Photovoltaic module and preparation method thereof

Abstract

The present application discloses a photovoltaic module and its preparation method, which first collects the surface image of the cell to be detected through a camera, and extracts the cell surface comparison standard template image from the background database, then uses a neural network model based on deep learning to extract the deep features of the cell surface of the cell surface image to be detected and the cell surface comparison standard template image, and further combines the spatial-semantic dual enhancement algorithm to strengthen the representation difference between the defective area and the non-defective area, so as to realize the pixel-level difference detection of the cell surface state. In this way, the detection accuracy of the cell surface state is improved, thereby effectively improving the overall quality and long-term reliability of the photovoltaic module, reducing the problem of reduced photoelectric conversion efficiency caused by cell surface defects, and providing strong support for the efficient and reliable development of the photovoltaic industry.

Description

Technical Field

[0001] The present application relates to the field of intelligent manufacturing, and more specifically, to a photovoltaic module and a manufacturing method thereof. Background Art

[0002] In the field of photovoltaic module preparation, accurate detection of the surface state of the cell is the core link to ensure the photovoltaic conversion efficiency and long-term reliability of the module. In the existing preparation process, traditional cell inspection methods often rely on manual visual inspection or simple automated optical inspection equipment, which have certain limitations.

[0003] Traditional visual inspection methods are highly dependent on the operator's experience level and are prone to subjective judgment errors, especially when dealing with subtle defects such as hidden cracks and micro scratches. The missed detection rate is high; in addition, existing inspection systems mostly use overall image similarity comparison and lack the ability to semantically analyze pixel-level differences, resulting in insufficient positioning accuracy of defect boundaries. It cannot meet the stringent requirements of the new generation of high-efficiency photovoltaic modules for the surface integrity of battery cells. This not only limits the improvement of the overall quality of photovoltaic modules, but also to a certain extent hinders the development of the photovoltaic manufacturing industry towards high efficiency and reliability.

[0004] Therefore, an optimized method for preparing photovoltaic modules is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a photovoltaic module and a method for preparing the same, which first collects the surface image of the cell to be inspected through a camera, and extracts the cell surface comparison standard template image from the background database, then uses a neural network model based on deep learning to extract the deep features of the cell surface of the cell surface image to be inspected and the cell surface comparison standard template image, and further combines the spatial-semantic dual enhancement algorithm to strengthen the representation difference between the defective area and the non-defective area, so as to achieve pixel-level difference detection of the cell surface state. In this way, the detection accuracy of the cell surface state is improved, thereby effectively improving the overall quality and long-term reliability of the photovoltaic module, reducing the problem of reduced photoelectric conversion efficiency due to cell surface defects, and providing strong support for the efficient and reliable development of the photovoltaic industry.

[0006] According to one aspect of the present application, a method for preparing a photovoltaic module is provided, which includes: testing battery cells; after determining that the test results of the battery cells are qualified, performing single-cell welding and series welding of the battery cells; stacking and laminating various photovoltaic module materials, wherein the photovoltaic module materials include battery cells, glass plates, back plates and EVA; side-sealing and framing the laminated photovoltaic module materials to obtain a photovoltaic module.

[0007] According to another aspect of the present application, a wind-solar power generation energy storage management system is provided, which includes: an information acquisition module, used to obtain energy storage parameters of a stored battery during the energy storage process, wherein the energy storage parameters include a charging voltage value, a charging current value and a battery temperature value; and an SOC value generation module, used to determine the SOC value of the stored battery based on the energy storage parameters.

[0008] Compared with the prior art, the photovoltaic module and its preparation method provided by the present application firstly collect the surface image of the cell to be detected by the camera, and extract the surface comparison standard template image of the cell from the background database, then use the neural network model based on deep learning to extract the deep features of the cell surface of the cell surface image to be detected and the surface comparison standard template image of the cell, and further combine the spatial-semantic dual enhancement algorithm to strengthen the representation difference between the defective area and the non-defective area, so as to realize the pixel-level difference detection of the surface state of the cell. In this way, the detection accuracy of the surface state of the cell is improved, thereby effectively improving the overall quality and long-term reliability of the photovoltaic module, reducing the problem of reduced photoelectric conversion efficiency caused by surface defects of the cell, and providing strong support for the efficient and reliable development of the photovoltaic industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 Flow chart of a method for preparing a photovoltaic module according to an embodiment of the present application.

[0011] Figure 2 Schematic diagram of data flow of a method for preparing a photovoltaic module according to an embodiment of the present application.

[0012] Figure 3 Flow chart of sub-step S1 of the method for preparing a photovoltaic module according to an embodiment of the present application.

[0013] Figure 4 Flow chart of sub-step S14 of the method for preparing a photovoltaic module according to an embodiment of the present application. DETAILED DESCRIPTION

[0014] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0015] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0016] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0017] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0019] In the technical solution of the present application, a method for preparing a photovoltaic module is proposed. Figure 1 Flow chart of a method for preparing a photovoltaic module according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the method for preparing a photovoltaic module according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the preparation method of the photovoltaic module according to the embodiment of the present application includes the steps of: S1, testing the battery cells; S2, after determining that the test results of the battery cells are qualified, performing single-piece welding of the battery cells and series welding of the battery cells; S3, stacking and laminating the various photovoltaic module materials, wherein the photovoltaic module materials include battery cells, glass plates, back plates and EVA; S4, side-sealing and framing the laminated photovoltaic module materials to obtain a photovoltaic module.

[0020] In particular, the S1, testing the battery cell. In the technical solution of the present application, testing the battery cell includes: mapping the pixel-based semantic difference of the battery cell surface state image features between the battery cell surface image to be tested and the battery cell surface comparison standard template image to obtain the test result. Specifically, Figure 3As shown, the S1 includes: S11, acquiring a surface image of a battery cell to be detected collected by a camera; S12, extracting a battery cell surface comparison standard template image from a background database; S13, extracting battery cell surface state image features of the battery cell surface image to be detected and the battery cell surface comparison standard template image to obtain battery cell surface state detection feature coding features and battery cell surface state template feature coding features; S14, performing battery cell surface state significance enhancement processing on the battery cell surface state detection feature coding features and the battery cell surface state template feature coding features to obtain enhanced battery cell surface state detection feature coding features and enhanced battery cell surface state template feature coding features; S15, obtaining a detection result based on the battery cell surface state semantic difference between the enhanced battery cell surface state detection feature coding features and the enhanced battery cell surface state template feature coding features.

[0021] Specifically, the S11, obtains the surface image of the battery cell to be tested collected by the camera; extracts the battery cell surface comparison standard template image from the background database. The surface image of the battery cell to be tested is the digital representation data of the microstructure of the battery cell surface collected by a high-resolution industrial camera under specific lighting conditions, such as the physical morphology and material property information of the battery cell surface. The high-resolution image captured by the camera can reflect all the features of the battery cell surface in detail, including potential defects and contamination. In this way, an objective record of the surface state of the battery cell can be achieved, which ensures that each battery cell to be evaluated can be accurately analyzed, thereby improving the accuracy and reliability of the overall detection process.

[0022] Specifically, the S12 extracts the standard template image for comparing the surface of the battery cell from the background database. The background database is a data center that stores a large number of strictly screened and labeled high-quality battery cell surface images. These images represent the ideal state of the battery cell under different conditions and are the basis for comparative analysis. By extracting the standard template image for comparing the surface of the battery cell from the background database, the system can provide a reliable reference standard for the battery cell to be inspected, so as to accurately identify whether there are defects on its surface, thereby greatly improving the flexibility and adaptability of the inspection process.

[0023] Specifically, the S13 extracts the cell surface state image features of the cell surface image to be detected and the cell surface comparison standard template image to obtain the cell surface state detection feature coding features and the cell surface state template feature coding features. In the embodiment of the present application, first, the cell surface image is enhanced to obtain the cell enhanced surface image; it should be understood that due to the dynamic light fluctuations of the production line environment, the reflection interference of the glass plate and the noise of the camera itself, the original image often has local overexposure, low contrast area and high-frequency noise interference, resulting in poor image quality, and it is difficult to effectively capture the subtle features of microscopic defects on the cell surface. In a specific example of the present application, histogram equalization and Gaussian filtering are used to reduce noise on the surface image of the cell. Specifically, histogram equalization effectively stretches the grayscale distribution of low-contrast areas in the surface image of the cell by redistributing pixel grayscale values, so that the grayscale difference between weak texture defects such as hidden cracks and micro scratches and normal areas is amplified, solving the problem of feature blurring caused by uneven lighting or surface reflection. This global contrast optimization method can highlight the outline of subtle cracks that are difficult to detect with traditional visual inspection, and provide richer texture gradient information for deep learning models. Gaussian filtering denoising targets the salt and pepper noise and high-frequency interference generated during the acquisition process of industrial cameras. Through weighted average convolution kernel operations, it eliminates the interference of random noise on feature encoding while retaining the sharpness of defect edges, preventing the model from misjudging image sensor noise as physical defects. Through the synergistic effect of histogram equalization and Gaussian filtering noise reduction on the surface image of the cell, a dual mechanism of "enhancing effective signals and suppressing invalid interference" can be formed, so that the enhanced surface image of the cell has clear defect morphological characteristics and reduces artifact interference caused by complex reflective textures, laying a precise visual analysis foundation for subsequent deep feature extraction based on DenseNet, and ultimately achieving pixel-level precise positioning of surface contamination and structural defects of photovoltaic cells.

[0024] Next, the enhanced surface image of the cell is passed through a cell surface state image feature extractor based on the DenseNet model to obtain a cell surface state detection feature encoding feature map as a cell surface state detection feature encoding feature; it should be understood that the traditional CNN model, due to the unidirectionality of inter-layer feature transmission, leads to insufficient fusion of shallow texture details and deep semantic information, and it is difficult to capture the cross-scale correlation features between hidden cracks and substrates. Therefore, in order to break through the bottleneck of detail loss caused by inter-layer feature transmission attenuation in traditional convolutional neural networks, in the technical solution of the present application, the enhanced surface image of the cell is passed through a cell surface state image feature extractor based on the DenseNet model to obtain a cell surface state detection feature encoding feature map as a cell surface state detection feature encoding feature. In this process, the DenseNet model realizes cross-layer feature reuse and gradient optimization through a dense connection mechanism. Its core working principle is that the feature map output by each layer of convolution is directly connected to all subsequent layers to form a dense feature transmission path, so that the network can simultaneously capture the local detail features and global structural associations of the cell surface texture. Specifically, the densely connected architecture of DenseNet reuses fully connected cross-layer features, so that the multi-scale representation of microscopic defects (such as the sub-pixel edge gradient of hidden cracks and the reflectivity distribution of contaminated areas) can be continuously strengthened between each layer of the network, forming a progressively enhanced expression of defect features, thereby overcoming the problem of feature expression fragmentation in traditional models in photovoltaic surface detection. In this way, the model's ability to identify defects of different sizes and types can be improved, providing a solid foundation for subsequent comparative analysis.

[0025] Similarly, the cell surface comparison standard template image is passed through a cell surface state image feature extractor based on a DenseNet model to obtain a cell surface state template feature encoding feature map as a cell surface state template feature encoding feature. In the technical solution of the present application, in order to extract high-quality, deep-level feature information from the cell surface comparison standard template image as a basis for subsequent comparison with the cell surface image to be detected, the cell surface comparison standard template image is passed through a cell surface state image feature extractor based on a DenseNet model to obtain a cell surface state template feature encoding feature map. In this process, the dense connection mechanism unique to the DenseNet model constructs a multi-scale semantic representation system of the standard template through cross-layer feature reuse. Specifically, the feature map output by each convolutional layer in the DenseNet model establishes a direct connection channel with all subsequent layers to form a dense feature transfer network, so that the prior knowledge such as the periodic structure of the grid line of the battery cell and the optical characteristics of the surface passivation layer in the standard template image can be continuously iterated and optimized between different receptive field levels. Among them, the DenseNet model captures micron-level geometric topological features, such as the sharpness of the electrode edge, through the shallow convolution kernel, and analyzes the macroscopic optical reflection distribution pattern through the deep network, and finally forms a composite template feature encoding that integrates geometric constraints and physical characteristics. In this way, the extracted battery cell surface state template feature encoding features not only retain the core parameters of the process design specification (such as the hidden crack tolerance threshold), but also have the ability to adapt to the fluctuations of the production line environment, which improves the recognition accuracy of the battery cell surface compared with the standard template image, and builds a reference system with process tolerance capabilities for subsequent pixel-level difference detection.

[0026] Specifically, the S14 is to perform cell surface state significance enhancement processing on the cell surface state detection feature coding feature and the cell surface state template feature coding feature to obtain enhanced cell surface state detection feature coding feature and enhanced cell surface state template feature coding feature. It should be understood that after the cell surface state feature coding is extracted through the DenseNet model, although its coding feature already contains multiple layers of abstract information, subtle defects such as hidden cracks and micro scratches are still easily masked by the strong response of normal texture in the feature space, and the strong interference signal generated by the periodic grid line structure on the cell surface and the weak characteristic response of the hidden crack are easy to form feature confusion in traditional feature coding. Therefore, in order to enhance the signal-to-noise ratio of the defect response in the detection feature coding and suppress the periodic interference of the grid line texture, in the technical solution of the present application, the cell surface state detection feature coding feature and the cell surface state template feature coding feature are subjected to cell surface state significance enhancement processing to obtain enhanced cell surface state detection feature coding feature and enhanced cell surface state template feature coding feature. That is, by introducing the spatial-semantic joint explicit modeling mechanism to break through the limitations of single feature dimension analysis, a dual-path feature modulation system is constructed. Specifically, the sparse reference feature set constructed by random scanning can dynamically associate the typical pattern of the microstructure on the surface of the battery cell with the characteristics of the current detection area; the Poincare distance is used to calculate the spatial modulation matrix, which can effectively capture the hierarchical distribution characteristics of surface defects of the battery cell in the hyperbolic space, such as the tree-like expansion trend of hidden cracks or the radial gradient change of micro scratches. This nonlinear distance measurement method is more suitable for the complex geometric characteristics of defect morphology in industrial images than the traditional Euclidean distance; and the construction of the implicit semantic association coding matrix, By exploring the abstract connection between the features to be enhanced and the reference features beyond the pixel level (such as the potential mapping of the conductivity attenuation pattern of the contaminated area and the electrical performance indicators of the standard template), the feature enhancement process can break through the interference of surface texture similarity and focus on the essential attribute differences that affect the photoelectric conversion efficiency; in particular, in the technical solution of the present application, a double-layer mask modulation mechanism is adopted. The first-level semantic mask prioritizes the screening of reference feature patterns with functional associations (such as local oxidation features that are strongly correlated with the hot spot effect), and the second-level spatial mask performs spatial weight calibration for the defect diffusion path. This hierarchical screening strategy effectively avoids the defect that the traditional attention mechanism is easily misled by environmental noise in industrial scenarios. In this way, not only the detection accuracy of the surface state of the battery cell is improved, but also the gradient mutation response to the defect boundary is enhanced. In a specific example of the present application, such as Figure 4As shown, the S14 includes: S141, extracting the channel feature vector of the (i, j)th pixel position from the cell surface state detection feature coding feature map as the cell surface state channel feature vector to be enhanced; S142, performing n random scans on the cell surface state detection feature coding feature map to obtain n cell surface state channel feature vectors as a sparse set of cell surface state reference feature vectors; S143, based on the sparse set of cell surface state reference feature vectors, performing spatial explicit modulation and semantic association compensation on the cell surface state channel feature vector to be enhanced to obtain an enhanced cell surface state detection feature coding feature map as an enhanced cell surface state detection feature coding feature.

[0027] More specifically, the S141 extracts the channel feature vector of the (i, j)th pixel position from the cell surface state detection feature coding feature map as the channel feature vector of the cell surface state to be enhanced. It should be understood that although the cell surface state detection feature coding feature map generated by DenseNet aggregates the multi-scale features of the cell surface state, the channel vector of a single pixel position still mainly reflects the optical-geometric characteristics of the local area of ​​the cell surface, and it is difficult to independently carry the complete information of the cross-level propagation of hidden cracks or the scattering effect of contaminated particles. Therefore, in the technical solution of the present application, first, the channel feature vector of the (i, j)th pixel position is extracted from the cell surface state detection feature coding feature map as the channel feature vector of the cell surface state to be enhanced. That is, by anchoring the channel vector of a specific coordinate, a reference unit for pixel-level feature enhancement is constructed, so that the cross-layer features of each pixel position (such as the stress gradient at the crack tip and the reflection anomaly at the edge of the grid line) can be independently analyzed. In this process, the anchor vector is used as the radiation origin to construct a topological mapping relationship of the defect response in the feature space, so that the channel activation trajectory of the hidden crack propagation path is orthogonally decoupled from the periodic texture of the grid line, providing a physically interpretable feature operation interface for space-semantic joint modulation. In a specific example of the present application, the channel feature vector at the (i, j)th pixel position is extracted from the cell surface state detection feature coding feature map as the cell surface state channel feature vector to be enhanced; wherein the extraction formula is: ;in, It is the characteristic coding feature map of the battery cell surface state detection. is the set of real numbers, and They are The height and width of each feature matrix along the channel dimension, yes The number of channels, yes The channel feature vector of the (i, j)th pixel position in , is the characteristic vector of the surface state channel of the battery cell to be enhanced.

[0028] More specifically, the S142 performs n random scans on the cell surface state detection feature encoding feature map to obtain n cell surface state channel feature vectors as a sparse set of cell surface state reference feature vectors. It should be understood that when the detection system faces defects such as hidden cracks and micro scratches with random distribution characteristics, fixed area sampling can easily cause the feature reference set to over-focus on specific texture patterns and fail to adapt to the natural variation of the lattice arrangement of the cell surface in the production line. In the technical solution of the present application, by performing multi-region dynamic sampling in the high-dimensional feature space extracted by DenseNet, the constructed sparse reference set can cover both the ideal features of the standard grid line structure and the edge case feature patterns that may occur in the production process. The introduction of this spatial diversity effectively prevents the overfitting of local textures in the feature enhancement process, especially when encountering artifacts caused by uneven illumination or image distortion caused by equipment vibration, the randomly sampled reference vectors can form a more generalized context memory library. This intelligent feature sampling mechanism enables subsequent semantic association calculations to break through the local receptive field limitations, capture the deep correlation between the hidden crack extension path and the standard template structure deviation in the global feature space, and provide a dynamic adaptive feature enhancement benchmark for photovoltaic module preparation quality control. In a specific example of the present application, the following random scanning formula is used to randomly scan the cell surface state detection feature encoding feature map n times to obtain n cell surface state channel feature vectors as a sparse set of cell surface state reference feature vectors; wherein the random scanning formula is: ;in, is a sparse set of reference feature vectors of the cell surface state, and are the first, second, and third sparse sets of reference feature vectors of the cell surface state. and The cell surface state reference feature vector, and .

[0029] More specifically, in S143, based on the sparse set of cell surface state reference feature vectors, the cell surface state channel feature vector to be enhanced is subjected to spatial explicit modulation and semantic association compensation to obtain an enhanced cell surface state detection feature coding feature map as an enhanced cell surface state detection feature coding feature. In an embodiment of the present application, first, the spatial modulation degree and semantic association degree between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors are calculated to obtain a set of cell surface state spatial modulation degree matrix and cell surface state semantic association coding matrix; the specific steps are as follows: first, the Poincare distance between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors is calculated to obtain a cell surface state spatial modulation degree matrix; here, considering that when the detection system processes defects with tree-like expansion characteristics such as hidden cracks and lattice fractures, their morphological characteristics present a natural hierarchical distribution in the hyperbolic space, and the Poincare distance can accurately capture the relative position relationship in this non-Euclidean geometric structure. Compared with traditional distance metrics, this algorithm can more essentially reflect the gradient change law of defects radiating outward from the core point by constructing a geodesic distance model in hyperbolic space, such as the affiliation between the main branch of a hidden crack and a secondary crack, or the hierarchical characteristics of the attenuation of the conductivity of the contaminated area. This nonlinear spatial modeling capability makes the spatial modulation matrix no longer limited to the pixel-level similarity comparison of surface textures, but instead focuses on the topological differences between the defect growth pattern and the standard template structure. When the feature vector to be enhanced is quantified by the Poincare distance for similarity with the feature vector in the sparse reference set, the system essentially constructs a dynamic spatial attention weight map, in which high distance value areas correspond to feature patterns that have significant process deviations from the current detection point (such as abnormal grain boundary dislocations or non-standard grid line deformations), while low distance values ​​identify feature responses that meet production specifications. This hyperbolic geometry-based spatial modulation mechanism effectively solves the problem that traditional detection methods are prone to misjudging small fluctuations allowed by the process as defects under the interference of complex reflective textures, so that the feature enhancement process can adaptively distinguish normal production fluctuations from real quality anomalies, and provide an intelligent feature calibration benchmark that conforms to industrial physical characteristics for photovoltaic module production lines. In a specific example of the present application, the Poincare distance calculation formula is used to calculate the Poincare distance between the surface state channel feature vector of the cell to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain the cell surface state spatial modulation matrix; wherein, the Poincare distance calculation formula is: ;in, To calculate the square of the Euclidean norm, is the inverse hyperbolic cosine function, for and The Poincare distance between and are the eigenvalues ​​in the state space modulation matrix of the cell surface, It is the state space modulation matrix of the cell surface.

[0030] Then calculate the implicit semantic association between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain a set of cell surface state semantic association encoding matrices. It should be understood that the traditional pixel-level surface similarity measurement method cannot capture deep feature associations, especially when dealing with complex and subtle cell surface defects. Therefore, in the technical solution of the present application, by calculating the implicit semantic association between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector, higher-level cell surface state semantic information can be extracted from the image to extract information that is crucial to distinguishing defective areas from non-defective areas. For example, when evaluating whether the cell has subtle defects such as hidden cracks or micro scratches, this deep semantic analysis can help the system more accurately identify these difficult-to-detect problems, thereby improving the reliability of overall detection. In a specific example of the present application, the implicit semantic association between the channel feature vector of the cell surface state to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors is calculated using the following semantic association calculation formula to obtain a set of cell surface state semantic association encoding matrices; wherein the semantic association calculation formula is: ;in, is matrix-matrix multiplication, for The corresponding weight matrix is, is the transpose operation, yes and The length of the vector after multiplication, yes function, It is the first in the set of semantic association coding matrices of the battery cell surface state. The semantic association coding matrix of the surface state of each battery cell, is the set of semantic association coding matrices of the cell surface states, and The first, second and third sets of the semantic association coding matrices of the battery surface state are respectively The semantic association coding matrix of the surface state of each battery cell.

[0031] Next, each cell surface state semantic association coding matrix in the set of cell surface state semantic association coding matrices is used as a primary mask modulation unit and the cell surface state spatial modulation matrix is ​​used as a secondary mask modulation unit, and the cell surface state information compensation coding vector between each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors and the cell surface state channel feature vector to be enhanced is modulated explicitly to obtain the cell surface state enhancement component implicit coding vector. It should be understood that cell surface defects often present the dual characteristics of local spatial anomalies and global semantic shifts. For example, hidden cracks may cause optical scattering anomalies (spatial characteristics) in specific areas, while destroying the continuous topological structure (semantic characteristics) of the conductive layer on the cell surface. If only a single feature dimension is used for comparison, it is easy to cause misjudgment due to noise interference or feature confusion. In the technical solution of the present application, each cell surface state semantic association coding matrix in the set of cell surface state semantic association coding matrices is used as a primary mask modulation unit and the cell surface state spatial modulation matrix is ​​used as a secondary mask modulation unit, and the cell surface state information compensation coding vector between each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors and the cell surface state channel feature vector to be enhanced is explicitly modeled and modulated to obtain the cell surface state enhancement component implicit coding vector. That is, by constructing a double-layer mask modulation unit, specifically, using the semantic association coding matrix as a primary mask, semantic patterns with defect correlation are screened from a global perspective. For example, when hidden cracks appear in the area to be detected, the semantic mask will give priority to strengthening the feature responses related to the fracture texture and edge discontinuity, and then the spatial modulation matrix is ​​introduced as a secondary mask to geometrically correct the defect boundary at the local pixel level to eliminate the edge blurring effect that may be caused by Gaussian filtering. In this process, the topological structure of the feature space is constrained by the gauge field (semantic association), and the covariant field (spatial modulation) is used to compensate for the deformation of the cell surface state feature density distribution at the differential geometry level. The feature enhancement path is reconstructed through the interaction between the gauge field and the covariant field. This field-space coupling mechanism enables the enhancement process to break through the limitations of the receptive field of the traditional convolutional neural network: on the one hand, the semantic association matrix maintains the category invariance of the defect feature through the gauge transformation, avoiding the artifact interference introduced by image enhancement; on the other hand, the spatial modulation matrix dynamically adjusts the curvature of the feature manifold through the covariant derivative, and strengthens the differential feature contrast between the defect edge and the non-uniform background. The implicit coding vector of the cell surface state enhancement component generated in the end essentially constructs a defect-sensitive feature hyperplane, so that the subsequent differential feature map can highlight the pixel-level anomalies of the cell surface state, providing the classifier with a defect feature base with high discriminability.In a specific example of the present application, the specific steps of explicitly modeling and modulating the cell surface state information compensation coding vector between each cell surface state reference feature vector in a sparse set of cell surface state reference feature vectors and the cell surface state channel feature vector to be enhanced are as follows: calculating the positional difference vector between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain a set of cell surface state information compensation coding vectors; based on the primary mask modulation unit and the secondary mask modulation unit, performing cell surface state information quantization compensation based on a dual coupling mechanism on the set of cell surface state information compensation coding vectors to obtain an implicit coding vector of the cell surface state enhancement component.

[0032] In particular, in this example, each cell surface state semantic association coding matrix in the set of cell surface state semantic association coding matrices is used as a primary mask modulation unit and the cell surface state spatial modulation matrix is ​​used as a secondary mask modulation unit, and the cell surface state information compensation coding vector between each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors and the cell surface state channel feature vector to be enhanced is explicitly modeled and modulated using the following modulation formula to obtain an implicit coding vector of the cell surface state enhancement component; wherein the modulation formula is: ;in, yes and The cell surface state information compensation coding vector between yes and The cell surface state information compensation semantic enhancement coding vector between them, yes and The cell surface state information compensation space enhancement coding vector between, is the number of vectors in the sparse set of reference feature vectors of the cell surface state, It is the implicit coding vector of the enhanced component of the battery surface state.

[0033] Then, the cell surface state enhancement component implicit coding vector and the cell surface state channel feature vector to be enhanced are fused to obtain the cell surface state enhancement channel feature vector, wherein the cell surface state enhancement channel feature vector is the channel feature vector of the pixel position (i, j) of the enhanced cell surface state detection feature coding feature map. Since cell surface defects (such as hidden cracks or micro scratches) usually have local microscopic characteristics, their feature differences from normal areas may be hidden in the nonlinear relationship of the high-dimensional feature space. If only the enhancement component implicit coding vector is relied upon to directly replace the original channel feature vector, although the semantic response of the defective area can be enhanced, the discriminative texture information contained in the original feature (such as the cell grid line structure and the reflective characteristics of the surface passivation layer) may be over-covered, thereby weakening the model's ability to distinguish background noise caused by process fluctuations. Therefore, in the technical solution of the present application, the implicit coding vector of the cell surface state enhancement component and the channel feature vector of the cell surface state to be enhanced are fused to obtain the cell surface state enhancement channel feature vector, wherein the cell surface state enhancement channel feature vector is the channel feature vector of the pixel position (i, j) of the feature encoding feature map of the enhanced cell surface state detection feature. This coupling mechanism enables the fused cell surface state enhancement channel feature vector to have dual advantages in the photovoltaic detection scenario: on the one hand, by introducing the enhancement component, the activation intensity of the defect area in the feature space is enhanced; on the other hand, by retaining the underlying structural information of the original feature, the feature offset caused by excessive enhancement is avoided (for example, the cutting marks on the edge of the cell are misjudged as hidden cracks). In this way, the classifier can make decisions based on feature differences that can be explained by physical mechanisms, rather than simply relying on statistical image differences, thereby significantly improving the correlation between the detection results and the actual electrical performance degradation of the cell, and providing reliable data support for the closed-loop optimization of the photovoltaic module manufacturing process. In a specific example of the present application, the following fusion formula is used to fuse the implicit coding vector of the cell surface state enhancement component and the cell surface state channel feature vector to be enhanced to obtain the cell surface state enhancement channel feature vector; wherein the fusion formula is: ;in, and is a weighted hyperparameter, It is the channel feature vector of the enhanced cell surface state, that is, the channel feature vector of the (i, j)th pixel position of the enhanced cell surface state detection feature encoding feature map.

[0034] Specifically, the S15 is based on the semantic difference between the cell surface state detection feature coding feature and the enhanced cell surface state template feature coding feature to obtain the detection result. Specifically, in the embodiment of the present application, first, the position difference between the enhanced cell surface state detection feature coding feature map and the enhanced cell surface state template feature coding feature map is calculated to obtain the cell surface state comparison semantic coding feature map; here, since defects such as hidden cracks and micro scratches on the cell surface are often manifested as abnormal reflectivity or texture structure fracture in the local area in optical imaging, but they may overlap with normal process fluctuations in a single feature space, it is difficult for the traditional similarity comparison method to effectively separate the essential differences between the two. Therefore, in the technical solution of the present application, the position difference between the enhanced cell surface state detection feature coding feature map and the enhanced cell surface state template feature coding feature map is calculated to obtain the cell surface state comparison semantic coding feature map. By calculating the position difference between the two, the characteristic manifold deviation between the area to be tested and the standard template in the same spatial coordinates can be quantified through the vector field difference operation in the feature space. This deviation essentially maps the feature space distortion caused by abnormal microstructure on the cell surface. The vector value of each pixel position in the generated cell surface state comparison semantic coding feature map encodes the degree of electrical performance degradation and spatial distribution characteristics caused by defects. In this way, the subsequent classifier can intelligently distinguish between allowable process fluctuations and real defects based on process standards, providing a decision-making basis with both spatial accuracy and physical significance for the reliability assessment of photovoltaic modules.

[0035] Furthermore, the surface state of the cell is compared with the semantic coding feature map and input into the surface contamination detection module based on the classifier to obtain the detection result, and the detection result is used to indicate whether the surface cleanliness is qualified. That is, through the mapping relationship between the defect mode and the degree of electrical performance degradation learned during the training process, the physical information implicit in the comparison semantic coding feature map is converted into a specific detection result. It should be understood that in an actual production environment, even tiny pollutants may significantly reduce the performance of photovoltaic modules. Through such an automated detection method, these problems can be discovered in time and corresponding measures can be taken to clean or replace them to ensure that each cell can reach the best working state. In a specific example of the present application, the specific steps of inputting the surface state of the cell into the surface contamination detection module based on the classifier to obtain the detection result are as follows: the surface state of the cell is compared with the semantic coding feature map. Expand the classification feature vector based on the row vector or column vector; use multiple fully connected layers of the classifier to fully connect the classification feature vector to obtain the encoded classification feature vector; pass the encoded classification feature vector through the Softmax classification function of the classifier to obtain the detection result. It is worth mentioning that in the technical solution of the present application, the labels of the classifier include qualified surface cleanliness (first label), and unqualified surface cleanliness (second label), wherein the classifier determines to which classification label the cell surface state comparison semantic coding feature map belongs through a soft maximum function. It is worth noting that the first label p1 and the second label p2 here do not contain artificially set concepts. In fact, during the training process, the computer model does not have the concept of "whether the surface cleanliness is qualified". It just has two classification labels and the probability of the output feature under these two classification labels, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether the surface cleanliness is qualified is actually converted into a binary class probability distribution that conforms to the laws of nature through the classification label. In essence, the physical meaning of the natural probability distribution of the label is used, rather than the linguistic text meaning of "whether the surface cleanliness is qualified".

[0036] In particular, in S2, after determining that the test result of the battery cell is qualified, the single-piece welding of the battery cell and the series welding of the battery cell are performed. That is, after determining that the surface cleanliness of the battery cell is qualified, the battery cell is subjected to single-piece welding and series welding of the battery cell. Among them, the single-piece welding process mainly involves accurately welding the fine welding strips to the electrodes on the front and back of each individual battery cell. This step requires a high degree of precision to ensure good electrical connection and avoid any form of mechanical damage to the battery cell. After completing the single-piece welding, the next step is the series welding process of the battery cell, that is, connecting multiple battery cells that have completed single-piece welding in series through welding to form a battery string. Usually, in order to achieve the voltage output required by the photovoltaic module, a certain number of battery cells need to be arranged and connected in a specific manner. During the series welding process, it is also necessary to ensure the quality of each connection point to ensure that the entire battery string has good conductivity and mechanical strength.

[0037] In particular, the S3 is to stack and laminate the materials of each photovoltaic module, wherein the photovoltaic module materials include cells, glass plates, back plates and EVA. In a specific example of the present application, first, in the stacking stage, these materials need to be placed in a specific order. Usually, the bottom layer is the back plate, which provides a certain degree of protection for the entire component and helps to prevent moisture and electrical insulation; followed by the EVA film, which has good transparency and adhesion and is used to firmly combine the various component layers; then, the battery string that has been welded and connected in series is placed on the EVA film; here, in order to ensure the best photoelectric conversion efficiency, the position of the cell must be accurate; further, the EVA film is stacked on the surface to further protect the cell from the external environment and ensure that light can efficiently penetrate to the surface of the cell; finally, a layer of glass plate is covered on the top, which not only provides physical protection, but also enhances the overall structural strength of the component and allows enough sunlight to penetrate to stimulate the photoelectric effect in the cell.

[0038] After the stacking is completed, the materials of each photovoltaic module are laminated. During this process, all the stacked materials are placed in a specially designed laminator. By applying a certain temperature and pressure, the EVA film is melted and evenly distributed, so that all the component layers are tightly combined to form a solid and sealed whole. This process is crucial to ensure that the photovoltaic modules have good weather resistance, waterproofness and long-term reliability. It is worth mentioning that correct temperature and pressure control is also a key factor to ensure that the EVA film will not damage the battery cell due to overheating or excessive compression.

[0039] In particular, the S4 is to side-seal and frame the laminated photovoltaic module materials to obtain a photovoltaic module. Among them, side sealing refers to the sealing treatment on all sides of the laminated photovoltaic module to further enhance its waterproof and dustproof performance and protect the internal structure from the influence of the external environment. In the specific implementation process, a layer of special sealant is usually applied to the edge of the laminated component or a specially designed sealing strip is used. This step requires the sealing material to have good weather resistance, adhesion and flexibility to adapt to thermal expansion and contraction under different climatic conditions. By accurately controlling the application amount and uniformity of the sealing material, it is possible to effectively prevent moisture and other impurities from invading the interior of the component, thereby avoiding the degradation of electrical performance or failure caused by this.

[0040] Framing is the process of installing a sturdy frame for photovoltaic modules. Its main purpose is to provide additional mechanical protection and support, and also to facilitate the transportation, installation and maintenance of the modules. Commonly used frame materials include lightweight and high-strength metals such as aluminum alloys. These materials not only have good corrosion resistance, but can also withstand certain external impacts without damaging the modules. In the framing process, you first need to customize a suitable frame according to the size of the module and ensure that it fits tightly with the surface of the module. Then, fix the frame around the module using screws or other fasteners, while being careful not to cause any damage to the module. In addition, some holes or notches will be reserved on the frame for subsequent electrical connections and the installation of grounding wires.

[0041] In summary, the preparation method of the photovoltaic module according to the embodiment of the present application is explained, which first collects the surface image of the cell to be detected through a camera, and extracts the cell surface comparison standard template image from the background database, then uses a neural network model based on deep learning to extract the deep features of the cell surface of the cell surface image to be detected and the cell surface comparison standard template image, and further combines the spatial-semantic dual enhancement algorithm to strengthen the representation difference between the defective area and the non-defective area, so as to achieve pixel-level difference detection of the cell surface state. In this way, the detection accuracy of the cell surface state is improved, thereby effectively improving the overall quality and long-term reliability of the photovoltaic module, reducing the problem of reduced photoelectric conversion efficiency caused by cell surface defects, and providing strong support for the efficient and reliable development of the photovoltaic industry.

[0042] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for preparing a photovoltaic module, characterized in that: include: Test the battery cells; After the test result of the battery cell is determined to be qualified, the single-cell welding and series welding of the battery cell are carried out; Stacking and laminating various photovoltaic module materials, wherein the photovoltaic module materials include battery cells, glass plates, back plates and EVA; Performing side sealing and framing on the laminated photovoltaic module materials to obtain a photovoltaic module; The battery cell is inspected, including: performing pixel-level semantic difference mapping of the battery cell surface state image features on the surface image of the battery cell to be inspected and the battery cell surface comparison standard template image to obtain the inspection result, specifically including: Acquire the surface image of the battery cell to be inspected collected by the camera; Extract the battery cell surface comparison standard template image from the background database; Extracting the cell surface state image features of the cell surface image to be detected and the cell surface comparison standard template image to obtain the cell surface state detection feature coding features and the cell surface state template feature coding features, including: performing image enhancement on the cell surface image to obtain the cell enhanced surface image; passing the cell enhanced surface image through a cell surface state image feature extractor based on a DenseNet model to obtain a cell surface state detection feature coding feature map as the cell surface state detection feature coding feature; passing the cell surface comparison standard template image through a cell surface state image feature extractor based on a DenseNet model to obtain a cell surface state template feature coding feature map as the cell surface state template feature coding feature; Performing cell surface state significance enhancement processing on cell surface state detection feature coding features and cell surface state template feature coding features to obtain enhanced cell surface state detection feature coding features and enhanced cell surface state template feature coding features, including: extracting a channel feature vector at the (i, j)th pixel position from a cell surface state detection feature coding feature map as a cell surface state channel feature vector to be enhanced; performing n random scans on the cell surface state detection feature coding feature map to obtain n cell surface state channel feature vectors as a sparse set of cell surface state reference feature vectors; based on the sparse set of cell surface state reference feature vectors, performing spatial explicit modulation and semantic association compensation on the cell surface state channel feature vector to be enhanced to obtain an enhanced cell surface state detection feature coding feature map as an enhanced cell surface state detection feature coding feature; Based on the semantic difference of the cell surface state between the enhanced cell surface state detection feature coding feature and the enhanced cell surface state template feature coding feature, the detection result is obtained.

2. The method for preparing a photovoltaic module according to claim 1, characterized in that: The image enhancement includes performing histogram equalization and Gaussian filtering noise reduction on the surface image of the battery cell.

3. The method for preparing a photovoltaic module according to claim 2, characterized in that: Based on the sparse set of cell surface state reference feature vectors, spatial explicit modulation and semantic association compensation are performed on the cell surface state channel feature vectors to be enhanced to obtain an enhanced cell surface state detection feature coding feature map as an enhanced cell surface state detection feature coding feature, including: Calculate the spatial modulation and semantic association between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain a set of cell surface state spatial modulation matrix and cell surface state semantic association coding matrix; Using each cell surface state semantic association coding matrix in the set of cell surface state semantic association coding matrices as a primary mask modulation unit and using a cell surface state spatial modulation matrix as a secondary mask modulation unit, performing explicit modeling modulation on the cell surface state information compensation coding vector between each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors and the cell surface state channel feature vector to be enhanced to obtain a cell surface state enhancement component implicit coding vector; The implicit coding vector of the cell surface state enhancement component and the channel feature vector of the cell surface state to be enhanced are fused to obtain the cell surface state enhancement channel feature vector, wherein the cell surface state enhancement channel feature vector is the channel feature vector of the pixel position (i, j) of the enhanced cell surface state detection feature encoding feature map.

4. The method for preparing a photovoltaic module according to claim 3, characterized in that: Calculating the spatial modulation degree and semantic association degree between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain a set of cell surface state spatial modulation degree matrix and cell surface state semantic association coding matrix, including: Calculate the Poincare distance between the channel feature vector of the cell surface state to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain the cell surface state space modulation matrix; The implicit semantic association between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors is calculated to obtain a set of cell surface state semantic association encoding matrices.

5. The method for preparing a photovoltaic module according to claim 4, characterized in that: Using each cell surface state semantic association coding matrix in the set of cell surface state semantic association coding matrices as a primary mask modulation unit and using a cell surface state spatial modulation matrix as a secondary mask modulation unit, performing explicit modeling modulation on the cell surface state information compensation coding vector between each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors and the cell surface state channel feature vector to be enhanced to obtain a cell surface state enhancement component implicit coding vector, including: Calculating the position difference vector between the cell surface state channel feature vector to be enhanced and each cell surface state reference feature vector in the sparse set of cell surface state reference feature vectors to obtain a set of cell surface state information compensation coding vectors; Based on the primary mask modulation unit and the secondary mask modulation unit, the set of cell surface state information compensation coding vectors is subjected to the cell surface state information quantization compensation based on the dual coupling mechanism to obtain the cell surface state enhancement component implicit coding vector.

6. The method for preparing a photovoltaic module according to claim 5, characterized in that: Based on the semantic difference between the cell surface state detection feature coding feature and the enhanced cell surface state template feature coding feature, the detection results are obtained, including: Calculate the position difference between the enhanced cell surface state detection feature coding feature map and the enhanced cell surface state template feature coding feature map to obtain a cell surface state comparison semantic coding feature map; The surface state of the battery cell is compared with the semantic coding feature map and input into the surface contamination detection module based on the classifier to obtain the detection result, and the detection result is used to indicate whether the surface cleanliness is qualified.

7. A photovoltaic module, characterized in that: The photovoltaic module is prepared by the method for preparing the photovoltaic module according to any one of claims 1 to 6.

Images