Building photovoltaic integrated assembly defect intelligent detection method and system

Abstract

The invention discloses a building photovoltaic integrated assembly defect intelligent detection method and system, and relates to the technical field of photovoltaic assembly intelligent detection. According to the invention, by constructing a multi-source data fusion framework and a depth feature extraction technology, the accuracy and response speed of photovoltaic module defect detection are significantly improved, and the limitation of a traditional method is solved; further, through a multivariable regression analysis and time sequence analysis technology, a defect trend is accurately predicted, an operation and maintenance decision is optimized, and the defects in the aspects of defect cause analysis and predictive maintenance in the prior art are overcome; and finally, in combination with an intelligent operation and maintenance system and a real-time monitoring module, operation and maintenance resource allocation is optimized, the system reliability is improved, timely processing of a defect area is ensured, and the operation and maintenance efficiency and stability of the photovoltaic system are comprehensively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for photovoltaic modules, specifically to an intelligent inspection method and system for defects in building-integrated photovoltaic (BIPV) modules. Background Technology

[0002] Building-integrated photovoltaics (BIPV) is an important direction for the integration of new energy and the building industry. This field is not only related to the efficient use of energy, but also directly affects the safety and long-term stability of buildings. Its research and application have profound significance for socio-economic development and environmental protection. However, current traditional methods for detecting defects in photovoltaic modules generally have limitations. They mainly rely on a single data source and conventional algorithms, which are difficult to deal with diverse defect problems in complex environments. The detection accuracy and response speed cannot meet practical needs, especially in defect cause analysis and predictive maintenance.

[0003] Against this backdrop, the field faces multiple technical challenges. First, the detection of defects in photovoltaic modules requires processing large amounts of multi-source data, such as hyperspectral images and thermal imaging data. However, existing technologies are inefficient in feature extraction and real-time processing, which limits the timeliness of defect identification. Due to the bottleneck of data processing speed, the system struggles to accurately capture minute defects in a very short time. This further exacerbates the difficulty of defect cause analysis, making it impossible to delve into the root causes behind the defects and thus unable to provide a reliable basis for subsequent operation and maintenance decisions. This break in the chain from data processing to causal deduction to decision support has become a key obstacle to intelligent operation and maintenance. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent detection method and system for defects in building-integrated photovoltaic (BIPV) modules, which significantly improves the accuracy and response speed of photovoltaic module defect detection, optimizes operation and maintenance decisions and resource allocation, and enhances overall stability and reliability.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] This application provides an intelligent defect detection method for building-integrated photovoltaic (BIPV) modules, comprising the following steps:

[0007] Hyperspectral image data and thermal imaging data are preprocessed in a unified manner, and hierarchical sampling technology is used to reduce the dimensionality of the original data to obtain a fused dataset.

[0008] Based on the fused dataset, a convolutional neural network model is used to perform deep feature extraction on the data to obtain a feature map set containing potential defect patterns.

[0009] By using feature maps and a preset threshold, significant regions are initially screened to obtain a dataset of suspected defects.

[0010] A support vector machine classification model is used to perform fine classification of suspected defective regions, obtain a classification result set, and determine the regions that belong to real defects.

[0011] By combining historical operating data and environmental parameters, a defect cause analysis model is constructed. Multivariate regression analysis is used to mine the correlation between defect types and distribution information, and a cause analysis report is obtained.

[0012] By analyzing the causes, key influencing factors are extracted, and time series analysis is used to predict the development trend of defects, thus obtaining a prediction dataset.

[0013] Based on the predicted dataset, targeted operation and maintenance strategies are generated. Combined with the intelligent operation and maintenance system, risk areas are prioritized to obtain an operation and maintenance scheduling plan and determine the optimal path for resource allocation.

[0014] Furthermore, the fused dataset is obtained, specifically including:

[0015] Hyperspectral image data and thermal imaging data are acquired as raw multi-source data. The hyperspectral image data is filtered by a spectral band selection method. The information content of each band is determined based on the information entropy calculation results to obtain optimized hyperspectral data.

[0016] Thermal imaging data is preprocessed by temperature gradient analysis to calculate the temperature change rate of pixels and obtain key thermal imaging region data. Principal component analysis algorithm is used to reduce the dimensionality of the optimized hyperspectral data and key thermal imaging region data to obtain a set of dimensionality-reduced feature vectors.

[0017] The dimensionality-reduced feature vector set is geometrically corrected by spatial registration technology. Transformation parameters are calculated based on the coordinates of control points to obtain spatially aligned multi-source feature data. Then, a weighted fusion method is used to fuse the spatially aligned multi-source feature data to obtain a fused dataset.

[0018] Furthermore, a feature map set containing potential defect patterns is obtained, specifically including:

[0019] Acquire photovoltaic module image data from the fusion dataset, and use a convolutional neural network model to perform deep feature extraction on the standard image data to determine the initial feature mapping containing defect information;

[0020] A multi-scale analysis method is used to construct feature pyramids at different scales. Multi-scale feature representations are obtained through scale transformation and feature fusion operations, resulting in comprehensive feature data covering the features of minute defects.

[0021] The comprehensive feature data is processed through feature aggregation and feature recombination operations. Feature mapping technology is used to generate a feature map set containing potential defect patterns. The spatial distribution information in the feature map set is determined. Then, the saliency detection method is used to calculate the saliency score of each region in the feature map set. Based on the saliency score, the region is judged as a candidate defect region.

[0022] Candidate defect regions are precisely located through region screening and region optimization operations. Boundary detection technology is used to determine the precise boundaries of the defect regions, and the final coordinate information of the significant defect regions is obtained. Region labeling data containing defect location and defect features is generated, thus completing the in-depth feature analysis and significant region identification of photovoltaic module defects.

[0023] Furthermore, a dataset of suspected defects was obtained, specifically including:

[0024] The feature response intensity values ​​of each pixel in the feature map set are obtained. The response intensity values ​​are compared by a preset threshold mechanism to obtain preliminary screening results. Then, based on the pixel coordinates in the screening results, the boundary contour of the significant region is determined by the connected component analysis method to obtain the geometric shape parameters of the candidate region.

[0025] The candidate region is subjected to noise filtering through morphological operations to obtain an optimized set of suspected targets. Based on the center coordinates of each target in the set of suspected targets, the Euclidean distance matrix between the targets is calculated to determine the spatial distribution density characteristics of the targets.

[0026] Clustering algorithms are used to group the spatial distribution density features to obtain the distribution pattern of defect clusters. Statistical analysis methods are used to count the frequency of the location coordinates of each defect cluster to obtain a probability distribution map of the defect locations.

[0027] Based on the coordinates of high-frequency regions in the probability distribution map and the texture information of the original feature map set, a suspected defect dataset is obtained, and the final defect location distribution result is determined.

[0028] Furthermore, the classification result set is obtained, specifically including:

[0029] The image preprocessing module performs noise filtering and contrast enhancement on the image data in the suspected defect dataset to obtain standardized defect candidate region images. The edge detection algorithm is then used to extract the contour and texture features of the defect candidate regions to obtain a multi-dimensional feature vector combination.

[0030] The support vector machine classification model is used for binary classification. When the feature vector is located on the positive side of the decision boundary, the region is judged as a real defect region. When it is located on the negative side of the decision boundary, it is judged as a non-defect region. Based on the classification result of the real defect region, the defect type label information is determined.

[0031] The location coordinates of the real defect region in the original image are obtained by pixel coordinate extraction technology. The center point coordinates and boundary range parameters of the defect region are calculated. Then, a spatial distribution matrix is ​​constructed to record the distribution density and clustering degree of the defects, forming a complete set of classification results.

[0032] The defect type label information is associated and mapped with the spatial distribution matrix using a data structured storage method to generate a comprehensive database containing defect identification status, type attributes, and location information.

[0033] Furthermore, a causal analysis report was obtained, which specifically includes:

[0034] Obtain the defect type identifier code and corresponding defect feature vector from the classification result set, and extract the equipment operation status records and environmental parameter monitoring values ​​from the historical operation database to obtain a comprehensive dataset containing timestamps, equipment numbers, and operation parameters;

[0035] The comprehensive dataset is subjected to feature dimensionality reduction. Principal component analysis is used to select feature variables with high contribution. A mapping relationship table between defect types and environmental parameters is established. Data standardization is used to normalize parameters of different dimensions to obtain a standardized training sample set.

[0036] By fitting and calculating the standardized training sample set through multivariate regression analysis, a defect cause prediction model is constructed. The contribution score of each environmental parameter to defect formation is calculated using weight coefficients, and a ranking list of influencing factors is obtained.

[0037] Analyze the interaction strength among various factors, calculate the synergistic effect value among factors through the correlation matrix, determine the distribution ratio of compound influence paths and single influence paths, and generate a causal analysis report based on the distribution ratio and synergistic effect value.

[0038] Furthermore, the prediction dataset is obtained, specifically including:

[0039] Obtain defect records from the cause analysis report, and construct a time series sample set using the sliding window technique based on the time node information in the cause data to determine the time series feature vectors of defect occurrence frequency and intensity.

[0040] The ARIMA model is used to train the time series feature vectors to obtain the prediction parameters of the defect evolution trend. Then, the prediction parameters are used to calculate the probability of defect occurrence in future periods. Based on the probability distribution, a prediction dataset containing time dimension and probability value is generated.

[0041] The spatial interpolation algorithm is used to process the location coordinate information in the dataset to obtain the spatial distribution matrix of defect risk. Then, based on the risk probability value in the spatial distribution matrix, a threshold segmentation method is used to divide the risk into high, medium and low risk levels, and the boundary coordinates of the risk area corresponding to each level are determined.

[0042] Obtain the boundary coordinates of risk areas, identify the propagation paths of influencing factors between regions through a correlation pattern matching algorithm, and establish triggering conditions for a multi-regional joint early warning mechanism.

[0043] Furthermore, an operation and maintenance scheduling plan is obtained, which specifically includes:

[0044] The equipment status parameters and historical fault records in the prediction dataset are used to construct an equipment risk assessment matrix. The random forest algorithm is used to calculate the fault probability value and impact range coefficient of each equipment node to determine the risk area distribution map.

[0045] By using the failure probability value and impact range coefficient in the risk area distribution map, a priority score is established to obtain the priority ranking result of each area. Based on the priority ranking result and the total amount of currently available operation and maintenance resources, an initial scheduling plan is generated.

[0046] If there are resource conflicts or time overlaps in the initial scheduling scheme, the conflict nodes are identified by the scheduling conflict detection module, and the execution time or resource configuration of low-priority tasks are adjusted to obtain a conflict-free operation and maintenance scheduling scheme.

[0047] Based on the resource configuration data in the operation and maintenance scheduling plan, calculate the resource utilization rate and response time indicators of each region. If the resource utilization rate is lower than the preset threshold, reallocate idle resources to high priority regions and determine the final resource allocation optimization path.

[0048] Based on the resource allocation optimization path, an operation and maintenance task execution sequence and resource scheduling instructions are generated and transmitted to each operation and maintenance terminal device to complete the intelligent operation and maintenance scheduling deployment.

[0049] Furthermore, after obtaining the operation and maintenance scheduling plan, it also includes: continuously tracking the operating status of photovoltaic modules through a real-time monitoring module; if the monitoring data deviates from the preset normal range, a dynamic adjustment mechanism is triggered to obtain the adjusted scheduling instructions, determine the final execution plan, and then send the instructions to the field equipment through the automation control interface to complete the precise maintenance operation of the defective area, obtain execution feedback data, and judge the completion status and effect of the maintenance task.

[0050] This application provides an intelligent defect detection system for building-integrated photovoltaic (BIPV) modules, which implements an intelligent defect detection method for BIPV modules, including:

[0051] The data preprocessing module performs format standardization, band selection, temperature gradient analysis, principal component analysis for dimensionality reduction, spatial registration, and weighted fusion processing on hyperspectral image data and thermal imaging data to generate a fused dataset.

[0052] The feature extraction module uses a convolutional neural network to perform deep feature extraction on the fused dataset. Through multi-scale analysis and feature aggregation and recombination, it generates a feature map set containing potential defect patterns and identifies significant regions related to defects in the feature map. Through region filtering and optimization operations, it locates the coordinates of significant defect regions.

[0053] The preliminary defect screening module performs preliminary screening of salient regions in the feature map set according to a preset threshold, determines the boundary contours of salient regions through connected component analysis, filters out noise and optimizes the set of suspected targets using morphological operations, analyzes the spatial distribution density characteristics of targets through clustering algorithms, generates a probability distribution map of defect locations, and determines the final defect location distribution results.

[0054] The defect fine classification module uses a support vector machine classification model to finely classify defect regions. Through image preprocessing, feature extraction, and classification judgment, it determines the real defect regions and their types, and then constructs a spatial distribution matrix to record defect distribution information, generating a comprehensive database containing defect identification status, type attributes, and location information.

[0055] The defect cause analysis module combines the classification result set, historical operating data and environmental parameters, and constructs a defect cause prediction model through multivariate regression analysis. It analyzes the contribution and interaction of each factor to the formation of defects and generates a cause analysis report that includes the main influencing factors, secondary influencing factors and interaction relationships.

[0056] The defect trend prediction module, based on the key influencing factors in the cause analysis report, uses time series analysis technology to predict the development trend of defects. Through ARIMA model training and spatial interpolation algorithm, it generates a prediction dataset containing time dimension and probability value, divides risk level areas, and establishes a multi-area linkage early warning mechanism.

[0057] The operation and maintenance strategy generation module generates targeted operation and maintenance strategies based on the predicted dataset. Combined with the intelligent operation and maintenance system, it prioritizes risk areas, generates operation and maintenance scheduling plans, and optimizes resource allocation through real-time monitoring and dynamic adjustment mechanisms.

[0058] The beneficial effects of this invention are as follows:

[0059] By constructing a multi-source data fusion framework that integrates hyperspectral imagery and thermal imaging data, and utilizing convolutional neural networks for deep feature extraction, the problem of insufficient detection accuracy and response speed caused by traditional methods relying on a single data source and conventional algorithms is effectively solved. This not only improves the accuracy of defect detection but also significantly accelerates the detection speed, providing a high-quality data foundation for subsequent defect analysis and processing, and ensuring that defects can be detected and processed in a timely manner.

[0060] This invention addresses the shortcomings of existing technologies in defect cause analysis and predictive maintenance by constructing a defect cause analysis model, combining historical operating data and environmental parameters, and employing multivariate regression analysis and time series analysis techniques. This approach can accurately predict defect development trends, identify risk areas in advance, provide a scientific basis for operation and maintenance decisions, optimize the allocation of operation and maintenance resources, improve operation and maintenance efficiency, and ensure the long-term stable operation of photovoltaic systems.

[0061] By generating targeted operation and maintenance strategies through an intelligent operation and maintenance system, and combining risk area priority ranking and resource optimization allocation, the shortcomings of existing technologies in operation and maintenance resource allocation and execution efficiency are solved. The real-time monitoring module dynamically adjusts the operation and maintenance process to ensure the accurate execution of operation and maintenance tasks, significantly improves the utilization efficiency of operation and maintenance resources, optimizes the operation and maintenance process, ensures the timely handling of defective areas, and improves the overall stability and reliability of the photovoltaic system. Attached Figure Description

[0062] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0063] Figure 1 This is a flowchart illustrating an intelligent defect detection method for building-integrated photovoltaic (BIPV) modules provided in Embodiment 1 of this application.

[0064] Figure 2 This is a flowchart illustrating the process of obtaining a cause analysis report using an intelligent defect detection method for building-integrated photovoltaic (BIPV) modules provided in Embodiment 1 of this application.

[0065] Figure 3 This is a flowchart illustrating the process of obtaining a predicted dataset using an intelligent defect detection method for building-integrated photovoltaic (BIPV) modules, as provided in Embodiment 1 of this application.

[0066] Figure 4 This is a structural schematic diagram of an intelligent defect detection system for building photovoltaic integrated modules provided in Embodiment 2 of this application. Detailed Implementation

[0067] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of methods and systems consistent with some aspects of this application as detailed in the appended claims.

[0068] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0069] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0070] Example 1

[0071] See also Figures 1-3 This embodiment provides an intelligent defect detection method for building-integrated photovoltaic (BIPV) modules, including the following steps:

[0072] S1. By constructing a multi-source data fusion framework, unified preprocessing is performed on hyperspectral image data and thermal imaging data. Hierarchical sampling technology is used to reduce the dimensionality of the original data to obtain a fused dataset, which is used for subsequent feature extraction. This ensures that key defect information can still be retained in complex environments, resulting in a pre-processed dataset.

[0073] Furthermore, the fused dataset is obtained, specifically including:

[0074] Hyperspectral image data and thermal imaging data are acquired as raw multi-source data. Data matrix with unified format is obtained through data format standardization. Then, the hyperspectral image data is filtered by spectral band selection method. The information content of each band is determined according to the information entropy calculation results to obtain optimized hyperspectral data.

[0075] Thermal imaging data is preprocessed by temperature gradient analysis to calculate the temperature change rate of pixels and mark them as key regions to obtain thermal imaging key region data. Principal component analysis algorithm is used to reduce the dimensionality of the optimized hyperspectral data and thermal imaging key region data. The number of principal components is determined according to the cumulative contribution rate to obtain the dimensionality-reduced feature vector set.

[0076] The dimensionality-reduced feature vector set is geometrically corrected by spatial registration technology. Transformation parameters are calculated based on the coordinates of control points to obtain spatially aligned multi-source feature data. Then, a weighted fusion method is used to fuse the spatially aligned multi-source feature data. The fusion weights are calculated based on the signal-to-noise ratio of each data source and the weight allocation is adjusted to obtain the fused dataset.

[0077] It also includes: extracting features from the first-level fusion dataset using a defect feature detection algorithm, calculating texture features and spectral feature parameters, and obtaining a set of feature descriptors containing key defect information.

[0078] Specifically, by constructing a multi-source data fusion framework, hyperspectral image data and thermal imaging data are preprocessed, reduced in dimensionality, and fused in a unified manner. This effectively preserves key defect information in complex environments, generates a fused dataset, and extracts a set of feature descriptors through a defect feature detection algorithm. This provides a high-quality data foundation for subsequent deep feature analysis and defect identification, significantly improving the accuracy and reliability of defect detection. It helps to promptly discover and address potential defects in photovoltaic modules, ensuring the stable operation and performance optimization of photovoltaic systems.

[0079] S2. Based on the fused dataset, a convolutional neural network model is used to perform deep feature extraction on the data. Multi-scale analysis is performed on the minute features of photovoltaic module defects to obtain a feature map set containing potential defect patterns and to determine the significant regions related to defects in the feature map.

[0080] Furthermore, a feature map set containing potential defect patterns is obtained, specifically including:

[0081] Photovoltaic module image data from the fusion dataset is acquired, and a convolutional neural network model is used to extract deep features from the standard image data. Through multi-layer convolution and pooling operations, feature representations at different levels are obtained, and the initial feature mapping containing defect information is determined.

[0082] A multi-scale analysis method is used to construct feature pyramids at different scales. Multi-scale feature representations are obtained through scale transformation and feature fusion operations, resulting in comprehensive feature data covering the features of minute defects.

[0083] The comprehensive feature data is processed through feature aggregation and feature recombination operations. Feature mapping technology is used to generate a feature map set containing potential defect patterns. The spatial distribution information in the feature map set is determined. Then, the saliency detection method is used to calculate the saliency score of each region in the feature map set. Based on the saliency score, the region is judged as a candidate defect region.

[0084] Candidate defect regions are precisely located through region screening and region optimization operations. Boundary detection technology is used to determine the precise boundaries of the defect regions, and the final coordinate information of the significant defect regions is obtained. Region labeling data containing defect location and defect features is generated, thus completing the in-depth feature analysis and significant region identification of photovoltaic module defects.

[0085] Specifically, by employing a convolutional neural network model to extract deep features from the fused dataset and combining it with a multi-scale analysis method to construct a feature pyramid, this method can effectively capture the subtle features of photovoltaic module defects, generate a feature map containing potential defect patterns, and further accurately locate significant defect regions and label defect locations and features through saliency detection and region optimization operations. This achieves deep feature analysis and saliency region identification of photovoltaic module defects, significantly improving the accuracy and reliability of defect detection and providing strong support for subsequent defect classification and cause analysis.

[0086] S3. Using the feature map set, a preset threshold is used to initially screen the significant regions. If the feature value of a significant region exceeds the preset threshold, it is marked as a suspected defect region, and a suspected defect dataset is obtained to determine the distribution of possible defect locations.

[0087] Furthermore, a dataset of suspected defects was obtained, specifically including:

[0088] The feature response intensity values ​​of each pixel in the feature map set are obtained. The response intensity values ​​are compared by a preset threshold mechanism to obtain preliminary screening results. Then, based on the pixel coordinates in the screening results, the boundary contour of the significant region is determined by the connected component analysis method to obtain the geometric shape parameters of the candidate region.

[0089] The candidate regions are subjected to noise filtering through morphological operations. If the area of ​​a region is less than a preset area threshold, the region is removed to obtain an optimized set of suspected targets. Based on the center coordinates of each target in the set of suspected targets, the Euclidean distance matrix between the targets is calculated to determine the spatial distribution density characteristics of the targets.

[0090] Clustering algorithms are used to group the spatial distribution density features. If the distance between targets is less than a preset distance threshold, they are grouped into the same defect cluster to obtain the distribution pattern of the defect cluster. Statistical analysis methods are used to count the frequency of the position coordinates of each defect cluster and calculate the probability of the defects appearing in different regions to obtain a probability distribution map of the defect locations.

[0091] Based on the coordinates of high-frequency regions in the probability distribution map and the texture information of the original feature map set, a suspected defect dataset is obtained, and the final defect location distribution result is determined.

[0092] Specifically, by using a feature map set and a preset threshold, significant regions are initially screened to mark suspected defect areas, forming a suspected defect dataset. The possible location distribution of defects is then determined. This process effectively reduces the amount of data and removes a large amount of interference from non-defect areas. At the same time, by using connected component analysis, morphological operations, clustering algorithms, and other methods, candidate regions are optimized, screened, and grouped to ultimately determine the location distribution characteristics of defects. This provides more accurate location information for further defect analysis and processing, improving the efficiency and accuracy of defect detection.

[0093] S4. For the suspected defect dataset, a support vector machine classification model is used to classify the suspected defect regions in detail to obtain a classification result set, determine the regions that belong to the real defects, and record the type and spatial distribution information of the defects.

[0094] Furthermore, the classification result set is obtained, specifically including:

[0095] Image data from a suspected defect dataset is obtained. The image preprocessing module performs noise filtering and contrast enhancement on the image data to obtain standardized defect candidate region images. An edge detection algorithm is used to extract the contour and texture features of the defect candidate regions to obtain a multi-dimensional feature vector combination.

[0096] For the combination of multidimensional feature vectors, a binary classification judgment is performed using a support vector machine classification model. When the feature vector is located on the positive side of the decision boundary, the region is judged to be a real defect region. When it is located on the negative side of the decision boundary, it is judged to be a non-defect region. Based on the classification result of the real defect region, a morphological analysis method is used to perform geometric shape recognition on the real defect region to determine the defect type label information.

[0097] The location coordinates of the real defect region in the original image are obtained by pixel coordinate extraction technology. The center point coordinates and boundary range parameters of the defect region are calculated. Then, a spatial distribution matrix is ​​constructed to record the distribution density and clustering degree of the defects, forming a complete set of classification results.

[0098] The defect type label information is associated and mapped with the spatial distribution matrix using a data structured storage method to generate a comprehensive database containing defect identification status, type attributes, and location information.

[0099] Specifically, by employing a support vector machine classification model to perform fine classification on the suspected defect dataset, this method can accurately distinguish between real defect areas and non-defect areas, determine the defect type, and record its spatial distribution information, forming a complete set of classification results. This process effectively improves the accuracy of defect identification. At the same time, by constructing a comprehensive database, it realizes the structured storage of defect information, providing detailed and accurate data support for subsequent defect cause analysis, trend prediction, and operation and maintenance strategy formulation, thereby improving the overall efficiency and scientific nature of photovoltaic module defect management.

[0100] S5. Based on the classification result set, combined with historical operating data and environmental parameters, construct a defect cause analysis model, and use multivariate regression analysis to mine the correlation between defect types and distribution information, obtain a cause analysis report, and determine the potential influencing factors of defect formation.

[0101] Furthermore, a causal analysis report was obtained, which specifically includes:

[0102] S51. Obtain the defect type identifier code and the corresponding defect feature vector from the classification result set, and at the same time extract the equipment operation status records and environmental parameter monitoring values ​​from the historical operation database to obtain a comprehensive dataset containing timestamps, equipment numbers, and operation parameters.

[0103] S52. Perform feature dimensionality reduction on the comprehensive dataset, select feature variables with high contribution through principal component analysis, determine the feature matrix after dimensionality reduction as the model input, establish a mapping relationship table between defect types and environmental parameters, and normalize parameters of different dimensions using data standardization methods to obtain a standardized training sample set.

[0104] S53. Using multivariate regression analysis, a defect cause prediction model is constructed by fitting the standardized training sample set. The regression equation is Y = β0 + β1X1 + β2X2 + ... + β n X n +ε, Y represents the probability of defect occurrence, X1 to X n Represents the various environmental parameter variables, from β0 to β n Let ε represent the regression coefficient and ε represent the error term, and then obtain the weight coefficients of each parameter. Use the weight coefficients to calculate the contribution score of each environmental parameter to the formation of defects, and determine whether it is a major influencing factor based on the contribution score, and obtain a ranking list of influencing factors.

[0105] S54. Analyze the interaction strength between factors, calculate the synergistic effect value between factors through the correlation matrix, determine the distribution ratio of compound influence path and single influence path, and generate a causal analysis report containing the main influencing factors, secondary influencing factors, and interaction relationships based on the distribution ratio and synergistic effect value. The report marks the risk level and warning threshold range of each influencing factor.

[0106] In this process, principal component analysis (PCA) is used to perform feature dimensionality reduction on the comprehensive dataset to identify high-contribution feature variables. Specifically, the data is first standardized to eliminate the influence of different feature dimensions and magnitudes. Then, PCA is used to transform the original feature variables into a set of uncorrelated principal components. Each principal component is a linear combination of the original features, and the variance contribution rate of each principal component is calculated. Principal components with high variance contribution rates retain more information from the original data. Therefore, by selecting principal components with cumulative variance contribution rates reaching a certain threshold (such as 85% or 90%), high-contribution feature variables can be identified, thus providing key data support for subsequent defect cause analysis.

[0107] Specifically, by constructing a defect cause analysis model, combining classification result sets, historical operating data, and environmental parameters, multivariate regression analysis is used to mine the correlation between defect types and distribution information, generating a cause analysis report. This process not only identifies the potential influencing factors of defect formation and their contribution, but also analyzes the interaction between various factors, clarifies the main and secondary influencing factors, provides a scientific basis for defect prevention and management, helps to formulate targeted operation and maintenance strategies in advance, reduces the risk of defect occurrence, and ensures the stable operation of photovoltaic systems.

[0108] S6. Through the cause analysis report, extract key influencing factors, use time series analysis technology to predict the development trend of defects, obtain a prediction dataset, and determine the defect risk areas that may appear in the future.

[0109] Furthermore, the prediction dataset is obtained, specifically including:

[0110] S61. Obtain the defect records in the cause analysis report to get structured cause data containing time nodes, location coordinates and defect types. Then, based on the time node information in the cause data, use the sliding window technique to construct a time series sample set and determine the time series feature vector of defect occurrence frequency and intensity.

[0111] S62. Train the time series feature vectors using the ARIMA model to obtain the prediction parameters of the defect evolution trend. Then use the prediction parameters to calculate the probability of defect occurrence in future periods and generate a prediction dataset containing time dimension and probability value based on the probability distribution.

[0112] S63. The location coordinate information in the dataset is processed by spatial interpolation algorithm. If the interpolation accuracy meets the preset standard, the spatial distribution matrix of defect risk is obtained. Then, according to the risk probability value in the spatial distribution matrix, the threshold segmentation method is used to divide the risk levels into high, medium and low risk levels, and the boundary coordinates of the risk area corresponding to each level are determined.

[0113] S64. Obtain the boundary coordinate data of the risk area, identify the propagation path of the influencing factors between regions through the correlation pattern matching algorithm, and establish the triggering conditions for the multi-regional linkage early warning mechanism.

[0114] Specifically, by extracting key influencing factors from the causal analysis report and applying time series analysis techniques, this method can effectively predict the development trend of defects, generating a predictive dataset containing time dimensions and probability values. Furthermore, through spatial interpolation algorithms and threshold segmentation methods, it identifies potential defect risk areas within a future period, classifying them into high, medium, and low risk levels. Simultaneously, it identifies the propagation paths of influencing factors between regions and establishes a multi-regional collaborative early warning mechanism. This process provides forward-looking decision support for the operation and maintenance management of photovoltaic systems, helping to deploy resources in advance, optimize operation and maintenance strategies, reduce the impact of defects on system performance, and ensure the long-term stable operation of photovoltaic systems.

[0115] S7. Based on the predicted dataset, generate targeted operation and maintenance strategies, prioritize risk areas using the intelligent operation and maintenance system, obtain an operation and maintenance scheduling plan, and determine the optimal path for resource allocation.

[0116] Furthermore, an operation and maintenance scheduling plan is obtained, which specifically includes:

[0117] The equipment status parameters and historical fault records in the prediction dataset are used to construct an equipment risk assessment matrix. The random forest algorithm is used to calculate the fault probability value and impact range coefficient of each equipment node and determine the risk area distribution map.

[0118] By using the failure probability value and impact range coefficient in the risk area distribution map, a priority scoring formula is established: P = α × failure probability value + β × impact range coefficient, where α represents the probability weight factor and β represents the impact weight factor. The priority ranking results of each area are obtained. Based on the priority ranking results and the total amount of available operation and maintenance resources, a greedy algorithm is used to allocate the number of operation and maintenance personnel and the equipment maintenance time window in order of priority from high to low to generate an initial scheduling plan.

[0119] If there are resource conflicts or time overlaps in the initial scheduling scheme, the conflict nodes are identified by the scheduling conflict detection module, and the execution time or resource configuration of low-priority tasks are adjusted to obtain a conflict-free operation and maintenance scheduling scheme.

[0120] Based on the resource configuration data in the operation and maintenance scheduling plan, calculate the resource utilization rate and response time indicators of each region. If the resource utilization rate is lower than the preset threshold, reallocate idle resources to high priority regions and determine the final resource allocation optimization path.

[0121] Based on the resource allocation optimization path, an operation and maintenance task execution sequence and resource scheduling instructions are generated and transmitted to each operation and maintenance terminal device to complete the intelligent operation and maintenance scheduling deployment.

[0122] Specifically, it has enabled intelligent operation and maintenance strategy generation and resource optimization scheduling based on predictive data, which has significantly improved operation and maintenance efficiency and resource utilization, ensured that risk areas are dealt with in a timely manner, and enhanced the reliability and stability of photovoltaic systems.

[0123] Furthermore, after obtaining the operation and maintenance scheduling plan, it also includes: continuously tracking the operating status of photovoltaic modules through a real-time monitoring module; if the monitoring data deviates from the preset normal range, a dynamic adjustment mechanism is triggered to obtain the adjusted scheduling instructions, determine the final execution plan, and then send the instructions to the field equipment through the automation control interface to complete the precise maintenance operation of the defective area, obtain execution feedback data, and judge the completion status and effect of the maintenance task.

[0124] Example 2

[0125] See also Figure 4 This embodiment provides an intelligent defect detection system for building-integrated photovoltaic (BIPV) modules, which is used to implement an intelligent defect detection method for BIPV modules, including:

[0126] The data preprocessing module performs format standardization, band selection, temperature gradient analysis, principal component analysis dimensionality reduction, spatial registration, and weighted fusion processing on hyperspectral image data and thermal imaging data to generate a fused dataset. Through these operations, it ensures that key defect information can still be preserved in complex environments, providing a high-quality data foundation for subsequent feature extraction.

[0127] The feature extraction module uses a convolutional neural network to perform deep feature extraction on the fused dataset. Through multi-scale analysis and feature aggregation and recombination, it generates a feature map set containing potential defect patterns and identifies salient regions related to defects in the feature map. Through region filtering and optimization operations, it accurately locates the coordinates of salient defect regions, completing the deep feature analysis and salient region identification of photovoltaic module defects.

[0128] The preliminary defect screening module performs preliminary screening of salient regions in the feature map set according to a preset threshold, determines the boundary contours of salient regions through connected component analysis, filters out noise and optimizes the set of suspected targets using morphological operations, analyzes the spatial distribution density characteristics of targets through clustering algorithms, generates a probability distribution map of defect locations, and determines the final defect location distribution results.

[0129] The defect fine classification module uses a support vector machine classification model to finely classify defect regions. Through image preprocessing, feature extraction, and classification judgment, it determines the real defect regions and their types, and then constructs a spatial distribution matrix to record defect distribution information, generating a comprehensive database containing defect identification status, type attributes, and location information.

[0130] The defect cause analysis module combines the classification result set, historical operating data and environmental parameters, and constructs a defect cause prediction model through multivariate regression analysis. It analyzes the contribution and interaction of each factor to the formation of defects, and generates a cause analysis report containing the main influencing factors, secondary influencing factors and interaction relationships, providing a scientific basis for defect prevention and management.

[0131] The defect trend prediction module, based on the key influencing factors in the cause analysis report, uses time series analysis technology to predict the development trend of defects. Through ARIMA model training and spatial interpolation algorithm, it generates a prediction dataset containing time dimension and probability value, divides high, medium and low risk level areas, and establishes a multi-area linkage early warning mechanism to provide forward-looking guidance for operation and maintenance strategies.

[0132] The operation and maintenance strategy generation module generates targeted operation and maintenance strategies based on the predicted dataset. Combined with the intelligent operation and maintenance system, it prioritizes risk areas and generates operation and maintenance scheduling plans. Through real-time monitoring and dynamic adjustment mechanisms, it optimizes resource allocation, ensures efficient execution of operation and maintenance tasks, completes precise maintenance operations in defective areas, and guarantees the stable operation of photovoltaic modules.

[0133] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent detection of defects in building-integrated photovoltaic (BIPV) modules, characterized in that: Includes the following steps: Hyperspectral image data and thermal imaging data are preprocessed in a unified manner, and hierarchical sampling technology is used to reduce the dimensionality of the original data to obtain a fused dataset. Based on the fused dataset, a convolutional neural network model is used to perform deep feature extraction on the data to obtain a feature map set containing potential defect patterns. By using feature maps and a preset threshold, significant regions are initially screened to obtain a dataset of suspected defects. A support vector machine classification model is used to perform fine classification of suspected defective regions, obtain a classification result set, and determine the regions that belong to real defects. By combining historical operating data and environmental parameters, a defect cause analysis model is constructed. Multivariate regression analysis is used to mine the correlation between defect types and distribution information, and a cause analysis report is obtained. By analyzing the causes, key influencing factors are extracted, and time series analysis is used to predict the development trend of defects, thus obtaining a prediction dataset. Based on the predicted dataset, targeted operation and maintenance strategies are generated. Combined with the intelligent operation and maintenance system, risk areas are prioritized to obtain an operation and maintenance scheduling plan and determine the optimal path for resource allocation.

2. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The resulting fused dataset includes: Hyperspectral image data and thermal imaging data are acquired as raw multi-source data. The hyperspectral image data is filtered by a spectral band selection method. The information content of each band is determined based on the information entropy calculation results to obtain optimized hyperspectral data. Thermal imaging data is preprocessed by temperature gradient analysis to calculate the temperature change rate of pixels and obtain key thermal imaging region data. Principal component analysis algorithm is used to reduce the dimensionality of the optimized hyperspectral data and key thermal imaging region data to obtain a set of dimensionality-reduced feature vectors. The dimensionality-reduced feature vector set is geometrically corrected by spatial registration technology. Transformation parameters are calculated based on the coordinates of control points to obtain spatially aligned multi-source feature data. Then, a weighted fusion method is used to fuse the spatially aligned multi-source feature data to obtain a fused dataset.

3. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The feature map set containing potential defect patterns is obtained, specifically including: Acquire photovoltaic module image data from the fusion dataset, and use a convolutional neural network model to perform deep feature extraction on the standard image data to determine the initial feature mapping containing defect information; A multi-scale analysis method is used to construct feature pyramids at different scales. Multi-scale feature representations are obtained through scale transformation and feature fusion operations, resulting in comprehensive feature data covering the features of minute defects. The comprehensive feature data is processed through feature aggregation and feature recombination operations. Feature mapping technology is used to generate a feature map set containing potential defect patterns. The spatial distribution information in the feature map set is determined. Then, the saliency detection method is used to calculate the saliency score of each region in the feature map set. Based on the saliency score, the region is judged as a candidate defect region. Candidate defect regions are precisely located through region screening and region optimization operations. Boundary detection technology is used to determine the precise boundaries of the defect regions, and the final coordinate information of the significant defect regions is obtained. Region labeling data containing defect location and defect features is generated, thus completing the in-depth feature analysis and significant region identification of photovoltaic module defects.

4. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The dataset of suspected defects was obtained, specifically including: The feature response intensity values ​​of each pixel in the feature map set are obtained. The response intensity values ​​are compared by a preset threshold mechanism to obtain preliminary screening results. Then, based on the pixel coordinates in the screening results, the boundary contour of the significant region is determined by the connected component analysis method to obtain the geometric shape parameters of the candidate region. The candidate region is subjected to noise filtering through morphological operations to obtain an optimized set of suspected targets. Based on the center coordinates of each target in the set of suspected targets, the Euclidean distance matrix between the targets is calculated to determine the spatial distribution density characteristics of the targets. Clustering algorithms are used to group the spatial distribution density features to obtain the distribution pattern of defect clusters. Statistical analysis methods are used to count the frequency of the location coordinates of each defect cluster to obtain a probability distribution map of the defect locations. Based on the coordinates of high-frequency regions in the probability distribution map and the texture information of the original feature map set, a suspected defect dataset is obtained, and the final defect location distribution result is determined.

5. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The classification result set obtained includes: The image preprocessing module performs noise filtering and contrast enhancement on the image data in the suspected defect dataset to obtain standardized defect candidate region images. The edge detection algorithm is then used to extract the contour and texture features of the defect candidate regions to obtain a multi-dimensional feature vector combination. The support vector machine classification model is used for binary classification. When the feature vector is located on the positive side of the decision boundary, the region is judged as a real defect region. When it is located on the negative side of the decision boundary, it is judged as a non-defect region. Based on the classification result of the real defect region, the defect type label information is determined. The location coordinates of the real defect region in the original image are obtained by pixel coordinate extraction technology. The center point coordinates and boundary range parameters of the defect region are calculated. Then, a spatial distribution matrix is ​​constructed to record the distribution density and clustering degree of the defects, forming a complete set of classification results. The defect type label information is associated and mapped with the spatial distribution matrix using a data structured storage method to generate a comprehensive database containing defect identification status, type attributes, and location information.

6. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The causal analysis report received includes: Obtain the defect type identifier code and corresponding defect feature vector from the classification result set, and extract the equipment operation status records and environmental parameter monitoring values ​​from the historical operation database to obtain a comprehensive dataset containing timestamps, equipment numbers, and operation parameters; The comprehensive dataset is subjected to feature dimensionality reduction. Principal component analysis is used to select feature variables with high contribution. A mapping relationship table between defect types and environmental parameters is established. Data standardization is used to normalize parameters of different dimensions to obtain a standardized training sample set. By fitting and calculating the standardized training sample set through multivariate regression analysis, a defect cause prediction model is constructed. The contribution score of each environmental parameter to defect formation is calculated using weight coefficients, and a ranking list of influencing factors is obtained. Analyze the interaction strength among various factors, calculate the synergistic effect value among factors through the correlation matrix, determine the distribution ratio of compound influence paths and single influence paths, and generate a causal analysis report based on the distribution ratio and synergistic effect value.

7. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The prediction dataset is obtained, specifically including: Obtain defect records from the cause analysis report, and construct a time series sample set using the sliding window technique based on the time node information in the cause data to determine the time series feature vectors of defect occurrence frequency and intensity. The ARIMA model is used to train the time series feature vectors to obtain the prediction parameters of the defect evolution trend. Then, the prediction parameters are used to calculate the probability of defect occurrence in future periods. Based on the probability distribution, a prediction dataset containing time dimension and probability value is generated. The spatial interpolation algorithm is used to process the location coordinate information in the dataset to obtain the spatial distribution matrix of defect risk. Then, based on the risk probability value in the spatial distribution matrix, a threshold segmentation method is used to divide the risk into high, medium and low risk levels, and the boundary coordinates of the risk area corresponding to each level are determined. Obtain the boundary coordinates of risk areas, identify the propagation paths of influencing factors between regions through a correlation pattern matching algorithm, and establish triggering conditions for a multi-regional joint early warning mechanism.

8. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 1, characterized in that: The operation and maintenance scheduling plan includes: The equipment status parameters and historical fault records in the prediction dataset are used to construct an equipment risk assessment matrix. The random forest algorithm is used to calculate the fault probability value and impact range coefficient of each equipment node and determine the risk area distribution map. By using the failure probability value and impact range coefficient in the risk area distribution map, a priority score is established to obtain the priority ranking result of each area. Based on the priority ranking result and the total amount of currently available operation and maintenance resources, an initial scheduling plan is generated. If there are resource conflicts or time overlaps in the initial scheduling scheme, the conflict nodes are identified by the scheduling conflict detection module, and the execution time or resource configuration of low-priority tasks are adjusted to obtain a conflict-free operation and maintenance scheduling scheme. Based on the resource configuration data in the operation and maintenance scheduling plan, calculate the resource utilization rate and response time indicators of each region. If the resource utilization rate is lower than the preset threshold, reallocate idle resources to high priority regions and determine the final resource allocation optimization path. Based on the resource allocation optimization path, an operation and maintenance task execution sequence and resource scheduling instructions are generated and transmitted to each operation and maintenance terminal device to complete the intelligent operation and maintenance scheduling deployment.

9. The intelligent defect detection method for building-integrated photovoltaic (BIPV) modules according to claim 8, characterized in that: After obtaining the operation and maintenance scheduling plan, it also includes: continuously tracking the operating status of photovoltaic modules through a real-time monitoring module; if the monitoring data deviates from the preset normal range, a dynamic adjustment mechanism is triggered to obtain the adjusted scheduling instructions, determine the final execution plan, and then send the instructions to the field equipment through the automation control interface to complete the precise maintenance operation of the defective area, obtain execution feedback data, and judge the completion status and effect of the maintenance task.

10. A building-integrated photovoltaic (BIPV) module defect intelligent detection system, used to implement the building-integrated photovoltaic (BIPV) module defect intelligent detection method as described in any one of claims 1-9, characterized in that: include: The data preprocessing module performs format standardization, band selection, temperature gradient analysis, principal component analysis dimensionality reduction, spatial registration, and weighted fusion processing on hyperspectral image data and thermal imaging data to generate a fused dataset. The feature extraction module uses a convolutional neural network to perform deep feature extraction on the fused dataset. Through multi-scale analysis and feature aggregation and recombination, it generates a feature map set containing potential defect patterns and identifies significant regions related to defects in the feature map. Through region filtering and optimization operations, it locates the coordinates of significant defect regions. The preliminary defect screening module performs preliminary screening of salient regions in the feature map set according to a preset threshold, determines the boundary contours of salient regions through connected component analysis, filters out noise and optimizes the set of suspected targets using morphological operations, analyzes the spatial distribution density characteristics of targets through clustering algorithms, generates a probability distribution map of defect locations, and determines the final defect location distribution results. The defect fine classification module uses a support vector machine classification model to finely classify defect regions. Through image preprocessing, feature extraction, and classification judgment, it determines the real defect regions and their types, and then constructs a spatial distribution matrix to record defect distribution information, generating a comprehensive database containing defect identification status, type attributes, and location information. The defect cause analysis module combines the classification result set, historical operating data and environmental parameters, and constructs a defect cause prediction model through multivariate regression analysis. It analyzes the contribution and interaction of each factor to the formation of defects and generates a cause analysis report. The defect trend prediction module, based on the key influencing factors in the cause analysis report, uses time series analysis technology to predict the development trend of defects. Through ARIMA model training and spatial interpolation algorithm, it generates a prediction dataset containing time dimension and probability value, divides risk level areas, and establishes a multi-area linkage early warning mechanism. The operation and maintenance strategy generation module generates targeted operation and maintenance strategies based on the predicted dataset. Combined with the intelligent operation and maintenance system, it prioritizes risk areas, generates operation and maintenance scheduling plans, and optimizes resource allocation through real-time monitoring and dynamic adjustment mechanisms.

Images