Horizontal rotating knife switch closing-in-place recognition device and closing state judgment method

Abstract

The application discloses a horizontal rotary knife switch closing-in-place recognition device and a closing-in-state judgment method, which are used for recognizing whether a horizontal rotary knife switch of a transformer substation is accurately closed in place. The device comprises a camera, a horizontal rotary knife switch recognition front-end device, a horizontal rotary knife switch recognition background learning and training device and a data and control bus. The horizontal rotary knife switch is extremely important equipment in the transformer substation, and long-term use can cause abnormal structure of the horizontal rotary knife switch, thereby causing closing-in failure, causing heating and even explosion and other serious accidents. The application adopts a cloud-edge fusion architecture, uses a YOLO-V5 algorithm for target recognition, uses an HED method for edge extraction, and further proposes a method for detecting closing-in failure, so that accurate recognition of the horizontal rotary knife switch state is realized, manual inspection can be replaced, manpower can be saved, and safety can be improved. The application can also be used for monitoring other equipment of the transformer substation.

Description

TECHNICAL FIELD

[0001] The present application relates to the power knife switch state detection technology, in particular to the horizontal rotary knife switch state accurate detection, and more particularly to a horizontal rotary knife switch closing-in-place recognition device and a closing-in state judgment method. BACKGROUND

[0002] The knife switch, also known as high-voltage disconnector, is a main equipment of a substation. The horizontal rotary knife switch described in the present application refers to a single-phase horizontal rotary knife switch. A three-phase horizontal rotary knife switch has one horizontal rotary knife switch for each phase, and the three horizontal rotary knife switches are linked together. In the present application, each horizontal rotary knife switch has two horizontally rotatable arms, and the front ends of the arms have contacts.

[0003] In long-term operation, the horizontal rotary knife switch may not be closed in place due to various reasons. The existing closing-in-place discrimination mainly relies on the auxiliary contact or position sensor provided by the horizontal rotary knife switch. Since the horizontal rotary knife switch is basically placed outdoors for work, it has to face the problems of sun exposure, rain and corrosion damage, and wear and tear. Abnormal auxiliary contact or its transmission part will cause misjudgment of the on-off position, upload false signals, threaten the safety and life cycle of the equipment, and even cause major power accidents, resulting in significant personnel casualties and property losses.

[0004] The traditional method of closing-in-place fault identification is to send inspection personnel to the scene to observe with naked eyes. However, since the substations are mostly distributed in remote areas, and the number of horizontal rotary knife switches in each substation is also large, on-site observation is not only time-consuming but also labor-intensive. In addition, with the continuous development of society, the number of substations and horizontal rotary knife switches used is also increasing. In order to quickly detect the closing-in conditions of a large number of horizontal rotary knife switches at the same time, more human resources are needed compared to the past. This manual double detection method consumes a large amount of manpower and has poor real-time performance, which is a short board for the realization of full automation of substations, and there is a certain risk. The inspection personnel need to observe at close range, and when the closing-in is not in place, the high-voltage disconnector will heat up, which may produce discharge sparks or even explode.

[0005] It is extremely necessary to develop a double-checking method for automatically identifying the closing-out-of-position fault, and the State Grid and the Southern Power Grid have recently issued relevant notices. The current on-off state identification method is only a simple binary logic judgment, and the characteristics of closing-out-of-position are not obvious, and the identification of closing-out-of-position is much more difficult than the identification of on-off, which requires accurate identification devices and methods. The existing image acquisition or video monitoring method can only determine whether the switch is on or off, but cannot accurately determine whether the switch is in position. Other methods such as infrared method and attitude sensor are being studied. The infrared method needs to be powered for a period of time after closing, and the horizontal rotating knife switch can be judged by the abnormal heating. At this time, if the power is off, it will cause certain economic loss and affect the normal operation of the power grid, which is a kind of after-the-fact remedy method. The attitude sensor needs to install equipment on the horizontal rotating knife switch, and the power grid needs to be powered off for installation, and the maintenance also needs to be powered off, which affects the operation of the power grid. These methods have not yet achieved the expected results, and better double-checking methods need to be developed.

[0006] At present, there are various types and models of knife switches in substations in China, and their working principles and appearances are quite different. The horizontal rotating knife switch, hereinafter referred to as the horizontal rotating knife switch, is one of the most widely used knife switches in substations at present. The present application studies the closing-in-position state identification of this kind of horizontal rotating knife switch. SUMMARY

[0007] In order to overcome the defects in the prior art, the present application provides a horizontal rotating knife switch closing-in-position identification device and a closing state judgment method to solve the above problems.

[0008] The technical scheme provided by the present application is:

[0009] A horizontal rotating knife switch closing-in-position identification device, comprising:

[0010] A camera, a horizontal rotating knife switch identification front-end device, a horizontal rotating knife switch identification back-end learning and training device, and a data and control bus.

[0011] The camera is used to shoot images of the horizontal rotating knife switch, and the images are output to the horizontal rotating knife switch identification front-end device through the data and control bus. Three cameras are used to shoot a three-phase horizontal rotating knife switch, and each camera shoots a single-phase horizontal rotating knife switch.

[0012] The horizontal rotating knife switch recognition front-end device sends a control signal to control the shooting action of the camera, and can also control the gimbal of the camera to pitch and rotate to adjust the shooting direction. The horizontal rotating knife switch recognition front-end device recognizes the image of the horizontal rotating knife switch and detects the bending degree of the two rotating arms to determine whether the closing is in place, and sends a signal to the horizontal rotating knife switch recognition back-end learning and training device whether the closing is in place. At the same time, the image of the horizontal rotating knife switch is forwarded to the horizontal rotating knife switch recognition back-end learning and training device. The recognition parameters of the horizontal rotating knife switch recognition front-end device are provided and refreshed by the horizontal rotating knife switch recognition back-end learning and training device.

[0013] The horizontal rotating knife switch recognition back-end learning and training device performs deep learning training according to the received image of the horizontal rotating knife switch, and sends the trained parameters to the horizontal rotating knife switch recognition front-end device. At the same time, the signal whether the closing is in place is forwarded to the dispatching center. Multiple horizontal rotating knife switch recognition front-end devices can share one horizontal rotating knife switch recognition back-end learning and training device.

[0014] The data and control bus is used to transmit data and control signals between the camera, the horizontal rotating knife switch recognition front-end device and the horizontal rotating knife switch recognition back-end learning and training device.

[0015] The horizontal rotating knife switch refers to a single-phase horizontal rotating knife switch. A three-phase horizontal rotating knife switch includes three single-phase horizontal rotating knife switches connected in series.

[0016] Preferably, the camera is arranged as follows: the bases of the left, middle and right three cameras are installed in a straight line in parallel, the left and right cameras are fixed-angle cameras, and the middle camera is a camera with a gimbal, which are respectively used for shooting each phase of a three-phase horizontal rotating knife switch. The center of each camera is respectively aligned with the contact joint of the two rotating arms of a horizontal rotating knife switch. In addition to shooting the middle knife switch, the middle camera can also be used to shoot other power equipment.

[0017] A closing state judgment method using the horizontal rotating knife switch closing in place recognition device, characterized in that it comprises the following steps:

[0018] Horizontal rotating knife switch image acquisition and preprocessing: obtaining an image containing a horizontal rotating knife switch from a camera and performing image preprocessing;

[0019] Horizontal rotating knife switch target recognition: finding the required horizontal rotating knife switch in an image containing background information, filtering out the background, extracting the horizontal rotating knife switch from the photographed picture, and selecting the required horizontal rotating knife switch to be detected with a smallest rectangular frame;

[0020] Edge extraction: the profile lines of the two horizontal rotating arms of the horizontal rotating knife switch are extracted within the range selected by the rectangular frame, so as to accurately detect the closing state by using a geometric method;

[0021] Closing bending detection: the error of the profile lines of the two horizontal rotating arms of the horizontal rotating knife switch and the linear 180 degrees is calculated, if the error exceeds a given threshold, a closing misalignment signal is sent, otherwise a closing success signal is sent, if any horizontal rotating knife switch of the three-phase horizontal rotating knife switch is not closed in place, the entire three-phase horizontal rotating knife switch is regarded as closing misalignment.

[0022] Preferably, the horizontal rotating knife switch image acquisition and preprocessing comprises:

[0023] Data acquisition: a shooting instruction is sent, and the horizontal rotating knife switch image shot by the camera is read;

[0024] Data cropping: different cameras have different resolutions, in order to adapt to different cameras and reduce unnecessary calculation, after obtaining the photos from the camera, the photos are uniformly cropped to fixed resolution pixels;

[0025] Image noise removal: a filtering algorithm is used to filter the cropped image, the value of each pixel point is obtained by weighted average of itself and other pixel values in the neighborhood, each pixel in the image is scanned by a Gaussian kernel, and the weighted average gray value of the pixels in the field determined by the Gaussian kernel is calculated, and the value of the convolution center pixel is replaced by the weighted average gray value, the Gaussian kernel is an odd-sized Gaussian template.

[0026] Preferably, the horizontal rotating knife switch target recognition comprises: using YOLO-V5 algorithm to recognize the target.

[0027] Preferably, the edge extraction comprises the following steps:

[0028] Constructing an HED network model;

[0029] Using a multi-scale deep learning algorithm on the HED network model.

[0030] Preferably, the closing bending detection comprises:

[0031] Edge cropping: the image after edge extraction is further cropped to an image with the edge profile line as the center, saving calculation time and speeding up system response;

[0032] Key pixel extraction: the pixels of the upper half profile line of the left and right halves of the horizontal rotating knife switch are called key pixels, according to the positive and negative of the inclination angle, the corresponding positive and negative diagonal lines are used to further filter out irrelevant areas, and finally the key pixels are obtained;

[0033] Pixel normalization: traverse the remaining area in the picture after key pixel extraction, assign 1 to the pixel weight greater than the threshold value, and assign 0 to the pixel weight less than the threshold value;

[0034] Bending degree calculation: take the longest two line segments AB and CD in the straight line part of the two arms, A, B, C and D are the end points of the straight line part of the contour line, then the bending degree = (AB+BC+CD) / AD, wherein AB, BC, CD and AD are the pixel distances between two points, if the bending degree is less than the set threshold value, the closing is not in place, otherwise it is in place.

[0035] Preferably, the pan-tilt control method of the horizontal rotary knife switch closing-in-place recognition device adopts a fuzzy control algorithm, and comprises the following steps:

[0036] Recursive adjustment rule formula method fuzzy control;

[0037] Local continuous formula method;

[0038] The locally continuous formula method is optimized by using a spherical gap migration global optimization method.

[0039] The horizontal rotary knife switch closing-in-place recognition device and the closing state judgment method provided by the application have the following advantages and effects compared with the prior art:

[0040] 1. The closing-in-place of the horizontal rotary knife switch can be accurately judged;

[0041] 2. Good safety, non-contact optical video detection;

[0042] 3. No need to stop power installation and maintenance, and no influence on the normal operation of the power system.

[0043] 4. Multi-purpose, in addition to monitoring the state of the knife switch, other power equipment can also be monitored. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a module composition structure and principle schematic diagram of the horizontal rotary knife switch closing-in-place recognition device of the embodiment of the application;

[0045] Figure 2 is a flowchart of the closing state judgment method of the embodiment of the application;

[0046] Figure 3 is a schematic diagram of the target recognition processing of the horizontal rotary knife switch of the embodiment of the application;

[0047] Figure 4 is a schematic diagram of the edge extraction processing of the horizontal rotary knife switch of the embodiment of the application;

[0048] Figure 5is a schematic diagram of the bending degree calculation of the embodiment of the present application. DETAILED DESCRIPTION

[0049] The present application will be further described in conjunction with the embodiments and drawings, but the embodiments of the present application are not limited thereto.

[0050] Embodiments

[0051] Reference Figure 1 The horizontal rotary knife switch closing-in-place recognition device provided by the embodiment includes a camera, a horizontal rotary knife switch recognition front-end device, a horizontal rotary knife switch recognition background learning and training device, a data and control bus, and a router.

[0052] Suppose that a certain transformer substation has n horizontal rotary knife switches, Figure 1 The left, middle, and right three cameras for shooting the horizontal rotary knife switch 1 and the left, middle, and right three cameras for shooting the horizontal rotary knife switch n are shown in the middle, and the middle camera has a gimbal that can be tilted and rotated for shooting. For the sake of simplicity, Figure 1 A dotted line is used to represent other horizontal rotary knife switches. The bases of the left, middle, and right three cameras are installed in a straight line, and each camera shoots each phase of a three-phase horizontal rotary knife switch. The center of each camera is aligned with the contact joint of the two rotating arms of a horizontal rotary knife switch of each phase. The image of a horizontal rotary knife switch of a certain phase of a three-phase horizontal rotary knife switch is shown in Figure 3 The green box has two rotating arms, and the center is the closed contact. A three-phase horizontal rotary knife switch has three such horizontal rotary knife switches.

[0053] Figure 1 In the middle, a three-phase horizontal rotary knife switch is equipped with a horizontal rotary knife switch recognition front-end device. When the horizontal rotary knife switch recognition front-end device sends a shooting signal, the horizontal rotary knife switch images shot by the left, middle, and right three cameras are output to the horizontal rotary knife switch recognition front-end device through the data and control bus. The horizontal rotary knife switch recognition front-end device can also control the gimbal of the middle camera to tilt and rotate to adjust the shooting direction.

[0054] The horizontal rotary knife switch recognition front-end device uses a deep learning algorithm to recognize the horizontal rotary knife switch image, analyzes the closing bending degree value of the two rotating arm contour lines, judges whether the closing is in place, and sends the signal of whether the closing is in place to the horizontal rotary knife switch recognition background learning and training device. At the same time, the image of the horizontal rotary knife switch is also forwarded to the horizontal rotary knife switch recognition background learning and training device. The deep learning network parameters of the horizontal rotary knife switch recognition front-end device are provided and refreshed by the horizontal rotary knife switch recognition background learning and training device. Figure 1There are n three-phase horizontal rotating knife switches, but only one horizontal rotating knife switch identification background learning training device is needed, because the deep learning network parameters of these horizontal rotating knife switch identification front-end devices are the same, and further, multiple substations can share one horizontal rotating knife switch identification background learning training device, so Figure 1 The horizontal rotating knife switch identification front-end device is connected to the data and control bus through a router, which is a kind of advanced cloud edge fusion architecture at present.

[0055] The horizontal rotating knife switch identification background learning training device performs deep learning training according to the received image of the horizontal rotating knife switch, and sends the learned and trained parameters to the horizontal rotating knife switch identification front-end device, and simultaneously forwards the signal of whether the closing is in place to the dispatching center.

[0056] Figure 1 The data and control bus in the above can adopt general wired Ethernet or wireless wifi, the cameras are all network cameras, the horizontal rotating knife switch identification front-end device and the horizontal rotating knife switch identification background learning training device have network interfaces, and data and control signals can be transmitted between any devices through the network. The dispatching center can also directly obtain the image of the horizontal rotating knife switch to monitor the horizontal rotating knife switch without passing through the horizontal rotating knife switch identification front-end device.

[0057] Referring to Figure 2 The flowchart, the horizontal rotating knife switch closing position recognition device provided by the embodiment of the application is used to judge the closing state, and specifically includes the following steps:

[0058] S1, the horizontal rotating knife switch identification front-end device sends a shooting instruction to the left, middle and right three cameras, the middle camera returns to the position, and the three cameras shoot a certain three-phase horizontal rotating knife switch after closing together, and the three cameras send the images of the three horizontal rotating knife switches of the three-phase horizontal rotating knife switch to the horizontal rotating knife switch identification front-end device through the data and control bus;

[0059] S2, the horizontal rotating knife switch identification front-end device calls an image preprocessing subprogram module to perform data cropping and noise removal on the image;

[0060] S3, the horizontal rotating knife switch identification front-end device calls a horizontal rotating knife switch target recognition subprogram module to find the required horizontal rotating knife switch in an image containing background information, filter out the background, extract the horizontal rotating knife switch from the photographed picture, and select the horizontal rotating knife switch to be detected with a smallest rectangular frame, as shown in Figure 3

[0061] ​S4. The horizontal rotary knife switch identification front-end device calls the edge extraction subroutine module to extract the contour lines of the two horizontal rotating arms of the horizontal rotary knife switch within the range selected by the rectangular frame. The present invention adopts an HED network model and a multi-scale deep learning algorithm on the HED network model;

[0062] S5. The horizontal rotary knife switch identification front-end device calls the closing curvature detection subroutine module to calculate the contour line of the two horizontal rotating arms of the horizontal rotary knife switch and the linear 180-degree error. If the error exceeds a given threshold, a closing failure signal is issued, otherwise a closing success signal is issued. If any one of the three-phase horizontal rotary knife switches is not closed in place, the entire three-phase horizontal rotary knife switch is classified as not closed in place.

[0063] S6. The horizontal rotary knife switch identification background learning and training device collects all the closing judgment signals of the horizontal rotary knife switch identification front-end device, presents them through the interface, and forwards them to the dispatching center.

[0064] exist Figure 2 In the process, the image preprocessing subroutine module specifically includes the following steps:

[0065] Data cropping: Different cameras have different resolutions. To adapt to different cameras and reduce unnecessary calculations, after obtaining photos from the camera, they are uniformly cropped to images with fixed resolution pixels;

[0066] Image noise removal: Use a well-known filtering algorithm to filter the cropped image and perform weighted averaging on the entire image. The value of each pixel is obtained by weighted averaging the value of itself and other pixels in the neighborhood. Use a Gaussian kernel to scan each pixel in the image, and then calculate the weighted average grayscale value of the pixels in the area determined by the Gaussian kernel, and use it to replace the value of the convolution center pixel. The Gaussian kernel is an odd-sized Gaussian template. Here, a 5*5 Gaussian template is used.

[0067] above Figure 2 In the process, the horizontal rotary knife switch target recognition subroutine module adopts the YOLO-V5 algorithm. Figure 3 The green box is a schematic diagram. For more details, see the YOLO-V5 algorithm.

[0068] Figure 4 Obtained by the edge extraction subroutine module, see HED algorithm.

[0069] above Figure 2 In the process, the closing curvature detection subroutine module also includes the following steps:

[0070] Edge clipping: the image after edge extraction is further clipped as the image centered on the edge contour line, saving calculation time and speeding up system response;

[0071] Key pixel extraction: the pixels of the upper half of the profile line of the left and right halves of the horizontally rotating knife shutter are called key pixels, and according to the positive and negative of the inclination angle, the corresponding positive and negative diagonal lines are used to further filter out irrelevant areas, and finally the key pixels are obtained;

[0072] Pixel normalization: traverse the remaining area in the picture after key pixel extraction, and assign 1 to the pixel points with a weight value greater than the threshold, and set 0 to the points less than the threshold;

[0073] Bending degree calculation: Figure 5 In the straight line part of the two arms, the longest two line segments AB and CD are taken, A, B, C, and D are the end points of the straight line part of the profile line, and the bending degree = (AB+BC+CD) / AD, where AB, BC, CD, and AD are the pixel distances between two points. If the bending degree is less than the set threshold, it is not in place, otherwise it is in place. The final result is shown in Figure 5 When the closing is in place, the two rotating arms form a straight line, and at this time there is the equation AD = AB + BC + CD, and the bending degree = (AB+BC+CD) / AD = 1, which is the ideal case. In practice, sometimes the rotating arm stops before it is in place, i.e. under-closing, at which time AD<(AB+BC+CD), and the bending degree is greater than 1; sometimes the rotating arm still rotates before it is in place, i.e. over-closing, at which time AD<(AB+BC+CD), and the bending degree is greater than 1. Regardless of the situation, the bending degree criterion is applicable. Of course, the bending degree = 1 is the ideal case, and in practice there is always an error, so a threshold can be set, and when the threshold is exceeded, a signal is sent that it is not in place.

[0074] The gimbal control method of the horizontal rotating knife shutter closing in place recognition device described above, specifically uses a fuzzy control algorithm, including the following steps:

[0075] Recursive adjustment rule formula method fuzzy control;

[0076] Local continuous formula method is used for local continuous formula method;

[0077] The locally continuous formula method is optimized by using the spherical gap migration global optimization method.

[0078] Figure 1The horizontal rotary knife switch identification background learning training device in the utility model is provided with a depth learning server as hardware, and the depth learning software related to the horizontal rotary knife switch in the utility model is installed, including horizontal rotary knife switch target identification and edge extraction.

[0079] Another function of the horizontal rotary knife switch identification background learning training device is to collect the signals of whether the closing is in place from the horizontal rotary knife switch identification front-end devices, present through an interface, and forward to the dispatching center.

[0080] Figure 1 In the utility model, the two cameras can adopt a gimbal integrated spherical camera, and the advantages of the scheme are as follows:

[0081] 1) Because the gun type camera does not have a gimbal and a control mechanism, its price is much lower than that of the spherical camera when the optical magnification is the same;

[0082] 2) The gun type camera has fixed position and angle, and has no movement and rotation, so its identification accuracy is high and the error is small;

[0083] 3) The spherical camera is expensive, and because it rotates and pitches, its positioning accuracy is relatively low, but it can monitor the surrounding area at 360 degrees, and can shoot all other equipment in the substation except the knife switch, so by installing a spherical camera at each intermediate position, the purpose of monitoring other equipment in the substation can be achieved with fewer spherical cameras, thus partially replacing manual inspection and inspection robots, and when the left and right cameras are blocked by birds or other faults occur, the intermediate spherical camera can also temporarily replace them.

[0084] This scheme realizes multiple functions and saves costs. The key is the control algorithm of the spherical camera, which is a combination of a camera and a gimbal. It can rotate horizontally and pitch up and down, so it can shoot a wide range of surrounding scenes, and its field of view is much larger than that of the fixed-angle gun type camera. However, it also has two disadvantages:

[0085] Its rotation angle is digitally adjusted in steps, and cannot be infinitely positioned, so it is difficult to accurately position to a certain angle; its gimbal is mechanically driven by a motor gear, and has a gap, usually with a certain virtual position error, so after shooting the surrounding equipment, it may not be able to accurately return to the position of shooting the knife switch.

[0086] And the closing state recognition of the knife gap must reach the positioning level of high precision, otherwise it is difficult to accurately identify and judge whether the closing is in place, for this, the local continuous surface high precision fuzzy control algorithm is adopted to realize, see reference 1.

[0087] To realize the precise and rapid control of the spherical camera, according to the current mainstream automatic control theory, the differential equation of the spherical camera needs to be written, and the related mathematical model needs to be established, obviously, this needs a long time and experiment, which is difficult and too expensive, not suitable for this project. The local continuous surface high precision fuzzy control algorithm does not need to establish the model of the controlled object, it controls the local continuity of the control surface by table lookup method and formula method, while maintaining the fast speed characteristics, the precision even exceeds the rule-based method, and the algorithm is not complex.

[0088] The local continuous surface high precision fuzzy control algorithm has many parameters, which greatly affects the control precision and the speed of return, the ball gap migration algorithm is adopted to optimize the control parameters, see reference 2 for details.

[0089] Reference:

[0090] 1) Hu Jinsong, Zheng Qilun, Wu Jie, Time-varying correction factor two-level self-optimizing fuzzy controller, Automatic Control Report, 2002, 06, pp.1006-1011.

[0091] 2) Hu Jinsong, Zheng Qilun, Ball gap migration algorithm realizes global optimization, Computer Journal, Vol. 35, No. 2, February 2012, pp.193-201.

[0092] The above embodiments of the present application focus on adopting the cloud edge fusion architecture, using the YOLO-V5 algorithm for target recognition, then using the HED method for edge extraction, and on this basis, a method for detecting the closing position is proposed, so as to realize the precise identification of the horizontal rotating knife gap state, which can replace manual inspection, save manpower and improve safety; The present application can also be used for monitoring other equipment of the substation.

[0093] The above only describes the exemplary embodiments of the present application, and does not limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for judging the closing state of a horizontal rotary knife switch closing position identification device, characterized in that: The steps include: Horizontal rotary knife switch image acquisition and preprocessing: The image containing the horizontal rotary knife switch is acquired from the camera and image preprocessing is performed. The arrangement of the camera is as follows: The bases of the three cameras on the left, center, and right are installed in a straight line. The left and right cameras have fixed shooting angles, and the center camera has a pan / tilt head. They are used to shoot each phase of a three-phase horizontal rotary knife switch. The center of each camera is aimed at the contact junction of the two rotating arms of a horizontal rotary knife switch. In addition to photographing the middle switch, the middle camera can also be used to photograph other power equipment; Horizontal rotary knife switch target recognition: Find the required horizontal rotary knife switch in an image containing background information, filter out the background, extract the horizontal rotary knife switch from the captured image, and select the horizontal rotary knife switch to be tested using a minimum rectangular frame; Edge extraction: A HED network model is constructed and a multi-scale deep learning algorithm is used on the HED network model to extract the contours of the two horizontal rotating arms of the horizontal rotary knife switch within the range selected by the rectangular box, so as to accurately detect the closing state using a geometric method; Closing curvature detection: Calculate the error between the contour line and the linear 180-degree error of the two horizontal rotating arms of the horizontal rotary knife switch. If the error exceeds a given threshold, a closing failure signal is issued; otherwise, a closing success signal is issued. If any of the three-phase horizontal rotary knife switches fails to close properly, the entire three-phase horizontal rotary knife switch is classified as failing to close properly. The detection includes the following steps: Edge cropping: The edge-extracted image is further cropped into an image centered on the edge contour line, saving computing time and speeding up system response; Key pixel extraction: The pixels of the upper half of the left and right halves of the horizontal rotary knife switch are called key pixels. Irrelevant areas are filtered out to finally obtain the key pixels. Pixel normalization: traverse the remaining area of ​​the image after key pixel extraction, assign a value of 1 to pixels with a weight greater than the threshold, and set the weight of pixels with a weight less than the threshold to 0; Curvature calculation: Take the two longest line segments AB and CD in the straight part of the two arms. A, B, C, and D are the endpoints of the straight part of the contour line. The curvature = (AB + BC + CD) / AD, where AB, BC, CD, and AD are the pixel distances between the two points. If the curvature is less than the set threshold, the circuit breaker is not closed in place; otherwise, it is closed in place. The pan / tilt control method of the horizontal rotary knife switch closing position recognition device is a fuzzy control algorithm, comprising the following steps: Formula method fuzzy control with recursive adjustment rules; Local continuation of the formula method; The spherical gap migration global optimization method is used to optimize the local continuation formula method.

2. The method for judging the closing state according to claim 1, characterized in that: The horizontal rotary knife switch image acquisition and preprocessing includes the following steps: Data acquisition: Send out shooting instructions to read the horizontal rotary knife switch image taken by the camera; Data cropping: Different cameras have different resolutions. To adapt to different cameras and reduce unnecessary calculations, after obtaining photos from the camera, they are uniformly cropped to images with fixed resolution pixels; Image noise removal: A filtering algorithm is used to filter the cropped image and perform a weighted average on the entire image. The value of each pixel is obtained by weighted averaging the value of itself and other pixels in its neighborhood. A Gaussian kernel is used to scan each pixel in the image, and then the weighted average grayscale value of the pixels in the area determined by the Gaussian kernel is calculated and used to replace the value of the convolution center pixel. The Gaussian kernel is an odd-sized Gaussian template.

3. The method for judging the closing state according to claim 2, characterized in that: The horizontal rotary knife switch target recognition adopts the YOLO-V5 algorithm.

4. A horizontal rotary knife switch closing position recognition device, characterized in that: A method for determining a closing state according to any one of claims 1 to 3, comprising: Camera, horizontal rotary knife switch recognition front-end device, horizontal rotary knife switch recognition back-end learning and training device, data and control bus; The camera is used to capture images of the horizontal rotary knife switch. The images are output to the horizontal rotary knife switch identification front-end device via the data and control bus. Three cameras are used to capture images of a three-phase horizontal rotary knife switch, with each camera capturing images of a single-phase horizontal rotary knife switch. The horizontal rotary knife switch recognition front-end device sends a control signal to control the shooting action of the camera, and can also control the pan / tilt and rotation of the camera to adjust the shooting direction. The horizontal rotary knife switch recognition front-end device recognizes the image of the horizontal rotary knife switch and detects the closing curvature of the two rotating arms to determine whether the closing is in place, and sends a signal indicating whether the closing is in place to the horizontal rotary knife switch recognition background learning and training device. At the same time, the image of the horizontal rotary knife switch is also forwarded to the horizontal rotary knife switch recognition background learning and training device. The recognition parameters of the horizontal rotary knife switch recognition front-end device are provided and refreshed by the horizontal rotary knife switch recognition background learning and training device. The horizontal rotary knife switch recognition background learning and training device performs deep learning and training based on the received horizontal rotary knife switch image, sends the learned and trained parameters to the horizontal rotary knife switch recognition front-end device, and at the same time forwards the signal indicating whether the switch is closed in place to the dispatching center. Multiple horizontal rotary knife switch recognition front-end devices can share one horizontal rotary knife switch recognition background learning and training device; The data and control bus is used to transmit data and control signals between the camera, the horizontal rotary knife switch identification front-end device and the horizontal rotary knife switch identification back-end learning and training device; The horizontal rotary knife switches mentioned above all refer to single-phase horizontal rotary knife switches. A three-phase horizontal rotary knife switch includes three linked single-phase horizontal rotary knife switches.

Images