Early warning method for external damage sources of power transmission lines based on 4D millimeter wave radar

Through the density point cloud data processing and positive sample identification network analysis of 4D millimeter wave radar, a warning signal for breaking hazard sources outside the transmission line is directly generated, solving the sensor fusion problem in the existing technology and improving monitoring timeliness and accuracy.

CN118759486BActive Publication Date: 2024-12-03JIANGSU HAOHAN INFORMATION TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411237978.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-03
Estimated Expiration
2044-09-05

AI Technical Summary

Technical Problem

In the prior art, 4D millimeter wave radar needs to be fused with various sensors to achieve accurate positioning of dangerous sources of breaking out of power transmission lines, resulting in poor monitoring timeliness and accuracy stability.

Method used

By receiving the density point cloud data set of 4D millimeter wave radar, spatial point cloud positioning is performed, point cloud positioning timing information is generated, anchor points are traversed and identified, positive sample identification network analysis is used to generate positive sample point positioning sequences in the future time zone, when it is When intersecting with the boundary area of ​​the transmission line, an early warning signal is generated.

Benefits of technology

It does not need to be integrated with other sensors, which realizes accurate identification and early warning of dangerous sources outside the transmission line, and improves the timeliness and accuracy of monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118759486B_ABST
    Figure CN118759486B_ABST
Patent Text Reader

Abstract

The present invention discloses a transmission line external damage hazard source early warning method based on 4D millimeter wave radar, which relates to the field of radar early warning technology. The method includes: receiving a density point cloud data set of a 4D millimeter wave radar, performing spatial point cloud positioning, and generating point cloud positioning time sequence information; traversing the point cloud positioning time sequence information to perform anchor point identification and generate an anchor point positioning sequence; obtaining a positive sample point positioning sequence through a positive sample recognition network, and generating a future time zone positive sample point positioning sequence; when the future time zone positive sample point positioning sequence intersects with the transmission line boundary area, generating a transmission line external damage hazard source early warning signal for alarming. The present invention solves the technical problem that the existing technology requires the application of 4D millimeter wave radar to be fused with various sensors to achieve accurate positioning of the target, resulting in poor monitoring timeliness and accuracy stability, and achieves the technical effect of improving monitoring timeliness and accuracy without fusion with various sensors.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of radar early warning technology, and in particular to a method for early warning of external damage hazards of power transmission lines based on 4D millimeter wave radar. Background Art

[0002] With the continuous development of power systems, the safe operation of transmission lines has become crucial. Transmission lines are often threatened by various external damage sources, such as vehicle collisions, construction damage, and foreign objects approaching, which may cause transmission line failures, power outages, and even safety accidents. Traditional transmission line monitoring methods mainly rely on manual inspections, which are inefficient, costly, and difficult to detect potential danger sources in real time.

[0003] At present, the application of 4D millimeter-wave radar on the market still needs to be combined with various sensors to achieve accurate positioning of the target. However, there may be various disturbances in the sensor data, and the algorithm complexity of multi-source data fusion leads to poor timeliness and accuracy stability of monitoring results. Summary of the invention

[0004] The present application provides a power transmission line external damage hazard source early warning method based on 4D millimeter wave radar, which is used to solve the technical problem that the application of 4D millimeter wave radar in the prior art needs to be integrated with various sensors to achieve accurate positioning of the target, resulting in poor monitoring timeliness and accuracy stability.

[0005] In view of the above problems, the present application provides a power transmission line external damage hazard source early warning method based on 4D millimeter wave radar, the method comprising:

[0006] A density point cloud data set of a 4D millimeter wave radar is received, wherein the density point cloud data set includes a distance label data set, a speed label data set, a horizontal angle data set, and a pitch angle data set; the distance label data set, the speed label data set, the horizontal angle data set, and the pitch angle data set are used for spatial point cloud positioning to generate point cloud positioning time series information; the point cloud positioning time series information is traversed to perform anchor point identification to generate an anchor point positioning sequence; through a positive sample recognition network, based on the anchor point positioning sequence, the point cloud positioning time series information is traversed for analysis to obtain a positive sample point positioning sequence; motion simulation is performed according to the positive sample point positioning sequence to generate a future time zone positive sample point positioning sequence; when the future time zone positive sample point positioning sequence intersects with a transmission line boundary area, a transmission line external damage hazard source warning signal is generated to alarm.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Receive the density point cloud data set of 4D millimeter wave radar, perform spatial point cloud positioning, and generate point cloud positioning time sequence information; traverse the point cloud positioning time sequence information to identify anchor points and generate anchor point positioning sequence; traverse the point cloud positioning time sequence information through the positive sample recognition network to analyze and obtain the positive sample point positioning sequence; perform motion simulation based on the positive sample point positioning sequence to generate the future time zone positive sample point positioning sequence; when the future time zone positive sample point positioning sequence intersects with the transmission line boundary area, generate a transmission line external damage hazard source warning signal for alarm. It achieves the technical effect of improving the timeliness and accuracy of monitoring without the need for fusion with various sensors. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a process flow of a power transmission line external damage hazard source early warning method based on 4D millimeter wave radar provided in an embodiment of the present application;

[0011] Figure 2 A schematic diagram of the process of generating a point cloud density data set at the first moment for a method for early warning of a power transmission line external damage hazard source based on a 4D millimeter wave radar provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application provides a power transmission line external damage hazard source early warning method based on 4D millimeter wave radar, which is used to solve the technical problem that the application of 4D millimeter wave radar in the prior art needs to be integrated with various sensors to achieve accurate positioning of the target, resulting in poor monitoring timeliness and accuracy stability.

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0014] Embodiment 1

[0015] like Figure 1 As shown, the present application provides a method for early warning of external damage hazards of power transmission lines based on 4D millimeter wave radar, the method comprising:

[0016] Step S100: receiving a density point cloud data set of a 4D millimeter wave radar, wherein the density point cloud data set includes a distance label data set, a speed label data set, a horizontal angle data set, and a pitch angle data set.

[0017] Specifically, a connection is established with the 4D millimeter-wave radar through a specific data transmission interface and protocol. The 4D millimeter-wave radar detects and senses the surrounding environment by continuously emitting millimeter-wave signals to the surrounding space and receiving the reflected signals. After processing and calculation inside the radar, rich density point cloud data is generated. Among them, the distance label data set records in detail the precise straight-line distance between each detected target object and the radar. These distance values ​​are calculated by measuring the time difference between the transmitted signal and the received echo, combined with the propagation speed of the millimeter wave in the air. The speed label data set contains the movement speed information of the target object relative to the radar. This is determined by analyzing the frequency change of the reflected signal based on the principle of the Doppler effect, which can reflect whether the object is approaching or moving away from the radar, as well as the magnitude of the movement speed. The horizontal angle data set accurately identifies the angular position of the target object relative to the radar in the horizontal direction. It is obtained by scanning and measuring the radar antenna in the horizontal direction to determine the left and right direction of the object in the horizontal plane. The pitch angle data set clearly indicates the angle of the target object relative to the radar in the vertical direction. This is obtained by detecting and measuring the radar antenna in the vertical direction, which helps to understand the up and down position of the object in the vertical plane. These different types of data sets work together to form a comprehensive and detailed density point cloud data set. Combined with data such as distance and angle, the position of an object in three-dimensional space can be determined more accurately, positioning accuracy can be improved, and the position and trajectory of an object can be accurately tracked, enabling earlier detection and early warning, improving monitoring efficiency, and improving the accuracy and reliability of early warnings.

[0018] Step S200: performing spatial point cloud positioning on the distance label data set, the speed label data set, the horizontal angle data set and the pitch angle data set to generate point cloud positioning time series information.

[0019] Specifically, for each distance value in the distance label data set, combined with the corresponding horizontal angle and pitch angle data, it is converted into a coordinate point in three-dimensional space through trigonometric functions and spatial geometry calculations. At the same time, according to the speed information and time interval in the speed label data set, the position change of the object at different times is calculated. Then, each data point is processed in chronological order, and the coordinate points at different times are combined to form a series of spatial point clouds that change over time. During the processing, the synchronization and accuracy of the data are considered, and abnormal or missing data are properly processed and compensated. Finally, the generated point cloud positioning time series information contains the precise spatial position of the object at different time points, presenting the object's motion trajectory and position change trend in an orderly and coherent manner. For example, assuming that at a certain moment, the distance is 10 meters, the horizontal angle is 30 degrees, the pitch angle is 45 degrees, and the speed is 5 meters / second, the specific coordinates (x, y, z) of the object in three-dimensional space at that moment can be obtained by calculation. As time goes by, such calculations and combinations are repeated continuously to form a complete point cloud positioning time series information.

[0020] Step S300: traverse the point cloud positioning time sequence information to identify anchor points and generate an anchor point positioning sequence.

[0021] Specifically, first obtain the point cloud positioning time sequence information collected and processed by the 4D millimeter wave radar. This information contains the time-varying data of the position, speed, horizontal angle and pitch angle of the objects in the monitoring area. Then, start processing these point cloud data one by one. For the point currently being processed, extract all its detailed information including position, speed, angle, etc. The rules for determining whether it is an anchor point will be based on factors related to the safety of the transmission line. For example, if the distance between the object represented by a point and the transmission line is less than a certain safety threshold, then the point is identified as an anchor point. Alternatively, if the speed direction of the object is toward the transmission line and the speed value is large, there is a risk of rapidly approaching the transmission line, and the point will also be marked as an anchor point. For another example, when the movement trajectory of the object shows that it will enter the protection area of ​​the transmission line at a certain point in the future, the corresponding point will also be identified as an anchor point. When identifying an anchor point, a specific identifier will be added to it in the data structure, such as setting a specific field and assigning a specific value to indicate that it is an anchor point. The index of the current processing point is continuously increased, and the above extraction, judgment and identification steps are repeated until all the point cloud positioning time series information is traversed. Finally, all the points identified as anchor points form an anchor point positioning sequence according to their order in the original point cloud positioning time series information. This sequence highlights the key positions and status information of objects that pose a potential threat to the transmission line at different times, providing an important basis for the subsequent accurate judgment of whether there is an external damage hazard source, which helps to improve monitoring efficiency and improve the accuracy and reliability of early warning.

[0022] Step S400: using a positive sample recognition network and based on the anchor point positioning sequence, traversing the point cloud positioning time sequence information for analysis to obtain a positive sample point positioning sequence.

[0023] Specifically, based on the acquired anchor point positioning sequence, the positive sample recognition network is enabled. This positive sample recognition network is pre-trained and has the ability to identify features related to the source of external damage to the transmission line. Traversal starts from the starting position of the point cloud positioning time series information. For each point, the positive sample recognition network will comprehensively consider multiple features such as the position, speed, direction, and relationship with the anchor point of the point. By analyzing these features and matching them with the positive sample feature patterns preset in the network, if the point meets the conditions for being identified as a positive sample (for example, the object represented by the point is moving quickly and directly toward the transmission line), it will be marked as a positive sample point. Continue to perform such analysis and processing on the entire point cloud positioning time series information, and finally filter out all points identified as positive samples, and arrange them in the order of their appearance in the time series information to form a positive sample point positioning sequence. This positive sample point positioning sequence concentrates the location information of objects that are considered to be most likely to cause external damage to the transmission line, providing key data support for subsequent accurate judgment and timely warning.

[0024] Step S500: Perform motion simulation according to the positive sample point positioning sequence to generate a future time zone positive sample point positioning sequence.

[0025] Specifically, in the process of early warning of the external damage source of the transmission line, firstly, according to the existing positive sample point positioning sequence, the speed direction sequence and speed magnitude sequence corresponding to each point are further extracted. For the speed direction sequence, it is decomposed into a horizontal direction sequence and a vertical direction sequence. Then, the horizontal direction sequence and the vertical direction sequence are subjected to in-depth feature analysis respectively. In the horizontal direction, by analyzing its changing rules, trends and the relationship with other factors, the predicted horizontal direction feature is obtained, which can reflect the tendency and possible range of the future movement of the object in the horizontal direction. Similarly, the vertical direction sequence is processed similarly to obtain the predicted vertical direction feature. For the speed magnitude sequence, feature analysis is also performed to obtain the predicted motion speed feature to understand the changing trend and possible value range of the object's motion speed. Finally, these predicted horizontal direction features, predicted vertical direction features and predicted motion speed features are comprehensively used to accurately predict the positive sample point positioning information at the end time in the positive sample point positioning sequence. That is to say, based on the current motion state and features, the possible locations of these positive sample points in the future time zone are inferred, and these predicted positions are arranged in chronological order, thereby generating the future time zone positive sample point positioning sequence. The future time zone positive sample point positioning sequence generated in this way can provide key forward-looking information for judging whether potential danger sources will threaten transmission lines, so as to take corresponding early warning and prevention measures in time and improve the accuracy and reliability of early warning.

[0026] Step S600: When the future time zone positive sample point positioning sequence intersects with the transmission line boundary area, a transmission line external damage hazard source warning signal is generated to issue an alarm.

[0027] Specifically, after the future time zone positive sample point positioning sequence has been generated, the position information in this sequence will be compared and judged with the position information of the transmission line boundary area. If some of the predicted positions in the future time zone positive sample point positioning sequence overlap or intersect with the positions of the transmission line boundary area, this means that potential dangerous sources may invade the safety range of the transmission line in the future and cause damage to the transmission line. Once such an intersection is detected, an early warning signal of the transmission line external damage dangerous source is immediately generated. This early warning signal may contain key information such as the location of the dangerous object, the direction of movement, and the expected time to reach the transmission line boundary. Subsequently, the alarm mechanism will be triggered. The alarm can be in various forms, such as issuing a loud alarm, displaying eye-catching warning signs and text on the monitoring screen, and sending emergency notifications to the mobile devices of relevant staff, so as to ensure that relevant personnel can receive the danger warning in time and take corresponding measures to avoid the damage to the transmission line and ensure the safety and stability of power transmission. Without the need to fuse with other sensors, accurate identification and early warning of external damage dangerous sources around the transmission line can be achieved, thereby improving the timeliness and accuracy of monitoring.

[0028] In a possible implementation, step S200 further includes:

[0029] Step S210: extracting first distance label data, first speed label data, first horizontal angle data and first pitch angle data at a first moment according to the distance label data set, the speed label data set, the horizontal angle data set and the pitch angle data set.

[0030] Step S220: Taking the 4D millimeter-wave radar position as the origin position, positioning is performed based on the first distance label data, the first speed label data, the first horizontal angle data and the first pitch angle data to obtain the first point positioning information at the first moment, and add it to the point cloud positioning information at the first moment.

[0031] Step S230: adding the point cloud positioning information at the first moment into the point cloud positioning timing information.

[0032] Specifically, the distance label data set, speed label data set, horizontal angle data set and pitch angle data set are defined. For the first moment, the corresponding value is accurately extracted from the distance label data set, which is the first distance label data. It represents the straight-line distance between the target object and the determined origin (usually the position of the 4D millimeter wave radar). The speed value at the first moment is extracted from the speed label data set, that is, the first speed label data, which is used for subsequent analysis of the change in the motion state of the object. The angle value at the first moment is obtained from the horizontal angle data set to obtain the first horizontal angle data, which determines the orientation of the object relative to the origin in the horizontal direction. The angle value at the first moment is extracted from the pitch angle data set, which is the first pitch angle data, which indicates the inclination angle of the object relative to the origin in the vertical direction. Once these data are determined, further calculations can be performed to achieve the specific positioning of the point. Given the Euclidean distance (first distance label data) and height (calculated by the first pitch angle data and the first distance label data) of the known point and the origin, according to the Pythagorean theorem, the square of the Euclidean distance minus the square of the height is used, and then the square root operation is performed to obtain the horizontal distance. Combined with the determined horizontal direction (first horizontal angle data), the specific position of the object in space can be accurately determined, that is, the first point positioning information at the first moment is obtained. Then, this precise first point positioning information is added to the point cloud positioning information at the first moment. In this way, the point cloud positioning information at the first moment is richer and more accurate, and can more comprehensively describe the position of the object in space at the first moment. This provides key and accurate basic data for subsequent analysis work such as whether the object will approach the transmission line and whether it will constitute a source of external damage. Such precise positioning is crucial to judging whether the object will pose a threat to the transmission line, and provides key basic data for subsequent early warning analysis.

[0033] After completing the acquisition and determination of the point cloud positioning information at the first moment, this step adds the information to the point cloud positioning time series information. The point cloud positioning time series information is a collection of object positioning data at each moment in chronological order. Adding the point cloud positioning information at the first moment means that in this ordered collection, a suitable position is found and inserted for the positioning data at the first moment. In this way, the point cloud positioning time series information can be continuously enriched and improved, including the object position information at more moments, forming a continuous, time-varying object position record sequence. This is of great significance for the subsequent comprehensive and dynamic analysis of the object's motion trajectory, judging its relationship with the transmission line, and issuing timely and accurate warnings of external damage hazards. By combining the distance, speed, horizontal angle and pitch angle data of the 4D millimeter wave radar for accurate point cloud positioning, there is no need to rely on data fusion of other sensors, which effectively improves the accuracy of the point cloud positioning time series information and ensures the timeliness of monitoring.

[0034] In a possible implementation, step S300 further includes:

[0035] Step S310: extracting the point cloud positioning information at the first moment according to the point cloud positioning timing information.

[0036] Step S320: traverse the point cloud positioning information at the first moment to perform density analysis to generate a point cloud density data set at the first moment.

[0037] Step S330: According to the maximum value of the point cloud density data set at the first moment, the point cloud anchor point positioning information at the first moment is configured and added into the anchor point positioning sequence.

[0038] Specifically, the point cloud positioning timing information is like a detailed record table arranged in chronological order, in which each row records the relevant positioning information of objects at different moments. First, clarify the specific time mark of the first moment. Then, check the record table of the point cloud positioning timing information line by line. Each line has a corresponding time tag, and compare these time tags with the time mark of the first moment. When the row of data with a time tag that matches the first moment is found, all the detailed information related to positioning contained in this row will be extracted, such as the precise position coordinates of the object at the first moment, the magnitude and direction of the movement speed, the horizontal and pitch angles, etc. These extracted complete and accurate information constitute the point cloud positioning information at the first moment, which is ready for the subsequent in-depth analysis and processing of this specific moment.

[0039] The point cloud positioning information at the first moment has been obtained. Next, we need to conduct a comprehensive traversal and analysis of this information. The various position points in the point cloud positioning information at the first moment are regarded as points in space. Then, density analysis is performed by calculating the distribution density of these points within a certain spatial range. A fixed-size spatial region can be set to count the number of points in this region. Such statistics and calculations are performed on different spatial regions in turn. Combining the statistical results of the number of points in each spatial region, a data set that can reflect the density of the point cloud at different spatial positions at the first moment is generated. This is the point cloud density data set at the first moment. This data set can help understand the degree of aggregation of objects at different positions at the first moment, which is of great significance for judging the movement trend of objects and possible dangerous areas, especially in the early warning of external damage hazards of transmission lines, which can assist in determining locations where high risks may exist.

[0040] Find the maximum value from the first-moment point cloud density data set generated earlier. The position corresponding to this maximum value indicates the area with the most densely distributed point cloud at the first moment. Then, according to the position indicated by this maximum value, configure the first-moment point cloud anchor point positioning information. This anchor point positioning information usually contains the precise location coordinates of the dense area and related feature descriptions. In the anchor point identification process, the key areas are accurately located by analyzing the point cloud density without relying on data from additional sensors. The generated anchor point positioning sequence can improve the accuracy of hazard source identification and thus enhance the timeliness of monitoring.

[0041] In one possible implementation, Figure 2 As shown, step S320 also includes:

[0042] Step S321: extracting first point positioning information according to the point cloud positioning information at the first moment.

[0043] Step S322: taking the first point positioning information as a reference point, sorting a preset number of point positioning information from near to far from the point cloud positioning information at the first moment, setting a first point local domain, and adding the first point local domain to the local domain set.

[0044] Step S323: when the first point cloud positioning information is positioned, extract the second point local domain until the point in the Nth point local domain having the first point positioning information is set as the first point reverse local domain.

[0045] Step S324: traverse the points of the first point local domain and the first point local domain, calculate the Euclidean distance with the first point positioning information, set it as the first point distribution sparse factor, and add a distribution sparse factor set.

[0046] Step S325: Calculate the ratio of the distribution sparse factor set mean to the first point distribution sparse factor, set it as the first point distribution density, and add it to the point cloud density data set at the first moment.

[0047] Specifically, in the point cloud positioning information at the first moment, this contains a lot of detailed data about the location of the object. At this time, it is necessary to select a specific point as the starting point for subsequent analysis. The positioning information of this selected point is the first point positioning information. The selection method may be random or based on some preset rules, such as selecting the point closest to the 4D millimeter-wave radar, or a point at a specific position (such as the center position) in the point cloud distribution. By explicitly extracting this first point positioning information, the foundation is laid for subsequent operations such as local domain sorting, distance calculation, and density analysis around this point.

[0048] The first point positioning information has been determined. Next, use this first point positioning information as a reference benchmark. Then, sort the point cloud positioning information at the first moment according to the distance from this reference point. The distance calculation usually uses some distance metric, such as Euclidean distance. Starting from the closest point, select a preset number of point positioning information. These selected points and the first point positioning information together constitute a specific area, namely the first point local domain. Finally, add this first point local domain to the local domain set. This local domain set will continuously accumulate local domains corresponding to different reference points, providing a data basis for a more comprehensive analysis of the distribution characteristics of the point cloud in the future.

[0049] After completing the positioning operation as described in step S322 for all points in the point cloud positioning information at the first moment, it means that we have determined the corresponding local domain for each point. Next, we start to filter from these determined local domains. First, find the local domain of the second point, and then check the subsequent local domains of the third point, the fourth point, and so on. When checking each local domain, check whether it contains the previously determined first point positioning information. If it does, extract this local domain. Define all local domains containing the first point positioning information as the reverse local domain of the first point. Through this reverse screening method, the distribution and influence of the first point positioning information in the entire first moment point cloud positioning can be more comprehensively understood. For example, by observing the range, number and distribution pattern of the points in these reverse local domains, it can be inferred whether the location of the first point positioning information is of special importance or whether it is closely related to other points, thereby providing richer and more valuable information and clues for the subsequent judgment and analysis of the source of danger of external damage to the transmission line. Assume that in an actual scenario, the first point positioning information represents the position of an object suspected to be close to the transmission line. By determining the reverse local domain of the first point, we can further understand the distribution of other point clouds around the object and determine whether there are other objects or environmental factors that may cause the object to get closer to the transmission line, so as to make more accurate warnings and preventive measures in advance.

[0050] The first point local domain and each point contained therein are processed one by one. Starting from the first point in the first point local domain, the spatial distance between this point and the first point location information is calculated using the Euclidean distance calculation formula. The calculation of the Euclidean distance is usually based on the coordinate value of the point, and the straight-line distance between the two points is obtained. Such a distance calculation is performed for each point in the first point local domain, and the calculated Euclidean distance between each point and the first point location information is set as the first point distribution sparse factor. These distribution sparse factors represent the distribution discreteness of the points in the first point local domain. All the calculated first point distribution sparse factors are added to a special distribution sparse factor set. In this way, through traversal calculation and collection, the distribution sparse factor set contains quantitative information about the distribution of points in the first point local domain. Further analysis and processing can be performed based on the data in this set, such as evaluating the density characteristics of the point cloud, judging the distribution trend of the object, etc., so as to provide a more accurate and valuable basis for the early warning of the danger source of external damage of the transmission line.

[0051] Calculate the average value of all distribution sparse factors in the distribution sparse factor set. This average value reflects the general level of distribution sparse factors in the entire first point local domain. Then, divide this average value by the specific first point distribution sparse factor, and the obtained ratio is defined as the first point distribution density. This ratio can be understood as the relative density of the first point relative to the average distribution sparseness of the entire first point local domain. Finally, add the calculated first point distribution density to the first moment point cloud density dataset. The purpose of this is to more comprehensively describe the distribution characteristics of the points in the first moment point cloud. By incorporating the first point distribution density into the first moment point cloud density dataset, the density distribution of the point cloud at the first moment is analyzed more carefully, thereby providing a more accurate and detailed basis for determining whether there is a transmission line external damage hazard source. For example, if the first point distribution density is large, it means that the first point is denser than the average level, suggesting that there is a gathering of objects in the area where the point is located; conversely, if the first point distribution density is small, it means that the point is relatively sparse and there are fewer objects distributed. In practical applications, by analyzing and comprehensively considering the distribution density of multiple points, potential dangerous areas can be identified more effectively, and corresponding measures can be taken in advance to ensure the safe operation of transmission lines. The method of calculating the distribution sparsity factor and distribution density using Euclidean distance can improve the accuracy of the point cloud density data set without fusing data from other sensors, thereby ensuring the reliability of anchor point identification and the accuracy of monitoring.

[0052] In a possible implementation, step S400 further includes:

[0053] Step S410: extracting the first moment anchor point positioning information and the first moment anchor point velocity information according to the anchor point positioning sequence.

[0054] Step S420: extracting the point cloud positioning information at the first moment and the point cloud velocity information at the first moment according to the point cloud positioning timing information.

[0055] Step S430: construct an input quaternion array based on the anchor point positioning information at the first moment, the anchor point speed information at the first moment, the point cloud positioning information at the first moment and the point cloud speed information at the first moment, wherein the first dimension element attribute of the input quaternion array is the anchor point positioning information, the second dimension element attribute is the comparison point positioning information, the third dimension element attribute is the anchor point speed information, the fourth dimension element attribute is the comparison point speed information, and the comparison point belongs to the point cloud at the first moment.

[0056] Step S440: Analyze the input quaternion array according to the positive sample recognition network to obtain an output value.

[0057] Step S450: set the output value equal to 0, delete the comparison point positioning information from the point cloud positioning information at the first moment, and when the point cloud positioning information at the first moment is traversed, output the positive sample point positioning information at the first moment and add it to the positive sample point positioning sequence.

[0058] Specifically, the anchor point positioning sequence stores a series of anchor point positioning and speed-related data arranged in chronological order. First, it is clear that the data to be processed at the first moment is clear. Then, by searching and filtering the anchor point positioning sequence, the data portion corresponding to the first moment is found according to the time mark. From this part of the data, the information that can describe the position of the anchor point at the first moment, that is, the first moment anchor point positioning information, and the information that can reflect the speed and direction of the anchor point at the first moment, that is, the first moment anchor point speed information, are extracted respectively. The extracted information will be used for subsequent comparison and analysis with the point cloud data at the same moment to determine whether there is a potential source of external damage to the transmission line.

[0059] Point cloud positioning timing information is a collection of information about the position and speed of the point cloud recorded in chronological order. Determine that the relevant information to be obtained is the first moment. Then, search and filter in the point cloud positioning timing information according to the time tag. Extract the specific position information of the point cloud recorded at the first moment, that is, the first moment point cloud positioning information. At the same time, the movement speed information of the point cloud at the same moment will also be extracted, that is, the first moment point cloud speed information. These extracted first moment point cloud positioning and speed information will be subsequently processed and analyzed together with the anchor point information at the same moment, so as to more accurately evaluate and predict situations that may pose a threat to the transmission line.

[0060] The positioning information of the anchor point at the first moment, the anchor point velocity information, and the positioning information and point cloud velocity information of the point cloud at the first moment have been obtained, and these information are combined to construct an input quaternion array. The first dimension element is specifically used to store the anchor point positioning information at the first moment, which represents the specific position of the anchor point in space. The second dimension element is used to place the comparison point positioning information. Here, the comparison point is selected from the point cloud at the first moment, and its positioning information describes the position of the point in space. The third dimension element is used to save the anchor point velocity information at the first moment, reflecting the motion state of the anchor point. The fourth dimension element stores the velocity information of the comparison point, showing the motion of the comparison point. By constructing the input quaternion array in this way, the position and velocity information of the anchor point and the point cloud at the first moment are orderly organized and integrated, and are prepared for subsequent input into the positive sample recognition network for analysis and processing. This structured organization method helps to clearly present and compare the relevant information of the anchor point and the point cloud, so as to more effectively analyze and judge the data to determine whether there is a situation that may affect the safety of the transmission line.

[0061] The positive sample recognition network is a trained model that can make judgments and analyses based on the characteristics of the input data. The constructed input quaternion array is input into this network, and the network will perform comprehensive calculations and evaluations on each element in the array and finally give an output value. This output value represents the judgment result of the network on the input data.

[0062] By analyzing the input quaternion array through the positive sample recognition network, an output value will be obtained. If the output value is equal to 0, it means that the current comparison point positioning information does not meet the characteristics of the positive sample. Next, the comparison point positioning information that does not meet the characteristics of the positive sample is deleted from the point cloud positioning information at the first moment. After that, the same judgment and processing are continued for other comparison points in the point cloud positioning information at the first moment, that is, the above operation is repeated. When all the comparison points in the point cloud positioning information at the first moment have been judged and processed in this way, the remaining point cloud positioning information that has not been deleted is the positive sample point positioning information at the first moment. Finally, the positive sample point positioning information at the first moment is output and added to the positive sample point positioning sequence. The purpose of this is to filter out the point positioning information that meets the positive sample characteristics from the original point cloud positioning information at the first moment, and organize them into the positive sample point positioning sequence for subsequent further analysis, processing or use in specific applications. For example, in the early warning of the danger source of external damage of the transmission line, these positive sample point positioning information are of great significance for identifying potential dangerous situations. Through the positive sample recognition network, without relying on other sensor data, it is possible to accurately identify and exclude irrelevant point cloud information, improve the accuracy of the positive sample point positioning sequence, and thus achieve higher timeliness and accuracy in the monitoring process.

[0063] In a possible implementation, step S400 further includes:

[0064] Step S460: Collecting an anchor point positioning data set, a positive sample positioning data set, and a negative sample point positioning data set.

[0065] Step S470: constructing an unsupervised learning scoring function, wherein the unsupervised learning scoring function is a maximization scoring function:

[0066] ,

[0067] ,

[0068] ,

[0069] in, Characterize the scoring function for each M training, Characterize the anchor feature vector of the i-th training, Characterize the feature vector of the positive sample point of the i-th training, Characterize the feature vector of negative sample points in the i-th training, Represents the hyperparameters that need to be learned, and Characterizes the preset weight parameters, Characterize the similarity parameters of the feature vector of the positive sample point in the i-th training, Characterizes the similarity parameters of the feature vector of the positive sample points in the i-th training.

[0070] Step S480: Based on the scoring function threshold and according to the unsupervised learning scoring function, the anchor point positioning data set, the positive sample positioning data set and the negative sample point positioning data set are retrieved to train the positive sample recognition network.

[0071] Specifically, three important data sets are collected. The first is the anchor point positioning data set, which contains the location information of specific anchor points set in the monitoring area. These anchor points serve as reference points to help us locate and analyze other objects. Next is the positive sample positioning data set, which collects the positioning data of objects or situations marked as positive samples. In the early warning of transmission line external damage hazards, positive samples usually represent objects that are close to the transmission line and may pose a threat to it. Finally, there is the negative sample point positioning data set, which covers the positioning information of objects or situations marked as negative samples. These negative samples are usually objects or situations that are unlikely to cause external damage hazards to the transmission line. By collecting these three data sets, we provide sufficient sample data for the training of the positive sample recognition network, so that the network can learn the characteristic patterns of positive and negative samples, so as to more accurately identify positive samples related to transmission line external damage hazards.

[0072] The goal of the unsupervised learning scoring function is to maximize the score to distinguish between positive and negative samples. , Represents the score for each M training run. For each training run (from 1 to M), an internal maximum is calculated. The maximum value is calculated by adding the difference between two similar parameters. value.

[0073] The similarity parameter calculation formula of the positive sample point feature vector is: , Characterize the anchor point feature vector (position and speed) of the i-th training, Characterize the feature vector of the positive sample points in the i-th training. and Represents the preset weight parameter used to adjust the similarity calculation. The inner product of the anchor point feature vector and the positive sample point feature vector is calculated and normalized by the norm of the vector to measure their similarity. The square of the difference between the anchor point feature vector and the positive sample point feature vector is calculated and normalized by the norm of the vector to measure their degree of difference.

[0074] The similarity parameter calculation formula of the negative sample point feature vector is: , the similarity parameter calculation method of the positive sample point feature vector is similar, except that the positive sample point feature vector is replaced by the negative sample point feature vector. Through this construction method, the unsupervised learning scoring function can calculate the score based on the sample's feature vector, thereby providing a basis for the positive sample recognition network to judge the sample category. When training the positive sample recognition network, by adjusting the network's parameters, the network can learn how to accurately identify positive and negative samples based on the scoring function.

[0075] Next, determine the scoring function threshold. This threshold will be used to determine whether the sample is a positive sample or a negative sample. For example, if the score is higher than the threshold, the sample is considered more likely to be a positive sample; conversely, if the score is lower than the threshold, it may be classified as a negative sample. According to the unsupervised learning scoring function, the anchor point positioning data set, the positive sample positioning data set, and the negative sample point positioning data set are retrieved. These data sets contain rich positioning information and will provide data support for network training. During the training process, the samples in the data set are input into the positive sample recognition network. The network will calculate the corresponding score using the scoring function based on the input anchor point feature vector, positive sample point feature vector, and negative sample point feature vector. Then, the calculated score is compared with the set threshold. Based on the comparison results, the network will adjust its own parameters to improve the recognition accuracy of positive samples. If a sample is judged as a positive sample, but it is actually a negative sample, the network will adjust the parameters to reduce the occurrence of such misjudgment in future judgments. Similarly, if a positive sample is misjudged as a negative sample, the network will also adjust the parameters accordingly to improve the recognition ability of positive samples. Through continuous iterative training, the positive sample recognition network will gradually learn how to accurately identify positive samples. During the training process, a large amount of sample data will be used to ensure that the network can learn the features of various situations, thereby improving its generalization ability and accuracy. By constructing an unsupervised learning scoring function, the positive sample recognition network can enhance its ability to accurately identify anchor points and positive sample points through training and optimization without the need for data fusion from other sensors, thereby improving the overall monitoring effect.

[0076] In a possible implementation, step S500 further includes:

[0077] Step S510: extracting a velocity direction sequence and a velocity magnitude sequence according to the anchor point positioning sequence.

[0078] Step S520: Decompose the speed direction sequence to obtain a horizontal direction sequence and a vertical direction sequence.

[0079] Step S530: performing feature analysis on the horizontal direction sequence to obtain predicted horizontal direction features.

[0080] Step S540: performing feature analysis on the vertical direction sequence to obtain predicted vertical direction features.

[0081] Step S550: performing feature analysis on the speed magnitude sequence to obtain predicted motion speed features.

[0082] Step S560: predicting the positive sample point positioning information at the end time of the positive sample point positioning sequence according to the predicted horizontal direction feature, the predicted vertical direction feature and the predicted motion speed feature, to generate the future time zone positive sample point positioning sequence.

[0083] Specifically, the anchor point positioning sequence contains various information about the anchor point, including the speed direction and speed magnitude. By analyzing and extracting the anchor point positioning sequence, we can obtain the speed direction sequence and speed magnitude sequence, which reflect the changes in the movement direction and speed magnitude of the anchor point at different times.

[0084] The velocity direction sequence can be further decomposed into a horizontal direction sequence and a vertical direction sequence. Such decomposition helps to analyze the motion characteristics of the anchor point in the horizontal and vertical directions in more detail. By decomposing the velocity direction into two mutually perpendicular directions, the motion trajectory and trend of the anchor point can be understood more accurately.

[0085] The purpose of feature analysis of horizontal sequences is to extract information that can describe the characteristics of horizontal motion. This includes analyzing the characteristics of speed change and directional stability in the horizontal direction. By analyzing these characteristics, the predicted horizontal characteristics are obtained, that is, the prediction of future horizontal motion.

[0086] Similarly, feature parsing is performed on the vertical sequence to obtain predicted vertical features, which helps to understand the movement trend and possible changes of the anchor points in the vertical direction.

[0087] The purpose of feature analysis of the speed sequence is to extract the features of speed changes, including the average, maximum, minimum, and frequency of speed changes. By analyzing these features, we can obtain the predicted motion speed features, that is, the prediction of future motion speed.

[0088] Based on the predicted horizontal features, we can infer the horizontal movement trend and possible position changes of the positive sample points. For example, if the predicted horizontal features indicate that the positive sample points will move to the right, then we can predict the horizontal position of the positive sample points at the end time based on this trend. Similarly, based on the predicted vertical features, we can predict the vertical position changes of the positive sample points. Then, combined with the predicted motion speed features, the moving speed of the positive sample points in the future time zone can be determined. Taking these factors into consideration, the positioning information of the positive sample points at the end time of the positive sample point positioning sequence is predicted. Based on the current position of the positive sample points, the predicted movement speed and direction, the possible positions of the positive sample points in the future time zone are calculated, and these position information are combined into the future time zone positive sample point positioning sequence. The future time zone positive sample point positioning sequence generated in this way can help us understand the movement trend of the positive sample points in advance, so as to timely discover the situation that may cause external damage to the transmission line, and take corresponding early warning and prevention measures.

[0089] In a possible implementation, step S530 further includes:

[0090] Step S531: When the horizontal direction sequence has a gradient feature, the mean value of the horizontal direction change angle is obtained and set as the predicted horizontal direction feature.

[0091] Step S532: when the horizontal direction sequence does not have a gradient feature, a horizontal direction mode value is obtained and set as the predicted horizontal direction feature.

[0092] Step S533: The gradient feature refers to a continuous increase or decrease along a certain direction.

[0093] Specifically, we first need to determine whether the horizontal sequence has a gradient feature. The gradient feature refers to the continuous increase or decrease of data in a certain direction in the horizontal direction. If the horizontal sequence has this feature, calculate the horizontal change angle. For each data point in the horizontal sequence, calculate the horizontal change angle between the previous data point. Calculate the mean of the change angles, average all the calculated horizontal change angles, and obtain the mean of the horizontal change angles. Set it as the predicted horizontal feature, and set this mean as the predicted horizontal feature for subsequent analysis and prediction.

[0094] If it is determined that the horizontal sequence does not have a gradient feature, the horizontal mode value is counted, and all the data in the horizontal sequence are counted to find the horizontal value with the most occurrences. Set it as the predicted horizontal feature, and set the horizontal mode value with the most occurrences as the predicted horizontal feature. By performing feature analysis on the horizontal sequence, the horizontal change characteristics of the positive sample points can be accurately predicted without relying on other sensor data, thereby improving the accuracy of positioning the positive sample points in the future time zone and ensuring the timeliness and accuracy of monitoring.

[0095] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0096] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0097] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A 4D millimeter wave radar based early warning method for external damage hazards of power transmission lines, characterized in that: include: Receive a density point cloud data set of a 4D millimeter wave radar, wherein the density point cloud data set includes a distance label data set, a speed label data set, a horizontal angle data set, and a pitch angle data set; The distance label data set, the speed label data set, the horizontal angle data set and the pitch angle data set perform spatial point cloud positioning to generate point cloud positioning time series information; Traversing the point cloud positioning time sequence information to identify anchor points, and generating an anchor point positioning sequence, wherein the anchor points are determined by anchor point rules determined based on factors related to power transmission line safety, and all points identified as anchor points form an anchor point positioning sequence according to their order in the original point cloud positioning time sequence information; Through the positive sample recognition network, based on the anchor point positioning sequence, the point cloud positioning time series information is traversed for analysis to obtain a positive sample point positioning sequence, wherein the positive sample represents an object close to the transmission line and may pose a threat to it, and the entire point cloud positioning time series information is continuously analyzed and processed, and finally all points identified as positive samples are screened out and arranged in the order of their appearance in the time series information to form a positive sample point positioning sequence; Perform motion simulation according to the positive sample point positioning sequence to generate a future time zone positive sample point positioning sequence; When the future time zone positive sample point positioning sequence intersects with the transmission line boundary area, a transmission line external damage hazard source warning signal is generated to give an alarm.

2. The method according to claim 1, characterized in that The distance label data set, the speed label data set, the horizontal angle data set and the pitch angle data set are subjected to spatial point cloud positioning to generate point cloud positioning time series information, including: Extracting first distance label data, first speed label data, first horizontal angle data and first pitch angle data at a first moment according to the distance label data set, the speed label data set, the horizontal angle data set and the pitch angle data set; Taking the position of the 4D millimeter-wave radar as the origin position, positioning is performed based on the first distance label data, the first speed label data, the first horizontal angle data, and the first pitch angle data to obtain first point positioning information at a first moment, and add the first point positioning information to the point cloud positioning information at the first moment; The point cloud positioning information at the first moment is added to the point cloud positioning timing information.

3. The method according to claim 1, characterized in that Traversing the point cloud positioning time sequence information to identify anchor points and generate an anchor point positioning sequence, including: Extracting point cloud positioning information at a first moment according to the point cloud positioning time sequence information; Traversing the point cloud positioning information at the first moment to perform density analysis to generate a point cloud density data set at the first moment; According to the maximum value of the point cloud density data set at the first moment, the point cloud anchor point positioning information at the first moment is configured and added to the anchor point positioning sequence.

4. The method according to claim 3, characterized in that Traversing the point cloud positioning information at the first moment to perform density analysis to generate a point cloud density data set at the first moment, including: Extracting first point positioning information according to the point cloud positioning information at the first moment; Taking the first point positioning information as a reference point, sorting a preset number of point positioning information from near to far from the point cloud positioning information at the first moment, setting a first point local domain, and adding the first point local domain to the local domain set; When the first point cloud positioning information is located, extract the second point local domain until the point in the Nth point local domain that has the first point positioning information, and set it as the first point reverse local domain; Traversing the points of the first point local domain and the first point local domain, calculating the Euclidean distance to the first point positioning information, setting it as the first point distribution sparse factor, and adding a distribution sparse factor set; The ratio of the distribution sparse factor set mean to the first point distribution sparse factor is calculated, set as the first point distribution density, and added to the point cloud density dataset at the first moment.

5. The method according to claim 1, characterized in that Through the positive sample recognition network, based on the anchor point positioning sequence, the point cloud positioning time sequence information is traversed for analysis to obtain a positive sample point positioning sequence, including: Extracting the first moment anchor point positioning information and the first moment anchor point speed information according to the anchor point positioning sequence; Extracting the point cloud positioning information at a first moment and the point cloud velocity information at a first moment according to the point cloud positioning timing information; Constructing an input quaternion array according to the first-time anchor point positioning information, the first-time anchor point speed information, the first-time point cloud positioning information and the first-time point cloud speed information, wherein the first-dimensional element attribute of the input quaternion array is the anchor point positioning information, the second-dimensional element attribute is the comparison point positioning information, the third-dimensional element attribute is the anchor point speed information, the fourth-dimensional element attribute is the comparison point speed information, and the comparison point belongs to the first-time point cloud; Analyzing the input quaternion array according to the positive sample recognition network to obtain an output value; The output value is set equal to 0, and the comparison point positioning information is deleted from the point cloud positioning information at the first moment. When the point cloud positioning information at the first moment is traversed, the positive sample point positioning information at the first moment is output and added to the positive sample point positioning sequence.

6. The method according to claim 5, characterized in that The positive sample recognition network construction step includes: Collect anchor point positioning data sets, positive sample positioning data sets and negative sample point positioning data sets, wherein the anchor point positioning data sets contain the location information of specific anchor points set in the monitoring area, and these anchor points can be used as reference points to locate and analyze other objects; the positive sample positioning data sets collect the positioning data of objects or situations marked as positive samples; the negative sample point positioning data sets cover the positioning information of objects or situations marked as negative samples, which are objects or situations that are unlikely to cause external damage to the transmission line; Construct an unsupervised learning scoring function, wherein the unsupervised learning scoring function is a maximization scoring function: , , , in, Characterize the scoring function for each M training, Characterize the anchor feature vector of the i-th training, Characterize the feature vector of the positive sample point of the i-th training, Characterize the feature vector of negative sample points in the i-th training, Represents the hyperparameters that need to be learned, and Characterizes the preset weight parameters, Characterize the similarity parameters of the feature vector of the positive sample point in the i-th training, Characterize the similarity parameters of the feature vector of the negative sample points in the i-th training; Based on the scoring function threshold and according to the unsupervised learning scoring function, the anchor point positioning data set, the positive sample positioning data set and the negative sample point positioning data set are retrieved to train the positive sample recognition network.

7. The method according to claim 1, characterized in that Performing motion simulation according to the positive sample point positioning sequence to generate a future time zone positive sample point positioning sequence includes: Extracting a velocity direction sequence and a velocity magnitude sequence according to the anchor point positioning sequence; Decomposing the speed direction sequence to obtain a horizontal direction sequence and a vertical direction sequence; Performing feature analysis on the horizontal direction sequence to obtain predicted horizontal direction features; Performing feature analysis on the vertical direction sequence to obtain predicted vertical direction features; Performing feature analysis on the speed magnitude sequence to obtain predicted motion speed features; The positive sample point positioning information at the end point of the positive sample point positioning sequence is predicted according to the predicted horizontal direction feature, the predicted vertical direction feature and the predicted motion speed feature to generate the future time zone positive sample point positioning sequence.

8. The method according to claim 7, characterized in that Performing feature analysis on the horizontal sequence to obtain predicted horizontal features includes: When the horizontal direction sequence has a gradient feature, a mean value of the horizontal direction change angle is obtained and set as the predicted horizontal direction feature; When the horizontal direction sequence does not have a gradient feature, obtaining a horizontal direction mode value and setting it as the predicted horizontal direction feature; Among them, the gradient feature refers to the continuous increase or decrease along a certain direction.