Power transmission line insulator fault detection method

Abstract

The invention relates to the technical field of data analysis, in particular to a power transmission line insulator fault detection method. The method comprises the following steps: S1, acquiring an insulator data set, and enhancing an image background in the insulator data set to obtain an enhanced data set; s2, constructing an improved Faster R-CNN network, wherein the improved Faster R-CNN network comprises a residual neural network ResNet50, a feature pyramid network FPN, a region generation network RPN, a region of interest Ro I pooling layer and a classification regression network; and S3, based on the enhanced data set, training an improved Faster R-CNN network to obtain a fault diagnosis model, the fault diagnosis model being used for determining the defect type of the insulator based on the collected to-be-diagnosed image data. According to the method, the original feature extraction network VGG16 of the Faster R-CNN is abandoned, the ResNet50 network and the FPN network structure are utilized to replace the VGG16 network, and the number of anchor points in the RPN network and the scale of the anchor points are increased, so that the structure of the Faster R-CNN model is optimized, and insulator identification and fault type detection under a complex background are facilitated.

Description

Technical Field

[0001] The present application relates to the technical field of data analysis, and more specifically, to a method for detecting insulator faults in a transmission line. Background Art

[0002] Insulators are essential components in high-voltage power transmission systems to ensure electrical isolation and mechanical performance. Once an insulator fails, it will pose a direct threat to the stability and safety of the power system. From the statistical data, the vast majority of power system accidents are caused by insulator defects. Therefore, fast and accurate detection of insulator defects is currently the most important.

[0003] The patent with publication number CN116203462A discloses an insulator fault detection system, method and medium. It includes a multi-channel leakage current signal acquisition module, which is used to collect the leakage current signal of at least one insulator string and send the leakage current signal to the signal processing module; the signal processing module is used to perform signal preprocessing on the leakage current signal to obtain the data to be analyzed, and send the data to be analyzed to the fault monitoring base station; the fault monitoring base station is used to perform insulator fault analysis on the data to be analyzed according to the target algorithm to obtain the fault analysis result. It can be seen that the insulator detection method in this patent adopts a power detection method. The principle of power detection is to detect the leakage current of the insulator and use an electrical method to detect whether the insulator has leakage current. However, this detection method is susceptible to electromagnetic interference caused by electric arc, which affects the accuracy of judgment.

[0004] The invention patent with the authorization announcement number CN113112489B discloses an insulator string drop fault detection method based on a cascade detection model. First, aerial images of transmission line insulators are collected to produce insulator data sets and insulator string drop data sets; insulator positioning detection models and insulator fault detection models are established and trained; the insulator position is first located by the trained insulator positioning detection model, and then the located insulator is detected for faults using the insulator fault detection model, and finally the positioning and fault detection results are output on the test image. It can be seen that the patent adopts a non-electrical detection method, and based on the image information of the insulator, an image processing method is used to detect faulty insulators. At present, it is very difficult to manually perform power supply inspections to effectively obtain a large number of pictures of insulator defects. In order to overcome the limitations of manual inspections, using unmanned aerial vehicles (UAVs) and high-definition cameras to obtain images of high-altitude insulators is currently the most reliable solution. However, how to quickly and accurately identify insulator images taken during drone inspections has become a key issue. Summary of the invention

[0005] In view of this, the present application provides a method for detecting insulator faults in a power transmission line, which improves the detection accuracy and detection performance of insulator detection while ensuring the detection capability of small targets and obscured targets.

[0006] The technical solutions provided by this application are as follows:

[0007] A method for detecting insulator faults in a transmission line, the method comprising:

[0008] S1. Acquire an insulator dataset, and simulate a weather scene to enhance the image background in the insulator dataset to obtain an enhanced dataset;

[0009] S2. Build an improved Faster R-CNN network, including the residual neural network ResNet50, the feature pyramid network FPN, the region generation network RPN, the region of interest RoI pooling layer and the classification regression network;

[0010] S3. Based on the enhanced data set, the improved Faster R-CNN network is trained to obtain a fault diagnosis model, where the fault diagnosis model is used to determine the defect type of the insulator based on the collected image data to be diagnosed.

[0011] In one possible implementation, in S1, after obtaining the insulator data set, the method further includes:

[0012] The insulator dataset is preprocessed to remove similar images in the insulator dataset.

[0013] In one possible implementation, in S1, enhancing the image background in the insulator dataset includes:

[0014] Establish a simulated weather scene library, including rain, snow, fog and dark light scenes;

[0015] Based on any simulated weather scene in the simulated weather scene library, the background environment and morphology of the image background in the insulator dataset are enhanced.

[0016] In one possible implementation, the improved Faster R-CNN network uses the ResNet50 network and the FPN network to extract image features;

[0017] The ResNet50 network consists of 50 layers and is divided into seven parts;

[0018] The first part is used to perform convolution, regularization, activation function and maximum pooling operations on the image input to the ResNet50 network;

[0019] The second to fifth parts contain residual blocks, which are used to continue the convolution operation;

[0020] The sixth part is used to perform the average pooling operation;

[0021] The seventh part is the fully connected layer, which is used to obtain the feature map of the image input into the network.

[0022] In one possible implementation, the FPN network is used to fuse the features on different feature maps obtained, and predict the image depth of the fused feature maps;

[0023] The feature fusion includes:

[0024] Using ResNet50 as the backbone network, we get a series of feature maps, including conv1 to conv5.

[0025] Feature fusion is performed on conv2~conv5, and the feature maps P2~P5 after feature fusion are obtained by upsampling the integer multiples of 2.

[0026] In one possible implementation, after feature fusion and upsampling on conv5, it also includes performing maximum pooling on the feature map after feature fusion to obtain feature map P6. The feature map P6 is used for the region generation network RPN part, which is the shared weight between the region generation network RPN and the improved Faster R-CNN.

[0027] In one possible implementation, the ratios of the anchor points of feature maps P2, P3, P4, P5, and P6 are 32 2 ,64 2 ,128 2 ,256 2 and 512 2 , the anchor frame ratios are uniform {1:2, 1:1, 2:1}.

[0028] In one possible implementation, the improved Faster R-CNN uses the RPN network as the candidate region generation part to create candidate regions and obtain candidate boxes with specified aspect ratios;

[0029] The RPN network obtains a specified number of anchor boxes through pre-set anchor point parameter values, and determines the positive anchors of the candidate boxes through the normalized softmax function.

[0030] In one possible implementation, the region of interest (RoI) pooling layer is used to adjust the size of the candidate region generated by the RPN network, and input the feature data in the candidate region into the classification regression network, and the classification regression network is used to determine whether it is an insulator based on the feature data; if so, the defect type of the insulator is determined based on the feature data.

[0031] In one possible implementation, in S3, based on the enhanced data set, training the improved Faster R-CNN network includes:

[0032] Input the training samples in the enhanced dataset into the improved Faster R-CNN network;

[0033] The improved Faster R-CNN network outputs the calculation results and compares them with the preset expected output results;

[0034] If the calculation result is inconsistent with the preset expected output result, the improved Faster R-CNN network will learn weights based on the training samples, determine the weights through multiple iterations, generate new training samples to train the improved Faster R-CNN network, until the output calculation result is consistent with the preset expected output result, and obtain a fault diagnosis model.

[0035] Compared with the prior art, the technical solution of this application has the following beneficial effects:

[0036] By abandoning the original feature extraction network VGG16 of Faster R-CNN, replacing the VGG16 network with the ResNet50 network and the FPN network structure, increasing the number of anchor points in the RPN network, and increasing the anchor point scale, the structure of the Faster R-CNN model is optimized.

[0037] The complexity and required parameters of the improved Faster R-CNN network are reduced. The residual structure is used to build an ultra-deep network structure to alleviate the gradient vanishing phenomenon. At the same time, the FPN network solves the problem of low semantic level of low-level features through top-down and lateral connections, so that low-level features are conducive to small target detection and improve detection accuracy, which helps to improve the recognition of insulators and fault type detection in complex backgrounds. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0039] Figure 1 A flow chart of a method for detecting insulator faults in a transmission line provided in Example 1 of the present application.

[0040] Figure 2 This is a network structure diagram of a ResNet50 neural network provided in Example 1 of the present application.

[0041] Figure 3 A schematic diagram of the network structure of an FPN network provided in this application.

[0042] Figure 4 Schematic diagram of image feature extraction using the resnet50 network and FPN network provided in Example 1 of the present application.

[0043] Figure 5 A flowchart of training an improved Faster R-CNN network is provided in Example 1 of the present application.

[0044] Figure 6 A structural diagram of an improved Faster R-CNN network structure provided in Example 2 of the present application. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] Embodiment 1:

[0047] See also Figure 1 , is a flow chart of a method for detecting a transmission line insulator fault provided in Example 1 of the present application. Figure 1 As shown in , the specific implementation steps of the above method include:

[0048] S1. Obtain an insulator dataset, and enhance the image background in the insulator dataset to obtain an enhanced dataset.

[0049] This application considers that the insulator dataset has a single image background, which makes it difficult for the Faster R-CNN network to extract features. In order to solve the above problem, the imgaug library is used to simulate rain, snow, fog and dark light scenes, and an insulator dataset WI based on the insulator dataset (PI) is created in AI Studio, and the defect type is insulator shedding.

[0050] For example, the insulator dataset PI has a total of 600 images, some of which are similar. Our work is as follows: First, we remove similar images from PI, leaving 420 images, and then randomly select a simulated weather and 1-3 conventional enhancement methods to obtain an enhanced dataset with rich background environment and morphology. Each image in the enhanced dataset is enhanced twice, resulting in a total of 1260 images in the enhanced dataset.

[0051] S2. Build an improved Faster R-CNN network, including the residual neural network ResNet50, the feature pyramid network (FPN network), the region generation network RPN, the region of interest RoI pooling layer and the classification regression network.

[0052] Specifically, the improved Faster R-CNN network introduces the residual neural network ResNet50 and the FPN network on the basis of the traditional Faster R-CNN network as the image feature extraction part of the improved Faster R-CNN network.

[0053] See also Figure 2 , is a network structure diagram of a ResNet50 neural network provided in Example 1 of the present application. The ResNet50 network consists of 50 layers, such as Figure 2 As shown in the figure, the ResNet-50 network consists of seven parts. The first part is used to perform convolution, regularization, activation function and maximum pooling operations on the input network information. The second, third, fourth and fifth parts continue to perform convolution operations, and these parts contain residual blocks. The sixth part is used to perform average pooling operations, and the seventh part is a complete fully connected layer used to obtain feature maps of all images input to the network.

[0054] The principle of the FPN network is to fuse the features on different feature maps and then predict the fused feature maps. Figure 3 , Figure 3 A schematic diagram of the network structure of an FPN network provided in this application. Figure 3 As shown in the figure, firstly, a deep convolution is performed on the input feature map, and the features above the second layer are reduced in dimension, that is, a 1x1 convolution layer is added. The features above the fourth layer are upsampled. By upsampling the integer multiples of 2, the feature map on the right side of the FPN network after feature fusion is obtained. The number of 1x1 convolution kernels is 256, and the final predicted feature map channel (image depth) is 256.

[0055] See also Figure 4, is a schematic diagram of image feature extraction using the resnet50 network and the FPN network provided in Example 1 of the present application. Figure 4 As shown in , ResNet50 is used as the backbone network to obtain a series of feature maps, including conv1 to conv5, which are denoted as C1, C2, C3, C4, and C5 respectively. Figure 4 As shown in , feature fusion is performed only on the first 4 layers, so C2, C3, C4, and C5 are taken.

[0056] Feature fusion is performed on the feature maps of conv2~conv5, and the feature maps Predict2~Predict5 after feature fusion are obtained by upsampling the integer multiples of 2, which are denoted as P2, P3, P4, and P5.

[0057] exist Figure 4 In the above figure, feature map P6 is only used in the RPN part, not in the Faster R-CNN part and different prediction feature layers, and is the shared weight between RPN and Faster R-CNN. P6 is obtained by performing the maximum pooling operation on the feature map. The ratio of the anchor points of P2, P3, P4, P5 and P6 is 32. 2 ,64 2 ,128 2 ,256 2 and 512 2 , their anchor frame ratios are uniform {1:2, 1:1, 2:1}.

[0058] The RPN network is used to create candidate regions, which is the candidate region generation part of the improved Faster R-CNN. First, the specified number of anchor boxes can be obtained by setting the anchor point parameter value, and then the positive anchors of the candidate box are determined by the normalized softmax function.

[0059] The region of interest RoI pooling layer is specifically used to adjust the size of the candidate region generated by the RPN network and input the feature data in the candidate region into the classification regression network.

[0060] The classification regression network is specifically used to determine whether it is an insulator based on the characteristic data. If so, the defect type of the insulator is determined based on the characteristic data.

[0061] S3. Based on the enhanced data set, the improved Faster R-CNN network is trained to obtain a fault diagnosis model, where the fault diagnosis model is used to determine the defect type of the insulator based on the collected image data to be diagnosed.

[0062] Specifically, see Figure 5, is a flowchart of training an improved Faster R-CNN network provided in Example 1 of the present application.

[0063] like Figure 5 As shown in , based on the enhanced dataset, training the improved Faster R-CNN network specifically includes:

[0064] The training samples in the enhanced data set are input into the improved Faster R-CNN network. The improved Faster R-CNN network outputs the calculation results and compares them with the preset expected output results. If the calculation results are inconsistent with the preset expected output results, the improved Faster R-CNN network will learn weights based on the training samples, determine the weights through multiple iterations, generate new training samples to train the improved Faster R-CNN network, until the output calculation results are consistent with the preset expected output results, and obtain a fault diagnosis model.

[0065] In summary, this application provides a high-precision detection method for insulator fault defect problems. By abandoning the original feature extraction network VGG16 of Faster R-CNN, using the ResNet50 network and FPN network structure to replace the VGG16 network, increasing the number of anchor points in the RPN network, and increasing the anchor point scale, the structure of the Faster R-CNN model is optimized.

[0066] The advantage of the technical solution of this application is that the complexity and required parameters of the improved Faster R-CNN neural network are reduced. The residual structure is used to build an ultra-deep network structure to alleviate the gradient vanishing phenomenon. At the same time, the FPN network solves the problem of low semantic level of low-level features through top-down and lateral connections, so that low-level features are conducive to small target detection, while ensuring the detection ability of small targets and occluded targets, and improving the detection accuracy and detection performance of insulator detection.

[0067] Embodiment 2:

[0068] See also Figure 6 , which is a structural diagram of an improved Faster R-CNN network structure provided in Example 2 of the present application. As a basic network for transmission line insulator defect detection, the improved Faster R-CNN network consists of four parts, namely image feature extraction, candidate region generation, region of interest pooling, classification, and regression estimation.

[0069] The improved Faster R-CNN first uses convolutional layers and pooling layers to obtain feature maps of all input images, and inputs the feature maps into the subsequent RPN layer and fully connected layer. RPN is used to create candidate regions. First, a specified number of anchor boxes can be obtained by setting the anchor point parameter value, and then the candidate boxes are judged by the softmax function. RoI pooling obtains candidate feature regions of a certain size by adjusting the candidate regions generated by RPN. Finally, the candidate feature region data is injected into the subsequent classification regression network to determine whether it is an insulator, thereby further determining the defect type of the insulator.

[0070] Specifically, Figure 6 As shown in , for any image, first scale the original image to the specified size. Then, send the image to the network. Figure 6 As shown in the figure, the conv layers contain 13 conv layers, 13 relu layers and 4 pooling layers to generate feature maps.

[0071] Entering the RPN network, it first undergoes a 3×3 convolution, generates positive anchors and corresponding bounding box regression offsets, and then calculates proposals.

[0072] Enter the Roi Pooling layer, use proposals to extract proposals from feature maps and send them to the subsequent softmax network for classification, that is, to classify what object the proposal is. Specifically, the candidate feature area data is injected into the subsequent classification regression network to determine whether it is an insulator, thereby further determining the defect type of the insulator.

[0073] Finally, it should be noted that, in this article, 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 sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0074] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0075] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting insulator faults in a transmission line, characterized in that: The method comprises: S1. Acquire an insulator dataset, and simulate a weather scene to enhance the image background in the insulator dataset to obtain an enhanced dataset; S2. Build an improved Faster R-CNN network, including the residual neural network ResNet50, the feature pyramid network FPN, the region generation network RPN, the region of interest RoI pooling layer and the classification regression network; S3. Based on the enhanced data set, the improved Faster R-CNN network is trained to obtain a fault diagnosis model, where the fault diagnosis model is used to determine the defect type of the insulator based on the collected image data to be diagnosed.

2. A method for detecting insulator faults in a transmission line according to claim 1, characterized in that: In S1, after obtaining the insulator data set, the method further includes: The insulator dataset is preprocessed to remove similar images in the insulator dataset.

3. A method for detecting insulator faults in a transmission line according to claim 1, characterized in that: In S1, the image background in the insulator dataset is enhanced, including: Establish a simulated weather scene library, including rain, snow, fog and dark light scenes; Based on any simulated weather scene in the simulated weather scene library, the background environment and morphology of the image background in the insulator dataset are enhanced.

4. A method for detecting insulator faults in a power transmission line according to claim 1, characterized in that: The improved Faster R-CNN network uses the ResNet50 network and the FPN network to extract image features; The ResNet50 network consists of 50 layers and is divided into seven parts; The first part is used to perform convolution, regularization, activation function and maximum pooling operations on the image input to the ResNet50 network; The second to fifth parts contain residual blocks, which are used to continue the convolution operation; The sixth part is used to perform the average pooling operation; The seventh part is the fully connected layer, which is used to obtain the feature map of the image input into the network.

5. A method for detecting insulator faults in a transmission line according to claim 4, characterized in that: The FPN network is used to fuse the features obtained from different feature maps and predict the image depth of the fused feature maps; The feature fusion includes: Using ResNet50 as the backbone network, we get a series of feature maps, including conv1 to conv5. Feature fusion is performed on conv2~conv5, and the feature maps P2~P5 after feature fusion are obtained by upsampling the integer multiples of 2.

6. A method for diagnosing transmission line insulator faults according to claim 5, characterized in that: After feature fusion and upsampling on conv5, it also includes maximum pooling of the feature map after feature fusion to obtain feature map P6. The feature map P6 is used for the regional generation network RPN part, which is the shared weight between the regional generation network RPN and the improved Faster R-CNN.

7. A method for diagnosing transmission line insulator faults according to claim 6, characterized in that: The ratios of the anchor points of feature maps P2, P3, P4, P5, and P6 are 32 2 ,64 2 ,128 2 ,256 2 and 512 2 , the anchor frame ratios are uniform {1:2, 1:1, 2:1}.

8. A method for diagnosing transmission line insulator faults according to claim 1, characterized in that: The improved Faster R-CNN uses the RPN network as the candidate region generation part to create candidate regions and obtain candidate boxes with specified length and width ratios; The RPN network obtains a specified number of anchor boxes through the pre-set anchor point parameter values, and determines the positive anchors of the candidate boxes through the normalized softmax function.

9. A method for diagnosing transmission line insulator faults according to claim 8, characterized in that: The region of interest (RoI) pooling layer is used to adjust the size of the candidate region generated by the RPN network and input the feature data in the candidate region into the classification regression network. The classification regression network is used to determine whether it is an insulator based on the feature data; if so, the defect type of the insulator is determined based on the feature data.

10. A method for diagnosing transmission line insulator faults according to claim 1, characterized in that: In S3, based on the enhanced data set, the improved Faster R-CNN network is trained, including: Input the training samples in the enhanced dataset into the improved Faster R-CNN network; The improved Faster R-CNN network outputs the calculation results and compares them with the preset expected output results; If the calculation result is inconsistent with the preset expected output result, the improved Faster R-CNN network will learn weights based on the training samples, determine the weights through multiple iterations, generate new training samples to train the improved Faster R-CNN network, until the output calculation result is consistent with the preset expected output result, and obtain a fault diagnosis model.

Images