Photovoltaic module quality detection method based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence and photovoltaic detection, in particular to a photovoltaic module quality detection method based on artificial intelligence, and the method comprises the steps: building a spectrum-geometric model and a multi-field coupling model through multi-modal data collection, carrying out the feature fusion through the combination of a hierarchical attention mechanism, and carrying out the detection of the quality of a photovoltaic module. A generative adversarial network is utilized to extract key deep features, a defect area is accurately positioned, and severity is evaluated. In addition, according to the scheme, the assembly operation condition is simulated through the digital twinning technology, the defect expansion trend is predicted in combination with a time sequence coupling prediction network, and high efficiency and foresight of defect detection are achieved. The method can be widely applied to the field of photovoltaic power generation, the precision and efficiency of module defect detection are remarkably improved, and technical support is provided for efficient operation and maintenance of a photovoltaic system.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and photovoltaic detection technology, and in particular to a photovoltaic component quality detection method based on artificial intelligence. Background Art

[0002] Photovoltaic modules are the core equipment of photovoltaic power generation systems, and their quality directly affects power generation efficiency and system life. However, photovoltaic modules are prone to defects such as hidden cracks, hot spots, broken grids, and cold solder joints when they are operated in complex environments (such as high temperature, ultraviolet rays, humidity, etc.) for a long time. These defects will lead to reduced power generation efficiency and even cause more serious failures.

[0003] In the existing technology (China Public Invention Patent, Publication No.: CN114399484A, Name: A Photovoltaic Module Defect Detection Method, Device, Equipment and Medium), electroluminescence (EL) image detection and thermal imaging detection methods are commonly used. These methods mainly rely on single-mode image features, such as EL images to identify surface defects such as surface cracks and broken grids of photovoltaic modules, and thermal imaging mainly identifies hot spots. However, these technologies have the following defects: they cannot fully identify internal defects (such as circuit short circuits) and dynamic defects (such as hidden crack extension); they are limited to single-mode data and lack comprehensive characterization of complex defects, resulting in a high rate of missed reports or false alarms; they rely on manual inspection and sampling detection, and the detection efficiency is difficult to meet the large-scale detection needs of photovoltaic power stations. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior art, the present invention provides a photovoltaic module quality detection method based on artificial intelligence. The present invention constructs a spectral-geometric model and a multi-field coupling model through multimodal data acquisition (including spectral, thermal, current and geometric data), and uses feature mapping and deep learning to perform multimodal feature fusion and defect detection. At the same time, combined with digital twin and time series prediction technology, the operating status of photovoltaic modules is dynamically simulated, defects are accurately identified, and defect expansion trends are predicted. The present invention significantly improves the accuracy and efficiency of photovoltaic module defect detection and provides a scientific basis for the operation and maintenance management of modules.

[0005] A photovoltaic module quality detection method based on artificial intelligence includes the following steps: Collect spectral polarization data, thermal imaging data, current imaging data and stereo depth data of photovoltaic modules, perform time synchronization and spatial calibration on all data, and generate preliminary multi-modal feature data; Based on the preliminary multimodal feature data, a spectral-geometric model and a multi-field coupling model are constructed to generate model data respectively; the generated model data are fused through a feature mapping algorithm, and input into a generative adversarial network for feature abstraction to obtain key deep feature data; Input the key deep feature data into the hierarchical attention mechanism to fuse global and local features to generate fused defect feature data; use the region detection algorithm to locate the defect region of the fused defect feature data, classify the defect region and evaluate the severity to generate defect classification data and defect evaluation data; The model data is loaded into the digital twin platform, the operating conditions are simulated, and virtual detection data is generated; a time series analysis is performed on the virtual detection data to predict the expansion trend of defects and optimize the detection system parameters.

[0006] Preferably, the spectral polarization data is acquired by adjusting the polarization angle to capture the multi-angle spectral characteristics of the surface of the photovoltaic module, the thermal imaging data is acquired by recording the temperature distribution characteristics of the surface of the photovoltaic module, the current imaging data is acquired by applying a small current to the photovoltaic module and synchronously recording the current distribution characteristics, and the stereo depth data is acquired by generating the three-dimensional morphological characteristics of the component surface through multi-angle ray tracing.

[0007] Preferably, the time synchronization is based on timestamp technology, and a unified time base is used to mark the collection time of spectral polarization data, thermal imaging data, current imaging data and stereo depth data; the spatial calibration uses a feature point matching algorithm to align the thermal imaging data with the abnormal areas in the current imaging data, and maps the aligned results to the three-dimensional coordinate space of the stereo depth data to generate preliminary multimodal feature data after spatial calibration.

[0008] Preferably, the feature mapping algorithm realizes dynamic mapping of spectral-geometric model data by calculating the correlation matrix between spectral reflectance characteristics and geometric morphology, and extracts the correlation characteristics of the internal stress field and electric field of the component by performing multi-field tensor decomposition on thermal imaging data and current imaging data, and jointly models the correlation characteristics to generate fusion model data after feature mapping.

[0009] Preferably, the fusion model data is input into a generative adversarial network, which includes a generator and a discriminator. The generator generates deep abstract features through back-propagation optimization, and the discriminator optimizes the feature data output by the generator by judging the difference between the generated features and the real features, thereby generating enhanced key deep feature data.

[0010] Preferably, the hierarchical attention mechanism performs weighted processing on the overall spectral distribution, thermal distribution and current characteristics of the photovoltaic module through a global feature extraction module, and at the same time focuses on the spectral anomaly points, hot spot locations and current anomaly points in the defect area through a local feature extraction module, so as to achieve efficient fusion of global features and local features and generate fused defect feature data.

[0011] Preferably, the region detection algorithm performs multi-level feature extraction on the fused defect feature data, first roughly determines the region where the defect exists through an edge detection algorithm, then performs fine-grained boundary optimization on the defect region through an improved region extraction algorithm, and finally outputs the precise position coordinates and region boundary information of the defect region.

[0012] Preferably, the timing analysis includes time series decomposition of virtual detection data, extracting dynamic change patterns of defect areas, including hot spot expansion rate, resistance change rate and spectral reflectivity fluctuations, and extrapolating defect expansion trends based on a timing coupled prediction network to generate defect trend prediction data.

[0013] Preferably, the optimization of detection system parameters includes adjusting the acquisition frequency of the spectral imaging device, dynamically adjusting the temperature sensitivity threshold of the thermal imaging device, optimizing the sampling interval of the current imaging system, and updating the learning rate of the generative adversarial network, so as to adapt to the photovoltaic module defect detection needs under different operating environments.

[0014] Compared with the prior art, the advantages and beneficial effects of the present invention are: The present invention realizes a comprehensive characterization of surface and internal defects of photovoltaic modules through multi-modal data acquisition and fusion technology; The present invention realizes the efficient fusion of global and local features through the hierarchical attention mechanism technology, thus improving the accuracy of defect recognition; The present invention uses the time-series coupling prediction network technology to achieve accurate prediction of defect expansion trends, providing forward-looking support for preventive maintenance of photovoltaic modules. The present invention realizes adaptive adjustment of the detection system to complex operating environments through dynamic parameter optimization technology, thereby improving detection efficiency and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 This is a flow chart of feature fusion and generative adversarial network in the present invention; Figure 3 It is a defect detection and evaluation flow chart of the present invention; Figure 4 It is a flow chart of virtual working condition simulation and timing analysis in the present invention. DETAILED DESCRIPTION

[0016] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0017] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0018] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0019] like Figure 1 As shown, a photovoltaic module quality detection method based on artificial intelligence includes the following steps: Collect spectral polarization data, thermal imaging data, current imaging data and stereo depth data of photovoltaic modules, perform time synchronization and spatial calibration on all data, and generate preliminary multi-modal feature data; Preferably, the spectral polarization data is acquired by adjusting the polarization angle to capture the multi-angle spectral characteristics of the surface of the photovoltaic module, the thermal imaging data is acquired by recording the temperature distribution characteristics of the surface of the photovoltaic module, the current imaging data is acquired by applying a small current to the photovoltaic module and synchronously recording the current distribution characteristics, and the stereo depth data is acquired by generating the three-dimensional morphological characteristics of the component surface through multi-angle ray tracing.

[0020] Spectral polarization data collection principle and application Principle: Spectral polarization imaging technology uses the polarization characteristics of light to capture the reflection spectrum of different angles on the surface of photovoltaic modules by changing the polarization angle of the incident light. The interaction between polarized light and the surface structure and optical properties of the material can reveal subtle differences in surface cracks, scratches, and coating defects. By adjusting the polarization angle (such as 0°, 45°, 90°, etc.) and the incident direction, the full-angle spectral characteristics of the surface can be obtained to form a set of polarization spectral data.

[0021] Implementation: In the present invention, an automated polarization angle adjuster is used to control the angle change of the polarizer, and a spectral camera is used to record the reflected spectrum intensity at each polarization angle. The generated data can be expressed by the following formula:

[0022] in, Represents a certain polarization angle and wavelength The spectral intensity under is the reflectivity; The captured spectral polarization data can accurately identify the cracks and scratches on the surface of the component, especially the subtle differences in optical properties that are difficult to detect with the naked eye.

[0023] Thermal imaging technology records the temperature distribution on the surface of photovoltaic modules in a non-contact manner, and uses the local temperature rise phenomenon caused by internal cracks, hot spots or material aging during the operation of the module to generate a thermal image of the module. Infrared sensors capture infrared energy on the surface of materials based on the principle of thermal radiation and convert it into temperature distribution data. In the present invention, the thermal imaging camera collects infrared images in the operating state at a fixed frequency. By calculating the radiation brightness of each pixel , converted to temperature :

[0024] in, is the emissivity of the material, is the Stefan-Boltzmann constant, The infrared radiation brightness. Thermal imaging data can quickly identify abnormal heating areas caused by hidden cracks, hot spots or cold solder joints, and provide basic data for defect classification.

[0025] Current imaging applies a tiny current to the photovoltaic module and records its distribution characteristics on the surface of the module, which is used to capture local resistance anomalies caused by hidden cracks or cold solder joints. The current distribution is collected synchronously by the current sensor and the thermal infrared imaging device, and the integrity of the current path can be analyzed. The present invention uses a highly sensitive current sensor to work in conjunction with the thermal infrared imaging device to collect current imaging data under the condition of tiny current injection, and uses the following formula to calculate the resistance distribution:

[0026] in, is the pixel position The local resistance, is the voltage at that location, The generated current imaging data can accurately locate the area with abnormal local resistance, and combined with thermal imaging data analysis, it can effectively improve the reliability of hidden crack and cold solder joint detection.

[0027] Stereo depth data generates three-dimensional morphological features through ray tracing technology, and uses multi-angle light fields to capture component surface structure information. The light field sensor calculates the depth information of the component surface by recording the light distribution at different incident angles. The present invention adopts a multi-view light field imaging system, and the cameras with fixed angle intervals collect the light distribution on the component surface from different perspectives, and use triangulation to calculate the depth:

[0028] in, For depth, is the focal length of the camera, is the baseline distance, The stereo depth data can be used to identify macroscopic geometric defects on the component surface, such as warpage, dents, and cracks, and provide a geometric basis for constructing spectral-geometric models.

[0029] Example: In a certain actual application scenario, the photovoltaic module has power attenuation after long-term operation, and no obvious abnormality is seen on the surface by naked eyes. The following tests are carried out through the above data collection technology: Spectral polarization data was used to identify surface crack areas, and it was found that the spectral reflectivity anomalies were associated with geometric characteristics. Thermal imaging data was used to locate hot spot areas, verifying that there was local temperature rise at the crack points. Current imaging data was combined to further confirm the abnormal resistance areas, which were consistent with the thermal imaging data. Stereo depth data was used to find tiny surface warping, further verifying that the cracks affected the mechanical stability of the components.

[0030] Finally, based on the above data, multimodal feature data is generated to complete the defect assessment of PV modules and provide support for subsequent optimized operation and maintenance decisions.

[0031] Preferably, the time synchronization is based on timestamp technology, and a unified time base is used to mark the collection time of spectral polarization data, thermal imaging data, current imaging data and stereo depth data; the spatial calibration uses a feature point matching algorithm to align the thermal imaging data with the abnormal areas in the current imaging data, and maps the aligned results to the three-dimensional coordinate space of the stereo depth data to generate preliminary multimodal feature data after spatial calibration.

[0032] Time synchronization is to align the acquisition time of multimodal data to a unified time reference to ensure the temporal consistency of data of different modalities, thereby avoiding timing errors caused by acquisition delays or hardware clock drift. The present invention uses timestamp technology to give each acquired data a high-precision time tag, so that multimodal data can accurately correspond to the component status at the same time point.

[0033] A unified high-precision time server is used to align the internal clock of each sensor (including spectral camera, thermal imager, current sensor and light field sensor) with the time reference using NTP (Network Time Protocol). During data collection, a timestamp is used to record the collection time of each frame of data. For example, the collection time of spectral data is marked as , the thermal imaging data is annotated as , by comparing the time difference , to determine whether the synchronization conditions are met. Through time synchronization, it is ensured that the spectrum, heat, current and stereo depth data reflect the real state of the photovoltaic module at the same time point, providing accurate time series association for subsequent fusion analysis.

[0034] Spatial calibration is to align the spatial coordinate systems of different modal data to a unified three-dimensional coordinate system to ensure the consistency of the data in space. The present invention adopts a feature point matching algorithm to first match the position of abnormal areas in thermal imaging data and current imaging data, and then maps the alignment results to the three-dimensional coordinate space of stereo depth data to generate spatially consistent multi-modal feature data.

[0035] Through spatial calibration, data of different modes are aligned in space, which can accurately identify the spatial correlation between hot spots, resistance anomalies and three-dimensional geometric features of photovoltaic modules, laying a spatial foundation for multi-modal feature fusion.

[0036] like Figure 2 As shown, based on the preliminary multimodal feature data, a spectral-geometric model and a multi-field coupling model are constructed to generate model data respectively; the generated model data are fused through a feature mapping algorithm, and input into a generative adversarial network for feature abstraction to obtain key deep feature data; The spectral-geometric model combines spectral polarization data with stereo depth data to achieve a comprehensive characterization of surface cracks, coating defects and geometric anomalies of photovoltaic modules. The multi-field coupling model combines thermal imaging data with current imaging data to simulate the interaction between the stress field and the electric field inside the module, thereby reflecting the internal characteristics of the module under operation.

[0037] The temperature distribution in the thermal imaging data is combined with the resistance distribution in the current imaging data using tensor decomposition technology to generate a multi-field tensor. The interaction characteristics between the stress field and the electric field are calculated using the finite element analysis method to generate multi-field coupling model data.

[0038] The spectral-geometric model can comprehensively characterize the optical properties and geometric features of the surface of photovoltaic modules, while the multi-field coupling model reveals the thermal-electric coupling relationship inside the module, providing comprehensive data support for defect detection.

[0039] The feature mapping algorithm fuses the spectral-geometric model data with the multi-field coupling model data by constructing a cross-mapping matrix between multimodal features, thereby generating a unified multimodal feature representation.

[0040] The Generative Adversarial Network (GAN) consists of a generator and a discriminator. The generator generates abstract feature data through optimization, and the discriminator is used to judge the quality of the generated features, thereby achieving deep feature abstraction of multimodal fusion data.

[0041] The generative adversarial network performs deep abstraction of multimodal data to obtain key deep feature data. These features can accurately reflect the defect characteristics of photovoltaic modules and significantly improve the accuracy of subsequent detection and evaluation.

[0042] Preferably, the feature mapping algorithm realizes dynamic mapping of spectral-geometric model data by calculating the correlation matrix between spectral reflectance characteristics and geometric morphology, and extracts the correlation characteristics of the internal stress field and electric field of the component by performing multi-field tensor decomposition on thermal imaging data and current imaging data, and jointly models the correlation characteristics to generate fusion model data after feature mapping.

[0043] Spectral reflectance reflects the optical changes of the surface coating, cracks or scratches of photovoltaic modules, while geometric morphology reveals macroscopic characteristics such as surface deformation or warping. By calculating the correlation matrix between spectral reflectance and geometric morphology, the coupling relationship between spectral and geometric features can be dynamically mapped to reveal the comprehensive manifestation of cracks or surface defects. The specific implementation steps include: Generate spectral reflectance matrix for spectral polarimetric data ,in is the two-dimensional coordinate, is the wavelength; Calculate the 3D shape coordinates of components using stereo depth data ; Constructing a correlation matrix between spectral reflectance characteristics and geometric morphology :

[0044] in, It is the surface characteristic function of the geometric point, reflecting the local changes in the surface morphology.

[0045] The incidence matrix Mapping to spectral-geometric model data , complete dynamic mapping. Through dynamic mapping, the spectral characteristics and geometric morphology are expressed in a unified manner, providing a comprehensive basis for analyzing the multi-modal characteristics caused by surface cracks or warping of photovoltaic modules.

[0046] Thermal imaging data reflects the temperature distribution characteristics of the component surface, and current imaging data reflects the internal resistance distribution characteristics. Multi-field tensor decomposition technology can extract the coupling relationship between these two modes and reveal the interactive characteristics of heat and electric fields, such as the combined effect of hot spots caused by local hidden cracks and current anomalies. The specific implementation steps include: Using thermal imaging data and current imaging data Constructing the initial tensor :

[0047] Perform principal component decomposition on the tensor and extract the main feature matrix :

[0048] in, and are the left and right singular value matrices of the tensor, respectively. is a diagonal matrix of singular values; Main feature matrix Represent the thermal-electric field interaction characteristics and convert them into multi-field coupling model data Multi-field tensor decomposition can accurately extract the common characteristics of hot spots and resistance anomalies, providing key data support for analyzing internal hidden cracks or cold solder joints.

[0049] Through the joint modeling technology of the feature mapping algorithm, the spectral-geometric model data and the multi-field coupling model data are comprehensively expressed to generate fusion model data to uniformly characterize the surface and internal defect characteristics of the components.

[0050] Using spectral-geometric model data and multi-field coupled model data , construct the feature mapping matrix :

[0051] in, Represents the similarity calculation function, which is used to measure the correlation between the two model data; Generate fusion model data based on the weight of the feature mapping matrix :

[0052] The fusion model data comprehensively integrates the surface features and internal characteristics of photovoltaic modules, and can uniformly analyze surface cracks, coating defects and internal thermal-electric interaction characteristics, providing optimized input for subsequent deep feature abstraction.

[0053] Example: During the operation inspection of a photovoltaic module, it was found that the optical reflection was abnormal on the surface, and hot spots and resistance abnormalities appeared in local areas. The following operations were completed through the feature mapping algorithm: Constructing spectral-geometric model data , analyze surface hidden cracks and warping areas; Extract the thermal-electric interaction characteristics of thermal imaging data and current imaging data through multi-field tensor decomposition to generate multi-field coupling model data ; The above two types of model data are jointly modeled using the feature mapping matrix to generate fusion model data ; Analyze the fusion model data, identify the key defect areas of PV modules, verify the interaction between surface cracks and internal hot spots and resistance anomalies, and finally provide repair plan recommendations.

[0054] Preferably, the fusion model data is input into a generative adversarial network, which includes a generator and a discriminator. The generator generates deep abstract features through back-propagation optimization, and the discriminator optimizes the feature data output by the generator by judging the difference between the generated features and the real features, thereby generating enhanced key deep feature data.

[0055] Generative Adversarial Network (GAN) is a dual network structure in which the generator is responsible for generating feature representations close to the real features, and the discriminator is responsible for distinguishing between the generated features and the real features, thereby improving the quality of the generated features through adversarial training.

[0056] In the present invention, the input of GAN is the fusion model data, and the goal is to abstract key deep features from it and enhance the characterization ability of photovoltaic module defects. Through the collaborative training of the generator and the discriminator, the generator optimizes the accuracy of the generated features, and the discriminator optimizes the evaluation criteria for the generated features.

[0057] Generator Design: Input Fusion Model Data , extract multi-level features through convolutional neural network (CNN). Use back propagation algorithm to optimize the generated features The goal is to make the generated features as close to the real features as possible The objective function of generating features is defined as:

[0058] in, is the loss function of the generator, is the quadratic norm, which is used to measure the difference between the generated features and the real features.

[0059] Discriminator design: input is generated features and real features , the similarity between the two is evaluated through the discriminant network. The goal of the discriminator is to maximize the difference between the generated features and the real features, and the optimization objective function is:

[0060] in, is the loss function of the discriminator, Represents the output of the discriminative network.

[0061] The generator and discriminator are continuously improved through adversarial optimization: the generator aims to deceive the discriminator and generate data that is closer to the real features; the discriminator aims to distinguish between real features and generated features and strengthen the judgment criteria for generated features. Finally, after multiple rounds of iterations, the quality of generated features reaches the ideal standard and key deep feature data is obtained. .

[0062] A dynamic learning rate adjustment strategy is adopted to optimize the learning efficiency of the generator and discriminator according to the gradient changes during the training process; Batch Normalization technology is introduced to stabilize the training process and prevent mode collapse.

[0063] Through generative adversarial networks, deep multimodal features can be abstracted from fusion model data. These features have higher semantic expression capabilities and can accurately characterize the complex patterns of surface and internal defects of photovoltaic modules. The deep feature data optimized by the generator removes the noise and redundant information that may exist in the multimodal data, and strengthens the expression of the key characteristics of the defects. The evaluation criteria optimized by the discriminator improve the authenticity and reliability of the generated features, providing higher quality input data for subsequent defect detection and trend prediction.

[0064] In a photovoltaic module quality inspection task, the collected fusion model data M is fused , perform the following steps: Will Input generator, extract multi-level features through convolutional network, and generate feature representation M initially ; Will With real characteristics The discriminator evaluates the quality of the generated features and returns the loss value. ; The back-propagation algorithm is used to optimize the parameters of the generator and the discriminator, so that the generated features gradually approach the real features. After multiple rounds of training, the final key deep feature data is generated , which accurately characterizes the component surface cracks, coating defects and internal thermal-electric field interaction characteristics; use The data successfully identified the association between a hidden crack on the component surface and a local hot spot inside, providing a direct basis for subsequent maintenance and optimization.

[0065] like Figure 3 As shown, the key deep feature data is input into the hierarchical attention mechanism for global and local feature fusion to generate fused defect feature data; the defect region of the fused defect feature data is located using a region detection algorithm, and the defect region is classified and the severity is evaluated to generate defect classification data and defect evaluation data; Hierarchical Attention Mechanism (HAM) is a feature fusion technology based on deep learning. The global attention module focuses on the overall feature pattern of photovoltaic modules to detect macroscopic anomalies, such as large-scale coating wear or deformation; the local attention module focuses on the detailed features of local abnormal areas to capture small defects such as hidden cracks and hot spots. The two are fused in a hierarchical weighted manner to achieve complementary expression of global and local information.

[0066] Through the hierarchical attention mechanism, global features and local features can be efficiently fused, making the defect feature expression more comprehensive and accurate, significantly improving the accuracy of subsequent defect area positioning and classification.

[0067] The region detection algorithm divides the fused defect feature data into regions through a convolutional neural network (CNN), identifies areas where defects may exist, and determines the specific location of the defects through bounding box positioning technology. At the same time, the classification module is used to evaluate the category and severity of each defect area.

[0068] The regional detection algorithm can not only accurately locate the position of the defective area, but also comprehensively evaluate the defect type and severity, providing a reliable basis for fault diagnosis of PV modules.

[0069] Preferably, the hierarchical attention mechanism performs weighted processing on the overall spectral distribution, thermal distribution and current characteristics of the photovoltaic module through a global feature extraction module, and at the same time focuses on the spectral anomaly points, hot spot locations and current anomaly points in the defect area through a local feature extraction module, so as to achieve efficient fusion of global features and local features and generate fused defect feature data.

[0070] Hierarchical Attention Mechanism (HAM) is a deep learning method that combines global and local features. The global feature extraction module is responsible for capturing the overall characteristic distribution pattern of photovoltaic modules, such as spectral reflectance uniformity, temperature gradient, etc.; the local feature extraction module focuses on abnormal points in specific areas, such as spectral anomalies caused by hidden cracks, high temperature areas of hot spots, current abnormal points, etc. By fusing global and local features through a hierarchical structure, HAM can generate more accurate defect feature data.

[0071] The global feature extraction module uses a self-attention mechanism to weight the spectral, thermal, and current data of the entire photovoltaic module to capture the distribution pattern of the overall characteristics of the module.

[0072] Input key deep feature data , generate the query vector through linear transformation , key vector , value vector :

[0073]

[0074]

[0075] in, , , is the trainable weight matrix.

[0076] Calculate the attention weights:

[0077] in, is the dimension of the key vector.

[0078] Weighted global features :

[0079] The local feature extraction module uses sliding window technology to focus on the defective area of ​​the component and extract high-dimensional features of spectral anomalies, hot spot locations, and current anomalies. Fixed window size Divide into several sub-areas , calculate the local weighted features for each sub-region:

[0080] in, is the local weight matrix, It is a bias term. It focuses on the abnormal point features in the defect area and enhances the feature expression through high weight.

[0081] Generate fused defect feature data by weighted fusion of global and local features , which comprehensively expresses the overall characteristics and local abnormalities of the component. Perform weighted summation of global features and local features:

[0082] in, , dynamically adjust the weights to adapt to different detection requirements.

[0083] Through the hierarchical attention mechanism, global features can capture the overall optical, electrical and thermodynamic characteristics of the component and identify macroscopic abnormal areas; local features can accurately capture the detailed characteristics of the defect area, such as the spectral changes of hidden cracks, the high temperature characteristics of hot spots and the local interruption of the current path. The fused defect feature data finally generated has higher expressive power and provides optimized input for subsequent defect location, classification and severity assessment.

[0084] In a photovoltaic module detection task, the following steps are completed through a hierarchical attention mechanism: The component's spectral, thermal imaging and current imaging data were input into the global feature extraction module to obtain the global spectral distribution, temperature gradient and current path characteristics. The results showed that the spectral reflectivity in the lower left corner of the component was widely offset, and the temperature in this area was slightly higher than normal.

[0085] Local feature extraction is performed on the lower left corner of the component, focusing on the spectral anomalies of hidden cracks and the high-temperature positions of hot spots, and multiple groups of local high-weight features are extracted.

[0086] The global features and local features are weighted and fused to generate fused defect feature data. The analysis found that the spectral changes in the hidden crack area are highly correlated with the hot spot position, which is the key defect area.

[0087] Subsequent regional inspections and assessments confirmed that hidden cracks in the lower left corner of the PV module had caused local overheating, and it was recommended to replace the module or strengthen monitoring.

[0088] Preferably, the region detection algorithm performs multi-level feature extraction on the fused defect feature data, first roughly determines the region where the defect exists through an edge detection algorithm, then performs fine-grained boundary optimization on the defect region through an improved region extraction algorithm, and finally outputs the precise position coordinates and region boundary information of the defect region.

[0089] Multi-level feature extraction is a regional detection method that refines features layer by layer. First, the edge detection algorithm is used to quickly identify areas in photovoltaic modules that may have defects, and the location and range of these areas are roughly determined; then, an improved regional extraction algorithm is used to perform a refined analysis of these areas, optimize boundary information, and output precise coordinates of defective areas. This method combines the advantages of coarse positioning and fine optimization, and can significantly improve detection accuracy while ensuring detection efficiency.

[0090] The edge detection algorithm identifies the edges of abnormal areas of spectral, thermal or current characteristics in photovoltaic modules by calculating the grayscale change or gradient change of the image. Input edge detection module to calculate the gradient strength of each pixel :

[0091] in, is the pixel intensity value, and is the horizontal and vertical gradient.

[0092] Setting the Threshold , the gradient strength is greater than The pixel points are marked as edge points to generate a preliminary defect area mask map . Quickly identify areas where defects may exist in PV panels and provide input for subsequent refined processing.

[0093] The improved region extraction algorithm refines the boundary information of the rough region by optimizing the defect region boundary to ensure the accuracy of region positioning. Optimize and dynamically adjust the region boundaries based on the spectral reflectance, temperature distribution and current path characteristics within the region; Calculate the boundary point set , the optimized boundary is updated by the following formula:

[0094] in, is the characteristic distribution in the region, It is the distribution of features outside the region.

[0095] Map the optimized boundary information back to the original image and output the precise coordinates of the defect area and boundary information The improved region extraction algorithm can refine the boundaries of rough regions and ensure the accuracy of defect region positioning, which is particularly suitable for the detection of complex defects in photovoltaic modules.

[0096] Through the multi-level feature extraction method, the edge detection algorithm can quickly identify the location and range of the defect area, saving the computational cost for subsequent refinement processing; the improved region extraction algorithm accurately optimizes the boundaries of the rough region, effectively improving the positioning accuracy of the defect area. The final output of the defect area coordinates and boundary information provides reliable data support for defect classification and severity assessment.

[0097] In an embodiment, in a component detection task of a photovoltaic power station, possible hidden cracks and hot spots in the components are identified through a regional detection algorithm, and the following operations are performed: Fusion of defect feature data Input edge detection module to calculate gradient strength The results show that there is a high gradient area with obvious edges in the middle of the component, which is marked as a preliminary defect area.

[0098] The preliminary mask The input area extraction module optimizes the boundary, eliminates some redundant boundary points, and refines the boundary range of the real defect. Finally, an accurate defect area boundary is generated. .

[0099] The output defect area coordinates and boundary information show that there are significant anomalies in the spectral reflectivity, temperature distribution and current path of the area, confirming the coexistence of hidden cracks and hot spots, which has a significant impact on the power output of the component.

[0100] like Figure 4 As shown, the model data is loaded into the digital twin platform, the operating conditions are simulated, and virtual detection data is generated; the virtual detection data is subjected to time series analysis to predict the expansion trend of defects and optimize the detection system parameters.

[0101] Digital twin technology reproduces the operating status and behavior of physical objects in a virtual environment through dynamic mapping between virtual models and actual physical systems. In the present invention, the spectral-geometric model data and multi-field coupling model data are loaded into the digital twin platform to build a virtual model of the photovoltaic module, and the actual operating conditions are simulated based on the input environment and operating parameters (such as temperature, humidity, light intensity, etc.). The digital twin platform can reflect the operating characteristics of photovoltaic modules under different working conditions in real time, provide high-precision defect evolution data, and lay the foundation for subsequent analysis.

[0102] Time series analysis is a method based on mining the changing rules of time series data. It predicts the expansion trend of defects by analyzing the dynamic changes of defect areas in virtual inspection data over time. The present invention processes virtual inspection data through a time series coupling prediction network (TCP-Net) to obtain the future development path and expansion rate of defects. Time series analysis can accurately predict the expansion trend of defects, provide forward-looking data support, and help formulate preventive maintenance strategies.

[0103] Based on the prediction results of defect expansion trends, key parameters of the detection system (such as acquisition equipment resolution, feature extraction algorithm hyperparameters, etc.) are adjusted to optimize system performance to adapt to different operating environments and defect characteristics. The optimized detection system can adapt to complex operating environments more efficiently, and provide early warnings before defects occur, reducing the risk of component failure.

[0104] Preferably, the timing analysis includes time series decomposition of virtual detection data, extracting dynamic change patterns of defect areas, including hot spot expansion rate, resistance change rate and spectral reflectivity fluctuations, and extrapolating defect expansion trends based on a timing coupled prediction network to generate defect trend prediction data.

[0105] Time series analysis aims to reveal the evolution pattern of defect area features over time by mining the dynamic changes of virtual detection data. The present invention uses the time series decomposition method to decompose the multimodal data of the defect area (such as hot spot temperature, resistance change and spectral reflectance fluctuation) into three parts: long-term trend, periodic fluctuation and random fluctuation, and extract the dynamic change pattern of each feature. Subsequently, the extracted time series features are modeled using the Temporal Coupling Prediction Network (TCP-Net) to predict the defect expansion trend.

[0106] Time series decomposition Decomposition into long-term trends , Cyclical Fluctuations and random fluctuations :

[0107] This decomposition can help identify long-term expansion trends and short-term fluctuation characteristics of defects.

[0108] Virtual test data The features of each defect area are sampled to generate a time series feature matrix:

[0109] in , , Represents the change of spectral reflectance, hot spot temperature and resistance value over time; decomposed using additive model to extract long-term trend and cyclical fluctuations , and filter random fluctuations through denoising methods .

[0110] The expansion rate of the hot spot is calculated by analyzing the change of the hot spot area over time. :

[0111] in, is the change in the hot spot area, For the time interval.

[0112] Resistance change rate, by analyzing the change of resistance value, calculate the resistance change rate :

[0113] in, is the resistance change. Spectral reflectivity fluctuation, calculate the standard deviation of spectral reflectivity of defect area , quantifying the intensity of its fluctuations.

[0114] The time series coupling prediction network (TCP-Net) is used to combine multimodal time series features to predict the expansion trend of the defect area in the future. TCP-Net extracts the spatial correlation of time series features through convolutional networks and models time series correlation through recurrent neural networks (RNNs).

[0115] Will Enter TCP-Net and calculate future time points The characteristic values ​​of:

[0116] in, is the predicted future feature matrix. The prediction results include the expansion range of the hot spot area, the resistance change trend and the spectral reflectivity fluctuation intensity.

[0117] Through time series analysis, the dynamic change pattern of the defect area can be accurately extracted, such as the hot spot expansion rate, resistance change rate and spectral reflectivity fluctuation, so as to identify the development trend and potential risks of the defect. Combined with the time series coupling prediction network, the future evolution of the defect can be accurately predicted and forward-looking maintenance suggestions can be provided.

[0118] In the embodiment, in the component detection of a photovoltaic power station, the following operations are implemented through timing analysis and prediction: Collecting virtual inspection data , the spectral reflectance, hot spot temperature and resistance value of the defect area are decomposed in time series to extract long-term trends and periodic fluctuations. The results show that the temperature of the hot spot area shows an upward trend, the resistance value gradually increases, and the spectral reflectance fluctuation increases.

[0119] Calculate the hot spot expansion rate , resistance change rate The standard deviation of spectral reflectance is .

[0120] TCP-Net predicts that the area of ​​the hot spot will increase by 50%, the resistance value will increase by 20%, and the spectral reflectance fluctuation will increase by 10% in the next 4 hours.

[0121] Based on the prediction results, it is recommended to cool the hot spot area and conduct further inspections on related components to prevent further expansion of hidden cracks.

[0122] Preferably, the optimization of detection system parameters includes adjusting the acquisition frequency of the spectral imaging device, dynamically adjusting the temperature sensitivity threshold of the thermal imaging device, optimizing the sampling interval of the current imaging system, and updating the learning rate of the generative adversarial network, so as to adapt to the photovoltaic module defect detection needs under different operating environments.

[0123] The acquisition frequency of spectral imaging equipment determines the timeliness and precision of data acquisition. In dynamic conditions (such as when photovoltaic modules are exposed to strong light or shadows change rapidly), a higher acquisition frequency can capture rapidly changing spectral reflectance characteristics, while in stable conditions, the frequency can be appropriately reduced to reduce data redundancy and processing burden.

[0124] Dynamically adjust the acquisition frequency according to the operating environment of the photovoltaic components :

[0125] in, is the threshold of the light change rate. Dynamically adjusting the acquisition frequency can balance the timeliness and efficiency of data acquisition, ensuring that the spectrum changes of photovoltaic modules are captured in a dynamic environment, while reducing the cost of data storage and processing.

[0126] The temperature sensitivity threshold of thermal imaging equipment determines the device's ability to detect temperature differences. For small hot spots or slight temperature rises caused by hidden cracks, higher sensitivity is required; in environments with large temperature fluctuations, the sensitivity can be appropriately reduced to avoid false alarms.

[0127] Dynamically adjust temperature sensitivity threshold :

[0128] in, is the threshold of the ambient temperature change rate. Through dynamic adjustment of sensitivity, tiny hot spots can be captured more accurately, while avoiding false alarms caused by ambient temperature fluctuations, and improving the accuracy of thermal imaging detection.

[0129] The sampling interval of the current imaging system determines the temporal resolution of the current distribution characteristics. In the case of rapid current changes caused by hidden cracks or poor welding, a shorter sampling interval is required to capture transient signals; while under steady-state conditions, a longer sampling interval can reduce redundant data.

[0130] Optimize sampling interval :

[0131] in, is the threshold of the current change rate. The optimization of the sampling interval ensures that key features are captured under dynamic current change conditions, while improving the efficiency and accuracy of data acquisition.

[0132] The learning rate of the generative adversarial network (GAN) directly affects the optimization speed and convergence of the feature abstraction process. A higher learning rate can speed up the training of the network, but may cause model oscillation; a lower learning rate helps the model to converge stably, but the training speed is slower. The present invention dynamically updates the learning rate according to the training error.

[0133] Dynamically update the learning rate based on the Learning Rate Scheduler (LRS) :

[0134] in, is the initial learning rate, is the learning rate decay coefficient, is the number of training steps. Dynamic learning rate adjustment can balance training speed and stability, ensuring that the generative adversarial network can efficiently extract key deep features and improve the robustness of defect detection.

[0135] By optimizing the key parameters of the detection system, the present invention can flexibly adapt to different operating environments, improve the accuracy of data collection and feature extraction, and enhance detection efficiency and precision. Dynamically adjust the spectral imaging frequency, thermal imaging sensitivity, current sampling interval, and GAN learning rate, so that the system can respond to changes in the operating status of components in real time, optimize defect detection performance, and reduce false alarm and missed alarm rates.

[0136] In the embodiment, in a dynamic working condition detection task of a photovoltaic module, the system performs the following optimization operations: When the light intensity fluctuates greatly, the system increases the acquisition frequency from 30 frames / second to 50 frames / second to ensure that the details of the spectral changes are captured; under stable lighting conditions, the frequency is reduced to 20 frames / second to reduce redundant data.

[0137] In response to a weak hot spot appearing on a certain component, the system optimized the temperature sensitivity threshold from 0.1°C to 0.05°C, successfully capturing the early characteristics of the hot spot expansion.

[0138] Under the condition of rapid changes in resistance in the hidden crack area, the system shortens the sampling interval from 1 second to 0.5 seconds to accurately record the current fluctuation characteristics.

[0139] During the feature abstraction process, the initial learning rate is 0.01 and gradually decays to 0.001 as the number of training steps increases, ensuring that the model converges stably while being efficiently trained. The generated key deep feature data significantly improves the accuracy of defect detection.

[0140] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0141] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A photovoltaic module quality detection method based on artificial intelligence, characterized in that: The following steps are involved: Collect spectral polarization data, thermal imaging data, current imaging data and stereo depth data of photovoltaic modules, perform time synchronization and spatial calibration on all data, and generate preliminary multi-modal feature data; Based on the preliminary multimodal characteristic data, a spectral-geometric model and a multi-field coupling model are constructed to generate model data respectively; The generated model data is fused through the feature mapping algorithm and input into the generative adversarial network for feature abstraction to obtain key deep feature data; Inputting the key deep feature data into a hierarchical attention mechanism to perform global and local feature fusion to generate fused defect feature data; Using a region detection algorithm to locate the defect region of the fused defect feature data, and classify and evaluate the severity of the defect region to generate defect classification data and defect evaluation data; Loading the model data into the digital twin platform, simulating the operating conditions, and generating virtual detection data; The virtual detection data is subjected to time series analysis to predict the expansion trend of defects and optimize detection system parameters.

2. The method according to claim 1, characterized in that: The spectral polarization data is collected by adjusting the polarization angle to capture the multi-angle spectral characteristics of the surface of the photovoltaic module, the thermal imaging data is collected by recording the temperature distribution characteristics of the surface of the photovoltaic module, the current imaging data is obtained by applying a small current to the photovoltaic module and synchronously recording the current distribution characteristics, and the stereo depth data is generated by multi-angle ray tracing to generate three-dimensional morphological characteristics of the component surface.

3. The method according to claim 2, characterized in that The time synchronization is based on the time stamp technology, and a unified time reference is used to mark the acquisition time of the spectral polarization data, thermal imaging data, current imaging data and stereo depth data; The spatial calibration aligns the thermal imaging data with the abnormal areas in the current imaging data through a feature point matching algorithm, and maps the aligned result to the three-dimensional coordinate space of the stereo depth data to generate preliminary multimodal feature data after spatial calibration.

4. The method according to claim 1, characterized in that The feature mapping algorithm realizes dynamic mapping of spectral-geometric model data by calculating the correlation matrix between spectral reflectance characteristics and geometric morphology. At the same time, by performing multi-field tensor decomposition on thermal imaging data and current imaging data, the correlation features between the stress field and the electric field inside the component are extracted, and the correlation features are jointly modeled to generate fusion model data after feature mapping.

5. The method according to claim 4, characterized in that The fusion model data is input into a generative adversarial network, which includes a generator and a discriminator. The generator generates deep abstract features through back-propagation optimization, and the discriminator optimizes the feature data output by the generator by judging the difference between the generated features and the real features, thereby generating enhanced key deep feature data.

6. The method according to claim 1, characterized in that The hierarchical attention mechanism performs weighted processing on the overall spectral distribution, thermal distribution and current characteristics of photovoltaic modules through a global feature extraction module, and focuses on the spectral anomaly points, hot spot locations and current anomaly points in the defective area through a local feature extraction module, so as to achieve efficient fusion of global features and local features and generate fused defect feature data.

7. The method according to claim 6, characterized in that The region detection algorithm performs multi-level feature extraction on the fused defect feature data, firstly roughly determines the region where the defect exists through the edge detection algorithm, then performs fine-grained boundary optimization on the defect region through the improved region extraction algorithm, and finally outputs the precise position coordinates and region boundary information of the defect region.

8. The method according to claim 1, characterized in that: The timing analysis includes time series decomposition of virtual detection data, extraction of dynamic change patterns of defect areas, including hot spot expansion rate, resistance change rate and spectral reflectivity fluctuations, and extrapolation of defect expansion trends based on a timing coupled prediction network to generate defect trend prediction data.

9. The method according to claim 1, characterized in that: The optimization of detection system parameters includes adjusting the acquisition frequency of the spectral imaging device, dynamically adjusting the temperature sensitivity threshold of the thermal imaging device, optimizing the sampling interval of the current imaging system, and updating the learning rate of the generative adversarial network, so as to adapt to the photovoltaic module defect detection needs under different operating environments.

Images