Method and device for predicting service life of photovoltaic module

Abstract

The invention relates to the technical field of data processing, and discloses a photovoltaic module life prediction method and device. The method comprises the following steps: carrying out characteristic curve modeling and normalization processing on an operation data set to obtain an actual working characteristic curve and performance parameters of a component; feature vector construction and pattern classification are carried out on the actual working characteristic curve and the performance parameters, and a feature library containing multiple degradation types is obtained; performing mode matching on the real-time operation features of the component and the feature library through a dynamic time warping algorithm to obtain a degradation state judgment result of the component; analyzing and modeling the degradation state judgment result and the historical data through a deep learning algorithm to obtain a residual life prediction value of the component; and carrying out risk level evaluation and maintenance cost benefit analysis on the residual life prediction value to obtain a maintenance decision suggestion of the component. According to the invention, the efficiency and accuracy of photovoltaic module life prediction are improved.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method and device for predicting the lifespan of photovoltaic modules. Background Art

[0002] Existing methods for predicting the lifespan of photovoltaic modules usually rely on a single monitoring index, such as a simple trend analysis of voltage or current. However, this analysis method based on a single parameter has great limitations and often cannot accurately reflect the degradation state of photovoltaic modules. Traditional methods mainly rely on regular manual inspections or the calculation of lifespan based on fixed-point historical data. However, due to the long-term exposure of photovoltaic modules in the outdoor environment and being affected by multiple environmental factors such as temperature, humidity, and irradiance, their performance changes are complex and diverse. Existing lifespan prediction technologies lack the integration and dynamic analysis of multi-dimensional data, making it difficult to achieve real-time and accurate lifespan assessment, resulting in components not being able to obtain early warnings in time before potential failures, affecting the long-term stability and power generation benefits of photovoltaic systems.

[0003] The deficiencies of the existing technology are that it is impossible to systematically distinguish and judge the types of performance degradation of photovoltaic modules. Especially when multiple degradation types (such as PID effect, hot spot effect, etc.) appear intertwined, it is difficult to accurately identify the degradation state of the components. In addition, traditional degradation detection methods usually rely on simple empirical formulas or statistical models and cannot adapt to the complex dynamic environment during the operation of photovoltaic modules. These methods lack accuracy and flexibility in actual operation, resulting in a low accuracy of lifespan prediction and making it difficult to provide effective support for operation and maintenance decisions. Therefore, it is particularly important to develop a method for predicting the lifespan of photovoltaic modules based on multi-channel data acquisition and dynamic analysis to achieve accurate prediction of the component lifespan and provide reasonable maintenance suggestions. Summary of the Invention

[0004] This application provides a method and device for predicting the lifespan of photovoltaic modules to improve the efficiency and accuracy of predicting the lifespan of photovoltaic modules.

[0005] In a first aspect, the present application provides a method for predicting the life of a photovoltaic module. The method for predicting the life of a photovoltaic module includes: performing multi-channel synchronous acquisition on the voltage, current, temperature, and environmental parameters of the photovoltaic module to obtain a component operation data set with a unified time stamp; performing characteristic curve modeling and normalization processing on the operation data set to obtain the actual working characteristic curve and performance parameters of the component; constructing a feature vector and performing pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library including various degradation types; performing pattern matching on the real-time operation characteristics of the component and the feature library through a dynamic time warping algorithm to obtain a discrimination result of the degradation state of the component; performing analysis and modeling on the discrimination result of the degradation state and historical data through a deep learning algorithm to obtain a predicted remaining life value of the component; performing risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value to obtain a maintenance decision recommendation for the component.

[0006] In a second aspect, the present application provides a device for predicting the life of a photovoltaic module. The device for predicting the life of a photovoltaic module includes:

[0007] An acquisition module, configured to perform multi-channel synchronous acquisition on the voltage, current, temperature, and environmental parameters of the photovoltaic module to obtain a component operation data set with a unified time stamp;

[0008] A modeling module, configured to perform characteristic curve modeling and normalization processing on the operation data set to obtain the actual working characteristic curve and performance parameters of the component;

[0009] A classification module, configured to construct a feature vector and perform pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library including various degradation types;

[0010] A matching module, configured to perform pattern matching on the real-time operation characteristics of the component and the feature library through a dynamic time warping algorithm to obtain a discrimination result of the degradation state of the component;

[0011] A modeling module, configured to perform analysis and modeling on the discrimination result of the degradation state and historical data through a deep learning algorithm to obtain a predicted remaining life value of the component;

[0012] An analysis module, configured to perform risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value to obtain a maintenance decision recommendation for the component.

[0013] In the technical solution provided by this application, by synchronously collecting the voltage, current, temperature and environmental parameters of photovoltaic modules through multiple channels, a component operation dataset with a unified timestamp is obtained, significantly improving the accuracy and consistency of data collection and laying a reliable data foundation for life prediction. By performing characteristic curve modeling and normalization processing on the collected data, the actual working characteristic curve and performance parameters of the components are obtained. This processing step effectively reduces the influence of environmental factors on the measurement results, enabling the performance parameters of the components under standardized conditions to more truly reflect the actual working state of the components. In addition, by constructing eigenvectors and performing pattern classification on the actual working characteristic curve and performance parameters, a feature library containing multiple degradation types is established, providing a rich reference basis for the system to distinguish and identify different degradation types and ensuring the accuracy and reliability of the degradation discrimination results. During the process of identifying the degradation state, this solution introduces the dynamic time warping algorithm to perform pattern matching between the real-time operation characteristics of the components and the feature library, and to judge the degradation state of the components in real time. Through the dynamic matching of the operation characteristics of the components by this algorithm, the system can more accurately identify the degradation patterns, overcoming the limitations of traditional static analysis methods in dealing with diverse degradation characteristics. At the same time, the solution analyzes and models the degradation state discrimination results and historical data through deep learning algorithms to construct a life prediction model for photovoltaic modules. Through this model, the system can dynamically predict the remaining life of the components, provide early warnings for the operation and maintenance teams, facilitate the adoption of timely maintenance measures, prevent sudden failures of the components, and ensure the long-term stable operation of the photovoltaic system.

[0014] In addition, based on life prediction, this solution further combines risk assessment and cost-benefit analysis to evaluate the risk level of the remaining life prediction value and analyze the maintenance cost-benefit, generating maintenance decision suggestions for the components. This function effectively balances the maintenance cost and system reliability. By comprehensively evaluating factors such as the degradation degree, risk level, and operation and maintenance cost of the components, it provides scientific decision support for the operation and maintenance of the photovoltaic system, extends the service life of the components, and reduces the operation cost. Generally speaking, through the integrated application of multi-channel data collection, dynamic time warping algorithm and deep learning model, this solution realizes the high-precision, real-time and intelligent life prediction of photovoltaic modules, not only improving the accuracy of degradation state discrimination, but also optimizing the economy and effectiveness of maintenance strategies, providing effective guarantee for the long-term stable operation of the photovoltaic system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic diagram of an embodiment of the photovoltaic module life prediction method in the embodiments of the present application;

[0017] Figure 2 It is a schematic diagram of an embodiment of the photovoltaic module life prediction device in the embodiments of the present application. Detailed implementation manners

[0018] The embodiments of the present application provide a photovoltaic module life prediction method and device. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and the above-mentioned drawings of the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "comprising" or "having" and any deformation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 An embodiment of the photovoltaic module life prediction method in the embodiments of the present application includes:

[0020] Step S101: Perform multi-channel synchronous acquisition on the voltage, current, temperature and environmental parameters of the photovoltaic module to obtain a component operation data set with a unified time stamp;

[0021] Step S102: Perform characteristic curve modeling and normalization processing on the operation data set to obtain the actual working characteristic curve and performance parameters of the component;

[0022] Step S103: Construct a feature vector and perform pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library containing multiple degradation types;

[0023] Step S104: Perform pattern matching on the real-time operation characteristics of the component and the feature library through the dynamic time warping algorithm to obtain the degradation state discrimination result of the component;

[0024] Step S105: Analyze and model the degradation state discrimination result and historical data through a deep learning algorithm to obtain the predicted remaining life value of the component;

[0025] Step S106: Perform risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value to obtain maintenance decision-making suggestions for the component.

[0026] It is understandable that the execution entity of this application can be a photovoltaic module life prediction device, or it can also be a terminal or a server, and specific limitations are not made here. In the embodiments of this application, the server is taken as an example of the execution entity for illustration.

[0027] Specifically, data such as the voltage, current, temperature, and environmental parameters of the photovoltaic module are obtained through multi-channel synchronous acquisition. Specifically, high-speed sampling is performed on the DC bus to obtain the original voltage data, and then the data is digitized through 16-bit AD conversion to obtain voltage values convenient for subsequent processing. At the same time, 50Hz power frequency interference filtering and zero-drift compensation processing are performed on the digitized voltage data to eliminate signal noise and drift errors, and the IEEE1588 protocol is used for time synchronization to make the data carry a unified timestamp. In current acquisition, a Hall current sensor is used to obtain the string current data, and its effective value calculation and temperature compensation are performed to obtain accurate calibrated current data. For temperature acquisition, a PT100 temperature sensor is used to obtain the distribution data of the surface temperature of the module in a scanning manner, and spatial interpolation and outlier rejection are performed on the temperature data to form the temperature field data of the module. After time sequence alignment, the voltage data, current data, and temperature field data are integrated into a synchronous acquisition data packet and transmitted to the main control unit through the POWERBUS bus, and finally a component operation data set with a unified timestamp is generated. Based on the obtained data set, characteristic curve modeling and normalization processing are performed on the operation data such as voltage and current to generate the actual working characteristic curve and performance parameters of the component. When modeling, the single-diode equivalent circuit equation is used to fit the voltage and current data, so as to obtain the reference I-V characteristic curve, and key point parameters are further extracted, including short-circuit current, open-circuit voltage, voltage and current values at the maximum power point. To eliminate the influence of temperature and irradiance changes on the data, the scheme performs temperature correction and irradiance normalization on these parameters to generate performance parameters under standard test conditions. Then, equivalent circuit parameters are calculated through series and parallel resistances, and the fill factor and power attenuation rate are calculated therefrom, thus completing the construction of the actual working characteristic curve of the component and providing reliable parameters for further analysis.

[0028] Subsequently, eigenvector construction and pattern classification are performed on the actual working characteristic curve and performance parameters to form a feature library containing various degradation types. Specifically, according to different degradation modes of components, such as PID effect, hot spot effect, hidden crack, and encapsulation material aging, etc., the data is classified to obtain a degradation mode feature matrix. Thresholds for characteristic parameters such as voltage anomaly, temperature anomaly, power attenuation, and resistance change are set for each degradation mode to form a systematic degradation discrimination criterion. By constructing eigenvectors for these degradation characteristics, various degradation modes are described mathematically, and then the weights of each characteristic parameter are calculated according to the correlation. Finally, a feature library covering various degradation types is formed, providing a detailed basis for subsequent degradation state discrimination. With the support of this feature library, real-time pattern matching is realized through the dynamic time warping algorithm (DTW) to discriminate the degradation state of components. Specifically, the operation characteristic data of components collected in real time is standardized to generate a standardized eigenvector, and the similarity between it and the degradation modes in the feature library is calculated through the DTW algorithm. The similarity calculation helps to identify whether the operation characteristics of the component are similar to a certain known degradation mode and generates a feature matching degree, indicating the possibility of degradation characteristics. Then, the feature matching degree is evaluated according to the fuzzy membership function to obtain the confidence level of the degradation mode, and the confidence level is compared with the set discrimination threshold to obtain a multi-level degradation state discrimination result.

[0029] Based on the obtained degradation state discrimination results and historical data, the life of the component is predicted through a deep learning algorithm. The solution uses a recurrent neural network containing three LSTM layers for time series feature extraction, constructs a health index data sequence, and gradually reduces the number of neurons (128, 64, 32) layer by layer to improve the accuracy of feature extraction. Through this model, the future change trend of the health index is predicted to generate a health index prediction sequence. Finally, through cross-analysis with the preset failure threshold, the remaining life prediction value of the photovoltaic component is obtained to ensure the dynamics and accuracy of life prediction. On the basis of life prediction, risk level assessment and maintenance cost-benefit analysis are performed on the remaining life prediction value to generate maintenance decision suggestions. The risk assessment first performs a risk matrix analysis on the remaining life value according to the failure probability and impact degree to obtain the risk level of the component. Based on this risk level, the costs and benefits of the maintenance plan are comprehensively calculated to generate an economic evaluation result. Finally, the economic evaluation result is combined with the operation plan and spare parts inventory situation of the power station to obtain comprehensive maintenance decision suggestions, providing data support and optimization solutions for the operation and management of the photovoltaic system.

[0030] For example, assume that the collected voltage data is 300V, the current data is 5A, and the temperature is 35°C. The timestamp data obtained through synchronous acquisition aligns the parameters. Based on these data, an I-V characteristic curve is constructed and the short-circuit current of 5.2A, open-circuit voltage of 305V, and maximum power point of 295V and 4.9A are extracted. After temperature correction of the key point parameters and normalization based on the standard irradiance, the performance parameters under standard test conditions are obtained. In the construction of the feature library, the identified features conform to the PID effect mode, and the feature matching degree reaches 90%. After the deterioration state is judged, the change trend of the health index is predicted through the LSTM network, and the remaining life prediction value for the next 6 months is obtained. According to this life value, it is determined that the component is in the medium risk level in the risk assessment. Combining the maintenance cost-benefit analysis, a maintenance decision recommendation for replacement half a year in advance is generated, thus ensuring the stable operation of the photovoltaic system.

[0031] In the embodiments of the present application, by synchronously collecting the voltage, current, temperature, and environmental parameters of the photovoltaic module through multiple channels, a component operation data set with a unified timestamp is obtained, significantly improving the accuracy and consistency of data collection and laying a reliable data foundation for life prediction. By performing characteristic curve modeling and normalization processing on the collected data, the actual working characteristic curve and performance parameters of the component are obtained. This processing step effectively reduces the influence of environmental factors on the measurement results, enabling the performance parameters of the component under standardized conditions to more truly reflect the actual working state of the component. In addition, by constructing feature vectors and performing pattern classification on the actual working characteristic curve and performance parameters, a feature library containing multiple deterioration types is established, providing a rich reference basis for the system to distinguish and identify different deterioration types and ensuring the accuracy and reliability of the deterioration discrimination results. In the process of identifying the deterioration state, this solution introduces the dynamic time warping algorithm to perform pattern matching between the real-time operation characteristics of the component and the feature library, and to judge the deterioration state of the component in real time. Through the dynamic matching of the operation characteristics of the component by this algorithm, the system can more accurately identify the deterioration mode, overcoming the limitations of traditional static analysis methods in dealing with diverse deterioration characteristics. At the same time, the solution analyzes and models the deterioration state discrimination results and historical data through deep learning algorithms, and constructs a life prediction model for the photovoltaic module. Through this model, the system can dynamically predict the remaining life of the component, providing an early warning for the operation and maintenance team, facilitating timely maintenance measures to be taken, preventing sudden failures of the component, and ensuring the long-term stable operation of the photovoltaic system.

[0032] In addition, based on the remaining useful life prediction, this solution further combines risk assessment and cost-benefit analysis to evaluate the risk level of the remaining useful life prediction value and analyze the maintenance cost-benefit, generating maintenance decision-making suggestions for the components. This function effectively balances the maintenance cost and system reliability. By comprehensively evaluating factors such as the deterioration degree, risk level, and operation and maintenance cost of the components, it provides scientific decision-making support for the operation and maintenance of the photovoltaic system, extends the service life of the components, and reduces the operation cost. Generally speaking, through the integrated application of multi-channel data acquisition, dynamic time warping algorithm, and deep learning model, this solution realizes the high-precision, real-time, and intelligent prediction of the remaining useful life of photovoltaic modules, not only improves the accuracy of the deterioration state discrimination, but also optimizes the economy and effectiveness of the maintenance strategy, providing an effective guarantee for the long-term stable operation of the photovoltaic system.

[0033] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0034] (1) Perform high-speed sampling on the DC bus of the photovoltaic module to obtain the original voltage sampling data, and perform 16-bit AD conversion on the original voltage sampling data to obtain the digital voltage data;

[0035] (2) Filter out the 50Hz power frequency interference and compensate for the zero drift of the digital voltage data to obtain the calibrated voltage data, and perform time synchronization processing on the calibrated voltage data according to the IEEE1588 protocol to obtain the voltage data with time stamps;

[0036] (3) Collect the string current through a Hall current sensor to obtain the original current data, and calculate the effective value and perform temperature compensation on the original current data to obtain the calibrated current data;

[0037] (4) Scan and collect the surface temperature of the module through a PT100 temperature sensor to obtain the temperature distribution data, and perform spatial interpolation and outlier rejection on the temperature distribution data to obtain the component temperature field data;

[0038] (5) Align the time sequence and pack the voltage data with time stamps, the calibrated current data, and the component temperature field data to obtain a synchronous acquisition data packet, and transmit the synchronous acquisition data packet to the main control unit through the POWERBUS bus to obtain a component operation data set with a unified time stamp.

[0039] Specifically, high-speed sampling is performed on the DC bus of the photovoltaic module to obtain accurate voltage data. Specifically, through high-speed sampling, the real-time changes in the bus voltage can be captured, and 16-bit AD (analog-to-digital) conversion is performed on these original voltage sampling data to generate more accurate digital voltage data. The 16-bit AD conversion can finely convert the sampled analog voltage signal into a digital signal for subsequent data processing. After obtaining the digital voltage data, to ensure the accuracy of the data, 50Hz power frequency interference filtering and zero-drift compensation operations are then carried out. 50Hz power frequency interference filtering can eliminate the noise interference from the power frequency, and zero-drift compensation is used to correct the voltage offset caused by circuit drift, making the calibrated voltage data more accurate. The calibrated voltage data is then processed for time synchronization according to the IEEE1588 protocol, attaching a timestamp to each piece of data, thereby ensuring the consistency of data acquisition time. This time synchronization processing is crucial because it strictly aligns the voltage data with other collected data in time, providing a reliable time marker for further analysis of the data.

[0040] In the acquisition of current data, a Hall current sensor is used to collect the current of the photovoltaic module string to obtain the original current data. The Hall sensor can detect the actual value of the current and avoid contact loss and loss problems in current transmission. The effective value of the original current data is calculated to obtain the average current value, and at the same time, temperature compensation is carried out to correct the influence of the ambient temperature on the current measurement. These processing operations generate calibrated current data to ensure the accuracy of the current data under different temperature conditions. In addition, the surface temperature of the photovoltaic module is scanned and collected by a PT100 temperature sensor to obtain the temperature distribution data of the module. The PT100 sensor is suitable for detecting the surface temperature of the module due to its high precision and good linearity. Subsequently, spatial interpolation processing is performed on the collected temperature distribution data to fill the blank areas of the surface temperature data of the module, ensuring the continuity and integrity of the temperature data of the entire surface. For possible abnormal temperature points, the temperature data is further optimized by eliminating the abnormal points, and finally, accurate component temperature field data is generated. This temperature field data reflects the temperature change of the surface of the photovoltaic module and is an important parameter for analyzing the thermal effect and attenuation of the photovoltaic module.

[0041] After the data acquisition is completed, the voltage data with timestamps, the calibrated current data, and the component temperature field data are aligned in time series and packed. The time series alignment precisely matches the data from different sources in time, ensuring that they form complete operating data points at the same time point. The aligned data is packed to generate a synchronous acquisition data packet, which contains the complete records of voltage, current, and temperature data at the same time point. Finally, the synchronous acquisition data packet is transmitted to the main control unit through the POWERBUS bus to ensure the stability and real-time performance of data transmission. After this step of processing, a component operating data set with a unified timestamp is obtained, laying a reliable foundation for subsequent characteristic analysis and life prediction.

[0042] For example, assume that the original voltage data collected from the DC bus of a photovoltaic module at a certain time point is 320V, and the digitized voltage value generated after 16-bit AD conversion is 65520. Subsequently, 50Hz interference filtering and zero-drift compensation are performed on this data to obtain a calibrated voltage value of 319.8V, with the timestamp "T1" attached. The original current data measured by the Hall current sensor is 8A, and after RMS calculation and temperature compensation, the calibrated current data is 7.95A. When collecting temperature, the temperature values collected at each point on the surface of the module are 25°C to 35°C. The data at unmeasured points is filled by spatial interpolation, and abnormal high-temperature points are removed to obtain an average temperature field data of 30°C. After aligning and packing these voltage, current, and temperature data with timestamps, a synchronous acquisition data packet is generated and transmitted to the main control unit through the POWERBUS bus to obtain a component operating data set with the timestamp "T1", providing high-precision initial data for the next analysis and prediction.

[0043] In a specific embodiment, the process of performing step S102 may specifically include the following steps:

[0044] (1) Fit the voltage and current data in the operating data set according to the single-diode equivalent circuit equation to obtain the reference I-V characteristic curve of the component, and extract the key point parameters in the reference I-V characteristic curve to obtain the short-circuit current, open-circuit voltage, maximum power point voltage, and current value;

[0045] (2) Perform temperature correction on the key point parameters according to the temperature coefficient to obtain the characteristic parameters at the standard temperature, and normalize the characteristic parameters at the standard temperature according to the irradiance ratio to obtain the performance parameters under standard test conditions;

[0046] (3) Calculate the series and parallel resistances of the performance parameters under the standard test conditions to obtain the equivalent circuit parameters of the component, and calculate the fill factor and power attenuation rate of the component through the equivalent circuit parameters to obtain the actual operating characteristic curve and performance parameters of the component.

[0047] Specifically, in the photovoltaic module life prediction method, based on the voltage and current data in the operation dataset, fitting is performed through the single-diode equivalent circuit equation to obtain the reference I-V characteristic curve of the module. The single-diode equivalent circuit model is a commonly used model to describe the working characteristics of photovoltaic modules, and its equation can accurately simulate the relationship between the current and voltage of photovoltaic modules. During the fitting process, according to the voltage and current data in the operation dataset, the parameters in the model are continuously adjusted through an optimization algorithm to minimize the fitting error as much as possible, so that the generated I-V characteristic curve can truly reflect the electrical characteristics of the photovoltaic module. The reference I-V characteristic curve can intuitively display the current and voltage characteristics of the photovoltaic module under specific conditions, which includes some key point parameters, including short-circuit current (the current when the two ends of the module are short-circuited), open-circuit voltage (the voltage when the module has no load), maximum power point voltage and current (the voltage and current when the module outputs the maximum power). After extracting these key point parameters, temperature correction is performed on them according to the temperature coefficient. The temperature coefficient is a scaling factor for the characteristics of photovoltaic modules to change with temperature, and the temperature coefficients of different modules are slightly different, usually provided by the module manufacturer. Through temperature correction, these key point parameters can be adjusted to the characteristic parameters at the standard temperature (such as 25°C), eliminating the influence of external temperature on the voltage and current data, thereby improving the accuracy of the analysis. Next, irradiance normalization is performed on the corrected characteristic parameters. Irradiance normalization is to eliminate the influence of the change in solar irradiance intensity on the performance of the module and ensure that all data are normalized to the parameter state under standard test conditions (such as 1000 W / m2). Through this normalization process, the performance parameters under standard test conditions are obtained, ensuring the consistency of data under different measurement conditions.

[0048] After obtaining the performance parameters under standard test conditions, further calculation of the series and parallel resistances is performed on these parameters to determine the equivalent circuit parameters of the photovoltaic module. The equivalent circuit of a photovoltaic module usually includes a series resistance and a parallel resistance. The series resistance mainly affects the short-circuit current of the module, while the parallel resistance affects the open-circuit voltage. By calculating these two resistances, a more accurate circuit model of the photovoltaic module can be established, and these equivalent circuit parameters are important indicators reflecting the internal losses of the photovoltaic module. Then, the fill factor and power attenuation rate of the module are calculated using these equivalent circuit parameters. The fill factor is a parameter to measure the output efficiency of a photovoltaic module, defined as the ratio of the maximum power to the product of the open-circuit voltage and the short-circuit current. The power attenuation rate reflects the power loss of the module during long-term operation and is a direct manifestation of the degradation state of the module. Finally, through this series of calculations, the actual working characteristic curve and performance parameters of the module are obtained, providing accurate basic data for further degradation analysis and life prediction.

[0049] For example, assume that on the I-V characteristic curve of a certain photovoltaic module, the key point parameters extracted include: short-circuit current of 8 A, open-circuit voltage of 600 V, maximum power point voltage of 550 V, and corresponding current of 7.5 A. First, according to the module temperature coefficient, these data are corrected to 25 °C, resulting in a short-circuit current of 7.8 A and an open-circuit voltage of 595 V. Then, irradiance normalization is performed on these temperature-corrected parameters. Assuming the current irradiance is 800 W / m2, the normalized short-circuit current is 9.75 A, and the open-circuit voltage remains unchanged. Next, based on the equivalent circuit calculation, the series resistance is 0.4 Ω and the parallel resistance is 500 Ω. Using these parameters, the fill factor is calculated to be approximately 0.75, and the power attenuation rate is 5%. These calculation results provide accurate data support for the actual working characteristics of the module and reflect the degradation state of the module, laying a foundation for subsequent life prediction and degradation analysis.

[0050] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0051] (1) Classify the actual working characteristic curve according to degradation types such as PID effect, hot spot effect, module microcrack, and encapsulation material aging to obtain a degradation mode feature matrix, and set threshold values for characteristic parameters such as voltage anomaly, temperature anomaly, power attenuation, and resistance change for the degradation mode feature matrix to obtain a degradation discrimination criterion;

[0052] (2) Construct a feature vector for the degradation discrimination criterion to obtain a mathematical description of the degradation mode, and calculate the parameter weights according to the correlation for the mathematical description of the degradation mode to obtain a feature library containing multiple degradation types.

[0053] Specifically, by classifying the actual working characteristic curves of photovoltaic modules, a detailed deterioration mode feature library is established to facilitate the accurate discrimination and analysis of different types of deterioration. First, according to the actual working characteristic curves of photovoltaic modules, they are classified into common deterioration types such as PID effect, hot spot effect, module microcracks, and encapsulation material aging, obtaining a deterioration mode feature matrix. The PID effect is a performance degradation phenomenon caused by the surface potential difference of photovoltaic modules; the hot spot effect is that the local temperature is too high, resulting in local heating and affecting the module performance; module microcracks refer to the generation of fine cracks inside the module due to stress or manufacturing defects; encapsulation material aging is the degradation of the performance of the encapsulation material under the long-term action of environmental factors such as light, temperature, and humidity. By classifying the characteristic curves according to these deterioration types, the formed feature matrix can intuitively reflect the mode features of various deteriorations. Next, specific characteristic parameter thresholds are set for the deterioration mode feature matrix to form a deterioration discrimination criterion. The characteristic parameters include voltage anomaly, temperature anomaly, power attenuation, and resistance change, etc., which are the key manifestations of various deterioration modes. For example, the PID effect usually shows a rapid voltage drop, while the hot spot effect will cause a sharp rise in temperature. By setting thresholds for these parameters, the triggering conditions for each deterioration mode can be defined, thus forming a systematic deterioration discrimination criterion, enabling different types of deterioration to be identified through specific parameter anomalies. This discrimination criterion provides a unified and objective evaluation basis for the judgment of deterioration modes in practical applications.

[0054] Based on the deterioration discrimination criterion, characteristic vectors are constructed for the characteristic parameters of each deterioration type to form a mathematical description of the deterioration mode. The characteristic vector is a multi-dimensional vector containing the characteristic parameters under each deterioration type. Through this mathematical description, the characteristics of different deterioration types can be numerically and vectorially represented, facilitating subsequent automated analysis. Subsequently, the parameter weight calculation is carried out for these characteristic vectors according to their correlation. The weight calculation sets the weight coefficients according to the importance of each parameter in a specific deterioration mode. For example, in the deterioration mode of the hot spot effect, the weight of temperature anomaly is relatively high, while in the PID effect, the weight of voltage change is relatively high. Through weight calculation, the key features can be highlighted when analyzing the characteristic vectors, ensuring the accuracy of the deterioration mode description. Finally, this series of data processing and calculations result in a feature library containing multiple deterioration types. The feature library includes the mathematical descriptions of various deteriorations and the corresponding weight information, which is an important reference basis for the deterioration analysis and life prediction of photovoltaic modules. Through the feature library, the subsequent system can quickly match the deterioration modes in the real-time operation data, realize the real-time discrimination of the module state and the assessment of the deterioration degree, providing a key basis for subsequent maintenance decisions.

[0055] For example, assume that the actual operating characteristic curve of the component shows that the voltage decay rate is higher than the normal level and there is a significant rise in the local temperature in a certain area. Through classification, it is determined to be a mixed degradation of the PID effect and the hot spot effect. In the degradation mode feature matrix, the voltage threshold for the PID effect is set to -5V, while the temperature threshold for the hot spot effect is set to +10°C. In actual measurement, the voltage change is -6V and the temperature rise is +12°C, both exceeding the threshold range. After vectorizing these parameter features and assigning weights according to the degradation type respectively, the update of the degradation feature library is obtained, and based on this library, the degradation mode of the component is quickly identified as the combined state of the PID effect and the hot spot effect, providing a reliable data basis for life prediction and maintenance.

[0056] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0057] (1) Perform data standardization processing on the real-time operation characteristics of the component to obtain a standardized feature vector, and calculate the similarity between the standardized feature vector and the degradation modes in the feature library through the dynamic time warping algorithm to obtain the feature matching degree;

[0058] (2) Evaluate the feature matching degree according to the fuzzy membership function to obtain the confidence level of the degradation mode, and perform multi-level discrimination on the confidence level through a set discrimination threshold to obtain the discrimination result of the degradation state of the component.

[0059] Specifically, accurate discrimination of the degradation state of photovoltaic components is achieved through data standardization processing and the dynamic time warping algorithm (DTW). First, perform data standardization processing on the real-time operation characteristics of the component to convert each feature data into a standardized feature vector. Data standardization is to normalize data with different dimensions to ensure that each data in the feature vector is within the same numerical range, thereby avoiding weight deviation caused by data range differences. In the standardization process, methods such as mean-variance normalization can be used to adjust the data to the same standard level for subsequent algorithm processing. After obtaining the standardized feature vector, calculate the similarity between this feature vector and the degradation modes in the feature library through the dynamic time warping algorithm (DTW). DTW is an algorithm for measuring the similarity between two time series, especially suitable for processing data with time axis changes. In this solution, the DTW algorithm is used to compare the similarity between the real-time operation characteristics of the component and the known degradation mode sequences in the feature library. Specifically, DTW calculates the minimum "distance" between the real-time feature sequence and the degradation mode sequence through dynamic programming, thereby obtaining the feature matching degree. The feature matching degree represents the similarity between the real-time operation data and the known degradation mode, and is the basic data for degradation state judgment.

[0060] After calculating the feature matching degree, a fuzzy membership function is used to evaluate the feature matching degree to obtain the confidence level of the degradation mode. The fuzzy membership function is a method for processing fuzzy data. By defining different membership functions, the matching degree values are fuzzily evaluated to generate a confidence level between 0 and 1. The higher the confidence level, the closer the current operating state of the component is to the matching degree of a specific degradation mode. After obtaining the confidence level, a multi-level discrimination is performed on the confidence level according to a preset discrimination threshold. The multi-level discrimination can be divided into levels such as "mild degradation", "moderate degradation", and "severe degradation". Through these level divisions, the system can further refine the degradation state of the component and generate a clear discrimination result of the degradation state.

[0061] For example, assume that in real-time monitoring, the feature data of the component forms a feature vector \([0.8, 0.6, 0.75]\) after standardization, and this vector is matched with a set of PID effect degradation modes in the feature library. Through the calculation of the DTW algorithm, the feature matching degree between the two is 0.85. Subsequently, based on the fuzzy membership function, the matching degree is evaluated, and the confidence level of the degradation mode is generated as 0.78. Finally, the discrimination threshold is set to 0.7, and through multi-level discrimination, this confidence level is classified as the "moderate degradation" state, thereby generating the discrimination result of the component's degradation state. This result can be used to determine whether the component needs maintenance or further detection, thus providing strong support for subsequent life prediction and operation and maintenance decisions.

[0062] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0063] (1) Extract the time series features of the degradation state discrimination result and historical data to obtain a health index data sequence, and construct a recurrent neural network containing three LSTM layers based on the health index data sequence, with the number of neurons in each layer being 128, 64, and 32 respectively, to obtain a deep learning model;

[0064] (2) Predict the future change trend of the health index through the deep learning model to obtain a health index prediction sequence, and perform cross-analysis on the health index prediction sequence and a preset failure threshold to obtain the remaining life prediction value of the component.

[0065] Specifically, an effective life prediction model is established through the analysis of the deterioration state discrimination results and historical data. First, time series features are extracted from the deterioration state discrimination results and historical data to construct a health index data sequence. The health index is a numerical indicator reflecting the current health state of the component, which is obtained based on the change patterns of the real-time deterioration state and historical data. The time series feature extraction aims to extract the characteristic trends of the health index changing over time, unify the deterioration states at different times into a sequence for analysis, and provide continuous and dynamic input data for the subsequent neural network model. After obtaining the health index data sequence, a recurrent neural network containing three LSTM (Long Short-Term Memory) layers is constructed based on this sequence. LSTM is a neural network structure suitable for time series data and is widely used in time series prediction tasks because it can learn the long-term dependence features of the data. In this solution, the first layer of the LSTM network is set to 128 neurons to capture the long-term trend of the health index change; the second layer contains 64 neurons to further extract the medium-term dynamic features; the third layer contains 32 neurons to focus on the detailed processing of short-term changes. Through the gradually decreasing number of neurons in each layer, the model can gradually filter information, retain the key features in the health index data, and finally obtain the deep learning model.

[0066] Based on this deep learning model, the future change trend of the health index of the photovoltaic module is predicted to generate a health index prediction sequence. The health index prediction sequence is the estimated future health state output by the model through analyzing the current data sequence, reflecting the health changes of the component during the prediction period. After generating this prediction sequence, it is cross-analyzed with a preset failure threshold. The failure threshold is a fixed reference point, and when the health index is lower than this threshold, the component is considered to be approaching failure. Through cross-analysis, it can be determined when the prediction sequence touches or approaches the failure threshold, thereby obtaining the remaining life prediction value of the component. This prediction value is the estimated remaining service time of the component without taking intervention measures and is an important reference for formulating maintenance and replacement plans.

[0067] For example, assume that the health index data sequence shows as: [95, 92, 88, 85, 82] during the operation of a certain photovoltaic module. This data indicates that the health index is gradually decreasing, reflecting the continuous deterioration state of the component. Inputting this sequence into the constructed LSTM model, the health index prediction sequence for the next 3 months is obtained as: [78, 75, 71]. Setting the failure threshold to 70, through cross-analysis, it can be found that the last item in the health index prediction sequence is close to 70, indicating that the component may reach the failure threshold within the next 3 months. Therefore, the remaining life prediction value is approximately 3 months. Based on this prediction result, replacement or maintenance strategies can be planned in advance to avoid sudden failures of the component before it fails.

[0068] In a specific embodiment, the process of performing step S106 may specifically include the following steps:

[0069] (1) Perform a risk matrix analysis on the remaining life prediction value according to the failure probability and impact degree to obtain the component risk level, and calculate the cost and benefit of the maintenance plan according to the component risk level to obtain the economic evaluation result of the maintenance plan;

[0070] (2) Combine the economic evaluation result of the maintenance plan with the power station operation plan and spare part inventory situation for comprehensive evaluation to obtain the maintenance decision-making suggestion for the component.

[0071] Specifically, by performing a failure risk assessment and economic analysis on the remaining life prediction value of the photovoltaic module, scientific maintenance decision-making suggestions are generated. Specifically, first, a risk matrix analysis is performed on the remaining life prediction value according to the failure probability and impact degree. The risk matrix classifies potential failure events according to their occurrence probability and the severity of the impact, so as to determine the risk level of the component. The failure probability represents the probability that the photovoltaic module may fail within a certain period of time, and the impact degree measures the impact of the component failure on the overall performance and economic benefits of the photovoltaic system. Through the risk matrix, the components can be divided into different risk levels, such as "low risk", "medium risk" and "high risk", etc., providing a systematic risk identification tool. After obtaining the risk level of the component, maintenance plans are designed according to different risk levels, and the costs and benefits of these plans are calculated to obtain the economic evaluation result of the maintenance plan. Specifically, for components with a high risk level, the maintenance plan may include early replacement or repair to avoid high losses caused by sudden failures; while for components with a low risk level, a way to extend the service life may be adopted to reduce unnecessary maintenance costs. When calculating the economy of the maintenance plan, the balance between the maintenance cost and the expected benefit will be considered. The maintenance cost includes expenses such as required labor, spare parts and downtime, while the benefit refers to the economic benefit brought by extending the component life and improving the power generation efficiency through maintenance. Through cost-benefit analysis, the economic effects of each maintenance plan can be clearly seen, providing data support for formulating the optimal maintenance strategy.

[0072] Next, comprehensively evaluate the economic assessment results of the maintenance plan in combination with the power plant's operation plan and spare part inventory situation to form final maintenance decision suggestions. The operation plan is the production and operation and maintenance arrangement of the power plant in a certain future period, and the spare part inventory refers to the quantity and type of spare parts currently available for replacement in the power plant. By comprehensively considering the power plant's operation plan, frequent maintenance during peak power periods can be avoided to ensure stable power generation; and based on the spare part inventory situation, the currently implementable maintenance plan can be determined to avoid delays in maintenance due to insufficient spare parts. The final maintenance decision suggestions not only consider the risk level and economy, but also combine the actual operation situation of the power plant, making the maintenance strategy more reasonable and efficient.

[0073] For example, assume that the predicted remaining life of a component is 6 months. Through risk matrix analysis, the failure probability of this component is high and the impact level is medium, so its risk level is "medium risk". For this risk level, two maintenance plans are designed: Plan A is to replace the component in advance, and Plan B is to conduct regular inspections and monitor the status. The economic assessment results show that the cost of Plan A is 1000 yuan and the expected revenue is 1500 yuan; the cost of Plan B is 300 yuan and the expected revenue is 600 yuan. After combining with the power plant operation plan, it is found that this component is located in the high-light area and is suitable for replacement to ensure continuous power generation, and the spare part inventory also meets the replacement requirements. Therefore, the final maintenance decision suggestion is to choose Plan A, that is, to replace the component in advance to ensure the stable operation and efficient power generation of the power plant.

[0074] The above describes the photovoltaic component life prediction method in the embodiments of the present application. Next, the photovoltaic component life prediction device in the embodiments of the present application will be described. Please refer to Figure 2 , an embodiment of the photovoltaic component life prediction device in the embodiments of the present application includes:

[0075] An acquisition module 201, configured to perform multi-channel synchronous acquisition on the voltage, current, temperature, and environmental parameters of the photovoltaic component to obtain a component operation data set with a unified time stamp;

[0076] A modeling module 202, configured to perform characteristic curve modeling and normalization processing on the operation data set to obtain the actual working characteristic curve and performance parameters of the component;

[0077] A classification module 203, configured to construct feature vectors and perform pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library containing multiple degradation types;

[0078] A matching module 204, configured to perform pattern matching on the real-time operation characteristics of the component and the feature library through the dynamic time warping algorithm to obtain a discrimination result of the degradation state of the component;

[0079] A modeling module 205, configured to analyze and model the deterioration state discrimination result and historical data through a deep learning algorithm to obtain a predicted remaining life value of the component;

[0080] An analysis module 206, configured to perform risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value to obtain maintenance decision suggestions for the component.

[0081] Through the collaborative cooperation of the above-mentioned various components, the voltage, current, temperature and environmental parameters of the photovoltaic component are synchronously collected through multiple channels to obtain a component operation data set with a unified time stamp, significantly improving the accuracy and consistency of data collection and laying a reliable data foundation for life prediction. By performing characteristic curve modeling and normalization processing on the collected data, the solution obtains the actual working characteristic curve and performance parameters of the component. This processing step effectively reduces the influence of environmental factors on the measurement results, enabling the performance parameters of the component under standardized conditions to more truly reflect the actual working state of the component. In addition, by constructing feature vectors and pattern classification for the actual working characteristic curve and performance parameters, a feature library containing various deterioration types is established, providing a rich reference basis for the system to discriminate and distinguish different deterioration types and ensuring the accuracy and reliability of the deterioration discrimination result. In the process of identifying the deterioration state, this solution introduces the dynamic time warping algorithm to perform pattern matching between the real-time operation characteristics of the component and the feature library, and to judge the deterioration state of the component in real time. Through the dynamic matching of the operation characteristics of the component by this algorithm, the system can more accurately identify the deterioration pattern, overcoming the limitation of the traditional static analysis method in dealing with diverse deterioration characteristics. At the same time, the solution analyzes and models the deterioration state discrimination result and historical data through a deep learning algorithm to construct a life prediction model of the photovoltaic component. Through this model, the system can dynamically predict the remaining life of the component, provide early warnings for the operation and maintenance team, facilitate the adoption of timely maintenance measures, prevent sudden failures of the component, and ensure the long-term stable operation of the photovoltaic system.

[0082] In addition, based on life prediction, this solution further combines risk assessment and cost-benefit analysis, performs risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value, and generates maintenance decision suggestions for the component. This function effectively balances the maintenance cost and system reliability. By comprehensively evaluating factors such as the deterioration degree, risk level, and operation and maintenance cost of the component, it provides scientific decision support for the operation and maintenance of the photovoltaic system, extends the service life of the component and reduces the operation cost. Generally speaking, through the integrated application of multi-channel data collection, dynamic time warping algorithm and deep learning model, this solution realizes the high-precision, real-time and intelligent life prediction of photovoltaic components, not only improves the accuracy of deterioration state discrimination, but also optimizes the economy and effectiveness of the maintenance strategy, providing an effective guarantee for the long-term stable operation of the photovoltaic system.

[0083] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting the lifespan of a photovoltaic module, characterized in that, The photovoltaic module life prediction method includes: Performing multi-channel synchronous acquisition on the voltage, current, temperature, and environmental parameters of the photovoltaic module to obtain a component operation dataset with a unified timestamp; Performing characteristic curve modeling and normalization processing on the operation dataset to obtain the actual working characteristic curve and performance parameters of the component; Constructing feature vectors and performing pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library containing multiple degradation types; Performing pattern matching on the real-time operation characteristics of the component and the feature library through the dynamic time warping algorithm to obtain the degradation state discrimination result of the component; Performing analysis and modeling on the degradation state discrimination result and historical data through a deep learning algorithm to obtain the predicted remaining life value of the component; Performing risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value to obtain maintenance decision-making suggestions for the component.

2. The photovoltaic module life prediction method according to claim 1, wherein The performing multi-channel synchronous acquisition on the voltage, current, temperature, and environmental parameters of the photovoltaic module to obtain a component operation dataset with a unified timestamp includes: Performing high-speed sampling on the DC bus of the photovoltaic module to obtain the original voltage sampling data, and performing 16-bit AD conversion on the original voltage sampling data to obtain digitized voltage data; Filtering out 50Hz power frequency interference and compensating for zero drift on the digitized voltage data to obtain calibrated voltage data, and performing time synchronization processing on the calibrated voltage data according to the IEEE1588 protocol to obtain timestamped voltage data; Collecting the series current through a Hall current sensor to obtain the original current data, and performing effective value calculation and temperature compensation on the original current data to obtain calibrated current data; Performing scanning acquisition on the surface temperature of the component through a PT100 temperature sensor to obtain temperature distribution data, and performing spatial interpolation and outlier removal on the temperature distribution data to obtain the component temperature field data; Performing time series alignment and data packaging on the timestamped voltage data, calibrated current data, and component temperature field data to obtain a synchronous acquisition data packet, and transmitting the synchronous acquisition data packet to the main control unit through the POWERBUS bus to obtain a component operation dataset with a unified timestamp.

3. The photovoltaic module life prediction method according to claim 1, characterized in that The performing characteristic curve modeling and normalization processing on the operation dataset to obtain the actual working characteristic curve and performance parameters of the component includes: Fitting the voltage and current data in the operation dataset according to the single diode equivalent circuit equation to obtain the reference I-V characteristic curve of the component, and extracting the key point parameters in the reference I-V characteristic curve to obtain the short-circuit current, open-circuit voltage, maximum power point voltage, and current value; Performing temperature correction on the key point parameters according to the temperature coefficient to obtain the characteristic parameters at the standard temperature, and performing normalization processing on the characteristic parameters at the standard temperature according to the irradiance ratio to obtain the performance parameters under the standard test conditions. Calculate the series and parallel resistances of the performance parameters under the standard test conditions to obtain the equivalent circuit parameters of the component, and calculate the fill factor and power attenuation rate of the component through the equivalent circuit parameters to obtain the actual working characteristic curve and performance parameters of the component.

4. The photovoltaic module life prediction method according to claim 1, wherein Construct eigenvectors and perform pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library containing multiple degradation types, including: Classify the actual working characteristic curve according to degradation types such as PID effect, hot spot effect, component crack, and encapsulation material aging to obtain a degradation mode feature matrix, and set threshold values for feature parameters such as voltage anomaly, temperature anomaly, power attenuation, and resistance change for the degradation mode feature matrix to obtain a degradation discrimination criterion; Construct eigenvectors for the degradation discrimination criterion to obtain a mathematical description of the degradation mode, and calculate the parameter weights according to the correlation of the mathematical description of the degradation mode to obtain a feature library containing multiple degradation types.

5. The method for predicting the service life of a photovoltaic module according to claim 1, characterized in that, Match the real-time operation characteristics of the component with the feature library through the dynamic time warping algorithm to obtain the degradation state discrimination result of the component, including: Perform data standardization processing on the real-time operation characteristics of the component to obtain a standardized eigenvector, and calculate the similarity between the standardized eigenvector and the degradation mode in the feature library through the dynamic time warping algorithm to obtain a feature matching degree; Evaluate the feature matching degree according to the fuzzy membership function to obtain the confidence level of the degradation mode, and perform multi-level discrimination on the confidence level through the set discrimination threshold to obtain the degradation state discrimination result of the component.

6. The method for predicting the service life of a photovoltaic module according to claim 1, characterized in that Analyze and model the degradation state discrimination result and historical data through a deep learning algorithm to obtain the predicted remaining life value of the component, including: Extract the time series characteristics of the degradation state discrimination result and historical data to obtain a health index data sequence, and construct a recurrent neural network containing three LSTM layers based on the health index data sequence, with the number of neurons in each layer being 128, 64, and 32 respectively, to obtain a deep learning model; Predict the future change trend of the health index through the deep learning model to obtain a health index prediction sequence, and perform cross-analysis on the health index prediction sequence and the preset failure threshold to obtain the predicted remaining life value of the component.

7. The photovoltaic module life prediction method according to claim 1, wherein Evaluate the risk level and analyze the maintenance cost-benefit of the predicted remaining life value to obtain maintenance decision-making suggestions for the component, including: Perform a risk matrix analysis on the predicted remaining life value according to the failure probability and impact degree to obtain the component risk level, and calculate the cost and benefit of the maintenance plan according to the component risk level to obtain the economic evaluation result of the maintenance plan; Comprehensively evaluate the economic evaluation result of the maintenance plan in combination with the power station operation plan and spare parts inventory situation to obtain maintenance decision-making suggestions for the component.

8. A photovoltaic module life prediction device for implementing the photovoltaic module life prediction method according to any one of claims 1-7, characterized in that, The photovoltaic component life prediction device includes: An acquisition module for performing multi-channel synchronous acquisition of the voltage, current, temperature, and environmental parameters of the photovoltaic component to obtain a component operation data set with a unified time stamp; A modeling module, which is used to perform characteristic curve modeling and normalization processing on the operation data set to obtain the actual working characteristic curve and performance parameters of the component; A classification module, which is used to construct feature vectors and perform pattern classification on the actual working characteristic curve and performance parameters to obtain a feature library containing multiple deterioration types; A matching module, which is used to perform pattern matching on the real-time operation characteristics of the component and the feature library through the dynamic time warping algorithm to obtain the discrimination result of the deterioration state of the component; A modeling module, which is used to analyze and model the discrimination result of the deterioration state and historical data through a deep learning algorithm to obtain the predicted remaining life value of the component; An analysis module, which is used to perform risk level assessment and maintenance cost-benefit analysis on the predicted remaining life value to obtain maintenance decision-making suggestions for the component.

Images