LED module breakpoint detection and positioning method and device

Abstract

The invention relates to the technical field of LED modules, and discloses an LED module breakpoint detection and positioning method and device, and the method comprises the steps: carrying out the voltage and current data collection of series and parallel nodes of an LED module, and obtaining an electrical feature matrix and a node topological graph; performing iterative calculation on the electrical characteristic matrix, and processing a voltage difference value and a current variation between adjacent nodes to obtain a breakpoint characteristic vector; inputting the breakpoint feature vector and the node topological graph into a multi-task neural network for multilayer convolution operation to obtain an electrical fault distribution graph; performing feature segmentation and depth feature extraction on the electrical fault distribution map to obtain fault area feature data; and establishing an electric field distribution model for the fault area feature data, calculating an electric field singular point through a least square method, and obtaining breakpoint space coordinate data, thereby improving the accuracy of LED module breakpoint detection, and realizing accurate positioning of the LED module breakpoint.

Description

Technical Field

[0001] The present application relates to the technical field of LED modules, and in particular to a method and device for detecting and locating breakpoints of LED modules. Background Art

[0002] Due to the complexity of the series-parallel structure of LED modules, breakpoint failures are prone to occur during long-term use. This failure not only affects the display effect, but may also cause the function of the entire module to fail. Traditional LED module breakpoint detection methods mainly rely on manual visual inspection or simple voltage detection, which is inefficient and inaccurate. In recent years, although some researchers have proposed detection methods based on image processing and electrical characteristics analysis, these methods can often only identify obvious faults and lack effective recognition capabilities for potential breakpoint hazards and early fault characteristics.

[0003] The existing LED module breakpoint detection technology faces three main problems: the series-parallel structure of the LED module is complex, and the traditional single voltage or current detection method is difficult to accurately locate the breakpoint position; the electrical characteristics of the breakpoint fault are time-varying and nonlinear, and the simple feature extraction method cannot effectively capture the fault characteristics; the existing detection algorithms generally lack the ability to fuse and analyze multi-source information, making it difficult to achieve high-precision fault location. Summary of the invention

[0004] The present application provides a method and device for detecting and locating breakpoints of an LED module, thereby improving the accuracy of breakpoint detection of the LED module and achieving precise positioning of the breakpoints of the LED module.

[0005] A first aspect of the present application provides a method for detecting and locating a breakpoint of an LED module, the method comprising: Collect voltage and current data of the series and parallel nodes of the LED module to obtain the electrical characteristic matrix and node topology diagram; Iteratively calculating the electrical characteristic matrix, processing the voltage difference and current change between adjacent nodes, and obtaining a breakpoint characteristic vector; Inputting the breakpoint feature vector and the node topology map into a multi-task neural network for multi-layer convolution operation to obtain an electrical fault distribution map; Performing feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data; An electric field distribution model is established for the characteristic data of the fault area, and the electric field singular points are calculated by the least square method to obtain the breakpoint space coordinate data.

[0006] A second aspect of the present application provides a LED module breakpoint detection and positioning device, the LED module breakpoint detection and positioning device comprising: The acquisition module is used to collect voltage and current data of the series and parallel nodes of the LED module to obtain the electrical characteristic matrix and node topology diagram; A calculation module, used for iteratively calculating the electrical characteristic matrix, processing the voltage difference and current variation between adjacent nodes, and obtaining a breakpoint characteristic vector; A calculation module, used for inputting the breakpoint feature vector and the node topology map into a multi-task neural network to perform multi-layer convolution operations to obtain an electrical fault distribution map; An extraction module, used for performing feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data; A module is established to establish an electric field distribution model for the characteristic data of the fault area, calculate the electric field singular points by the least square method, and obtain the breakpoint space coordinate data.

[0007] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned LED module breakpoint detection and positioning method.

[0008] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned LED module breakpoint detection and positioning method.

[0009] Compared with the prior art, the present application has the following beneficial effects: by establishing a multi-task neural network and a deep feature extraction model, combined with a voltage domain iterative learning strategy, the accuracy of breakpoint detection is significantly improved, and the electric field distribution model and the least squares optimization algorithm are used to achieve accurate positioning of the breakpoint position and improve the spatial resolution of fault location. The combination of electrical feature analysis and deep learning methods breaks through the limitations of traditional single detection methods and achieves rapid identification of breakpoint faults in complex series-parallel structures. By introducing adaptive feature extraction and multi-level data fusion technology, the system's resistance to environmental interference is enhanced, making the detection results more stable and reliable. Based on the distributed node data acquisition solution, combined with an efficient feature processing algorithm, real-time monitoring of large-scale LED modules is achieved. Through the application of deep separable convolution modules and attention mechanisms, the computational complexity of the algorithm is reduced, while the detection performance is guaranteed. The electric field curvature distribution analysis method is used to improve the system's ability to identify early fault signs, and potential breakpoint fault risks can be warned in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0011] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0012] Figure 1 It is a flow chart of a method for detecting and locating breakpoints of an LED module provided by an embodiment of the present invention; Figure 2 is a schematic block diagram of the structure of an LED module breakpoint detection and positioning device provided by an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0015] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0016] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In the embodiment of the present application, an embodiment of the LED module breakpoint detection and positioning method includes: Step 100, collecting voltage and current data of the series-parallel nodes of the LED module to obtain an electrical characteristic matrix and a node topology diagram; It is understandable that the execution subject of the present application may be a LED module breakpoint detection and positioning device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0017] Specifically, a coordinate system is established for each series-parallel node in the LED module, and a spatial mapping calculation is performed to form a node coordinate data set. By analyzing the physical structure of the module and combining the spatial geometric mapping rules, the position of each node in the actual space is represented by a unified coordinate system. The voltage and current of each node in the node coordinate data set are collected and synchronously sampled and converted, and the collected analog signals are converted into digital signals to form original electrical data, reflecting the real-time electrical characteristics of the LED module at each node. The original electrical data is denoised. The original signal is decomposed at multiple scales using wavelet decomposition operations to extract high-frequency components, and the high-frequency coefficients are threshold processed to effectively filter out noise and obtain the denoised signal data. The denoised signal data is divided into multiple time periods and segmented, and the characteristic parameters of each data segment are extracted from it to generate segmented feature data. The voltage and current values ​​in the segmented feature data are subjected to zero-mean standardization and variance normalization calculations, so that the data distribution has a uniform mean and standard deviation, eliminating the calculation deviation caused by dimensional differences or absolute value changes. The standardized electrical features and node coordinate data sets are spatially associated, and the feature mapping transformation is completed using geometric mapping and node matching technology to generate a node feature association table. The node feature association table describes the relationship between the coordinate information of each node in the spatial position and its corresponding standardized electrical features in the form of a matrix. Based on the node feature association table, combined with the series-parallel topological structure of the LED module, connectivity analysis and adjacency matrix calculation are performed. By analyzing the physical connection relationship between nodes, a node topology diagram reflecting the internal network structure of the module is established. The node topology diagram uses graph theory to show the connection path, relationship strength, and distribution law of electrical characteristics of nodes in the module in the network. The node feature association table is reconstructed into a feature structure with M×N dimensions. The node features and topological information are organized in a unified format through data rearrangement operations to obtain an electrical feature matrix.

[0018] Step 200, iteratively calculating the electrical characteristic matrix, processing the voltage difference and current change between adjacent nodes, and obtaining the breakpoint characteristic vector; Specifically, according to the voltage data in the electrical characteristic matrix, the voltage ratios of adjacent nodes are calculated one by one to form a node voltage ratio sequence, which reflects the relative change relationship of the voltage between each node. The node voltage ratio sequence is subjected to time series analysis, the dynamic characteristics of voltage change are extracted from the time dimension, and the voltage change rate matrix is ​​generated to quantify the change speed of voltage at different time points. At the same time, according to the current data in the electrical characteristic matrix, a current time series model is established, and the current change data is extracted by performing a first-order difference operation on the time series. The first-order difference can effectively eliminate the trend interference in the absolute value of the current, highlight the abnormal characteristics such as mutation and jump, and make the current change data a key indicator for characterizing the breakpoint characteristics. The voltage change rate matrix is ​​subjected to Fourier transform to extract its frequency domain information, and the voltage periodic fluctuation data is obtained to reveal the implicit laws in the voltage signal, such as the periodic anomaly caused by the breakpoint. The current change data is subjected to adaptive threshold analysis, and the threshold is automatically adjusted according to the signal distribution characteristics to extract the current mutation point data, thereby accurately capturing the abnormal behavior of the current at the breakpoint. Based on the extracted voltage periodic fluctuation data, a voltage domain iterative learning model is constructed, and the voltage feature weight is continuously optimized through multiple weight update operations. The weight update process is carried out in an iterative manner, and each round of update is adjusted based on the previous feature extraction results, so that the final voltage feature weight can accurately reflect the voltage abnormality characteristics caused by the breakpoint. The voltage feature weight and the current mutation point data are subjected to feature cross operation, and the information of two different electrical features is fused to generate fused feature data. The synergistic relationship between multidimensional features is introduced through cross operation, so that the fused feature data not only contains a single voltage or current feature, but also captures the interaction characteristics between the two. The fused feature data is subjected to feature dimensionality reduction and normalization. The purpose of dimensionality reduction is to remove redundant features, reduce computational complexity, and retain important breakpoint-related features in the data. Normalization eliminates the influence of dimensional differences between features by standardizing the data with zero mean and unit variance, and improves the consistency of data in subsequent calculations. The fused feature data after dimensionality reduction and normalization is the breakpoint feature vector.

[0019] Step 300: Input the breakpoint feature vector and the node topology map into a multi-task neural network for multi-layer convolution operation to obtain an electrical fault distribution map; It should be noted that the breakpoint feature vector and the node topology map are processed by feature mapping to construct the input data of the multi-task neural network. The input data is effectively organized into a format suitable for the multi-task neural network in the design to ensure that the spatial relationship and electrical characteristics between the features are preserved. The architecture of the multi-task neural network includes a shared feature extraction network and two parallel task branch networks, which are used for node feature learning and electrical boundary feature learning respectively. In the feature extraction stage, the shared feature extraction network generates shared feature data by processing the input data. The network consists of 4 convolutional layers, each of which uses a 3×3 convolution kernel and combines the ReLU activation function to ensure the full expression of nonlinear features. In order to stabilize the training process and accelerate convergence, a BatchNorm layer is introduced after each convolution layer for batch normalization processing, and some neurons are randomly shielded through the Dropout layer to avoid overfitting problems. The shared feature extraction network can extract multi-dimensional electrical features and generate shared feature data with high expression ability. The shared feature data is input into the first task branch network and the second task branch network respectively. In the first task branch network, node feature learning is performed. Its structure consists of three fully connected layers. Each layer uses the LeakyReLU activation function to enhance the gradient sensitivity to negative areas through the characteristics of small negative slopes. Weight regularization technology is added to the network to limit the scale of parameters to improve generalization ability and generate node feature mapping results. In the second task branch network, electrical boundary feature learning is focused on. The network contains two transposed convolutional layers to improve the resolution of the feature map, and the Tanh activation function is used to output smooth boundary features. At the same time, the attention layer is introduced into the network to enhance the expression ability of key features by calculating the importance weights of each feature point, and generate boundary feature mapping results. The node feature mapping results and the boundary feature mapping results are integrated through the feature fusion step. The attention mechanism is used to dynamically adjust the weights of the two types of features to obtain more representative weighted feature data. The weighted feature data is input into the decoder network for feature decoding. The decoder network contains three deconvolution layers, each of which uses a 4×4 convolution kernel and a ReLU activation function to gradually restore the original spatial resolution of the feature map while maintaining a high feature expression ability. The output of the decoder network is feature distribution data, which contains the spatial distribution information of electrical faults in the LED module. In order to optimize the feature expression, data reconstruction and residual connection operations are performed on the feature distribution data. The residual connection retains the key information in the initial features while enhancing the expressiveness of the global features. After the feature fusion is completed, the global feature data is reduced in dimension through a 1×1 convolutional layer to generate a feature representation with low dimensionality and high expressiveness. Through the node-level feature classification step, the reduced global feature data is parsed into an electrical fault distribution map. The distribution map shows the fault location and characteristics of the LED module in the form of spatial and topological information.

[0020] Step 400: performing feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data; Specifically, the voltage gradient calculation and statistical analysis are performed on the electrical fault distribution map. During the calculation process, the optimal segmentation threshold is determined by quantifying the voltage change rate and the spatial gradient distribution characteristics. This threshold is obtained through multiple iterative optimization and statistical analysis, and can distinguish the fault area from the normal area to the greatest extent. Based on the optimal segmentation threshold, the electrical fault distribution map is converted into a binary feature matrix to obtain the marked fault area data. The electrical boundary extraction operation is performed on the marked fault area data. The data of the fault area is processed by the voltage gradient operator, including calculating the voltage amplitude change and change direction to extract the key feature data of the electrical boundary. The voltage amplitude reflects the strength of the electrical signal, while the change direction provides the correlation between the geometric information of the boundary and the electrical characteristics. These data together constitute the electrical boundary feature data. The gradient operator calculation can capture the local details of the electrical changes in space and ensure the accuracy and completeness of the boundary extraction. The extracted electrical boundary feature data is input into the deep feature network, which is designed to contain a 5-layer stacked depth-separable convolution module. This module uses efficient convolution operations to extract and reduce features. Each module is composed of 3×3 deep convolution and 1×1 point convolution. The deep convolution is responsible for extracting spatial features, and the point convolution completes the channel fusion of features, so as to obtain highly expressive feature description vectors at a low computational cost. The feature description vector is enhanced by the self-attention mechanism. The self-attention mechanism dynamically calculates the weight distribution of feature vectors to improve the expressiveness of key features, while suppressing redundant or irrelevant information to generate enhanced feature data. The enhanced feature data is fused with the original fault area data. The feature fusion process combines the global electrical characteristics and local deep features of the fault area to form a comprehensive expression form and generate fused feature data. The fused feature data is reconstructed for electrical features, and the data of the fault area is restored with high fidelity through the reconstruction operation to generate reconstructed regional data. The original features and deep features are combined in the reconstruction process to ensure that the data contains both local high-precision information and global electrical characteristics. After the data reconstruction is completed, the reconstructed regional data is subjected to feature optimization and data fitting to eliminate existing noise or redundant features and improve the continuity and accuracy of the data. The obtained fault area characteristic data not only has high-quality electrical characteristic description, but also contains complete spatial and boundary information.

[0021] Step 500: Establish an electric field distribution model for the fault area characteristic data, calculate the electric field singular points by the least square method, and obtain the breakpoint space coordinate data.

[0022] Specifically, the characteristic data of the fault area is divided into potential grids, and a potential sampling point matrix is ​​generated in the form of a uniformly distributed grid. By spatially discretizing the electrical characteristics of the fault area, it is ensured that the potential distribution of the sampling points can fully cover the entire fault area. The Gauss-Seidel iterative calculation is performed on the potential sampling point matrix to obtain discrete potential data in a fast convergence manner, which locally reflects the distribution law of the potential in the fault area. According to the second-order central difference format, the electric field potential energy equation is established for the discrete potential data, and the electric field intensity distribution matrix is ​​calculated by finite difference operation. By accurately solving the potential gradient, the electric field intensity value of each grid point is generated, which can reflect the change trend and spatial distribution characteristics of the electric field. After obtaining the electric field intensity distribution, each grid point in the distribution matrix is ​​substituted into the Poisson equation, and the constraint operation is performed in combination with the boundary conditions of the fault area to generate the electric field Poisson distribution data. The Poisson equation can describe the relationship between the potential and the charge distribution, and enhance the physical consistency and accuracy of the distribution data through the constraints of the boundary conditions. The least square fitting objective function is established for the electric field Poisson distribution data. The optimization goal is to minimize the error between the electric field intensity and the model prediction value. The electric field fitting coefficient matrix is ​​optimized by the gradient descent algorithm. The optimized fitting coefficient matrix can fully describe the local characteristics and overall distribution trend of the electric field. The electric field fitting coefficient matrix is ​​input into the Laplace operator model for second-order partial derivative calculation to generate the electric field curvature distribution map. The electric field curvature distribution map can reveal the direction and rate of electric field change and highlight the abnormal points in the electric field distribution. The electric field curvature distribution map is subjected to singular value decomposition, and the main eigenvector is identified by eigenvalue sorting to extract the key characteristics of the electric field distribution. The local extreme points in the main eigenvector are identified by the extreme point detection method, and a singular point candidate set is generated. These candidate points are concentrated in the area where the electric field changes abnormally, and the potential position of the breakpoint is preliminarily located. The singular point candidate set is subjected to spatial aggregation analysis. Through the clustering algorithm, the singular points close in space are aggregated to obtain the cluster center points. These center points are used as candidate positions for breakpoints, which significantly reduces the redundancy of candidate points. The cluster center points are coordinate transformed and precision corrected to eliminate the deviation introduced by coordinate system transformation or calculation error. The corrected cluster center points accurately represent the space of the breakpoints in the LED module.

[0023] The electric field Poisson distribution data is expanded by basis functions through N-order Legendre polynomials. The polynomials are characterized by their orthogonality and can efficiently decompose the complex components in the electric field Poisson distribution. By orthogonally expanding the Poisson distribution data, the expansion coefficients of each order basis function are calculated, and a set of basis function expansion coefficients are obtained to describe the projection of the Poisson distribution on each basis function. Based on the basis function expansion coefficients, the least squares fitting objective function is constructed to optimize the electric field fitting coefficient matrix. In this process, in order to avoid overfitting and improve the generalization ability of the model, the objective function introduces the L2 norm regularization constraint to form a loss function with a penalty term. The loss function is designed to minimize the electric field distribution fitting error and limit the parameter scale through the penalty term. The loss function with the penalty term is differentiated by variables, and the gradient direction is calculated by forward difference and backward difference operations to obtain the complete gradient direction vector. After the gradient direction vector is determined, the optimization step size is dynamically adjusted by setting the adaptive learning rate to ensure the balance between convergence speed and accuracy. The model is iteratively optimized using the back-propagation algorithm, and the electric field Poisson distribution data is continuously fitted and updated to obtain the optimized electric field fitting coefficient matrix. The matrix reflects the global trend and local details of the electric field distribution. The electric field fitting coefficient matrix is ​​input into the two-dimensional Laplace operator model, and the second-order partial derivatives are calculated by the finite difference method to generate the second-order derivative tensor. The high-precision characteristics of the finite difference method ensure the reliability of the tensor calculation, and the second-order derivative tensor describes the local change rate and bending characteristics of the electric field. The principal curvature of the second-order derivative tensor is calculated. According to the formulas of Gaussian curvature and mean curvature, the data in the second-order derivative tensor are processed respectively to extract the key curvature feature data. Gaussian curvature reflects the overall curvature of the surface at each point, while the mean curvature reveals the smoothness and trend of the local change of the electric field. The curvature feature data is reconstructed according to the electric field distribution, and the reconstructed data is smoothed and spatially expanded by the bicubic interpolation method to generate the final electric field curvature distribution map. The high-order smoothing characteristics of bicubic interpolation ensure the continuity and accuracy of the curvature distribution map in space. The data at each point in the map clearly shows the changing characteristics of the electric field at the corresponding position.

[0024] In the embodiment of the present application, by establishing a multi-task neural network and a deep feature extraction model, combined with a voltage domain iterative learning strategy, the accuracy of breakpoint detection is significantly improved. The electric field distribution model and the least squares optimization algorithm are used to achieve accurate positioning of the breakpoint position and improve the spatial resolution of fault location. The electrical feature analysis is combined with the deep learning method to break through the limitations of the traditional single detection method and achieve rapid identification of breakpoint faults in complex series-parallel structures. By introducing adaptive feature extraction and multi-level data fusion technology, the system's resistance to environmental interference is enhanced, making the detection results more stable and reliable. Based on the distributed node data acquisition solution, combined with an efficient feature processing algorithm, real-time monitoring of large-scale LED modules is achieved. Through the application of deep separable convolution modules and attention mechanisms, the computational complexity of the algorithm is reduced, while ensuring the detection performance. The electric field curvature distribution analysis method is used to improve the system's ability to identify early fault signs, and can provide early warning of potential breakpoint fault risks.

[0025] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Establish a coordinate system and perform space mapping calculations for each series-parallel node in the LED module to obtain a node coordinate data set; The voltage and current of each node in the node coordinate data set are collected and synchronously sampled and converted to obtain the original electrical data; Perform wavelet decomposition and high-frequency coefficient threshold processing on the original electrical data to obtain the signal data after noise reduction; The signal data after noise reduction is processed in segments to obtain segmented characteristic data, and the voltage and current values ​​in the segmented characteristic data are subjected to zero-mean normalization processing and variance normalization calculation to obtain standardized electrical characteristics; Perform spatial correlation operation and feature mapping transformation on the standardized electrical features and the node coordinate data set to obtain a node feature correlation table; Based on the series-parallel topological structure of the LED module, the node feature association table is subjected to connectivity analysis and adjacency matrix calculation to obtain a node topological graph; The node feature association table is reconstructed into an M×N dimensional feature structure, and the electrical feature matrix is ​​obtained through data rearrangement operation.

[0026] Specifically, the coordinate system of the series and parallel nodes of the LED module is established and the spatial mapping is calculated to obtain a complete node coordinate data set. By analyzing the physical structure of the LED module, the geometric position of each node in the module is determined and mapped to the three-dimensional coordinate system. Assume that the physical position of the node is represented by Indicates that Respectively represent The position of the node on the three coordinate axes, then the spatial mapping result of the entire node is expressed as ,in is the total number of nodes, Represents the node coordinate data set. The voltage and current of each node in the node coordinate data set are collected, and the analog signal is converted into a digital signal through synchronous sampling conversion technology to obtain the original electrical data. Assume that the collected voltage and current signals are and ,in Indicates time, represents the node number, and the raw electrical data of each node is represented as ,in Is a node The original electrical data is subjected to wavelet decomposition and high-frequency coefficient threshold processing to obtain the signal data after noise reduction. Wavelet decomposition effectively separates different frequency components by projecting the signal onto a series of time-frequency bases. , the formula of wavelet decomposition is: ; in, are the wavelet decomposition coefficients, It is Tier wavelet basis functions, is the number of decomposition levels, is the number of coefficients per layer. High frequency noise passes through the threshold Filter out, the processing is: ; After treatment Generate the signal data after noise reduction. After segmenting the signal data after noise reduction, obtain the segment feature data. Assume that the segment time window is , then the data of each signal segment is expressed as The voltage and current values ​​in each signal segment are normalized to zero mean and variance to one to ensure the consistency and comparability of data between different nodes. The normalization formula is: ; in and are the mean and standard deviation of voltage, and are the mean and standard deviation of the current respectively. The standardized data and Construct standardized electrical features. Perform spatial correlation operations on the standardized electrical features and the node coordinate data set, combine the geometric positions of the nodes, and generate a node feature correlation table through feature mapping transformation. The feature mapping formula is: ; in is the mapping function, Representation Node Based on the generated node feature association table, combined with the series-parallel topology of the LED module, the node connectivity is analyzed and the adjacency matrix is ​​calculated. Construct a node topology graph. The adjacency matrix is ​​defined as: ; The node topology diagram reflects the physical connection relationship of the nodes in the LED module. The node feature association table is reconstructed into dimensional feature structure, where is the feature dimension, is the number of nodes. Through data rearrangement operation, it adapts to the input format of subsequent analysis and generates the final electrical characteristic matrix. The rearrangement formula is: ; in is the electrical characteristics matrix, which contains the spatial and electrical characteristics of all nodes.

[0027] In a specific embodiment, the process of executing step 200 may specifically include the following steps: The voltage ratios of adjacent nodes are calculated according to the voltage data in the electrical characteristic matrix to obtain a node voltage ratio sequence, and the node voltage ratio sequence is subjected to a time series analysis to obtain a voltage change rate matrix; A current time series model is established based on the current data in the electrical characteristic matrix, and the current variation data is obtained through first-order difference operation; The voltage change rate matrix is ​​subjected to Fourier transformation to obtain voltage periodic fluctuation data, and the current change amount data is subjected to adaptive threshold analysis to obtain current mutation point data; A voltage domain iterative learning model is constructed based on voltage periodic fluctuation data, and the voltage feature weight is obtained through weight update operation; The current mutation point data and the voltage feature weights are cross-operated to obtain fused feature data, and the fused feature data are subjected to feature dimension reduction and normalization to obtain the breakpoint feature vector.

[0028] Specifically, the voltage ratio of adjacent nodes is calculated based on the voltage data in the electrical characteristic matrix. Assume that the voltage data in the electrical characteristic matrix is ,in Indicates The voltage value of each node, is the total number of nodes. The voltage ratio of adjacent nodes is expressed as: ; in Is a node and The first voltage ratio in the series The voltage ratios of all adjacent nodes are calculated using the above formula to obtain the node voltage ratio sequence. The time series analysis of the ratio sequence is carried out to analyze its changing pattern over time. The time series analysis is realized by calculating the rate of change of the ratio over time. Assuming that The ratio sequence is , then the voltage change rate matrix It is expressed as: ; in Is a node In time The voltage change rate, is the time interval. The calculated voltage change rate matrix reflects the dynamic change of the voltage ratio of adjacent nodes. At the same time, according to the current data in the electrical characteristic matrix , establish a current time series model for each node. The current time series is expressed as ,in Indicates time, Represents the node number. The current change is calculated by performing the first-order difference operation on the time series: ; in Representation Node In time The first-order difference can highlight the dynamic characteristics such as current mutation. The voltage change rate matrix Perform Fourier transform to extract its frequency domain features, especially periodic fluctuation information. The Fourier transform formula is: ; in Indicates frequency The Fourier coefficients of is the length of the time series, is an imaginary unit. The result of Fourier transform is voltage periodic fluctuation data , reflecting the main periodic component of the voltage ratio change. At the same time, the current change data Perform adaptive threshold analysis to automatically filter out mutation point data. The core of adaptive threshold analysis is to dynamically adjust the threshold , so that it can adapt to the signal distribution. For the current change data , the identification rule of mutation point is: Mutation point: , ; in and are the mean and standard deviation of the current variation, is the control parameter. In this way, the current mutation point data J is effectively extracted. Based on the extracted voltage periodic fluctuation data , constructing a voltage domain iterative learning model. This model optimizes the voltage feature weights through a weight update mechanism , the objective function is to minimize the weight error: ; in is the target value, are input features, is the weight. Update the weight by gradient descent: ; in is the learning rate, Represents the number of iterations. After multiple iterations, the optimized voltage characteristic weight is obtained. . The current mutation point data and voltage characteristic weight Perform feature cross operation to generate fused feature data The operation rules of feature crossover are: ; in is the first Elements. Fusion feature data By integrating voltage and current information, the breakpoint characteristics are fully reflected. The fused feature data is reduced in dimension and normalized. Dimension reduction is done through principal component analysis, the goal of which is to retain the direction with the largest data variance, and normalization is achieved through zero mean standardization: ; in and are the feature mean and standard deviation respectively. The normalized data The breakpoint feature vector , used for subsequent breakpoint detection and positioning.

[0029] In a specific embodiment, the process of executing step 300 may specifically include the following steps: The breakpoint feature vector and the node topology map are processed by feature mapping to construct the input data of the multi-task neural network, which includes a shared feature extraction network and two parallel task branch networks; The shared feature extraction network is used to extract features from the input data to obtain shared feature data. The shared feature extraction network contains 4 convolutional layers. Each convolutional layer uses a 3×3 convolution kernel and a ReLU activation function. BatchNorm and Dropout layers are added between each layer. The shared feature data is input into the first task branch network for node feature learning to obtain the node feature mapping result. The first task branch network contains 3 fully connected layers, each of which uses the LeakyReLU activation function and weight regularization. The shared feature data is input into the second task branch network for electrical boundary feature learning to obtain the boundary feature mapping result. The second task branch network contains 2 transposed convolutional layers and 1 attention layer, and each layer uses the Tanh activation function. Perform feature fusion on the node feature mapping results and the boundary feature mapping results, calculate the feature weights through the attention mechanism, and obtain weighted feature data; The weighted feature data is input into the decoder network for feature decoding to obtain feature distribution data. The decoder network contains 3 deconvolution layers, each of which uses a 4×4 convolution kernel and a ReLU activation function. Data reconstruction and residual connection operations are performed on the feature distribution data to obtain global feature data, and the global feature data is reduced in dimension through a 1×1 convolution layer. After node-level feature classification, the electrical fault distribution map is obtained.

[0030] Specifically, feature mapping is performed on the breakpoint feature vector and the node topology map to construct the input data of the multi-task neural network. Assume that the breakpoint feature vector is represented by the matrix Indicates that is the number of nodes, is the feature dimension; the node topology graph uses the adjacency matrix express, Representation Node and There is a connection between Indicates no connection. Through the feature mapping function , combining the breakpoint features with the topological structure to generate the input tensor ,in is the number of channels after mapping. Input shared feature extraction network, the network contains 4 convolutional layers, each convolutional layer uses size The convolution kernel ,in is the convolutional layer number, and Represent the number of channels of the previous layer and the current layer respectively. The convolution operation formula is: ; in It is The activation output of the layer, is the bias term. A BatchNorm layer is added after each convolutional layer to normalize the activation value to improve the stability of training. At the same time, the Dropout layer randomly blocks some neurons to prevent overfitting. After the shared feature extraction network, the shared feature data is obtained. ,in and is the size and number of channels of the feature after dimensionality reduction. Input into two task branch networks respectively. In the first task branch network, node feature learning is performed. The network contains 3 fully connected layers, and the output of each layer is calculated by the formula: ; in and are the weight and bias matrices, respectively, It is The input features of the layer. The LeakyReLU activation function introduces a small negative slope , to keep the negative gradient non-zero, the formula is: ; To prevent overfitting, the weight matrix Introduction Regularization constraint. In the second task branch network, electrical boundary feature learning is performed. The network contains 2 transposed convolution layers and 1 attention layer. The purpose of transposed convolution is to upsample the features and restore the original spatial resolution. The calculation formula is: ; in is the output feature map, and the Tanh activation function ensures that the output value is in the range of [-1,1]. The attention layer learns the feature weights To perform feature enhancement, the formula is: ; in Indicates bitwise multiplication of elements. The node feature mapping results and the boundary feature mapping results are feature fused. The fusion weight is calculated through the attention mechanism , the weighted formula is: ; in and They are node features and boundary features respectively. The fusion feature data Input the decoder network for feature decoding. The decoder network contains 3 deconvolution layers, each layer is calculated by the formula: ; in It is The decoded output of the layer. The decoded feature data is reconstructed into global feature data, and the original information of the feature is maintained through a residual connection. The convolution layer reduces the dimension of the global feature data, outputs the node-level feature data, and generates the electrical fault distribution map after classification operation. ,in is the number of categories.

[0031] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Perform voltage gradient calculation and statistical analysis on the electrical fault distribution map to obtain the optimal segmentation threshold, and convert the electrical fault distribution map into a binary feature matrix according to the optimal segmentation threshold to obtain the marked fault area data; The electrical boundary of the marked fault area data is extracted, and the voltage amplitude and change direction are calculated by the voltage gradient operator to obtain the electrical boundary feature data; The electrical boundary feature data is input into the deep feature network to obtain the feature description vector. The deep feature network contains 5 layers of stacked depthwise separable convolution modules. Each depthwise separable convolution module uses a combination of 3×3 depthwise convolution and 1×1 pointwise convolution. The feature description vector is enhanced through the self-attention mechanism to obtain enhanced feature data, and the enhanced feature data is fused with the original fault area data to obtain fused feature data; The electrical characteristics of the fused feature data are reconstructed to obtain reconstructed regional data, and the reconstructed regional data are subjected to feature optimization and data fitting to obtain fault area feature data.

[0032] Specifically, for the electrical fault distribution map Perform voltage gradient calculation and statistical analysis to determine the optimal segmentation threshold. The gradient calculation uses the Sobel operator and The voltage gradients in the transverse and longitudinal directions are calculated respectively, and the formulas are: ; ; in and Respectively indicate points The voltage gradient component at and is the Sobel kernel matrix, which is used to detect changes in the horizontal and vertical directions. The comprehensive gradient amplitude is: ; Through statistical analysis The histogram distribution of , select the segmentation point that makes the gradient change most obvious as the optimal segmentation threshold , which is solved using Otsu's algorithm: ; in is the threshold The corresponding inter-class variance. Using the optimal segmentation threshold , the electrical fault distribution map Convert to a binary feature matrix: ; Among them, the binary matrix Mark the location of the fault area. Electrical boundary extraction is performed, and the voltage gradient operator is used to calculate the voltage amplitude and change direction at the fault area boundary. Calculated by the gradient, and the direction of change It is expressed as: ; The obtained boundary feature data consists of two parts: amplitude and direction, which represent the strength and direction characteristics of the boundary position respectively. The electrical boundary feature data is input into the deep feature network for high-dimensional feature extraction. The deep feature network consists of 5 layers of deep separable convolution modules, each of which includes a A depthwise convolution and a Point convolution. Depth convolution is responsible for extracting local features in the spatial dimension, and its calculation formula is: ; in is the depthwise convolution output, are input features, is the depth convolution kernel. Point convolution is used for inter-channel feature fusion, and the formula is: ; in is the point convolution kernel weight. After 5 layers of stacking, the deep feature network outputs a high-dimensional feature description vector . For the feature description vector Apply the self-attention mechanism for feature enhancement. The self-attention mechanism calculates the similarity weight of each feature with other features. Improve the expressive power of key features. The formula is: ; in and is the query and key representation of the feature vector. The enhanced feature data is: ; in It is the eigenvalue representation. The enhanced feature data With the original fault area data Perform feature fusion, the fusion formula is: ; in and is the fusion weight. Reconstruct electrical features. Upsample the feature data through the deconvolution layer to restore the original resolution. The formula is: ; in is the upsampling factor, is the reconstructed feature. Perform feature optimization and data fitting using the least squares fitting formula: ; in is the target data, by optimizing the loss function , and obtain the optimized fault area characteristic data.

[0033] In a specific embodiment, the process of executing step 500 may specifically include the following steps: The characteristic data of the fault area is divided into electric potential grids to obtain an electric potential sampling point matrix, and the electric potential sampling point matrix is ​​subjected to Gauss-Seidel iterative calculation to obtain discrete electric potential data; The electric field potential energy equation is constructed for discrete electric potential data according to the second-order central difference format, and the electric field intensity distribution matrix is ​​obtained through finite difference operation. Substitute each grid point in the electric field intensity distribution matrix into the Poisson equation, and obtain the electric field Poisson distribution data through boundary condition constraint calculation; A least square fitting objective function is established for the electric field Poisson distribution data, and the electric field fitting coefficient matrix is ​​obtained through gradient descent optimization. The electric field fitting coefficient matrix is ​​input into the Laplace operator model for second-order partial derivative calculation to obtain the electric field curvature distribution map. Perform singular value decomposition on the electric field curvature distribution diagram, obtain the main eigenvector by eigenvalue sorting, and perform extreme point detection on the main eigenvector to obtain a singular point candidate set; The singular point candidate set is spatially aggregated through cluster analysis to obtain the cluster center point, and the cluster center point is subjected to coordinate transformation and precision correction to obtain the breakpoint spatial coordinate data.

[0034] Specifically, the fault area characteristic data Divide into potential sampling point matrix according to uniform grid ,in and is the number of rows and columns of the grid. Each grid point The initial potential value By interpolation or mapping from Get the potential sampling point matrix Perform Gauss-Seidel iteration to solve the distribution of electric potential in equilibrium. The update formula of Gauss-Seidel iteration is: ; in Indicates Iteration midpoint The potential value is iterated until the convergence condition is met. , is the error threshold. After the iteration is completed, the electric field potential energy equation is constructed for the discrete electric potential data using the second-order central difference format. and The components are calculated as: ; in and is the grid spacing. The calculated electric field components are combined into the electric field intensity distribution matrix , each element of which The size is: ; Substitute each grid point in the electric field intensity distribution matrix into the Poisson equation , the numerical solution of Poisson's equation is calculated by finite difference, and the electric field Poisson distribution data is adjusted in combination with boundary condition constraints The boundary conditions are set to fixed potential values ​​(Dirichlet boundary conditions) or fixed electric field strength (Neumann boundary conditions), and the appropriate constraint form is selected according to the actual scenario. The least squares fitting objective function is established to fit the optimal electric field distribution parameters. The objective function is defined as: ; Update the fitting parameter matrix through gradient descent optimization : ; in is the learning rate, is the fitting coefficient matrix. After the fitting is completed, the electric field fitting coefficient matrix is ​​input into the Laplace operator model for second-order partial derivative calculation to generate the electric field curvature distribution diagram. The curvature calculation formula is: ; in is the curvature value, which reflects the local change rate of the electric field distribution. After obtaining the electric field curvature distribution diagram, the characteristic analysis is performed through singular value decomposition. The curvature matrix The decomposition form is: ; in It is a diagonal matrix whose diagonal elements are singular values, representing the main characteristic intensity of the electric field distribution. By sorting the singular values, the main eigenvector is extracted , then Perform extreme point detection. The detection conditions of extreme points are: ; The detected extreme points form a singular point candidate set. Cluster analysis is performed on the singular point candidate set. The candidate points in space are aggregated using algorithms such as K-Means or DBSCAN to obtain the cluster center point. Suppose the singular point set is , the clustering goal is to minimize the intra-cluster distance: ; in It is clusters, is its center point. After clustering is completed, for each center point Perform coordinate transformation, convert it from the grid coordinate system to the actual space coordinate system, and perform accuracy correction in combination with the error model to obtain the space coordinates of the breakpoint

[0035] In a specific embodiment, the execution step establishes a least square fitting objective function for the electric field Poisson distribution data, obtains the electric field fitting coefficient matrix through gradient descent optimization, and inputs the electric field fitting coefficient matrix into the Laplace operator model to calculate the second-order partial derivative. The process of obtaining the electric field curvature distribution map can specifically include the following steps: The electric field Poisson distribution data is expanded by basis function through N-order Legendre polynomials, and the basis function expansion coefficients are obtained according to orthogonality calculation; The least squares fitting objective function is constructed based on the basis function expansion coefficients. The loss function with penalty term is obtained through L2 norm regularization constraint. The loss function with penalty term is differentiated and the gradient direction vector is obtained through forward difference and backward difference operations. The adaptive learning rate is set according to the gradient direction vector, and the electric field fitting coefficient matrix is ​​obtained by iterative operation through the back propagation algorithm. The electric field fitting coefficient matrix is ​​input into the two-dimensional Laplace operator for finite difference to obtain the second-order derivative tensor; The principal curvature of the second-order derivative tensor is calculated, and the curvature characteristic data is obtained through the Gaussian curvature and mean curvature formulas. The curvature characteristic data is reconstructed according to the electric field distribution, and the electric field curvature distribution map is obtained through bicubic interpolation.

[0036] Specifically, the electric field Poisson distribution data is expanded by the Nth-order Legendre polynomials. Legendre polynomials are a set of orthogonal polynomials in the interval [-1,1], which are used to represent the decomposition of complex functions. Assume that the electric field Poisson distribution data is ,in is a two-dimensional coordinate, Legendre polynomial The recursive relation is: ; Will Expanded into the basis function form of Legendre polynomials, it is expressed as: ; in is the coefficient of the basis function expansion, which represents the projection of the Poisson distribution on the Legendre basis. Using the orthogonality of the Legendre polynomials, its expansion coefficients The calculation formula is: ; The above formula is realized by numerical integration. As the core parameter of the basis function expansion. Based on the calculated basis function expansion coefficient , construct the least squares fitting objective function to minimize the electric field Poisson distribution The reconstruction error of . The objective function is defined as: ; To prevent overfitting, we introduce The norm regularization constraint adds a penalty term to the loss function and corrects it to: ; in is the regularization coefficient, which is used to balance the fitting error and the complexity of the coefficient. Derivative the variables to get the gradient direction of each coefficient. Assume Express The gradient of is: ; Gradient calculation is achieved through finite difference operations of forward difference and backward difference to ensure the accuracy of numerical gradient. Adaptive learning rate is set according to the gradient direction , to optimize the update coefficient matrix : ; Learning Rate Dynamic adjustment is performed to ensure iterative convergence. After multiple back-propagation optimizations, the electric field fitting coefficient matrix is ​​finally obtained. . The coefficient matrix obtained by fitting Input the two-dimensional Laplace operator model, use the finite difference method to calculate the second-order partial derivatives, and generate the second-order derivative tensor The two-dimensional Laplacian operator is defined as: ; After discretization, the finite difference calculation of the second-order derivative is: ; ; The calculation results are stored as second-order derivative tensors , each element of which represents the second-order change of the local electric field distribution. The principal curvature is calculated for the second-order derivative tensor, and the curvature feature data is extracted using the Gaussian curvature and mean curvature formulas. Gaussian curvature and mean curvature The formula is: ; Through curvature feature data and Reconstruct the geometric distribution of the electric field. In order to improve the smoothness of the reconstruction, the bicubic interpolation method is used to interpolate the curvature feature data, and the low-resolution curvature data is upgraded to high resolution to generate the electric field curvature distribution map.

[0037] This embodiment also includes: constructing an error compensation model according to the breakpoint space coordinate data, and optimizing the parameters through the deep reinforcement learning algorithm to obtain the corrected detection results, specifically: performing time series analysis and spatial distribution statistics on the breakpoint space coordinate data to obtain coordinate error feature data; inputting the coordinate error feature data into the deep reinforcement learning network, performing feature mapping through the state encoder to obtain the state vector; constructing a Q learning network based on double time difference, performing action value evaluation on the state vector, and obtaining compensation strategy parameters; performing exploration noise injection on the compensation strategy parameters, generating a compensation action sequence through the ε-greedy strategy, and obtaining a candidate compensation plan; inputting the candidate compensation plan into the experience replay buffer pool , empirical data is screened through priority sampling to obtain a training sample set; the target network is updated for the training sample set, and the temporal difference target is calculated through a soft update strategy to obtain a network optimization target; a policy gradient loss function is constructed based on the network optimization target, and the compensation gradient is calculated through a dual Q network architecture to obtain a parameter update vector; the parameter update vector is input into an adaptive moment estimation optimizer, and the compensation model parameters are obtained through a gradient normalization operation; the stability of the compensation model parameters is evaluated, and the convergence domain is calculated through Lyapunov function analysis to obtain a parameter feasible domain; the optimal compensation parameters in the parameter feasible domain are applied to the original coordinate data, and the corrected detection results are obtained through a coordinate transformation operation.

[0038] The above describes the LED module breakpoint detection and positioning method in the embodiment of the present application. The following describes the LED module breakpoint detection and positioning device 10 in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the LED module breakpoint detection and positioning device 10 includes: The acquisition module 11 is used to collect voltage and current data of the series-parallel nodes of the LED module to obtain an electrical characteristic matrix and a node topology diagram; The calculation module 12 is used to iteratively calculate the electrical characteristic matrix, process the voltage difference and current change between adjacent nodes, and obtain the breakpoint characteristic vector; The operation module 13 is used to input the breakpoint feature vector and the node topology map into the multi-task neural network for multi-layer convolution operation to obtain an electrical fault distribution map; The extraction module 14 is used to perform feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data; The building module 15 is used to build an electric field distribution model for the fault area characteristic data, calculate the electric field singular points by the least square method, and obtain the breakpoint space coordinate data.

[0039] Through the cooperation of the above components, by establishing a multi-task neural network and a deep feature extraction model, combined with the voltage domain iterative learning strategy, the accuracy of breakpoint detection is significantly improved. The electric field distribution model and the least squares optimization algorithm are used to achieve accurate positioning of the breakpoint position and improve the spatial resolution of fault location. The combination of electrical feature analysis and deep learning methods breaks through the limitations of traditional single detection methods and achieves rapid identification of breakpoint faults in complex series-parallel structures. By introducing adaptive feature extraction and multi-level data fusion technology, the system's resistance to environmental interference is enhanced, making the detection results more stable and reliable. Based on the distributed node data acquisition solution, combined with an efficient feature processing algorithm, real-time monitoring of large-scale LED modules is achieved. Through the application of deep separable convolution modules and attention mechanisms, the computational complexity of the algorithm is reduced while ensuring detection performance. The electric field curvature distribution analysis method is used to improve the system's ability to identify early fault signs and can provide early warning of potential breakpoint fault risks.

[0040] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0041] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned LED module breakpoint detection and positioning methods.

[0042] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0043] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned LED module breakpoint detection and positioning methods.

[0044] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0045] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0046] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned LED module breakpoint detection and positioning method, and will not be repeated here.

[0047] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the LED module breakpoint detection and positioning method provided in the embodiment of the present application.

[0048] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0049] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0050] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0051] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting and locating breakpoints of an LED module, characterized in that: The method comprises: Collect voltage and current data of the series and parallel nodes of the LED module to obtain the electrical characteristic matrix and node topology diagram; Iteratively calculating the electrical characteristic matrix, processing the voltage difference and current change between adjacent nodes, and obtaining a breakpoint characteristic vector; Inputting the breakpoint feature vector and the node topology map into a multi-task neural network for multi-layer convolution operation to obtain an electrical fault distribution map; Performing feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data; An electric field distribution model is established for the characteristic data of the fault area, and the electric field singular points are calculated by the least square method to obtain the breakpoint space coordinate data.

2. The LED module breakpoint detection and positioning method according to claim 1, characterized in that: The voltage and current data of the series-parallel nodes of the LED modules are collected to obtain an electrical characteristic matrix and a node topology diagram, including: Establish a coordinate system and perform space mapping calculations for each series-parallel node in the LED module to obtain a node coordinate data set; Performing voltage and current acquisition and synchronous sampling conversion on each node in the node coordinate data set to obtain original electrical data; Performing wavelet decomposition operation and high-frequency coefficient threshold processing on the original electrical data to obtain signal data after noise reduction; The signal data after noise reduction is processed in segments to obtain segmented characteristic data, and voltage and current values ​​in the segmented characteristic data are subjected to zero mean standardization processing and variance normalization calculation to obtain standardized electrical characteristics; Performing spatial correlation operation and feature mapping transformation on the standardized electrical features and the node coordinate data set to obtain a node feature correlation table; Based on the series-parallel topological structure of the LED modules, the node feature association table is subjected to connectivity analysis and adjacency matrix calculation to obtain a node topological graph; The node feature association table is reconstructed into an M×N dimensional feature structure, and an electrical feature matrix is ​​obtained through data rearrangement operation.

3. The LED module breakpoint detection and positioning method according to claim 2, characterized in that: The iterative calculation of the electrical characteristic matrix, processing the voltage difference and current change between adjacent nodes, and obtaining the breakpoint characteristic vector includes: Calculating the voltage ratios of adjacent nodes according to the voltage data in the electrical characteristic matrix to obtain a node voltage ratio sequence, and performing a time series analysis on the node voltage ratio sequence to obtain a voltage change rate matrix; Establishing a current time series model according to the current data in the electrical characteristic matrix, and obtaining current variation data through first-order difference operation; Performing Fourier transformation on the voltage change rate matrix to obtain voltage periodic fluctuation data, and performing adaptive threshold analysis on the current change amount data to obtain current mutation point data; Building a voltage domain iterative learning model based on the voltage periodic fluctuation data, and obtaining a voltage feature weight through a weight update operation; The current mutation point data and the voltage feature weight are subjected to feature cross operation to obtain fused feature data, and the fused feature data are subjected to feature dimension reduction and normalization processing to obtain a breakpoint feature vector.

4. The LED module breakpoint detection and positioning method according to claim 3, characterized in that: The step of inputting the breakpoint feature vector and the node topology map into a multi-task neural network for multi-layer convolution operation to obtain an electrical fault distribution map includes: The breakpoint feature vector and the node topology map are processed by feature mapping to construct input data of a multi-task neural network, wherein the multi-task neural network includes a shared feature extraction network and two parallel task branch networks; Extracting features from the input data through the shared feature extraction network to obtain shared feature data, wherein the shared feature extraction network includes 4 convolutional layers, each convolutional layer uses a 3×3 convolution kernel and a ReLU activation function, and a BatchNorm layer and a Dropout layer are added between each layer; Inputting the shared feature data into the first task branch network for node feature learning to obtain a node feature mapping result, wherein the first task branch network comprises three fully connected layers, each layer using a LeakyReLU activation function and weight regularization; Inputting the shared feature data into a second task branch network for electrical boundary feature learning to obtain a boundary feature mapping result, wherein the second task branch network comprises two transposed convolutional layers and one attention layer, and each layer uses a Tanh activation function; Performing feature fusion on the node feature mapping result and the boundary feature mapping result, calculating feature weights through an attention mechanism, and obtaining weighted feature data; Inputting the weighted feature data into a decoder network for feature decoding to obtain feature distribution data, wherein the decoder network comprises three deconvolution layers, each layer using a 4×4 convolution kernel and a ReLU activation function; Data reconstruction and residual connection operations are performed on the feature distribution data to obtain global feature data, and the global feature data is subjected to dimensionality reduction processing through a 1×1 convolution layer. After node-level feature classification, an electrical fault distribution map is obtained.

5. The LED module breakpoint detection and positioning method according to claim 4, characterized in that: The performing feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data includes: Performing voltage gradient calculation and statistical analysis on the electrical fault distribution map to obtain an optimal segmentation threshold, and converting the electrical fault distribution map into a binary feature matrix according to the optimal segmentation threshold to obtain marked fault area data; Extracting electrical boundaries from the marked fault area data, calculating voltage amplitude and change direction using a voltage gradient operator, and obtaining electrical boundary feature data; Inputting the electrical boundary feature data into a deep feature network to obtain a feature description vector, wherein the deep feature network comprises 5 layers of stacked depthwise separable convolution modules, each of which uses a combination of 3×3 depthwise convolution and 1×1 pointwise convolution; Performing feature enhancement on the feature description vector through a self-attention mechanism to obtain enhanced feature data, and performing feature fusion on the enhanced feature data with the original fault area data to obtain fused feature data; The fused feature data is subjected to electrical feature reconstruction to obtain reconstructed regional data, and the reconstructed regional data is subjected to feature optimization and data fitting to obtain fault region feature data.

6. The LED module breakpoint detection and positioning method according to claim 5, characterized in that: The method of establishing an electric field distribution model for the fault area characteristic data, calculating the electric field singular points by the least square method, and obtaining the breakpoint space coordinate data includes: Dividing the fault area characteristic data into electric potential grids to obtain an electric potential sampling point matrix, and performing Gauss-Seidel iterative calculation on the electric potential sampling point matrix to obtain discrete electric potential data; constructing an electric field potential energy equation for the discrete electric potential data according to a second-order central difference format, and obtaining an electric field intensity distribution matrix through a finite difference operation; Substituting each grid point in the electric field intensity distribution matrix into the Poisson equation, and obtaining the electric field Poisson distribution data through boundary condition constraint calculation; A least square fitting objective function is established for the electric field Poisson distribution data, an electric field fitting coefficient matrix is ​​obtained by gradient descent optimization, and the electric field fitting coefficient matrix is ​​input into a Laplace operator model for second-order partial derivative calculation to obtain an electric field curvature distribution diagram; Performing singular value decomposition on the electric field curvature distribution diagram, obtaining a main eigenvector by eigenvalue sorting, and performing extreme point detection on the main eigenvector to obtain a singular point candidate set; The singular point candidate set is spatially aggregated through cluster analysis to obtain cluster center points, and the cluster center points are subjected to coordinate transformation and precision correction to obtain breakpoint spatial coordinate data.

7. The LED module breakpoint detection and positioning method according to claim 6, characterized in that: The least square fitting objective function is established for the electric field Poisson distribution data, an electric field fitting coefficient matrix is ​​obtained by gradient descent optimization, and the electric field fitting coefficient matrix is ​​input into the Laplace operator model for second-order partial derivative calculation to obtain an electric field curvature distribution diagram, including: Performing basis function expansion on the electric field Poisson distribution data using N-order Legendre polynomials, and obtaining basis function expansion coefficients according to orthogonality calculation; A least squares fitting objective function is constructed based on the basis function expansion coefficients, a loss function with a penalty term is obtained through an L2 norm regularization constraint, and the loss function with a penalty term is differentiated, and a gradient direction vector is obtained through forward difference and backward difference operations; An adaptive learning rate is set according to the gradient direction vector, an iterative operation is performed through a back propagation algorithm to obtain an electric field fitting coefficient matrix, and the electric field fitting coefficient matrix is ​​input into a two-dimensional Laplace operator for finite difference to obtain a second-order derivative tensor; The principal curvature of the second-order derivative tensor is calculated, and curvature characteristic data is obtained through Gaussian curvature and mean curvature formulas. The curvature characteristic data is reconstructed according to the electric field distribution, and the electric field curvature distribution map is obtained through bicubic interpolation.

8. A LED module breakpoint detection and positioning device, characterized in that: The device is used to perform the LED module breakpoint detection and positioning method according to any one of claims 1 to 7, comprising: The acquisition module is used to collect voltage and current data of the series and parallel nodes of the LED module to obtain the electrical characteristic matrix and node topology diagram; A calculation module, used for iteratively calculating the electrical characteristic matrix, processing the voltage difference and current variation between adjacent nodes, and obtaining a breakpoint characteristic vector; A calculation module, used for inputting the breakpoint feature vector and the node topology map into a multi-task neural network to perform multi-layer convolution operations to obtain an electrical fault distribution map; An extraction module, used for performing feature segmentation and deep feature extraction on the electrical fault distribution map to obtain fault area feature data; A module is established to establish an electric field distribution model for the characteristic data of the fault area, calculate the electric field singular points by the least square method, and obtain the breakpoint space coordinate data.

9. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the LED module breakpoint detection and positioning method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, the LED module breakpoint detection and positioning method according to any one of claims 1 to 7 is implemented.

Images