Methods and devices for detecting the operating status of disconnecting switches

Abstract

This invention discloses a method and apparatus for detecting the operating state of a disconnecting switch. The method includes: acquiring a point cloud dataset; constructing an initial detection model and training it using a target loss function to obtain a target detection model; inputting test data into the target detection model to obtain first point cloud data corresponding to the conductive arm of the disconnecting switch, second point cloud data corresponding to the insulating support of the disconnecting switch, third point cloud data corresponding to the transmission cable of the disconnecting switch, and fourth point cloud data corresponding to the base of the disconnecting switch; obtaining the vector corresponding to the first point cloud data, and determining the operating state of the disconnecting switch based at least on the vector corresponding to the first point cloud data. This invention solves the technical problems of low efficiency and safety hazards associated with manual inspections in related technologies, and the need for complex models and high computational resources in image processing techniques.

Description

Technical Field

[0001] This invention relates to the field of sensing technology, and more specifically, to a method for detecting the operating state of a disconnecting switch, a device for detecting the operating state of a disconnecting switch, and a computer-readable storage medium. Background Technology

[0002] LiDAR (Light Detection and Ranging) is a long-range measurement technology that acquires the position, shape, and surface features of a target object by sending laser pulses and measuring their return time and other properties. With the development of LiDAR 3D scanning technology, the acquisition of 3D point cloud models has become increasingly sophisticated, and these models are widely used in fields such as autonomous driving, machine vision, and medical research. Compared to traditional 2D images, LiDAR-acquired 3D point cloud data models offer advantages such as more object information, real-time performance, and less susceptibility to lighting conditions.

[0003] A disconnecting switch is in the closed state when it is energized, and in the open state when it is de-energized. The process of changing the disconnecting switch from the open / de-energized state to the closed / energized state is the closing process, and the process of changing the disconnecting switch from the closed / energized state to the open / de-energized state is the opening process. The conductive arms of a double-column horizontal rotary disconnecting switch are fixed to the upper end of an insulating support. The operating mechanism drives the two conductive arms to rotate, thereby completing the opening and closing action.

[0004] When the disconnecting switch is in the open state, the left and right conductive arms are parallel to each other. When the disconnecting switch is in the closing process, the operating mechanism drives the two conductive arms to rotate. When the two conductive arms rotate to the same horizontal line, the contacts at the front end of the conductive arms can contact each other and form a stable conductive path. At this time, the disconnecting switch is in the closed state. When the disconnecting switch is in the opening process, the operating mechanism drives the two conductive arms to rotate. When the two conductive arms rotate to a parallel position, the disconnecting switch breaks the conductive path. At this time, the disconnecting switch is in the open state.

[0005] Substations, as a crucial component of the power system, play a vital role in transforming voltage and current, and concentrating and distributing electrical energy. The normal operation of substation power facilities is critical to the safety of the entire power grid. Disconnect switches are one of the most important power facilities in a substation, and ensuring their proper functioning is a crucial aspect of substation power operations. Traditionally, the status of substation disconnect switches is primarily determined through manual inspections. This method suffers from low efficiency, high workload, and certain safety hazards for inspection personnel. With the increasing intelligence and automation of substation safety measures, disconnect switch status detection technology based on machine vision and image processing has developed rapidly. Using image processing technology to determine the disconnect switch's posture ensures the safety of inspection personnel and the reliability of disconnect switch status assessments.

[0006] There is currently no effective solution to the above problems. Summary of the Invention

[0007] This invention provides a method for detecting the operating state of a disconnecting switch, a device for detecting the operating state of a disconnecting switch, and a computer-readable storage medium, in order to at least solve the technical problems of low efficiency and safety hazards of manual inspection in related technologies, and the need for complex models and high computing resources in image processing technology.

[0008] According to one aspect of the present invention, a method for detecting the operating state of a disconnecting switch is provided, comprising: acquiring a point cloud dataset, the point cloud dataset including multiple point cloud data corresponding to the disconnecting switch, each point cloud data including multiple frames of point cloud images obtained by scanning the disconnecting switch to be detected by a lidar, each point cloud image including multiple point cloud data points, a portion of the point cloud dataset being test data and another portion being training data, the disconnecting switch including at least a conductive arm, an insulating support, a power transmission cable, and a base; constructing an initial detection model and training the initial detection model using a target loss function to obtain a target detection model, wherein the target detection model includes a feature extraction module, a classification module, and a segmentation module, the input end of the feature extraction module being the input end of the target detection model, the output end of the feature extraction module being connected to the input end of the classification module and the input end of the segmentation module respectively, the output end of the classification module and the output end of the segmentation module being the output end of the target detection model, the feature extraction module including... The system includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract point cloud features from the input data of the target detection model. The classification module is used to classify the output data of the feature extraction module. The segmentation module is used to segment the output data of the feature extraction module. The test data is input into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the power transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector. The vector corresponding to the first point cloud data is obtained. Based on at least the vector corresponding to the first point cloud data, the action state of the disconnector is determined. The action state includes an open state, a closed state, a disconnection process, and a closing process. The open state is the state of the disconnector when it is not energized, and the closed state is the state of the disconnector when it is energized.

[0009] According to another aspect of the present invention, a detection device for the operating state of a disconnecting switch is also provided, comprising: an acquisition module for acquiring a point cloud dataset, the point cloud dataset including multiple point cloud data corresponding to the disconnecting switch, each point cloud data including multiple frames of point cloud images obtained by scanning the disconnecting switch to be detected by a lidar, each point cloud image including multiple point cloud data points, a portion of the point cloud dataset being test data and another portion being training data, the disconnecting switch including at least a conductive arm, an insulating support, a power transmission cable, and a base; and a first processing module for constructing an initial detection model and training the initial detection model using a target loss function to obtain a target detection model, wherein the target detection model includes a feature extraction module, a classification module, and a segmentation module, the input end of the feature extraction module being the input end of the target detection model, the output end of the feature extraction module being connected to the input end of the classification module and the input end of the segmentation module respectively, the output end of the classification module and the output end of the segmentation module being the output end of the target detection model, the feature extraction module... The system includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module extracts point cloud features from the input data of the target detection model. The classification module classifies the output data of the feature extraction module, and the segmentation module segments the output data of the feature extraction module. An input module inputs the test data into the target detection model to obtain first point cloud data corresponding to the conductive arm of the disconnector, second point cloud data corresponding to the insulating support of the disconnector, third point cloud data corresponding to the power transmission cable of the disconnector, and fourth point cloud data corresponding to the base of the disconnector. A second processing module obtains the vector corresponding to the first point cloud data and determines the operating state of the disconnector based at least on the vector corresponding to the first point cloud data. The operating state includes an open state, a closed state, a disconnection process, and a closing process. The open state is the state of the disconnector when it is not energized, and the closed state is the state of the disconnector when it is energized.

[0010] Optionally, the acquisition module includes: a scanning processing unit, used to scan the disconnector switch using a lidar to obtain multiple initial point cloud data; and a filtering and denoising processing unit, used to filter and denoise the multiple initial point cloud data respectively to obtain the point cloud dataset.

[0011] Optionally, the input module includes: a first input unit, configured to input the test data into the first local feature extraction submodule to perform a first downsampling and first feature extraction process on the test data, obtaining multiple first point cloud data points; a second input unit, configured to sequentially input the multiple first point cloud data points into the first channel attention submodule to perform a first feature dimension weighting process on the multiple first point cloud data points, obtaining multiple second point cloud data points; a third input unit, configured to input the multiple second point cloud data points into the second local feature extraction submodule to perform a second downsampling and second feature extraction process on the multiple second point cloud data points, obtaining multiple third point cloud data points; and a fourth input unit, configured to sequentially input the multiple third point cloud data points into the second channel attention submodule to perform a first feature dimension weighting process on the test data, obtaining multiple second point cloud data points. Multiple third point cloud data points are weighted by a second feature dimension to obtain multiple fourth point cloud data points; a fifth input unit is used to input multiple test data, multiple second point cloud data points, and multiple fourth point cloud data points into the segmentation module to segment the multiple fourth point cloud data points respectively to obtain multiple fifth point cloud data points, the number of fifth point cloud data points being the same as the number of point cloud data in the test data; a sixth input unit is used to input multiple fourth point cloud data points into the classification module to classify the multiple fourth point cloud data points respectively to obtain multiple sixth point cloud data points, the multiple fifth point cloud data points and the multiple sixth point cloud data points forming the first point cloud data, the second point cloud data, the third point cloud data, and the fourth point cloud data.

[0012] Optionally, the first input unit includes: a first input subunit, configured to input the test data into the downsampling layer VFS to perform voxel-farthest-point downsampling processing on the test data to obtain a first sampling point set, the first sampling point set including multiple first sampling points; a second input subunit, configured to input the test data and the first sampling point set into the weighted mean grouping layer PAG to perform grouping processing with the first sampling points as centroids to obtain a second sampling point set, the second sampling point set being a local feature point set corresponding to each first sampling point, the second sampling point set including multiple second sampling points; a third input subunit, configured to input the second sampling point set into the local feature point set input into the point network layer to perform feature extraction processing on the multiple second sampling points respectively to obtain multiple first point cloud data points; and a fourth input subunit, configured to input the multiple second point cloud data points into the second local feature extraction submodule VFS. PP performs second downsampling and second feature extraction processing on multiple second point cloud data points to obtain multiple third point cloud data points, including: a fifth input subunit, used to sequentially input multiple second point cloud data points into the downsampling layer to perform voxel farthest point downsampling processing on the second point cloud data points to obtain a third sampling point set, the third sampling point set including multiple third sampling points; a sixth input subunit, used to input multiple second point cloud data points and the third sampling point set into the weighted mean grouping layer to group the third sampling points with the third sampling points as centroids to obtain a fourth sampling point set, the fourth sampling point set being the local feature point set corresponding to each of the third sampling points, the fourth sampling point set including multiple fourth sampling points; and a seventh input subunit, used to input the fourth sampling point set into the local feature point set input point network layer to perform feature extraction processing on multiple fourth sampling points respectively to obtain multiple third point cloud data points.

[0013] Optionally, the second input unit includes: an eighth input subunit, configured to sequentially input multiple first point cloud data points to the activation function layer to calculate the weights of the feature dimensions of each first point cloud data point, thereby obtaining multiple first weight values; a ninth input subunit, configured to input multiple first weight values ​​and corresponding first point cloud data points to the calculation layer to perform dot product processing on the first weight values ​​and corresponding first point cloud data points, thereby obtaining multiple second point cloud data points; a tenth input subunit, configured to sequentially input multiple third point cloud data points to the second channel attention submodule to perform second feature dimension weighting processing on the multiple third point cloud data points, thereby obtaining multiple fourth point cloud data points, including: sequentially inputting multiple third point cloud data points to the activation function layer to calculate the weights of the feature dimensions of each third point cloud data point, thereby obtaining multiple second weight values; and an eleventh input subunit, configured to input multiple second weight values ​​and corresponding third point cloud data points to the calculation layer to perform dot product processing on the second weight values ​​and corresponding third point cloud data points, thereby obtaining multiple fourth point cloud data points.

[0014] Optionally, the fifth input unit includes: a twelfth input subunit, used to input multiple second point cloud data points and multiple fourth point cloud data points into the first interpolation layer, using the second point cloud data points as interpolation points and the fourth point cloud data points as nearest neighbor feature points, to perform a first interpolation process on the second point cloud data points and the fourth point cloud data points to obtain a first data point set; a thirteenth input subunit, used to input the first data point set into the first convolutional layer to perform feature extraction processing on the first data point set to obtain a second data point set, the second data point set including multiple second data points; a fourteenth input subunit, used to input the second data point set and the test data into the second interpolation layer, using the test data as interpolation points and the second data points as nearest neighbor feature points, to perform a second interpolation process on the second data point set and the test data to obtain a third data point set, the third data point set including multiple third data points; and a fifteenth input subunit, used to input the third data point set into the second convolutional layer to perform feature extraction processing on the multiple third data points to obtain multiple fifth point cloud data points.

[0015] Optionally, the sixth input unit includes: a sixteenth input subunit, used to sequentially input multiple fourth point cloud data points into the third convolutional layer to perform feature extraction processing on the multiple fourth point cloud data points respectively, to obtain multiple seventh point cloud data points; and a seventeenth input subunit, used to input multiple seventh point cloud data points into the fully connected layer to perform classification processing on the seventh point cloud data points, to obtain multiple sixth point cloud data points.

[0016] Optionally, the second processing module includes: a calculation unit, used to calculate according to the formula Calculate the angle α between the conductive arm corresponding to the point cloud image in the i-th frame and the conductive arm in the combined state. i , where u i =((x0,y0),(x i ,y i (x0, y0) are the coordinates of the end of the conductive arm that is fixed to the stationary end of the insulating support. i ,y i Let ) be the coordinates of the end of the conductive arm that rotates around the insulating post, v = ((x0, y0), (x v ,y v )), (x v ,y v ) represents the coordinate of the end of the conductive arm that rotates around the insulating support when in the closed state, u i ·v is a vector u i The dot product of vector v and vector v, |u i | is vector u i The modulus, |v| is the modulus of vector v; the first determining unit is used to determine α. i Is it greater than α? i+n , in α i Greater than α i+n In the case of α, determine i+n The corresponding operating state of the disconnecting switch is the closing process, where α i+n The angle between the conductive arm corresponding to the point cloud image in the (i+n)th frame and the conductive arm in the combined state, where n is greater than 10; the second determining unit is used to determine α i Less than α i+n In the case of α, determine i The corresponding disconnecting switch's operating state is the disconnection process; the third determining unit is used to determine α i equal to α i+n In the case of α, determine i and α i+n Does α satisfy? i+n =α i =0, in α i+n =α i When α = 0, determine α i The corresponding disconnecting switch's operating state is the closed state; the fourth determining unit is used to determine α i+n =α i In the case that α ≠ 0, determine α i The corresponding disconnector switch is in the open state.

[0017] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the detection method for the operation state of the disconnecting switch as described above.

[0018] According to another aspect of the present invention, a computer program product is also provided, including computer instructions, which, when executed by a processor, perform a method for detecting the operating state of an isolating switch as described in any one of the above embodiments.

[0019] According to another aspect of the present invention, a processor is also provided, the processor being configured to run a program, wherein the program, when running, executes the method for detecting the operating state of the disconnecting switch as described in any of the above embodiments.

[0020] In this embodiment of the invention, a point cloud dataset is acquired. The point cloud dataset includes multiple point cloud data corresponding to the disconnector switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnector switch to be detected using a LiDAR. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data, and the other part is training data. The disconnector switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base. An initial detection model is constructed, and the initial detection model is trained using a target loss function to obtain a target detection model. The target detection model includes a feature extraction module, a classification module, and a segmentation module. The input of the feature extraction module is the input of the target detection model. The output of the feature extraction module is connected to the input of the classification module and the input of the segmentation module, respectively. The outputs of the classification module and the segmentation module are the outputs of the target detection model. The feature extraction module includes stacked local feature extraction. The system includes a submodule and an attention submodule. A feature extraction module extracts point cloud features from the input data of the target detection model. A classification module classifies the output data of the feature extraction module, and a segmentation module segments the output data of the feature extraction module. Test data is input into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector. The system obtains the vector corresponding to the first point cloud data and determines the operating state of the disconnector based on at least the vector corresponding to the first point cloud data. The operating state includes the open state, closed state, disconnection process, and closing process. The open state is the state of the disconnector when it is not energized, and the closed state is the state of the disconnector when it is energized. The technical solution provided by this invention achieves the goal of acquiring three-dimensional point cloud data of disconnect switches using LiDAR, and combining voxel farthest point sampling and weighted mean grouping techniques. At the same time, the point cloud channel attention mechanism module is used to enhance the weight of feature dimensions useful for the current task, thereby realizing the technical effect of real-time automated monitoring of the disconnect switch's operating status, improving the accuracy and efficiency of monitoring, and solving the technical problems of low efficiency and safety hazards of manual inspection in related technologies, and the need for complex models and high computing resources in image processing technology. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0022] Figure 1 This is a flowchart of a method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention;

[0023] Figure 2 This is a technical roadmap of a method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention;

[0024] Figure 3 This is a structural diagram of the VPP module of the method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram of the PSE module of the method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention;

[0026] Figure 5 This is a schematic diagram of point cloud feature information of a method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of the coordinates of both ends of the conductive arm of the disconnector switch according to an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of the conductive arm of the disconnector switch according to an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the disconnector switch action monitoring model structure according to an embodiment of the present invention;

[0030] Figure 9 This is a schematic diagram of a device for detecting the operating state of a disconnecting switch according to an embodiment of the present invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] As described in the background section, manual inspection in related technologies is inefficient and poses safety hazards, while image processing technology requires complex models and high computing resources. To address these shortcomings, embodiments of the present invention provide a method for detecting the operating state of a disconnecting switch, a device for detecting the operating state of a disconnecting switch, and a computer-readable storage medium.

[0034] According to an embodiment of the present invention, a method embodiment for detecting the operating state of a disconnecting switch is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0035] Figure 1 This is a flowchart of a method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0036] Step S202: Obtain a point cloud dataset. The point cloud dataset includes multiple point cloud data corresponding to the disconnecting switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnecting switch to be detected by LiDAR. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data, and the other part is training data. The disconnecting switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base.

[0037] In this embodiment, a point cloud dataset including multiple point cloud data corresponding to the disconnect switch can be obtained, which facilitates subsequent analysis of the point cloud data, extraction of the disconnect switch features, and automatic identification and positioning of the disconnect switch.

[0038] Step S204: Construct an initial detection model and train it using a target loss function to obtain an object detection model. The object detection model includes a feature extraction module, a classification module, and a segmentation module. The input of the feature extraction module is the input of the object detection model, and the output of the feature extraction module is connected to the inputs of the classification module and the segmentation module, respectively. The outputs of the classification module and the segmentation module are the outputs of the object detection model. The feature extraction module includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract point cloud features from the input data of the object detection model. The classification module is used to classify the output data of the feature extraction module, and the segmentation module is used to segment the output data of the feature extraction module.

[0039] Figure 2 This is a technical roadmap of a method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention, such as... Figure 2 As shown, after constructing the initial disconnector switch action monitoring model, the model is trained to obtain the target monitoring model.

[0040] In this embodiment, an initial detection model is constructed and trained using a cross-entropy loss function (i.e., the target loss function) to improve the model's accuracy and generalization ability, resulting in a disconnect switch action monitoring model that includes a feature extraction module, a classification module, and a segmentation module. The feature extraction module can extract key disconnect switch action features, the classification module can classify these features, and finally the segmentation module can segment the monitoring area to determine the specific location of the disconnect switch action.

[0041] Step S206: Input the test data into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the power transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector.

[0042] In this embodiment, all point cloud data points in a frame of disconnector switch point cloud data are input into the target detection model to obtain point cloud data divided into four components: disconnector switch conductive arm, insulating support, power transmission cable, and disconnector switch base. This enables automatic identification and segmentation of the disconnector switch and its components in the test data for further analysis and processing, thereby improving the accuracy and generalization ability of the target detection model.

[0043] Step S208: Obtain the vector corresponding to the first point cloud data. Based on at least the vector corresponding to the first point cloud data, determine the action state of the disconnecting switch. The action state includes the open state, the closed state, the disconnection process, and the closing process. The open state is the state of the disconnecting switch when it is not energized, and the closed state is the state of the disconnecting switch when it is energized.

[0044] like Figure 2 As shown, a trained disconnector switch action monitoring model is applied to determine the disconnector switch action status.

[0045] In this embodiment, the operating status of the disconnecting switch can be determined based on the vector corresponding to the first point cloud data. This allows for the timely detection of abnormalities in the disconnecting switch, such as abnormal opening or closing processes or inconsistent opening and closing states. Consequently, timely measures can be taken for maintenance or repair to ensure the safe and stable operation of the power system.

[0046] As described above, in this embodiment of the invention, a point cloud dataset is obtained, which includes multiple point cloud data corresponding to the disconnector switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnector switch to be detected using a LiDAR. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data, and the other part is training data. The disconnector switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base. An initial detection model is constructed, and the initial detection model is trained using a target loss function to obtain a target detection model. The target detection model includes a feature extraction module, a classification module, and a segmentation module. The input of the feature extraction module is the input of the target detection model, and the output of the feature extraction module is connected to the input of the classification module and the input of the segmentation module. The outputs of the classification module and the segmentation module are the outputs of the target detection model. The feature extraction module includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract the point cloud features of the input data of the target detection model, and the classification module is used to process the output data of the feature extraction module. The classification and segmentation module is used to segment the output data of the feature extraction module. The test data is input into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the power transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector. The vector corresponding to the first point cloud data is obtained. Based on at least the vector corresponding to the first point cloud data, the action state of the disconnector is determined. The action state includes the open state, the closed state, the disconnection process, and the closing process. The open state is the state of the disconnector when it is not energized, and the closed state is the state of the disconnector when it is energized. This achieves the technical effect of using LiDAR to obtain the three-dimensional point cloud data of the disconnector, combined with voxel farthest point sampling and weighted mean grouping technology. At the same time, the point cloud channel attention mechanism module is used to improve the weight of the feature dimensions that are useful to the current task, thereby realizing the technical effect of real-time automated monitoring of the action state of the disconnector and improving the accuracy and efficiency of monitoring.

[0047] The technical solution provided by this invention solves the technical problems of low efficiency and safety hazards of manual inspection in related technologies, and the need for complex models and high computing resources in image processing technology.

[0048] According to the above embodiments of the present invention, obtaining a dataset includes: scanning the disconnector switch with a lidar to obtain multiple initial point cloud data; and filtering and denoising the multiple initial point cloud data to obtain a point cloud dataset.

[0049] In this embodiment, a point cloud dataset of disconnect switches can be constructed and preprocessed. By scanning the disconnect switches with LiDAR, the three-dimensional spatial information of the disconnect switches can be obtained, resulting in multiple initial point cloud data. After filtering and denoising, some noise points can be removed, making the point cloud data more accurate and clear. The obtained point cloud dataset can be used for subsequent applications such as modeling, recognition, and positioning, helping to achieve precise control and monitoring of disconnect switches.

[0050] Specifically, a lidar scanner is used to scan the disconnector switch to acquire its 3D point cloud data, constructing a disconnector switch point cloud dataset. Each point cloud image in the dataset consists of consecutive 3D point cloud images of the disconnector switch, containing a minimum of 100 frames and a maximum of 200 frames. The point cloud data set in each frame is P = {p1, p2, ..., p...} n}, p represents a point cloud data point in P, and the size of P is n×c, where n is the number of point cloud data points contained in P, and c is the feature dimension of the point cloud data point. At this time, c=3, that is, each point p contains three-dimensional coordinate information of (x,y,z).

[0051] In addition, the point cloud images in the disconnector switch point cloud dataset are filtered and denoised, and the point cloud positions of the disconnector switch conductive arms in the point cloud images are labeled.

[0052] According to the above embodiments of the present invention, the feature extraction module includes a first local feature extraction submodule, a first channel attention submodule, a second local feature extraction submodule, and a second channel attention submodule connected in sequence. Test data is input into the target detection model to obtain first point cloud data corresponding to the conductive arm of the disconnector, second point cloud data corresponding to the insulating support of the disconnector, third point cloud data corresponding to the transmission cable of the disconnector, and fourth point cloud data corresponding to the base of the disconnector. The module includes: inputting test data into the first local feature extraction submodule to perform a first downsampling and first feature extraction process on the test data to obtain multiple first point cloud data points; inputting the multiple first point cloud data points sequentially into the first channel attention submodule to perform a first feature dimension weighting process on the multiple first point cloud data points to obtain multiple second point cloud data points; and inputting the multiple first point cloud data points sequentially into the first channel attention submodule to perform a first feature dimension weighting process on the multiple first point cloud data points to obtain multiple second point cloud data points; and inputting the multiple first point cloud data points into the first channel attention submodule to perform a first feature dimension weighting process on the multiple first point cloud data points to obtain multiple second point cloud data points; and inputting the multiple first point cloud data points into the first channel attention submodule to obtain multiple second point cloud data points. The two point cloud data points are input into the second local feature extraction submodule to perform second downsampling and second feature extraction processing on the multiple second point cloud data points, resulting in multiple third point cloud data points. The multiple third point cloud data points are then sequentially input into the second channel attention submodule to perform second feature dimension weighting processing on the multiple third point cloud data points, resulting in multiple fourth point cloud data points. The multiple test data, multiple second point cloud data points, and multiple fourth point cloud data points are input into the segmentation module to perform segmentation processing on the multiple fourth point cloud data points respectively, resulting in multiple fifth point cloud data points. The number of fifth point cloud data points is the same as the number of point cloud data points in the test data. The multiple fourth point cloud data points are input into the classification module to perform classification processing on the multiple fourth point cloud data points respectively, resulting in multiple sixth point cloud data points. The multiple fifth point cloud data points and the multiple sixth point cloud data points form the first point cloud data, the second point cloud data, the third point cloud data, and the fourth point cloud data.

[0053] In this embodiment, a frame of isolation switch point cloud data point set P = {p1, p2, ..., p nAll point cloud data points in the set are input into the disconnector switch action monitoring model. The size of P is n×c, where n is the number of point cloud data points contained in the point cloud data set P, and c is the feature dimension of the point cloud data points. After P is processed by the first VPP module (i.e., the first local feature extraction submodule) for downsampling (i.e., first downsampling) and feature extraction (i.e., first feature extraction), a point cloud data set P1 (i.e., multiple first point cloud data points) is obtained. P1 is then processed by the first VSE module (i.e., the first channel attention submodule) for feature dimension weighting (i.e., first feature dimension weighting), resulting in a point cloud dataset P2 (i.e., multiple second point cloud data points). P2 is then processed by the second VPP module (i.e., the second local feature extraction submodule) for downsampling (i.e., second downsampling) and feature extraction (i.e., second feature extraction). After processing, a point cloud data set P3 (i.e., multiple third point cloud data points) is obtained. P3 is then processed by the second VSE module (i.e., the second channel attention submodule) to perform feature dimension weighting (i.e., second feature dimension weighting) on ​​the point cloud data points, resulting in a point cloud data set P4 (i.e., multiple fourth point cloud data points). Finally, P4 is input into the segmentation module ST and the classification module CF respectively for segmentation and classification, resulting in point cloud data divided into four components: disconnector conductive arm, insulating support, power transmission cable, and disconnector base, namely, the aforementioned first point cloud data, second point cloud data, third point cloud data, and fourth point cloud data.

[0054] In the above embodiments of the present invention, the first local feature extraction submodule and the second local feature extraction submodule are the same. The first local feature extraction submodule and the second local feature extraction submodule respectively include a downsampling layer, a weighted mean grouping layer, and a local feature point set input point network layer. The test data is input to the first local feature extraction submodule to perform first downsampling and first feature extraction processing on the test data to obtain multiple first point cloud data points, including: inputting the test data to the downsampling layer to perform voxel farthest point downsampling processing on the test data to obtain a first sampling point set, which includes multiple first sampling points; inputting the test data and the first sampling point set into the weighted mean grouping layer to group the test data with the first sampling points as centroids to obtain a second sampling point set, which is a local feature point set corresponding to each first sampling point, and includes multiple second sampling points; inputting the second sampling point set into the local feature point set input point network layer. The process involves: performing feature extraction on multiple second sampling points to obtain multiple first point cloud data points; inputting these multiple second point cloud data points into a second local feature extraction submodule to perform second downsampling and second feature extraction on them to obtain multiple third point cloud data points; sequentially inputting these multiple second point cloud data points into a downsampling layer to perform voxel-farthest point downsampling on them to obtain a third sampling point set, which includes multiple third sampling points; inputting the multiple second point cloud data points and the third sampling point set into a weighted mean grouping layer to group the data points with the third sampling points as centroids to obtain a fourth sampling point set, which is a local feature point set corresponding to each third sampling point and includes multiple fourth sampling points; and inputting the fourth sampling point set into a local feature point set into a point network layer to perform feature extraction on the multiple fourth sampling points to obtain multiple third point cloud data points.

[0055] Figure 3 This is a structural diagram of the VPP module of the method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention, as shown below. Figure 3 As shown, the point cloud data set P of the disconnector switch is input into the VPP module. After the input disconnector switch point cloud information passes through the voxel farthest point downsampling layer VFS, n′ data points are sampled from P. Then, with the sampled n′ data points as the center points, the data is grouped by the weighted mean grouping layer PAG to obtain n′ local feature point sets. Finally, the n′ sampled points and the corresponding local feature point sets are input into the Pointnet layer to extract features.

[0056] Obtain the first set of sampling points mentioned above, that is, input the point cloud data point set P = {p1, p2, ..., p} of the isolation switch with a size of n×c in the above frame. nIn the VFS layer, n is the number of point cloud data points in P, and c is the number of feature dimensions of the point cloud data points in P. The point cloud space containing the point cloud data points in P is divided into 4×4×4 voxel spaces. The distance between each point in each voxel space is calculated using the following formula:

[0057] Where, d ij For point p i and point p j The distance between them, (x i ,y i ,z i Let p be a point. i Position coordinates, (x j ,y j ,z j Let p be a point. j Position coordinates, p i and p j For any point in voxel space, sample from each voxel space. The specific steps are as follows: 1) Randomly select an unsampled voxel space and create a query point set T to temporarily store the sampled points. Randomly select a point a from the selected voxel space and put it into T. From the remaining points in the voxel space (i.e., points not in T), select a point b that is farthest from point a and put it into T. 2) Put the extracted points a and b into T. From the remaining points in the voxel space, select the point c that is farthest from T and put it into T. The method for solving c is as follows: Calculate the distance d from all remaining points in the voxel space to T. The point with the largest distance d from T is point c. The method for calculating the distance d is as follows: For any point t among the remaining points in the voxel space, calculate the distance from t to all points in T. Then, take the minimum distance between t and the points in T as the distance d from t to T. t 3) Repeat step two above until samples are collected within the voxel space. 4) Repeat the above steps until all voxel spaces have been sampled. Merge all sampled points from all voxel spaces to obtain a total of n′ sampled points, forming a sampled data point set V′={v1′,v2′,…,v n "} (i.e., the first set of sampling points) is the output of the VFS layer, and V′ has a size of n′×c.

[0058] It should be noted that the aforementioned voxel space refers to cubes or cube mesh units uniformly distributed in three-dimensional space.

[0059] To obtain the second set of sampling points, that is, input the point cloud data set P = {p1, p2, ..., p} of the isolation switch with a size of n×c in the above frame. n} and the above sampling data point set V′={v1′,v2′,…,v n "}, using the sampling points in V′ as centroids, perform grouping operations to obtain the local feature point set corresponding to the sampling points. The specific steps are as follows: 1) Randomly select an ungrouped sampling point v in V′. i Let ' be the centroid, and set the radius r of the spherical neighborhood used for grouping operations and the number of local feature points k. 2) Find v in P. i Within the sphere's neighborhood of v i The k nearest neighbors are taken as local feature points, and these k neighbors constitute v. i Find v from the local feature point set of ′. i The formula for the local feature point set corresponding to ′ is as follows:

[0060] Where q(r,k) is v i The local feature point set of ′, ∥vv i ′∥ represents the neighborhood points v and v i The distance between ' and ', where s is the number of points in set q(r,k), and s≤k indicates that in v i Select k neighborhoods of v from the sphere of ′ i If v' is the nearest neighbor point that is smaller than the radius r of the sphere's neighborhood, then... i If the number of neighborhood points within the sphere's neighborhood of ' is greater than or equal to k, select the first k neighborhood points that are related to v. i The nearest neighbor points are taken as local feature points, and these k neighbor points constitute v. i The local feature point set of ′, based on point v i The grouping operation of ' is complete; if v i If the number of neighborhood points within the sphere's neighborhood is less than k, then v can be calculated. i The weighted mean of all points in the sphere's neighborhood is given by the following formula:

[0061] in, v is the weighted mean point. j For v i For any point s in the sphere's neighborhood of ', v i The number of points in the sphere's neighborhood, d j For v i Any point v in the neighborhood of the sphere ′ j to v i The distance d' s For v i All points within the sphere's neighborhood to v i The sum of distances between ′ and ′. The weighted mean point. Copy multiple times to get k points as v i The local feature points of ′ are used to obtain v iThe local feature point set of ′, based on point v i The grouping operation is complete. 3) Repeat the above steps until the local feature point set corresponding to all sampling points in V′ is obtained. The grouping operation ends, and a point cloud data point set V″ (i.e., the second sampling point set) of size n′×k×c containing sampling points and their corresponding local feature point sets is obtained, where n′ is the number of sampling points, k is the number of local feature points of the sampling points, and c is the feature dimension of the sampling points; output the point cloud data point set V″.

[0062] It should be noted that the centroid point mentioned above is the center point of the spherical neighborhood used for grouping; the weighted mean point mentioned above is a data point obtained by calculating the weight value based on the distance of the neighborhood point from the centroid, and then using the weight value to calculate the mean value of the neighborhood points.

[0063] To obtain the specific first point cloud data points mentioned above, the point cloud data point set V″ is input into the Pointnet layer. After the Pointnet's fully connected layer, ReLU activation function, and MaxPool layer extract features, the feature dimension of the isolation switch point cloud data points is increased to c′, and the output isolation switch point cloud data point set P1 with the expanded feature dimension is n′×c′.

[0064] It should be noted that the PointNet described above is a lightweight convolutional network suitable for point cloud data. It extracts features by fusing local information and consists of a fully connected layer, a ReLU activation function, and a max pooling function.

[0065] Taking the PAG layer parameters of spherical neighborhood radius r = 5 and neighborhood points k = 10 as an example, this embodiment is illustrated. A set of point cloud data points P is input, assuming a size of 4096 × 3. After downsampling by the VFS layer, a set of sampled point data points V′ is obtained, with a size of 1024 × 3. A single ungrouped point cloud data point v is randomly selected from V′. i Grouping operations are performed using ' as the centroid point to determine v. i Given a spherical neighborhood with centroid ′ and radius r, find the distance v in P. i 'Place v at the k nearest points in the spherical neighborhood' i If the number of points in the local feature point set of ′ is less than k within the spherical neighborhood, then calculate the weighted mean point of all points within the spherical neighborhood and copy the weighted mean point into v. i If the local feature points of ′ are gathered to fill k points, then v is obtained. iThe local feature point set of V′ is repeatedly selected from the sampling point dataset V′ and grouped. After all sampling points in the sampling point dataset V′ have obtained their corresponding local feature point sets, the grouping operation is completed, resulting in a point cloud dataset V″ containing the sampling points and their corresponding local feature point sets with a size of 1024×10×3. Finally, after passing through PointNet and extracting features, the point cloud dataset P1 is obtained as the output of the VPP module, with a size of 1024×128.

[0066] In the above embodiments of the present invention, the first channel attention submodule and the second channel attention submodule are identical. The first channel attention submodule and the second channel attention submodule each include an activation function layer and a computation layer. Multiple first point cloud data points are sequentially input into the first channel attention submodule to perform first feature dimension weighting processing on the multiple first point cloud data points to obtain multiple second point cloud data points. This includes: sequentially inputting the multiple first point cloud data points into the activation function layer to calculate the weight of the feature dimension of each first point cloud data point, obtaining multiple first weight values; and inputting the multiple first weight values ​​and the corresponding first point cloud data points into the computation layer to respectively perform first feature dimension weighting processing on the multiple first point cloud data points. The first weight value and the corresponding first point cloud data point are multiplied to obtain multiple second point cloud data points. The multiple third point cloud data points are then sequentially input into the second channel attention submodule to perform second feature dimension weighting on the multiple third point cloud data points to obtain multiple fourth point cloud data points. This includes: sequentially inputting the multiple third point cloud data points into the activation function layer to calculate the weight of the feature dimension of each third point cloud data point to obtain multiple second weight values; and inputting the multiple second weight values ​​and the corresponding third point cloud data points into the calculation layer to perform dot product on the second weight values ​​and the corresponding third point cloud data points to obtain multiple fourth point cloud data points.

[0067] In this embodiment, the Feature Dimension Attention (PSE) module, which is suitable for point cloud data, is used to obtain the importance of each point cloud data point's information dimension, that is, to obtain the weight value of each point cloud data point's feature dimension. Then, based on this weight value, the feature dimension of each point cloud data point is weighted and calculated to increase the weight of feature dimensions that are useful for the current task and decrease the weight of feature dimensions that are not very useful for the current task.

[0068] Figure 4 This is a schematic diagram of the PSE module in the method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention, as shown below. Figure 4 As shown, input the above isolation switch point cloud data point set P1, calculate the feature dimension of each point cloud data point through the sigmoid activation function to obtain the weight value of the feature dimension of the point cloud data point, and finally multiply the weight value of the feature dimension of the point cloud data point with its corresponding feature dimension to obtain the weighted isolation switch point cloud data point set P2.

[0069] The specific steps are as follows: 1) Input a point cloud data set P1 of size n′×c′ to the PSE module, where n′ is the number of point cloud data points in P1, and c′ is the number of feature dimensions for each point cloud data point. Use the sigmoid activation function to obtain the weight value of the feature dimension of each point cloud data point. The calculation formula is as follows: W i =σ(p i ), where W i For p i The weight values ​​of the feature dimensions, p i For the input point cloud data points, σ(·) represents the sigmoid function, and the formula for calculating σ(·) is as follows: 2) The weights W of the feature dimensions of the point cloud data points in P1 are compared with those of the point cloud data points. i The corresponding dot product operation is then performed to weight the feature dimensions of the point cloud data points, resulting in feature dimension-weighted point cloud data points. The calculation formula is as follows: Where, p i (That is, the first point cloud data point) is the point cloud data point in P1, W i (That is, the first weight value) is p i The corresponding feature dimension weight value, p i ′ represents the point cloud data points weighted by the feature dimension. After all point cloud data points are calculated, the point cloud data point set P2 of the isolation switch weighted by the feature dimension is obtained.

[0070] Figure 5 This is a schematic diagram of point cloud feature information of a method for detecting the operating state of a disconnecting switch according to an embodiment of the present invention, such as... Figure 5 As shown, point clouds are generally 3-dimensional, but after feature extraction by the network, many more feature information will be added, expanding the feature dimension of the point cloud. The number of feature dimensions of the point cloud will also increase. Moreover, when extracting features from point cloud data, the xyz coordinate information of the point cloud data is not included in the calculation, and the extracted feature information is concatenated after the xyz coordinate information.

[0071] Taking a point cloud data set P1 of size 1024×128 as an example, this embodiment is illustrated by inputting the point cloud data set P1 into the PSE module and using the sigmoid function to calculate the weight value W of the feature dimension of each point cloud data point. i Then the weight value W i Perform a dot product with the feature dimension of the corresponding point cloud data point, and weight the feature dimension of the point cloud data point to obtain the feature dimension-weighted point cloud dataset P2.

[0072] In addition, multiple fourth point cloud data points are obtained by stacking the VPP module and the PSE module. The above point cloud data point set P2 is input into the second VPP module for downsampling and feature extraction to obtain the point cloud data point set P3. P3 is input into the second PSE module (i.e., the second channel attention submodule) for feature dimension weighting of the point cloud data points to obtain the point cloud data point set P4.

[0073] It should be noted that the operation steps of the second VPP module are the same as those of the first VPP module, and the operation steps of the second PSE module are the same as those of the first PSE module, so they will not be repeated here.

[0074] In the above embodiments of the present invention, the segmentation module includes a first interpolation layer, a first convolutional layer, a second interpolation layer, and a second convolutional layer connected in sequence. Multiple test data points, multiple second point cloud data points, and multiple fourth point cloud data points are input into the segmentation module to segment the multiple fourth point cloud data points respectively, obtaining multiple fifth point cloud data points. This includes: inputting the multiple second point cloud data points and the multiple fourth point cloud data points into the first interpolation layer, using the second point cloud data points as interpolation points and the fourth point cloud data points as nearest neighbor feature points, to perform a first interpolation process on the second point cloud data points and the fourth point cloud data points, obtaining... The first data point set is input into the first convolutional layer to perform feature extraction processing, resulting in a second data point set, which includes multiple second data points. The second data point set and test data are input into the second interpolation layer, with the test data as the interpolation point and the second data points as the nearest neighbor feature points, to perform second interpolation processing, resulting in a third data point set, which includes multiple third data points. The third data point set is input into the second convolutional layer to perform feature extraction processing on the multiple third data points, resulting in multiple fifth point cloud data points.

[0075] In this embodiment, the point cloud dataset P4 is input to the segmentation module ST for segmentation, so as to classify different point cloud data points into different categories, enabling more in-depth and accurate analysis and processing of point cloud data.

[0076] Specifically, firstly, the (x, y, z) coordinates of all point cloud data points are obtained from the point cloud data point set P2 mentioned above. The (x, y, z) coordinates of all point cloud data points form a point cloud coordinate set N, with a size of n′. Points within N are used as interpolation points, and points within the point cloud data point set P4 are used as nearest neighbor feature points. These points are input into the interpolation layer Interpolate for interpolation operations. Finally, features are extracted through a Unit Pointnet network to obtain a point cloud data point set P5 (i.e., multiple second data point sets) of size n′×c′.

[0077] Next, the (x, y, z) coordinates of all point cloud data points are obtained from the above data point set P. The (x, y, z) coordinates of all point cloud data points form a point cloud coordinate set N1 with a size of n. The points in N1 are used as interpolation points, and the points in the point cloud data point set P5 are used as nearest neighbor feature points. The points are input into the interpolation layer Interpolate for interpolation. Finally, features are extracted through a UnitPointnet network to obtain a point cloud data point set P6 of size n×(c+s) (i.e., multiple fifth point cloud data points), where s is the number of different categories. At this time, the number of point cloud data points in P6 is restored to the same as the number of point cloud data points in the above point cloud data point set P, and the point clouds in P6 have been divided into multiple different component categories.

[0078] It should be noted that the Interpolate interpolation layer mentioned above is an interpolation method based on the k-nearest neighbor backward distance weighted average, which is a common method in point cloud processing; PointNet mentioned above is a lightweight convolutional network suitable for point cloud data, which extracts features by fusing local information, and consists of a fully connected layer, a ReLU activation function, and a max pooling function; Unit PointNet mentioned above is similar to a 1×1 convolution. Unit PointNet is PointNet without pooling operations. It uses a fully connected layer and ReLU to reconstruct the feature vector of each point in the input point cloud data set without changing the number of points.

[0079] In the above embodiments of the present invention, the classification module includes a third convolutional layer and a fully connected layer connected in sequence. Multiple fourth point cloud data points are input into the classification module to classify the multiple fourth point cloud data points respectively, resulting in multiple sixth point cloud data points. This includes: inputting the multiple fourth point cloud data points sequentially into the third convolutional layer to perform feature extraction processing on the multiple fourth point cloud data points respectively, resulting in multiple seventh point cloud data points; and inputting the multiple seventh point cloud data points into the fully connected layer to classify the seventh point cloud data points, resulting in multiple sixth point cloud data points.

[0080] In this embodiment, the point cloud data set P4 is input to the classification module CF for classification. Multiple data points can be classified simultaneously to obtain multiple classification results, thereby achieving rapid classification processing of multiple data points.

[0081] Specifically, the point cloud data set P4 is input into the classification module CF. After feature extraction through a Pointnet layer (i.e., the third convolutional layer), the classification is completed through a fully connected layer FC, resulting in four categories of information corresponding to the point cloud data: disconnector conductive arm, insulating support, transmission cable, and disconnector base. This information is then combined with the point cloud data set P6, which is segmented into different components, to obtain point cloud data that is segmented and classified into four components: disconnector conductive arm, insulating support, transmission cable, and disconnector base.

[0082] Taking a point cloud dataset P of size 4096×3 as an example, this embodiment is illustrated. A frame of isolation switch point cloud dataset P is input. After downsampling and feature extraction by the first VPP module, a point cloud dataset P1 of size 1024×128 is obtained. After passing through the second PSE module, a point cloud dataset P2 with weighted feature dimensions is obtained. After downsampling and feature extraction by the second VPP module, a point cloud dataset P3 of size 256×512 is obtained. After passing through the second PSE module, a point cloud dataset P4 with weighted feature dimensions is obtained. Point cloud dataset P4 is input to the segmentation module. First, a point cloud coordinate set N of size 1024×3 is obtained from point cloud dataset P2. Interpolation is performed using 1024 points in point cloud coordinate set N as interpolation points. The result is obtained through Unit... After feature extraction using the PointNet network, a point cloud dataset P5 with a size of 1024×128 is obtained. A point cloud coordinate set N1 with a size of 4096×3 is obtained from the disconnector switch point cloud dataset P. The 4096 points in the point cloud coordinate set N1 are used as interpolation points for interpolation. After feature extraction using the Unit PointNet network, a point cloud data point set P6 with a size of 4096×(3+4) is obtained, where 3 represents the coordinate information of the point cloud and 4 represents the number of the four categories into which the point cloud data is divided. Point cloud data set P4 is input to the classification module. After feature extraction through a Pointnet layer, it is then passed through a fully connected layer FC with a size of (512, 128, 4), where 512 is the number of the first hidden layer of FC, 128 is the number of the second hidden layer of FC, and 4 is the number of the output layer of FC (i.e., the number of categories). Finally, the point cloud data is used to obtain information on four categories: disconnector conductive arm, insulating support, transmission cable, and disconnector base. This information is then combined with point cloud data set P6, which is segmented into different components, to obtain point cloud data that is segmented and classified into four components: disconnector conductive arm, insulating support, transmission cable, and disconnector base.

[0083] In addition, the disconnector switch action monitoring model needs to be trained. The specific steps are as follows: First, the disconnector switch point cloud dataset is divided into a training set, a validation set, and a test set in a 6:2:2 ratio for training and validation of the disconnector switch action monitoring model. A batch of disconnector switch 3D point cloud data from the training set is input into the disconnector switch action monitoring model for training. The batch size is set to batch_size = 16, the training epoch is set to 200, and the initial learning rate is 0.001. The cross-entropy loss function is used to calculate the loss value. After training all batches of disconnector switch point cloud data from the training set is complete, the disconnector switch point cloud data from the validation set is input into the disconnector switch action monitoring model in batches to obtain the corresponding batch loss value, batch_loss. During training and validation, the disconnector switch action monitoring model will automatically learn and adjust its parameters based on each loss and batch_loss. When the loss and batch_loss values ​​converge and the difference between them is small, the disconnector switch action monitoring model training ends.

[0084] It should be noted that the cross-entropy loss function mentioned above is a commonly used loss function in deep learning model training. It is a loss function used to measure the difference between two probability distributions.

[0085] According to the above embodiments of the present invention, determining the operating state of the disconnecting switch based at least on the vector corresponding to the first point cloud data includes: according to the formula Calculate the angle α between the conductive arm corresponding to the i-th frame point cloud image and the conductive arm in the closed state. i , where u i =((x0,y0),(x i ,y i (x0, y0) are the coordinates of the end of the conductive arm that is fixed to the stationary end of the insulating support. i ,y i Let be the coordinates of the end of the conductive arm that rotates around the insulating support, v = ((x0, y0), (x...) v ,y v )), (x v ,y v ) represents the coordinates of the end of the conductive arm that rotates around the insulating support when in the closed state, u i ·v is a vector u i The dot product of vector v and vector v, |u i | is vector u i The modulus of |v| is the modulus of vector v; determine α. i Is it greater than α? i+n , in α i Greater than α i+n In the case of α, determine i+nThe corresponding disconnecting switch operates in a closing process, where α i+n Let α be the angle between the conductive arm corresponding to the (i+n)th frame point cloud image and the conductive arm in the closed state, where n is greater than 10; i Less than α i+n In the case of α, determine i The corresponding disconnecting switch is in the opening process; in α i equal to α i+n In the case of α, determine i and α i+n Does α satisfy? i+n =α i =0, in α i+n =α i When α = 0, determine α i The corresponding disconnector switch is in the closed state; in α i+n =α i In the case that α ≠ 0, determine α i The corresponding disconnect switch is in the open state.

[0086] Figure 6 This is a schematic diagram showing the coordinates of the two ends of the conductive arm of the disconnector according to an embodiment of the present invention, as shown below. Figure 6 As shown, when the double-column horizontal rotary disconnector is in the opening or closing process, the conductive arm will only translate within a single plane. The input is a point cloud video containing continuous point cloud images of the disconnector (a point cloud video of the disconnector point cloud images can be acquired in real time using LiDAR). Each frame of the point cloud image in the video is input into the disconnector action monitoring model in chronological order for processing, resulting in a point cloud dataset G = {g1, g2, ..., g...} of the conductive arm of the disconnector in each frame. m}, g1 represents the point cloud data of the disconnector switch's conductive arm in the first frame point cloud image, and m represents the number of frames contained in the point cloud video. The coordinates (x0, y0) and (x...) of the two ends of the conductive arm can be obtained from the point cloud data of the disconnector switch's conductive arm. i ,y i ), where i represents the conductive arm of the disconnector corresponding to the i-th frame of the point cloud image. Furthermore, the vector u of the conductive arm of the disconnector corresponding to the i-th frame of the point cloud image in the x,y coordinate plane can be determined using the above coordinates. i :u i =((x0,y0),(x i ,y i Let the vector v of the disconnecting switch conductive arm in the closed state be: v = ((x0, y0), (x v ,y v The vector u is calculated using the formula described above. iThe angle between the contact arm and vector v can be used to obtain the angle α between the contact arm of the disconnector switch in the i-th frame point cloud image and the contact arm of the disconnector switch in the closed state. i Compare the included angle α corresponding to the i-th frame of the point cloud image. i The angle α between the i+10th frame point cloud image and the corresponding point cloud image i+10 If α i+10 Greater than α i At this time, the disconnecting switch is in the process of opening; if α i+10 Less than α i Then, at this time, the isolation switch corresponding to the point cloud image of the (i+10th)th frame is in the closing process. If α i+10 =α i If =0, then the conductive arm of the disconnecting switch corresponding to the (i+10)th frame point cloud image coincides with the conductive arm when it is in the closed state. At this time, the disconnecting switch corresponding to the (i+10)th frame point cloud image is in the closed state; otherwise, it is in the open state.

[0087] It should be noted that the frame rate of the point cloud images acquired by the aforementioned lidar is approximately 10 to 20 frames per second. Therefore, comparing point cloud images with a difference of 10 frames per second makes the displacement of the conductive arm more obvious. The aforementioned disconnecting switch is in the closed state when it is energized and in the open state when it is de-energized. The disconnecting switch is in the closing process when it changes from the de-energized state to the closed energized state, and in the opening process when it changes from the closed energized state to the de-energized state.

[0088] This embodiment is illustrated using a point cloud video containing 100 consecutive frames of 3D point cloud images of a disconnector switch as an example. The 3D point cloud images of the disconnector switch are sequentially input into the disconnector switch action monitoring model in chronological order. After processing the 3D point cloud images, the disconnector switch action monitoring model outputs the point cloud data of the disconnector switch's conductive arm and its corresponding vector u. i Calculate vector u i The angle α between the vector v of the disconnecting switch conductive arm and the set closed state. i .

[0089] Figure 7 This is a schematic diagram of the conductive arm of the disconnecting switch according to an embodiment of the present invention, as shown below. Figure 7 As shown, the straight line is an abstract representation of the conductive arm of the disconnector switch from a top-down view. This is compared to the vector u corresponding to the point cloud image in frame 10. 10 =((x0,y0),(x 10 ,y 10 The angle α between )) and v 10 The vector u corresponding to the point cloud image of frame 20 20 =((x0,y0),(x 20 ,y 20The angle α between )) and v 20 At this time α 20 Greater than α 10 The disconnect switch is in the process of opening; compare the vector u corresponding to the point cloud image in frame 50. 50 =((x0,y0),(x 50 ,y 50 The angle α between )) and v 50 The vector u corresponding to the point cloud image of frame 60 60 =((x0,y0),(x 60 ,y 60 The angle α between )) and v 60 At this time α 60 =α 50 , and α 60 The value is not 0, so the disconnect switch is in the open state.

[0090] Figure 8 This is a schematic diagram of the disconnector switch action monitoring model structure according to an embodiment of the present invention, as shown below. Figure 8 As shown, point cloud features are extracted by stacking VPP and PSE modules, and then the classification module ST and segmentation module CF are used to classify and segment the point cloud data of the disconnector switch to realize the identification of the point cloud region of the disconnector switch conductive arm. Among them, the segmentation module ST extracts features through interpolation upsampling Interpolate and Unit Pointnet network to realize the point cloud segmentation function, and the classification module CF realizes the point cloud classification function through a point net layer Pointnet and a fully connected layer FC.

[0091] It should be noted that the above-mentioned disconnector operation monitoring model is used to identify and classify the information of the disconnector's conductive arm and determine the operating status of the disconnector.

[0092] Therefore, the technical solution provided by the above embodiments of the present invention solves the following problems: 1) Traditional detection of the action status of disconnecting switches relies on manual inspection, which is inefficient and poses certain safety hazards to inspection personnel; 2) Judging the action status of disconnecting switches through image processing technology has limited information levels and often requires complex image processing models and high computing resources. It also has the following beneficial effects: 1) A voxel farthest point sampling method is designed, which divides the point cloud space into multiple voxel spaces before performing farthest point downsampling, preserving local feature information during downsampling; 2) A weighted mean grouping method is designed, in which, when it is necessary to copy neighboring feature points during grouping, the neighboring feature points are weighted and mean-weighted according to their distance before copying; 3) A point cloud channel attention mechanism module (PSE) is designed, which obtains the weight value of the feature dimension of each point cloud data point, and then weights the feature dimensions of the point cloud data points according to this weight value, increasing the weight of feature dimensions useful for the current task and decreasing the weight of feature dimensions not useful for the current task.

[0093] That is, the above-mentioned technical solution provided by the embodiments of the present invention takes into account a variety of factors and uses LiDAR to obtain the three-dimensional point cloud data of the disconnector. The three-dimensional point cloud data contains the three-dimensional coordinate information of the disconnector, with rich information layers and simple data structure, which is more efficient and intuitive for real-time judgment of the disconnector action information.

[0094] Furthermore, it should be noted that this invention discloses a method for monitoring the action of a disconnector switch based on laser point clouds. A Local Feature Extraction (VPP) module is designed, employing a voxel farthest point sampling method and a weighted mean grouping method to better preserve local feature information of the disconnector switch's conductive arm while extracting point cloud features. A Feature Dimension Attention (PSE) module suitable for point cloud data is designed to obtain the weight value of the feature dimension of each point cloud data point, and then perform weighted calculations on the feature dimensions of each point cloud data point based on this weight value, increasing the weight of feature dimensions useful for the current task and decreasing the weight of feature dimensions less useful for the current task. A method for determining the open / closed state of the disconnector switch is designed to accurately determine the action state of the disconnector switch. By calculating the vector corresponding to the point cloud information of the disconnector switch's conductive arm and calculating the angle between it and the vector of the disconnector switch's conductive arm in the set closed state, the angle change in consecutive frames is compared to determine the action state of the disconnector switch. This method can accurately determine the action state of the disconnector switch in real time, solving the problems of insufficient monitoring efficiency and high manual costs, and achieving real-time automated monitoring of the disconnector switch's action state.

[0095] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0097] According to an embodiment of the present invention, a device for detecting the operating state of a disconnecting switch for implementing the above-described method for detecting the operating state of a disconnecting switch is also provided. Figure 9 This is a schematic diagram of a detection device for the operating state of a disconnecting switch according to an embodiment of the present invention, as shown below. Figure 9 As shown, the device includes: an acquisition module 1001, a first processing module 1003, an input module 1005, and a second processing module 1007. The following describes the device for detecting the operating status of the disconnecting switch.

[0098] The acquisition module 1001 is used to acquire a point cloud dataset. The point cloud dataset includes multiple point cloud data corresponding to the disconnecting switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnecting switch to be detected by LiDAR. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data, and the other part is training data. The disconnecting switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base.

[0099] The first processing module 1003 is used to construct an initial detection model and train the initial detection model using a target loss function to obtain an object detection model. The object detection model includes a feature extraction module, a classification module, and a segmentation module. The input of the feature extraction module is the input of the object detection model, and the output of the feature extraction module is connected to the input of the classification module and the input of the segmentation module, respectively. The output of the classification module and the output of the segmentation module are the outputs of the object detection model. The feature extraction module includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract point cloud features from the input data of the object detection model. The classification module is used to classify the output data of the feature extraction module, and the segmentation module is used to segment the output data of the feature extraction module.

[0100] The input module 1005 is used to input test data into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the power transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector.

[0101] The second processing module 1007 is used to obtain the vector corresponding to the first point cloud data, and at least based on the vector corresponding to the first point cloud data, determine the action state of the disconnecting switch. The action state includes the open state, the closed state, the disconnection process, and the closing process. The open state is the state of the disconnecting switch when it is not energized, and the closed state is the state of the disconnecting switch when it is energized.

[0102] It should be noted that the above-mentioned acquisition module 1001, first processing module 1003, input module 1005 and second processing module 1007 correspond to steps S202 to S208 in the above embodiments. The four modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in the above embodiments.

[0103] As can be seen from the above, in the scheme described in the above embodiments of the present invention, a point cloud dataset can be acquired first using an acquisition module. The point cloud dataset includes multiple point cloud data corresponding to the disconnector switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnector switch to be detected using a lidar. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data, and the other part is training data. The disconnector switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base. Next, an initial detection model can be constructed using a first processing module, and the initial detection model can be trained using a target loss function to obtain a target detection model. The target detection model includes a feature extraction module, a classification module, and a segmentation module. The input end of the feature extraction module is the input end of the target detection model. The output end of the feature extraction module is connected to the input end of the classification module and the input end of the segmentation module, respectively. The output end of the classification module and the output end of the segmentation module are the output ends of the target detection model. The feature extraction module includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract the point cloud features of the input data of the target detection model. The classification module is used to process the input data of the feature extraction module. The output data is classified and processed. The segmentation module is used to segment the output data of the feature extraction module. Then, the input module is used to input the test data into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the power transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector. Finally, the second processing module can be used to obtain the vector corresponding to the first point cloud data. Based on the vector corresponding to the first point cloud data, the action state of the disconnector is determined. The action state includes the open state, the closed state, the disconnection process, and the closing process. The open state is the state of the disconnector when it is not energized, and the closed state is the state of the disconnector when it is energized. This achieves the technical effect of using LiDAR to obtain the three-dimensional point cloud data of the disconnector, combined with voxel farthest point sampling and weighted mean grouping technology. At the same time, the point cloud channel attention mechanism module is used to improve the weight of the feature dimensions useful for the current task, thereby realizing the technical effect of real-time automated monitoring of the action state of the disconnector and improving the accuracy and efficiency of monitoring.

[0104] The technical solution provided by this invention solves the technical problems of low efficiency and safety hazards of manual inspection in related technologies, and the need for complex models and high computing resources in image processing technology.

[0105] In one optional embodiment, the acquisition module includes: a scanning processing unit, used to scan the disconnector switch using a lidar to obtain multiple initial point cloud data; and a filtering and denoising processing unit, used to filter and denoise the multiple initial point cloud data respectively to obtain a point cloud dataset.

[0106] In one optional embodiment, the input module includes: a first input unit for inputting test data into a first local feature extraction submodule to perform first downsampling and first feature extraction processing on the test data, obtaining multiple first point cloud data points; a second input unit for sequentially inputting the multiple first point cloud data points into a first channel attention submodule to perform first feature dimension weighting processing on the multiple first point cloud data points, obtaining multiple second point cloud data points; a third input unit for inputting the multiple second point cloud data points into a second local feature extraction submodule to perform second downsampling and second feature extraction processing on the multiple second point cloud data points, obtaining multiple third point cloud data points; and a fourth input unit for sequentially inputting the multiple third point cloud data points into a second channel. The attention submodule performs weighted processing on multiple third point cloud data points using the second feature dimension to obtain multiple fourth point cloud data points; the fifth input unit is used to input multiple test data, multiple second point cloud data points, and multiple fourth point cloud data points into the segmentation module to segment the multiple fourth point cloud data points respectively to obtain multiple fifth point cloud data points, the number of fifth point cloud data points being the same as the number of point cloud data in the test data; the sixth input unit is used to input the multiple fourth point cloud data points into the classification module to classify the multiple fourth point cloud data points respectively to obtain multiple sixth point cloud data points, the multiple fifth point cloud data points and the multiple sixth point cloud data points forming the first point cloud data, the second point cloud data, the third point cloud data and the fourth point cloud data.

[0107] In one optional embodiment, the first input unit includes: a first input subunit, configured to input test data into a downsampling layer (VFS) to perform voxel-farthest-point downsampling on the test data, obtaining a first sampling point set, the first sampling point set including multiple first sampling points; a second input subunit, configured to input the test data and the first sampling point set into a weighted mean grouping layer (PAG) to group the data with the first sampling points as centroids, obtaining a second sampling point set, the second sampling point set being a local feature point set corresponding to each first sampling point, the second sampling point set including multiple second sampling points; a third input subunit, configured to input the second sampling point set into a local feature point set into a point network layer to perform feature extraction processing on the multiple second sampling points respectively, obtaining multiple first point cloud data points; and a fourth input subunit, configured to input the multiple second point cloud data points into a second local feature extraction submodule. VPP performs second downsampling and second feature extraction on multiple second point cloud data points to obtain multiple third point cloud data points. It includes: a fifth input subunit, used to sequentially input multiple second point cloud data points into a downsampling layer to perform voxel-farthest point downsampling on the second point cloud data points, obtaining a third sampling point set, which includes multiple third sampling points; a sixth input subunit, used to input the multiple second point cloud data points and the third sampling point set into a weighted mean grouping layer to group the data points with the third sampling points as centroids, obtaining a fourth sampling point set, which is a set of local feature points corresponding to each third sampling point, including multiple fourth sampling points; and a seventh input subunit, used to input the fourth sampling point set into a local feature point set and input it into a point network layer to perform feature extraction on the multiple fourth sampling points, obtaining multiple third point cloud data points.

[0108] In an optional embodiment, the second input unit includes: an eighth input subunit, configured to sequentially input multiple first point cloud data points to an activation function layer to calculate the weights of the feature dimensions of each first point cloud data point, thereby obtaining multiple first weight values; a ninth input subunit, configured to input the multiple first weight values ​​and the corresponding first point cloud data points to a computation layer to perform dot product processing on the first weight values ​​and the corresponding first point cloud data points, thereby obtaining multiple second point cloud data points; a tenth input subunit, configured to sequentially input multiple third point cloud data points to a second channel attention submodule to perform second feature dimension weighting processing on the multiple third point cloud data points, thereby obtaining multiple fourth point cloud data points, including: sequentially inputting the multiple third point cloud data points to an activation function layer to calculate the weights of the feature dimensions of each third point cloud data point, thereby obtaining multiple second weight values; and an eleventh input subunit, configured to input the multiple second weight values ​​and the corresponding third point cloud data points to a computation layer to perform dot product processing on the second weight values ​​and the corresponding third point cloud data points, thereby obtaining multiple fourth point cloud data points.

[0109] In an optional embodiment, the fifth input unit includes: a twelfth input subunit, configured to input multiple second point cloud data points and multiple fourth point cloud data points into a first interpolation layer, using the second point cloud data points as interpolation points and the fourth point cloud data points as nearest neighbor feature points, to perform a first interpolation process on the second point cloud data points and the fourth point cloud data points to obtain a first data point set; a thirteenth input subunit, configured to input the first data point set into a first convolutional layer to perform feature extraction processing on the first data point set to obtain a second data point set, the second data point set including multiple second data points; a fourteenth input subunit, configured to input the second data point set and test data into a second interpolation layer, using the test data as interpolation points and the second data points as nearest neighbor feature points, to perform a second interpolation process on the second data point set and the test data to obtain a third data point set, the third data point set including multiple third data points; and a fifteenth input subunit, configured to input the third data point set into a second convolutional layer to perform feature extraction processing on the multiple third data points to obtain multiple fifth point cloud data points.

[0110] In an optional embodiment, the sixth input unit includes: a sixteenth input subunit, used to sequentially input multiple fourth point cloud data points into the third convolutional layer to perform feature extraction processing on the multiple fourth point cloud data points respectively, to obtain multiple seventh point cloud data points; and a seventeenth input subunit, used to input multiple seventh point cloud data points into the fully connected layer to perform classification processing on the seventh point cloud data points, to obtain multiple sixth point cloud data points.

[0111] In one alternative embodiment, the second processing module includes: a calculation unit, configured to calculate according to the formula... Calculate the angle α between the conductive arm corresponding to the i-th frame point cloud image and the conductive arm in the closed state. i , where u i =((x0,y0),(x i ,y i (x0, y0) are the coordinates of the end of the conductive arm that is fixed to the stationary end of the insulating support. i y i Let be the coordinates of the end of the conductive arm that rotates around the insulating support, v = ((x0, y0), (x...) v y v )), (x v y v ) represents the coordinates of the end of the conductive arm that rotates around the insulating support when in the closed state, u i ·v is a vector u i The dot product of vector v and vector v, |u i | is vector u i The modulus, |v| is the modulus of vector v; the first determining unit is used to determine α. i Is it greater than α? i+n, in α i Greater than α i+n In the case of α, determine i+n The corresponding disconnecting switch operates in a closing process, where α i+n Let α be the angle between the conductive arm corresponding to the (i+n)th frame point cloud image and the conductive arm in the closed state, where n is greater than 10; the second determining unit is used to determine α i Less than α i+n In the case of α, determine i The corresponding disconnecting switch is in the disconnection process; the third determining unit is used to determine α i equal to α i+n In the case of α, determine i and α i+n Does α satisfy? i+n =α i =0, in α i+n =α i When α = 0, determine α i The corresponding disconnector switch is in the closed state; the fourth determining unit is used to determine α i+n =α i In the case that α ≠ 0, determine α i The corresponding disconnect switch is in the open state.

[0112] According to another aspect of the present invention, a computer program product is also provided, including computer instructions, wherein when the computer instructions are executed by a processor, a method for detecting the operating state of a disconnecting switch as described above is performed. According to another aspect of the present invention, a processor is also provided, the processor being configured to run a program, wherein the program, during runtime, executes a method for detecting the operating state of a disconnecting switch as described above.

[0113] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored program, wherein a method for detecting the operation state of an isolating switch that controls the device where the computer-readable storage medium is located to perform any of the above-described actions is provided when the program is running.

[0114] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting the operating state of a disconnecting switch, characterized in that, include: A point cloud dataset is obtained, which includes multiple point cloud data corresponding to the disconnect switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnect switch to be detected by LiDAR. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data and the other part is training data. The disconnect switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base. An initial detection model is constructed and trained using a target loss function to obtain an object detection model. The object detection model includes a feature extraction module, a classification module, and a segmentation module. The input of the feature extraction module is the input of the object detection model, and the output of the feature extraction module is connected to the inputs of the classification module and the segmentation module, respectively. The outputs of the classification module and the segmentation module are the outputs of the object detection model. The feature extraction module includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract point cloud features from the input data of the object detection model. The classification module is used to classify the output data of the feature extraction module, and the segmentation module is used to segment the output data of the feature extraction module. The test data is input into the target detection model to obtain the first point cloud data corresponding to the conductive arm of the disconnector, the second point cloud data corresponding to the insulating support of the disconnector, the third point cloud data corresponding to the power transmission cable of the disconnector, and the fourth point cloud data corresponding to the base of the disconnector. Obtain the vector corresponding to the first point cloud data. Based on at least the vector corresponding to the first point cloud data, determine the operating state of the disconnecting switch. The operating state includes an open state, a closed state, a disconnection process, and a closing process. The open state is the state of the disconnecting switch when it is not energized, and the closed state is the state of the disconnecting switch when it is energized. The feature extraction module includes a first local feature extraction submodule, a first channel attention submodule, a second local feature extraction submodule, and a second channel attention submodule connected in sequence. The test data is input into the target detection model to obtain first point cloud data corresponding to the conductive arm of the disconnector, second point cloud data corresponding to the insulating support of the disconnector, third point cloud data corresponding to the power transmission cable of the disconnector, and fourth point cloud data corresponding to the base of the disconnector. The test data is input into the first local feature extraction submodule to perform first downsampling and first feature extraction processing on the test data, thereby obtaining multiple first point cloud data points; Multiple first point cloud data points are sequentially input into the first channel attention submodule to perform weighted processing on the multiple first point cloud data points according to the first feature dimension, thereby obtaining multiple second point cloud data points; Multiple second point cloud data points are input into the second local feature extraction submodule to perform second downsampling and second feature extraction processing on the multiple second point cloud data points to obtain multiple third point cloud data points; Multiple third point cloud data points are sequentially input into the second channel attention submodule to perform second feature dimension weighting processing on the multiple third point cloud data points to obtain multiple fourth point cloud data points; Multiple test data, multiple second point cloud data points, and multiple fourth point cloud data points are input into the segmentation module to segment the multiple fourth point cloud data points respectively, resulting in multiple fifth point cloud data points. The number of fifth point cloud data points is the same as the number of point cloud data in the test data. Multiple fourth point cloud data points are input to the classification module to classify them respectively, resulting in multiple sixth point cloud data points. The multiple fifth point cloud data points and the multiple sixth point cloud data points form the first point cloud data, the second point cloud data, the third point cloud data, and the fourth point cloud data. The first local feature extraction submodule and the second local feature extraction submodule are the same. The first local feature extraction submodule and the second local feature extraction submodule each include a downsampling layer, a weighted mean grouping layer, and a local feature point set input point network layer. The test data is input to the first local feature extraction submodule to perform first downsampling and first feature extraction processing on the test data, resulting in multiple first point cloud data points, including: The test data is input into the downsampling layer to perform voxel farthest point downsampling processing on the test data to obtain a first sampling point set, which includes multiple first sampling points; The test data and the first set of sampling points are input into the weighted mean grouping layer and grouped with the first sampling point as the centroid to obtain the second set of sampling points. The second set of sampling points is the set of local feature points corresponding to each first sampling point. The second set of sampling points includes multiple second sampling points. The second set of sampling points is input into the local feature point set input point network layer to perform feature extraction processing on multiple second sampling points respectively, thereby obtaining multiple first point cloud data points; Multiple second point cloud data points are input into the second local feature extraction submodule to perform second downsampling and second feature extraction processing on the multiple second point cloud data points, resulting in multiple third point cloud data points, including: Multiple second point cloud data points are sequentially input into the downsampling layer to perform voxel farthest point downsampling processing on the second point cloud data points to obtain a third sampling point set, which includes multiple third sampling points. Multiple second point cloud data points and the third sampling point set are input into the weighted mean grouping layer and grouped with the third sampling point as the centroid to obtain a fourth sampling point set. The fourth sampling point set is the local feature point set corresponding to each third sampling point, and the fourth sampling point set includes multiple fourth sampling points. The fourth set of sampling points is input into the local feature point set input point network layer to perform feature extraction processing on multiple fourth sampling points, thereby obtaining multiple third point cloud data points. The first channel attention submodule and the second channel attention submodule are identical. The first channel attention submodule and the second channel attention submodule each include an activation function layer and a computation layer. Multiple first point cloud data points are sequentially input into the first channel attention submodule to perform weighted processing based on a first feature dimension, resulting in multiple second point cloud data points, including: Multiple first point cloud data points are sequentially input into the activation function layer to calculate the weight of the feature dimension of each first point cloud data point, thereby obtaining multiple first weight values. Multiple first weight values ​​and corresponding first point cloud data points are input to the computing layer to perform dot multiplication on the first weight values ​​and corresponding first point cloud data points respectively, so as to obtain multiple second point cloud data points. The process involves sequentially inputting multiple third point cloud data points into the second channel attention submodule to perform second feature dimension weighting on the multiple third point cloud data points to obtain multiple fourth point cloud data points, including: sequentially inputting multiple third point cloud data points into the activation function layer to calculate the weight of the feature dimension of each third point cloud data point to obtain multiple second weight values. Multiple second weight values ​​and corresponding third point cloud data points are input to the computing layer to perform dot product processing on the second weight values ​​and corresponding third point cloud data points respectively, so as to obtain multiple fourth point cloud data points.

2. The method according to claim 1, characterized in that, Obtain the dataset, including: Multiple initial point cloud data are obtained by scanning the disconnect switch using a lidar. The initial point cloud data are filtered and denoised to obtain the point cloud dataset.

3. The method according to claim 1, characterized in that, The segmentation module includes a first interpolation layer, a first convolutional layer, a second interpolation layer, and a second convolutional layer connected in sequence. Multiple test data points, multiple second point cloud data points, and multiple fourth point cloud data points are input into the segmentation module to segment the multiple fourth point cloud data points respectively, resulting in multiple fifth point cloud data points, including: Multiple second point cloud data points and multiple fourth point cloud data points are input into the first interpolation layer. The second point cloud data points are used as interpolation points, and the fourth point cloud data points are used as nearest neighbor feature points. The first interpolation process is performed on the second point cloud data points and the fourth point cloud data points to obtain the first data point set. The first data point set is input into the first convolutional layer to perform feature extraction processing on the first data point set to obtain a second data point set, the second data point set including multiple second data points; The second data point set and the test data are input into the second interpolation layer. The test data is used as the interpolation point and the second data point is used as the nearest neighbor feature point to perform a second interpolation process on the second data point set and the test data to obtain a third data point set. The third data point set includes multiple third data points. The third data point set is input into the second convolutional layer to perform feature extraction processing on multiple third data points, thereby obtaining multiple fifth point cloud data points.

4. The method according to claim 1, characterized in that, The classification module includes a third convolutional layer and a fully connected layer connected in sequence. Multiple fourth point cloud data points are input into the classification module to classify the multiple fourth point cloud data points respectively, resulting in multiple sixth point cloud data points, including: Multiple fourth point cloud data points are sequentially input into the third convolutional layer to perform feature extraction processing on the multiple fourth point cloud data points respectively, thereby obtaining multiple seventh point cloud data points; Multiple seventh cloud data points are input into the fully connected layer to classify the seventh cloud data points, thereby obtaining multiple sixth cloud data points.

5. The method according to claim 1, characterized in that, The operational state of the disconnecting switch is determined based at least on the vector corresponding to the first point cloud data, including: According to the formula Calculate the angle α between the conductive arm corresponding to the point cloud image in the i-th frame and the conductive arm in the combined state. i , where u i =((x0,y0),(x i ,y i (x0, y0) are the coordinates of the end of the conductive arm that is fixed to the stationary end of the insulating support. i y i Let be the coordinates of the end of the conductive arm that rotates around the insulating post, v = ((x0, y0), (x...) v y v )), (x v y v ) represents the coordinate of the end of the conductive arm that rotates around the insulating support when in the closed state, u i ·v is a vector u i The dot product of vector v and vector v, |u i | is vector u i The modulus of |v| is the modulus of vector v; Determine α i Is it greater than α? i+n , in α i Greater than α i+n In the case of α, determine i+n The corresponding operating state of the disconnecting switch is the closing process, where α i+n The angle between the conductive arm corresponding to the point cloud image in the (i+n)th frame and the conductive arm in the combined state, where n is greater than 10; In α i Less than α i+n In the case of α, determine i The corresponding disconnecting switch's operating state is the disconnection process; In α i equal to α i+n In the case of α, determine i and α i+n Does α satisfy? i+n =α i =0, in α i+n =α i When α = 0, determine α i The corresponding disconnector switch is in the closed state. In α i+n =α i In the case that α ≠ 0, determine α i The corresponding disconnector switch is in the open state.

6. A device for detecting the operating state of a disconnecting switch, characterized in that, include: The acquisition module is used to acquire a point cloud dataset, which includes multiple point cloud data corresponding to the disconnect switch. Each point cloud data includes multiple frames of point cloud images obtained by scanning the disconnect switch to be detected by LiDAR. Each point cloud image includes multiple point cloud data points. Part of the point cloud dataset is test data and the other part is training data. The disconnect switch includes at least a conductive arm, an insulating support, a power transmission cable, and a base. The first processing module is used to construct an initial detection model and train the initial detection model using a target loss function to obtain a target detection model. The target detection model includes a feature extraction module, a classification module, and a segmentation module. The input of the feature extraction module is the input of the target detection model, and the output of the feature extraction module is connected to the inputs of the classification module and the segmentation module, respectively. The outputs of the classification module and the segmentation module are the outputs of the target detection model. The feature extraction module includes a stacked local feature extraction submodule and an attention submodule. The feature extraction module is used to extract point cloud features from the input data of the target detection model. The classification module is used to classify the output data of the feature extraction module, and the segmentation module is used to segment the output data of the feature extraction module. An input module is used to input the test data into the target detection model to obtain first point cloud data corresponding to the conductive arm of the disconnector, second point cloud data corresponding to the insulating support of the disconnector, third point cloud data corresponding to the power transmission cable of the disconnector, and fourth point cloud data corresponding to the base of the disconnector. A second processing module is used to obtain the vector corresponding to the first point cloud data and, based at least on the vector corresponding to the first point cloud data, determine the operating state of the disconnector. The operating state includes an open state, a closed state, a disconnection process, and a closing process. The open state is the state of the disconnector when it is not energized, and the closed state is the state of the disconnector when it is energized. The feature extraction module includes a first local feature extraction submodule, a first channel attention submodule, a second local feature extraction submodule, and a second channel attention submodule connected in sequence. The input module includes: a first input unit, used to input the test data into the first local feature extraction submodule to perform a first downsampling and a first feature extraction process on the test data to obtain multiple first point cloud data points; a second input unit, used to input the multiple first point cloud data points into the first channel attention submodule to perform a first feature dimension weighting process on the multiple first point cloud data points to obtain multiple second point cloud data points; a third input unit, used to input the multiple second point cloud data points into the second local feature extraction submodule to perform a second downsampling and a second feature extraction process on the multiple second point cloud data points to obtain multiple third point cloud data points; and a fourth input unit, used to input the multiple... The third point cloud data points are sequentially input to the second channel attention submodule to perform second feature dimension weighting processing on the multiple third point cloud data points, resulting in multiple fourth point cloud data points; the fifth input unit is used to input multiple test data, multiple second point cloud data points, and multiple fourth point cloud data points to the segmentation module to segment the multiple fourth point cloud data points respectively, resulting in multiple fifth point cloud data points, the number of which is the same as the number of point cloud data in the test data; the sixth input unit is used to input multiple fourth point cloud data points to the classification module to classify the multiple fourth point cloud data points respectively, resulting in multiple sixth point cloud data points. The multiple fifth point cloud data points and the multiple sixth point cloud data points form the first point cloud data, the second point cloud data, the third point cloud data, and the fourth point cloud data. The first local feature extraction submodule and the second local feature extraction submodule are the same. The first local feature extraction submodule and the second local feature extraction submodule respectively include a downsampling layer, a weighted mean grouping layer, and a local feature point set input point network layer. The first input unit includes: a first input subunit, used to input the test data into the downsampling layer to perform voxel farthest point downsampling processing on the test data to obtain a first sampling point set, the first sampling point set including multiple first sampling points; a second input subunit, used to input the test data and the first sampling point set into the weighted mean grouping layer to group the data with the first sampling points as centroids to obtain a second sampling point set, the second sampling point set being a local feature point set corresponding to each first sampling point, the second sampling point set including multiple second sampling points; and a third input subunit, used to input the second sampling point set into the local feature point set input point network layer to perform feature extraction processing on the multiple second sampling points respectively to obtain multiple first point cloud data points. The third input unit includes: a fourth input subunit, used to sequentially input multiple second point cloud data points into the downsampling layer to perform voxel farthest point downsampling processing on the second point cloud data points to obtain a third sampling point set, the third sampling point set including multiple third sampling points; a fifth input subunit, used to input multiple second point cloud data points and the third sampling point set into the weighted mean grouping layer to group the data points with the third sampling points as centroids to obtain a fourth sampling point set, the fourth sampling point set being a local feature point set corresponding to each of the third sampling points, the fourth sampling point set including multiple fourth sampling points; a sixth input subunit, used to input the fourth sampling point set into the local feature point set input network layer to perform feature extraction processing on the multiple fourth sampling points respectively to obtain multiple third point cloud data points; and a seventh input subunit, used to input the fourth sampling point set into the local feature point set input network layer to perform feature extraction processing on the multiple fourth sampling points respectively to obtain multiple third point cloud data points. The first channel attention submodule and the second channel attention submodule are identical. The first channel attention submodule and the second channel attention submodule each include an activation function layer and a computation layer. The second input unit includes: an eighth input subunit, used to sequentially input multiple first point cloud data points to the activation function layer to calculate the weights of the feature dimensions of each first point cloud data point, obtaining multiple first weight values; a ninth input subunit, used to input multiple first weight values ​​and corresponding first point cloud data points to the computation layer to perform dot product processing on the first weight values ​​and corresponding first point cloud data points, obtaining multiple second point cloud data points; and a tenth input subunit, used to sequentially input multiple third point cloud data points to the second channel attention submodule to perform second feature dimension weighting processing on the multiple third point cloud data points, obtaining multiple fourth point cloud data points, including: Multiple third point cloud data points are sequentially input into the activation function layer to calculate the weights of the feature dimensions of each third point cloud data point, thereby obtaining multiple second weight values; the eleventh input subunit is used to input multiple second weight values ​​and the corresponding third point cloud data points into the calculation layer to perform dot product processing on the second weight values ​​and the corresponding third point cloud data points, thereby obtaining multiple fourth point cloud data points.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program executes the method for detecting the operating state of the disconnecting switch according to any one of claims 1 to 5.

Images