Distribution line fault identification model training method, fault identification method and system

Abstract

The invention provides a distribution line fault identification model training method, a distribution line fault identification method and a distribution line fault identification system. The training method comprises the following steps: acquiring visible light images and infrared thermal imaging image data containing various distribution network line faults; and enhancing, preprocessing and classifying the image data to form a distribution line fault data set for training, and training the model based on the data set. According to the method, the advantages of visible light and infrared thermal imaging images are fully exerted, the two kinds of data are used for model training at the same time, the accuracy of fault detection and recognition is improved, and the fault detection range is widened. Data quality and model generalization ability are improved through data enhancement and preprocessing operation, the constructed model is based on a VIT transformer architecture and is integrated with a classification model of a CBAM attention mechanism, fault features can be effectively extracted, noise interference is removed, detection precision is improved, and operation and maintenance and fault diagnosis efficiency of a distribution network line is powerfully guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of electric power, and in particular to a distribution line fault identification model training method, a distribution line fault identification method and a distribution line fault identification system. Background Art

[0002] With the continuous development of the power system, the inspection and maintenance of distribution lines are crucial to ensure the safe and stable supply of electricity.

[0003] Traditional inspection methods have problems such as low efficiency, large errors, and high risks, and cannot meet the complex and diverse inspection needs of modern distribution networks. To this end, intelligent inspection methods based on image recognition technology have gradually become mainstream, which automatically identify faults and defects in distribution lines by building target detection models, thereby improving inspection efficiency and accuracy.

[0004] In the prior art, the detection effect of the target detection model is highly dependent on the quality and diversity of the training data. In a complex distribution network environment, the collected training data often has problems of insufficient quality or imbalance, resulting in insufficient accuracy and reliability of fault identification. Summary of the invention

[0005] In view of the problems in the prior art, the embodiments of the present invention provide a distribution line fault identification model training method, a distribution line fault identification method and a system to solve the problems in the prior art that the training data often has insufficient quality or imbalance.

[0006] Illustratively, the embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a distribution line fault identification model training method, comprising:

[0008] Acquire image data containing multiple distribution network line faults, wherein the image data includes visible light images and infrared thermal imaging images;

[0009] Performing data enhancement on the image data;

[0010] Preprocessing the image data after data enhancement;

[0011] Classifying the preprocessed image data according to the type of distribution line fault to form a distribution line fault data set as a training data set; and

[0012] A distribution line fault identification model is trained based on the training data set.

[0013] In a possible implementation, the performing data enhancement on the image data includes: enhancing the visible light image and / or enhancing the infrared thermal imaging image;

[0014] Wherein, the enhancement of the visible light image includes: random cropping, rotation, flipping operations and / or brightness and contrast adjustment operations;

[0015] The enhancement of the infrared thermal imaging image includes: random cropping, rotation, flipping operations and / or stretching operations of the thermal spectrum range.

[0016] In a possible implementation, the stretching operation of the thermal spectrum range includes: using power stretching to stretch the temperature range by performing power transformation on pixel values ​​in the infrared thermal imaging image data.

[0017] In a possible implementation manner, the preprocessing of the image data after data enhancement includes: performing size normalization processing and / or grayscale processing on the image data.

[0018] In a possible implementation, the distribution line fault type includes at least one of the following: a line short circuit fault, a strand break fault, an insulator flashover fault, and a conductor connection point overheating fault.

[0019] In a possible implementation manner, after the image data is preprocessed and before the preprocessed image data is classified, it includes: performing a filtering and denoising operation on the image data.

[0020] In a possible implementation, the training of the distribution line fault identification model based on the training data set includes:

[0021] The training data set is input into the distribution line fault identification model, and the following cross entropy loss function is used to measure the difference between the model prediction result and the true label:

[0022]

[0023] Among them, L is the difference between the model prediction result and the true label, N is the total number of samples, C is the total number of categories, and y ij is the true label, p ij is the predicted label of the model.

[0024] In a possible implementation, the distribution line fault identification model is constructed based on a VIT transformer architecture. In the VIT transformer architecture, a convolutional block-based attention mechanism CBAM is introduced before the classification head. The convolutional block-based attention mechanism CBAM is used to complete the attention allocation in the channel dimension and the spatial dimension for the fault feature-related areas in the visible light image and the infrared thermal imaging image during the fault identification process.

[0025] In a second aspect, an embodiment of the present invention provides a method for identifying a distribution line fault, comprising:

[0026] Acquire image data of the power distribution line, wherein the image data includes a visible light image and an infrared thermal imaging image;

[0027] Determining the fault condition of the distribution line according to the image data using a distribution line fault identification model;

[0028] The distribution line fault identification model is obtained by training using any of the distribution line fault identification model training methods described above.

[0029] In a third aspect, an embodiment of the present invention provides a distribution line fault identification system, including:

[0030] An image data acquisition module, used to acquire image data of the power distribution line, wherein the image data includes a visible light image and an infrared thermal imaging image;

[0031] A fault condition determination module, configured to determine the fault condition of the distribution line according to the image data using a distribution line fault identification model;

[0032] The distribution line fault identification model is obtained by training using any of the distribution line fault identification model training methods described above.

[0033] The distribution line fault identification model training method, distribution line fault identification method and system provided by the embodiment of the present invention, in the model training method, by acquiring visible light and infrared thermal imaging images, the image data is enhanced, the enhanced image data is preprocessed, the preprocessed image data is classified according to the distribution line fault type to form a distribution line fault data set for training, the model is trained by the distribution line fault data set, and the distribution line fault is identified and detected based on the trained model. The present invention gives full play to the advantages of both visible light and infrared thermal imaging images, and the two data are used for model training at the same time, thereby improving the fault detection and identification accuracy of the trained model, improving the accuracy of fault detection and identification, and in addition, broadening the fault detection range, and being able to comprehensively capture various types of fault information. Data enhancement and preprocessing operations improve data quality and model generalization ability. The constructed classification model based on VIT transformer and integrated with CBAM attention mechanism can effectively extract fault features, remove noise interference, improve detection accuracy, effectively guarantee the operation and maintenance of distribution network lines and fault diagnosis efficiency, and reduce the risk of power supply interruption caused by faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flowchart of an exemplary distribution line fault identification model training method provided by an embodiment of the present invention;

[0036] Figure 2 A flowchart of another exemplary distribution line fault identification model training method provided by an embodiment of the present invention;

[0037] Figure 3 A flow chart of a method for intelligently identifying a power distribution line fault provided by an embodiment of the present invention;

[0038] Figure 4 A block diagram of a power distribution line fault intelligent identification system provided by an embodiment of the present invention;

[0039] Figure 5 A flow chart of a distribution line fault intelligent identification and alarm program provided by an embodiment of the present invention;

[0040] Figure 6 A block diagram of a computer device including a distribution line fault intelligent identification and alarm program provided for an embodiment of the present invention; and

[0041] Figure 7 The physical structure and connection relationship of a computer device including a distribution line fault intelligent identification and alarm program provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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.

[0043] Traditional inspection methods have problems such as low efficiency, large errors, and high risks, and cannot meet the complex and diverse inspection needs of modern distribution networks. To this end, intelligent inspection methods based on image recognition technology have gradually become mainstream. By building a target detection model to automatically identify faults and defects in distribution lines, the inspection efficiency and accuracy are improved. However, in the prior art, the detection effect of the target detection model is highly dependent on the quality and diversity of the training data. In a complex distribution network environment, affected by factors such as lighting, equipment form, and shooting angle, the collected images often have problems of insufficient or unbalanced quality, which makes it difficult to meet the actual needs of model training, resulting in insufficient accuracy and reliability of fault identification.

[0044] Therefore, there is an urgent need for an improved model training method, distribution network fault identification technology, and corresponding software and equipment to optimize the model training process, improve the accuracy and adaptability of fault identification, and provide an intelligent solution for distribution network inspection.

[0045] In order to solve the problems of insufficient or unbalanced quality, insufficient accuracy and reliability of fault identification, and low precision and adaptability of fault identification in existing collected images, the present invention provides a distribution network line fault identification model training method, a distribution network line fault identification method and system, which are of great significance to the inspection and maintenance of distribution lines.

[0046] The present invention combines visible light and infrared thermal imaging images for model training and distribution network line fault identification and detection, giving full play to the advantages of both. Both data are used for model training at the same time, which improves the fault detection and identification accuracy of the trained model, improves the accuracy of fault identification, broadens the scope of fault identification and detection, and can comprehensively capture various types of fault information. In the data processing link, data enhancement and preprocessing operations improve data quality and model generalization capabilities. The constructed classification model based on VIT transformer and integrated with CBAM attention mechanism can effectively extract fault features, remove noise interference, improve detection accuracy, effectively guarantee the operation and maintenance of distribution network lines and fault diagnosis efficiency, and reduce the risk of power supply interruption caused by faults.

[0047] First, refer to Figure 1, which shows a distribution line fault identification model training method provided by an exemplary embodiment of the present invention, including: step 1) obtaining image data containing multiple distribution network line faults, wherein the image data includes visible light images and infrared thermal imaging images; step 2) performing data enhancement on the image data; step 3) preprocessing the image data after data enhancement; step 4) classifying the preprocessed image data according to the distribution line fault type to form a distribution line fault data set as a training data set; step 5) training the distribution line fault identification model based on the training data set. Among them, the above steps 1)-4) constitute the process of obtaining the training data set. Step 5) is the process of training the model.

[0048] In step 1), by obtaining image data of visible light images and infrared thermal imaging images containing various distribution network line faults, and combining visible light and infrared thermal imaging images for model training and distribution network line fault detection, the advantages of both are fully utilized, the fault recognition accuracy of the trained model is improved, the fault recognition accuracy is improved, the fault detection range is expanded, and various types of fault information can be fully captured.

[0049] In step 2), enhancing the image data includes: enhancing the visible light image and / or enhancing the infrared thermal imaging image. In one example, the enhancement of the visible light image includes: random cropping, rotation, flipping operations and / or brightness and contrast adjustment operations; in one example, the enhancement of the infrared thermal imaging image includes: random cropping, rotation, flipping operations and / or stretching operations of the thermal spectrum range. In one example, the stretching operation of the thermal spectrum range includes: using power stretching to stretch the temperature range by power transforming the pixel values ​​in the infrared thermal imaging image data. Enhancing the image data can improve the diversity of the image data and the generalization ability of the model, improve the robustness of the model, and improve the accuracy of the model in identifying faults.

[0050] In step 3), preprocessing the enhanced image data includes: size normalization and / or grayscale processing of the image data. In one example, size normalization is to uniformly adjust visible light and infrared thermal imaging images of different resolutions to a standard size suitable for model input to ensure consistent processing of data in the model. The image is then grayscaled to convert the color information of the visible light image into grayscale information, simplifying the amount of data while highlighting the texture and structural features of the image. Grayscale infrared thermal imaging images help to more clearly show the distribution gradient of thermal radiation intensity. The above data enhancement and preprocessing operations improve data quality and model generalization capabilities.

[0051] In step 4), the preprocessed image data is classified according to the distribution line fault type to form a distribution line fault data set for training.

[0052] In one example, the fault types may include line short circuit faults, strand break faults (surface and internal), insulator flashover faults, conductor connection point overheating faults, etc. For each fault type, the corresponding image samples are selected from the image data after data enhancement and preprocessing, and a distribution network line fault dataset with clear fault type identification is constructed, providing a solid data foundation for subsequent model training, testing, and intelligent detection and diagnosis of distribution network line faults based on the dataset.

[0053] Among them, in step 5), training the distribution line fault identification model based on the training data set is to input the distribution line fault data set obtained after enhancement and preprocessing into the distribution line fault identification model to train the distribution line fault identification model.

[0054] The distribution line fault recognition model is based on the VIT transformer architecture, which takes the VIT transformer as the basic architecture. The VIT transformer architecture has excellent feature extraction and image understanding capabilities, and can effectively process the complex information structure in the distribution line fault image. The basic framework of the VIT transformer model is built using the deep learning framework. The convolutional block-based attention mechanism (CBAM) is integrated before the classification head. The CBAM attention mechanism can perform refined attention allocation on the channel dimension and spatial dimension of the image. In this way, the model can more accurately capture the key features of the distribution line fault, thereby significantly improving the detection accuracy of the model, greatly improving the accuracy and reliability of the distribution line fault image classification, and providing strong technical support for the efficient operation and maintenance of the distribution line and rapid fault diagnosis.

[0055] This embodiment combines visible light and infrared thermal imaging images to train the model before the distribution network line fault detection, so as to give full play to the advantages of both, broaden the training scope and improve the accuracy of fault identification, and can comprehensively capture various types of fault information. In the data processing link, data enhancement and preprocessing operations improve the data quality and model generalization ability. The constructed model is based on VIT transformer and integrates CBAM attention mechanism. CBAM attention mechanism can perform refined attention allocation on the channel dimension and spatial dimension of the image. The above scheme can effectively extract fault features, remove noise interference, improve detection accuracy, effectively guarantee the operation and maintenance of distribution network lines and fault diagnosis efficiency, and reduce the risk of power supply interruption caused by faults.

[0056] The following is another exemplary embodiment of the present application, referring to Figure 2 The following embodiment of the present invention provides a distribution network line fault identification model training method, including:

[0057] Step a) obtaining visible light and infrared thermal imaging image data containing multiple distribution network line faults;

[0058] Step b) performing random cropping, rotation, flipping, brightness and contrast adjustment, and thermal spectrum range stretching data enhancement based on the original data;

[0059] Step c) performing normalization, grayscale, denoising, and classification preprocessing on the data after data enhancement to form a distribution network line fault data set;

[0060] Step d) building an image classification model based on the self-attention mechanism and introducing an attention mechanism based on convolutional blocks;

[0061] Step e) training the classification model based on the distribution network line fault data set to obtain a distribution network line fault identification model.

[0062] In step a), for visible light images, it can clearly present the appearance of the distribution network line, connecting components, and surrounding environmental conditions. For example, the aging, damage, foreign matter adhesion of the line, and the rust and deformation of the tower can be intuitively reflected in the visible light image. Infrared thermal imaging images, by capturing the thermal radiation of objects, effectively highlight the abnormal heating areas caused by faults in the distribution network line. Fault phenomena such as heating at the broken strands of the internal line and overall or local overheating caused by line overload will show thermal feature distribution that is significantly different from normal parts in infrared thermal imaging images. Assuming the ambient temperature is 25 degrees Celsius, the temperature of the normally operating conductor may reach about 30-40°C. If it is higher than this value, it can be determined as fault overheating.

[0063] In the principle of infrared thermal imaging, according to the Stefan-Boltzmann law, the total power M radiated per unit area of ​​an object's surface is proportional to the fourth power of the object's thermodynamic temperature T, as shown in the following formula.

[0064] M=εσT 4

[0065] Where ε is the radiation coefficient of the object (for a specific distribution line material, the radiation coefficient is a relatively fixed value), and the proportionality coefficient σ is the Stefan-Boltzmann constant (approximately 5.67×10 -8 W·m -2 ·K -4 ).

[0066] Under normal operating conditions, the distribution network line has a relatively stable operating temperature T 正常 , its total radiated power is When a line is broken, the cross-sectional area at the break decreases, causing the resistance to increase, as shown in the formula.

[0067]

[0068] Among them, R is the resistance of the line at the broken strand, ρ is the line resistivity, L is the length of the line at the broken strand, and S is the cross-sectional area of ​​the line.

[0069] According to Joule's law, excessive heat will be generated at the broken strand, causing the temperature to rise, as shown in the following formula.

[0070] Q=I 2 R

[0071] Among them, Q represents the heat generated by the current passing through the conductor, I represents the current intensity, R represents the resistance at the break of the line, and t represents the power-on time.

[0072] When the line is overloaded, the current is too large, and according to Joule's law, a large amount of heat is generated in the whole or part of the line, causing the temperature to rise. Therefore, when the line is broken or overloaded, the resistance change and other factors will cause the local or overall temperature of the line to rise to T 故障 , then the total radiated power is Due to T 故障 >T 正常 , so M 故障 >M 正常 , that is, the thermal radiation intensity of the fault area will be higher than that of the normal area, thus presenting a characteristic distribution different from that of the normal part in the infrared thermal imaging image.

[0073] In step b), in order to improve the diversity of image data and the generalization ability of the model, data enhancement operations are performed after the image data is acquired.

[0074] Specifically, for visible light images, a variety of technical means such as random cropping, rotation, flipping, and brightness and contrast adjustment are used. Random cropping helps to capture different local areas of the image, so that the model can learn the characteristics of line faults at different viewing angles; rotation and flipping operations simulate the changes in images under different shooting angles, enhancing the model's adaptability to the diversity of line fault morphology; brightness and contrast adjustments can cope with changes in line images under different lighting conditions, further improving the robustness of the model. For infrared thermal imaging images, in addition to the same cropping, rotation, and flipping operations as visible light images, it is also necessary to stretch the thermal spectrum range to highlight the temperature anomaly characteristics of different degrees, thereby helping the model to more accurately identify subtle temperature differences and the corresponding fault conditions. The stretching operation of the thermal spectrum range includes: using power stretching to stretch the temperature range by power transforming the pixel values ​​in the infrared thermal imaging image data.

[0075] The present invention adopts power stretching to stretch the temperature range by power transforming the pixel values ​​in the image, thereby optimizing the visual effect of the image. Images with nonlinear temperature distribution can be better processed, especially when the temperature difference in the image is uneven. The power stretching principle is shown in the following formula.

[0076]

[0077] Among them, T(x,y) is the temperature value in the original image, T max is the maximum temperature value in the image, γ is the exponent of power stretching. In order to highlight the characteristic information of abnormal high temperature parts in the line, γ takes a positive number greater than 1, and T'(x, y) is the processed temperature value.

[0078] In step c), after completing data enhancement, image preprocessing is performed. First, the image is size normalized, and the visible light and infrared thermal imaging images of different resolutions are uniformly adjusted to a standard size of 640*640 suitable for model input to ensure the consistency of data processing in the model. Of course, the standard size can be adjusted according to the model, and the sizes of other numerical values ​​also fall within the scope of the present invention. Then the image is grayed, and the color information of the visible light image is converted into grayscale information, which simplifies the amount of data while highlighting the texture and structural features of the image. Graying the infrared thermal imaging image helps to more clearly show the distribution gradient of the thermal radiation intensity. Then the image is filtered and denoised to remove noise interference introduced by factors such as shooting equipment and environment, so as to improve the clarity and quality of the image, make the key features of the distribution network line failure more prominent, and facilitate the accurate identification and analysis of subsequent models.

[0079] The image data is classified according to the distribution line fault type to form a distribution line fault dataset. The fault types may include line short circuit fault, strand break fault (surface and internal), insulator flashover fault, conductor connection point overheating fault, etc. For each fault type, the corresponding image samples are selected from the image data after data enhancement and preprocessing, and a distribution line fault dataset with clear fault type identification is constructed, which provides a solid data foundation for subsequent model training, testing, and intelligent detection and diagnosis of distribution line faults based on this dataset.

[0080] In step d), the distribution line fault recognition model in this embodiment is a special model for classifying distribution network line fault images. The construction method of the model is as follows: the distribution line fault recognition model is based on the VIT transformer architecture, in which a convolutional block-based attention mechanism CBAM is introduced before the classification head, that is, a convolutional block-based attention mechanism (CBAM) is integrated before the classification head, and the convolutional block-based attention mechanism CBAM is used to complete the attention allocation of the channel dimension and the spatial dimension to the fault feature-related area in the image.

[0081] Based on the advanced VIT Transformer architecture, it can effectively process the complex information structure in the distribution network line fault image. In terms of architectural design, the convolutional block attention mechanism (CBAM) is innovatively introduced before the classification head, which can accurately allocate attention to the spatial dimension and channel dimension of the image.

[0082] For the convolutional block-based attention mechanism CBAM, in one example, in the spatial dimension, CBAM focuses on the areas in the image that are closely related to the fault features, such as the distribution network line fault point, abnormal heating area, etc., to ensure that the features of these key areas are strengthened; in the channel dimension, CBAM can screen out image feature channels that have a large contribution to fault classification, and filter out invalid channels caused by noise or irrelevant background information, thereby reducing noise interference. Through the above mechanism, the model can more accurately capture the key features of distribution network line faults, significantly improve detection accuracy, enhance the accuracy and reliability of distribution network line fault image classification, and provide a solid technical guarantee for efficient operation and maintenance of distribution network lines and rapid fault diagnosis.

[0083] For model building, in one example, when building a distribution network line fault image classification model, VIT transformer was selected as the basic architecture, and the deep learning framework pytorch was used to build the basic framework structure of the VIT transformer model, and key parameters such as the number of layers, number of heads, and hidden layer dimensions of the model were set to ensure that it can adapt to the processing requirements of distribution network line fault image data.

[0084] For the implementation of CBAM, in one example, in the spatial attention module, the input feature map is first convolved through a convolution layer to generate a spatial attention feature map. The convolution kernel size, step size and other parameters of the convolution layer are carefully tuned to obtain the best spatial feature extraction effect. Then, the generated spatial attention feature map is multiplied with the original input feature map to achieve weighted attention in the spatial dimension and highlight the key spatial area. , to achieve weighted attention in the spatial dimension, focus on the areas in the image that are closely related to the fault features, such as the fault point on the distribution network line, the precise location of the abnormal heating area in space, etc., so as to strengthen the feature representation of these key areas. As shown in the following formula.

[0085]

[0086] Among them, F 7×7 Represents a convolution operation with a convolution kernel size of 7×7.

[0087] In the channel attention module, the input feature map is aggregated in the channel dimension through global average pooling and global maximum pooling operations to obtain two different channel description vectors. In the channel dimension, the image feature channels that contribute more to fault classification are selected, and the channel information generated by noise or irrelevant background information is filtered out to remove useless noise interference in the image. This is shown in the following formula.

[0088]

[0089] Among them, δ represents the activation function, F is the feature map, W0 and W1 represent two convolution operations respectively. and They represent average pooling and maximum pooling respectively.

[0090] After that, the two vectors are processed by a multi-layer perceptron (MLP) respectively, and the processed results are added and passed through a Sigmoid activation function to generate a channel attention weight vector. Finally, the channel attention weight vector is multiplied with the feature map after spatial attention weighting to complete the attention allocation in the channel dimension. As shown in the following formula.

[0091]

[0092] Among them, F′ and F″ represent the output feature maps after channel attention and spatial attention, respectively. Represents dot product.

[0093] In step e), during the model training phase, the distribution network line fault dataset is input into the constructed model. In one example, a cross entropy loss function is used to measure the difference between the model prediction result and the true label, as shown in the following formula.

[0094]

[0095] Among them, L represents the difference between the model prediction result and the true label, N is the total number of samples, C is the total number of categories, and y ij is the true label, p ij is the predicted label of the model.

[0096] During the training process, in one example, the Adam optimization algorithm is used to iteratively update the parameters of the model. By setting appropriate hyperparameters such as the learning rate, learning rate decay strategy, and number of training rounds, the weight of the model is gradually optimized, so that the loss value of the model on the training data set continues to decrease, while the accuracy on the verification data set continues to improve until the model reaches a convergence state, thereby obtaining a distribution network line fault image classification model with high detection accuracy, which can accurately classify and identify distribution network line fault images, and provide reliable technical support for the actual distribution network line operation and maintenance work.

[0097] In summary, the present invention has at least the following beneficial effects:

[0098] The present invention combines visible light and infrared thermal imaging images for model training and distribution network line fault detection, giving full play to the advantages of both. Both data are used for model training at the same time, which improves the fault detection and identification accuracy of the trained model, improves the accuracy of fault detection and identification, broadens the scope of fault detection, and can comprehensively capture various types of fault information. In the data processing link, data enhancement and preprocessing operations improve data quality and model generalization capabilities. The constructed classification model based on VITtransformer and integrated with CBAM attention mechanism can effectively extract fault features, remove noise interference, improve detection accuracy, effectively guarantee the operation and maintenance of distribution network lines and fault diagnosis efficiency, and reduce the risk of power supply interruption caused by faults.

[0099] Referring to 3, the present invention also provides a method for identifying a distribution line fault, comprising:

[0100] Acquire image data of the power distribution line, wherein the image data includes a visible light image and an infrared thermal imaging image;

[0101] The distribution line fault identification model is used to determine the fault condition of the distribution line based on the image data.

[0102] Among them, the distribution line fault identification model is obtained by training using any of the above-mentioned distribution line fault identification model training methods, and the model construction method is also constructed based on the above-mentioned description method, which will not be repeated here.

[0103] The distribution line fault identification model is constructed based on the method described above. For the specific distribution line fault identification process, please refer to the above description.

[0104] The present invention provides a distribution network line fault intelligent identification program product, including a computer program, the program including computer instructions, and the computer instructions implement the steps of the above-mentioned distribution network line fault intelligent identification method when executed by a processor.

[0105] Reference Figure 4 The present invention also provides a distribution network line fault identification system 400, comprising:

[0106] An image data acquisition module 410 is used to acquire image data of a power distribution line, wherein the image data includes a visible light image and an infrared thermal imaging image;

[0107] The fault condition determination module 420 is used to determine the fault condition of the distribution line according to the image data using the distribution line fault identification model.

[0108] Among them, the distribution line fault identification model is obtained by training using any of the above-mentioned distribution line fault identification model training methods, and the model construction method is also constructed based on the above-mentioned description method, which will not be repeated here.

[0109] Reference Figure 5 The present invention also provides a distribution network line fault intelligent identification and alarm method, the distribution network line fault intelligent identification and alarm method comprising:

[0110] By acquiring and processing the video signal stream of all or part of the key areas of the distribution network line in real time, in one example, by docking with a camera or other video source, the video data frames of all or part of the key areas of the distribution network line are read in real time;

[0111] The video data is transmitted to a convolutional neural network (CNN) model trained using the training method of the present application.

[0112] After receiving the video frame data, the model extracts and analyzes its features, and performs content detection and classification based on the trained model parameters and algorithm logic. By analyzing the video frames, the model can identify whether there is a distribution network line failure or other specific categories of abnormal conditions.

[0113] When the model determines that the video content belongs to the category that requires an alarm, the pop-up alarm mechanism will be triggered to promptly warn the monitoring personnel. In one example, the pop-up content includes information such as the alarm type, alarm time, and screenshots of related video frames, ensuring that the monitoring personnel can quickly understand the fault situation and take appropriate measures. The key area is the area where the distribution network line has a fault. The model is trained using two types of images, and the model is input with video data during actual fault identification.

[0114] The present invention also provides a distribution network line fault intelligent identification and alarm program product, including a computer program, the program including computer instructions, and the computer instructions, when executed by a processor, implement the steps of the above-mentioned distribution network line fault intelligent identification and alarm method.

[0115] The above-mentioned distribution network line fault intelligent identification and alarm program, by integrating fault intelligent identification and alarm, realizes autonomous detection and pop-up alarm of real-time video streams, significantly improves the intelligence level of distribution network line video inspection, and enhances the timeliness of early warning, effectively ensuring the safety and stable operation of distribution network lines; it can realize autonomous detection and pop-up alarm of real-time video streams, effectively improves the intelligence level of video inspection and the timeliness of early warning, facilitates monitoring personnel to respond and take measures in time, ensures the stable operation of the distribution network system, and improves the operation and maintenance management efficiency, and is provided in the form of computer program products and computer equipment, which is convenient for wide promotion and application in related fields.

[0116] Reference Figure 6 The present invention also provides a computer device, including a processor, a memory, a graphics and parallel computing processor (GPU), and a computer program stored in the memory and configured to be executed by the processor. When the computer device executes the computer program, the steps of the above-mentioned distribution network line fault intelligent identification and alarm method are implemented.

[0117] Reference Figure 7 As shown, the physical structure and connection relationship of the computer equipment of the distribution line fault intelligent identification and alarm program include: the line to be tested 1, the visible light camera 2 and the infrared sensor 3, the computer equipment 4, and the audio alarm 5. Among them, the cameras 2 and 3 transmit the image of the line to be tested 1 back to the computer equipment 4, the computer equipment 4 detects the image by running the computer program, and the computer equipment 4 displays the detection results to the staff through the screen and the audio alarm 5.

[0118] It can be understood that the detailed functions and extended functions that can be executed by the computer program can refer to the description of the above embodiments.

[0119] Based on the same inventive concept, another embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, all steps of the above-mentioned distribution line fault identification and alarm method are implemented.

[0120] It can be understood that the detailed functions and extended functions that can be executed by the computer program can refer to the description of the above embodiments.

[0121] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0122] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention. Those of ordinary skill in the art may understand and implement them without creative effort.

[0123] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution 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 ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the security defense method described in each embodiment or some parts of the embodiment.

[0124] In addition, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0125] In addition, in the present invention, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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 invention.

Claims

1. A distribution line fault identification model training method, characterized in that: include: Acquire image data containing multiple distribution network line faults, wherein the image data includes visible light images and infrared thermal imaging images; Performing data enhancement on the image data; Preprocessing the image data after data enhancement; Classifying the preprocessed image data according to the type of distribution line fault to form a distribution line fault data set as a training data set; and A distribution line fault identification model is trained based on the training data set.

2. The distribution line fault identification model training method according to claim 1, characterized in that: The data enhancement of the image data includes: enhancing the visible light image and / or enhancing the infrared thermal imaging image; Wherein, the enhancement of the visible light image includes: random cropping, rotation, flipping operations and / or brightness and contrast adjustment operations; The enhancement of the infrared thermal imaging image includes: random cropping, rotation, flipping operations and / or stretching operations of the thermal spectrum range.

3. The distribution line fault identification model training method according to claim 2 is characterized in that: The stretching operation of the thermal spectrum range includes: adopting power stretching to stretch the temperature range by performing power transformation on pixel values ​​in the infrared thermal imaging image data.

4. The distribution line fault identification model training method according to claim 1, characterized in that: The preprocessing of the image data after data enhancement includes: performing size normalization processing and / or grayscale processing on the image data.

5. The distribution line fault identification model training method according to claim 1, characterized in that: The distribution line fault type includes at least one of the following: line short circuit fault, strand break fault, insulator flashover fault and conductor connection point overheating fault.

6. The distribution line fault identification model training method according to claim 1, characterized in that: After the image data is preprocessed and before the preprocessed image data is classified, the method includes: performing a filtering and denoising operation on the image data.

7. The distribution line fault identification model training method according to claim 1, characterized in that: The training of the distribution line fault identification model based on the training data set includes: The training data set is input into the distribution line fault identification model, and the following cross entropy loss function is used to measure the difference between the model prediction result and the true label: Among them, L is the difference between the model prediction result and the true label, N is the total number of samples, C is the total number of categories, and y ij is the true label, p ij is the predicted label of the model.

8. The distribution line fault identification model training method according to any one of claims 1 to 7, characterized in that: The distribution line fault recognition model is built based on the VIT transformer architecture. In the VIT transformer architecture, a convolutional block-based attention mechanism CBAM is introduced before the classification head. The convolutional block-based attention mechanism CBAM is used to complete the attention allocation in the channel dimension and the spatial dimension for the fault feature-related areas in the visible light image and the infrared thermal imaging image during the fault recognition process.

9. A method for identifying a distribution line fault, characterized in that: include: Acquire image data of the power distribution line, wherein the image data includes a visible light image and an infrared thermal imaging image; Determining the fault condition of the distribution line according to the image data using a distribution line fault identification model; The distribution line fault identification model is obtained by training using the distribution line fault identification model training method described in any one of claims 1 to 8.

10. A distribution network line fault identification system, characterized in that: include: An image data acquisition module, used to acquire image data of the power distribution line, wherein the image data includes a visible light image and an infrared thermal imaging image; A fault condition determination module, configured to determine the fault condition of the distribution line according to the image data using a distribution line fault identification model; The distribution line fault identification model is obtained by training using the distribution line fault identification model training method described in any one of claims 1 to 8.

Images