Transformer vibration detection methods

Abstract

A transformer vibration detection method is disclosed. The method involves acquiring real-time video signals of the transformer, using a Faster R-CNN model for transformer detection, extracting image features of the transformer based on the video signal using a shared basic convolutional layer, generating candidate detection boxes based on image features using a candidate detection box generation network connected to the shared basic convolutional layer, and classifying the candidate detection boxes using a region of interest pooling and classification network. The classified candidate detection boxes are then adjusted to output the detection results. Feature points are extracted within the target region using the Shi-Tomasi method; vibration vectors of the feature points are calculated using the pyramid L-K optical flow method; K-means clustering is used to remove vibration vectors with deviations greater than a predetermined deviation as noise; and the transformer vibration vector is obtained by averaging the vibration vectors of the feature points.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of transformer vibration detection, and particularly relates to a transformer vibration detection method. BACKGROUND

[0002] In recent years, large-scale renewable energy access substations bring problems such as overload, harmonics, short circuit, and further cause the increase of fault rate and the shortening of service life of important electrical equipment such as transformers. The transformer is one of the key equipment of the power system, and its operating condition has an important influence on the safe and stable operation of the power grid. Therefore, using various measures to monitor the operating state of the transformer and identify hidden dangers is an important work to ensure the safe and stable operation of the power system. With the progress of technology, the maintenance of the transformer gradually adopts condition-based maintenance instead of the traditional periodic maintenance and post-accident maintenance. At present, the transformer state evaluation method is mainly divided into online monitoring and offline detection. Online monitoring can make the transformer not need to stop running, can save manpower and material resources, and has obvious advantages.

[0003] Obtaining rich online monitoring data is the key to implementing state evaluation of the power transformer. According to different state information obtained, the method of online monitoring of the power transformer includes: oil dissolved gas analysis, oil temperature monitoring, partial discharge measurement, oil furfural content, frequency response analysis, short-circuit impedance analysis, and vibration signal monitoring. Among them, one of the advantages of the vibration analysis method is that the detection system does not have any form of electrical connection with the measured transformer, and does not affect the normal operation of the power grid, fully ensuring the safety of online monitoring.

[0004] The detection of vibration signals is an important prerequisite for analyzing and evaluating the operating state and fault diagnosis of the transformer by using the vibration analysis method. The detection method of vibration signals can be divided into contact type vibration measurement and non-contact type vibration measurement. The traditional contact type vibration measurement mainly adopts the method of installing sensors on site, which has many defects such as limited measurement range and the need for point-by-point arrangement. The commonly used non-contact vibration measurement mainly includes laser vibration measurement and visual vibration measurement based on image video. Although the laser vibration measurement method has the advantages of high precision and sensitivity, long measurement distance, and high measurement frequency, the related equipment is very expensive, and the requirement for the professionalism of the operator is also high, which greatly restricts its wide application and promotion.

[0005] As a new emerging vibration measurement method, the visual-based vibration detection has attracted widespread attention from scholars at home and abroad. Video image monitoring is a non-contact monitoring method that can not only measure displacement, strain and other static parameters, but also be suitable for dynamic characteristic measurement. It has the advantages of simple operation, non-contact, non-destructive, no additional mass, and can realize remote, wide-range and multi-point monitoring. However, it also has some disadvantages such as the need to set up a target, high requirements for monitoring equipment, high requirements for environmental light and background, etc.

[0006] The above information disclosed in the Background section is only for enhancing the understanding of the background of the present application, and therefore can contain information that does not constitute the prior art that is already known to those of ordinary skill in the art. SUMMARY

[0007] The object of the present application is to provide a transformer vibration detection method. In order to achieve the above object, the present application provides the following technical solutions:

[0008] The transformer vibration detection method comprises the following steps:

[0009] Step S100: Collecting a real-time video signal of a transformer, and detecting the transformer by using a Faster RCNN model, wherein the Faster RCNN model comprises,

[0010] a backbone network VGG-16, using the VGG-16 to extract features of the video signal to generate a feature map layer,

[0011] a region proposal network, the feature map layer being input into the region proposal network, and a plurality of candidate boxes being generated by using a sliding window mode,

[0012] a region of interest pooling, the input feature map layer and the candidate boxes of different scales being converted into an output of a unified scale, and then being input into a fully connected layer to obtain a feature vector,

[0013] softmax classification and bounding box regression, the feature vector being input into a softmax classifier to perform target classification, and a final position of the target being obtained by using bounding box regression;

[0014] Step S200: Extracting feature points in a target region by using a Shi-Tomasi method;

[0015] Step S300: Calculating vibration vectors of the feature points by using a pyramid L-K optical flow method;

[0016] Step S400: Removing vibration vectors with a deviation greater than a predetermined deviation as noise by using a K-means clustering algorithm;

[0017] Step S500: Obtaining a transformer vibration vector by averaging the vibration vectors of the feature points.

[0018] In the transformer vibration detection method, the pyramid L-K optical flow method is expressed as:

[0019] I x (q1)V x +I y (q1)V y =-I t (q1)

[0020] I x (q2)V x +I y (q2)V y =-I t (q2)

[0021]

[0022] I x (q n V x +I y (q n V y =-I t (q n ),

[0023] Where q1, q2, ..., q n I represents the pixels contained within a small window, which delineates an N×N region around a feature point. x (q i ), I y (q i ) and I t (q i ) represent pixels q i The partial derivatives of the light intensity I with respect to the x, y, and t directions, V x V represents the speed at which a pixel moves in the x-direction. y The speed at which the pixel moves in the y-direction.

[0024] In the aforementioned transformer vibration detection method, a camera captures real-time video signals from the front of the transformer.

[0025] In the transformer vibration detection method described above, the imaging device includes a CCD camera.

[0026] In the above technical solution, the transformer vibration detection method provided by the present invention has the following beneficial effects: This method acquires real-time video signals of the transformer, uses the Faster RCNN model to detect the transformer as the region of interest, thereby avoiding manual target setting. The Shi-Tomasi method is used to extract feature points in the region of interest of the transformer and calculate the transformer vibration vector, thereby avoiding interference from ambient light and background factors. The present invention uses the pyramid LK optical flow method to calculate the vibration vector of all feature points in the region of interest, and further uses the K-means algorithm to perform cluster analysis on the vibration vector of the feature points to remove vibration vectors with large deviations, thereby improving the calculation accuracy of the transformer vibration vector. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0028] Figure 1 This is a flowchart illustrating one embodiment of the transformer vibration detection method of the present invention;

[0029] Figure 2 A schematic diagram of the Faster RCNN model structure is provided for one embodiment of the transformer vibration detection method of the present invention;

[0030] Figure 3 This is a schematic diagram of the detection results of a transformer image using a Faster RCNN model, which is an embodiment of the transformer vibration detection method of the present invention.

[0031] Figure 4 This is a schematic diagram of feature points extracted within the transformer region of interest using the Shi-Tomasi method in the transformer vibration detection method of this invention.

[0032] Figure 5 This is a schematic diagram illustrating the estimation of optical flow for black pixels using the Lucas-Kanade method, as provided in an embodiment of the transformer vibration detection method of the present invention.

[0033] Figure 6 This is a schematic diagram illustrating feature point tracking from the highest level to the lowest level of the pyramid, as provided in an embodiment of the transformer vibration detection method of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Therefore, the following is an explanation of the figures. Figures 1 to 6 The detailed description of the embodiments of the present invention provided herein is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0036] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0037] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0038] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0039] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0040] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0041] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings. A transformer vibration detection method includes,

[0042] Step S100: Acquire real-time video signals of the transformer and use the Faster R-CNN model for transformer detection. The Faster R-CNN model includes...

[0043] Step S100: Acquire real-time video signals of the transformer and use the Faster R-CNN model for transformer detection. The Faster R-CNN model includes...

[0044] The backbone network is VGG-16, which is used to extract features from the video signal and generate a feature layer.

[0045] A region candidate network is used, where the feature layer is input and multiple candidate boxes are generated using a sliding window approach.

[0046] Region-of-interest pooling transforms the input feature layers and candidate boxes of different scales into a uniform-scale output, which is then fed into a fully connected layer to obtain the feature vector.

[0047] Softmax classification and bounding box regression are used to classify the target by inputting the feature vector into the softmax classifier and to obtain the final position of the target by using bounding box regression.

[0048] Step S200: Extract feature points within the target region using the Shi-Tomasi method;

[0049] Step S300: Calculate the vibration vector of the feature point using the pyramid LK optical flow method;

[0050] Step S400: Use the K-means clustering algorithm to remove vibration vectors with deviations greater than a predetermined deviation as noise;

[0051] Step S500: Average the vibration vector of the feature point to obtain the transformer vibration vector.

[0052] In a preferred embodiment of the transformer vibration detection method, the pyramid LK optical flow method is expressed as:

[0053] I x (q1)V x +I y (q1)V y =-I t (q1)

[0054] I x (q2)V x +I y (q2)V y =-I t (q2)

[0055]

[0056] I x (q n V x +I y (q n V y =-I t (q n ),

[0057] Where q1, q2, ..., q n I represents the pixels contained within a small window, which delineates an N×N region around a feature point. x (q i ), I y (q i ) and I t (q i ) represent pixels q i The partial derivatives of the light intensity I with respect to the x, y, and t directions, V x V represents the velocity in the x-direction; y The velocity is the speed of movement in the y-direction.

[0058] In a preferred embodiment of the transformer vibration detection method, the imaging device acquires real-time video signals from the front of the transformer.

[0059] In a preferred embodiment of the transformer vibration detection method, the imaging device includes a CCD camera.

[0060] Preferably, the pyramid LK optical flow method is represented in matrix form as follows:

[0061]

[0062] In one embodiment, such as Figure 1 As shown, a transformer vibration detection method includes the following steps:

[0063] S100: Transformer detection using the Faster RCNN model;

[0064] In this step, Figure 2 This is a schematic diagram of the Faster RCNN model structure. Figure 3 The image shows the results of transformer detection using the Faster RCNN model. The area within the red box in the image represents the transformer of interest.

[0065] S200: Feature points are extracted within the transformer region of interest using the Shi-Tomasi method;

[0066] In this step, such asFigure 4 The green dots represent the feature points extracted from the transformer of interest region.

[0067] S300: Calculate the vibration vector of all feature points within the transformer region of interest using the pyramid LK optical flow method;

[0068] S400: The K-means clustering algorithm is used to remove vibration vectors with large deviations as noise;

[0069] S500: Calculate the transformer vibration vector by averaging the vibration vectors at characteristic points.

[0070] The above embodiments constitute the complete technical solution of this disclosure. This embodiment, based on real-time video signals of the transformer, utilizes transfer learning and the Faster R-CNN model to detect the transformer as the region of interest, thereby avoiding manual target setting. Secondly, this disclosure uses the Shi-Tomasi method to extract feature points within the region of interest and calculate the transformer vibration vector, thus avoiding interference from ambient light and background factors. Finally, this disclosure uses the pyramid LK optical flow method to calculate the vibration vectors of all feature points within the region of interest, and further utilizes the K-means algorithm to perform cluster analysis on the feature point vibration vectors, removing vibration vectors with large deviations, thereby improving the calculation accuracy of the transformer vibration vector.

[0071] In another embodiment, such as Figure 2 As shown, the Faster R-CNN model includes:

[0072] The backbone network VGG-16 is used to extract features from the input image and generate a feature layer.

[0073] The feature layer generated by the region candidate network is input into the region candidate network, and multiple candidate boxes are generated using a sliding window method.

[0074] Region of interest pooling converts input feature maps and candidate boxes of different scales into outputs of a uniform scale, which are then input into a fully connected layer to obtain feature vectors.

[0075] Softmax classification and bounding box regression are used to classify the target by inputting the feature vector into the softmax classifier, and at the same time, bounding box regression is used to obtain the final position of the target.

[0076] In another embodiment, step S100 includes the following steps:

[0077] (1) Environment Setup. A transformer target detection model was built using Faster R-CNN. The environment was set to Windows 10 64-bit operating system, a 32-core Intel Xeon E5-2695v3 CPU, 32GB of RAM, an NVIDIA Grid P40-24Q GPU with NVIDIA driver version 441.66 and 24GB of VRAM. The deep learning framework used was TensorFlow-GPU 2.2, CUDA version 10.1, and CUDNN version 7.6.5.32.

[0078] (2) Dataset Construction. Transformer images were selected, organized, and labeled to construct a transformer target detection dataset. The dataset contains 489 transformer images of different types, angles, and environments. The dataset was randomly divided into a training set, a validation set, and a test set, with 396 images in the training set, 44 images in the validation set, and 49 images in the test set. During model training, random data augmentation was performed on the training set data, including scaling, width and height distortion, flipping, and color gamut distortion.

[0079] (3) Model Training. The training of the Faster-RCNN transformer object detection model incorporates the concept of transfer learning, using pre-trained weights from Faster-RCNN on ImageNet to aid in model training. The model undergoes 100 iterations of training, divided into two phases: frozen training and unfrozen training. In the first 50 iterations of the first phase, the feature extraction layers of the pre-trained model, i.e., the weights of the first 17 layers, are frozen, with a batch size of 4 and an initial learning rate of l. r =0.0001. Also, after each iteration of the model, press l r =l r * The learning rate is automatically reduced by 0.96. In the last 50 iterations of the second phase, the weights of the model's feature extraction layer are unfrozen during training, with a batch size of 2 and an initial learning rate of l. r =0.00001, the learning rate decrease method is the same as in the first stage.

[0080] (4) Transformer Detection. The image of the transformer to be detected is scaled to the model input size, and the trained Faster-RCNN model is used for target detection. The output transformer detection result is as follows: Figure 3 As shown.

[0081] In another embodiment, step S300 includes the following steps:

[0082] (1) The implementation of the optical flow method is based on the premise that the image intensity remains unchanged and the target motion amplitude is small. Based on the first premise, that is, the light intensity remains unchanged after the pixel is displaced, the following constraint equation can be established:

[0083] I(x,y,t)=I(x+dx,y+dy,t+dt) (1)

[0084] (2) After removing the common terms, the right side of equation (1) is simplified by Taylor expansion to obtain:

[0085]

[0086] (3) Dividing formula (2) by dt, we get:

[0087]

[0088] in, Represent the two unknown variables that need to be solved; and These represent the partial derivatives of the gray level of a pixel in the image along the x, y, and t directions, respectively, and can be calculated based on the actual image data. Since there is only one constraint equation (3), it is impossible to solve for two unknowns. Therefore, the LK algorithm is introduced to solve this problem.

[0089] (4) The LK optical flow method adds the assumption of "spatial consistency". Based on this assumption, an N×N window can be drawn around the feature point, and it is assumed that all pixels within the window have the same movement, such as... Figure 5 As shown.

[0090] (5) The LK optical flow method described above can be described by the following formula:

[0091]

[0092] In the formula, q1, q2, ..., q n I represents the number of pixels contained within the small window. x (q i ), I y (q i ) and I t (q i ) represent pixels q i The partial derivatives of the light intensity I in the x, y, t directions.

[0093] (6) The system of equations (4) can be represented in the following matrix form:

[0094]

[0095] In the basic optical flow method, because there is only one constraint equation, it is impossible to solve for the two unknowns. However, in the LK optical flow method, there are n equations (n>2), which become an overdetermined system of equations.

[0096] (7) Solving the system of equations (5) using the least squares method yields:

[0097]

[0098] in, This represents the speed of movement in the x-direction; This represents the speed of movement in the y-direction.

[0099] (8) The LK optical flow method is based on the assumption that the object's motion amplitude is small. When the object's motion amplitude is large, a large error will occur. The pyramid method is used to improve the LK optical flow method. For example... Figure 6 As shown, the pyramid LK optical flow method reduces the large displacement in the high-level pyramid image by downsampling the image, thereby obtaining a more accurate optical flow vector at this scale.

[0100] (9) Then, when solving from the top layer down, the high-level vectors are enlarged proportionally as the initial guide for the next layer. At this time, there will be an error between the target position indicated by the optical flow vector enlarged from the high layer and the actual target position of the current layer. However, this error usually conforms to the scale of small motion. Therefore, the optical flow vector of the current layer can be calculated on this basis. When the process is repeated to reach the bottom layer original image, a more accurate optical flow vector under large motion can be obtained.

[0101] Finally, it should be noted that the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0102] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for detecting transformer vibration, characterized in that, It includes the following steps: Step S100: Acquire real-time video signals of the transformer and use the Faster R-CNN model for transformer detection. The Faster R-CNN model includes... The backbone network VGG-16 is used to extract features from the video signal and generate a feature layer. A region candidate network is used, where the feature layer is input into the region candidate network, and multiple candidate boxes are generated using a sliding window method. Region of interest pooling transforms the input feature layers and candidate boxes of different scales into a uniform scale output, which is then fed into a fully connected layer to obtain the feature vector. Softmax classification and bounding box regression are used to classify the target by inputting the feature vector into a softmax classifier, while bounding box regression is used to obtain the final position of the target. A camera captures real-time video signals from the front of the transformer. Pre-trained weights of Faster-RCNN on ImageNet are used to train the model. The model is trained iteratively 100 times, divided into two phases: frozen training and unfrozen training. In the first 50 iterations of the frozen training phase, the feature extraction layers of the pre-trained model (the first 17 layers) are frozen, and the batch size is set to 4 with an initial learning rate of... At the same time, after each iteration of the model, according to The learning rate is automatically reduced. In the last 50 iterations of the unfrozen training phase, the weights of the model's feature extraction layer are unfrozen and trained, with a batch size of 2 and an initial learning rate of 1. The learning rate reduction method is the same as in the first stage. The image of the transformer to be tested is scaled to the model input size, and the trained Faster-RCNN model is used for target detection. Step S200: Extract feature points within the target region using the Shi-Tomasi method; Step S300: Calculate the vibration vector of the feature point using the pyramid LK optical flow method; Step S400: Use the K-means clustering algorithm to remove vibration vectors with deviations greater than a predetermined deviation as noise; Step S500: Average the vibration vector of the feature point to obtain the transformer vibration vector.

2. The transformer vibration detection method according to claim 1, characterized in that, The pyramid LK optical flow method is represented as: ， in, , , , This represents the number of pixels contained within a small window, which is a boundary drawn around a feature point. Scope , and Representing pixels light intensity exist , , Partial derivatives in the direction, For pixels in Speed ​​of movement in a certain direction; For pixels in The speed of movement in a certain direction.

3. The transformer vibration detection method according to claim 1, characterized in that, The shooting equipment included a CCD camera.

Images