Photovoltaic module early fault early warning method based on electric-thermal characteristic conjoint analysis

Abstract

The invention discloses a photovoltaic module early-stage fault early-warning method based on electric-thermal characteristic joint analysis, and belongs to the field of photovoltaic module early-stage fault early-warning, and the method comprises the steps: segmenting an irregular high-temperature or low-temperature region according to a temperature distribution image through employing a data segmentation algorithm, and extracting the boundary contour and area quantification characteristics of the abnormal region; performing joint feature vector construction by using the abnormal region boundary contour and area quantization feature and the current and voltage fluctuation amplitude in the electrical signal time sequence to obtain the comprehensive fault diagnosis vector; according to the preliminary fault classification label, extracting a temperature electrical corresponding association mode of a similar fault case from a historical fault diagnosis database, and obtaining an enhanced fault corresponding relation model; and generating a system risk assessment report by adopting the final fault diagnosis classification result, triggering a corresponding alarm mechanism, and judging the overall risk level of the renewable energy system.

Description

Technical Field

[0001] This invention belongs to the field of early fault warning technology for photovoltaic modules, and particularly relates to an early fault warning method for photovoltaic modules based on joint analysis of electro-thermal characteristics. Background Technology

[0002] Monitoring the operational status of photovoltaic (PV) modules is a crucial research area in the new energy sector. Its core function is to ensure the stability and efficiency of power generation systems, directly impacting energy utilization efficiency and economic benefits. With the widespread application of clean energy, the ability to promptly detect and address potential PV module failures through technological means has become a key requirement for the industry's development. Research in this field not only affects the operational safety of individual power plants but also has profound implications for the stability of the entire power grid.

[0003] However, many current monitoring methods often struggle to comprehensively capture the root causes of problems when faced with complex fault scenarios. Especially when analyzing component operating status, existing methods are often limited to single-dimensional observation, lacking the ability to comprehensively judge multiple characteristics, resulting in inaccurate early fault identification and classification. This limitation makes it difficult to effectively detect some hidden problems before they develop into serious issues, increasing system risks and maintenance costs.

[0004] Against this backdrop, the main technical challenges of the research lie in accurately identifying temperature anomaly regions on the component surface and correlating them with changes in electrical performance. First, the distribution of temperature anomaly regions is diverse and irregular; these regions may exhibit low or high temperatures and vary in shape and area, posing a challenge to automated and accurate identification. Second, such temperature anomalies are often closely related to fluctuations in electrical performance, such as abnormal changes in current or voltage. However, the correlation between the two is complex and not easily determined directly. Temperature anomalies may be caused by various faults, and changes in electrical performance may also be affected by external environmental interference, leading to biases in fault type identification.

[0005] Therefore, accurately extracting abnormal temperature areas in practical operations and effectively matching them with abnormal electrical performance characteristics has become a critical issue that urgently needs to be addressed. For example, in the routine monitoring of a photovoltaic power station, a localized high temperature may appear on the surface of a component. However, the temperature distribution alone cannot determine whether it is caused by shading or an internal short circuit. If a comprehensive analysis is not conducted in conjunction with changes in current and voltage, the optimal repair time may be missed, or even greater safety hazards may be caused. This problem directly affects the efficiency of early warning and handling of faults, becoming a key issue in improving the technical level of photovoltaic module operation monitoring. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an early fault warning method for photovoltaic modules based on joint analysis of electro-thermal characteristics, thereby resolving the issues existing in the prior art.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a method for early fault warning of photovoltaic modules based on joint analysis of electro-thermal characteristics, comprising: The surface temperature distribution data of renewable energy components is collected by thermal imaging equipment, and electrical performance data is collected simultaneously by electrical performance sensors to obtain the temperature distribution image and the electrical signal time series. Based on the temperature distribution image, an irregular high-temperature or low-temperature region is segmented using a data segmentation algorithm, and the boundary contour and area quantization features of the abnormal region are extracted. The comprehensive fault diagnosis vector is obtained by constructing a joint feature vector using the boundary contour and area quantization features of the abnormal region and the current and voltage fluctuation amplitude in the time series of the electrical signal. The comprehensive fault diagnosis vector is input into the machine learning classification model for model training and pattern matching. If the temperature anomaly area feature in the comprehensive fault diagnosis vector exceeds the threshold and the current fluctuation amplitude feature is greater than the threshold, it is judged to be an internal short circuit fault type, and the preliminary fault classification label is obtained. Based on the preliminary fault classification labels, temperature and electrical correspondence patterns of similar fault cases are extracted from the historical fault diagnosis database to obtain the enhanced fault correspondence model; The enhanced fault correspondence model is used to perform secondary pattern verification on the current comprehensive fault diagnosis vector. If the verification result shows that the corresponding correlation deviation is less than the threshold, the fault type is confirmed and the final fault diagnosis classification result is obtained. The final fault diagnosis classification results are used to generate a system risk assessment report and trigger corresponding alarm mechanisms to determine the overall risk level of the renewable energy system.

[0008] In a second aspect, the present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0009] Thirdly, the present invention also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0010] Compared with the prior art, the present invention has the following advantages and technical effects: This invention addresses the problem of "difficulty in accurately diagnosing internal short-circuit faults caused by abnormal surface temperatures and electrical performance fluctuations in renewable energy systems, leading to overall system risk and efficiency loss." It simultaneously acquires temperature distribution images and electrical signal time series using thermal imaging equipment and electrical performance sensors. A data segmentation algorithm is employed to extract the boundary contours and area quantification features of abnormal regions, which are then combined with current and voltage fluctuation amplitudes to construct a comprehensive fault diagnosis vector. This vector is input into a machine learning classification model for preliminary fault classification. If the abnormal temperature area and current fluctuation amplitude exceed thresholds, it is determined to be an internal short-circuit fault. Subsequently, similar cases are extracted from a historical database to construct an enhanced fault correspondence model for secondary pattern verification. If the deviation is less than a threshold, the fault type is confirmed. Finally, a risk assessment report is generated, and an alarm mechanism is triggered.

[0011] This invention enables the fusion analysis of temperature and electrical data, improves the accuracy and real-time performance of fault diagnosis, effectively reduces the risk level of renewable energy systems, and ensures the safe and stable operation of components. Attached Figure Description

[0012] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of an early fault warning method for photovoltaic modules based on joint analysis of electro-thermal characteristics, according to an embodiment of the present invention. Detailed Implementation

[0013] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0014] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0015] Example 1 like Figure 1 As shown, this embodiment provides a method for early fault warning of photovoltaic modules based on joint analysis of electro-thermal characteristics, including: Step S101: Collect surface temperature distribution data of renewable energy components using thermal imaging equipment, and simultaneously collect electrical performance data using electrical performance sensors to obtain the temperature distribution image and the electrical signal time series.

[0016] A correlation analysis process for temperature distribution images and electrical signal sequences is constructed using data synchronously acquired by thermal imaging equipment and electrical performance sensors. Initial temperature distribution images and signal time series are obtained, and their correspondence in the time dimension is determined. Based on the initial temperature distribution image, image processing techniques are used to denoise and enhance edges, resulting in a processed temperature distribution image. Temperature distribution features of key regions are extracted from the processed image. The relationship between the temperature distribution features of key regions and the fluctuations of the signal time series is analyzed using the processed temperature distribution image. If abnormal fluctuations occur in the signal time series within a certain time period, the temperature distribution features of the corresponding time period are used to determine if there are potential abnormalities in equipment operation. The fluctuation features of the signal time series are obtained, and a support vector machine algorithm is used to classify the fluctuation features to determine whether the fluctuations are within the normal operating range. If the classification result shows that the fluctuations exceed the normal range, the time period is marked as an abnormal interval. For the marked abnormal intervals, the temperature distribution features of the corresponding time period are extracted from the processed temperature distribution image, and the correlation between the temperature distribution features and the fluctuation features is analyzed to obtain a comprehensive feature description of the abnormal interval. Based on the comprehensive feature description, detailed feature records of the abnormal intervals are generated. Combined with historical data on equipment operating status, it is determined whether the abnormal intervals are related to the long-term trend of component surface data, thus identifying the specific category of the abnormality. Through the specific categories of abnormalities, a feature mapping relationship for the abnormal intervals is constructed. A preset threshold is used to classify the abnormal categories, yielding the final abnormality classification result.

[0017] For example, when simultaneously acquiring data using thermal imaging equipment and electrical performance sensors, we can consider a photovoltaic module within a photovoltaic power station as the research object. The thermal imaging equipment generates a temperature distribution image every 5 seconds, while the electrical performance sensors simultaneously record current and voltage signals, forming time-series data. First, it is necessary to ensure time alignment between the two, for example, by matching timestamps to associate the temperature image at a certain moment with the current fluctuation data at the corresponding moment. In the image processing stage, for the initial temperature distribution image, median filtering can be used to reduce noise in the temperature data caused by environmental interference. Then, edge enhancement is performed using the Laplacian operator to highlight temperature change areas in key components of the photovoltaic module, such as joints. After processing, suppose that the temperature in a certain joint area is found to be significantly higher than the surrounding area, reaching 85 degrees Celsius, while the normal value should be around 50 degrees Celsius. This feature is extracted as a temperature anomaly point in the key area. For the signal time series, suppose that within a certain 10-minute time period, the current value suddenly jumps from a stable 100 amperes to 150 amperes, with the fluctuation amplitude exceeding the normal range. Using a support vector machine algorithm, the fluctuation feature is classified as an anomaly, and this time period is marked as the anomaly interval. Further analysis of the temperature images for the corresponding time period revealed a synchronous increase in temperature in the joint area, characterized by "abnormal current fluctuations accompanied by localized high temperatures." In classifying the anomaly, based on historical data, assuming a slow upward trend in the joint area over the past six months, accompanied by repeated current fluctuations, the anomaly can be inferred to be "joint aging leading to poor contact." Using a preset threshold, such as a temperature exceeding 80 degrees Celsius and current fluctuations exceeding 30% constituting a severe anomaly, this anomaly was classified as "high-risk."

[0018] Specifically, the construction of feature mapping relationships can associate temperature values, current fluctuation amplitudes, and anomaly categories to form a data table, facilitating rapid subsequent queries and early warnings. This process not only accurately locates anomalies but also guides maintenance priorities through tiered results, significantly improving equipment operational reliability and reducing the risk of sudden failures.

[0019] For example, from another perspective, if the temperature does not rise significantly within the abnormal range, but only the current fluctuates noticeably, analysis based on historical data suggests that the false alarm may be caused by sensor drift. In this case, it is classified as "low-risk," requiring only sensor calibration rather than immediate equipment repair. This multi-dimensional analysis ensures comprehensive judgment and avoids wasting resources.

[0020] Understandably, the above method, through the coordinated analysis of thermal imaging and electrical signals, forms a complete closed loop from data acquisition to anomaly classification, taking into account both real-time performance and accuracy, and providing strong technical support for equipment condition monitoring.

[0021] Step S102: Based on the temperature distribution image, use a data segmentation algorithm to segment irregular high-temperature or low-temperature regions, and extract the boundary contours and area quantization features of the abnormal regions.

[0022] Step 1: For the temperature distribution image, high-temperature and low-temperature regions are separated using data segmentation technology to obtain the specific location information of irregular regions, resulting in a preliminary set of anomalous regions. Step 2: From the preliminary set of anomalous regions, boundary contour data of each anomalous region is extracted using a boundary tracing method to determine the closed path information of the boundary contour. Step 3: Based on the closed path information of the boundary contour, the area quantization value of each anomalous region is calculated to obtain the area quantization distribution data. Step 4: For the area quantization distribution data, if the area quantization value of a certain anomalous region exceeds a preset threshold range, it is marked as a key focus object, resulting in a list of key focus anomalous regions. Step 5: From the list of key focus anomalous regions, corresponding temperature distribution features are extracted, and these features are classified using a support vector machine algorithm to determine the category of the anomalous region. Step 6: Based on the category of the classified anomalous regions and combined with distribution analysis technology, the positional proportion of the anomalous regions in the overall temperature distribution image is obtained, determining their priority ranking in the global distribution. Step 7: Based on the priority ranking results, corresponding region extraction schemes are generated for high-priority anomalous regions to obtain a detailed anomalous region distribution view.

[0023] Specifically, when processing temperature distribution images of renewable energy components, the system first processes the image using a data segmentation algorithm to identify irregular high-temperature or low-temperature regions. Assuming the input temperature distribution image has a resolution of 1024x768 pixels and the temperature range is set from 10℃ to 90℃, the system uses a threshold-based segmentation algorithm to mark areas with temperatures exceeding 70℃ as high-temperature areas and areas below 15℃ as low-temperature areas. The thresholds are determined based on historical data statistical analysis. Subsequently, the system uses the Canny edge detection algorithm to extract the boundary contours of abnormal regions, setting a low threshold of 50 and a high threshold of 150 to ensure boundary continuity. The generated boundary contours are stored in pixel coordinates. Assuming a high-temperature region boundary contains approximately 300 pixels, forming a closed curve, the system then quantifies the area of ​​the abnormal region by calculating the number of pixels within the boundary. Assuming the high-temperature region area is 12,000 pixels, combined with the image scale (each pixel corresponds to 0.01 square meters), this translates to an actual area of ​​approximately 0.12 square meters. Simultaneously, the system calculates the geometric features of the region, such as aspect ratio (assumed to be 2.5) and roundness (assumed to be 0.7), to further describe the region's morphology. After the above data processing is completed, the system stores the extracted boundary contours and area features in the database and automatically associates them with component location information (such as number A-001) to form a data traceability chain, providing a basis for subsequent anomaly area location and maintenance strategy formulation. The entire process is automatically completed by the image processing module, with the algorithm running time controlled within 2 seconds to ensure real-time performance. The generated feature data is transmitted to the analysis module through an internal interface for further anomaly pattern recognition.

[0024] Step S103: The comprehensive fault diagnosis vector is obtained by constructing a joint feature vector using the boundary contour and area quantization features of the abnormal region and the current and voltage fluctuation amplitude in the electrical signal time series.

[0025] Step 1: Extract geometric morphology information from the boundary contour and area quantification data of the abnormal region. Use data extraction techniques to separate the boundary point set and area values ​​to obtain a preliminary geometric feature set. Step 2: Based on the preliminary geometric feature set, combine it with time-series electrical signal data to obtain current fluctuation and voltage amplitude information. Use data alignment techniques to match the geometric features with the signal features to determine the joint feature matrix. Step 3: For the joint feature matrix, use a support vector machine algorithm to classify the multidimensional data in the matrix, determine the fault category corresponding to each group of features, and obtain the classified fault feature groups. Step 4: Based on the classified fault feature groups, extract the current fluctuation and voltage amplitude data from each group. If the fluctuation amplitude of a certain group exceeds a preset threshold range, mark it as a high-risk category to determine the high-risk fault set. Step 5: For the high-risk fault set, obtain the corresponding abnormal region boundary contour information. Use regional positioning techniques to analyze its distribution location in the overall system to obtain a distribution location dataset. Step 6: Based on the distribution location dataset and the signal analysis results in the time series, if the signal fluctuation frequency of a certain area is higher than a preset threshold, its priority is increased, and a list of priority fault areas is determined. Step 7: For the priority fault area list, data integration technology is used to generate a corresponding fault area distribution view. View mapping technology is then used to associate the distribution view with the electrical signal fluctuation data to obtain comprehensive fault distribution information.

[0026] Specifically, in the fault diagnosis process of renewable energy equipment, the system first extracts the boundary contour data and area quantization features of the abnormal area from the database. Assuming the boundary of an abnormal area of ​​a certain device consists of approximately 400 pixels, the area quantization value is 15,000 pixels, corresponding to an actual area of ​​0.15 square meters. Simultaneously, the geometric parameters of this area are obtained, such as an eccentricity value of 0.8. Subsequently, the system synchronously collects electrical signal time-series data. Assuming a collection period of 1 hour, the current fluctuation range is obtained as 2.5 amperes to 3.8 amperes, and the voltage fluctuation range is 210 volts to 235 volts. The system performs frequency domain analysis on the signals using a Fourier transform algorithm, extracting the dominant frequency component and calculating the fluctuation frequencies of current and voltage as 5 Hz and 3 Hz, respectively, as signal features. Next, the system normalizes the number of boundary contour points, area value, eccentricity, and the current and voltage fluctuation amplitudes and frequencies. Principal component analysis is used for dimensionality reduction, generating a 6-dimensional feature vector. Assuming the normalized area feature value is 0.65 and the current fluctuation feature value is 0.72. Finally, the system uses a weighted fusion algorithm to construct a comprehensive fault diagnosis vector, setting the area feature weight to 0.3 and the electrical signal feature weight to 0.7, generating the final diagnostic vector value, for example [0.45, 0.68, 0.33, 0.51, 0.29, 0.62]. This value is then stored in the fault analysis database and automatically associated with the equipment identifier to form a data link, providing support for subsequent fault mode matching. The entire process is automatically executed by the data processing module, with computation time controlled within 1.5 seconds. The vector data is then transmitted to the diagnostic model via an internal interface to complete feature integration.

[0027] Step S104: Input the comprehensive fault diagnosis vector into the machine learning classification model for model training and pattern matching. If the temperature anomaly area feature in the comprehensive fault diagnosis vector exceeds the threshold and the current fluctuation amplitude feature is greater than the threshold, it is judged to be an internal short circuit fault type, and the preliminary fault classification label is obtained.

[0028] A fault diagnosis vector is constructed using raw data acquired from equipment sensors, incorporating key features such as the area of ​​temperature anomalies and the amplitude of current fluctuations, resulting in an initial dataset. Based on this initial dataset, a pre-established machine learning classification model is used for pattern matching, and a support vector machine algorithm is employed to classify the features, determining a preliminary fault classification label. If the preliminary fault classification label indicates an internal short-circuit fault, a secondary comparison is performed on the area of ​​temperature anomalies and the amplitude of current fluctuations against preset threshold standards, yielding comparison result data. Using this comparison result data, specific deviation values ​​for the area of ​​temperature anomalies and the amplitude of current fluctuations are obtained, and it is determined whether they exceed the preset threshold standards, confirming the deviation. Based on the deviation confirmation results, if all deviation values ​​exceed the preset threshold standards, the fault category is confirmed as an internal short-circuit fault, combined with fault type determination logic, resulting in a final fault type label. Using the final fault type label, a comprehensive analysis result is generated, recording the specific feature data of the internal short-circuit fault, and outputting the fault diagnosis conclusion data.

[0029] Specifically, in the process of processing and classifying the comprehensive fault diagnosis vector, the system first collects equipment operating data, such as temperature and current data, through sensors. Assuming the collected temperature anomaly area feature value is 25.5 square centimeters, a threshold of 20.0 square centimeters is set, while the current fluctuation amplitude feature value is 8.3 amperes, a threshold of 5.0 amperes is set. Next, this data is used to construct a comprehensive fault diagnosis vector in the form [25.5, 8.3], and input into a pre-trained machine learning classification model. This model uses the Support Vector Machine (SVM) algorithm, which maps the data to a high-dimensional space through a kernel function to achieve non-linear classification. The training dataset contains 1000 sets of historical fault data, with internal short-circuit faults accounting for 30%. Cross-validation is used to optimize parameters during model training to ensure a classification accuracy of over 92%. Subsequently, the model performs pattern matching on the input vector and, combined with feature value analysis, finds that the temperature anomaly area feature of 25.5 exceeds the threshold of 20.0, and the current fluctuation amplitude feature of 8.3 is greater than the threshold of 5.0, meeting the criteria for internal short-circuit faults. Therefore, the initial fault classification label is output as "internal short circuit". To further verify the results, the system automatically invoked the historical data comparison module to analyze the fault distribution of similar feature vectors over the past 24 hours. It found that 90% of similar cases were confirmed as internal short circuits, enhancing the credibility of the classification labels. Simultaneously, the system correlated the results with equipment operating environment data (such as humidity and voltage stability). If humidity exceeds 80%, it may exacerbate the short circuit risk, requiring the generation of alarm information and its push to the maintenance system. This forms a complete logical chain from data acquisition to fault determination, ensuring the rigor and reliability of the diagnostic process.

[0030] Step S105: Extract the temperature and electrical correspondence patterns of similar fault cases from the historical fault diagnosis database based on the preliminary fault classification labels to obtain the enhanced fault correspondence model.

[0031] Step 1: Filter historical fault records from the diagnostic database to obtain a set of records matching the fault labels related to the preliminary classification, resulting in a preliminary fault dataset. Step 2: Analyze the distribution of similar cases within the preliminary fault dataset and use a comparison method to determine the subset of cases closest to the current fault label. Step 3: Extract raw data of temperature and electrical parameters from the case subset, remove outliers through data cleaning, and obtain a standardized parameter dataset. Step 4: Analyze the correlation patterns between temperature and electrical parameters in the standardized parameter dataset. If the correlation of a certain set of parameters exceeds a preset threshold, it is identified as a key correlation pattern. Step 5: Based on the key correlation patterns, construct the mapping relationship between faults, obtain feature combinations matching the current fault label, and determine the fault feature library. Step 6: By comparing the fault feature library with historical faults, use the support vector machine algorithm to classify the feature combinations, obtaining an enhanced fault correspondence model. Step 7: Based on the enhanced fault correspondence model, perform pattern matching on newly input fault data to determine its fault category and obtain the final classification result.

[0032] Specifically, in implementing the enhanced fault diagnosis model, the system first automatically extracts similar fault cases from the historical fault diagnosis database. Assuming the initial fault classification label is "overheating fault," the system uses a text matching algorithm based on cosine similarity to compare the label with 100,000 historical fault records stored in the database, filtering out cases with a similarity greater than 0.85, extracting 500 relevant records containing temperature and electrical parameter data. Next, the system mines the temperature-electrical correlation patterns in these cases, using principal component analysis (PCA) to reduce the dimensionality of temperature data (ranging from 50 to 150 degrees Celsius) and electrical parameters (such as voltage fluctuations ranging from 2.5 to 5.0 volts), extracting the main feature vectors, and calculating a correlation coefficient of 0.78 between temperature and voltage, indicating a strong correlation between the two. Subsequently, the system constructs an enhanced fault correspondence model, using a Support Vector Machine (SVM) algorithm to train the extracted association patterns. The kernel function is set to Radial Basis Function (RBF), the penalty parameter C is 10, and the training set to test set ratio is 8:2. The model's prediction accuracy reaches 92.3%. To ensure logical rigor, the system further incorporates business scenarios, inputting ambient humidity (range 30% to 80%) as an auxiliary variable into the model to analyze its impact on temperature-electrical correlations. The correlation coefficient between humidity and fault probability was found to be 0.45; after incorporating humidity into the model, the overall prediction accuracy improved to 93.5%. Through this process, the system automatically completes the entire process from data extraction to model optimization, forming an enhanced fault diagnosis model based on historical data, providing a reliable basis for subsequent fault prediction.

[0033] Step S106: Perform secondary pattern verification on the current comprehensive fault diagnosis vector using the enhanced fault correspondence model. If the verification result shows that the corresponding correlation deviation is less than the threshold, then confirm the fault type and obtain the final fault diagnosis classification result.

[0034] By initially collecting comprehensive vector data, standardized processing methods are used to clean and format the raw data, resulting in a standardized fault data set. Based on this standardized fault data set, a pre-established enhancement model is used for feature extraction to obtain key fault feature parameters. For these key fault feature parameters, a pattern analysis process is implemented to determine the degree of matching between the feature parameters and known fault types. If the matching degree is higher than a preset threshold, the preliminary fault type is determined. A secondary verification mechanism is used to calculate the correlation deviation between the preliminary fault types. If the calculated deviation value is lower than a preset threshold, the fault type is confirmed, and intermediate diagnostic results are obtained. Based on the intermediate diagnostic results, a comparison is performed with a corresponding relational database to determine if there are any potential hidden faults. If hidden faults exist, relevant features are extracted to obtain extended diagnostic information. For this extended diagnostic information, a support vector machine algorithm is used for classification confirmation to obtain the final fault diagnosis classification result. Through structured storage of the final fault diagnosis classification result, a traceable diagnostic process record is generated, establishing a complete fault analysis archive.

[0035] Specifically, in the process of performing secondary pattern verification and confirming the fault type of the current comprehensive fault diagnosis vector through the enhanced fault correspondence model, we first obtain a comprehensive fault diagnosis vector with values ​​[0.75, 0.32, 0.18], corresponding to the probability distributions of three possible fault modes A, B, and C, respectively. The system inputs this vector into the enhanced fault correspondence model, which constructs a mapping matrix between fault modes and feature vectors based on historical data. For example, matrix M is [[0.8, 0.3, 0.1], [0.2, 0.7, 0.2], [0.1, 0.2, 0.6]], representing the correlation strength between the three fault modes and the vector components. The system calculates the predicted vector P=M×[0.75, 0.32, 0.18]=[0.713, 0.314, 0.221] through matrix multiplication and compares it with the original vector to calculate the correlation deviation. The deviation was calculated using the Euclidean distance formula, yielding a deviation value D = sqrt((0.75-0.713)^2 + (0.32-0.314)^2 + (0.18-0.221)^2) = 0.057. The system's preset threshold is 0.1. The verification result shows that the deviation of 0.057 is less than the threshold of 0.1, thus confirming the fault type as Mode A (corresponding to a maximum probability value of 0.75). Subsequently, the system cross-validates this result with a historical fault database. The typical feature vector for Mode A in the database is [0.78, 0.30, 0.15]. The similarity between the current vector and the typical vector (using the cosine similarity algorithm) is calculated, yielding a result of 0.98, exceeding the similarity threshold of 0.95. This further confirms the fault type is correct, and the final fault diagnosis classification result is Mode A. The entire process is automated by the algorithm, logically forming a closed loop from vector input, model calculation, deviation verification to final classification, ensuring the accuracy of the diagnostic results.

[0036] Step S107: Use the final fault diagnosis classification results to generate a system risk assessment report and trigger the corresponding alarm mechanism to determine the overall risk level of the renewable energy system.

[0037] By obtaining classification results from fault diagnosis data and analyzing abnormal patterns, a preliminary system risk category is determined. Based on this preliminary risk category, a pre-established risk assessment model is used, combined with historical data, to determine the overall risk level. If the overall risk level exceeds a preset threshold, a corresponding alarm mechanism is triggered, generating detailed notification information containing the risk level and abnormal patterns. For this detailed notification, relevant energy system operating parameters are obtained, and the presence of persistent anomalies is analyzed to identify potential fault sources. Through the analysis of potential fault sources, a support vector machine algorithm is used to classify the operating parameters, obtaining the specific risk impact range. Based on the specific risk impact range, a system risk assessment report containing targeted recommendations is generated to determine the priority of subsequent processing. If the priority of subsequent processing is high, the operating mode of the renewable energy system is automatically adjusted, and the adjusted operating status data is obtained to determine system stability.

[0038] Specifically, in the risk assessment and alarm mechanism of renewable energy systems, a system risk assessment report is first generated based on the final fault diagnosis classification results. Assuming the system uses a machine learning model to classify photovoltaic module fault data into three fault categories: mechanical faults (40%), electrical faults (35%), and environmental impact faults (25%), a weighted risk assessment algorithm is used to combine the severity and probability of occurrence of each type of fault to calculate the risk value. The formula is: Risk Value = Severity × Probability of Occurrence × Weight. For mechanical faults, the severity is 8, the probability of occurrence is 0.4, and the weight is 0.5, resulting in a risk value of 8 × 0.4 × 0.5 = 1.6; for electrical faults, the severity is 7, the probability of occurrence is 0.35, and the weight is 0.4, resulting in a risk value of 7 × 0.35 × 0.4 = 0.98; and for environmental impact faults, the severity is 5, the probability of occurrence is 0.25, and the weight is 0.3, resulting in a risk value of 5 × 0.25 × 0.3 = 0.375. The total risk value is calculated as 1.6 + 0.98 + 0.375 = 2.955. The system sets a risk threshold of 3. If the total risk value is below the threshold, it is assessed as low risk; otherwise, it is considered high risk. In this case, it is low risk. The system then automatically generates a risk assessment report, including the calculated results and risk level. It also generates a risk distribution map using data visualization tools and stores it in a cloud database for subsequent analysis. Subsequently, the system triggers an alarm mechanism. Based on the risk value of 2.955 being below the threshold of 3, it automatically sends a low-risk notification to the monitoring center. It also interfaces with the internal communication system via API to send a briefing email containing the risk value and the percentage of failures, ensuring real-time information transmission. Finally, the system assesses the overall risk level. Combining historical data analysis, it assumes the average risk value over the past 30 days is 2.8, and the current value of 2.955 is slightly higher than the average. The system automatically marks this as a state requiring attention and inputs the data into the risk trend analysis model. Using a linear regression algorithm, it predicts that the risk value may reach 3.1 in the next 7 days. If this exceeds the threshold, the system generates an early warning suggestion and adjusts the equipment inspection frequency from once a week to once every three days. The relevant instructions are issued to the equipment management system through the IoT platform, forming a closed-loop control to ensure that the risk is controllable.

[0039] Example 2 This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0040] Example 3 This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0041] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for early fault warning of photovoltaic modules based on joint analysis of electro-thermal characteristics, characterized in that, Includes the following steps: The surface temperature distribution data of renewable energy components is collected by thermal imaging equipment, and electrical performance data is collected simultaneously by electrical performance sensors to obtain the temperature distribution image and the electrical signal time series. Based on the temperature distribution image, an irregular high-temperature or low-temperature region is segmented using a data segmentation algorithm, and the boundary contour and area quantization features of the abnormal region are extracted. The comprehensive fault diagnosis vector is obtained by constructing a joint feature vector using the boundary contour and area quantization features of the abnormal region and the current and voltage fluctuation amplitude in the time series of the electrical signal. The comprehensive fault diagnosis vector is input into the machine learning classification model for model training and pattern matching. If the temperature anomaly area feature in the comprehensive fault diagnosis vector exceeds the threshold and the current fluctuation amplitude feature is greater than the threshold, it is judged to be an internal short circuit fault type, and the preliminary fault classification label is obtained. Based on the preliminary fault classification labels, temperature and electrical correspondence patterns of similar fault cases are extracted from the historical fault diagnosis database to obtain the enhanced fault correspondence model; The enhanced fault correspondence model is used to perform secondary pattern verification on the current comprehensive fault diagnosis vector. If the verification result shows that the corresponding correlation deviation is less than the threshold, the fault type is confirmed and the final fault diagnosis classification result is obtained. The final fault diagnosis classification results are used to generate a system risk assessment report and trigger corresponding alarm mechanisms to determine the overall risk level of the renewable energy system.

2. The method according to claim 1, characterized in that, The process of acquiring surface temperature distribution data of renewable energy components using thermal imaging equipment, and simultaneously acquiring electrical performance data using electrical performance sensors, to obtain the temperature distribution image and the electrical signal time series, includes: A correlation analysis process for temperature distribution images and electrical signal sequences is constructed using data synchronously collected by thermal imaging equipment and electrical performance sensors. Acquire the initial temperature distribution image and signal time series, and determine the correspondence between the two in the time dimension; Based on the initial temperature distribution image, image processing techniques are used to denoise and enhance the edges of the image to obtain the processed temperature distribution image. Extract temperature distribution features from key areas of the processed image; By analyzing the temperature distribution characteristics of key areas and their relationship with the fluctuation of the signal time series through the processed temperature distribution image; If the signal time series shows abnormal fluctuations within a certain period of time, the temperature distribution characteristics of the corresponding period of time can be combined to determine whether there are potential abnormalities in the equipment's operating status. The fluctuation characteristics of the signal time series are obtained, and the support vector machine algorithm is used to classify the fluctuation characteristics to determine whether the fluctuation is within the normal operating range. If the classification results show that the fluctuation exceeds the normal range, then mark that time period as an abnormal interval; For the marked abnormal intervals, the temperature distribution features of the corresponding time period are extracted from the processed temperature distribution image, and the correlation between the temperature distribution features and the fluctuation features is analyzed to obtain a comprehensive feature description of the abnormal intervals. Based on the comprehensive feature description, detailed feature records of the abnormal interval are generated. Combined with historical data of equipment operating status, it is determined whether the abnormal interval is related to the long-term change trend of component surface data, and the specific category of the abnormality is determined. By constructing a feature mapping relationship for anomaly intervals based on the specific categories of anomalies, and using a preset threshold to classify the anomaly categories, the final anomaly classification result is obtained.

3. The method according to claim 1, characterized in that, The step of segmenting irregular high-temperature or low-temperature regions based on the temperature distribution image using a data segmentation algorithm, and extracting the boundary contours and area quantization features of the abnormal regions, includes: For the temperature distribution image, the high temperature region and the low temperature region are separated by data segmentation technology to obtain the specific location information of the irregular region and obtain a preliminary set of abnormal regions. From the initially defined set of abnormal regions, the boundary contour data of each abnormal region is extracted using the boundary tracing method to determine the closure path information of the boundary contour; Based on the closed path information of the boundary contour, calculate the area quantization value of each abnormal region and obtain the distribution data of the area quantization; For the area quantization distribution data, if the area quantization value of an abnormal region exceeds the preset threshold range, it is marked as a key focus object, and a list of key focus abnormal regions is obtained. From the list of key anomaly areas, extract the corresponding temperature distribution features, and use the support vector machine algorithm to classify these features and determine the category of the anomaly area. Based on the classification of abnormal regions, combined with distribution analysis techniques, the positional proportion of abnormal regions in the overall temperature distribution image is obtained, and their priority ranking in the global distribution is determined. Based on the priority sorting results, corresponding region extraction schemes are generated for high-priority abnormal regions, resulting in a detailed abnormal region distribution view.

4. The method according to claim 1, characterized in that, The comprehensive fault diagnosis vector is constructed by combining the boundary contour and area quantization features of the abnormal region with the current and voltage fluctuation amplitudes in the electrical signal time series to form a joint feature vector, including: Based on the boundary contour and area quantization data of the abnormal region, geometric morphology information is obtained, and data extraction techniques are used to separate the boundary point set and area value to obtain a preliminary set of geometric features. Based on the preliminary set of geometric features, combined with the time series data of electrical signals, the current fluctuation and voltage amplitude information are obtained. The geometric features are matched with the signal features through data alignment technology to determine the joint feature matrix. For the joint feature matrix, the support vector machine algorithm is used to classify the multidimensional data in the matrix, determine the fault category corresponding to each group of features, and obtain the classified fault feature groups. Based on the classified fault characteristics, the current fluctuation and voltage amplitude data of each group are extracted. If the fluctuation amplitude of a certain group exceeds the preset threshold range, it is marked as a high-risk category, and the high-risk fault set is determined. For a set of high-risk faults, the boundary contour information of the corresponding abnormal areas is obtained, and the distribution location of the faults in the overall system is analyzed by regional positioning technology to obtain a distribution location dataset. Based on the distribution location dataset and the signal analysis results in the time series, if the signal fluctuation frequency of a certain area is higher than the preset threshold, its priority is increased, and a list of fault areas to be processed first is determined. For the list of priority fault areas, data integration technology is used to generate a corresponding fault area distribution view. The distribution view is then associated with electrical signal fluctuation data through view mapping technology to obtain comprehensive fault distribution information.

5. The method according to claim 1, characterized in that, The comprehensive fault diagnosis vector is input into a machine learning classification model for model training and pattern matching. If the temperature anomaly area feature in the comprehensive fault diagnosis vector exceeds a threshold and the current fluctuation amplitude feature is greater than a threshold, it is determined to be an internal short-circuit fault type, and the preliminary fault classification label is obtained, including: By using raw data obtained from the device's sensors, a fault diagnosis vector is constructed, which includes key features such as the area of ​​temperature anomalies and the amplitude of current fluctuations, thus obtaining an initial dataset. Based on the initial dataset, a pre-established machine learning classification model is input for pattern matching, and the support vector machine algorithm is used to classify the features to determine the preliminary fault classification labels. If the initial fault classification label indicates an internal short circuit fault, a second comparison is made between the abnormal temperature area and the current fluctuation amplitude and the preset threshold standard to obtain the comparison result data. By comparing the results data, the specific deviation values ​​of the temperature anomaly area and current fluctuation amplitude are obtained, and it is determined whether they exceed the preset threshold standard to confirm the deviation. Based on the deviation confirmation results, if all deviation values ​​exceed the preset threshold standard, the fault category is confirmed as an internal short circuit fault by combining the fault type determination logic, and the final fault type label is obtained. The final fault type label generates a comprehensive analysis result, records the specific characteristic data of the internal short circuit fault, and outputs the fault diagnosis conclusion data.

6. The method according to claim 1, characterized in that, The step of extracting temperature and electrical correlation patterns of similar fault cases from the historical fault diagnosis database based on the preliminary fault classification labels to obtain the enhanced fault correspondence model includes: The historical fault records in the diagnostic database are filtered to obtain a set of records that match the fault labels related to the preliminary classification, thus obtaining the preliminary filtered fault dataset. For the initially screened fault dataset, the distribution of similar cases is analyzed, and a comparison method is used to determine the subset of cases that are closest to the current fault label; Raw data of temperature and electrical parameters were extracted from the case subset, and outliers were removed through data cleaning to obtain a standardized parameter dataset. For the standardized parameter dataset, the correlation pattern between temperature parameters and electrical parameters is analyzed. If the correlation of a certain set of parameters exceeds a preset threshold, it is judged as a key correlation pattern. Based on key association patterns, construct the mapping relationship corresponding to the fault, obtain the feature combination that matches the current fault label, and determine the fault feature library; By comparing the fault feature database with historical faults, the support vector machine algorithm is used to classify the feature combinations and obtain an enhanced fault correspondence model. Based on the enhanced fault correspondence model, pattern matching is performed on the newly input fault data to determine its fault category and obtain the final classification result.

7. The method according to claim 1, characterized in that, The enhanced fault correspondence model is used to perform secondary pattern verification on the current comprehensive fault diagnosis vector. If the verification result shows that the corresponding correlation deviation is less than a threshold, the fault type is confirmed, and the final fault diagnosis classification result is obtained, including: By initially collecting data from the comprehensive vector, the collected raw data is cleaned and formatted using standardized processing methods to obtain a standardized fault data set. Based on the standardized fault data set, a pre-established enhanced model is invoked to extract features and obtain key fault feature parameters; For key fault characteristic parameters, a pattern analysis process is implemented to determine the degree of matching between the characteristic parameters and known fault types. If the degree of matching is higher than a preset threshold, the preliminary fault type is determined. Through a secondary verification mechanism, the correlation deviation of the preliminary fault type is calculated. If the calculated deviation value is lower than the preset threshold, the fault type is confirmed and intermediate diagnostic results are obtained. Based on the intermediate diagnostic results, a comparison is made with the corresponding relational database to determine whether there are any potential hidden faults. If there are hidden faults, relevant features are extracted to obtain extended diagnostic information. For the extended diagnostic information, a support vector machine algorithm is used for classification and confirmation to obtain the final fault diagnosis classification result; By structurally storing the final fault diagnosis and classification results, a traceable diagnostic process record is generated, and a complete fault analysis archive is established.

8. The method according to claim 1, characterized in that, The process of generating a system risk assessment report based on the final fault diagnosis classification results and triggering a corresponding alarm mechanism to determine the overall risk level of the renewable energy system includes: By obtaining classification results from fault diagnosis data and analyzing the abnormal patterns within them, a preliminary system risk category can be determined. Based on the preliminary system risk categories, a pre-established risk assessment model is used, combined with historical data, to determine the overall risk level; If the overall risk level exceeds the preset threshold, the corresponding alarm mechanism will be triggered to generate detailed notification information including the risk level and abnormal pattern. Based on the generated detailed notification information, relevant energy system operating parameters are obtained, and it is analyzed whether there are any persistent anomalies to determine potential sources of failure. By analyzing potential sources of failure, the support vector machine algorithm is used to classify the operating parameters and obtain the specific risk impact range. Based on the specific scope of risk impact, generate a system risk assessment report containing targeted recommendations to determine the priority of subsequent handling; If the subsequent processing has a higher priority, the operating mode of the renewable energy system will be automatically adjusted, the adjusted operating status data will be obtained, and the system stability will be determined.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-8.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-8.

Images