Method and system for testing attenuation rate of outdoor photovoltaic module

Abstract

The invention relates to the technical field of photovoltaic power generation detection, in particular to an outdoor photovoltaic module attenuation rate test method and test system, and the method comprises the steps: a multi-dimensional environment and operation data collection process; constructing and updating a digital twin reference model; a self-adaptive environment compensation and normalization process; and attenuation rate calculation and state evaluation. Nonlinear measurement errors caused by spectrum red shift / blue shift, large-angle incidence and a temperature hysteresis effect are effectively stripped, and real-time output data are mapped to equivalent performance data under a standard test condition, so that the measurement uncertainty is remarkably reduced, and sensitive capture and accurate calculation of a micro power attenuation trend are realized.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation testing technology, and in particular to a method and system for testing the degradation rate of outdoor photovoltaic modules. Background Technology

[0002] As photovoltaic (PV) power generation technology plays an increasingly important role in the global energy structure, PV modules, as the core power generation unit, have become a key technical indicator for evaluating the levelized cost of electricity (LCOE) of PV power plants, formulating precise operation and maintenance strategies, and determining the quality and stability of module products throughout their entire life cycle. During their 25 to 30 years of outdoor service, PV modules are exposed to a complex and ever-changing atmospheric environment, continuously subjected to the coupled effects of multiple environmental stress fields, including cumulative ultraviolet radiation, high and low temperature thermal cycling, humidity and heat alternation, mechanical loads, and potential-induced degradation. This causes irreversible physicochemical aging of the encapsulation materials, cell structure, and electrical connection interfaces, resulting in a non-linear dynamic decline in photoelectric conversion efficiency over time. Accurately quantifying this degradation rate is not only the data foundation for PV asset valuation and financing guarantees but also a necessary prerequisite for optimizing power plant system design and increasing power generation. Therefore, developing a testing method and system that can overcome the effects of environmental disturbances and achieve high-precision degradation data acquisition and objective evaluation under actual outdoor conditions is of significant engineering value for ensuring the long-term reliability and economic benefits of the PV industry.

[0003] For assessing the degradation rate of outdoor photovoltaic (PV) modules, existing technologies mainly present two typical approaches: online monitoring systems and offline field testing. However, these existing solutions still have significant shortcomings in practical outdoor applications, failing to meet the current demand for high-precision, low-interference measurement of PV module degradation rates. Firstly, regarding environmental adaptability and measurement accuracy, the spectral distribution, incident angle variation, scattered radiation ratio, and atmospheric turbidity of the outdoor environment exhibit highly dynamic nonlinear characteristics. Existing conversion methods based on simplified steady-state mathematical models often fail to accurately decouple the impact of these complex environmental factors on photoelectric conversion efficiency, resulting in significant dispersion and systematic errors in the converted power data. The measurement uncertainty (typically between 2% and 5%) is often higher than the actual annual average degradation rate of the module (approximately 0.5% to 0.8%), leading to data overload and an inability to accurately capture early, subtle degradation trends. Secondly, regarding data consistency and defect diagnosis, existing monitoring systems are limited by long-term zero-point drift of sensors and module surface... The non-uniformity of surface ash pollution and the lag in temperature measurement lead to a lack of comparability of test data across different time dimensions. Simultaneously, single electrical parameter monitoring lacks the ability to specifically identify the physical mechanisms causing degradation (such as the PID effect, LeTID effect, or yellowing of encapsulation materials), making it difficult to achieve the diagnostic leap from "knowing the degradation" to "knowing why the degradation occurs." Furthermore, traditional IV scanning tests are typically invasive, requiring the components to be temporarily removed from maximum power point tracking or disconnected from the grid. This not only results in substantial power generation loss but also, under cloudy weather conditions with drastic light fluctuations, creates an extremely narrow testing window, severely limiting the sample size and continuity of test data, making it impossible to construct a degradation evolution model under all weather and operating conditions. Summary of the Invention

[0004] The objective of this application is to provide a method for testing the degradation rate of outdoor photovoltaic (PV) modules, comprising: a multi-dimensional environmental and operational data acquisition process, which acquires in real time the current and voltage output data of the PV module under test, as well as synchronous on-site environmental parameters, including at least solar irradiance, module backsheet temperature, environmental spectral distribution, and light incident angle; a digital twin benchmark model construction and updating process, which constructs a digital twin benchmark model characterizing the output performance of the PV module under test in an ideal, degradation-free state based on the initial factory characteristic parameters and historical operational data of the PV module under test; and an adaptive environmental compensation and normalization process, which uses machine learning algorithms to analyze the nonlinear impact of current environmental parameters on module performance, and uses the digital twin benchmark model... The system dynamically corrects the real-time acquired current and voltage output data to eliminate measurement deviations caused by environmental fluctuations and generate equivalent performance data under standard test conditions. The attenuation rate calculation and state assessment process compares the equivalent performance data with the initial rated power of the component to calculate the current power attenuation rate and generates an attenuation trend curve based on time series analysis. Specifically, the adaptive environmental compensation and normalization process includes: establishing a dynamic correlation matrix between the environmental spectral response mismatch factor and the component temperature coefficient; and, within each measurement cycle, weighting the static temperature correction formula based on real-time monitored spectral distribution data and incident angle data to obtain the true photoelectric conversion efficiency after removing environmental interference.

[0005] By adopting the above technical solution, a multi-dimensional environmental and operational data acquisition process is used to simultaneously acquire all environmental parameters, including irradiance, backplane temperature, spectral distribution, and incident angle, overcoming the model bias caused by traditional methods that rely solely on single irradiance and temperature corrections. A digital twin benchmark model construction and update mechanism is introduced, using the initial characteristics of the component and historical data to construct a virtual image representing the ideal no-attenuation state, providing a dynamic and accurate reference benchmark for attenuation calculation. Through an adaptive environmental compensation and normalization process based on machine learning, a dynamic correlation matrix between the environmental spectral response mismatch factor and the component temperature coefficient is established, and the static correction formula is weighted and compensated in each measurement cycle. This application effectively eliminates nonlinear measurement errors caused by spectral redshift / blueshift, large-angle incident, and temperature hysteresis effects, mapping real-time output data to equivalent performance data under standard test conditions, thereby significantly reducing measurement uncertainty and achieving keen capture and accurate calculation of minute power attenuation trends.

[0006] Optionally, the process also includes a multimodal fusion diagnostic process for microscopic defects, specifically including: acquiring non-contact image data of the photovoltaic module under test, the image data including infrared thermal imaging spectra and electroluminescence images; extracting hot spot feature regions and lattice defect texture features from the image data, spatially mapping and temporally aligning them with the fill factor variation features in the current and voltage output data; using a deep neural network model to identify multiple concurrent decay modes, the decay modes at least covering potential-induced decay, yellowing of encapsulation materials, and microcracks in the cell, and quantifying the contribution weight of each decay mode to the total power decay rate.

[0007] By adopting the above technical solution, by acquiring infrared thermal imaging and electroluminescence images, texture features such as hot spots and lattice defects are extracted, and these features are spatially mapped and temporally aligned with electrical features such as fill factor. A deep neural network model is used to identify multiple concurrent decay modes such as potential-induced decay, material yellowing, and microcracks. This not only realizes cross-scale correlation from macroscopic power loss to microscopic physical defects, but also accurately quantifies the contribution weight of each decay mode to the total power decay rate.

[0008] Optionally, the process of constructing and updating the digital twin benchmark model specifically includes: setting a self-learning period at the initial stage of component operation, collecting all-weather operation data within the self-learning period as a benchmark dataset; fitting the output characteristic surfaces of the component in different irradiance and temperature ranges through regression analysis to generate an initial digital twin model; in subsequent operation, identifying and removing abnormal data points caused by shading or surface dust accumulation, and using effective data to continuously update the output characteristic surfaces to distinguish between recoverable performance degradation and irreversible material aging degradation.

[0009] By adopting the above technical solution, a self-learning cycle is set to generate an initial model in the early stage of component operation. In subsequent operation, abnormal data points caused by shadow occlusion or surface dust are intelligently identified and removed. Only valid data is used to iteratively optimize the characteristic surface, ensuring that the digital twin model always represents the current ideal physical state of the component. This effectively distinguishes between recoverable performance degradation caused by environmental interference and irreversible degradation caused by material aging, avoiding false alarms and missed alarms.

[0010] Optionally, the step of acquiring the environmental spectral distribution during the multidimensional environmental and operational data acquisition process specifically includes: using a group of spectral sensors deployed in the photovoltaic array area to monitor in real time the energy distribution ratio of the solar spectrum in the ultraviolet, visible, and near-infrared bands; calculating the mismatch between the current spectral distribution and the standard solar spectrum, and inputting the mismatch as an independent variable into the algorithm model of the adaptive environmental compensation and normalization process to correct the current measurement error caused by the change in the spectral response characteristics of the photovoltaic material.

[0011] By adopting the above technical solution, the spectral sensor group is used to monitor the energy distribution of multiple bands in real time, calculate the spectral mismatch degree and use it as an independent variable to input the compensation algorithm. By correcting the current measurement error caused by the change of spectral response characteristics of photovoltaic materials, especially in scenarios with drastic spectral changes such as dawn, dusk or cloudy weather, the fidelity of the equivalent performance data is greatly improved and the systematic error introduced by spectral mismatch is eliminated.

[0012] Optionally, the quantification of the contribution weight of each attenuation mode to the total power attenuation rate specifically includes: constructing an equivalent circuit model of the component, converting the identified micro-defect features into the increase in series resistance, decrease in parallel resistance, or loss of photocurrent in the equivalent circuit; and calculating the maximum power point power drop caused by the changes in the above-mentioned circuit parameters through a sensitivity analysis algorithm, thereby determining the proportion of each defect type in the current total attenuation rate.

[0013] By adopting the above technical solution, the identified micro-defect characteristics are transformed into changes in circuit parameters such as increased series resistance and decreased parallel resistance. This enables the quantitative calculation of the maximum power point power drop caused by each type of defect, thereby achieving a refined deconstruction of the attenuation source and solving the technical pain point that traditional black-box testing cannot quantify the degree of defect harm.

[0014] Optionally, the multi-dimensional environmental and operational data acquisition process adopts a distributed edge collaborative online monitoring mode, specifically including: deploying edge computing nodes at the junction box end of the photovoltaic module or the micro-inverter end; performing volt-ampere characteristic scanning at a millisecond frequency during the intervals of normal grid-connected power generation of the photovoltaic module or through micro-perturbation; extracting and compressing features from the high-frequency scanning data at the edge end, and uploading only the feature vector representing the health status of the module to the central processing unit to achieve continuous attenuation rate monitoring under zero downtime; and using the edge computing nodes of adjacent photovoltaic modules to perform horizontal data comparison, automatically identifying and filtering transient power fluctuation interference caused by local cloud cover.

[0015] By adopting the above technical solution, edge computing nodes deployed at junction boxes or micro-inverters perform volt-ampere characteristic scanning at millisecond-level frequencies, and complete feature extraction and compression at the edge. By using lateral data comparison between adjacent nodes to filter local cloud interference, continuous monitoring under zero downtime is achieved, which not only ensures the timeliness and integrity of the data, but also avoids the impact on the power grid and power generation loss.

[0016] Optionally, the continuous degradation rate monitoring under zero-downtime conditions is specifically achieved in the following way: by utilizing the conduction characteristics of the module's bypass diode or the maximum power point tracking disturbance of the power electronic converter, the operating point voltage of the module is instantaneously changed without interrupting the main circuit current; the current and voltage response sequence during this instantaneous change is collected, the knee region of the photovoltaic module's volt-ampere characteristic curve is reconstructed, and the open-circuit voltage and short-circuit current are calculated based on the characteristics of this knee region, thereby completing the performance evaluation.

[0017] By adopting the above technical solution, utilizing the conduction characteristics of bypass diodes or the disturbance mechanism of power electronic converters, the operating point voltage is instantaneously changed without disconnecting the main circuit, reconstructing the knee region of the volt-ampere characteristic curve, and calculating global parameters through local features. This ensures continuous system operation while completing high-precision performance evaluation, fundamentally resolving the contradiction between online testing and grid-connected power generation.

[0018] Optionally, the attenuation rate calculation and condition assessment process further includes a dust accumulation effect decoupling step: obtaining the surface cleanliness index through a dust accumulation monitoring sensor installed on the component frame; establishing a dust accumulation transmittance loss model, and before calculating the attenuation rate, performing transmittance compensation on the current data based on the cleanliness index to separate the power loss caused by external physical shading from the power attenuation caused by the aging of the internal materials of the component.

[0019] By adopting the above technical solution, the cleanliness index is obtained through the dust accumulation monitoring sensor, and the current data is pre-compensated based on the dust accumulation transmittance loss model. At the algorithm level, the transmittance loss caused by external physical obstruction is separated from the aging of the internal materials of the component, ensuring that the attenuation rate calculation result purely reflects the health status of the component itself.

[0020] Optionally, it also includes a degradation trend prediction step based on federated learning: establishing an encrypted communication mechanism between photovoltaic power plants in different geographical locations, sharing the degradation feature model parameters trained by edge computing nodes without sharing the original collected data; optimizing the local degradation prediction algorithm using the aggregated global model parameters, and predicting the remaining service life and performance degradation trajectory within a specific time window in the future based on the historical degradation rate curve of the components.

[0021] By adopting the above technical solution, the parameters of the attenuation feature model trained on the edge nodes are shared through an encrypted communication mechanism. The global model is optimized under the premise of protecting data privacy. The aggregated model is used to analyze the local historical attenuation curve, which can more accurately predict the remaining life and degradation trajectory of the components under specific geographical and climatic conditions. This achieves knowledge sharing and improved prediction accuracy across power plants.

[0022] The second objective of this application is to provide an outdoor photovoltaic module degradation rate testing system, comprising multiple functional modules configured to perform the aforementioned methods. These functional modules include at least: a distributed sensing module installed at the outdoor photovoltaic module site for collecting multi-dimensional data including the photovoltaic module's port voltage, loop current, surface temperature, ambient light irradiance, module backsheet temperature, environmental spectral distribution, and light incident angle; an edge computing terminal connected to the distributed sensing module for performing preliminary data cleaning, feature extraction, and zero-downtime online scanning control; and a cloud analysis platform comprising a digital twin engine, an adaptive compensation unit, and an intelligent diagnostic unit. The digital twin engine is used to run the digital twin baseline model; the adaptive compensation unit is used to perform the adaptive environmental compensation and normalization process; and the intelligent diagnostic unit is used to perform the microscopic defect multimodal fusion diagnostic process, ultimately outputting a precise degradation rate and defect diagnosis report for the photovoltaic module after removing environmental interference. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the outdoor photovoltaic module degradation rate test method of this application.

[0024] Figure 2 This is a block diagram of the outdoor photovoltaic module degradation rate testing system of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the technical details in the embodiments of the present application; obviously, the described embodiments are only some embodiments of the present application, and not all embodiments; based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present application.

[0026] like Figure 1 As shown in the figure, this application provides a method for testing the degradation rate of outdoor photovoltaic modules, including the following steps.

[0027] S01: Multidimensional environmental and operational data acquisition process, real-time acquisition of current and voltage output data of the photovoltaic module under test and synchronous on-site environmental parameters, including at least solar irradiance, module backsheet temperature, environmental spectral distribution and light incident angle.

[0028] Understandably, in complex and dynamic outdoor engineering applications, the multi-dimensional environmental and operational data acquisition process is the fundamental physical link to overcome the technical bottleneck of environmental noise masking the true attenuation signal in traditional testing. In actual outdoor scenarios, solar irradiance is not a constant light source; it not only varies sinusoidally with the solar altitude angle but is also subject to transient interference from cloud flow, aerosol scattering, and changes in atmospheric mass. This causes drastic fluctuations in the photon energy density received by photovoltaic modules on a millisecond timescale. Simultaneously, due to wind speed non-uniformity and the thermal inertia of the encapsulation materials, the module backsheet temperature often lags behind changes in ambient temperature, forming a non-uniform temperature gradient field on the module surface. Therefore... To overcome measurement errors caused by the aforementioned environmental nonlinear fluctuations, this embodiment employs a high-frequency synchronous sampling and multi-source heterogeneous data fusion strategy. The specific data acquisition system consists of a high-precision Hall effect current sensor, a voltage divider resistor network, a fast-response thermocouple array, and a spectroradiometer. The logical flow of data acquisition follows a causal chain of "physical signal sensing → analog signal conditioning → analog-to-digital conversion → timestamp synchronization → raw data caching." Specifically, the acquisition of current and voltage output data requires a high-speed data acquisition card with a sampling frequency of at least 1 kilohertz, with the data type set to double-precision floating-point, the current unit in amperes, and the measurement range covering zero to 120% of the component short-circuit current, and the voltage unit in... The measurement range covers volts from zero to 120% of the module open-circuit voltage to ensure complete capture of transient electrical responses during maximum power point tracking. Irradiance, a key environmental parameter, is obtained using a combination of total and direct radiation meters, with sampling frequencies synchronized with electrical data at millisecond levels. Data is measured in watts per square meter. Module backsheet temperature is acquired using T-type or K-type thermocouple probes attached to the geometric center and edges of the module backsheet. Temperature data resolution must reach 0.1 degrees Celsius, and a multi-point weighted averaging algorithm is used to eliminate measurement bias caused by localized hot spots. Environmental spectral distribution data is obtained using a spectrometer mounted on the same plane as the photovoltaic module. The spectrum analyzer needs to have a wavelength response range of 300 nanometers to 1100 nanometers and be able to scan the energy distribution of the solar spectrum at a frequency of no less than once per minute. The output data structure is a two-dimensional array containing wavelength and corresponding irradiance. The incident angle of light is calculated by a high-precision tilt sensor combined with local latitude and longitude and time algorithm. The data unit is degrees and is used to correct Fresnel reflection loss. Before entering the processing stage, all the above-mentioned raw data can be filtered by Kalman filtering algorithm to remove high-frequency noise interference and strictly time-aligned according to the timestamp of the Global Positioning System to form a high-dimensional feature vector containing time, electrical parameters, thermodynamic parameters and optical parameters.

[0029] S02: The process of constructing and updating the digital twin benchmark model: Based on the initial factory characteristic parameters and historical operating data of the photovoltaic module under test, a digital twin benchmark model is constructed to characterize the output performance of the module under ideal no-attenuation conditions.

[0030] Specifically, the construction and updating process of the digital twin benchmark model serves as a time-scale reference system to distinguish between the theoretically expected performance of a component and its actual aging performance. Its technical bottleneck lies in how to eliminate the interference of environmental fluctuations during the initial commissioning period on the benchmark fitting and how to define the model's update boundary to prevent misjudging actual degradation as model error. In specific engineering implementation, the construction and updating process of the digital twin benchmark model includes: setting a self-learning cycle at the initial stage of component operation and collecting all-weather operating data within that cycle as a benchmark dataset; fitting the output characteristic surfaces of the component in different irradiance and temperature ranges through regression analysis to generate an initial digital twin model; and in subsequent operation, identifying and eliminating abnormal data points caused by shading or surface dust accumulation, and using effective data to continuously update the output characteristic surfaces to distinguish between recoverable performance degradation and irreversible material aging degradation.

[0031] To build a high-confidence initial model, the self-learning period is typically set to be no less than seven consecutive cloudless sunny days or an operating time with a cumulative irradiance of 500 kWh / m², during which the system locks in the assumption of zero attenuation rate. The data processing engine employs a polynomial surface fitting algorithm, using component backplane temperature (ranging from -20°C to 85°C) and planar irradiance (ranging from 100 to 1200 watts / m²) as dual independent variables and normalized power as the dependent variable to construct a three-dimensional response surface. To address the problem of dirty data contaminating the model, the algorithm... Density-based spatial clustering and noise algorithms are introduced to automatically identify and remove outliers that deviate from the master response surface by more than 5% (usually caused by cloud shadows or bird droppings). In the rolling update mechanism, the system sets a forgetting factor (usually 0.95 to 0.99) to make the model more sensitive to recent cleaning status data. When the deviation between the actual power and the model's predicted power shows a monotonically increasing trend and cannot be recovered by cleaning, the deviation is locked as irreversible material aging degradation and is no longer included in the model's baseline update.

[0032] S03: Adaptive environmental compensation and normalization process. It uses machine learning algorithms to analyze the nonlinear impact of current environmental parameters on component performance, and dynamically corrects the real-time acquired current and voltage output data through a digital twin benchmark model to eliminate measurement deviations caused by environmental fluctuations and generate equivalent performance data under standard test conditions.

[0033] The adaptive environmental compensation and normalization process specifically includes: establishing a dynamic correlation matrix between the environmental spectral response mismatch factor and the component temperature coefficient; and in each measurement cycle, weighting compensation is applied to the static temperature correction formula based on real-time monitored spectral distribution data and incident angle data to obtain the true photoelectric conversion efficiency after removing environmental interference.

[0034] Understandably, in real-world outdoor testing, redshift or blueshift in the spectral distribution can significantly alter the effective photocurrent of photovoltaic materials. Traditional methods typically simplify the spectral effect to a constant or ignore it, leading to severe deviations in attenuation rate calculations during dawn / dusk or in cloudy weather. This embodiment employs a deep neural network-based digital twin benchmark model to perform this complex nonlinear compensation task. The input layer of this digital twin benchmark model receives a high-dimensional feature vector output from the aforementioned acquisition process, specifically including real-time irradiance, backplane temperature, spectral distribution characteristic values ​​(such as average photon energy), incident angle cosine, and measured current and voltage data. The digital twin... The baseline model's network architecture includes an input layer, several hidden layers, and an output layer. The hidden layers employ a fully connected structure, and the number of neurons in each layer is dynamically adjusted based on the dimension of the input features, typically set between 64 and 256. Modified linear units are used as activation functions between layers to introduce non-linear mapping capabilities and prevent the vanishing gradient problem. The training process of this digital twin baseline model adopts a supervised learning model, utilizing all-weather operational data accumulated during the initial deployment phase (assuming no decay period) as the training set. The connection weights and bias parameters between neurons are continuously adjusted through backpropagation. During the feature engineering stage, environmental spectral response loss is specifically introduced. The mismatch factor, a key feature, is obtained by calculating the integral ratio of the real-time spectral distribution to the standard solar spectrum within the module's spectral response band. This ratio is then used in conjunction with module temperature data to construct a dynamic "spectrum-temperature" coupling term. The algorithm's operational logic follows the sequence of "feature extraction → weight allocation → nonlinear mapping → bias correction." Specifically, the digital twin baseline model first identifies the current environmental state, such as low irradiance, high temperature, and spectral redshift. Then, based on the trained parameter matrix, it calculates the theoretical fill factor, open-circuit voltage, and short-circuit current correction coefficients for the module under this state. Particularly for the dynamic correlation matrix, the digital twin... During the inference phase, the digital twin benchmark model outputs a weighted correction value for the temperature coefficient at the current moment in real time. This correction value dynamically adjusts the linear coefficient in the standard temperature correction formula, transforming it into a nonlinear function that adapts to the current spectral and incident angle conditions. Finally, the output layer of the digital twin benchmark model generates a set of normalized coefficients, mapping the measured current and voltage data to equivalent performance data under standard test conditions. This process not only eliminates the conventional effects of temperature and irradiance but also deeply removes environmental noise caused by spectral mismatch and optical reflection, ensuring that the input data for subsequent attenuation rate calculations only reflects the physical aging state of the component materials themselves, rather than a projection of environmental fluctuations.

[0035] S04: The attenuation rate calculation and condition assessment process compares the equivalent performance data with the initial rated power of the component, calculates the current power attenuation rate, and generates an attenuation trend curve by combining time series analysis.

[0036] Specifically, the system first retrieves the initial calibration power data of the photovoltaic module under test. This initial calibration power data can be obtained from the flash test report at the time of module delivery or from a benchmark value obtained by a third-party authoritative institution during the initial installation of the system. The data fields include key indicators such as maximum power, short-circuit current, and open-circuit voltage under standard test conditions. The system then performs an arithmetic calculation of the instantaneous attenuation rate, subtracting the initial calibration power from the equivalent maximum power under standard test conditions generated at the current moment, and then dividing the difference by the initial calibration power to obtain a percentage value representing the performance at the current moment relative to the initial state. However, considering the uncontrollable nature of the outdoor environment, the attenuation rate value at a single moment may still contain uncompensated occasional interference (such as the instantaneous shading from birds). Therefore, this embodiment introduces a time series analysis algorithm to construct a moving average window with a configurable length (e.g., seven or thirty days). Within this moving average window, for all... Effective instantaneous decay rate data undergoes probability density distribution analysis to remove outliers deviating from the center of the normal distribution by more than three standard deviations. The weighted average of the remaining data is then calculated as the effective decay rate for that time period. Furthermore, to generate a decay trend curve with predictive value, the system employs a trend decomposition algorithm (e.g., STL decomposition) to decompose the long-term decay rate sequence into a trend term, a seasonal periodic term, and a residual term. The trend term directly reflects the irreversible power loss trajectory caused by component materials (such as yellowing of the encapsulation film and corrosion of the solar cells). The final evaluation report output by the system includes not only the current numerical decay rate but also a predicted future lifespan based on historical trends. The physical meaning of the output data clearly indicates the health status of the components, providing power plant maintenance personnel with a quantitative basis for decision-making regarding whether cleaning, repair, or replacement of components is necessary. The entire calculation process is executed automatically and periodically on edge computing terminals or cloud servers.

[0037] Understandably, this method also includes a multimodal fusion diagnostic process for microscopic defects. In actual engineering deployments, this process is typically triggered when the photovoltaic array stops generating electricity at night or during periods of low light, and is executed by automated inspection equipment (such as tracked robots or drones) that integrates a high-resolution infrared thermal imager and an electroluminescent camera. The acquired non-contact image data has multimodal properties. The infrared thermal imaging spectrum reflects the temperature distribution gradient on the surface of the component and can capture the hot spot effect caused by local current overload. The data format is a thermal image file with a temperature value matrix. The electroluminescent image, on the other hand, is generated by injecting a reverse current into the component to excite near-infrared radiation, visually presenting lattice fractures, microcracks, and sintering defects inside the silicon wafer. The data format is a high-grayscale monochrome bitmap. The feature extraction stage uses a convolutional neural network to perform pixel-level segmentation and texture analysis on the above images, extracting the geometric shape and temperature gradient features of the hot spot region, as well as the dark lines and black spots features in the electroluminescent image. Then, the images are processed... The defect coordinate system is registered with the physical dimension coordinate system of the component, and the image acquisition time is synchronized with the current and voltage data acquisition time at the microsecond level. Subsequently, a deep neural network model is used for multimodal data fusion. The input of this deep neural network model includes image feature vectors and synchronized fill factor change features, such as the decrease in fill factor caused by the increase in series resistance. The output layer of this deep neural network model is designed as a multi-label classifier, which can identify a variety of concurrent decay modes, including but not limited to potential-induced decay, yellowing of encapsulation materials, and microcracks in the battery cell. The algorithm for quantifying contribution weights is based on the Shapley value interpretation method, which calculates the marginal contribution of each identified defect feature to the predicted total power decay value, thus clearly indicating in the output results: for example, if the total decay rate is 5%, 3% is caused by potential-induced decay and 2% is caused by microcracks. This directly links the macroscopic electrical performance degradation with the microscopic physical damage, achieving a precise assessment that not only knows what happened but also why it happened.

[0038] Understandably, the difficulty in the multimodal fusion diagnosis process of micro-defects lies in the accurate spatial mapping of heterogeneous data and the decoupling and quantization of the composite attenuation mechanism. In specific engineering implementation, the process first faces the natural differences in spatial resolution and field of view between infrared thermal imaging spectra and electroluminescent images. In order to achieve pixel-level spatial mapping and temporal alignment, the system can adopt an automatic registration algorithm based on feature points.

[0039] Specifically, firstly, edge detection operators (such as the Canny operator) are used to extract the component border and cell grid lines as rigid feature anchor points, constructing an affine transformation matrix to resample the low-resolution thermal imaging data and project it into the high-resolution EL image coordinate system, with the error controlled at the sub-pixel level. At the same time, time alignment not only requires synchronization of image acquisition time, but also requires that the component's electrical operating point at the time of image acquisition strictly correspond to the time of current and voltage data acquisition. For example, the grayscale value of the EL image is directly affected by the magnitude of the injected current. Therefore, the system synchronously records the instantaneous injected current value during the exposure cycle of the captured image and uses this current value to perform radiometric correction on the image grayscale, eliminating the nonlinear error of image brightness caused by the fluctuation of injected current, and ensuring that the image features only reflect the physical properties of the material itself.

[0040] After data preprocessing, the model proceeds to the recognition stage of the deep neural network. This embodiment constructs a dual-stream convolutional neural network architecture. One network specifically handles thermal features, focusing on extracting the temperature gradient, diffusion morphology, and center temperature of hot spots. The other network handles optical texture features, utilizing the deep structure of the residual network to capture lattice fracture textures, black core shapes, and hidden crack orientations in the EL image. The feature vectors of the two networks are weighted and fused before the fully connected layer, and an attention mechanism is introduced to automatically focus on overlapping regions that exhibit abnormalities in both modalities. For concurrent decay modes such as potential-induced decay, yellowing of encapsulation materials, and hidden cracks in the battery cell, the model employs a multi-label regression strategy. For example, the PID effect typically manifests as blackening of the battery cell edges in the EL image, while in the thermal image it may not show obvious local hot spots but is accompanied by a decrease in overall shunt resistance. Hidden cracks appear as sharp black lines in the EL image, while in the thermal image they may correspond to weak point-like temperature rises. By learning from massive amounts of historical labeled data, the neural network can identify these complex feature combinations. To quantify the contribution weight of each attenuation mode to the total power attenuation rate, the system introduces a physical constraint layer. The model not only outputs the defect category but also inversely calculates the equivalent parameter drift in the single diode model of the component. For example, the model maps the identified microcrack features to the percentage loss of the effective illumination area, and then calculates the theoretical attenuation value of the photocurrent. It maps the PID features to the exponential decrease of the parallel resistance. Through sensitivity analysis, the system calculates the theoretical power reduction when only microcracks or only PID exist. Using the Shapley addition method interpretation principle, the total measured power attenuation value (e.g., 5% attenuation) is mathematically decomposed into: microcrack contribution 0.8%, PID contribution 3.2%, and material yellowing contribution 1.0%.

[0041] It is understandable that the multi-dimensional environment and operation data acquisition process can adopt a distributed edge collaborative online monitoring mode. Traditional centralized inspection often faces the problems of data transmission bandwidth bottleneck and excessively long inspection cycle. In this embodiment, by embedding high-computing-power edge computing nodes inside the junction box of each photovoltaic module or at the input end of the micro-inverter, the monitoring logic is reconstructed from the bottom layer of the architecture. The core of the node contains a microcontroller that integrates a high-precision analog-to-digital converter and a digital signal processing unit, which has the ability to execute complex algorithms locally. In order to achieve continuous degradation rate monitoring under zero downtime, the edge node adopts a micro-scanning technology based on power electronic disturbance. During the normal grid-connected power generation of the module, the node uses the maximum power point tracking adjustment gap or actively injects a small voltage disturbance signal. The disturbance amplitude is usually controlled within 1% of the open circuit voltage. Under the premise of not triggering inverter protection and having almost no impact on power generation, the node quickly scans the knee area of ​​the volt-ampere characteristic curve near the maximum power point.

[0042] Understandably, edge data processing follows the principles of feature extraction and compression. Since the original high-frequency sampled IV data is massive, directly uploading it to the cloud would exhaust communication bandwidth. Therefore, edge nodes run feature extraction algorithms locally, retaining only the key feature vectors of the volt-ampere curve. These feature vectors include: the radius of curvature at the curve's knee, the rate of change of the first derivative of the fill factor, and the current intercept at a specific voltage. This high-dimensional feature data is only one-thousandth the size of the original data, significantly reducing the communication load. In real-world scenarios, a passing cloud can cause a sudden drop in component power in a certain area, which can easily be misinterpreted as a sudden component failure. To filter out such transient interference, this system establishes a distributed consensus protocol based on geographical topology. When an edge node detects abnormal power fluctuations, it will not immediately... Instead of triggering an alarm, the edge computing cluster uses a local area network (such as ZigBee or LoRa) to query physically adjacent nodes. If neighboring nodes report similar power drops within the same time window, the edge computing cluster will determine the event as environmental shading based on majority rule, automatically marking and filtering the data for that period, excluding it from the attenuation rate calculation. Conversely, if the power of surrounding nodes is normal and only the data of this node is abnormal, it will be determined as an internal component fault, triggering a detailed diagnostic process. This distributed horizontal comparison logic is equivalent to building a neural network with self-awareness and self-correction capabilities on the photovoltaic array side, ensuring that the data uploaded to the central processing unit is cleaned and verified as valid characterization data, thereby guaranteeing the extremely high robustness of the final attenuation rate assessment results under complex weather conditions.

[0043] To further enhance the input dimensions of environmental parameters and address the issue that a single irradiance sensor cannot detect the impact of spectral redshift or blueshift on photovoltaic materials with different bandgap, the environmental spectral distribution acquisition step in the multidimensional environmental and operational data acquisition process of this application specifically includes: using a group of spectral sensors deployed within the photovoltaic array area to monitor the energy distribution ratio of the solar spectrum in the ultraviolet, visible, and near-infrared bands in real time; calculating the mismatch between the current spectral distribution and the standard solar spectrum, and inputting this mismatch as an independent variable into the algorithm model of the adaptive environmental compensation and normalization process to correct the current measurement error caused by changes in the spectral response characteristics of the photovoltaic material.

[0044] Specifically, in terms of hardware selection, the spectral sensor group consists of multiple narrowband photodiode arrays, with center wavelengths set at 360 nm, 550 nm, 850 nm, and 1050 nm, respectively, corresponding to the ultraviolet aging sensitive region, the visible light main energy region, and the spectral response cutoff region of crystalline silicon materials. The data acquisition unit synchronously reads the photocurrent signal of each channel once per second and converts it into a relative spectral irradiance value. This process compares the weighted energy ratio of the real-time acquired spectral distribution curve with the international standard solar spectrum curve within the effective response band of the module through integral calculation. If a significant shift to longer wavelengths (redshift) in the spectrum is detected during dawn and dusk, the system will automatically lower the expected value of the current output of the polycrystalline silicon module to prevent the natural decrease in current caused by the spectral redshift from being misjudged as power degradation of the module itself. The introduction of this high-dimensional feature improves the testing accuracy of the system by at least 15% during non-midday periods, significantly expanding the effective outdoor testing time window.

[0045] Once the macroscopic power attenuation is confirmed, in order to delve deeper into its internal physical causes, this application embodiment also performs a refined quantitative analysis. The quantification of the contribution weight of each attenuation mode to the total power attenuation rate specifically includes: constructing an equivalent circuit model of the component, converting the identified microscopic defect features into the increase in series resistance, the decrease in parallel resistance, or the loss of photocurrent in the equivalent circuit; and using a sensitivity analysis algorithm to calculate the maximum power point power decrease value caused by the changes in the above-mentioned circuit parameters, thereby determining the proportion of each defect type in the current total attenuation rate.

[0046] Specifically, the logical starting point of this process is translating the image recognition results into circuit language. The system adopts an improved single-diode or dual-diode equivalent circuit model. For example, when infrared thermal imaging identifies three hot spots of a component with a unit area, the algorithm calculates the incremental value of the equivalent series resistance in ohms based on the thermosensitive relationship between the hot spot temperature and the local resistance. When the electroluminescent image shows a potential-induced decay black area covering 10% of the area, the algorithm derives the exponential decay of the equivalent parallel resistance based on the semiconductor physics model. Subsequently, sensitivity analysis is performed using numerical differentiation. Keeping other parameters constant, the algorithm simulates and calculates the specific impact of changing the series resistance, parallel resistance, or photocurrent parameter on the maximum power point. The final diagnostic report will present the following structured data: of the current total decay of 3%, 60% is due to the increase in series resistance (mainly caused by solder ribbon aging), 30% is due to the decrease in parallel resistance (mainly caused by the PID effect), and 10% is due to the decrease in photocurrent (mainly caused by the decrease in glass transmittance). This quantitative attribution transforms the operation and maintenance strategy from blind cleaning to targeted maintenance.

[0047] To obtain high-frequency electrical characteristics without affecting the power generation revenue of the power plant, the multi-dimensional environmental and operational data acquisition process in this application adopts a distributed edge collaborative online monitoring mode. Specifically, it includes: deploying edge computing nodes at the junction box end of the photovoltaic module or the micro-inverter end; performing volt-ampere characteristic scanning at a millisecond frequency during the intervals of normal grid-connected power generation of the photovoltaic module or through micro-perturbation; extracting and compressing features from the high-frequency scanning data at the edge end, and uploading only the feature vector representing the health status of the module to the central processing unit to achieve continuous degradation rate monitoring under zero-downtime conditions; using the edge computing nodes of adjacent photovoltaic modules to perform horizontal data comparison, automatically identifying and filtering transient power fluctuation interference caused by local cloud cover; the continuous degradation rate monitoring under zero-downtime conditions is specifically achieved in the following way: using the conduction characteristics of the module's bypass diode or the maximum power point tracking disturbance of the power electronic converter, the operating point voltage of the module is instantaneously changed without interrupting the main circuit current; collecting the current and voltage response sequence during this instantaneous change process, reconstructing the knee region of the volt-ampere characteristic curve of the photovoltaic module, and calculating the open-circuit voltage and short-circuit current based on the characteristics of the knee region, thereby completing the performance evaluation.

[0048] Understandably, during the disturbance period of maximum power point tracking, the controller actively injects a tiny voltage step signal, the amplitude of which is controlled within one percent of the open-circuit voltage and the duration is no more than fifty milliseconds. Within this extremely short time, the high-speed analog-to-digital converter captures the current-voltage pair at a frequency of ten kilohertz, thereby capturing the knee region with the largest curvature in the volt-ampere curve. Since a complete volt-ampere curve scan requires disconnecting the load, which would cause power generation loss, the embodiments of this application can use a curve fitting algorithm based on a physical model, using only local data segments near the knee, combined with the analytical equations of the single diode model, to mathematically inversely calculate the complete open-circuit voltage and short-circuit current values, with the error controlled within 0.5 percent. This measurement method avoids the mechanical life problem of frequent relay operation and achieves non-intrusive monitoring of power grid supply.

[0049] In outdoor environments, dust accumulation is the external variable that causes the greatest error in power measurement. Therefore, in this embodiment of the application, the attenuation rate calculation and state assessment process also includes a dust accumulation effect decoupling step: obtaining the surface cleanliness index through a dust accumulation monitoring sensor installed on the component frame; establishing a dust accumulation transmittance loss model, and before calculating the attenuation rate, performing transmittance compensation on the current data based on the cleanliness index to separate the power loss caused by external physical shading from the power attenuation caused by the aging of the internal materials of the component.

[0050] Specifically, the dust accumulation monitoring sensor adopts a photoelectric reflection or image grayscale analysis structure and is installed on the non-cell-covered area of ​​the module glass surface, such as the dead corner of the frame, outputting a dimensionless cleanliness coefficient between zero and one. The dust accumulation transmittance loss model is based on the Beer-Lambert law and is modified. Considering that dust particles of different sizes (such as sand, clay, and bird droppings) have different scattering effects on the spectrum, the dust accumulation transmittance loss model is calibrated by associating power jump data before and after historical cleaning with machine learning algorithms. In terms of calculation logic, the system first uses the cleanliness index to correct the measured short-circuit current to the theoretical clean state, and then substitutes the corrected current value into the attenuation rate formula, thereby ensuring that the output attenuation rate value purely reflects the aging degree of the cells and encapsulation materials inside the module, avoiding false attenuation alarms caused by long-term uncleaning.

[0051] As the scale of power plants expands and the number of monitoring nodes increases dramatically, a single data center can hardly handle massive amounts of privacy data. Therefore, this application embodiment also includes a degradation trend prediction step based on federated learning: an encrypted communication mechanism is established between photovoltaic power plants in different geographical locations, and the degradation feature model parameters trained by edge computing nodes are shared without sharing the original collected data; the local degradation prediction algorithm is optimized using the aggregated global model parameters, and the remaining service life and performance degradation trajectory within a specific time window in the future are predicted based on the historical degradation rate curve of the components.

[0052] Specifically, this step is logically at the top of the data processing chain, adopting a cyclical iterative mode of "local training - cloud aggregation - parameter distribution". The edge servers of each photovoltaic power station act as clients of federated learning, using locally accumulated attenuation data to train long short-term memory networks or gated recurrent unit models, and only uploading the gradient information or weight parameters of the model update to the central aggregation server through a homomorphic encrypted channel. The central server performs a weighted average of the model parameters from different climate zones (such as hot and dry deserts, hot and humid coastal areas, and high-altitude plateaus) to generate a globally universal attenuation model with extremely strong generalization ability. This globally universal attenuation model integrates the wind and sand abrasion characteristics of desert areas and the salt spray corrosion characteristics of coastal areas. When its parameters are distributed back to the local area, even a newly built power station can accurately predict the attenuation trend of the next five to ten years using the experience data of the entire industry, significantly improving the confidence of the remaining service life prediction.

[0053] Finally, all the above functional modules are organically integrated in physical and logical space to form a complete system architecture: an outdoor photovoltaic module degradation rate testing system, comprising: a distributed sensing module, installed at the outdoor photovoltaic module site, used to collect multi-dimensional data including the port voltage, loop current, surface temperature, ambient light irradiance, module backsheet temperature, environmental spectral distribution, and light incident angle of the photovoltaic module under test; an edge computing terminal, connected to the distributed sensing module, used to perform preliminary data cleaning, feature extraction, and zero-downtime online scanning control; and a cloud analysis platform, including a digital twin engine, an adaptive compensation unit, and an intelligent diagnostic unit; wherein, the digital twin engine is used to run the digital twin benchmark model; the adaptive compensation unit is used to perform adaptive environmental compensation and normalization processes; and the intelligent diagnostic unit is used to perform a multi-modal fusion diagnostic process for microscopic defects, ultimately outputting an accurate degradation rate and defect diagnosis report of the photovoltaic module after removing environmental interference. At the system integration level, the distributed sensing module adopts industrial-grade wireless IoT communication protocols (such as LoRaWAN or ZigBee) for networking to ensure the stability of data transmission in substations with complex electromagnetic environments; the edge computing terminal uses embedded AI chips (such as NPU or GPU modules) with trillions of floating-point operations per second to support real-time IV curve reconstruction and image preprocessing; the cloud analysis platform is deployed based on a microservice architecture, supports elastic scaling, and provides a visual operation and maintenance dashboard interface to display the attenuation heatmap of all station components, fault warning list, and asset health score.

[0054] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0055] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0056] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0057] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0058] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0059] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for testing the degradation rate of outdoor photovoltaic modules, characterized in that, include: The multi-dimensional environmental and operational data acquisition process acquires the current and voltage output data of the photovoltaic module under test in real time, as well as the synchronous field environmental parameters. The environmental parameters include at least the solar irradiance, module backsheet temperature, environmental spectral distribution, and light incident angle. The process of constructing and updating the digital twin benchmark model involves building a digital twin benchmark model that characterizes the output performance of the photovoltaic module under ideal, non-degradation conditions, based on the initial factory characteristic parameters and historical operating data of the photovoltaic module under test. The adaptive environmental compensation and normalization process uses machine learning algorithms to analyze the nonlinear impact of current environmental parameters on component performance. The digital twin benchmark model is used to dynamically correct the real-time collected current and voltage output data, eliminate measurement deviations caused by environmental fluctuations, and generate equivalent performance data under standard test conditions. The attenuation rate calculation and state assessment process compares the equivalent performance data with the initial rated power of the component, calculates the current power attenuation rate, and generates an attenuation trend curve by combining time series analysis. Specifically, the adaptive environmental compensation and normalization process includes: establishing a dynamic correlation matrix between the environmental spectral response mismatch factor and the component temperature coefficient; and in each measurement cycle, weighting the static temperature correction formula based on the real-time monitored spectral distribution data and incident angle data to obtain the true photoelectric conversion efficiency after removing environmental interference.

2. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 1, characterized in that, It also includes a multimodal fusion diagnostic process for microscopic defects, specifically including: Non-contact image data of the photovoltaic module under test is acquired, including infrared thermal imaging spectra and electroluminescence images; Hot spot feature regions and lattice defect texture features are extracted from the image data, and spatially mapped and temporally aligned with the fill factor variation features in the current and voltage output data. A deep neural network model is used to identify multiple concurrent degradation modes, which at least include potential-induced degradation, yellowing of encapsulation materials, and microcracks in the battery cells, and the contribution weight of each degradation mode to the total power degradation rate is quantified.

3. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 1, characterized in that, The process of constructing and updating the digital twin benchmark model specifically includes: In the early stages of component deployment, a self-learning period is set, and all-day operational data within that self-learning period is collected as a benchmark dataset. An initial digital twin model is generated by fitting the output characteristic surfaces of the component in different irradiance and temperature ranges through regression analysis. In subsequent operation, abnormal data points caused by shadow occlusion or surface dust are identified and removed. The output characteristic surface is then updated using valid data to distinguish between recoverable performance degradation and irreversible material aging degradation.

4. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 1, characterized in that, The steps for obtaining the environmental spectral distribution during the multidimensional environmental and operational data acquisition process specifically include: By utilizing a group of spectral sensors deployed within the photovoltaic array area, the energy distribution ratio of the solar spectrum in the ultraviolet, visible, and near-infrared bands can be monitored in real time. The mismatch between the current spectral distribution and the standard solar spectrum is calculated, and this mismatch is input as an independent variable into the algorithm model of the adaptive environmental compensation and normalization process to correct the current measurement error caused by the change in the spectral response characteristics of photovoltaic materials.

5. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 2, characterized in that, The specific weights for the contribution of each attenuation mode to the total power attenuation rate include: Construct an equivalent circuit model of the component and transform the identified micro-defect features into the increase in series resistance, decrease in parallel resistance, or loss of photocurrent in the equivalent circuit. The sensitivity analysis algorithm is used to calculate the maximum power point power drop caused by the changes in the above circuit parameters, and then determine the proportion of each defect type in the current total attenuation rate.

6. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 1, characterized in that, The multi-dimensional environment and operational data acquisition process adopts a distributed edge collaborative online monitoring mode, specifically including: Edge computing nodes are deployed at the junction box end of the photovoltaic module or the micro-inverter end to perform volt-ampere characteristic scanning at a millisecond frequency during the intervals when the photovoltaic module is generating electricity normally on the grid or through micro-perturbation. At the edge, high-frequency scanning data is extracted and compressed, and only the feature vectors representing the health status of the components are uploaded to the central processing unit to achieve continuous attenuation rate monitoring under zero downtime. By using the edge computing nodes of adjacent photovoltaic modules to perform horizontal data comparison, transient power fluctuation interference caused by local cloud cover can be automatically identified and filtered.

7. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 6, characterized in that, The continuous attenuation rate monitoring under zero downtime condition is specifically achieved through the following methods: By utilizing the conduction characteristics of the component's bypass diode or the maximum power point tracking disturbance of the power electronic converter, the operating point voltage of the component can be changed instantaneously without interrupting the main circuit current. The current and voltage response sequence during this instantaneous change is collected, the knee region of the photovoltaic module's current-voltage characteristic curve is reconstructed, and the open-circuit voltage and short-circuit current are calculated based on the characteristics of this knee region, thereby completing the performance evaluation.

8. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 6, characterized in that, The attenuation rate calculation and condition assessment process also includes a decoupling step for the effects of dust accumulation: The surface cleanliness index is obtained by a dust accumulation monitoring sensor installed on the component frame; A dust accumulation transmittance loss model is established. Before calculating the attenuation rate, transmittance compensation is performed on the current data based on the cleanliness index to separate the power loss caused by external physical obstruction from the power attenuation caused by the aging of internal materials of the component.

9. The method for testing the degradation rate of outdoor photovoltaic modules according to claim 6, characterized in that, It also includes a decay trend prediction step based on federated learning: Establish an encrypted communication mechanism between photovoltaic power plants in different geographical locations, and share the attenuation characteristic model parameters trained by edge computing nodes without sharing the original collected data; The local degradation prediction algorithm is optimized by using aggregated global model parameters, and the remaining service life and performance degradation trajectory within a specific time window are predicted based on the historical degradation rate curve of the component.

10. An outdoor photovoltaic module degradation rate testing system, characterized in that, This includes a plurality of functional modules configured to perform the method as described in any one of claims 1 to 9, said functional modules comprising at least: The distributed sensing module is installed at the outdoor photovoltaic module site to collect multi-dimensional data, including the port voltage, loop current, surface temperature, ambient light irradiance, module backsheet temperature, ambient spectral distribution, and light incident angle of the photovoltaic module under test. An edge computing terminal, connected to the distributed sensing module, is used to perform preliminary data cleaning, feature extraction, and zero-downtime online scanning control. The cloud-based analytics platform includes a digital twin engine, an adaptive compensation unit, and an intelligent diagnostic unit. The digital twin engine is used to run the digital twin baseline model; the adaptive compensation unit is used to execute the adaptive environmental compensation and normalization process; and the intelligent diagnostic unit is used to execute the microscopic defect multimodal fusion diagnostic process, ultimately outputting the accurate attenuation rate and defect diagnosis report of the photovoltaic module after removing environmental interference.

Images