A device and method for accurately identifying the state of a high-voltage disconnector

Abstract

The present invention discloses a device and method for accurately identifying the state of a high-voltage disconnector, which is used to identify whether the high-voltage disconnector in a substation is accurately closed in place. The high-voltage disconnector is an extremely important device in the substation. Long-term use can cause abnormalities in the structure of the high-voltage disconnector, resulting in incomplete closing, leading to serious accidents such as overheating and even explosion. The present invention adopts an integrated collector of a camera and a lidar. First, the camera is used to preliminarily identify and target-locate the high-voltage disconnector, and then the lidar is used for accurate identification. The two-dimensional image captured by the camera is easy to identify, which can help the lidar narrow the scanning range and improve the response time, and the advantages of the two complement each other.

Description

Technical Field

[0001] The present invention relates to the technical field of switch detection, and in particular to a device and method for accurately identifying the state of a high-voltage isolating switch. Background Art

[0002] High-voltage disconnectors, also known as knife switches in literature, are a major device in substations. Since high-voltage disconnectors are basically placed outdoors for operation, they have to face the problems of sun and rain, metal corrosion damage, and wear. In long-term operation, various reasons may cause the high-voltage disconnectors to fail to close properly. The existing judgment of closing properly mainly relies on the auxiliary contacts of the high-voltage disconnectors. Abnormalities in the auxiliary contacts or their transmission parts will lead to misjudgment of the opening and closing positions, upload wrong signals, and cause major power accidents.

[0003] The traditional detection method is to send inspection personnel to the site to observe with the naked eye, but since most substations are distributed in remote areas and each substation has a large number of high-voltage disconnectors, on-site observation is not only time-consuming and laborious. This manual double detection method consumes a lot of manpower, has poor real-time performance, and is somewhat dangerous.

[0004] It is extremely necessary to develop a double verification method for automatically identifying incomplete closing faults. At present, there are methods such as infrared method and attitude sensor. The infrared method requires power to be turned on for a period of time after closing the switch to judge the abnormal heating of the high-voltage disconnector. If the power is cut off at this time, it will cause certain economic losses and affect the normal operation of the power grid. It is a post-event remedial method. The attitude sensor needs to be installed on the high-voltage disconnector separately. The power grid must be cut off for installation. Its maintenance also requires the power grid to be cut off, which affects the operation of the power grid. These methods have not yet achieved the expected results, and there is an urgent need to develop a better double verification method.

[0005] Capturing images with a camera and identifying whether the switch is in place can be achieved without contact and without power outage. In theory, this is a more ideal method. However, the images captured by the camera are two-dimensional and lack depth information. There are high requirements for the shooting angle, and there are many restrictions on the installation location of the substation. Therefore, in practice, the recognition accuracy of the image method is not high. It is usually used to determine whether the switch is closed or not. However, the change of whether the switch is in place or not is not obvious, making it difficult to accurately identify whether the switch is in place. Summary of the invention

[0006] At present, there are a wide variety of models and types of high-voltage disconnectors in substations in China. Their working principles and appearances vary greatly, and the relevant terms and definitions are not unified. Therefore, the present invention defines that the high-voltage disconnector refers to a single-phase high-voltage disconnector. Each phase of a three-phase high-voltage disconnector has a high-voltage disconnector, and the three high-voltage disconnectors are interlocked. In the present invention, each high-voltage disconnector has a conductive arm, which is also called a conductive rod, a conductive plate, or a knife switch. It has contacts and wires and can rotate (in the present invention, only the movable conductive arm is referred to). During the closing or opening process of the high-voltage disconnector, the angle of the conductive arm changes.

[0007] The object of the present invention is to overcome the shortcomings and deficiencies of the prior art and propose a safe and reliable device and method for accurately identifying the state of a high-voltage disconnector.

[0008] To achieve the above object, the technical solution provided by the present invention is: a device for accurately identifying the state of a high-voltage disconnector, including: a camera, a lidar, and a front-end device for identifying the high-voltage disconnector;

[0009] The camera is used to capture images of the high-voltage disconnector, and the images are output to the front-end device for identifying the high-voltage disconnector. The shooting action of the camera is controlled by a control signal sent by the front-end device for identifying the high-voltage disconnector;

[0010] The lidar collects point cloud data of the high-voltage disconnector and outputs it to the front-end device for identifying the high-voltage disconnector. The start of the lidar is controlled by the front-end device for identifying the high-voltage disconnector;

[0011] The front-end device for identifying the high-voltage disconnector comprehensively identifies the images and point cloud data of the high-voltage disconnector to determine whether the closing is in place.

[0012] Preferably, the camera and the lidar are fixed on a base, and the central axes of the two are parallel. The central axis of the camera is the optical axis of its lens, and the central axis of the lidar is a straight line passing through the center point of its sensor plane and perpendicular to this plane. A device integrating the camera, the lidar, and the base is called an integrated collector.

[0013] Preferably, the device for accurately identifying the state of the high-voltage disconnector further includes: a back-end learning and training device for identifying the high-voltage disconnector, a data and control bus;

[0014] The front-end device for identifying the high-voltage disconnector forwards the images and point cloud data of the high-voltage disconnector to the back-end learning and training device for identifying the high-voltage disconnector. The identification parameters of the front-end device for identifying the high-voltage disconnector are provided and refreshed by the back-end learning and training device for identifying the high-voltage disconnector;

[0015] The high-voltage disconnector identification background learning and training device performs deep learning training based on the received images and point cloud data of the high-voltage disconnector, and sends the learned and trained parameters to the high-voltage disconnector identification front-end device. Multiple high-voltage disconnector identification front-end devices can share one high-voltage disconnector identification background learning and training device;

[0016] The data and control bus are used to transfer data and control signals between the camera, lidar, high-voltage disconnector identification front-end device, and high-voltage disconnector identification background learning and training device.

[0017] Preferably, the integrated collector is arranged as follows:

[0018] Three integrated collectors are installed in a straight line side by side, respectively aligned with each phase of a three-phase high-voltage disconnector, that is, one integrated collector corresponds to one phase of the high-voltage disconnector. The center line of each integrated collector is respectively aligned with the center of the conductive arm of one high-voltage disconnector in each phase. The center line of the integrated collector is parallel to the central axis of the camera and the central axis of the lidar and is located in the middle of the two central axes. The high-voltage disconnector refers to a single-phase high-voltage disconnector. A three-phase high-voltage disconnector includes three interlocked single-phase high-voltage disconnectors. The conductive arm is also called a conductive rod, conductive plate, or knife switch, and has contacts and wires on it.

[0019] Preferably, the high-voltage disconnector state accurate identification device further includes:

[0020] An image target recognition module for processing the entire image captured by the camera, identifying the conductive arm of the high-voltage disconnector from it, and positioning the conductive arm of the high-voltage disconnector with a frame or contour line;

[0021] An image contour extraction module for extracting the contour line of the conductive arm within the range located by the frame, and skipping if the contour line has already been obtained;

[0022] A closing and opening action judgment module for judging the current state of the high-voltage disconnector, calculating the angle of the conductive arm and its change trend over time from the contour line, and inferring which of the following states the high-voltage disconnector is currently in: closing, opening, closed, or open;

[0023] A point cloud discrimination module for discriminating whether the closing is in place. If the current state transitions from closing to closed, the lidar is started to scan to obtain point cloud data, and whether the closing is in place is directly discriminated by the deep learning algorithm.

[0024] Preferably, the point cloud discrimination module further includes:

[0025] The point cloud preprocessing module uses the positioning box generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise in the point cloud data, so as to reduce the calculation amount and improve the recognition accuracy.

[0026] The present invention also provides a closing state judgment method for the above-mentioned high-voltage disconnector state precise recognition device, including:

[0027] Image target recognition: Find the conductive arm of the required high-voltage disconnector in an image containing background information, filter out the background, extract the conductive arm from the captured picture, and frame it with a smallest rectangular box;

[0028] Image contour extraction: Extract the contour line of the conductive arm within the range framed by the rectangular box;

[0029] Closing and opening operation judgment: Calculate the angle of the conductive arm and its change trend over time from the contour line, and infer which of the following states the high-voltage disconnector is currently in: closing, opening, closed, or opened;

[0030] Lidar startup: If the current state transitions from closing to closed, a signal is sent by the front-end device for high-voltage disconnector recognition to start lidar scanning to obtain point cloud data;

[0031] Point cloud preprocessing: Use the positioning box generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise in the point cloud data;

[0032] Point cloud discrimination: Directly discriminate whether the closing is in place through the preprocessed point cloud data by a deep learning algorithm.

[0033] Preferably, the method for removing the outlier point cloud noise is as follows:

[0034] Let: The given point cloud set P = {p i |p i ∈R 3 , i = 1, 2, 3,..., N}, p i represents the i-th point, the total number of points is N, the outlier point cloud set Deletion = {d i |d i ∈R 3}, d i represents the i-th point in the outlier point cloud set, the total number of which is to be determined, R 3 represents a bounded three-dimensional space domain, the number of points in the neighborhood range is K, and the variance coefficient is β;

[0035] 1) Read the point cloud file C and convert it to generate the point cloud set P;

[0036] 2) Construct a kd - Tree to store the point cloud set P;

[0037] 3) Use the kd - Tree to calculate the K - nearest neighbor neighborhoods of each point in the point cloud set P;

[0038] 4) According to the formula Calculate the average distance d of point p i in its K - nearest neighbor neighborhood, where i , is the distance between point p j and p i , p j is the j - th point in the K - nearest neighbor neighborhood of p i , and the coordinates of p i , p j are (x i , y i , z i ) and (x j , y j , z j );

[0039] 5) Calculate the average K - nearest neighbor neighborhood distance μ and variance σ of the entire point cloud set P according to the formula ; 2 ;

[0040] 6) Traverse the average distance d of each point's K - nearest neighbor neighborhood. If i , then consider this point as an abnormal outlier and delete it from the point cloud set P and add it to the outlier point cloud set Deletion;

[0041] 7) Distance information neighborhood transfer: Traverse the elements in the point cloud set P and use the formula to calculate the score s of each point i , where the intermediate variable If , then delete it from the point cloud set P and add it to the outlier point cloud set Deletion;

[0042] 8) Output the remaining points in the point cloud set P as the normal point cloud set and output the final outlier point cloud set.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] 1. Using video images to assist lidar in recognition and positioning, with complementary advantages, speeds up the response time and improves the recognition accuracy.

[0045] 2. Judging the opening and closing actions from video images and starting the lidar as needed, which extends the service life of the lidar.

[0046] 3. Non-contact optical video detection, with good safety.

[0047] 4. It can be installed and maintained without power outage, without affecting the normal operation of the power system. Description of the Drawings

[0048] Figure 1 It is a schematic diagram of the principle of the device for accurately identifying the state of a high-voltage disconnector.

[0049] Figure 2 It is a schematic diagram after the target recognition and processing of the high-voltage disconnector.

[0050] Figure 3 It is a flowchart for judging the fully-closed state of the high-voltage disconnector. Detailed Implementation Modes

[0051] The present invention will be further described in detail below in conjunction with the embodiments and the drawings, but the implementation modes of the present invention are not limited thereto.

[0052] See Figure 1 As shown, this embodiment discloses a device for accurately identifying the state of a high-voltage disconnector. Its hardware part includes: an integrated collector (including a camera, a lidar, and a base), a front-end device for identifying the high-voltage disconnector, a back-end learning and training device for identifying the high-voltage disconnector, a data and control bus, and a router; the camera and the lidar are fixed on a base, and the central axes of the two are parallel. The central axis of the camera is the optical axis of its lens, and the central axis of the lidar is a straight line passing through the center point of its sensor plane and perpendicular to this plane. A device integrating a camera, a lidar, and a base is called an integrated collector.

[0053] Its software part includes: an image target recognition module, an image contour extraction module, a closing and opening action judgment module, and a point cloud discrimination module;

[0054] The image target recognition module is used to process the entire image captured by the camera, identify the conductive arm of the high-voltage disconnector therefrom, and locate the conductive arm of the high-voltage disconnector with a frame or a contour line;

[0055] The image contour extraction module is used to extract the contour line of the conductive arm within the range located by the frame, and skip it if the contour line has been obtained;

[0056] The closing and opening action judgment module is used to judge the current state of the high-voltage disconnector, calculate the angle of the conductive arm and its change trend over time from the contour line, and infer which of the following states the high-voltage disconnector is currently in: closing, opening, fully closed, or fully open;

[0057] The point cloud discrimination module is used to determine whether the closing is in place. If the current state transitions from closing to closed, the lidar is activated to scan and obtain point cloud data, and a deep learning algorithm is used to directly determine whether the closing is in place.

[0058] Preferably, the point cloud discrimination module further includes: a point cloud preprocessing module, which uses the positioning frame generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise in the point cloud data, so as to reduce the calculation amount and improve the recognition accuracy.

[0059] Suppose there are n high-voltage disconnectors in a certain substation. Figure 1 The left, middle, and right 3 integrated collectors for collecting the data of the high-voltage disconnector 1 (including image data and point cloud data) and the left, middle, and right 3 integrated collectors for collecting the data of the high-voltage disconnector n are shown in. For simplicity, Figure 1 a dashed line is used to indicate that there are other high-voltage disconnectors. The bases of the left, middle, and right 3 integrated collectors are installed side by side in a straight line, and are respectively aligned with each phase of a three-phase high-voltage disconnector. The centers of each camera and lidar are respectively aligned with the center of the conductive arm of a high-voltage disconnector of each phase. A three-phase high-voltage disconnector has three such high-voltage disconnectors. The image of a high-voltage disconnector of a certain phase of a three-phase high-voltage disconnector is as Figure 2 shown. The conductive arm is in the frame, which has two telescopic arms, and the center is the connecting shaft of the two arms.

[0060] Figure 1 In, a high-voltage disconnector recognition front-end device is equipped for a three-phase high-voltage disconnector. When the high-voltage disconnector recognition front-end device sends a shooting signal, the 3 high-voltage disconnector images taken by the left, middle, and right 3 cameras are output to the high-voltage disconnector recognition front-end device through the data and control bus. The high-voltage disconnector recognition front-end device uses a deep learning algorithm to recognize the high-voltage disconnector image, and then calculates the included angle of the telescopic arm contour line. Since the image is a kind of planar two-dimensional data, lacking depth information, its accuracy is limited by the installation position, and it is difficult to obtain an accurate included angle value. However, the change trend of the conductive arm angle can be judged by the image method, so as to infer which of the following states the high-voltage disconnector is currently in: closing, opening, closed, open.

[0061] If the closing operation has been completed and the device enters the closed state, the front-end device for identifying the high-voltage disconnector sends a scanning signal to the lidar, and the lidar starts scanning to obtain the point cloud data of the high-voltage disconnector. This measure is taken because the semiconductor laser diode used in the lidar is a power device with a limited lifespan. If it scans continuously for 24 hours a day, it will be scrapped within a few years or even one year, and the lidar is expensive. In addition, the high-voltage disconnector is usually closed only a few times a year. Therefore, in the present invention, the lidar scanning is only started after the closing operation of the high-voltage disconnector is completed, which not only saves costs but also improves the reliability of the lidar.

[0062] The front-end device for identifying the high-voltage disconnector first preprocesses the point cloud obtained by the lidar: using the positioning frame generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise; then, the PointNet algorithm based on deep learning is used to determine whether the closing is in place. If the discrimination result is that the closing is in place, it only means that the single-phase high-voltage disconnector is closed in place. However, the three-phase high-voltage disconnector is composed of three single-phase switches. If any one of the single-phase high-voltage disconnectors is not closed in place, the entire three-phase high-voltage disconnector is classified as not closed in place.

[0063] The front-end device for identifying the high-voltage disconnector sends the signal indicating whether the closing is in place to the back-end learning and training device for identifying the high-voltage disconnector, and at the same time forwards the image and point cloud data of the high-voltage disconnector to the back-end learning and training device for identifying the high-voltage disconnector. The deep learning network parameters of the front-end device for identifying the high-voltage disconnector are provided and refreshed by the back-end learning and training device for identifying the high-voltage disconnector; Figure 1 There are n three-phase high-voltage disconnectors, but only one back-end learning and training device for identifying the high-voltage disconnector is needed because the similar deep learning network parameters of these front-end devices for identifying the high-voltage disconnectors are the same. Further, multiple substations can share one back-end learning and training device for identifying the high-voltage disconnector. Therefore Figure 1 In the front-end device for identifying the high-voltage disconnector is connected to the data and control bus through a router, which is an advanced cloud-edge fusion architecture at present.

[0064] The back-end learning and training device for identifying the high-voltage disconnector performs deep learning training based on the received image and point cloud data of the high-voltage disconnector, sends the trained parameters to the front-end device for identifying the high-voltage disconnector, and at the same time forwards the signal indicating whether the closing is in place to the power dispatching center. At present, there are many types and models of high-voltage disconnectors in substations in China, and their working principles and appearances vary greatly. The same type of disconnector can share parameters, but different types of disconnectors require multiple deep learning networks and multiple sets of parameters.

[0065] Figure 1The data and control buses therein can use general wired Ethernet or wireless Wi-Fi. Cameras and lidars can all communicate directly through the network. The front-end device for identifying high-voltage disconnectors and the back-end learning and training device for identifying high-voltage disconnectors both have network interfaces, and data and control signals can be transmitted between any devices through the network. The power dispatching center can also directly obtain the images and point clouds of high-voltage disconnectors without passing through the front-end device for identifying high-voltage disconnectors to monitor high-voltage disconnectors.

[0066] Figure 3 It is a flowchart for judging the closing-in-place state of the high-voltage disconnector state precise identification device, including:

[0067] Image target recognition: Find the required conductive arm of the high-voltage disconnector in an image containing background information, filter out the background, extract the conductive arm from the captured picture, and frame it with a smallest rectangular box;

[0068] Image contour extraction: Extract the contour line of the conductive arm within the range framed by the rectangular box;

[0069] Closing and opening operation judgment: Calculate the angle of the conductive arm and its changing trend over time from the contour line, and infer which of the following states the high-voltage disconnector is currently in: closing, opening, closed, opened;

[0070] Lidar startup: If the current state transitions from closing to closed, then the front-end device for identifying high-voltage disconnectors sends a signal to start lidar scanning to obtain point cloud data;

[0071] Point cloud preprocessing: Use the positioning box generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise in the point cloud data;

[0072] Point cloud discrimination: Use a deep learning algorithm to directly discriminate whether it is closed in place through the preprocessed point cloud data.

[0073] The above Figure 3 In the above process, the details of removing the outlier point cloud noise in the point cloud data are as follows:

[0074] Suppose: The given point cloud set P = {p i |p i ∈R 3 , i = 1, 2, 3, …, N}, p i represents the i-th point, the total number of points is N, the outlier point cloud set Deletion = {d i |d i ∈R 3}, d i represents the i-th point in the outlier point cloud set, the total number of which is to be determined, R3 represents a bounded three-dimensional spatial domain with the number of neighborhood range points being K and the coefficient of variance being β; a kd-tree (short for k-dimensional tree) is a tree-shaped data structure for storing instance points in a K-dimensional space for fast retrieval.

[0075] 1) Read the point cloud file C and convert it to generate a point cloud set P.

[0076] 2) Construct a kd-Tree to store the point cloud set P.

[0077] 3) Use the kd-Tree to calculate the K-nearest neighbor neighborhoods of each point in the point cloud set P.

[0078] 4) According to the formula calculate the average distance d of point p i in its K-nearest neighbor neighborhood, where i , is the distance between point p j and p i , p j is the j-th point in the K-nearest neighbor neighborhood of p i , and the coordinates of p i , p j are (x i , y i , z i ) and (x j , y j , z j ) respectively.

[0079] 5) Calculate the average K-nearest neighbor neighborhood distance μ and variance σ of the entire point cloud set P according to the formula ; 2 ;

[0080] 6) Traverse the average distance d of the K-nearest neighbor neighborhoods of each point. i If then consider this point as an abnormal outlier and delete it from the point cloud set P and add it to the outlier point cloud set Deletion.

[0081] 7) Distance information neighborhood transfer: Traverse the elements in the point cloud set P and use the formula to calculate the score s of each point i , where the intermediate variable If then delete it from the point cloud set P and add it to the outlier point cloud set Deletion.

[0082] 8) Output the remaining points in the point cloud set P as the normal point cloud set and output the final outlier point cloud set.

[0083] The above embodiments are preferred embodiments of the present invention. However, the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. An accurate recognition device for the state of a high-voltage disconnector, characterized in that, Including: a camera, a lidar, a front-end device for identifying high-voltage disconnectors, a back-end learning and training device for identifying high-voltage disconnectors, a data and control bus, an image target recognition module, an image contour extraction module, a closing and opening action judgment module, and a point cloud discrimination module; The camera is used to capture images of the high-voltage disconnector, and the images are output to the front-end device for identifying high-voltage disconnectors. The shooting action of the camera is controlled by a control signal sent by the front-end device for identifying high-voltage disconnectors; The lidar collects point cloud data of the high-voltage disconnector and outputs it to the front-end device for identifying high-voltage disconnectors. The startup of the lidar is controlled by the front-end device for identifying high-voltage disconnectors; The front-end device for identifying high-voltage disconnectors comprehensively identifies the images and point cloud data of the high-voltage disconnector to judge whether the closing is in place; The front-end device for identifying high-voltage disconnectors forwards the images and point cloud data of the high-voltage disconnector to the back-end learning and training device for identifying high-voltage disconnectors. The identification parameters of the front-end device for identifying high-voltage disconnectors are provided and refreshed by the back-end learning and training device for identifying high-voltage disconnectors; The back-end learning and training device for identifying high-voltage disconnectors performs deep learning training based on the received images and point cloud data of the high-voltage disconnector, and sends the trained parameters to the front-end device for identifying high-voltage disconnectors. Multiple front-end devices for identifying high-voltage disconnectors can share one back-end learning and training device for identifying high-voltage disconnectors; The data and control bus is used to transfer data and control signals between the camera, the lidar, the front-end device for identifying high-voltage disconnectors, and the back-end learning and training device for identifying high-voltage disconnectors; The image target recognition module is used to process the entire image captured by the camera, identify the conductive arms of the high-voltage disconnector from it, and locate the conductive arms of the high-voltage disconnector with a frame or contour line; The image contour extraction module is used to extract the contour line of the conductive arm within the range located by the frame, and skip it if the contour line has been obtained; The closing and opening action judgment module is used to judge the current state of the high-voltage disconnector, calculate the angle of the conductive arm and its change trend over time from the contour line, and infer which of the following states the high-voltage disconnector is currently in: closing, opening, closed, or open; The point cloud discrimination module is used to discriminate whether the closing is in place. If the current state transitions from closing to closed, the lidar is started to scan and obtain point cloud data, and whether the closing is in place is directly discriminated by a deep learning algorithm.

2. The precise identification device for the state of the high-voltage disconnector according to claim 1, characterized in that, The camera and the lidar are fixed on a base, and the central axes of the two are parallel. The central axis of the camera is the optical axis of its lens, and the central axis of the lidar is a straight line passing through the center point of its sensor plane and perpendicular to this plane. A device integrating the camera, the lidar, and the base is called an integrated collector.

3. The high-voltage disconnector state precise identification device according to claim 2, wherein The arrangement of the integrated collector is as follows: Three integrated collectors are installed side by side in a straight line, each aligned with a phase of a three-phase high-voltage disconnector, that is, one integrated collector corresponds to one phase of the high-voltage disconnector. The center line of each integrated collector is respectively aligned with the center of the conductive arm of a high-voltage disconnector of each phase. The center line of the integrated collector is parallel to the central axes of the camera and the lidar and is located between the two central axes. The high-voltage disconnector refers to a single-phase high-voltage disconnector. A three-phase high-voltage disconnector includes three interlocked single-phase high-voltage disconnectors. The conductive arm is also called a conductive rod, a conductive plate or a knife switch, and has contacts and wires on it.

4. The high-voltage disconnector state precise identification device according to claim 1, characterized in that, The point cloud discrimination module further includes: A point cloud preprocessing module, which uses the positioning frame generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise in the point cloud data, so as to reduce the calculation amount and improve the recognition accuracy.

5. The closing state judgment method of the high-voltage disconnector state precise recognition device according to any one of claims 1-4, characterized in that, It includes: Image target recognition: Find the conductive arm of the required high-voltage disconnector in an image containing background information, filter out the background, extract the conductive arm from the captured picture, and frame it with a smallest rectangular box; Image contour extraction: Extract the contour line of the conductive arm within the range framed by the rectangular box; Closing and opening action judgment: Calculate the angle of the conductive arm and its change trend over time from the contour line, and infer which of the following states the high-voltage disconnector is currently in: closing, opening, closed or open; Lidar startup: If the current state transitions from closing to closed, a signal is sent by the front-end device for high-voltage disconnector recognition to start lidar scanning and obtain point cloud data; Point cloud preprocessing: Use the positioning frame generated by the image target recognition module to narrow the sampling and processing range of the point cloud data and remove the outlier point cloud noise in the point cloud data; Point cloud discrimination: Directly discriminate whether the closing is in place through the preprocessed point cloud data by a deep learning algorithm.

6. The closing state determination method of the high-voltage disconnector state precise identification device according to claim 5, characterized in that The method for removing outlier point cloud noise is as follows: Given: The given point cloud set \(P = \{p i |p i \in\mathbb{R} 3 , i = 1, 2, 3, \ldots, N\}\), where \(p i \) represents the \(i\)-th point, the total number of points is \(N\), the outlier point cloud set \(Deletion=\{d i |d i \in\mathbb{R} 3 \}\), and \(d i \) represents the \(i\)-th point in the outlier point cloud set, the total number of which is to be determined. \(\mathbb{R} 3 \) represents a bounded three-dimensional space domain, the number of points in the neighborhood range is \(K\), and the variance coefficient is \(\beta\). 1) Read the point cloud file C and convert it to generate a point cloud set P; 2) Construct a kd-Tree to store the point cloud set P; 3) Use the kd-Tree to calculate the K-nearest neighbor neighborhood of each point in the point cloud set P; 4) Calculate the average distance d of point p in its K-nearest neighbor neighborhood according to the formula i where i is the distance between point p and p j , p i is the j-th point in the K-nearest neighbor neighborhood of p j , and the coordinates of p i , p i are (x j , y i , z i ) and (x i , y j , z j , z j ), respectively; 5) Calculate the average K-nearest neighbor neighborhood distance μ and variance σ in the entire point cloud set P according to the formula 2 ;​ 6) Traverse the average distance di of the K-nearest neighbor neighborhoods of each point. If then this point is considered an abnormal outlier, delete it from the point cloud set P, and add it to the outlier point cloud set Deletion; 7) Distance information neighborhood transfer: Traverse the elements in the point cloud set P and use the formula to calculate the score si of each point, where the intermediate variable j = 1, 2, …, K. If then delete it from the point cloud set P and add it to the outlier point cloud set Deletion; 8) Output the remaining points in the point cloud set P as the normal point cloud set and output the final outlier point cloud set.

Images