Photovoltaic power generation real-time monitoring method and system based on multi-source data fusion

Abstract

The application discloses a photovoltaic power generation real-time monitoring method and system based on multi-source data fusion, and belongs to the photovoltaic power generation field, and comprises the following steps: photovoltaic information acquisition, physical fault classification, image defect classification and fatigue test. The application records irradiance and module backboard temperature, judges whether quality optimization processing is carried out on the I-V curve obtained by scanning, then carries out nonlinear fitting on the I-V curve by using a preset photovoltaic cell physical model, inputs the extracted key physical parameters into a fault classification model, outputs a fault classification result, then inputs the cell piece image of the photovoltaic module string identified as a fault into an instance segmentation model, outputs a defect classification result, and finally obtains the power loss of the photovoltaic module based on the fault classification result and the defect classification result, and formulates a power dispatching strategy, so that the fault attribution effectiveness in photovoltaic power generation real-time monitoring is improved, and the problem of low fault attribution effectiveness in photovoltaic power generation real-time monitoring in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and in particular to a method and system for real-time monitoring of photovoltaic power generation based on multi-source data fusion. Background Technology

[0002] Photovoltaic power generation systems, based on the photovoltaic effect of semiconductors, directly convert solar energy into direct current (DC) electricity, which is then converted into alternating current (AC) by power electronic devices such as inverters for use by loads or fed into the power grid. To improve the flexibility and reliability of energy utilization, modern photovoltaic systems are often equipped with energy storage devices, controllers, and intelligent management systems to achieve optimized storage and dispatch of electrical energy.

[0003] However, photovoltaic (PV) power plants are typically deployed in complex outdoor environments, exposed to prolonged conditions such as sunlight, rain, and sandstorms. This inevitably leads to performance degradation of the PV modules and the potential for various malfunctions. For instance, PV arrays may experience structural vibrations under wind loads and other external factors, which can cause fatigue damage or even failure of the modules over time. These malfunctions not only reduce the power plant's efficiency but can also pose serious safety hazards. Therefore, efficient and accurate condition monitoring and fault diagnosis of PV strings and modules are crucial for ensuring the safe, stable, and efficient operation of PV power plants.

[0004] Among them, the scanning analysis based on the IV characteristic curve can obtain a current-voltage curve (IV curve) containing rich electrical information by scanning the current response of the photovoltaic string under different voltages, which can macroscopically reflect the overall health status of the string. Using physical models (such as single-diode models and dual-diode models) to fit the measured IV curve, and inputting the IV curve into a classification model (such as random forest or convolutional neural network), the fault category to which it belongs can be directly output.

[0005] Meanwhile, drones equipped with visible light cameras, infrared thermal imaging cameras, or electroluminescent cameras are used to acquire large-scale, high-resolution images of photovoltaic arrays. The massive amounts of image data acquired are processed by image segmentation models (such as U-Net and Mask R-CNN) to outline the precise contours of defects such as hot spots, cracks, and occlusions with pixel-level accuracy, providing more detailed defect information.

[0006] For example, Chinese patent application CN116111951B discloses a data monitoring system based on photovoltaic power generation. The retrieval unit obtains the operating parameters of each component in the photovoltaic power generation unit through the data detection unit to monitor the photovoltaic power generation system in real time and promptly identify the cause of the photovoltaic power generation system's failure. The first analysis unit preliminarily determines the operating status of the photovoltaic power generation unit based on the voltage frequency of the power consumption terminal. The second analysis unit further determines the cause of the failure of the photovoltaic power generation unit based on the average temperature value of each photovoltaic panel, the battery charge value, and the battery consumption rate. The alarm unit issues corresponding alarm information based on the failure cause output by the second analysis unit, thus judging the operating status of the photovoltaic power generation unit from multiple aspects.

[0007] For example, Chinese patent application CN119070740A discloses a photovoltaic power generation efficiency monitoring system and method, a monitoring center, and photovoltaic module data acquisition module, environmental data acquisition module, power generation anomaly analysis module, power generation anomaly processing module, power generation processing module, and power generation efficiency visualization module connected to the monitoring center; the power generation anomaly analysis module is connected to the photovoltaic module data acquisition module and the environmental data acquisition module respectively; the power generation anomaly analysis module, power generation anomaly processing module, power generation processing module, and power generation efficiency visualization module are connected sequentially.

[0008] The above-mentioned technology has at least the following technical problems: In existing technologies, photovoltaic power generation performance is strongly correlated with the external environment. However, current diagnostic systems generally lack effective linkage with dynamic environments, resulting in fundamental flaws in the data foundation for diagnosis. Therefore, it is difficult to distinguish whether a decrease in power generation is due to external environmental factors such as weather changes or internal module faults, leading to low accuracy in fault diagnosis. Furthermore, the acquisition process of IV curves is highly susceptible to interference from non-ideal operating conditions. Rapid changes in irradiance at dawn / dusk, random cloud shadows, localized temperature differences on the module surface, and uneven dust settling all mean that the acquired IV curves are not obtained under the theoretically assumed steady-state conditions. Diagnosis based on non-steady-state data lacks reliability.

[0009] Given the already weak data foundation, early fault characteristics of photovoltaic modules (such as initial microcracks and slight potential-induced degradation effects) are weakly reflected on the IV curve and are highly susceptible to interference from changes in operating conditions such as illumination, temperature, and load, making it difficult to effectively identify early fault characteristics of photovoltaic modules. Furthermore, when multiple faults occur concurrently (such as a module simultaneously exhibiting potential-induced degradation effects, microcracks, and poor wiring), its IV curve will display extremely complex and atypical distortions. Traditional single / dual diode models, as idealized physical models, cannot accurately fit this complex morphology caused by the coupling of multiple mechanisms. Although theoretically more complex three-diode models or higher-order models can be constructed, this results in enormous computational costs, making it difficult to apply to real-time online analysis of large-scale power plants, and leading to low effectiveness of fault attribution in real-time monitoring of photovoltaic power generation. Summary of the Invention

[0010] This invention provides a method and system for real-time monitoring of photovoltaic power generation based on multi-source data fusion. This method improves the accuracy of parameter acquisition in real-time photovoltaic power generation monitoring, thereby enhancing the effectiveness of fault attribution in real-time photovoltaic power generation monitoring. The technical solution provided by this application is as follows: Firstly, a real-time monitoring method for photovoltaic power generation based on multi-source data fusion is provided. The specific implementation of this method is as follows: S1. Scan the IV curves of all photovoltaic strings in the photovoltaic power station, and simultaneously record the irradiance and module backsheet temperature during the scanning process. Determine whether to discard the scan data. If discarding the scan data, re-scan the IV curves; otherwise, perform quality optimization processing on the IV curves obtained from the scan, which characterize the output characteristics of the photovoltaic strings under the current operating conditions. S2. Use a preset photovoltaic cell physical model to perform nonlinear fitting on the quality-optimized IV curves, and input the extracted key physical parameters into the fault classification model. Output the fault classification results, including the fault type and the corresponding fault category probability. S3. Obtain the cell images of the photovoltaic strings identified as faulty, and input the cell images into the instance segmentation model. Output the defect classification results, including the defect type, coordinates, segmentation mask, and the corresponding defect category probability. S4. Based on the fault classification results and defect classification results, obtain the power loss of the photovoltaic modules and formulate power dispatch strategies.

[0011] Secondly, a real-time monitoring system for photovoltaic power generation based on multi-source data fusion is provided. This system is applied to a real-time monitoring method for photovoltaic power generation based on multi-source data fusion, and includes: The photovoltaic information acquisition module scans the IV curves of all photovoltaic strings in the photovoltaic power station, simultaneously recording the irradiance and module backsheet temperature during the scanning process. It determines whether to discard the scan data; if discarded, the IV curve scan is repeated. Otherwise, the IV curves representing the output characteristics of the photovoltaic strings under the current operating conditions are optimized. The physical fault classification module uses a pre-defined photovoltaic cell physical model to perform nonlinear fitting on the optimized IV curves, inputting the extracted key physical parameters into the fault classification model and outputting fault classification results including fault type and corresponding fault category probability. The image defect classification module acquires images of the cells in photovoltaic strings identified as faulty, inputs the cell images into an instance segmentation model, and outputs defect classification results including defect type, coordinates, segmentation mask, and corresponding defect category probability. The power dispatch module obtains the power loss of photovoltaic modules based on the fault classification and defect classification results and formulates power dispatch strategies.

[0012] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. By scanning the IV curves of all photovoltaic strings and simultaneously recording irradiance and module backsheet temperature during the scanning process, this invention provides more comprehensive environmental information for subsequent analysis of photovoltaic string output characteristics. Existing technologies often focus only on IV curve scanning, resulting in a single data dimension. This invention effectively avoids analytical errors caused by insufficient consideration of environmental factors. It determines whether to discard data and rescan based on irradiance, ensuring the reliability of the acquired IV curve data. Furthermore, in actual data acquisition, various interference factors may present outliers. This invention optimizes the IV curves by removing outliers, improving data accuracy and usability, laying the foundation for subsequent fault analysis and parameter extraction. While existing technologies may rely on a single machine learning model or simple empirical judgment methods, this invention combines the theoretical accuracy of the physical model with the efficient classification capability of the machine learning model. The nonlinear fitting algorithm accurately matches the IV curves with the physical model, thereby extracting data reflecting photovoltaic power generation. The key physical parameters of performance provide a solid theoretical foundation for fault analysis. The fault classification model, based on these key physical parameters, can quickly and accurately identify whether a photovoltaic string has a fault and the type of fault, providing fault category probabilities and detailed information for subsequent fault handling. Next, images of the solar cells in the identified faulty photovoltaic string are acquired and input into the instance segmentation model. This model can accurately locate and identify defects on the solar cells, providing detailed information about the defects. This helps to gain a deeper understanding of the specific circumstances of the fault, improving the efficiency and accuracy of fault handling, and thus enhancing the effectiveness of fault attribution in real-time photovoltaic power generation monitoring. Finally, based on the fault classification and defect classification results, the power loss of the photovoltaic module is calculated, enabling a more accurate assessment of the performance degradation of the photovoltaic module. This allows for the development of power dispatch strategies based on actual conditions, ensuring the reliability of power resource allocation, reducing losses caused by photovoltaic module failures, and ultimately achieving reliable real-time monitoring of photovoltaic power generation.

[0013] 2. Since environmental variables such as irradiance and temperature in the actual environment can affect the extracted key physical parameters, by correcting the key physical parameters to standard test conditions, the interference of environmental variables can be eliminated, enabling accurate evaluation of photovoltaic power generation performance and making the key physical parameters measured under different environments comparable. Then, an irradiance correction factor and a temperature compensation factor are introduced to correct the current photocurrent, obtaining a corrected photocurrent. This takes into account the influence of actual irradiance and temperature on the photocurrent, improving the accuracy of obtaining the photocurrent parameter. A series resistance correction factor is introduced to reverse compensate the current series resistance, obtaining a corrected series resistance, which can accurately obtain the series resistance value of the photovoltaic cell at standard temperature. Finally, a parallel resistance correction factor is introduced to correct the current parallel resistance, obtaining... Correcting the parallel resistance yields a more accurate value, allowing for a deeper understanding of photovoltaic module performance and potential faults. Compared to existing technologies that only make simple parameter corrections without fully considering the comprehensive impact of complex environmental variables, this invention comprehensively and accurately eliminates the interference of environmental variables on key physical parameters. Finally, the corrected key physical parameters are used as input to a pre-trained fault classification model. The model classifies and judges based on the characteristics of the input parameters, outputting the possible fault types of photovoltaic cells and the probability of each fault type. This enables rapid and accurate identification of photovoltaic cell fault types and provides the probability of fault occurrence. Compared to existing technologies, this significantly reduces the occurrence of misjudgments and omissions due to parameter errors, improving the accuracy and reliability of fault diagnosis.

[0014] 3. Obtain photovoltaic module power loss through fault classification results and defect classification results. If the fault type and defect type are inconsistent, prompt the preset personnel for decision-making and generate a detailed report to avoid misjudgment due to inconsistent classification. If the fault type and defect type are consistent, obtain physical power loss and image power loss separately and calculate them together. This allows for a comprehensive assessment of the photovoltaic module's power loss from different perspectives, improving the accuracy of the assessment results. Then, obtain the physical-image power loss deviation value and select an appropriate power loss acquisition method based on the deviation value. If the physical-image power loss deviation value is not greater than the lower limit of power loss deviation, take the average of the physical power loss and image power loss as the power loss. This allows for the integration of both information when the difference between the two is small, resulting in a more reasonable power loss. If the physical-image power loss deviation value is greater than the lower limit of power loss deviation and... If the power loss deviation is not greater than the upper limit, then through weighted coupling processing, the power loss can be calculated more accurately based on the relative importance of the fault category probability and the defect category probability. If the physical-image power loss deviation value is greater than the upper limit, then a larger power loss value is selected by comparison and judgment. This ensures that when the difference between the two is large, the value that has a more significant impact on the performance of the photovoltaic module is selected as the power loss, which is more in line with the actual situation. Compared with the existing technology that estimates power loss based on simple parameters, this invention can more accurately assess the degree of performance degradation of photovoltaic modules, obtain a more realistic power loss, and finally obtain the expected available power, providing a key basis for subsequent power dispatch decisions. Power dispatch based on the expected available power and the state of charge of energy storage realizes the power balance between photovoltaic power plants and the grid, and improves energy utilization efficiency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a real-time monitoring method for photovoltaic power generation based on multi-source data fusion provided in an embodiment of the present invention; Figure 2 This is a flowchart of photovoltaic power generation monitoring provided in an embodiment of the present invention; Figure 3 This is an architecture diagram of the cell defect feature extraction module in a photovoltaic module provided in an embodiment of the present invention; Figure 4 This is a flowchart of obtaining the power loss of a photovoltaic module provided in an embodiment of the present invention; Figure 5This is a schematic diagram of the structure of a real-time monitoring system for photovoltaic power generation based on multi-source data fusion provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0018] Before providing a detailed explanation of the embodiments of this application, the application scenarios in the embodiments of this application will be described first.

[0019] The real-time photovoltaic (PV) power generation monitoring method provided in this application is applied to the PV power generation operation and maintenance phase. It quickly identifies abnormal strings through IV curve scanning and then precisely locates specific components and defect types using UAV image recognition. This enables step-by-step troubleshooting from the PV power plant to the solar cells, recognizing early, subtle fault characteristics and issuing warnings before the fault evolves into a serious failure. This transforms passive maintenance into proactive maintenance, minimizing PV power generation losses. The real-time PV power generation monitoring method provided in this application can also be applied to the energy management phase of PV power generation. Based on power loss quantification, PV power plants can more accurately predict available power, automatically determine power surplus or deficit, and intelligently select energy storage charging / discharging or grid interaction, providing data support for grid dispatching and energy storage charging / discharging strategies.

[0020] This application provides a method for real-time monitoring of photovoltaic power generation based on multi-source data fusion. For example... Figure 1 The flowchart shown is for a real-time monitoring method for photovoltaic power generation based on multi-source data fusion. The processing flow of this method may include the following steps: S1 performs IV curve scanning on all photovoltaic strings in the photovoltaic power station, simultaneously recording irradiance and module backsheet temperature during the scanning process. It then determines whether to discard the current scan data. If discarded, the IV curve scan is repeated; otherwise, the obtained IV curves, representing the output characteristics of the photovoltaic strings under the current operating conditions, undergo quality optimization to remove outliers. The IV curve describes the relationship between the output current of the photovoltaic string under different output voltages. Irradiance refers to the solar radiation energy projected onto a unit area per unit time. Module backsheet temperature refers to the temperature of the photovoltaic module's backsheet; since photovoltaic modules generate heat during power generation, the backsheet temperature affects the power generation efficiency and performance of the photovoltaic cells. Recording irradiance and module backsheet temperature provides more comprehensive environmental information for subsequent analysis of the photovoltaic string output characteristics. Determining whether to discard data and rescan ensures the reliability of the acquired IV curve data. Quality optimization of the IV curves, removing outliers, improves the accuracy and usability of the data, laying the foundation for subsequent fault analysis and parameter extraction.

[0021] like Figure 2 The flowchart for photovoltaic power generation monitoring shown in S2 involves using a pre-set photovoltaic cell physical model to perform nonlinear fitting on the IV curve after quality optimization, extracting key physical parameters, and inputting these parameters into a fault classification model. The output includes fault type and corresponding fault category probability. In this application, a single-diode model (or a dual-diode model) is used as the photovoltaic cell physical model. The single-diode model is relatively simple, while the dual-diode model considers more physical effects and can more accurately describe the component characteristics. The nonlinear fitting algorithm is a mathematical algorithm used to handle nonlinear relationships between data. In this application, it is used to fit the IV curve to the pre-set photovoltaic cell physical model. The fault classification model is a model built based on machine learning (such as random forest or neural network). By analyzing the input key physical parameters, it determines whether there is a fault in the photovoltaic string and the fault type, and outputs the corresponding fault category probability, providing detailed information for subsequent fault handling. Through comprehensive data collection and processing, and by combining the physical model and the machine learning model, the faults in the photovoltaic string can be identified more accurately, reducing misjudgments and omissions.

[0022] S3. Acquire images of the solar cells in the photovoltaic string identified as faulty, and input the solar cell images into an instance segmentation model for automatically identifying and segmenting defects in the images. Output defect classification results including defect type, coordinates, segmentation mask, and corresponding defect category probability. The instance segmentation model is a computer vision model that can automatically identify target objects in the image and accurately segment each target object, outputting information such as defect type, coordinates, segmentation mask, and corresponding defect category probability. In this application, the target object is a solar cell defect, which improves the efficiency and accuracy of fault handling during photovoltaic power generation monitoring.

[0023] S4. Based on the fault classification results and defect classification results, obtain the power loss of photovoltaic modules and formulate power dispatch strategies. Power loss is the difference between the actual power generation of photovoltaic modules and the theoretical power generation due to faults, defects and other factors. It is an important indicator for evaluating the performance of photovoltaic modules in photovoltaic power generation and formulating power dispatch strategies. By comprehensively considering fault and defect information to calculate power loss, the degree of performance degradation of photovoltaic modules can be assessed more accurately.

[0024] Further, the process for determining whether to discard the scanned data is as follows: The irradiance change is obtained, which reflects the degree of deviation between the irradiance at the start and end of the scan. If the irradiance change is greater than the preset irradiance change threshold, it indicates that the environmental conditions during the scan are unstable, which may have a significant impact on the IV curve data obtained by the scan, resulting in inaccurate data. Therefore, the scan data is discarded to avoid inaccurate data interfering with subsequent analysis and processing, thereby improving the quality and reliability of the data from the source. Otherwise, the IV curve obtained by the scan is optimized.

[0025] The IV curves obtained from the scan are subjected to quality optimization processing, specifically as follows: The IV curve is checked for power non-negativity. This involves multiplying the current and voltage values ​​at each data point on the IV curve to calculate the power at each data point and removing data points with negative power values. Under normal circumstances, the output power of photovoltaic equipment should be non-negative. Data points with negative power values ​​may be due to measurement errors or abnormal conditions, thus ensuring the rationality and accuracy of the IV curve data.

[0026] Sliding window statistical filtering is applied to the IV curve. Specifically, sliding window statistical filtering involves setting a sliding window containing a preset number of data points, calculating the median and standard deviation of the current values ​​of all data points within the sliding window, and recording the absolute value of the difference between the current value of the center data point of the sliding window and the median of the current values ​​of all data points within the sliding window as the absolute deviation of the current value. The preset number is set by the preset personnel based on experience. Since it is necessary to take the center data point within the sliding window, the preset number can be set to an odd number, such as 7.

[0027] If the absolute deviation of the current value is greater than the standard deviation of the current values ​​of all data points within a preset multiple sliding window, the data point at the center of the sliding window is marked as an outlier and removed; otherwise, the outlier is not removed. The preset multiple is set by the preset personnel based on experience, for example, it can be set to 3 times. The median is not sensitive to outliers; by comparing it with the median, it can be preliminarily determined whether the center data point deviates from the normal range. The standard deviation reflects the dispersion of the data; combined with the preset multiple, it can more accurately determine whether the center data point is an outlier. Removing outliers can effectively reduce the impact of noise and outliers on the IV curve and improve the quality of the IV curve.

[0028] Interpolation and moving average processing are performed on the IV curve after data point filtering.

[0029] To fill the data gaps left after outlier removal, appropriate interpolation methods (such as linear interpolation, spline interpolation, etc.) are used to obtain the data point values ​​at the gaps, thereby restoring the continuity of the curve.

[0030] Set a moving average window, slide the window on the IV curve data sequence, calculate the average value of the current or voltage values ​​of the data points within the window, and use the average value to replace the data point value at the center of the window to further smooth out residual high-frequency random noise, making the curve smoother, facilitating subsequent accurate fitting, and improving the accuracy of photovoltaic module performance evaluation during photovoltaic power generation.

[0031] Furthermore, a pre-defined photovoltaic cell physical model is used to perform nonlinear fitting on the quality-optimized IV curve, and key physical parameters are extracted. This process also includes: The Levenberg-Marquardt (LM) algorithm is used to iteratively optimize the extracted key physical parameters. By fitting the objective function between the predicted and measured currents through the sum of squared residuals, the error between the model-predicted current and the measured current is minimized. The LM algorithm is a nonlinear least squares optimization algorithm that combines the advantages of the Gauss-Newton method and the gradient descent method. Its core idea is to dynamically adjust the iteration step size by introducing a damping factor, and to minimize the sum of squared errors between the theoretical and measured curves by iteratively adjusting the parameters. The initial damping factor in the LM algorithm is set by the pre-set personnel based on experience, for example, it can be set to 0.01. This balances the convergence speed and stability, avoids the divergence problem caused by matrix singularity or poor initial values ​​in the Gauss-Newton method, and overcomes the slow convergence of the gradient descent method.

[0032] Key physical parameters are corrected to standard test conditions to eliminate the influence of environmental variables on these parameters, ensuring comparability under different measurement conditions and facilitating accurate evaluation and comparison of photovoltaic module performance. Key physical parameters include photocurrent, diode reverse saturation current, diode ideality factor, series resistance, and parallel resistance. Standard test conditions are set by experienced personnel, such as a temperature of 25°C, irradiance of 1000 W / m², and atmospheric mass of 1.5G (G represents global). Photocurrent is the current generated by the cells in a photovoltaic module under sunlight, representing the conversion of light energy into electricity. The main component generating current during the electrical energy process; the diode reverse saturation current is the current of the diode in the equivalent circuit of the cell in the photovoltaic module when it is reverse biased; the diode ideality factor is used to describe the difference between the diode characteristics and the ideal diode characteristics. The closer the ideality factor is to 1, the closer the diode characteristics are to the ideal situation; the series resistance represents the resistance generated inside the cell in the photovoltaic module due to factors such as materials and electrodes. The presence of series resistance will reduce the output power of the cell in the photovoltaic module; the parallel resistance represents the resistance reflecting the leakage current of the cell in the photovoltaic module. If the parallel resistance is too small, it will lead to an increase in leakage current and affect the performance of the cell in the photovoltaic module.

[0033] An irradiance correction factor and a temperature compensation factor are introduced to correct the current photocurrent, resulting in a corrected photocurrent. The differences between actual irradiance and temperature and standard conditions are considered, making the corrected photocurrent more reflective of the photovoltaic cell's performance under standard test conditions, thus improving the accuracy of the photocurrent. The irradiance correction factor is obtained by calculating the ratio of the average irradiance during the scanning process to the standard irradiance, and is used to correct the impact of the difference between actual and standard irradiance on the photocurrent. The temperature compensation factor is obtained by combining a preset short-circuit current temperature coefficient to correct the module backsheet temperature deviation, which reflects the degree of deviation between the average module backsheet temperature and the standard module backsheet temperature during the scanning process, and is used to correct the photocurrent to eliminate the influence of temperature differences. The module backsheet temperature deviation represents the difference between the average module backsheet temperature and the standard module backsheet temperature during the scanning process, reflecting the temperature change. In this application, "correction" refers to a product operation. The short-circuit current temperature coefficient is set by the pre-set personnel according to industry standards and material characteristics; for example, the short-circuit current temperature coefficient of silicon-based photovoltaic cells is between 0.04% / ℃ and 0.06% / ℃.

[0034] A series resistance correction factor is introduced to compensate for the current series resistance in reverse. This is achieved by calculating the ratio between the current series resistance and the series resistance correction factor to obtain the corrected series resistance. The series resistance correction factor is obtained by correcting the temperature deviation of the module backsheet in conjunction with the temperature coefficient of resistance. The temperature coefficient of resistance is set by the pre-set personnel based on the material characteristics. This eliminates the influence of temperature on the series resistance, making the corrected series resistance more accurately reflect the internal resistance characteristics of the photovoltaic cell, which helps to more accurately analyze the power loss of the cells in the photovoltaic module.

[0035] A parallel resistance correction factor is introduced to correct the current parallel resistance, resulting in a corrected parallel resistance. This correction factor is obtained by combining an empirical coefficient with an exponential compensation for the module backsheet temperature deviation. Specifically, the product of the empirical coefficient and the module backsheet temperature deviation is substituted into the natural exponential function. The empirical coefficient is set by the pre-set personnel based on experimental data or engineering experience. This eliminates the influence of temperature on the parallel resistance, making the corrected parallel resistance more accurately reflect the leakage current of the cells in the photovoltaic module, which helps in evaluating the performance stability of the photovoltaic module.

[0036] Furthermore, the battery cell image is input into an instance segmentation model for automatically identifying and segmenting defects in the image, which previously included: If the fault classification model outputs a fault type of "no fault," it indicates that the photovoltaic module is in normal working condition, directly outputting a normal photovoltaic power generation result, thus improving the operational efficiency of real-time monitoring of photovoltaic power generation. Otherwise, in order to further analyze the fault situation, infrared imaging (or electroluminescence detection) is used to obtain images of the corresponding photovoltaic string cells to fully understand the scope of the fault, providing richer information for further fault location and maintenance, thereby reducing the impact of the fault on photovoltaic power generation.

[0037] Due to various factors affecting image acquisition, solar cell images may exhibit uneven brightness. Therefore, brightness uniformity correction is performed on the solar cell images. Brightness uniformity correction includes flat-field correction and background subtraction based on polynomial fitting. Flat-field correction is used to eliminate image sensor inhomogeneities. Due to inconsistent light responses of pixels in an image sensor (such as a camera) and vignetting effects (where the brightness at the image edges is lower than the center brightness) in optical systems (such as lenses), uneven brightness is present in the acquired images. Flat-field correction eliminates this inhomogeneity by acquiring a reference image (flat-field image) under uniform illumination and performing calculations with the actual acquired image, thereby improving image brightness. More uniform; In image processing, for battery cell images, the background usually has a relatively smooth trend of change. Polynomial fitting can be used to simulate the brightness distribution of the background, and then the fitted background image is subtracted from the original image. Polynomial fitting is a mathematical method that uses appropriate polynomial functions (such as quadratic polynomials, cubic polynomials, etc.) to approximate the trend of data change. Background subtraction based on polynomial fitting further removes background interference in the image, highlights the features of the battery cell and its defects, improves the contrast of the image, makes the defects more obvious, and is conducive to the subsequent instance segmentation model working more accurately, thus improving the recognition accuracy and segmentation precision of the instance segmentation model.

[0038] During image acquisition, random noise may be introduced into the battery cell image due to various interference factors (such as electronic device noise, ambient light interference, etc.). Therefore, median filtering is used to filter out random noise in the battery cell image after brightness uniformity correction to suppress random noise generated during image acquisition. Compared with linear filtering methods such as mean filtering and Gaussian filtering, which tend to blur image edges and details while removing noise, median filtering is a non-linear image filtering method based on ordination statistics theory. For each pixel in the image, a neighborhood (such as a 3×3 or 5×5 square neighborhood) is selected with the pixel as the center. The gray values ​​of all pixels in the neighborhood are sorted, and the median value is taken as the new gray value of the pixel. Median filtering can effectively suppress random noise in the image, especially salt-and-pepper noise (random black and white pixels in the image), while also preserving the edge information of the image to a certain extent, which helps to improve the defect recognition ability and segmentation accuracy of the instance segmentation model.

[0039] It should be added that, such as Figure 3 The diagram shown illustrates the architecture of the solar cell defect feature extraction module in a photovoltaic module. It demonstrates the module's workflow, adapted to instance segmentation models such as Mask Region-based Convolutional Neural Networks (Mask R-CNN) and Selective Kernel Convolution (SK convolution). Specifically, it describes how the model takes an image of a solar cell in the photovoltaic module (such as an electroluminescent image or an infrared thermal imaging image) as input X, and first, in the segmentation phase (Split Phase),... (This refers to convolution operations specifically involving feature extraction from the input feature map using a 3x3 convolution kernel) and (This also represents a convolution operation, corresponding to feature extraction from the input feature map using a 5x5 convolution kernel.) Two types of convolution kernels extract fine-grained (e.g., micrometer-level hidden cracks) and coarse-grained (e.g., large-area hot spots) features respectively, generating dual-branch feature maps. After entering the fusion phase (Fuse phase), the dual-branch features are first fused into an intermediate feature map using element-wise summation, and then... (Global feature compression, typically achieved by compressing a high-dimensional feature map into a vector using global average pooling for subsequent weight calculations, yields a compressed feature vector s (containing the fused global feature information), which is then processed...) (Feature transformation operations typically convert compressed feature vectors into weight vectors corresponding to different convolutional kernel branches through fully connected layers.) A weight vector z is generated (which will subsequently be converted into weights for each branch using softmax). Element-wise product is then used to weight the dual-branch feature maps. Finally, in the Select stage (the selection operation, which involves filtering and integrating effective information based on the weighted multi-scale features to generate the output feature map), the weighted feature maps are integrated, outputting an image V labeled with the defect type, the probability of the corresponding defect type, and the segmentation mask. This result can be directly input into the subsequent Mask R-CNN model to complete instance segmentation. This module effectively adapts to the characteristics of large scale differences and complex features in photovoltaic module cell defects through multi-scale adaptive feature extraction and fusion, and is a core component for improving defect recognition accuracy. SK convolution is used to optimize the performance of Mask R-CNN by replacing or supplementing the original fixed-scale convolutional layers in Mask R-CNN. Through multi-scale parallel convolution and adaptive weighting, it improves the feature capture capability for defects of different scales and types, thereby enhancing the accuracy of subsequent instance segmentation.

[0040] like Figure 4 The flowchart shown here illustrates how the power loss of a photovoltaic module is obtained based on fault classification and defect classification results. The specific process is as follows: If the obtained fault type and defect type are inconsistent, the power loss cannot be directly determined, prompting the pre-set personnel to make a decision and generating a report including IV curves, extracted key physical parameters, cell images, and defect types. This helps the pre-set personnel to fully understand the condition of the photovoltaic modules, thereby making more accurate decisions and avoiding misjudgments caused by inconsistent classifications.

[0041] If the obtained fault type and the obtained defect type are consistent, the physical power loss and image power loss of the photovoltaic module are obtained respectively, and the power loss of the photovoltaic module is obtained based on the physical power loss and image power loss of the photovoltaic module. This allows for a comprehensive assessment of the power loss of the photovoltaic module from different perspectives, thereby improving the accuracy of real-time monitoring of photovoltaic power generation.

[0042] Specifically, the power loss of a photovoltaic module is obtained based on its physical power loss and image power loss. The specific process is as follows: The difference between physical power loss and image power loss is denoted as the physical-image power loss deviation value, which is used to measure the degree of difference between physical power loss and image power loss, and provides a basis for selecting an appropriate power loss calculation method according to different deviation conditions.

[0043] If the physical-image power loss deviation is not greater than the lower limit of the power loss deviation, it means that the physical power loss and the image power loss are relatively close. The average value of the physical power loss and the image power loss is taken as the power loss of the photovoltaic module. Taking the average value as the power loss is simple and easy to implement, and when the difference between the two is small, it can combine the information of the two to obtain a more reasonable power loss value. The upper limit and the lower limit of the power loss deviation are set by the preset personnel based on experience. For example, the upper limit of the power loss deviation can be set as the sum of the absolute value of the physical-image power loss deviation value in the historical time period and three times the standard deviation, and the lower limit of the power loss deviation can be set as the difference between the absolute value of the physical-image power loss deviation value in the historical time period and three times the standard deviation.

[0044] If the physical-image power loss deviation is greater than the lower limit of power loss deviation, but not greater than the upper limit of power loss deviation, then the power loss of the photovoltaic module is obtained by weighted coupling of the physical power loss and image power loss, combined with the corresponding coupling weights. The output fault category probability and defect category probability are normalized to obtain the corresponding coupling weights, that is, the ratio of the fault category probability (or defect category probability) to the sum of the fault category probability and defect category probability is used to obtain the corresponding coupling weight. Through weighted coupling, the power loss can be calculated more accurately according to the relative importance of the fault category probability and defect category probability, thus improving the accuracy of power loss acquisition in photovoltaic power generation monitoring.

[0045] If the physical-image power loss deviation is greater than the upper limit of the power loss deviation, the physical power loss and the image power loss are compared and judged. Specifically, if the physical power loss is greater than the image power loss, the physical power loss of the photovoltaic module is taken as the power loss; if the physical power loss is less than the image power loss, the image power loss of the photovoltaic module is taken as the power loss. By comparing and judging and selecting the larger power loss value, it can ensure that when the difference between the two is large, the value that has a more significant impact on the performance of the photovoltaic module is selected as the power loss, which is more in line with the actual situation.

[0046] Specifically, the physical power loss and image power loss of photovoltaic modules are obtained using the following methods: The obtained defect classification results are input into a preset image power loss mapping set for mapping, resulting in the power loss of each cell in the photovoltaic module. The maximum power loss among all cells is taken as the image power loss of the photovoltaic module. The image power loss mapping set provides a threshold standard for adaptive adjustment of image power loss based on defect classification results. The image power loss mapping set analyzes the image power loss training data based on a nonlinear model (such as a neural network) to find the mathematical relationship between the image power loss training data and the actual image power loss. The image power loss training data includes defect classification results based on historical time periods and image power losses set by professional technicians based on empirical rules. By using the image power loss mapping set and the method of taking the maximum power loss, the image power loss of the photovoltaic module can be quickly and accurately assessed according to the defect situation, reflecting the impact of defects on the power of the photovoltaic module.

[0047] The difference between the standard power of the module and the current theoretical maximum power is denoted as the physical power loss. The theoretical maximum power represents the maximum power at each data point on the IV curve, obtained by substituting the corrected key parameters into the photovoltaic cell physical model and solving the model using Newton's iteration method. Based on the physical model and actual parameters, this accurately reflects the theoretical power loss of the photovoltaic module. Calculating the theoretical maximum power based on the physical model and actual parameters, and then obtaining the physical power loss, accurately reflects the power loss of the photovoltaic module from a physical principle perspective, providing a reliable foundation for subsequent comprehensive evaluation.

[0048] Specifically, the process for formulating power dispatch strategies is as follows: The average power loss is the average power loss of all photovoltaic modules. The difference between the standard power and the average power loss of all photovoltaic modules is recorded as the expected available power, which reflects the actual available power that can be provided under the current conditions and provides a key basis for subsequent power dispatch decisions.

[0049] If the expected available power is greater than the current load power, and the energy storage state of charge is not greater than the charging limit, it means that the generated power has a surplus in addition to meeting the current load demand, and the energy storage device has not reached the charging limit. In this case, the energy storage device is charged first, and the excess electrical energy is stored for later use. The energy storage state of charge represents the ratio of the current stored electrical energy of the energy storage device (such as a battery) to its rated capacity. The charging limit is the highest state of charge that the energy storage device is allowed to reach when charged. Continuing to charge beyond this value may damage the energy storage device. The discharge limit is the lowest state of charge that the energy storage device is allowed to reach when discharged. Continuing to discharge below this value may affect the lifespan and performance of the energy storage device.

[0050] If the expected available power is greater than the current load power and the energy storage state of charge is greater than the charging limit, it indicates that there may be a power surplus. However, since the energy storage device is already fully charged, it cannot continue to charge. Therefore, it outputs electricity to the grid, feeding the excess power back to the grid.

[0051] If the expected available power is not greater than the current load power and the energy storage state of charge is not greater than the discharge limit, it means that the generated power cannot meet the current load demand and the energy storage device has insufficient power to discharge and replenish. In this case, power is received from the grid to ensure the normal operation of the load.

[0052] If the expected available power is not greater than the current load power, and the energy storage state of charge is greater than the discharge limit, it indicates a power deficit. However, the energy storage device still has some charge. In this case, the energy storage device discharges to supplement the insufficient power generated by photovoltaic power generation and meet the load demand. By rationally allocating the power generated during photovoltaic power generation according to different operating conditions, priority is given to meeting the load demand, while also taking into account the charging and discharging management of the energy storage device, thus improving energy utilization efficiency and the stability of photovoltaic power generation.

[0053] like Figure 5 The diagram shows the structure of a real-time photovoltaic power generation monitoring system based on multi-source data fusion. The real-time photovoltaic power generation monitoring system based on multi-source data fusion provided in this application includes: The photovoltaic information acquisition module is used to scan the IV curves of all photovoltaic strings in the photovoltaic power station, and simultaneously record the irradiance and module backsheet temperature during the scanning process. It determines whether to discard the scan data. If the scan data is discarded, the IV curve scan is repeated. Otherwise, the IV curves obtained from the scan, which characterize the output characteristics of the photovoltaic strings under the current operating conditions, are subjected to quality optimization processing to remove outliers. Irradiance directly affects the amount of solar energy received by the photovoltaic modules, while the module backsheet temperature affects the conversion efficiency of the cells in the photovoltaic modules. This makes the subsequent analysis and processing of the IV curves more accurate and reliable, and can more realistically reflect the performance of the photovoltaic strings under actual operating conditions.

[0054] The physical fault classification module uses a pre-set photovoltaic cell physical model to perform nonlinear fitting on the IV curve after quality optimization and extracts key physical parameters. These extracted parameters are then input into the fault classification model, which outputs fault classification results including fault type and corresponding fault category probability. The fault classification model is trained using IV curves labeled with fault types over historical time periods. It can quickly and accurately classify the fault types of photovoltaic strings based on the input parameters and provide fault category probabilities. This allows for more targeted subsequent image defect classification and fault investigation, improving the efficiency of fault handling in photovoltaic power generation monitoring.

[0055] The image defect classification module acquires images of solar cells from photovoltaic strings identified as faulty. This provides intuitive visual information for further analysis of cell defects, helping to discover defects that are difficult to detect through electrical data. It provides an important basis for comprehensively assessing the health status of photovoltaic modules. The module inputs the cell images into an instance segmentation model for automatic identification and segmentation of defects in the images. The output includes defect type, coordinates, segmentation mask, and corresponding defect category probability. The defect type clarifies the type of defect, the coordinates and segmentation mask accurately describe the location and shape of the defect in the image, and the defect category probability reflects the likelihood of each defect occurring. This provides comprehensive and accurate information for personnel to develop targeted maintenance plans, improving maintenance efficiency and quality.

[0056] The power dispatch module is used to obtain the power loss of photovoltaic modules based on fault classification results and defect classification results, and to formulate power dispatch strategies. It can rationally allocate power resources according to the actual power generation capacity of photovoltaic power generation, supplement power from other power sources in a timely manner when photovoltaic power generation is insufficient, and store or transmit excess power to other areas when photovoltaic power generation is excessive, thereby improving the stability and flexibility of power supply.

[0057] Example 2: The extracted key physical parameters are input into the fault classification model, and the output includes fault classification results including fault type and corresponding fault category probability, followed by: Poor data quality (e.g., excessive noise, severe curve distortion) or improper initial value estimation can cause the LM algorithm to fail to converge during iteration, or the output parameters may be physically unreasonable (e.g., series resistance less than 0, diode ideality factor greater than 10). Furthermore, extreme faults that the physical model cannot describe may occur, such as large-area component breakage or severe burnout of the junction box leading to intermittent contact. The shape of these curves may completely exceed the applicability of the single / dual diode model; for example, the coefficient of determination may be extremely low, such as less than 0.95, indicating a complete mismatch between the theoretical and measured curves, thus rendering Example 1 inapplicable.

[0058] The data point with the minimum voltage value is obtained and recorded as the short-circuit current point, with the corresponding current value recorded as the short-circuit current. An irradiance correction factor and a short-circuit temperature correction factor are introduced to correct the short-circuit current, resulting in the corrected short-circuit current. The irradiance correction factor is expressed as the ratio of standard irradiance to average irradiance. The short-circuit temperature correction factor is obtained by correcting the difference between the average component temperature and the standard component temperature, in conjunction with a set temperature coefficient for the short-circuit current. This accurately determines the short-circuit current point. By considering the influence of irradiance and temperature on the short-circuit current through correction factors, the corrected short-circuit current is made closer to the true value, improving the accuracy of fault diagnosis.

[0059] The data point with the largest voltage value is obtained and recorded as the open-circuit voltage point, and the corresponding voltage value is recorded as the open-circuit voltage. After correcting the open-circuit voltage by introducing an open-circuit temperature correction factor, it is coupled with the open-circuit voltage to obtain the corrected open-circuit voltage. The open-circuit temperature correction factor is obtained by correcting the difference between the standard module temperature and the average module temperature by combining the temperature coefficient of the open-circuit voltage. In this application, the coupling process represents a summation operation. Thus, the open-circuit voltage point is accurately determined. The influence of temperature on the open-circuit voltage is considered for correction and coupling processing, so that the corrected open-circuit voltage more accurately reflects the characteristics of the photovoltaic module and is conducive to fault diagnosis.

[0060] The power values ​​of each data point are obtained, and the data point corresponding to the maximum power value is recorded as the maximum power point. Irradiance correction factors and short-circuit temperature correction factors are introduced to correct the current value at the maximum power point, resulting in the corrected current value. An open-circuit temperature correction factor is introduced to correct the voltage value at the maximum power point, and then coupled with the voltage value at the maximum power point to obtain the corrected voltage value. By determining the maximum power point and considering the effects of irradiance and temperature on current and voltage for correction and coupling, more accurate maximum power point parameters are obtained, providing reliable data for subsequent fill factor calculations and fault location.

[0061] The ratio of the corrected power at the maximum power point to the ideal maximum power is recorded as the fill factor, which reflects the photovoltaic module's ability to convert light energy into electrical energy. Based on the fill factor, the corrected power at the maximum power point, the corrected short-circuit current, and the corrected open-circuit voltage, the fault type is obtained by querying a preset rule base, thus achieving more comprehensive and accurate fault diagnosis.

[0062] Meanwhile, the theoretical maximum power represents the product of the corrected short-circuit current and the corrected open-circuit voltage. By obtaining the theoretical maximum power, the subsequent power loss can be quantified.

[0063] It should be added that the specific steps for setting up the preset rule base are as follows: Based on common fault types of photovoltaic modules, such as hot spots, shading, microcracks, and power degradation, determine the fault types to be identified in the rule base; for each fault type, analyze its performance characteristics in parameters such as fill factor, corrected power at maximum power point, corrected short-circuit current, and corrected open-circuit voltage. For example, hot spot faults may lead to a decrease in fill factor and a reduction in corrected short-circuit current; based on the fault characteristics, set specific rule conditions for each fault type. The rule conditions should include threshold or range judgments for parameters such as fill factor, corrected power at maximum power point, corrected short-circuit current, and corrected open-circuit voltage; set a corresponding fault type output for each rule. When the input parameters meet the conditions of a certain rule, the rule base should output the corresponding fault type. Through actual testing and data analysis, the rule conditions in the rule base are continuously optimized to improve the accuracy and reliability of fault diagnosis. For example, when the fill factor is less than 0.7, the maximum power point correction power is less than 80% of the rated power, the correction short-circuit current is less than 90% of the rated short-circuit current, and the correction open-circuit voltage is greater than 90% of the rated open-circuit voltage, it is judged as a hot spot fault. When the fill factor is not less than 0.8, the maximum power point correction power is not less than 90% of the rated power, the correction short-circuit current is not less than the rated short-circuit current, and the correction open-circuit voltage is not less than the rated open-circuit voltage, it is judged as a normal state.

[0064] The above-disclosed embodiments are merely some of the embodiments of the present invention, and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A real-time monitoring method for photovoltaic power generation based on multi-source data fusion, characterized in that, The method includes: S1. Perform IV curve scanning on all photovoltaic strings in the photovoltaic power station, and simultaneously record the irradiance and module backsheet temperature during the scanning process. Determine whether to discard the data of this scan. If the data of this scan is discarded, perform IV curve scanning again. Otherwise, perform quality optimization processing on the IV curve obtained by scanning, which represents the output characteristics of the photovoltaic string under the current operating conditions. S2, using a pre-set photovoltaic cell physical model to perform nonlinear fitting on the IV curve after quality optimization, and inputting the extracted key physical parameters into the fault classification model, outputting fault classification results including fault type and corresponding fault category probability; S3, acquire the cell images of the photovoltaic strings identified as faulty, input the cell images into the instance segmentation model, and output the defect classification results including defect type, coordinates, segmentation mask and corresponding defect category probability; S4. Based on the fault classification results and defect classification results, obtain the power loss of photovoltaic modules and formulate power dispatch strategies; The specific process for obtaining the power loss of photovoltaic modules based on fault classification results and defect classification results is as follows: If the obtained fault type and the obtained defect type are inconsistent, the preset personnel will be prompted to make a decision and a report including the IV curve, extracted key physical parameters, cell images and defect type will be generated. If the obtained fault type and the obtained defect type are consistent, then the physical power loss and image power loss of the photovoltaic module are obtained, and the power loss of the photovoltaic module is obtained based on the physical power loss and image power loss of the photovoltaic module. The specific process for obtaining the power loss of a photovoltaic module based on its physical power loss and image power loss is as follows: The difference between physical power loss and image power loss is denoted as the physical-image power loss deviation value; If the physical-image power loss deviation is not greater than the lower limit of the power loss deviation, then the average value of the physical power loss and the image power loss is taken as the power loss of the photovoltaic module. If the physical-image power loss deviation is greater than the lower limit of the power loss deviation and the physical-image power loss deviation is not greater than the upper limit of the power loss deviation, then the power loss of the photovoltaic module is obtained by weighted coupling of the physical power loss and the image power loss. The output fault category probability and defect category probability are normalized to obtain the corresponding coupling weights. If the physical-image power loss deviation is greater than the upper limit of the power loss deviation, then the physical power loss and the image power loss are compared and judged. The comparison and judgment between physical power loss and image power loss is as follows: if the physical power loss is greater than the image power loss, then the physical power loss of the photovoltaic module is taken as the power loss; if the physical power loss is less than the image power loss, then the image power loss of the photovoltaic module is taken as the power loss. The physical power loss and image power loss of the photovoltaic module are obtained using the following methods: The obtained defect classification results are input into the preset image power loss mapping set for mapping to obtain the power loss of each cell in the photovoltaic module. The maximum power loss in each cell is recorded as the image power loss of the photovoltaic module. The difference between the standard power of the component and the current theoretical maximum power is recorded as the physical power loss. The theoretical maximum power represents the maximum power at each data point of the IV curve obtained by substituting the corrected key parameters into the photovoltaic cell physical model and solving the photovoltaic cell physical model through Newton's iteration method.

2. The real-time monitoring method for photovoltaic power generation based on multi-source data fusion as described in claim 1, characterized in that, The specific process for determining whether to discard the scanned data is as follows: Obtain the irradiance change, which reflects the degree of deviation between the irradiance at the start of the scan and the irradiance at the end of the scan; If the change in irradiance is greater than the preset threshold for the change in irradiance, the data from this scan will be discarded; otherwise, the IV curve obtained from the scan will be optimized. The quality optimization process for the IV curves obtained from the scan is specifically as follows: Perform a power non-negativity check on the IV curve, that is, obtain the power of each data point on the IV curve and remove data points with negative power. The sliding window statistical filtering of the IV curve is performed as follows: a sliding window containing a preset number of data points is set, and the absolute value of the difference between the current value of the center data point of the sliding window and the median of the current values ​​of all data points in the sliding window is recorded as the absolute deviation of the current value. If the absolute deviation of the current value is greater than the standard deviation of the current values ​​of all data points within the sliding window by a preset multiple, then the data point at the center of the sliding window is marked as an abnormal data point and removed; otherwise, the abnormal data point is not removed. Interpolation and moving average processing are performed on the IV curve after data point filtering.

3. The real-time monitoring method for photovoltaic power generation based on multi-source data fusion as described in claim 1, characterized in that, The process of using a preset photovoltaic cell physical model to perform nonlinear fitting on the IV curve after quality optimization further includes: The key physical parameters were corrected to standard test conditions. These key physical parameters include photocurrent, diode reverse saturation current, diode ideality factor, series resistance, and parallel resistance. By introducing an irradiance correction factor and a temperature compensation factor, the current photocurrent is corrected to obtain the corrected photocurrent. The irradiance correction factor is obtained by calculating the ratio of the average irradiance and the standard irradiance during the scanning process, and the temperature compensation factor is obtained by correcting the component backplane temperature deviation during the scanning process by combining a preset short-circuit current temperature coefficient. By introducing a series resistance correction factor, the current series resistance is reverse-compensated to obtain the corrected series resistance. The series resistance correction factor is obtained by combining the resistance temperature coefficient to correct the temperature deviation of the component backplane. By introducing a parallel resistance correction factor, the current parallel resistance is corrected to obtain the corrected parallel resistance; The parallel resistance correction factor is obtained by exponentially compensating for the temperature deviation of the component backplane by combining empirical coefficients.

4. The real-time monitoring method for photovoltaic power generation based on multi-source data fusion as described in claim 1, characterized in that, The extracted key physical parameters are input into the fault classification model, and the output includes a fault classification result comprising the fault type and the corresponding fault category probability, followed by: The data point with the minimum voltage value is obtained and recorded as the short-circuit current point. The corresponding current value is recorded as the short-circuit current. The short-circuit current is corrected by introducing the irradiance correction factor and the short-circuit temperature correction factor to obtain the corrected short-circuit current. The data point with the largest voltage value is obtained and recorded as the open circuit voltage point, and the corresponding voltage value is recorded as the open circuit voltage. After the open circuit voltage is corrected by introducing an open circuit temperature correction factor, it is coupled with the open circuit voltage to obtain the corrected open circuit voltage. Obtain the power value of each data point, and record the data point corresponding to the maximum power value as the maximum power point; The current value at the maximum power point is corrected by introducing an irradiance correction factor and a short-circuit temperature correction factor, and the corrected current value at the maximum power point is obtained. An open-circuit temperature correction factor is introduced to correct the voltage value at the maximum power point. The correction is then coupled with the voltage value at the maximum power point to obtain the corrected voltage value at the maximum power point. The ratio of the corrected power at the maximum power point to the ideal maximum power is recorded as the fill factor. The fault type is obtained by querying the preset rule base based on the fill factor, the corrected power at the maximum power point, the corrected short-circuit current, and the corrected open-circuit voltage. The irradiance correction factor is expressed as the ratio of standard irradiance to average irradiance; The short-circuit temperature correction factor is obtained by correcting the difference between the average component temperature and the standard component temperature by combining the set temperature coefficient of the short-circuit current. The open-circuit temperature correction factor is obtained by correcting the difference between the standard component temperature and the average component temperature by combining the temperature coefficient of the open-circuit voltage.

5. The real-time monitoring method for photovoltaic power generation based on multi-source data fusion as described in claim 1, characterized in that, The step of inputting the battery cell image into the instance segmentation model, prior to which includes: If the output fault type is no fault, the photovoltaic power generation is normal. Otherwise, the corresponding photovoltaic string cell image is obtained. Brightness uniformity correction is performed on the battery cell image, which includes flat field correction and background subtraction based on polynomial fitting; Median filtering was used to remove random noise from the battery cell image after brightness uniformity correction.

6. The real-time monitoring method for photovoltaic power generation based on multi-source data fusion as described in claim 1, characterized in that, The specific process for formulating power dispatch strategies is as follows: The difference between the standard power and the average power loss of all photovoltaic modules is recorded as the expected available power; If the expected available power is greater than the current load power and the energy storage state of charge is not greater than the charging limit, then the energy storage should be charged first. If the expected available power is greater than the current load power and the energy storage state of charge is greater than the charging limit, then power will be output to the grid. If the expected available power is not greater than the current load power and the energy storage state of charge is not greater than the discharge limit, then the power is received from the grid. If the expected available power is not greater than the current load power, and the energy storage state of charge is greater than the discharge lower limit, then the energy storage will discharge.

7. A photovoltaic power generation real-time monitoring system based on multi-source data fusion, used to implement the photovoltaic power generation real-time monitoring method based on multi-source data fusion as described in any one of claims 1-6, characterized in that, The system includes: The photovoltaic information acquisition module is used to scan the IV curves of all photovoltaic strings in the photovoltaic power station, and simultaneously record the irradiance and module backsheet temperature during the scanning process. It also determines whether to discard the data of this scan. If the data of this scan is discarded, the IV curve scan is repeated. Otherwise, the quality optimization processing is performed on the IV curves obtained by the scan, which represent the output characteristics of the photovoltaic strings under the current operating conditions. The physical fault classification module is used to perform nonlinear fitting on the IV curve after quality optimization using a preset photovoltaic cell physical model, and input the extracted key physical parameters into the fault classification model to output fault classification results including fault type and corresponding fault category probability. The image defect classification module is used to acquire images of solar cells in photovoltaic strings that are identified as faulty, and inputs the solar cell images into the instance segmentation model to output defect classification results including defect type, coordinates, segmentation mask and corresponding defect category probability. The power dispatch module is used to obtain the power loss of photovoltaic modules based on fault classification results and defect classification results, and to formulate power dispatch strategies.

Images