and meteorological data. The sensor data includes irradiance data and temperature data, and the UAV inspection data includes visible light image data, infrared image data, and spectral image data. The image analysis and defect identification module is used to identify the boundaries of photovoltaic modules based on the visible light image data in the initial dataset through an image segmentation algorithm, extract the geographical location markers of each module, calculate the transmittance loss value of each module, generate a transmittance loss feature matrix including the transmittance loss value and the corresponding geographical location markers of the modules, and identify modules with transmittance degradation and temperature anomalies based on the infrared image data in the initial dataset and the transmittance loss feature matrix, thereby generating a set of potentially defective modules. The data fusion and matrix construction module is used to obtain the measured current and voltage data of the associated strings based on the set of potential defective components, calculate the power loss contribution weight value caused by the defect through the mismatch power analysis method, generate the defect influence weight matrix, extract the terrain feature parameters from the terrain data, and fuse the terrain feature parameters, the degradation value in the transmittance loss feature matrix and the weight value in the defect influence weight matrix to construct a dynamic feature matrix. The performance prediction and maintenance decision module is used to input the dynamic feature matrix into a machine learning model based on time series analysis for processing, generate a power generation performance prediction curve, and calculate a maintenance priority score based on the power generation performance prediction curve and the defect impact weight matrix through a weighted scoring algorithm, thereby generating a maintenance decision list with geographic coordinates.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images