Arrester abnormal flashover identification method, system and device and storage medium

Abstract

The invention discloses a lightning arrester abnormal flashover identification method, system and device and a storage medium, and the method comprises the steps: collecting lightning arrester image data of a power transmission line under different operation conditions, and carrying out the preprocessing of the lightning arrester image data; and presetting a lightning arrester fault identification model, wherein the lightning arrester fault identification model detects the abnormal discharge condition of the lightning arrester based on the preprocessed image data. Spatial features are extracted from each frame of image based on a convolutional neural network, and a channel attention mechanism is introduced to generate a feature channel attention weight matrix so as to enhance important feature map channel features; adopting a Pedal algorithm to optimize multiple parameters of the new model; according to the lightning arrester abnormal discharge detection system, the pictures are classified through the output layer, the discharge phenomenon is detected, high precision and robustness can be displayed in a complex environment, abnormal discharge of the lightning arrester can be effectively recognized and detected, and the safety and reliability of a power system can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system equipment status monitoring and fault diagnosis, and in particular to a lightning arrester abnormal flashover identification method, system, equipment and storage medium. Background Art

[0002] With the continuous development of power systems and the increasing demand for lightning arrester reliability, traditional detection methods are unable to meet the needs of real-time monitoring. Early reliance on manual and simple electrical testing was inefficient and inaccurate. Although online monitoring technology has made progress, it is susceptible to interference and has difficulty identifying early faults.

[0003] In recent years, image processing and deep learning techniques have been applied to lightning arrester fault detection. However, image recognition-based methods are significantly affected by the environment, and traditional image recognition algorithms have low accuracy. While convolutional neural networks (CNNs), a deep learning framework, have certain advantages, they also have limitations in processing complex backgrounds and subtle fault features. Traditional CNNs treat all image regions equally, making it difficult to focus on the key features of abnormal lightning arrester flashovers in complex environments. For example, in situations with drastic lighting changes and complex backgrounds (such as the presence of debris or other equipment), background information can be misinterpreted as fault features, making it difficult to extract abnormal flashover features and affecting recognition accuracy. In complex environments, the instability of image quality necessitates more complex image preprocessing, such as dynamic denoising and illumination normalization, to improve the robustness of subsequent models. Most studies have employed traditional image recognition models. While these models are now mature, they also exhibit numerous issues, such as simple model structures, incomplete predictions, poor parameterization, and inaccurate recognition results. Continuing to use traditional models will undoubtedly inherit the flaws of previous research, and model algorithms require innovation. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a method, system, device and storage medium for identifying abnormal flashover of a lightning arrester to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a method for identifying abnormal flashover of a lightning arrester, comprising:

[0007] Collecting lightning arrester image data under different operating conditions of the transmission line and preprocessing the lightning arrester image data;

[0008] A lightning arrester fault identification model is preset, and the lightning arrester fault identification model is based on the preprocessed image data to detect abnormal discharge of the lightning arrester.

[0009] As a preferred solution of the method for identifying abnormal flashover of a lightning arrester according to the present invention, the preprocessing of the lightning arrester image data includes: integrating the collected lightning arrester images into an image data set, and cropping each image in the image data set;

[0010] The cropped images are scaled to a uniform size and the image pixel values ​​are normalized, and the generated image sequence is used as input data for the arrester fault recognition model.

[0011] As a preferred solution of the lightning arrester abnormal flashover identification method described in the present invention, the lightning arrester fault identification model includes: an input layer, an output layer, a convolution layer, a pooling layer, a fully connected layer and a Softmax layer, as well as an integrated channel attention module.

[0012] As a preferred solution of the method for identifying abnormal flashover of a lightning arrester according to the present invention, the lightning arrester fault identification model comprises: optimizing the parameters of the lightning arrester fault identification model before detecting abnormal discharge of the lightning arrester based on the preprocessed image data;

[0013] Initialize the model parameters, use accuracy as the objective function, and calculate the fitness of each individual;

[0014] According to the optimization algorithm update rules, the positions of individuals in the population are updated globally, including the local search in the first stage and the local search in the second stage;

[0015] If the current update iteration number does not reach the maximum number of iterations, continue global update and local search; if the current update iteration number reaches the maximum number of iterations, output the optimal parameter combination of the model;

[0016] An arrester fault identification optimization model is established based on the optimal parameter combination of the model.

[0017] As a preferred solution of the method for identifying abnormal flashover of a lightning arrester according to the present invention, the lightning arrester fault identification optimization model includes: performing spatial feature extraction and feature dimension reduction on input data;

[0018] The insulator and its discharge feature map of the input image are extracted and the spatial dimension of the insulator and its discharge feature map is compressed to obtain a one-dimensional feature vector of length C′ describing the insulator discharge state, which is used to characterize the operating state of the arrester.

[0019] As a preferred solution of the method for identifying abnormal flashover of a lightning arrester according to the present invention, the method further includes: based on different feature channels in the image, generating a feature vector after channel attention weighting to express it as follows:

[0020] A s =S(W s V+b s )

[0021] V′=A s V, V′∈R C′ ×H′×W′

[0022] Among them, W s ∈R C′ ×C′ is the weight matrix, b s is the bias term, S represents the Sigmoid function; the output A s ∈R C′ is the feature map channel attention weight, size is C′.

[0023] The beneficial effect of this preferred technical solution is that by generating an attention weight matrix and allocating attention to the feature map channels, the model can focus more on the characteristics of the arrester discharge phenomenon in a complex environment, thereby improving the accuracy of feature extraction.

[0024] As a preferred solution of the method for identifying abnormal flashover of a lightning arrester according to the present invention, the lightning arrester fault identification optimization model outputs the identification result expressed as:

[0025] y=Softmax(W f ·V′+b f )

[0026] Among them, W f is the weight matrix of the output layer, b f is the bias term of the output layer.

[0027] The beneficial effects of this preferred technical solution are that it extracts spatial features from each frame of the image based on a convolutional neural network, and introduces a channel attention mechanism to generate a feature channel attention weight matrix to enhance the channel features of important feature maps, further optimize the model parameters, and classify images through the output layer. It can show high accuracy and robustness in complex environments, effectively identify and detect abnormal discharge of lightning arresters, and help improve the safety and reliability of the power system.

[0028] In a second aspect, the present invention provides a lightning arrester abnormal flashover identification system, comprising:

[0029] A data acquisition module is used to collect lightning arrester image data under different operating conditions of the transmission line and pre-process the lightning arrester image data;

[0030] The fault identification module is used to preset a lightning arrester fault identification model, wherein the lightning arrester fault identification model is based on the preprocessed image data to detect abnormal discharge of the lightning arrester.

[0031] In a third aspect, the present invention provides an electronic device, comprising:

[0032] memory and processor;

[0033] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for identifying abnormal flashover of a lightning arrester are implemented.

[0034] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for identifying abnormal flashover of a lightning arrester.

[0035] Compared with the existing technology, the present invention has the following advantages: the present invention generates an attention weight matrix through a channel attention mechanism and allocates attention to the feature map channels, so that the model can focus more on the characteristics of the arrester discharge phenomenon in a complex environment, and can improve the accuracy of feature extraction. Compared with the traditional convolutional neural network, the attention mechanism can adjust the attention to different feature channels according to the feature weights, which solves the difficulty of traditional models in extracting key features in complex environments.

[0036] This paper uses a temporal attention mechanism to dynamically weight each time step, emphasizing the importance of discharge phenomena at different time points. Even in the presence of noise, this mechanism can filter out irrelevant temporal information, allowing insufficient preprocessing to be compensated by the model's own temporal modeling. Even if preprocessing yields suboptimal results, the attention mechanism can still automatically adjust focus across spatiotemporal dimensions, thereby improving recognition robustness.

[0037] The present invention introduces a channel attention mechanism into the convolutional neural network and performs weighted processing on the key features of the feature map channel through the channel attention mechanism. Compared with the traditional model, it has stronger expressive ability and can perform more detailed feature capture and accurate recognition in complex scenarios, further enhancing the effect of feature extraction. It solves the problem of inaccurate recognition results caused by the simple structure of the traditional model. Combined with the gazelle optimization algorithm to optimize the model parameters, the accuracy and reliability of abnormal flashover identification of lightning arresters are further improved.

[0038] In summary, the present invention extracts spatial features from each frame of image based on a convolutional neural network, and introduces a channel attention mechanism to generate a feature channel attention weight matrix to enhance the channel features of important feature maps; the gazelle algorithm is used to optimize multiple parameters of the new model; the output layer is used to classify images and detect discharge phenomena, which can show high accuracy and robustness in complex environments, can effectively identify and detect abnormal discharge of lightning arresters, and help improve the safety and reliability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0040] Figure 1 A schematic diagram of a method flow of a lightning arrester abnormal flashover identification method, system, device, and storage medium according to an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a process flow for implementing a lightning arrester abnormal flashover identification method, system, device, and storage medium according to an embodiment of the present invention;

[0042] Figure 3 This is a diagram of the CAM-CNN model architecture that introduces a channel attention mechanism for a lightning arrester abnormal flashover identification method, system, device, and storage medium described in one embodiment of the present invention. DETAILED DESCRIPTION

[0043] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0044] Example 1, reference Figures 1 to 3 , is an embodiment of the present invention, which provides a method for identifying abnormal flashover of a lightning arrester, comprising:

[0045] S100: collecting lightning arrester image data under different operating conditions of the transmission line and preprocessing the lightning arrester image data;

[0046] S200: Preset a lightning arrester fault recognition model, which is based on preprocessed image data to detect abnormal discharge of the lightning arrester.

[0047] It should be noted that traditional CNN treats all areas of the image equally, and it is difficult to focus on the key features of abnormal flashover of lightning arresters in complex environments. For example, when the lighting changes drastically and the background is complex (such as the presence of debris or interference from other equipment), it is easy to misjudge background information as fault features, making it difficult to extract abnormal flashover features and affecting recognition accuracy. In complex environments, due to the instability of image quality, more complex image preprocessing is required, such as dynamic denoising and lighting normalization, to improve the robustness of subsequent models. Most studies use traditional image recognition models. These model technologies are now mature, but they have also exposed many problems, such as simple model structure, incomplete prediction results, poor parameters, inaccurate recognition results, etc. The use of traditional models will undoubtedly inherit the defects of previous studies, and the model algorithm needs to be innovated.

[0048] This application allocates attention to feature map channels by generating an attention weight matrix. This allows the model to focus more closely on the characteristics of the arrester discharge phenomenon in complex environments, significantly improving the accuracy of feature extraction. Compared to traditional CNNs, the attention mechanism can adjust the focus on different feature channels based on feature weights, solving the difficulties faced by traditional models in extracting key features in complex environments. The temporal attention mechanism can dynamically weight each time step, emphasizing the importance of the discharge phenomenon at different time points. Even in the presence of noise, this mechanism can filter out irrelevant temporal information, allowing the problem of insufficient preprocessing to be compensated by the model's own temporal modeling. This means that even if the preprocessing stage is not ideal, the model can still automatically adjust the focus in the spatiotemporal dimensions through the attention mechanism, thereby improving the robustness of recognition. The Gazelle Optimization Algorithm (GOA) is used to optimize the CNN parameters that introduce the Channel Attention Mechanism (CAM) to construct a CAM-CNN model. The key features of the feature map channels are weighted using CAM. This multi-layered modeling approach offers greater expressiveness than traditional models, enabling more detailed feature capture and accurate recognition in complex scenarios. Through deep feature extraction and channel-wise attention weighting, it overcomes the inaccurate recognition results of traditional models due to their simple structure.

[0049] In the embodiment of the present application, pre-processing the arrester image data in step S100 includes: integrating the collected arrester images into an image data set, and cropping each image in the image data set;

[0050] The cropped images are scaled to a uniform size and the image pixel values ​​are normalized. The generated image sequence is used as the input data of the arrester fault recognition model.

[0051] In an optional embodiment, the acquisition of lightning arrester image data in step S100 is performed by deploying high-definition cameras, ultraviolet imagers and other equipment at key locations of the transmission line where the lightning arrester is installed. The high-definition camera is used to collect visible light images of the lightning arrester under different lighting and weather conditions to record the appearance of the lightning arrester; the ultraviolet image generated by the lightning arrester during discharge is collected by the ultraviolet imager to capture weak discharge signals. Data collection is performed in different time periods and different transmission line operating conditions (different loads, different seasons, etc.) to ensure that the collected data covers various situations in which the lightning arrester operates normally and may have abnormal flashovers.

[0052] In an optional embodiment, the pre-processing process in step S100 includes: integrating the taken photos together to form a unified image dataset {I}. For each image I in the dataset t Crop to focus on the insulator and its surrounding area, excluding irrelevant background.

[0053] Scale to a uniform size to ensure image consistency when inputting to the model, i.e.:

[0054] I∈R H×W×C

[0055] Among them, H and W are height and width respectively, and C is the number of channels, which can be 3.

[0056] Normalization is performed so that the pixel values ​​of the image are within the standard range of [0,1], that is:

[0057]

[0058] Where I′ is the normalized image, μ and σ are the mean and standard deviation of the image pixels, respectively.

[0059] Use the median filter image processing technique to remove the noise caused by thunderstorms and adjust the contrast to enhance the visibility of the discharge phenomenon, namely:

[0060] I″=MedianFilter(I′)

[0061] Finally, the processed image sequence {I″} is used as input.

[0062] In another optional embodiment, in the pre-processing of step S100 , Gaussian filtering or mean filtering may be used to remove noise caused by thunderstorms.

[0063] In an embodiment of the present application, the arrester fault identification model in step S200 includes: an input layer, an output layer, a convolution layer, a pooling layer, a fully connected layer and a Softmax layer, as well as an integrated channel attention module.

[0064] In an optional embodiment, in order to maintain the size consistency of the input and output feature maps, the convolutional layer is set to a zero-padded operation; in order to prevent the model from overfitting and improve the generalization ability of the model, the Dropout technology is adopted, and the deletion ratio is set to 0.1, that is, 10% of the neurons are randomly discarded during the training process.

[0065] In an optional embodiment, the model optimizer selects the Adam optimizer, which can automatically adjust the learning rate according to the training situation of the model, and the initial learning rate is set to 0.001. During the training process, the learning rate will be dynamically adjusted according to a certain strategy to balance the convergence speed and training effect of the model. The maximum number of training times of the model is set to 100 times. During the actual training process, the number of training times can be adjusted autonomously according to the actual results of the training such as the convergence of the loss function and the performance on the validation set to achieve the best training effect. At the same time, in order to further prevent the occurrence of overfitting, the L2 regularization method is adopted to constrain the parameters of the model to make the model more concise and generalized.

[0066] In the embodiment of the present application, before the arrester fault identification model is used to detect abnormal discharge of the arrester based on the pre-processed image data in step S200, the following steps are included: optimizing the arrester fault identification model parameters;

[0067] Initialize the model parameters, use accuracy as the objective function, and calculate the fitness of each individual;

[0068] According to the optimization algorithm update rules, the positions of individuals in the population are updated globally, including the local search in the first stage and the local search in the second stage;

[0069] If the current update iteration number does not reach the maximum number of iterations, continue global update and local search; if the current update iteration number reaches the maximum number of iterations, output the optimal parameter combination of the model;

[0070] An arrester fault identification optimization model is established based on the optimal parameter combination of the model.

[0071] In an optional embodiment, the process of optimizing the CAM-CNN model parameters using the Gazelle Optimization Algorithm (GOA) includes steps A1 to A6:

[0072] Step A1: GOA initialization and parameter setting: Assume the Gazelle optimization algorithm population size N and randomly generate an initial population. Each individual i represents a set of CAM-CNN parameter combinations, including the number of convolution kernels, the convolution kernel size, and the initial value of the weight matrix in the channel attention mechanism. Set the maximum number of iterations to T and the initial number of iterations t = 0.

[0073] Step A2: Construct a fitness function with the accuracy of arrester abnormal flashover recognition as the target. Divide the preprocessed data into a training set and a validation set. Use the current individual parameter combination to process the training set data. Use the CAM-CNN model to extract features and identify abnormal flashovers. Compare the recognition results with the true labels. Calculate the abnormal flashover recognition accuracy on the validation set and use this as the fitness value, which is expressed as:

[0074]

[0075] Among them, TP is the number of true positives, TN is the number of true negatives, FP is the number of false positives, and FN is the number of false negatives.

[0076] Step A3: Population Update, Global Search Phase: Based on the GOA algorithm's update rules, the positions of individuals in the population are updated globally, adjusting the CAM-CNN parameter combination. This phase simulates gazelles grazing freely without predators or being tracked by predators, and the gazelles adopt Brownian motion. The update formula is as follows:

[0077]

[0078] in, is the position (parameter combination) of individual gazelle i at the tth iteration, S represents the movement speed of individual gazelle i, and R is a random number between [0,1]. A new value is regenerated at each update to introduce randomness so that individuals move to different degrees during updates. Indicates the individual with the best position in the current population, and the rest of the individuals should move towards this position; R B Represents a random vector based on Brownian motion, each component of which satisfies the normal (Gaussian) probability distribution function as follows:

[0079]

[0080] Among them, x is a random variable, μ is the mean, which indicates the center position of the distribution, and σ is the standard deviation, which indicates the degree of dispersion of the data. The larger the σ value, the more dispersed the data, and the smaller the σ value, the more concentrated the data.

[0081] Step A4: The first phase of local search; this phase simulates the escape behavior of gazelles using the Wright flight method after discovering a predator. The Lévy flight allows individuals to move randomly with larger strides, which helps them escape the local optimum. The position update formula is as follows:

[0082]

[0083] Among them, θ is -1 or 1, indicating two directions of movement, which can be randomly selected to determine whether to move in the forward or reverse direction, R L represents a random vector based on the Wright distribution, whose probability distribution function and related parameters are determined according to the definition of the standard Levy flight as follows:

[0084]

[0085] Among them, γ is the stability index, which determines the tail behavior of the distribution, and α is the scale parameter, which affects the width of the distribution.

[0086] Step A5: The second phase of local search; this phase simulates the escape behavior of the gazelle using Brownian motion after discovering the predator. The position update formula is as follows:

[0087]

[0088] Among them, CF represents the cumulative effect of the predator. As the iteration progresses, the value of CF will change, reflecting the dynamic changes of the predator's influence and further refining the search.

[0089] Step A6: If the maximum number of iterations T has been reached, stop the iteration; otherwise, jump back to step A3 until the maximum number of iterations is reached to obtain the optimal CAM-CNN parameter combination.

[0090] In the embodiment of the present application, the arrester fault identification optimization model in step S200 includes: performing spatial feature extraction and feature dimensionality reduction on the input data;

[0091] The insulator and its discharge feature map of the input image are extracted and the spatial dimension of the insulator and its discharge feature map is compressed to obtain a one-dimensional feature vector of length C′ describing the insulator discharge state, which is used to characterize the operating state of the arrester.

[0092] In an optional embodiment, feature extraction: Select appropriate convolutional layers and pooling layers to capture subtle discharge signals and complex background features. Input each preprocessed frame image into the convolutional neural network to extract the discharge feature map F∈R of the insulator and its surroundings in the image. C′ ×H′×W′, where H′ and W′ are the height and width after processing by the convolutional layer and the pooling layer, respectively, and C′ is the number of feature map channels, that is:

[0093] F=CNN(I″)

[0094] It should be noted that the number of channels C′ in a feature map is determined by the designed CNN architecture. In convolutional neural networks, convolution kernels are used to extract specific features such as color, edge features, and texture features. Each kernel extracts only one feature. Therefore, the number of channels in a feature map is the number of features extracted and is determined by the number of kernels. There is no universal standard for determining the number of kernels; this is usually determined by balancing model performance and resource constraints, and can be adjusted through experience and experimentation.

[0095] Furthermore, the spatial dimension of the CNN output feature map F is compressed to obtain a one-dimensional feature vector V∈R with a length of C′ describing the discharge state of the insulator. C′ , this eigenvector can effectively describe the state of the arrester in a complex environment, namely:

[0096]

[0097] It is important to note that by focusing on different feature channels in the image, the model can enhance the importance of features in specific channels and highlight the areas of discharge phenomena by generating a feature vector V′ weighted by channel attention. The squeeze excitation module adaptively recalibrates the channel characteristics of the feature map, enabling the model to enhance the relevant features of insulator discharge.

[0098] In the embodiment of the present application, step S200 further includes: based on different feature channels in the image, generating a feature vector after channel attention weighting is expressed as:

[0099] A s =S(W s V+b s )

[0100] V′=A s V, V′∈R C′ ×H′×W′

[0101] Among them, W s ∈R C′ ×C′ is the weight matrix, b s is the bias term, S represents the Sigmoid function; the output A s ∈R C′ is the feature map channel attention weight, size is C′.

[0102] Furthermore, V′ is input into the fully connected layer and the category probability distribution h is generated through nonlinear transformation, which is expressed as:

[0103] h=ReLU(W o ·V′+b o )

[0104] Among them, W o is the weight matrix of the fully connected layer, b o is the bias term of the fully connected layer.

[0105] In the embodiment of the present application, the arrester fault identification optimization model output identification result in step S200 is expressed as:

[0106] y=Softmax(W f ·V′+b f )

[0107] Among them, W f is the weight matrix of the output layer, b f is the bias term of the output layer.

[0108] Example 2, reference Figures 1 to 3 , is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a lightning arrester abnormal flashover identification system, comprising:

[0109] The data acquisition module is used to collect the arrester image data under different operating conditions of the transmission line and pre-process the arrester image data;

[0110] The fault identification module is used to preset a lightning arrester fault identification model, which is based on preprocessed image data to detect abnormal discharge of the lightning arrester.

[0111] Specifically, when executed, each module of the lightning arrester abnormal flashover identification system of this embodiment implements the steps of the lightning arrester abnormal flashover identification method in Example 1, for example:

[0112] In one embodiment, the lightning arrester abnormal flashover identification system may perform the following steps:

[0113] The collected arrester images are integrated into an image dataset, and each image in the image dataset is cropped; the cropped images are scaled to a uniform size, and the image pixel values ​​are normalized. The generated image sequence is used as the input data of the arrester fault recognition model.

[0114] The arrester fault recognition model includes: input layer, output layer, convolution layer, pooling layer, fully connected layer and Softmax layer as well as an integrated channel attention module.

[0115] Optimize the parameters of the arrester fault identification model;

[0116] Initialize the model parameters, use accuracy as the objective function, and calculate the fitness of each individual;

[0117] According to the optimization algorithm update rules, the positions of individuals in the population are updated globally, including the local search in the first stage and the local search in the second stage;

[0118] If the current update iteration number does not reach the maximum number of iterations, continue global update and local search; if the current update iteration number reaches the maximum number of iterations, output the optimal parameter combination of the model;

[0119] An arrester fault identification optimization model is established based on the optimal parameter combination of the model.

[0120] Perform spatial feature extraction and feature dimensionality reduction on input data;

[0121] The insulator and its discharge feature map of the input image are extracted and the spatial dimension of the insulator and its discharge feature map is compressed to obtain a one-dimensional feature vector of length C′ describing the insulator discharge state, which is used to characterize the operating state of the arrester.

[0122] Based on the different feature channels in the image, the feature vector after generating channel attention weights is expressed as:

[0123] A s =S(W s V+b s )

[0124] V′=A s V, V′∈R C′ ×H′×W′

[0125] Among them, W s ∈R C′ ×C′ is the weight matrix, b s is the bias term, S represents the Sigmoid function; the output A s ∈R C′ is the feature map channel attention weight, size is C′.

[0126] The output identification result of the lightning arrester fault identification optimization model is expressed as:

[0127] y=Softmax(W f ·V′+b f )

[0128] Among them, W f is the weight matrix of the output layer, b f is the bias term of the output layer.

[0129] This embodiment further provides an electronic device applicable to the method for identifying abnormal flashover of a lightning arrester, including:

[0130] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method for identifying abnormal flashover of a lightning arrester as proposed in the above embodiment.

[0131] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the method for identifying abnormal flashover of a lightning arrester as proposed in the above embodiment.

[0132] The storage medium proposed in this embodiment and the method for identifying abnormal flashover of a lightning arrester proposed in the above embodiment belong to the same inventive concept. For technical details not described in detail in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0133] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0134] Example 3, reference Figure 3 , is an embodiment of the present invention, and this embodiment verifies the beneficial effects of the present invention through scientific comparative experiments.

[0135] The image acquisition process of this embodiment fully considers the diversity of environmental factors, and the scenes from different environments are captured by cameras at different angles, ensuring the richness and comprehensiveness of the data, which can truly reflect the status of the lightning arrester under various actual operating conditions. This embodiment collects a total of about 6,000 high-resolution images, covering a rich variety of lightning arrester image information. The image sources are wide-ranging and representative. Among them, 4,000 images are collected when the lightning arrester is in a discharge state, and the other 2,000 images correspond to normal states. In terms of data set division, 70% of the data is divided into a training set for the model training process, so that the model can learn the characteristic patterns of the normal and abnormal states of the lightning arrester from a large amount of data; 20% of the data is used as a validation set, which is used to adjust the model's hyperparameters during the model training process, evaluate the model's performance, and prevent the model from overfitting; the remaining 10% of the data is used as a test set for the final evaluation of the model's generalization ability and recognition accuracy on unseen data.

[0136] The model architecture constructed in this implementation is as follows Figure 3 As shown in the figure, the core is the CAM-CNN model that introduces the channel attention mechanism, and further optimizes it using the GOA algorithm to form the optimal model. Specifically, the model contains two convolutional layers, two pooling layers, one fully connected layer, and a softmax layer, and also integrates a channel attention module.

[0137] In order to verify the superiority and effectiveness of the proposed model, this example selects the CNN basic model, the CAM-CNN model with the introduction of the channel attention mechanism, and the CAM-CNN model with the introduction of channel and GOA optimization as comparison objects, as shown in Table 1. By comparing with these models under the same dataset and experimental conditions, the performance of the proposed model in the task of abnormal flashover identification of lightning arresters is evaluated.

[0138] Table 1 Performance index comparison data

[0139]

[0140]

[0141] Table 1 shows that the basic CNN model achieved an accuracy of 78.23% and a precision of 76.18%. The average recognition time was 2.23 seconds, and convergence required 80 iterations, resulting in poor overall performance. The introduction of the channel attention mechanism significantly improved the CAM-CNN model's feature extraction, increasing its accuracy to 86.49% and its precision to 84.36%. The recognition time was shortened to 1.61 seconds, and the number of convergence iterations was reduced to 65, significantly improving performance. The CAM-CNN model with channel and GOA optimization performed the best, achieving an accuracy of 91.57% and a precision of 89.42%. It took only 1.32 seconds to recognize the model and converged after only 50 iterations. This demonstrates that the combination of GOA optimization and the channel attention mechanism significantly improves the accuracy and efficiency of model recognition and significantly shortens the convergence time of model training, offering significant advantages in the field of lightning arrester abnormal flashover identification.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying abnormal flashover of a lightning arrester, characterized in that: include: Collecting lightning arrester image data under different operating conditions of the transmission line and preprocessing the lightning arrester image data; A lightning arrester fault identification model is preset, and the lightning arrester fault identification model is based on the preprocessed image data to detect abnormal discharge of the lightning arrester.

2. The method for identifying abnormal flashover of a lightning arrester according to claim 1, wherein: Preprocessing the arrester image data includes: integrating the collected arrester images into an image data set, and cropping each image in the image data set; The cropped images are scaled to a uniform size and the image pixel values ​​are normalized, and the generated image sequence is used as input data for the arrester fault recognition model.

3. The method for identifying abnormal flashover of a lightning arrester according to claim 2, wherein: The arrester fault recognition model includes: an input layer, an output layer, a convolution layer, a pooling layer, a fully connected layer and a Softmax layer, as well as an integrated channel attention module.

4. The method for identifying abnormal flashover of a lightning arrester according to claim 3, wherein: The arrester fault identification model comprises the following steps before detecting the abnormal discharge of the arrester based on the pre-processed image data: optimizing the arrester fault identification model parameters; Initialize the model parameters, use accuracy as the objective function, and calculate the fitness of each individual; According to the optimization algorithm update rules, the positions of individuals in the population are updated globally, including the local search in the first stage and the local search in the second stage; If the current update iteration number does not reach the maximum number of iterations, continue global update and local search; if the current update iteration number reaches the maximum number of iterations, output the optimal parameter combination of the model; An arrester fault identification optimization model is established based on the optimal parameter combination of the model.

5. The method for identifying abnormal flashover of a lightning arrester according to claim 2 or 4, wherein: The lightning arrester fault identification optimization model includes: spatial feature extraction and feature dimension reduction of input data; The insulator and its discharge feature map of the input image are extracted and the spatial dimension of the insulator and its discharge feature map is compressed to obtain a one-dimensional feature vector of length C′ describing the insulator discharge state, which is used to characterize the operating state of the arrester.

6. The method for identifying abnormal flashover of a lightning arrester according to claim 5, characterized in that: Also includes: Based on the different feature channels in the image, the feature vector after generating channel attention weights is expressed as: A s =S(W s ·V+b s ) V′=A s ·V,V′∈R C′ ×H′×W′ Among them, W s ∈R C′ ×C′ is the weight matrix, b s is the bias term, S represents the Sigmoid function; the output A s ∈R C′ is the feature map channel attention weight, size is C′.

7. The method for identifying abnormal flashover of a lightning arrester according to claim 6, wherein: The output identification result of the lightning arrester fault identification optimization model is expressed as: y=Softmax(W f ·V′+b f ) Among them, W f is the weight matrix of the output layer, b f is the bias term of the output layer.

8. A lightning arrester abnormal flashover identification system, applied to the method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to collect lightning arrester image data under different operating conditions of the transmission line and pre-process the lightning arrester image data; The fault identification module is used to preset a lightning arrester fault identification model, wherein the lightning arrester fault identification model is based on the preprocessed image data to detect abnormal discharge of the lightning arrester.

9. An electronic device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for identifying abnormal flashover of a lightning arrester according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method for identifying abnormal flashover of a lightning arrester according to any one of claims 1 to 7.

Images