Photovoltaic module fault self-diagnosis method and device

Abstract

The invention belongs to the technical field of photovoltaic power generation, and provides a photovoltaic module fault self-diagnosis method and device, and the method comprises the steps: constructing a temperature-irradiance-parameter drift three-dimensional database and a ternary nonlinear prediction model, calculating a module individual drift deviation coefficient, and calibrating a fault critical value; a dynamic threshold value is generated based on the model and the deviation coefficient, and real-time updating is achieved through edge-cloud cooperation; a high-temperature-resistant multi-source sensor network is deployed, and equivalent normal temperature parameters are reduced through data calibration, abnormity elimination and drift reverse compensation; a CNN + XGBoost fusion model is constructed, and normal drift and fault features are accurately distinguished through a 30-second lightweight diagnosis and abnormal depth verification layering mechanism. The device correspondingly comprises four functional modules. The method is suitable for high-temperature complex working conditions and assemblies of different ages, the fault recognition accuracy is larger than or equal to 95%, invalid operation and maintenance are greatly reduced, and the operation and maintenance efficiency and power generation stability of a photovoltaic power station in a high-temperature area are improved.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power generation technology, specifically a method and device for self-diagnosis of photovoltaic module faults. Background Technology

[0002] High-temperature and arid regions are important installation areas for photovoltaic power plants worldwide. These regions often experience continuous high temperatures, which significantly increase the operating temperature of the modules. High-temperature-related failures have become a key factor affecting the power generation of power plants.

[0003] The core technology for self-diagnosis of photovoltaic module faults is the fixed threshold method, which identifies faults by real-time acquisition of module output parameters and comparison with preset thresholds. This method performs well in normal temperature environments, but it completely fails in high-temperature environments due to module parameter drift.

[0004] Under high-temperature conditions, crystalline silicon solar cells exhibit two distinct characteristics: firstly, the material's bandgap decreases with increasing temperature; secondly, the series resistance within the module increases with increasing temperature. These dual factors lead to a significant drift in output parameters. This drift is also non-linearly related to irradiation intensity, and even within the same batch of modules, the degree of drift varies considerably due to differences in manufacturing processes.

[0005] The conventional fixed threshold method sets static thresholds based on room temperature parameters, which does not adapt to high-temperature drift characteristics and individual differences, leading to three major problems: First, normal high-temperature drift is misjudged as a fault, increasing the burden of ineffective operation and maintenance; second, the true fault characteristics are masked by drift, increasing safety risks; and third, it cannot adapt to the drift differences between new and old components, resulting in a two-way failure problem of false alarms for old components and missed alarms for new components.

[0006] Existing industry improvements, such as single temperature compensation, segmented thresholds, and offline calibration, have not completely solved the problems of radiation coupling effects, individual component differences, and real-time adaptation, resulting in limited diagnostic effectiveness.

[0007] In summary, the core contradiction between the strong coupling and drift of component parameters under high-temperature environments and individual differences, and the static threshold and lack of adaptation of conventional diagnostic methods, has become a key bottleneck restricting the operation and maintenance efficiency of photovoltaic power plants in high-temperature areas, and there is an urgent need to build a precise fault diagnosis technology system.

[0008] Therefore, the present invention provides a method and apparatus for self-diagnosis of photovoltaic module faults. Summary of the Invention

[0009] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0010] The technical solution adopted by this invention to solve its technical problem is: On the one hand, the present invention provides a self-diagnosis method for photovoltaic module faults, including: Step S10: Collect multidimensional data to form a three-dimensional database of temperature-irradiance-parameter drift, construct a ternary nonlinear prediction model, calculate the individual drift deviation coefficient of photovoltaic modules, and calibrate the critical fault parameter value through extreme high temperature test; Step S20: Based on the ternary nonlinear prediction model and the individual drift deviation coefficient of photovoltaic modules, a dynamic threshold that adapts to the high temperature environment and the individual state of photovoltaic modules is generated in real time. Through edge computing and cloud collaborative architecture, the dynamic threshold range is calculated and calibrated in real time. Step S30: Deploy a high-temperature resistant multi-source sensor network and complete the collection, cross-calibration and anomaly removal of multi-dimensional data under high-temperature scenarios, and then restore the equivalent room temperature parameters of the photovoltaic module through drift reverse compensation; Step S40: Construct a multi-dimensional fusion diagnostic model; through edge 30-second lightweight cycle diagnosis and abnormal triggering deep full model verification layering mechanism, achieve accurate differentiation between normal drift and fault characteristics in high temperature scenarios.

[0011] As a further improvement of the present invention, the specific process of collecting multidimensional data to form a three-dimensional database of temperature-irradiance-parameter drift is as follows: Thirty photovoltaic modules from the same batch but with different years of operation were selected. In an environmental chamber, a scenario of gradient high temperature and different irradiance combinations was simulated. The open-circuit voltage Voc, short-circuit current Isc, maximum power Pmax, and fill factor FF of the photovoltaic modules were collected every 5 minutes. At the same time, the backsheet temperature, junction box temperature, and ambient humidity of the photovoltaic modules were collected. The data were collected for 24 hours at each temperature point to form a three-dimensional database of temperature-irradiance-parameter drift.

[0012] As a further improvement of the present invention, the specific process of constructing the ternary nonlinear prediction model is as follows: A backpropagation (BP) neural network was chosen as the basic architecture, with 2 neurons and 3 hidden layers. The first layer has 16 neurons, the second layer has 8 neurons, and the third layer has 4 neurons. The ReLU activation function was used to extract nonlinear features layer by layer. Three neurons were set to correspond to open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax, using a linear activation function. A Dropout layer was added between the hidden and output layers to randomly disable some neurons to avoid overfitting. Momentum gradient descent was used instead of traditional gradient descent to accelerate the model's convergence speed. When the predicted MAE of the drift coefficients of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax are verified to be ≤2.1%, 1.8%, and 2.5% respectively, and the root mean square error RMSE is ≤3.0% on the test set, a ternary nonlinear prediction model is finally constructed.

[0013] As a further improvement of the present invention, the specific process for calculating the individual drift deviation coefficient of the photovoltaic module is as follows: Filter the full data of the target photovoltaic module by the photovoltaic module number. The filtering criteria are: covering all temperature gradients from 40℃ to 60℃ and all irradiance intensities from 800 to 1200W / ㎡; removing abnormal data; and retaining the actual drift of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax in each scenario, as well as the common drift of the corresponding scenario. For each temperature and radiation combination scenario, calculate the individual drift deviation coefficient of the photovoltaic module in a single scenario: Individual drift deviation coefficient of photovoltaic module = actual drift amount of the target photovoltaic module / common drift amount of the same temperature and radiation scenario. Calculate the individual drift deviation coefficients of the photovoltaic module for open circuit voltage Voc, short circuit current Isc, and maximum power Pmax respectively.

[0014] As a further improvement of the present invention, the specific process of calibrating the critical fault parameter value through extreme high temperature testing is as follows: Select differentiated photovoltaic modules with no initial faults and covering different years of operation, and deploy electrical parameter testers, infrared thermal imagers, and EL testers in the environmental chamber to simultaneously monitor core electrical parameters, photovoltaic module temperature field, and internal structure; The environmental chamber was stabilized at 60℃, 1000W / ㎡, and 60% humidity. After the photovoltaic modules were left to stand for 2 hours to reach thermal equilibrium, a 72-hour continuous test was started. Real-time recording of open-circuit voltage Voc, maximum power Pmax, fill factor FF, local temperature differences, and EL image changes; capturing parameter fluctuations and characteristic evolution from normal drift to fault precursors to fault failure. Calibrate the critical fault parameter values: Extract parameters corresponding to the failure precursors of hot spots and encapsulation aging; combine a ternary nonlinear prediction model to calculate the adaptation critical value under a gradient temperature of 40℃-55℃; increase safety redundancy by 5%-10% to avoid misjudgment due to instantaneous interference. A table comparing temperature, fault type, fault precursors, and critical values ​​for failure was created to clarify the critical parameter values ​​for different high temperatures.

[0015] As a further improvement of the present invention, the specific process of generating the dynamic threshold range of the photovoltaic module in real time to adapt to the high temperature environment is as follows: The core electrical parameters of the photovoltaic module are retrieved from the factory and used as the initial benchmark for threshold calculation; a high-temperature resistant fiber optic temperature sensor and a miniature irradiation sensor are integrated into the photovoltaic module to collect the current ambient temperature and surface irradiation intensity of the photovoltaic module in real time. Obtain the individual drift deviation coefficient of photovoltaic modules; The edge device performs real-time verification of the ambient temperature and surface irradiance of the photovoltaic modules, and removes outliers with individual drift deviation coefficients exceeding ±5% to ensure the reliability of the input data. The real-time collected ambient temperature and surface irradiance of the photovoltaic module are input into the ternary nonlinear prediction model, and the common drift coefficient range of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax is output. Based on the core electrical parameters of the components at the factory, the parameter range after theoretical common drift is calculated: Formula logic: Common drift range = core electrical parameters at the factory × common drift coefficient range; Based on the individual drift deviation coefficient of photovoltaic modules, the common drift range is modified to obtain the individual dynamic threshold range: Correction logic: Since the parameter drift of photovoltaic modules at high temperatures is mostly negative attenuation, only the lower limit of the dynamic threshold interval is corrected. The formula is: lower limit of dynamic threshold interval = lower limit of common drift interval × individual drift deviation coefficient of photovoltaic module, while the upper limit remains unchanged. Individual dynamic threshold range corrections were performed for open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax, respectively.

[0016] As a further improvement of the present invention, the specific process for realizing real-time calculation and calibration update of the dynamic threshold interval is as follows: At the edge, a high-temperature resistant sensor integrated into the photovoltaic module collects three types of core data at a frequency of once per minute: Current ambient temperature of photovoltaic modules and surface irradiance of photovoltaic modules; Obtain the open-circuit voltage Voc, short-circuit current Isc, maximum power Pmax, and individual drift deviation coefficient of the photovoltaic module; By using a ternary nonlinear prediction model, the common drift coefficient range is calculated by inputting ambient temperature and irradiance data. Then, based on the individual drift deviation coefficient of the photovoltaic module, a dynamic threshold range for individual photovoltaic module adaptation is generated. The dynamic threshold range is output to the edge diagnostic module in real time, compared with the collected photovoltaic module parameters, and the corresponding response is triggered. If the edge device loses network access, it will automatically activate the historical dynamic threshold range parameters cached locally to maintain the dynamic threshold range calculation, and re-upload the cached data after reconnecting to the network.

[0017] As a further improvement of the present invention, the specific process of restoring the equivalent room temperature parameters of the photovoltaic module through drift reverse compensation is as follows: Obtain the current ambient temperature, the surface irradiance of the photovoltaic module, and the measured electrical parameters of the module; Input real-time temperature and irradiance data into a ternary nonlinear prediction model and output the common drift coefficient range for the current scenario; The theoretical individual drift of a photovoltaic module is calculated by combining the individual drift amount with the individual drift deviation coefficient of the photovoltaic module: Formula: Theoretical individual drift amount = Common drift coefficient × Individual drift deviation coefficient; Convert the measured electrical parameters into percentages relative to a room temperature reference. Then, the equivalent room temperature parameters are restored through reverse compensation: Formula: Percentage of equivalent room temperature parameters = Percentage of measured parameters + Theoretical individual drift; As a further improvement of the present invention, the specific process for accurately distinguishing between normal drift and fault characteristics in high-temperature scenarios is as follows: At the edge, the equivalent room temperature parameters after compensation are retrieved in real time with a fixed period of 30 seconds. The current environmental characteristics and individual characteristics of photovoltaic modules are then correlated to form an input feature set for lightweight diagnosis. By simply calling the XGBoost lightweight inference module in the multi-dimensional fusion diagnostic model, parameter states can be quickly calculated based on the input feature set: The input feature set is compared with the normal dynamic threshold range and the normal drift boundary, and the preliminary judgment result is output: normal drift or suspected abnormality. When the multi-dimensional fusion diagnostic model outputs suspected abnormal parameters exceeding the dynamic threshold warning lower limit and local temperature exceeding the fault precursor threshold for two consecutive cycles, deep verification is initiated: The multi-dimensional fusion diagnostic model is invoked, and a time-series parameter sequence of equivalent room temperature parameters for 10 minutes is input. CNN extracts dynamic features such as voltage fluctuations and power trends, and XGBoost fuses multi-dimensional features to output fault classification probabilities. Infrared thermal images are acquired simultaneously, and secondary verification is performed using a high-temperature-specific fault feature library to distinguish between normal drift and fault features.

[0018] On the other hand, the present invention provides a self-diagnostic device for photovoltaic module faults, comprising: High-temperature drift basic data and model building module: Collect multi-dimensional data to form a three-dimensional database of temperature-irradiance-parameter drift, build a ternary nonlinear prediction model, calculate the individual drift deviation coefficient of photovoltaic modules, and calibrate the critical fault parameter values ​​through extreme high-temperature tests; High-temperature adaptation dynamic threshold management module: Based on the ternary nonlinear prediction model and the individual drift deviation coefficient of photovoltaic modules, it generates dynamic thresholds in real time to adapt to high-temperature environments and the individual states of photovoltaic modules. Through edge computing and cloud collaborative architecture, it realizes real-time calculation and calibration update of dynamic threshold range. High-temperature resistant multi-source data acquisition and compensation processing module: Deploys a high-temperature resistant multi-source sensor network and completes the acquisition, cross-calibration and anomaly removal of multi-dimensional data under high-temperature scenarios, and then restores the equivalent room temperature parameters of photovoltaic modules through drift reverse compensation; Multi-dimensional fusion fault diagnosis module: Constructs a multi-dimensional fusion diagnosis model; through edge 30-second lightweight cycle diagnosis and abnormal triggering deep full model verification layering mechanism, it achieves accurate differentiation between normal drift and fault characteristics in high temperature scenarios.

[0019] The beneficial effects of this invention are as follows: By quantifying the parameter drift law of high temperature-irradiation coupling through a ternary nonlinear prediction model and combining it with the individual drift deviation coefficient of photovoltaic modules to realize individual dynamic thresholds, the normal drift and fault characteristics can be accurately separated. This avoids the core pain points of conventional fixed threshold method, such as drift being misjudged as fault, fault being masked by drift, and bidirectional failure of new and old modules, and significantly reduces the burden of ineffective operation and maintenance and safety risks. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of the steps of a self-diagnosis method for photovoltaic module faults according to the present invention; Figure 2 This is a device module diagram of a photovoltaic module fault self-diagnosis device according to the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1 like Figure 1 As shown in the embodiment of the present invention, a self-diagnosis method for photovoltaic module faults includes: Step S10: Collect multidimensional data to form a three-dimensional database of temperature-irradiance-parameter drift, construct a ternary nonlinear prediction model, calculate the individual drift deviation coefficient of photovoltaic modules, and calibrate the critical fault parameter value through extreme high temperature test; In step S10, the specific process of acquiring multidimensional data to form a three-dimensional database of temperature-irradiance-parameter drift is as follows: Thirty photovoltaic modules from the same batch but with different operating years (0-10 years) were selected. In an environmental chamber, a combination of gradient high temperature (40℃, 45℃, 50℃, 55℃, 60℃) and different irradiance (800-1200W / ㎡) was simulated. The core parameters of the photovoltaic modules, namely open circuit voltage (Voc), short circuit current (Isc), maximum power (Pmax), and fill factor (FF), were collected every 5 minutes. At the same time, auxiliary data such as backsheet temperature, junction box temperature, and ambient humidity of the photovoltaic modules were collected. Each temperature point was collected for a total of 24 hours (the environmental chamber simulated the diurnal temperature fluctuation) to form a three-dimensional database of temperature-irradiance-parameter drift. In step S10, the specific process of constructing the ternary nonlinear prediction model is as follows: Based on a three-dimensional database of temperature-irradiance-parameter drift, a BP neural network was selected as the basic architecture, with two neurons corresponding to two core input features: temperature (gradient high temperature inside the environmental chamber) and irradiance (different irradiance intensities inside the environmental chamber), ensuring accurate matching between the input and feature dimensions.

[0024] Three hidden layers are set up (to alleviate the problem of insufficient fitting ability of a single hidden layer), with 16 neurons in the first layer, 8 neurons in the second layer, and 4 neurons in the third layer. The ReLU activation function is used (to solve the gradient vanishing problem) to achieve the extraction of nonlinear features layer by layer. Three neurons are set up to correspond to the drift coefficients of the three core parameters: open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax. A linear activation function is used (since the drift coefficients are continuous values, this is not a classification task) to directly output the quantized drift coefficient prediction values.

[0025] A Dropout layer (with a dropout probability of 0.2) is added between the hidden layer and the output layer to randomly disable some neurons and avoid overfitting; momentum gradient descent is used instead of traditional gradient descent to accelerate the model convergence speed. The mean squared error is used as the loss function, and the Adam optimizer is used (the learning rate is initially set to 0.001 and decays to 0.9 of the previous round every 100 rounds). The number of iterations is set to 1000 rounds. When the loss function of the validation set does not decrease for 20 consecutive rounds, an early stopping strategy is triggered to avoid overtraining. The core hyperparameters were optimized using a grid search method, with the number of hidden layer neurons (12-20), learning rate (0.0005-0.002), and Dropout probability (0.1-0.3). The mean absolute error (MAE) of the validation set was used as the evaluation metric to select the optimal combination of hyperparameters (e.g., 16-8-4 hidden layer neurons, learning rate 0.001, Dropout=0.2). During training, the prediction error under different temperature and radiation scenarios is monitored in real time. For scenarios with an error exceeding 5% (such as a low-irradiance high-temperature scenario of 60℃+800W / ㎡), the sample data of that scenario is supplemented (an additional 50 sets of data are collected) and the model is retrained to ensure the model's fitting accuracy for extreme scenarios. When the predicted MAE of the drift coefficients of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax are verified to be ≤2.1%, 1.8%, and 2.5% respectively on the test set, and the root mean square error (RMSE) is ≤3.0%, which meets the accuracy requirement of prediction error ≤5% in engineering applications, a ternary nonlinear prediction model is finally constructed.

[0026] In step S10, the specific process for calculating the individual drift deviation coefficient of the photovoltaic module is as follows: From the temperature-irradiance-parameter drift three-dimensional database, the full data of the target photovoltaic modules were filtered by module number. The filtering criteria were: covering all temperature gradients from 40℃ to 60℃ and all irradiance intensities from 800 to 1200 W / m²; abnormal data (such as sensor false alarms, transitional data during scene switching) were removed; and the actual drift of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax (the measured drift percentage of Voc / Isc / Pmax) and the common drift of the corresponding scene (the average drift percentage of all modules under the same temperature and irradiance) were retained. For each temperature and radiation combination scenario, the individual drift deviation coefficient of the photovoltaic module in a single scenario is calculated according to the following formula: Individual drift deviation coefficient of photovoltaic module = Actual drift amount of target photovoltaic module / Common drift amount of the same temperature and radiation scenario (Note: The drift amount is expressed as a percentage relative to the normal temperature reference. For example, in the 40℃+1000W / ㎡ scenario, the common maximum power Pmax drift amount of the same temperature and radiation scenario is 5%, and the actual maximum power Pmax drift amount of the target photovoltaic module is 6%, then the single-scenario deviation coefficient of maximum power Pmax in this scenario = 6% / 5% = 1.2). Following the above logic, the individual drift deviation coefficients of the photovoltaic module in all scenarios are calculated for the three core parameters: open circuit voltage Voc, short circuit current Isc, and maximum power Pmax. To avoid distortion of individual photovoltaic module drift deviation coefficients caused by data fluctuations in a single scenario, the individual drift deviation coefficients of a single photovoltaic module in each scenario are normalized: the mean of the individual drift deviation coefficients of the photovoltaic module is calculated (e.g., the mean of the individual drift deviation coefficients of the photovoltaic module under 5 temperature × 3 irradiation = 15 scenarios with maximum power Pmax); the individual drift deviation coefficients of the photovoltaic module in extreme scenarios within the mean ± 3σ range are removed (e.g., abnormal values ​​of the deviation coefficient caused by instantaneous failure of the photovoltaic module); the average value of the remaining valid individual drift deviation coefficients of the photovoltaic module is taken as the unified individual drift deviation coefficient of the module (e.g., final deviation coefficient of Pmax = 1.2, Voc = 1.1, Isc = 1.05), and the final individual drift deviation coefficient of the photovoltaic module is calculated; In step S10, the specific process of calibrating the critical fault parameter values ​​through extreme high temperature testing is as follows: Select differentiated photovoltaic modules with no initial faults and covering different years of operation. Deploy electrical parameter testers, infrared thermal imagers, and EL testers in the environmental chamber to simultaneously monitor core electrical parameters (open circuit voltage Voc, short circuit current Isc, maximum power Pmax, fill factor FF), temperature field, and internal structure of the photovoltaic modules. The environmental chamber was stabilized at 60℃ (extreme high temperature), 1000W / ㎡ (standard irradiation), and 60% humidity (harsh working conditions). After the photovoltaic modules were left to stand for 2 hours to reach thermal equilibrium, a 72-hour continuous test was started. Real-time recording of open-circuit voltage Voc, maximum power Pmax, fill factor FF, local temperature differences, and EL image changes; capturing parameter fluctuations and characteristic evolution from normal drift to fault precursors to fault failure. Calibrate the critical fault parameter values: Extract parameters corresponding to failures from hot spots and encapsulation aging precursors (such as the decrease in Pmax and local temperature difference); combine a ternary nonlinear prediction model to calculate the adaptation critical value under a gradient temperature of 40℃-55℃; increase safety redundancy by 5%-10% to avoid misjudgment due to instantaneous interference. A table is created to compare temperature, fault type, fault precursors, and critical failure values, clarifying the critical fault parameter values ​​at different high temperatures and linking them to the individual characteristic profiles of photovoltaic modules.

[0027] Step S20: Based on the ternary nonlinear prediction model and the individual drift deviation coefficient of photovoltaic modules, a dynamic threshold that adapts to the high temperature environment and the individual state of photovoltaic modules is generated in real time. Through edge computing and cloud collaborative architecture, the dynamic threshold range is calculated and calibrated in real time. In step S20, the specific process of generating a dynamic threshold range for high-temperature environments and individual photovoltaic modules in real time based on the ternary nonlinear prediction model and the individual drift deviation coefficient of the photovoltaic module is as follows: The core electrical parameters of the photovoltaic module (open-circuit voltage Voc, short-circuit current Isc, maximum power Pmax, and fill factor FF) are retrieved from the factory and used as the initial benchmark for threshold calculation. A high-temperature resistant fiber optic temperature sensor (accuracy ±0.5℃) and a miniature irradiation sensor (accuracy ±2%) are integrated into the photovoltaic module to collect the current ambient temperature and surface irradiance of the photovoltaic module in real time. Obtain the individual drift deviation coefficients of photovoltaic modules (e.g., Voc=1.1, Pmax=1.2, Isc=1.05). The edge device (inverter / module controller) performs real-time verification of the ambient temperature and surface irradiance of the photovoltaic module, and removes outliers with individual photovoltaic module drift deviation coefficients exceeding ±5% (such as instantaneous false alarms from the irradiance sensor) to ensure the reliability of the input data. The real-time collected ambient temperature and surface irradiance of the photovoltaic module are input into the ternary nonlinear prediction model, and the common drift coefficient range of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax is output (i.e., the normal drift range of all modules under the same temperature and radiation scenario). Based on the core electrical parameters of the components at the factory, the parameter range after theoretical common drift is calculated: Formula logic: Common drift range = core electrical parameters at the factory × common drift coefficient range; Example: The maximum power Pmax of a photovoltaic module is 320W. After real-time temperature and radiation data are input into the model, the common drift coefficient range is 95%-100%. Therefore, the theoretical common drift range is 320W×95%~320W×100%=304W~320W.

[0028] Based on the individual drift deviation coefficient of photovoltaic modules, the common drift range is modified to obtain the individual dynamic threshold range: Correction logic: Since the parameter drift of photovoltaic modules at high temperatures is mostly negative (the maximum power Pmax and the open-circuit voltage Voc decrease), only the lower limit of the dynamic threshold range is corrected (the upper limit remains consistent with the benchmark value to avoid excessive relaxation of the positive fluctuation threshold). The formula is: lower limit of dynamic threshold range = lower limit of common drift range × individual drift deviation coefficient of photovoltaic module, while the upper limit remains unchanged. Example: A photovoltaic module has an individual drift deviation coefficient of 1.2 for its maximum power Pmax, and a common drift range lower limit of 304W. After correction, the individual threshold lower limit = 304W × 1.2 = 300.8W. The final individual-adapted dynamic threshold range for Pmax is 300.8W~320W (corresponding to a percentage range of 94%~100%). Following the same logic, complete the individual dynamic threshold range correction for open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax respectively. In step S20, the specific process of realizing real-time calculation and calibration update of dynamic threshold range through edge computing and cloud collaborative architecture is as follows: At the edge, high-temperature resistant sensors (fiber optic temperature sensors and miniature radiation sensors) integrated into photovoltaic modules collect three types of core data at a frequency of 1 minute per scan: Current ambient temperature of photovoltaic modules and surface irradiance of photovoltaic modules; Obtain the open-circuit voltage (Voc), short-circuit current (Isc), maximum power (Pmax), and individual drift deviation coefficient of the photovoltaic module; Using a ternary nonlinear prediction model, the common drift coefficient range is calculated by inputting ambient temperature and irradiance data. Then, based on the individual drift deviation coefficient of the photovoltaic module, the dynamic threshold range (including normal / warning / fault three-level range) is generated according to the correction logic (only the lower limit of the dynamic threshold range is corrected, while the upper limit remains at the benchmark value). The dynamic threshold range is output to the edge diagnostic module in real time, compared with the collected photovoltaic module parameters, and the corresponding response is triggered (such as normal / warning / fault three-level range); at the same time, the calculation data of the whole day (including real-time ambient temperature, irradiance data, threshold results, parameter comparison records) is cached for subsequent cloud calibration. If the edge device loses network connection, the historical dynamic threshold range parameters of the local cache will be automatically activated to maintain the dynamic threshold range calculation (error ≤3%), and the cached data will be retransmitted after the network connection is restored. Step S30: Deploy a high-temperature resistant multi-source sensor network and complete the collection, cross-calibration and anomaly removal of multi-dimensional data under high-temperature scenarios, and then restore the equivalent room temperature parameters of the photovoltaic module through drift reverse compensation; In step S30, the specific process of deploying a high-temperature resistant multi-source sensor network and completing the acquisition, cross-calibration, and anomaly removal of multi-dimensional data under high-temperature scenarios is as follows: To address the parameter monitoring needs of components under high-temperature operating conditions, three types of dedicated sensors are precisely deployed at key locations on the components to ensure the coverage and reliability of data acquisition: Fiber optic temperature sensors: These are respectively mounted on the center of the photovoltaic module backsheet, the surface of the edge cells, and the inside of the junction box. They have a temperature range of -40℃ to 80℃ and collect temperature data from multiple areas of the photovoltaic module (accuracy ±0.5℃), covering hot spots and abnormal temperature rise risk points in the junction box. Miniature irradiance sensor: integrated above the front frame of the photovoltaic module (without blocking the light-receiving surface), temperature resistant from -20℃ to 70℃, collects the actual irradiance intensity on the module surface (accuracy ±2%), and matches the irradiance input dimension of the ternary nonlinear prediction model; High-frequency voltage and current sensor: connected in series at the output end of photovoltaic module, with a sampling frequency of 1kHz and a temperature resistance of -30℃~70℃, synchronously collects real-time open-circuit voltage (Voc), short-circuit current (Isc), and maximum power (Pmax), providing high-frequency data support for power verification; If the deviation is ≥5%, the surface irradiance value of the photovoltaic module is corrected based on the total irradiance meter data (to adapt to scenarios such as partial shading of the module). Temperature data calibration: Compare the module backsheet temperature collected by the fiber optic temperature sensor with the power station ambient temperature sensor data, calculate the photovoltaic module temperature compensation coefficient (formula: compensation coefficient = actual module backsheet temperature - (ambient temperature + 5℃)), correct the temperature collection values ​​of the cells and junction boxes, and eliminate the interference of ambient temperature fluctuations. Calculate the product of the data collected by the high-frequency voltage and current sensors (real-time power = voltage × current), compare it with the data of the power sensor integrated in the photovoltaic module, and if the deviation is ≥3%, then correct the power value based on the product of voltage and current to ensure the consistency of electrical parameters.

[0029] In step S30, the specific process of restoring the equivalent room temperature parameters of the photovoltaic module through drift reverse compensation is as follows: Obtain the current ambient temperature, the surface irradiance of the photovoltaic module, and the measured electrical parameters of the module (Voc, Isc, Pmax); Input real-time temperature and irradiance data into a ternary nonlinear prediction model and output the common drift coefficient range for the current scenario; The theoretical individual drift of a photovoltaic module is calculated by combining the individual drift amount with the individual drift deviation coefficient of the photovoltaic module: Formula: Theoretical individual drift amount = Common drift coefficient × Individual drift deviation coefficient; Example: If the individual drift deviation coefficient of this component Pmax is 1.2, then the theoretical individual drift amount = 5% × 1.2 = 6%; First, convert the measured electrical parameters into a percentage relative to a room temperature reference (e.g., if the measured Pmax = 300.8W and the room temperature reference Pmax = 320W, then the measured percentage = 300.8 / 320 × 100% = 94%). Then, the equivalent room temperature parameters are restored through reverse compensation: Formula: Percentage of equivalent room temperature parameters = Percentage of measured parameters + Theoretical individual drift; Example: The measured Pmax percentage is 94%, and the theoretical individual drift is 6%. Then the equivalent room temperature Pmax percentage = 94% + 6% = 100% (consistent with the room temperature baseline, indicating normal drift). Step S40: Construct a multi-dimensional fusion diagnostic model; through edge 30-second lightweight cycle diagnosis and abnormal triggering deep full model verification layering mechanism, achieve accurate differentiation between normal drift and fault characteristics in high temperature scenarios.

[0030] In step S40, the specific process of constructing the multi-dimensional fusion diagnostic model is as follows: Using time-series data with equivalent room temperature parameters (including 1-minute continuous sequences of Voc, Isc, Pmax, and FF) as the core samples, a training dataset with a ratio of normal drift samples to fault samples of 4:1 was constructed to avoid sample imbalance. The time-series data of equivalent room temperature parameters for 10 consecutive minutes (a total of 10 data points) are encapsulated into a single time-series sample and used as the input feature of the CNN layer; Feature normalization: Environmental features (temperature, humidity) and individual features (component operating years, individual drift deviation coefficient) are normalized in the range [0,1] to eliminate dimensional differences; Dataset partitioning: The dataset is divided into a training set (model fitting), a validation set (hyperparameter tuning), and a test set (performance verification) in an 8:1:1 ratio.

[0031] A concatenated architecture combining CNN temporal feature extraction and XGBoost multi-feature fusion is adopted to adapt to the fault feature characteristics in high-temperature scenarios: CNN layer (temporal feature extraction module): The input is a 10×4 temporal feature matrix (10 time steps × 4 core electrical parameters); the architecture is set to 2 convolutional layers + 1 pooling layer: the first convolutional layer (16 3×1 convolutional kernels, ReLU activation) extracts local temporal fluctuation features, the second convolutional layer (8 2×1 convolutional kernels, ReLU activation) enhances global trend features, followed by 1 max pooling layer (pooling kernel 2×1) to compress the feature dimension; finally, the convolutional output is transformed into a 16-dimensional temporal feature vector through a fully connected layer, which serves as one of the inputs to the XGBoost layer.

[0032] XGBoost layer (multi-feature fusion inference module): The input is a 16-dimensional temporal feature vector output by CNN + 2-dimensional environmental features + 2-dimensional individual features, for a total of 20-dimensional fused features; the architecture is set to 10 classification trees with a depth of 5, and the multi-dimensional features are non-linearly fused through a weighted splitting criterion to output the classification probability of the fault type (normal / hot spot / encapsulation aging, etc.).

[0033] On the test set, the model achieved an accuracy of ≥95%, a recall of ≥93%, and an F1-score of ≥94% for identifying high-temperature faults, meeting the accuracy requirements for engineering diagnosis and constructing a multi-dimensional fusion diagnostic model. In step S40, the specific process of accurately distinguishing between normal drift and fault characteristics in high-temperature scenarios through the edge 30-second lightweight cycle diagnosis and the anomaly triggering depth full model verification layering mechanism is as follows: The edge device (module controller) retrieves the compensated equivalent room temperature parameters (Voc, Isc, Pmax, FF) in real time with a fixed cycle of 30 seconds, and synchronously associates the current environmental characteristics (temperature, humidity) with the individual characteristics of the photovoltaic module (operating years, drift deviation coefficient) to form the input feature set for lightweight diagnosis. Only the XGBoost lightweight inference module in the multi-dimensional fusion diagnostic model (which removes the temporal feature layer of the CNN to reduce computation) is called to quickly calculate the parameter state based on the input feature set: The input feature set is compared with the normal dynamic threshold range and the normal drift boundary, and the preliminary judgment result is output: normal drift or suspected abnormality. When the multi-dimensional fusion diagnostic model outputs suspected abnormal parameters exceeding the dynamic threshold warning lower limit and local temperature exceeding the fault precursor threshold for two consecutive cycles (60 seconds), deep verification is initiated: The multi-dimensional fusion diagnostic model is retrieved, and the time-series parameter sequence of equivalent room temperature parameters for 10 minutes is input. CNN extracts dynamic features such as voltage fluctuation and power trend, and XGBoost fuses multi-dimensional features to output the fault classification probability. Infrared thermal images are acquired simultaneously, and secondary verification is performed using a high-temperature-specific fault feature library (e.g., hot spots must simultaneously meet the conditions of Pmax drop ≥10% + local temperature difference ≥15℃) to distinguish between normal drift (which conforms to theoretical drift rules) and fault features (which match specific fault conditions).

[0034] Example 2 like Figure 2 As shown in Embodiment 1, the present invention provides a photovoltaic module fault self-diagnosis device, comprising: High-temperature drift basic data and model building module: Collect multi-dimensional data to form a three-dimensional database of temperature-irradiance-parameter drift, build a ternary nonlinear prediction model, calculate the individual drift deviation coefficient of photovoltaic modules, and calibrate the critical fault parameter values ​​through extreme high-temperature tests; High-temperature adaptation dynamic threshold management module: Based on the ternary nonlinear prediction model and the individual drift deviation coefficient of photovoltaic modules, it generates dynamic thresholds in real time to adapt to high-temperature environments and the individual states of photovoltaic modules. Through edge computing and cloud collaborative architecture, it realizes real-time calculation and calibration update of dynamic threshold range. High-temperature resistant multi-source data acquisition and compensation processing module: Deploys a high-temperature resistant multi-source sensor network and completes the acquisition, cross-calibration and anomaly removal of multi-dimensional data under high-temperature scenarios, and then restores the equivalent room temperature parameters of photovoltaic modules through drift reverse compensation; Multi-dimensional fusion fault diagnosis module: Constructs a multi-dimensional fusion diagnosis model; through edge 30-second lightweight cycle diagnosis and abnormal triggering deep full model verification layering mechanism, it achieves accurate differentiation between normal drift and fault characteristics in high temperature scenarios.

[0035] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A self-diagnosis method for photovoltaic module faults, characterized in that: include: Step S10: Collect multidimensional data to form a three-dimensional database of temperature-irradiance-parameter drift, construct a ternary nonlinear prediction model, calculate the individual drift deviation coefficient of photovoltaic modules, and calibrate the critical fault parameter value through extreme high temperature test; Step S20: Based on the ternary nonlinear prediction model and the individual drift deviation coefficient of photovoltaic modules, a dynamic threshold that adapts to the high temperature environment and the individual state of photovoltaic modules is generated in real time. Through edge computing and cloud collaborative architecture, the dynamic threshold range is calculated and calibrated in real time. Step S30: Deploy a high-temperature resistant multi-source sensor network and complete the collection, cross-calibration and anomaly removal of multi-dimensional data under high-temperature scenarios, and then restore the equivalent room temperature parameters of the photovoltaic module through drift reverse compensation; Step S40: Construct a multi-dimensional fusion diagnostic model; through edge 30-second lightweight cycle diagnosis and abnormal triggering deep full model verification layering mechanism, achieve accurate differentiation between normal drift and fault characteristics in high temperature scenarios.

2. The photovoltaic module fault self-diagnosis method according to claim 1, characterized in that: The specific process of collecting multidimensional data to form a three-dimensional database of temperature-irradiance-parameter drift is as follows: Thirty photovoltaic modules from the same batch but with different years of operation were selected. In an environmental chamber, a scenario of gradient high temperature and different irradiance combinations was simulated. The open-circuit voltage Voc, short-circuit current Isc, maximum power Pmax, and fill factor FF of the photovoltaic modules were collected every 5 minutes. At the same time, the backsheet temperature, junction box temperature, and ambient humidity of the photovoltaic modules were collected. The data were collected for 24 hours at each temperature point to form a three-dimensional database of temperature-irradiance-parameter drift.

3. The self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process for constructing the ternary nonlinear prediction model is as follows: A backpropagation (BP) neural network was chosen as the basic architecture, with 2 neurons and 3 hidden layers. The first layer has 16 neurons, the second layer has 8 neurons, and the third layer has 4 neurons. The ReLU activation function was used to extract nonlinear features layer by layer. Three neurons were set to correspond to open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax, using a linear activation function. A Dropout layer was added between the hidden and output layers to randomly disable some neurons to avoid overfitting. Momentum gradient descent was used instead of traditional gradient descent to accelerate the model's convergence speed. When the predicted MAE of the drift coefficients of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax are verified to be ≤2.1%, 1.8%, and 2.5% respectively, and the root mean square error RMSE is ≤3.0% on the test set, a ternary nonlinear prediction model is finally constructed.

4. The self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process for calculating the individual drift deviation coefficient of photovoltaic modules is as follows: Filter the full data of the target photovoltaic module by the photovoltaic module number. The filtering criteria are: covering all temperature gradients from 40℃ to 60℃ and all irradiance intensities from 800 to 1200W / ㎡; removing abnormal data; and retaining the actual drift of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax in each scenario, as well as the common drift of the corresponding scenario. For each temperature and radiation combination scenario, calculate the individual drift deviation coefficient of the photovoltaic module in a single scenario: Individual drift deviation coefficient of photovoltaic module = actual drift amount of the target photovoltaic module / common drift amount of the same temperature and radiation scenario. Calculate the individual drift deviation coefficients of the photovoltaic module for open circuit voltage Voc, short circuit current Isc, and maximum power Pmax respectively.

5. The self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process for calibrating the critical fault parameter values ​​through extreme high-temperature testing is as follows: Select differentiated photovoltaic modules with no initial faults and covering different years of operation, and deploy electrical parameter testers, infrared thermal imagers, and EL testers in the environmental chamber to simultaneously monitor core electrical parameters, photovoltaic module temperature field, and internal structure; The environmental chamber was stabilized at 60℃, 1000W / ㎡, and 60% humidity. After the photovoltaic modules were left to stand for 2 hours to reach thermal equilibrium, a 72-hour continuous test was started. Real-time recording of open-circuit voltage Voc, maximum power Pmax, fill factor FF, local temperature differences, and EL image changes; capturing parameter fluctuations and characteristic evolution from normal drift to fault precursors to fault failure. Calibrate the critical fault parameter values: Extract parameters corresponding to the failure precursors of hot spots and encapsulation aging; combine a ternary nonlinear prediction model to calculate the adaptation critical value under a gradient temperature of 40℃-55℃; increase safety redundancy by 5%-10% to avoid misjudgment due to instantaneous interference. A table comparing temperature, fault type, fault precursors, and critical values ​​for failure was created to clarify the critical parameter values ​​for different high temperatures.

6. The self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process of generating real-time dynamic threshold ranges for adapting to high-temperature environments and individual photovoltaic modules is as follows: The core electrical parameters of the photovoltaic module are retrieved from the factory and used as the initial benchmark for threshold calculation; a high-temperature resistant fiber optic temperature sensor and a miniature irradiation sensor are integrated into the photovoltaic module to collect the current ambient temperature and surface irradiation intensity of the photovoltaic module in real time. Obtain the individual drift deviation coefficient of photovoltaic modules; The edge device performs real-time verification of the ambient temperature and surface irradiance of the photovoltaic modules, and removes outliers with individual drift deviation coefficients exceeding ±5% to ensure the reliability of the input data. The real-time collected ambient temperature and surface irradiance of the photovoltaic module are input into the ternary nonlinear prediction model, and the common drift coefficient range of open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax is output. Based on the core electrical parameters of the components at the factory, the parameter range after theoretical common drift is calculated: Formula logic: Common drift range = core electrical parameters at the factory × common drift coefficient range; Based on the individual drift deviation coefficient of photovoltaic modules, the common drift range is modified to obtain the individual dynamic threshold range: Correction logic: Since the parameter drift of photovoltaic modules at high temperatures is mostly negative attenuation, only the lower limit of the dynamic threshold interval is corrected. The formula is: lower limit of dynamic threshold interval = lower limit of common drift interval × individual drift deviation coefficient of photovoltaic module, while the upper limit remains unchanged. Individual dynamic threshold range corrections were performed for open-circuit voltage Voc, short-circuit current Isc, and maximum power Pmax, respectively.

7. The self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process for achieving real-time calculation and calibration update of the dynamic threshold range is as follows: At the edge, a high-temperature resistant sensor integrated into the photovoltaic module collects three types of core data at a frequency of once per minute: Current ambient temperature of photovoltaic modules and surface irradiance of photovoltaic modules; Obtain the open-circuit voltage Voc, short-circuit current Isc, maximum power Pmax, and individual drift deviation coefficient of the photovoltaic module; By using a ternary nonlinear prediction model, the common drift coefficient range is calculated by inputting ambient temperature and irradiance data. Then, based on the individual drift deviation coefficient of the photovoltaic module, a dynamic threshold range for individual photovoltaic module adaptation is generated. The dynamic threshold range is output to the edge diagnostic module in real time, compared with the collected photovoltaic module parameters, and the corresponding response is triggered. If the edge device loses network access, it will automatically activate the historical dynamic threshold range parameters cached locally to maintain the dynamic threshold range calculation, and re-upload the cached data after reconnecting to the network.

8. The self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process of restoring the equivalent room temperature parameters of photovoltaic modules through drift reverse compensation is as follows: Obtain the current ambient temperature, the surface irradiance of the photovoltaic module, and the measured electrical parameters of the module; Input real-time temperature and irradiance data into a ternary nonlinear prediction model and output the common drift coefficient range for the current scenario; The theoretical individual drift of a photovoltaic module is calculated by combining the individual drift amount with the individual drift deviation coefficient of the photovoltaic module: Formula: Theoretical individual drift amount = Common drift coefficient × Individual drift deviation coefficient; Convert the measured electrical parameters into percentages relative to a room temperature reference. Then, the equivalent room temperature parameters are restored through reverse compensation: Formula: Percentage of equivalent room temperature parameters = Percentage of measured parameters + Theoretical individual drift.

9. A self-diagnosis method for photovoltaic module faults according to claim 1, characterized in that: The specific process for achieving accurate differentiation between normal drift and fault characteristics in high-temperature scenarios is as follows: At the edge, the equivalent room temperature parameters after compensation are retrieved in real time with a fixed period of 30 seconds. The current environmental characteristics and individual characteristics of photovoltaic modules are then correlated to form an input feature set for lightweight diagnosis. By simply calling the XGBoost lightweight inference module in the multi-dimensional fusion diagnostic model, parameter states can be quickly calculated based on the input feature set: The input feature set is compared with the normal dynamic threshold range and the normal drift boundary, and the preliminary judgment result is output: normal drift or suspected abnormality. When the multi-dimensional fusion diagnostic model outputs suspected abnormal parameters exceeding the dynamic threshold warning lower limit and local temperature exceeding the fault precursor threshold for two consecutive cycles, deep verification is initiated: The multi-dimensional fusion diagnostic model is invoked, and a time-series parameter sequence of equivalent room temperature parameters for 10 minutes is input. CNN extracts dynamic features such as voltage fluctuations and power trends, and XGBoost fuses multi-dimensional features to output fault classification probabilities. Infrared thermal images are acquired simultaneously, and secondary verification is performed using a high-temperature-specific fault feature library to distinguish between normal drift and fault features.

10. A photovoltaic module fault self-diagnosis device, used to perform the method described in any one of claims 1-9, characterized in that: include: High-temperature drift basic data and model building module: Collect multi-dimensional data to form a three-dimensional database of temperature-irradiance-parameter drift, build a ternary nonlinear prediction model, calculate the individual drift deviation coefficient of photovoltaic modules, and calibrate the critical fault parameter values ​​through extreme high-temperature tests; High-temperature adaptation dynamic threshold management module: Based on the ternary nonlinear prediction model and the individual drift deviation coefficient of photovoltaic modules, it generates dynamic thresholds in real time to adapt to high-temperature environments and the individual states of photovoltaic modules. Through edge computing and cloud collaborative architecture, it realizes real-time calculation and calibration update of dynamic threshold range. High-temperature resistant multi-source data acquisition and compensation processing module: Deploys a high-temperature resistant multi-source sensor network and completes the acquisition, cross-calibration and anomaly removal of multi-dimensional data under high-temperature scenarios, and then restores the equivalent room temperature parameters of photovoltaic modules through drift reverse compensation; Multi-dimensional fusion fault diagnosis module: Constructs a multi-dimensional fusion diagnosis model; through edge 30-second lightweight cycle diagnosis and abnormal triggering deep full model verification layering mechanism, it achieves accurate differentiation between normal drift and fault characteristics in high temperature scenarios.

Images