Electrical equipment voltage heating type defect identification method based on temperature field distribution

Abstract

The invention relates to the technical field of voltage-induced heating type defect identification, in particular to an electrical equipment voltage-induced heating type defect identification method based on temperature field distribution, which comprises the following steps of: firstly, quickly identifying and positioning key parts of electrical equipment by adopting an optimized YOLOv4 algorithm, and automatically segmenting an image of a fault target part; then, color, edge and texture features are extracted from the segmented image, and the features are fused into a multi-feature fusion vector by using a DCA fusion algorithm; and finally, carrying out voltage-induced heating type equipment fault identification and classification on the fused feature data by using an SVM. According to the invention, the voltage-induced heating type defect can be effectively identified, and safe operation of a power system is guaranteed.

Description

Technical Field

[0001] The invention relates to the technical field of voltage-induced thermal defect recognition, and in particular to a method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution. Background Art

[0002] Heating failures of electrical equipment in substations account for a large proportion. In the industry specifications for infrared detection of electrical equipment, electrical equipment is divided into current-induced heating type, voltage-induced heating type and comprehensive heating type equipment. The heating state detection of voltage-induced heating type equipment such as high-voltage bushings, insulators, and lightning arresters is relatively difficult, and depends on the environment, the experience of operators, and the level of accurate detection. Substation electrical equipment is in a high-load operation state for a long time. Ensuring its safety and stability is an important research direction for ensuring power supply reliability. Therefore, it is very important to master the correct diagnosis method of voltage-induced heating defects of electrical equipment, monitor its thermal state, and complete thermal state analysis and diagnosis of key parts of electrical equipment, which can effectively ensure the safe and reliable operation of substation electrical equipment and the entire power system.

[0003] At present, the main status and problems of thermal defect identification technology for electrical equipment at home and abroad are summarized as follows:

[0004] 1) Based on the structured positioning method of CenterNet model, the infrared image defect detection method of power equipment is used. The algorithm model of CenterNet combined with structured positioning is adopted. Through the collection, training and verification of on-site infrared image data samples, different substation equipment and its components can be identified and located with high accuracy from complex infrared images. According to the surface temperature range of equipment components and the type of substation equipment identified and located, the infrared image defect detection of power equipment is realized in combination with relevant temperature specifications. This method improves the accuracy of infrared image defect detection of power equipment, and completes the thermal type defect detection of electric heating devices in combination with temperature specifications. However, the defect recognition rate of this method is low.

[0005] 2) Based on image enhancement and deep learning, the infrared thermal image recognition method of substation equipment is proposed. In view of the low contrast, blurred edges and complex background of substation infrared thermal images of substation equipment, the parameter self-adjustment Retinex infrared thermal image enhancement method combined with fast guided filtering is proposed to provide conditions for the accurate recognition of infrared thermal images of substation equipment, and the feature extraction network and loss function of YOLOv3 convolutional neural network are improved to solve the problem of low accuracy of infrared thermal image recognition of substation equipment. However, this method has the problem of high false alarm rate of defects.

[0006] 3) Infrared defect recognition method for substation equipment based on deep learning, using Faster RCNN algorithm to detect seven types of substation equipment, including transformers, bushings, circuit breakers, disconnectors, voltage transformers, current transformers, and lightning arresters, to achieve accurate positioning and identification of the equipment, and then using the temperature threshold discrimination method to identify thermal defects in the equipment area to determine whether the equipment has defects and the severity, and the experiment verified its feasibility and accuracy. However, this method fails to consider the impact of low contrast of infrared images on the recognition results, and the two-stage model detection such as Faster RCNN takes a long time and cannot meet the real-time requirements.

[0007] At present, the technology for identifying thermal defects in electrical equipment is difficult to apply to scenarios where the electrical equipment background is complex, multiple devices have similar appearances, and defects manifest in various forms. In addition, the current infrared image defect diagnosis basically identifies and diagnoses defects of single devices, making it difficult to detect voltage-induced thermal defects with smaller temperature rise. Summary of the invention

[0008] In order to realize intelligent diagnosis of voltage-induced thermal defects, the present invention proposes a method for identifying voltage-induced thermal defects of electrical equipment based on temperature field changes, and adopts infrared diagnosis of voltage-induced thermal defects by YOLOv4 and multi-feature fusion, which provides conditions and new ideas for voltage-induced thermal fault detection, and lays the foundation for real-time and accurate monitoring of substations, realizing intelligent and precise diagnosis of thermal faults of electrical equipment.

[0009] In order to achieve the purpose of the present invention, the technical solution adopted is: a method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution, comprising:

[0010] 1) The voltage-induced thermal defect recognition method for electrical equipment based on temperature field distribution uses the optimized YOLOv4 algorithm to quickly identify and locate the key parts of electrical equipment with voltage-induced thermal defects, and automatically segment and save the fault target parts;

[0011] 2) Then, the color, edge and texture features of the image of the segmented target part are extracted, and then the three features are fused through the DCA (discriminant correlation analysis) fusion algorithm to obtain a multi-feature fusion vector;

[0012] 3) Finally, based on the data set established by segmenting the target area image, support vector machine (SVM) is used to realize the recognition and classification of voltage-induced thermal equipment faults.

[0013] As an optimization scheme of the present invention, optimizing the YOLOv4 algorithm means: in view of the problem of uneven background complexity in infrared images, the YOLOv4 algorithm is optimized in parameters, loss function and inspection frame respectively; the Adam optimizer is selected for parameter optimization; the first-order matrix estimation and second-order matrix estimation of the gradient are used, the gradient algorithm and the RMS gradient descent algorithm are integrated, and the optimized loss function consists of two parts: a confidence loss function and a position regression loss function; there are relatively many candidate frames (Default boxes) in the infrared images of electrical equipment with voltage-induced heating defects, and they should be suppressed, and the frame with the highest score is retained as the inspection frame.

[0014] As an optimization scheme of the present invention, extracting the color, edge and texture features of the image of the segmented target part refers to: taking a color feature histogram based on the HSV (hue, saturation, value) space, and calculating the edge features of the target image, and extracting the edge feature histogram of the fault area of ​​the infrared image by calculating the gradient size and direction of each pixel in the infrared image; using the centrally symmetric local binary pattern (OCS-LBP), calculating the difference between relative pixels in the neighborhood, and extracting the texture features of the identified target. This process does not require too many parameters and has a strong ability to resist noise.

[0015] As an optimization scheme of the present invention, color feature extraction means: color histogram is important information of infrared image features, HSV (hue, saturation, brightness) is closer to the human eye's definition of color, and a two-dimensional histogram method is used to divide the HSV space into 8 tones, each tone is divided into 8 brightnesses, and each image can be represented as a 64-dimensional vector. The feature data of each component of H, S, and V are synthesized into a one-dimensional vector as an input vector in the fusion feature, thereby improving the recognition rate of various image features.

[0016] As an optimization scheme of the present invention, calculating the edge features of the target image refers to calculating the contour features using the gradient operator of the Sobel edge detector. Edge extraction simplifies the analysis of infrared images by reducing redundant information and retains the boundary structure information of key parts of the equipment. The Sobel edge detector is a better edge detector and is widely used in the field of image processing, based on the characteristics and analysis of infrared images of electrical equipment.

[0017] As an optimization scheme of the present invention, fusing three features through the DCA fusion algorithm means: fusing color, contour and texture features based on discriminant correlation analysis, and feature fusion combines two or more related features extracted from the image scene into a feature vector with more discriminative information than the input feature vector.

[0018] As an optimization scheme of the present invention, the recognition and classification of voltage-induced thermal equipment faults using a support vector machine (SVM) refers to: inputting the fused feature data of the sorted infrared image of the electrical equipment into the SVM classifier, and performing defect recognition and fault classification according to the following formula:

[0019]

[0020] Among them, sgn(.) is the classification function, ω0 is the hyperplane normal vector, X (t) is the test sample, b0 is the hyperplane constant term.

[0021] The present invention has positive effects: 1) The voltage-induced thermal device infrared image target detection algorithm of YOLOv4 optimized by the present invention optimizes the algorithm parameters and loss function, improves the speed and accuracy of the detection performance of the voltage-induced thermal defect target device, can weaken the background and other interference information, and can accurately locate and identify the target device in the infrared image, laying a foundation for the subsequent fault diagnosis of the voltage-induced thermal defect;

[0022] 2) The present invention adopts a method of extracting and fusing multiple features of images, and innovatively adopts a method of extracting multiple features to extract color features, contour features and texture features from the infrared image of the voltage-induced heating equipment, and fuses them using the DCA (discriminant correlation analysis) fusion method. A support vector machine suitable for small sample classification is used to perform accurate fault diagnosis on the inspected parts of the voltage-induced heating equipment.

[0023] 3) The DCA (discriminant correlation analysis) fusion method of the present invention can well express low-dimensional image scenes, reduce the influence of environmental factors on fault target feature recognition to a certain extent, and is superior to the method of using only a single feature, providing conditions for voltage-induced thermal defect recognition and detection, and laying a foundation for real-time and accurate monitoring of substations;

[0024] 4) The present invention can effectively identify voltage-induced thermal defects, monitor the thermal status of electrical equipment, and complete thermal status analysis and diagnosis of key parts of electrical equipment, which can effectively ensure the safe and reliable operation of substation electrical equipment and even the entire power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0026] Figure 1The fourth embodiment of the present invention provides a process for identifying voltage-type thermal defects and diagnosing faults of electrical equipment based on temperature field distribution. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0028] The terms "first" and "second" and the like in the specification and claims of the present invention and the above drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may include steps or units that are not listed.

[0029] Embodiment 1

[0030] At present, there are similar problems in infrared image target recognition of electrical equipment, especially voltage-induced thermal equipment. Some equipment is densely laid out and the distance is far when collecting infrared images of the equipment. There are many interference factors in the background environment. In most cases, the problem in infrared images is mainly the uneven background complexity. Therefore, the following optimizations are made for the YOLOv4 algorithm:

[0031] 1. Parameter optimization: Use the Adam Optimizer algorithm to update the parameters. The parameter update formula is as follows:

[0032] μ i-1 * =μ i-1 -β(α1+g(μ i-1 ))

[0033] Among them, α1 is a hyperparameter, β is the learning rate, μ i-1 is the parameter to be updated, μ i-1 * is the updated parameter, g(μ i-1 ) is the gradient of the loss function. Adam Optimize r uses the first-order matrix estimation and second-order matrix estimation of the gradient. Different parameters have corresponding adaptive learning rates. It can save the average gradient of the previous decay. It is an optimization algorithm that extracts the best global value and combines the advantages of the gradient algorithm and the RMS gradient descent algorithm.

[0034] 2. Loss function optimization: The optimized loss function of YOLOv4 consists of two parts: the confidence loss function and the position regression loss function in the following formula:

[0035]

[0036] Among them, L is the optimized YOLOV4 loss function, L loc is the position regression loss function, L conf is the confidence loss function, M is the number of candidate boxes matching the target area, It is used to adjust the ratio of the confidence loss function and the position regression loss function.

[0037] 3. Inspection frame optimization: In the infrared image of voltage-induced thermal equipment, there are very few reference frames and relatively more candidate frames, so the non-maximum suppression module is used for detection. The frame with the highest score is retained, and other detection frames that have obvious overlap with the selected frame are suppressed.

[0038] Embodiment 2

[0039] Compared with the improved YOLO disclosed in the first embodiment of the present invention v 4 algorithm, the second embodiment of the present invention also provides a specific implementation method for infrared image multi-feature extraction, the method comprising:

[0040] Single feature using improved YOLO v 4 algorithm has a low recognition rate when completing target recognition. It is difficult for a single feature to represent the target feature of the fault image. Therefore, the present invention proposes a voltage-induced thermal defect target recognition method based on multi-feature fusion, and extracts the following features respectively:

[0041] 1. Color feature extraction: Color histogram is an important information of infrared image features. HSV (hue, saturation, brightness) is closer to the human eye's definition of color. The two-dimensional histogram method is used to divide the HSV space into 8 tones, each tones is divided into 8 brightnesses, and each image can be represented as a 64-dimensional vector. The feature data of H, S, and V components are synthesized into a one-dimensional vector as an input vector in the fusion feature, which improves the recognition rate of various image features.

[0042] 2. Edge feature extraction: Edge extraction simplifies the analysis of infrared images by reducing redundant information and retains the boundary structure information of key parts of the equipment. The Sobel edge detector is a better edge detector and is widely used in the field of image processing. Based on the characteristics and analysis of infrared images of electrical equipment, the present invention uses the gradient operator of the Sobel edge detector to calculate the contour features. The theoretical basis is the following formula:

[0043]

[0044] Among them, G x , G Y are the gradient values ​​corresponding to the horizontal and vertical coordinates of the pixel point, and δf.δ× are the partial derivative values ​​of the horizontal and vertical coordinates of the pixel point.

[0045] 3. Texture feature extraction: Since infrared images often overlap with similar targets in the environment, the ability to describe infrared targets is reduced. Texture features do not rely on color and contour, and the target recognition effect is better. Because the original feature quantity will generate a very long histogram, the calculation time will increase. In addition, the feature does not contain the direction information of the pixel point, so the improved texture feature extraction method is used: the central symmetric local binary pattern (OCS-LBP). The method of extracting the texture features of the identified target at different environmental background scales is shown in the following formula:

[0046]

[0047] Among them, p is the number of all pixels, y c is the center pixel, g c is the center pixel y c The gray value size, g p is the gray value of the pixel associated with the central pixel, and T is a comparison threshold.

[0048] Embodiment 3

[0049] Corresponding to the infrared image multi-feature extraction disclosed in the second embodiment of the present invention, the third embodiment of the present invention further provides a specific implementation method for infrared image feature fusion, which includes:

[0050] Based on discriminant correlation analysis, the color, contour and texture features are fused. The purpose of feature fusion is to combine two or more related features extracted from the image scene into a feature vector with more identification information than the input feature vector. The fusion feature strategy adopted by the present invention is DCA (discriminant correlation analysis), which maximizes the correlation between feature sets and the difference between classes. Let D represent the image data set, X represent the data matrix of the feature set collected from the image data set D, c represent the feature vector extracted from the i-th image of the i-th category, and assume Represents the entire data matrix Category I The mean of a vector:

[0051]

[0052] Where: n represents the total number of image categories, n i Represents the number of images belonging to the i-th category. The data matrix after feature fusion is shown as follows:

[0053]

[0054] Embodiment 4

[0055] Corresponding to the improved YOLOv4 algorithm, infrared image multi-feature extraction method and infrared image feature fusion method disclosed in the first, second and third embodiments of the present invention, the fourth embodiment of the present invention further provides a specific implementation method of the electrical equipment voltage-type thermal defect recognition and fault diagnosis process based on temperature field distribution, which method includes:

[0056] The fused feature data of the organized infrared images of electrical equipment are input into the SVM classifier, and then defect recognition and fault classification can be performed according to the following formula:

[0057]

[0058] Among them, sgn(.) is the classification function, ω0 is the hyperplane normal vector, X (t) is the test sample, b0 is the hyperplane constant term.

[0059] The voltage-type thermal defect recognition and fault diagnosis process of electrical equipment based on temperature field distribution established by the present invention first uses the improved YOLOv4 target detection algorithm to extract and save the key equipment from the infrared image to be detected, then extracts the fusion features of the target part, and uses the SVM support vector machine to identify the defect. The specific process is as follows: Figure 1 shown.

[0060] 1. Identify and save the target parts of electrical equipment, based on the saved key equipment, and fuse the three features extracted through DCA (Discriminant Correlation Analysis).

[0061] 2. Perform fault diagnosis according to thermal fault diagnosis rules. Identify the type of electrical equipment through the established scheme and determine whether it has a voltage-induced thermal defect. Display the identification result on the software interface to complete the electrical equipment voltage-induced thermal defect identification scheme.

[0062] The technical focus of the present invention is to use the optimized YOLOv4 algorithm to quickly identify and locate the key parts of electrical equipment with voltage-induced heating defects, realize automatic segmentation and preservation of the fault target part, then extract the color, edge and texture features of the image of the segmented target part, and then fuse the three features through the DCA (discriminant correlation analysis) fusion algorithm to obtain a multi-feature fusion vector, and finally, according to the established data set, use the support vector machine SVM to realize the identification and classification of voltage-induced heating equipment faults.

[0063] The technical solution of the present invention is:

[0064] 1. Optimize the YOLOv4 algorithm. To address the problem of uneven background complexity in infrared images, the YOLOv4 algorithm is optimized for parameters, loss function, and test box. Adam Optimizer is used for parameter optimization. The first-order matrix estimation and second-order matrix estimation of the gradient are used to combine the advantages of the gradient algorithm and the RMS gradient descent algorithm. The optimized loss function consists of two parts: the confidence loss function and the position regression loss function. In the infrared images of electrical equipment with voltage-induced heating defects, there are relatively many default boxes, which should be suppressed and the boxes with the highest scores should be retained.

[0065] 2. Based on the infrared image feature extraction of voltage-induced thermal defective equipment, the color feature histogram based on the HSV (hue, saturation, brightness) space is extracted, and the edge features of the target image are calculated. By calculating the gradient size and direction of each pixel in the infrared image, the edge feature histogram of the fault area of ​​the infrared image is extracted. For texture features, OCS-LBP (centrosymmetric local binary pattern) is used to calculate the difference between relative pixels in the neighborhood. It does not require too many parameters and has a strong ability to resist noise. Then, the three features of color, texture and edge of the target are fused through DCA (discriminant correlation analysis) as a feature fusion strategy.

[0066] 3. To diagnose voltage-induced thermal defects, first use the improved YOLOv4 target detection algorithm to extract and save the key devices in the image to be inspected, then extract the fusion features of the target part, and finally give the diagnosis result through the support vector machine.

[0067] In the process of fault diagnosis of current-induced thermal equipment becoming more and more mature, the fault diagnosis effect of voltage-induced thermal equipment is not satisfactory. Therefore, the present invention studies the diagnosis method of voltage-induced thermal equipment. In order to cope with the problem that voltage-induced thermal defects with small temperature rise are difficult to find in scenarios where the background of electrical equipment is complex, the appearance of multiple devices is similar, and the defect manifestations are diverse, a method for extracting and fusing multiple features of infrared images is proposed, and a thermal fault diagnosis process of electrical equipment is established based on these studies.

[0068] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution, characterized in that: include: 1) The voltage-induced thermal defect recognition method for electrical equipment based on temperature field distribution uses the optimized YOLOv4 algorithm to quickly identify and locate the key parts of electrical equipment with voltage-induced thermal defects, and automatically segment and save the fault target parts; 2) Then, the color, edge and texture features of the image of the segmented target part are extracted, and then the three features are fused through the DCA fusion algorithm to obtain a multi-feature fusion vector; 3) Finally, based on the data set established by segmenting the target area image, support vector machine (SVM) is used to realize the recognition and classification of voltage-induced thermal equipment faults.

2. The method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution according to claim 1, characterized in that: Optimizing the YOLOv4 algorithm means: performing parameter optimization, loss function optimization and inspection frame optimization on the YOLOv4 algorithm respectively; selecting the Adam optimizer for parameter optimization; utilizing the first-order matrix estimation and the second-order matrix estimation of the gradient, the integrated gradient algorithm and the RMS gradient descent algorithm, and the optimized loss function is composed of two parts: the confidence loss function and the position regression loss function; for the candidate frames in the infrared images of electrical equipment with voltage-induced heating defects, retaining the frame with the highest score as the inspection frame.

3. The method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution according to claim 1, characterized in that: Extracting the color, edge and texture features of the image of the segmented target area means: taking the color feature histogram based on the HSV space and calculating the edge features of the target image, and extracting the edge feature histogram of the fault area of ​​the infrared image by calculating the gradient size and direction of each pixel in the infrared image; using the centrally symmetric local binary pattern, calculating the difference between relative pixels in the neighborhood, and extracting the texture features of the identified target.

4. The method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution according to claim 3 is characterized in that: Color feature extraction means: using a two-dimensional histogram method to divide the HSV space into 8 tones, each tone is divided into 8 brightnesses, each image can be represented as a 64-dimensional vector, and the feature data of each component of H, S, and V are synthesized into a one-dimensional vector as an input vector in the fusion feature.

5. The method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution according to claim 3, characterized in that: Calculating the edge features of the target image refers to calculating the contour features using the gradient operator of the Sobel edge detector.

6. The method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution according to claim 1, characterized in that: The fusion of three features through the DCA fusion algorithm means: fusing color, contour and texture features based on discriminant correlation analysis. Feature fusion combines two or more related features extracted from the image scene into a feature vector with identification information.

7. The method for identifying voltage-induced thermal defects of electrical equipment based on temperature field distribution according to claim 1, characterized in that: Using support vector machine (SVM) to realize the recognition and classification of voltage-induced thermal equipment faults means: inputting the fused feature data of the sorted infrared image of electrical equipment into the SVM classifier, and performing defect recognition and fault classification according to the following formula: Among them, sgn(.) is the classification function, ω0 is the hyperplane normal vector, X (t) is the test sample, b0 is the hyperplane constant term.

Images