Power cable monitoring analysis method and system

Abstract

The invention relates to the technical field of power inspection, and discloses a power cable monitoring analysis method and system. The method comprises the following steps: acquiring an overall topology network of a power system and the number of unmanned aerial vehicles, and partitioning nodes of the overall topology network to obtain a plurality of local topology networks; creating an inspection task and task data information based on the local topology network, and issuing the inspection task and the task data information to each unmanned aerial vehicle; controlling the unmanned aerial vehicle to go to the starting point of the corresponding partition, traversing and recording each node and branch based on a preset rule, and checking the electric wire in the traversing process; controlling the unmanned aerial vehicle to generate a new partition topology network based on the recorded nodes and branches; and after all the unmanned aerial vehicles complete the inspection tasks and return the inspection tasks, acquiring nodes and branches to be recorded by the unmanned aerial vehicles, summarizing the partition network as a total topology network, and deleting the repeatedly generated branches and nodes. The method has the advantage of reducing the problems of missing detection, re-detection and repeated path flight caused by positioning drift or line change.

Description

Technical Field

[0001] The present application relates to the technical field of power inspection, and in particular to a power cable monitoring and analysis method and system thereof. Background Art

[0002] The power system plays a vital role in the infrastructure of modern society, transmitting high-voltage electricity to meet the electricity needs of cities, industries and remote areas. Since transmission lines often span large geographical areas and are distributed in high-rise or complex terrain environments, the maintenance and troubleshooting of power lines are particularly important. If potential risks are not discovered and eliminated in a timely manner, such as loose conductors, damaged insulation, tilted poles and entangled foreign objects, they may cause large-scale power outages, safety accidents or economic losses; therefore, maintaining the real-time health of transmission lines has always been a key task faced by power grid companies.

[0003] In order to improve the inspection efficiency, the existing technology usually uses drones to carry cameras or sensors to inspect power lines and poles in a visual way. The common practice is to use GPS and geographic coordinates to fly drones autonomously or semi-autonomously along known routes, and the captured video and image data are then used by operation and maintenance personnel or recognition algorithms for fault detection. Although GPS is convenient for large-scale positioning, it often faces multiple challenges in the actual application scenarios of power transmission networks: on the one hand, the distance between poles may be relatively short, and the accuracy and stability of GPS signals are difficult to meet the requirements of accurate navigation at the meter or even sub-meter level; on the other hand, in the face of surrounding buildings, terrain shielding or signal interference, the drift phenomenon of GPS will be more obvious. In addition, the power network is constantly changing: new poles may be built, some old lines may be removed, or some branches may be temporarily disconnected due to maintenance. In addition, cable sags, curvature changes, etc., which will cause the method that simply relies on GPS path inspection to have problems of missed inspections or repeated flights. Summary of the invention

[0004] In order to ensure the comprehensiveness and efficiency of inspections and reduce the problems of missed inspections, re-inspections, and repeated path flights caused by positioning drift or line changes, the present application provides a power cable monitoring and analysis method and system.

[0005] In a first aspect, the present application provides a power cable monitoring and analysis method, which adopts the following technical solution: A power cable monitoring and analysis method comprises the following steps: S1. Obtaining an overall topological network of the power system and the number of drones, and partitioning the nodes of the overall topological network based on the number of drones to obtain multiple local topological networks, wherein the overall topological network includes nodes and branches, the nodes correspond to electric poles, and the branches correspond to electric wire branches; S2. Create inspection tasks and task data information based on the local topology network, and send them to each drone respectively. The task data information includes node information and branch information corresponding to a local topology network, and pre-recorded node image information corresponding to each node. The edge of the local topology network is a node, and adjacent local topology networks are connected through shared nodes. S3. Control the drone to go to the starting point of the corresponding partition, and then start from the starting point to traverse and record each node and branch based on the preset rules, and check the wires during the traversal process; S4. Control the UAV to generate a new partition topology network based on the recorded nodes and branches; S5. After all drones complete their inspection missions and return, obtain the nodes and branches recorded by the drones, determine the common intersections generated by drones in adjacent areas, summarize the partitioned networks into a total topological network, and delete the duplicated branches and nodes.

[0006] Optionally, the S1 comprises the following steps: S11. Obtain the overall topology network of the power system and the number of drones N, and determine the branch number distribution range and node number distribution range of the local topology network based on the total number of nodes, the total number of branches and the number of drones in the overall topology network; S12. According to the number of branches and the distance constraint, each node of the overall topological network is preliminarily judged and screened, and N candidate nodes are selected as possible centers of the local topological network; S13. Taking each candidate center point as the core, absorb the ownerless nodes into their corresponding sub-networks layer by layer or in a shortest path manner, wherein the absorption speed of each candidate center is the same; S14. When the number of nodes or branches in a sub-network reaches the upper limit, stop absorbing new nodes into the sub-network; S15. Allocate the remaining nodes that have not been assigned to avoid isolated nodes or redundant sub-networks; S16. In each local topology network, check the path length from the center point to each edge node. If it exceeds or falls below a preset threshold, adjust the boundary nodes to reassign some nodes to adjacent local topology networks so that the distance from the center point to the edge gradually returns to a reasonable range. S17. Re-evaluate the center point position so that the difference between the maximum node distance and the minimum node distance from the center point to the edge in the local topology network is less than a preset threshold; wherein the center point position is used as the flight starting point of the drone in the partition; S18. Check whether the constraints of the local topology network are met and perform step rollback for adjustment.

[0007] Optionally, the S3 includes: S31. The drone goes to the starting point of the corresponding partition; S32. The drone flies to the reference point corresponding to the electric pole based on image recognition; S33. The drone creates a blank new topological network diagram, and generates nodes corresponding to the electric poles in the new topological network diagram; S34. The drone is adjusted to a preset posture at the reference point and photographed at the node position to obtain a node position image; S35. The drone uses the pre-recorded node image information as a priori condition to identify the wires in the node position image; S36. The drone generates branches corresponding to the electric wires based on the node location image, and generates nodes corresponding to the electric wire poles in the new topological network diagram; S37. The UAV determines the branch inspection order of the node based on the new topology network diagram, and inspects the wires corresponding to the branches according to the branch inspection order and inspection strategy.

[0008] Optionally, the inspection strategy includes: The drone will first inspect the branches according to the branch inspection order set by the local topology network at the node. If there are still branches that have not been inspected after the local topology network corresponding to the wire inspection of the new topology network diagram is completed, the branch will be inspected; Based on image recognition, the drone determines whether the node reached along the branch inspection is the corresponding node in the local topology network. If so, it determines whether all branches corresponding to the node have been traversed. If not, it goes to the next branch in the branch inspection order. If so, it returns to the previous node. If not: calculate the distance between the node and the center point. If the distance is less than or equal to the distance to the edge, the node is determined to be a new node in the partition. If it is greater, the drone is controlled to return to the previous node.

[0009] Optionally, the S32 includes the following sub-steps: S321. When the UAV reaches the area corresponding to the starting point, it obtains an image of the electric pole at the node; S322. The drone determines the location of the electric pole in the image based on image recognition; S323. The drone selects a position at a preset height above the electric pole as a reference point; S324. The drone flies to the reference point.

[0010] Optionally, the S35 includes the following steps: S351. Obtaining pre-recorded node image information; wherein the pre-recorded node image includes a utility pole and a plurality of wires, and the pre-recorded node image is photographed at a reference point corresponding to the utility pole in a preset posture; S352. Acquire a node position image, wherein the node position image is captured at a reference point corresponding to the electric pole in a preset posture; S353. Align the node position image with the pre-recorded node image; S354. Performing wire detection on the node position image to identify thin and long lines or arcs; S355. Perform feature analysis on the detected line segments, and merge the approximately continuous short line segments to obtain the wire contour, wherein the feature analysis includes length analysis, angle analysis, and continuity analysis; S356. In the corresponding registered coordinate system, the starting point, end point, direction and length of the line segments in the two photos are compared, and when the similarity or matching degree meets the set threshold, it is determined to be the same wire; wherein the matching degree includes the endpoint distance, the average offset, and the angle difference; If no matching object can be found or the matching degree is low, it is determined that new wires are added or reduced; the new wires correspond to wires that exist in the node position image but do not exist in the pre-recorded node image; reduced wires correspond to wires that do not exist in the node position image but exist in the pre-recorded node image.

[0011] Optionally, the S37 includes the following steps: S371. Control the drone to obtain image information, and obtain wire image information based on instance segmentation; S372. Input to a pre-trained deep learning model to identify the risk type; S373. If a risk is identified, record the branch and mark the risk type, and take a photo of the risk location.

[0012] Among them, the training steps of the pre-trained deep learning model include: S3721. Determine the types of risks that need to be identified and define the identifiable characteristics of each risk type; S3722. Obtain various types of image samples containing identifiable risk features, and manually mark the identifiable features in the images using annotation tools; S3723. Divide the image samples into a training set and a test set based on a pre-selected partitioning strategy and a partitioning ratio, wherein the training set is used to train the model, and the test set is used to verify and adjust the model, and the pre-selected partitioning strategy is random partitioning, stratified partitioning, or cross-validation; S3724. Train a CNN model using the ResNet architecture based on the training set, using the cross entropy loss function during training; S3725. Input the test set to the trained model, input the risk type identified by the trained model into a preset reward function for calculation and obtain a reward value, wherein the output of the preset reward function is related to the difference between the risk type identified by the CNN model and the actual risk type, and the greater the difference, the lower the reward value, and the smaller the difference, the smaller the reward value; S3726. Adjust the parameters of the trained model based on the reinforcement learning algorithm and the change of the reward value to increase the reward value, thereby obtaining a new risk type, substituting it into the previous step and executing it until the risk type converges to a stable range.

[0013] In a second aspect, the present application provides a power cable monitoring and analysis system, which adopts the following technical solution: A power cable monitoring and analysis system comprises a processor, wherein a program of any one of the above-mentioned power cable monitoring and analysis methods is run in the processor.

[0014] In a third aspect, the present application provides a storage medium, which adopts the following technical solution: A storage medium stores a program of any one of the above-mentioned power cable monitoring and analysis methods.

[0015] In summary, the present application includes at least one of the following beneficial technical effects: it can avoid excessive reliance on GPS, and it can also quickly locate and identify nodes and branches that have been added or removed in a dynamic power system, so as to realize timely discovery and maintenance of potential fault hazards. Based on the original topological map of the power network, combined with the prior image information taken by the drone at each node, the added or removed wires and nodes can be updated locally, and finally an accurate and up-to-date overall topological network is formed after the data of multiple drones are aggregated, thereby ensuring the comprehensiveness and efficiency of the inspection, and reducing the problems of missed inspections, re-inspections, and repeated path flights caused by positioning drift or line changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of a power cable monitoring and analysis method in a certain embodiment of the present application. DETAILED DESCRIPTION

[0017] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings.

[0018] In the description of this specification, the description with reference to the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0019] The present application discloses a method for monitoring and analyzing a power cable. Figure 1 The method includes the following steps S1-S5.

[0020] S1. Obtain the overall topological network of the power system and the number of drones, and partition the nodes of the overall topological network based on the number of drones to obtain multiple local topological networks, wherein the overall topological network includes nodes and branches, the nodes correspond to utility poles, and the branches correspond to utility line branches.

[0021] In this step, it is first necessary to obtain the overall topological network of the power system and the number of drones in order to clarify all the nodes involved in the inspection process and the equipment resources available for inspection. The "overall topological network" mentioned here refers to a complete graphical or data structure consisting of electric poles and electric wire branches. Among them, "nodes" correspond to electric poles, which bear the function of power transmission support or information intersection; "branches" correspond to electric wire branches, connecting each electric pole and forming a complete power supply path. By mastering the global information of these nodes and branches, a more accurate division of labor can be made in the subsequent drone scheduling and inspection route planning. For example, if there are a total of 100 electric poles and 200 electric wire branches in an area, and there are only 5 drones on hand, these 100 nodes need to be reasonably divided into 5 groups so that each drone can complete the inspection of the block they are responsible for in a centralized and efficient manner.

[0022] After the overall network and the number of drones are acquired, the nodes will be partitioned according to the number of drones, splitting the huge power network into several small local topology networks that are easier to manage and inspect. The purpose of this is to reduce the workload of a single drone, so that each drone has a relatively balanced inspection range, while reducing the waste of resources caused by repeated flights or mutual interference between drones in complex environments. Through this partitioning method, on the one hand, it can ensure that all nodes can be included in the inspection range, reducing omissions and inspection blind spots; on the other hand, it can also be easier to locate specific branches or nodes when a fault occurs, which is very convenient for subsequent maintenance and correction.

[0023] Specifically, S1 includes the following steps S11-S18.

[0024] S11. Obtain the overall topology network of the power system and the number N of drones, and determine the branch number distribution range and node number distribution range of the local topology network based on the total number of nodes, the total number of branches and the number of drones in the overall topology network.

[0025] In this step, the system first obtains all the node and branch information that constitute the power system as a whole, that is, collects the connection relationship between each pole and branch line, and then confirms the number of drones N that can be put into inspection. By mastering these three types of key data, namely the total number of nodes, the total number of branches, and the number of drones, the system can formulate a reasonable inspection zoning plan at the planning level. The "node" here refers to the pole, and the "branch" refers to the branch line connecting the pole. For example, if a power system has 150 nodes and 280 branches, and the number of available drones is 4, the system will comprehensively consider the number of nodes and branches that each local network should contain, and determine a distribution range. For example, under the above scale, each drone can be responsible for 30 to 45 nodes and 50 to 90 branches, so that each local topology network can have a relatively balanced workload, which not only avoids the task of a single drone being too heavy, but also ensures the overall inspection efficiency of the system.

[0026] S12. According to the number of branches and distance constraints, each node of the overall topological network is preliminarily judged and screened, and N candidate nodes are selected as possible centers of the local topological network.

[0027] In this step, the system will preliminarily screen the nodes corresponding to each pole within the overall topological network according to the number of branches and distance restrictions. The goal is to select a suitable "center" node from many nodes to form the core of the local topological network in the future. In this process, it is necessary to comprehensively evaluate the number of branches around the node, the degree of connectivity, and the distribution of the distance from the node to other nodes. For example, if the poles in a certain area are very densely distributed and the distances between some nodes are far, it is necessary to give priority to those nodes that are relatively balanced with nearby nodes and where the distance is too far or too close. The situation will not appear on a large scale. Only in this way can it be ensured that each local network is neither too concentrated nor too dispersed during the subsequent inspection process.

[0028] When performing the above screening, the system often uses a certain algorithm to calculate the average distance, maximum distance and minimum distance between each node and its neighboring nodes to determine whether they meet the pre-set threshold. The N candidate nodes finally selected are considered to have the potential to become the center of the local network. Their locations usually cover the poles within a certain range and prevent the number of nodes and branches in the local network from being difficult to manage due to excessive concentration.

[0029] S13. Taking each candidate center point as the core, absorb the ownerless nodes into their corresponding sub-networks layer by layer or in a shortest path manner, wherein the absorption speed of each candidate center is the same.

[0030] In this step, each selected candidate center point will serve as the "core" of a local topological network, and gradually absorb the surrounding nodes that have not been divided into any sub-networks. The "layer-by-layer" method here means that you can start from the central node and expand radially to the adjacent nodes surrounding it; while the "shortest path" method is to classify other nodes under the name of the nearest center point according to the connection or path length between nodes, based on the principle of the shortest distance or the least number of hops. Ownerless nodes refer to nodes that have not been included in any local network before. They will be gradually allocated to more suitable or better center points in this process. In order to prevent some center points from "snatching away" too many nodes too quickly, the absorption speed of all candidate centers will be guaranteed to be consistent, so as to ensure that the division process of the entire topological network is relatively balanced and that the problem of excessive expansion of individual sub-networks will not occur. For example, if in an area, center point A and center point B are not far apart, and there are several nodes that have short paths to both of them, then the system will calculate the shortest path from each node to A and B for comparison, so as to decide which center point the node is assigned to. The purpose of doing this is to keep the sub-network structure stable and reasonably distributed, which can not only reduce the overlap of the subsequent inspection range of the drone, but also improve the efficiency and coverage of the overall inspection.

[0031] S14. When the number of nodes or branches in a sub-network reaches the upper limit, stop absorbing new nodes into the sub-network.

[0032] In this step, when the number of nodes or branches in a sub-network reaches a pre-defined upper limit, the system will temporarily stop adding new nodes to the sub-network to prevent it from over-expanding. The "upper limit" here usually refers to the reasonable load range set for each local topology network before. For example, when calculating the number of nodes and branches in the local topology network, the inspection capacity and inspection time of the drone have been estimated. If the burden of any sub-network exceeds this range, it will cause difficulties or unfair distribution in the subsequent actual inspection.

[0033] In the specific implementation, the system will always monitor the number of nodes and branches of the current sub-network during the network expansion process. Once it is detected that a sub-network has reached the preset upper limit, the sub-network will be marked as "full load" and new nodes will be prevented from joining. This approach can be figuratively likened to "stop when full": when the service range of a central point is full, the remaining nodes will no longer be accommodated to prevent an imbalance. For example, if a sub-network already accommodates 40 nodes and its threshold is exactly 40, the system will interrupt the expansion of the center and require the remaining unassigned nodes to seek other central points instead, thus maintaining a relatively balanced scale between the sub-networks.

[0034] S15. Allocate the remaining nodes that have not been assigned to avoid isolated nodes or redundant sub-networks.

[0035] During this process, the system will redistribute the remaining nodes that have not been absorbed by any sub-network to ensure the integrity of the entire network. The reason why "remaining nodes" appear is often because some nodes were not divided in time in the previous steps due to the limitation of distance or number of branches, or the adjacent center point has reached the upper limit and stopped absorbing. By reallocating these "ownerless" nodes to appropriate sub-networks, it is possible to avoid the phenomenon of "isolated nodes" or the formation of "extra small networks" in the topology structure, thereby maintaining the order and coherence of the overall division. For example, if there are several scattered electric poles in a specific area that are rejected by the adjacent sub-network (because the adjacent sub-network is saturated), then the system is required to identify these nodes and assign them to other central points that are closest to them and still have capacity. This not only ensures that all electric poles can be included in the inspection coverage, but also avoids the embarrassing situation that drones cannot be dispatched in the future. The final effect is to keep the number of local topological networks consistent with the previous design goals, reducing the risk of duplication or omissions in inspection and management.

[0036] Alternatively, in some other embodiments, when the system finds that there are remaining nodes that need to be allocated but are adjacent to a subnetwork that has reached its capacity limit, this problem can be solved by a "node replacement" method. Specifically, the remaining nodes are first temporarily assigned to the adjacent saturated subnetwork, and then a node is selected from the junction of this subnetwork and other subnetworks that still have surplus capacity to "release" it, and then it is divided into the adjacent subnetwork that still has capacity. In this way, it seems that a new node is added to the originally saturated subnetwork, but in fact, a node is transferred to the nearby subnetwork to ensure the overall network connectivity and load balance. For example, if a subnetwork has reached its node limit because a large number of electric poles have been absorbed in the early division, but it is adjacent to a subnetwork that still has room for expansion, the newly arrived remaining nodes can be placed in this fully loaded subnetwork first, and then a node suitable for transfer can be selected from the edge of it and the idle subnetwork for exchange. Doing so not only allows the two subnetworks to have a coherent pole layout, but also ensures that there will be no isolated or redundant nodes, while maintaining the overall physical and logical connectivity of the network structure.

[0037] S16. In each local topology network, check the path length from the center point to each edge node. If it exceeds or falls below a preset threshold, some nodes are reassigned to adjacent local topology networks by adjusting the boundary nodes, so that the distance from the center point to the edge gradually returns to a reasonable range.

[0038] In this step, the system will further examine the node distribution within each local topological network, focusing on whether the path length between the center point and the edge node exceeds or is lower than the established threshold. Here, the "center point" refers to the core node selected in the preliminary screening or calculation of the local topological network, and the "edge node" is distributed at the outermost part of the subnetwork, which may be connected to other subnetworks or far away from the center. The system will calculate the shortest path or average path distance from the center point to the edge node. If these values ​​deviate from the pre-set upper and lower limits, it means that the subnetwork is too large or too small, or there is an unreasonable node allocation at the boundary, and appropriate regulation is required. It should be noted that the reasonable interval mentioned in S16 is the average value ±1 of the distance from the center point to the edge node of each local topological network. Of course, other values ​​can also be used in other embodiments.

[0039] In specific implementation, non-compliant edge nodes and their related connections are usually reassigned to adjacent sub-networks. Through this "border node adjustment" method, nodes that were originally too far or too close are assigned to more appropriate center points. For example, if the center of a sub-network is located in an area with relatively dense electric poles, but there are still a few nodes on its outermost periphery that are too far away from the center and are not convenient for inspection and scheduling, these edge nodes can be transferred to adjacent sub-networks, so that all sub-networks are more balanced in terms of geographical distribution and connection relationships. The purpose of this is to maintain the compactness and manageability within each sub-network, to avoid the drone spending too much flight time in a single area due to excessive distance during subsequent inspections, and to reduce potential conflicts or resource waste caused by excessive overlap between sub-networks.

[0040] S17. Re-evaluate the center point position so that the difference between the maximum node distance and the minimum node distance from the center point to the edge in the local topology network is less than a preset threshold; wherein the center point position serves as the flight starting point of the drone in the partition.

[0041] The system will re-evaluate the location of the center point of the local topological network so that the difference between the maximum distance and the minimum distance between the center point and the edge node of the network is less than the preset threshold. The center point is the most critical reference position of a local network and the starting point of the drone's subsequent flight inspection mission. If the gap between the farthest and nearest nodes is too large, it means that the layout inside the network is not balanced enough, which may cause the drone to be inefficient or unevenly covered during flight and inspection. To solve this problem, the system calculates the actual distance distribution from the current center point to all edge nodes, and makes the edge distance more average by replacing or adjusting the center point. For example, in an irregularly distributed node set, if the initially selected center point is biased to one side of the network, then other edge nodes in diagonal positions will appear too far away. By moving the center point or reselecting a node closer to the overall geometric center, the far end distance can be effectively reduced, thereby keeping the distance distribution within a reasonable range.

[0042] During the implementation process, the system will combine the analysis results of the shortest path or average distance between nodes, and try to recalibrate the sub-network centers with excessive distance differences to a more balanced position, or select a pole that is more evenly distributed with the edge nodes as a new core node. This will not only make the structure of the local topology network more compact, the inspection range will be more reasonable, and the subsequent path planning effect will be improved. When the drone departs from this "updated" center point, whether it flies to nearby nodes or covers more distant nodes, it can complete the inspection well within the established range and time limit.

[0043] S18. Check whether the constraints of the local topology network are met and perform step rollback for adjustment.

[0044] At this stage, the system will conduct an overall check on all constraints of the local topology network to determine whether the number of nodes, the number of branches, the distance range between the center and the edge nodes have reached the previously set values. If it is found during the inspection that there are still sub-networks that fail to meet the requirements, or that the node allocation in some areas is unreasonable, it will fall back to the corresponding steps according to the problem type and re-perform fine-tuning. For example, if the system detects that the number of nodes in a sub-network exceeds the upper limit, it will fall back to the previous link responsible for network absorption and node movement; if it is found that the distance from the center to some edge nodes is still unbalanced, it will fall back to the step to adjust the center point position or allocation strategy. Through this iterative method, whether it is node allocation, branch division or distance correction, multiple correction opportunities can be obtained to gradually approach a balanced local topology structure that meets the inspection requirements. For example, if a sub-network is overloaded because it accidentally absorbs several more nodes, it can be transferred by falling back to the previous absorption or release process, or the problem of excessive distance of remote nodes can be alleviated by further adjusting the center point.

[0045] S2. Create inspection tasks and task data information based on the local topology network, and send them to each drone separately. The task data information includes the node information and branch information corresponding to a local topology network, as well as the pre-recorded node image information corresponding to each node. The edge of the local topology network is the node, and adjacent local topology networks are connected through shared nodes.

[0046] Based on the previously divided local topological network, the system will generate inspection tasks and the required data information for each drone. The so-called "local topological network" refers to the division of the entire power system into several relatively independent network areas, each of which contains several poles (i.e. nodes) and wire branches between them. This information has been clarified in the partitioning process in the previous stage. When creating an inspection task, it is necessary to organize the node information and branch information corresponding to the local topological network, as well as the pre-recorded node images (i.e. pre-stored or photographed image data) corresponding to each node to form "task information". In this way, after receiving the issued task, each drone can accurately locate the poles, branches and other key elements in combination with the graphic data of the local area.

[0047] In a more specific implementation, the system will determine a flight coverage range for each local topology network, and indicate in the mission information all the poles contained therein and the routing methods between the poles. For example, if a local topology network contains ten nodes and several branches, and each of the nodes already has one or more pre-taken photos, then these photos will be packaged into a drone inspection mission for the drone to refer to when comparing or identifying wires. The advantage of doing this is that on the one hand, the drone can compare known images with real-time footage during flight to identify whether there are new, missing or damaged wires; on the other hand, it can better ensure inspection coverage and prevent the omission of important nodes or branches.

[0048] S3. Control the drone to go to the starting point of the corresponding partition, and then start from the starting point to traverse and record each node and branch based on the preset rules, and check the wires during the traversal process.

[0049] At this stage, the system will first navigate the drone to the starting node of the corresponding partition, and then use this node as the starting point to traverse the entire sub-network along the wire branches one by one based on the pre-set inspection order. The so-called "preset rules" can be inspection route planning defined manually or by algorithms. One typical way is to select the surrounding branches for inspection in a clockwise direction with the current node as the center of the circle in the top view; in other words, when the drone arrives at a node, it will switch the various connection branches around the node from a "clockwise" angle in turn, thereby determining the direction of the next wire to be inspected.

[0050] Another method may also be used in the present application, which is similar to that of selecting surrounding branches for detection in a local topology network in a clockwise direction with the current node as the center of the circle.

[0051] Specifically, S3 includes sub-steps S31-S37.

[0052] S31. The drone goes to the starting point of the corresponding partition.

[0053] S32. Based on image recognition, the drone flies to the reference point corresponding to the utility pole.

[0054] Optionally, S32 includes the following sub-steps S321-S324.

[0055] S321. When the UAV reaches the area corresponding to the starting point, it obtains the image of the electric pole at the node.

[0056] S322. The drone determines the location of the utility poles in the image based on image recognition.

[0057] S323. The drone selects a position at a preset height above the electric pole as a reference point.

[0058] S324. The drone flies to the reference point.

[0059] The drone will first automatically fly to the starting position of the corresponding partition based on the previously divided partition information, and regard this position as the starting point for subsequent inspection activities. The significance of this step is to establish a fixed reference position for the drone so that the entire inspection mission can be carried out in a clear initial scenario. Next, the drone will use image recognition technology to go to the reference point corresponding to the utility pole. Here, "image recognition" refers to the use of machine vision algorithms to detect and locate targets within the field of view, such as utility poles, beams or other significant landmarks, thereby providing accurate coordinate references for drone navigation. The reference point usually refers to a position at a certain preset height above the utility pole, ensuring that the drone will not interfere with the power grid facilities and can obtain a stable and wide-view monitoring angle.

[0060] In the actual implementation process, once the drone reaches the area where the starting point is located, it will first take a bird's-eye or side view of the area, and then use a special image recognition model to detect whether there is a telephone pole in the image and its specific location. Assuming that the image recognition algorithm detects that the telephone pole is located in a certain area of ​​the picture, the drone will calculate its relative position and height relative to itself, and set the target height to a safe and easy-to-observe height value, such as three to five meters higher than the top of the telephone pole. Subsequently, the drone flies to this fixed point, the reference point, according to the calculation results, so that it hovers in a safe area directly above the telephone pole. Through this arrangement of steps, whether there are other obstacles in the environment or the shape of the telephone pole itself is irregular, the flight path can be effectively calibrated by image recognition and reference point setting, ensuring that the drone can complete various types of shooting and detection work at the appropriate height.

[0061] S33. The drone creates a blank new topological network diagram and generates nodes corresponding to the electric poles in the new topological network diagram.

[0062] S34. The UAV is adjusted to a preset posture at the reference point and takes photos at the node position to obtain a node position image.

[0063] When the drone arrives at the reference point of the electric pole, it will first create a new blank topological network diagram in its own system and generate nodes corresponding to the electric pole. The so-called "topological network diagram" refers to a digital diagram used to record the relationship between electric poles and their connections, which can more intuitively present the spatial structure and relationship of power lines. At this time, since the drone does not know the specific situation of the new and old wire branches in the actual environment at the beginning, it can start to record the nodes and related information seen in real time by creating such a blank map, which is equivalent to laying a scalable basic framework for subsequent inspection data. For example, if the drone arrives at the electric pole at position A, it will first record a "node A" on the new topological map and use it as the starting point for subsequent connection or node expansion. In this way, even if other poles or branches that have not been included are found during the subsequent inspection process, the records can be naturally expanded in this new map, avoiding confusion or redundancy in the inspection data.

[0064] After the nodes are generated in the new topology, the drone will shoot at the reference point according to the pre-set target flight attitude. The so-called "flight attitude" is usually provided by sensors such as gyroscopes in real time, including the pitch angle, yaw angle and roll angle of the drone. These data will be used by the system to adjust the drone to a specific angle to ensure that the shooting direction or the camera's top-down angle is consistent with the pre-set standard. For example, if a certain electric pole is best photographed from 10 degrees from directly above to clearly see the crossarm and the branch wires, then after reading the gyroscope data, the drone will automatically make fine adjustments and finally hover at the predetermined attitude before shooting.

[0065] 35. The drone uses the pre-recorded node image information as a priori conditions to identify the wires in the node location image.

[0066] Optionally, the S35 includes the following steps S351-S356: S351. Obtain pre-recorded node image information; wherein the pre-recorded node image includes a utility pole and a plurality of wires, and the pre-recorded node image is photographed at a reference point corresponding to the utility pole in a preset posture.

[0067] S352. Obtain a node position image, wherein the node position image is taken at a reference point corresponding to the utility pole in a preset posture.

[0068] S353. Align the node position image with the pre-recorded node image; S354. Performing wire detection on the node position image to identify thin and long lines or arcs; S355. Perform feature analysis on the detected line segments, and merge approximately continuous short line segments to obtain the wire contour, wherein the feature analysis includes length analysis, angle analysis, and continuity analysis.

[0069] S356. In the corresponding registered coordinate system, the starting point, end point, direction and length of the line segments in the two photos are compared, and when the similarity or matching degree meets the set threshold, it is determined to be the same wire; wherein the matching degree includes the endpoint distance, the average offset, and the angle difference; If no matching object can be found or the matching degree is low, it is determined that new wires are added or reduced; the new wires correspond to wires that exist in the node position image but do not exist in the pre-recorded node image; reduced wires correspond to wires that do not exist in the node position image but exist in the pre-recorded node image.

[0070] In this process, it is first necessary to obtain the pre-recorded node image information in S351, that is, the historical image taken by the drone at the reference point corresponding to the pole in a preset posture. Usually these images are stored together with metadata such as the shooting angle and flight altitude for subsequent comparison. When the drone flies to the same reference point again, the newly shot node image can be compared with this pre-recorded information. The reason why it is called a "prior condition" is that this image records the original layout of wires and poles, which can provide a direct reference for judging whether the wires are increased or decreased. For example, if there were only two wire branches on a certain pole, and now three or only one appears in the new image taken, that is, two of the three will still be in the predicted position, or the remaining one will be in one of the predicted two positions, then the system can more accurately determine the actual change in the number of wires based on the change of wires at that position.

[0071] In S352, the drone will obtain the current node position image under the same or similar flight attitude and use it as a new data source for comparison. To achieve this, the drone will calibrate the attitude in advance according to the established shooting angle and flight altitude to ensure that the "node position image" maintains a high degree of similarity with the pre-recorded image in terms of viewing angle and resolution. The advantage of this is that the error between the two photos in subsequent comparison is significantly reduced, while improving the accuracy of wire contour matching.

[0072] Then enter S353 to perform image registration on the node position image and the pre-recorded node image. The so-called "registration" refers to making precise corrections in translation, rotation or scale of the two photos based on several detectable feature points in the image (such as the obvious structure on the top of the pole, recognizable landmarks in the background, etc.) so that they overlap as much as possible in the coordinate system. If the shooting position and posture are almost exactly the same, this step of adjustment usually only requires fine-tuning; if there is a certain tilt or distance error in the shooting, the feature points can be detected and matched through algorithms such as SIFT and ORB, and homography transformation or other geometric correction methods can be applied to complete the alignment. For example, if dozens of matching feature points are found in the two images, the system will infer how to scale, rotate or translate one of the images based on these points so that the two images are highly overlapped at the pixel level.

[0073] When arriving at S354, the drone needs to perform wire detection on the currently captured node location image to identify the thin and long lines or curves therein. In actual operation, traditional image processing algorithms such as Hough transform can be used, or deep learning target detection models can be used to segment wires. Regardless of which algorithm is used, the ultimate goal is to obtain line segments or arc areas that may represent wires. In order to improve the accuracy of detection, the system often performs preliminary filtering or aggregation on the detection results, and removes stray short line segments or noise points in order to retain a more continuous wire shape.

[0074] In S355, the system performs further feature analysis and merging of the detected wire segments. Feature analysis usually includes quantifying indicators such as the length, direction (angle), and continuity of the segments; when several segments can be combined into a longer continuous wire in terms of angle and endpoint position, a merging operation will be performed to avoid recording multiple parts of a wire as multiple independent segments. For example, in a top view, several short lines that are closely connected to each other may be detected. If the splicing angles between them are not much different, and the distance between adjacent endpoints is within a certain threshold range, the system will merge them into a wire outline. Through this aggregation, subsequent comparison and judgment can more accurately lock the overall shape of the wire.

[0075] Finally, at S356, the system will perform substantial line segment matching on the two registered images, including comparing key information such as the starting point, end point, length, direction, angle, etc. of the line segment. If the matching degree of the two (such as endpoint offset, angle difference, average distance error, etc.) is within the preset threshold, it can be determined that this is the same wire; if the corresponding matching object cannot be found or the matching degree is very low, it is determined to be a "new" or "reduced" wire. New wires mean line segments that exist in the current image but are absent in the pre-recorded image. They may be connected to newly built poles or new branches extended from the original poles; reduced wires mean line segments that existed before but are no longer visible in the captured image, which are inferred to be disconnected or removed.

[0076] It should be noted that in this scheme, the reason why the drone must shoot at a specific reference point and angle is to achieve a one-to-one direct comparison between the captured node image and the pre-stored "same pole, same angle" image. Compared with fuzzy search in multiple photos, this "constant angle, fixed position" shooting method can minimize the impact of environmental variables and make the two images more consistent in terms of perspective, scale, shape distortion, etc. In this way, the matching algorithm can determine whether the image is indeed the same pole and its attached wires, or whether a new connection has appeared, at a lower cost and with higher accuracy. For example, if the drone's gimbal shooting angle deviates too much, the height and tilt angle of the pole in the image will be very different from the pre-recorded image, which can easily lead to the algorithm's difficulty in accurate identification or misclassification. By maintaining a consistent reference height, orientation and shooting posture, the difference between the two images mainly comes from the on-site equipment itself (such as the increase or decrease of new and old wires), rather than the distortion caused by the shooting posture. This not only reduces complex multi-image searches or large-scale matching, but also provides purer reference information for subsequent wire increase and decrease judgments.

[0077] S36. The drone generates branches corresponding to the electric wires based on the node location image, and generates nodes corresponding to the electric poles in the new topological network diagram.

[0078] In this step, the drone will generate corresponding wire branches in the new topology network diagram based on the captured node location images and the results of comparative analysis, and improve or update the corresponding pole nodes. Since the previous stage has completed the comparison of wire information in the old topology network with the current observation image, the system can easily identify those wires that have not changed, regard them as "known" and retain them directly; at the same time, it can quickly find the wires that have been added or removed, and add or remove the relevant nodes and branches in the new topology diagram accordingly. Without such a comparison, the drone is like flying in an unfamiliar environment. It is difficult to quickly determine which wires are original and which are new. All identification and generation must start from scratch, which is time-consuming and prone to omissions or confusion.

[0079] In fact, the shape of wires in real environments is often very complex. They may be in multiple parallel states, bundled and entangled, drooping due to gravity, or swaying in the wind. If there is no comparison with the old image, the system is very likely to make errors in recognition. Once the known wires are filtered out through comparison, it becomes efficient and accurate to identify those parts that have indeed changed, thereby greatly reducing the redundant process when re-building the map.

[0080] At the same time, the old topological network diagram also plays a key role in this link. It not only provides a clear flight range for the drone, so that it can know the inspection boundary without relying on GPS positioning; more importantly, the old map provides a reliable route guidance framework for the generation of the new map. Since the overall skeleton of the two is basically the same, the drone only needs to make fine adjustments on the originally set path according to the local differences detected in the field, which not only avoids "getting lost" in complex branches, but also reduces the possible repeated traversal. For example, if the new map completely abandons the old map and plans the flight independently, then each branch may be inspected back and forth more than twice, seriously wasting battery life and time; relatively speaking, if new nodes or branches are found in the new map, the range of change is usually not large, and it will not cause a fundamental impact on the overall inspection. It can be seen that by comparing the old images and the old topological data, the cost of recognition and flight can be significantly saved, and the perception accuracy of the increase and decrease of wires in a changing environment can be effectively improved.

[0081] S37. The UAV determines the branch inspection order of the node based on the new topology network diagram, and inspects the wires corresponding to the branches according to the branch inspection order and inspection strategy.

[0082] Specifically, the inspection strategies include: The drone will first inspect the branches according to the branch inspection order set by the local topology network at the node. If there are still branches that have not been inspected after the local topology network corresponding to the wire inspection of the new topology network diagram is completed, the branch will be inspected; Based on image recognition, the drone determines whether the node reached along the branch inspection is the corresponding node in the local topology network. If so, it determines whether all branches corresponding to the node have been traversed. If not, it goes to the next branch in the branch inspection order. If so, it returns to the previous node. If not: calculate the distance between the node and the center point. If the distance is less than or equal to the distance to the edge, the node is determined to be a new node in the partition. If it is greater, the drone is controlled to return to the previous node.

[0083] At this stage, the drone has identified each branch and the corresponding pole node based on the new topology network diagram, and needs to further plan its own inspection sequence to ensure that the inspection scope is comprehensive and avoid duplication. The so-called "branch inspection sequence" refers to how the drone makes a choice among many possible wire branches after arriving at a certain node to achieve a balance between efficiency and coverage. Usually, the local topology network will give priority to formulating a general branch sequence based on the original divisions and routes; and if it is found that there are still branches that are not included after the construction or addition and deletion of wires, the system will complete the established inspection tasks and then conduct a sweep inspection of the remaining branches.

[0084] In the specific implementation, the drone will first read the branch information connected to the current node in the new topology network diagram, and fly and inspect one by one in a predetermined order. When the drone arrives at the next node along a branch, it will determine whether the node it has arrived at is the known node recorded in the local network diagram based on image recognition technology. If the match is successful, it will check whether there are any branches that have not been inspected on this node. If so, it will continue to go to the next branch in the same way; if all branches of the node have been inspected, the drone will return to the previous node to avoid blindly shuttling back and forth in the network. If the image recognition result shows that the node reached by the drone is not in the existing records of the current local topology network, the system will also calculate the distance between this node and the center point to determine whether it should be regarded as a "new node". If the distance does not exceed the edge threshold, it will be added to the network diagram of the partition, otherwise the drone will determine that it is located in other partitions or is not within the effective inspection range, and thus choose to return to the previous node.

[0085] Through such a mechanism, the drone can always rely on the existing information in the new topology network map to navigate in an orderly manner during the inspection process. It can not only make full use of the inspection path framework provided by the old map, but also make flexible adjustments when new nodes or branches are discovered, reducing unnecessary repeated flights and route conflicts. For example, if there are three branches recorded on node A and they have all been inspected, then when the drone flies to an unknown node along a branch that may not be recorded, it can determine in real time after arriving whether the node needs to be included in the update of the current sub-network. If the distance is too large, the node is most likely a node in another partition (because the possibility of densely adding multiple poles in a small area within the partition is extremely low). At this time, choose to return directly to avoid wasting inspection time.

[0086] Optionally, S37 includes the following steps S371-S373.

[0087] S371. Control the drone to obtain image information, and obtain wire image information based on instance segmentation; S372. Input to a pre-trained deep learning model to identify the risk type; S373. If a risk is identified, record the branch and mark the risk type, and take a photo of the risk location.

[0088] Among them, the training steps of the pre-trained deep learning model include S3721-S3726.

[0089] S3721. Determine the types of risks that need to be identified and define the identifiable characteristics of each risk type; S3722. Obtain various types of image samples containing identifiable risk features, and manually mark the identifiable features in the images using annotation tools; S3723. Divide the image samples into a training set and a test set based on a pre-selected partitioning strategy and a partitioning ratio, wherein the training set is used to train the model, and the test set is used to verify and adjust the model, and the pre-selected partitioning strategy is random partitioning, stratified partitioning, or cross-validation; S3724. Train a CNN model using the ResNet architecture based on the training set, using the cross entropy loss function during training; S3725. Input the test set to the trained model, input the risk type identified by the trained model into a preset reward function for calculation and obtain a reward value, wherein the output of the preset reward function is related to the difference between the risk type identified by the CNN model and the actual risk type, and the greater the difference, the lower the reward value, and the smaller the difference, the smaller the reward value; S3726. Adjust the parameters of the trained model based on the reinforcement learning algorithm and the change of the reward value to increase the reward value, thereby obtaining a new risk type, substituting it into the previous step and executing it until the risk type converges to a stable range.

[0090] S4. Control the drone to generate a new partitioned topology network based on the recorded nodes and branches.

[0091] S5. After all drones complete their inspection missions and return, obtain the nodes and branches recorded by the drones, determine the common intersections generated by drones in adjacent areas, summarize the partitioned networks into a total topological network, and delete the duplicated branches and nodes.

[0092] After completing the inspection of each node and branch in the partition, the drone will organize the captured images and the real-time recorded branch connection information, and then generate a new partition topology network diagram locally. The so-called "new partition topology network diagram" here refers to the nodes, routings and their interconnected relationships obtained by the drone's current inspection in a digital form, which echoes or supplements the original topology network. For example, if the drone finds several new wires or new poles during the inspection, it will immediately record these findings in the current partition network diagram, forming an "incremental" update. The reason why the entire area is not fully photographed from a high altitude at the beginning and the entire topology network is constructed or reconstructed at one time is that due to ground environmental occlusion, cable droop and other complex factors, aerial images often make it difficult to clearly capture each wire. The branch-level inspection method can more effectively grasp the actual distribution, thereby avoiding errors or omissions to the greatest extent. At the same time, it also avoids the repeated secondary work of first building a large and rough network structure and then traversing it in detail, saving time and resources.

[0093] After all drones have completed their respective inspection tasks and returned, the system will summarize all new partition topology networks and captured images in the background. Since GPS positioning is not used (to prevent drift or signal instability caused by accuracy errors), the partition networks established by each drone during field operations and the images of the poles captured may overlap in the border areas. In order to accurately identify these overlapping or identical nodes as the same object, they need to be uniformly compared and deduplicated in the summary stage. In other words, if different drones report the same pole or branch, the system will confirm whether it is indeed pointing to the same target through image recognition and node attribute comparison. Once it matches, it will be merged into one node and duplicate records will be deleted. This ensures that the final total topology network has no redundant nodes or wire branches, and at the same time, it can also merge the differential findings reported by each drone (such as adding new wires or removing branches) into the overall topology map of the same power system. Finally, without the help of GPS, through the collaborative partition inspection of multiple drones, offline backhaul and data fusion, an accurate, detailed and up-to-date power system network structure can still be efficiently built and maintained.

[0094] An embodiment of the present application further discloses a power cable monitoring and analysis system, comprising a processor, wherein a program of any one of the above-mentioned power cable monitoring and analysis methods is running on the processor.

[0095] The embodiment of the present application further discloses a storage medium storing a program of any one of the above-mentioned power cable monitoring and analysis methods.

[0096] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A power cable monitoring and analysis method, characterized in that: The following steps are involved: S1. Obtaining an overall topological network of the power system and the number of drones, and partitioning the nodes of the overall topological network based on the number of drones to obtain multiple local topological networks, wherein the overall topological network includes nodes and branches, the nodes correspond to electric poles, and the branches correspond to electric wire branches; S2. Create inspection tasks and task data information based on the local topology network, and send them to each drone respectively. The task data information includes node information and branch information corresponding to a local topology network, and pre-recorded node image information corresponding to each node. The edge of the local topology network is a node, and adjacent local topology networks are connected through shared nodes. S3. Control the drone to go to the starting point of the corresponding partition, and then start from the starting point to traverse and record each node and branch based on the preset rules, and check the wires during the traversal process; S4. Control the UAV to generate a new partition topology network based on the recorded nodes and branches; S5. After all drones complete their inspection missions and return, obtain the nodes and branches recorded by the drones, determine the common intersections generated by drones in adjacent areas, summarize the partitioned networks into a total topological network, and delete the duplicated branches and nodes.

2. The power cable monitoring and analysis method according to claim 1, characterized in that: The S1 comprises the following steps: S11. Obtain the overall topology network of the power system and the number of drones N, and determine the branch number distribution range and node number distribution range of the local topology network based on the total number of nodes, the total number of branches and the number of drones in the overall topology network; S12. According to the number of branches and the distance constraint, each node of the overall topological network is preliminarily judged and screened, and N candidate nodes are selected as possible centers of the local topological network; S13. Taking each candidate center point as the core, absorb the ownerless nodes into their corresponding sub-networks layer by layer or in a shortest path manner, wherein the absorption speed of each candidate center is the same; S14. When the number of nodes or branches in a sub-network reaches the upper limit, stop absorbing new nodes into the sub-network; S15. Allocate the remaining nodes that have not been assigned to avoid isolated nodes or redundant sub-networks; S16. In each local topology network, check the path length from the center point to each edge node. If it exceeds or falls below a preset threshold, adjust the boundary nodes to reassign some nodes to adjacent local topology networks so that the distance from the center point to the edge gradually returns to a reasonable range. S17. Re-evaluate the center point position so that the difference between the maximum node distance and the minimum node distance from the center point to the edge in the local topology network is less than a preset threshold; wherein the center point position is used as the flight starting point of the drone in the partition; S18. Check whether the constraints of the local topology network are met and perform step rollback for adjustment.

3. The power cable monitoring and analysis method according to claim 2 is characterized in that: The S3 includes: S31. The drone goes to the starting point of the corresponding partition; S32. The drone flies to the reference point corresponding to the electric pole based on image recognition; S33. The drone creates a blank new topological network diagram, and generates nodes corresponding to the electric poles in the new topological network diagram; S34. The drone is adjusted to a preset posture at the reference point and photographed at the node position to obtain a node position image; S35. The drone uses the pre-recorded node image information as a priori condition to identify the wires in the node position image; S36. The drone generates branches corresponding to the electric wires based on the node location image, and generates nodes corresponding to the electric wire poles in the new topological network diagram; S37. The UAV determines the branch inspection order of the node based on the new topology network diagram, and inspects the wires corresponding to the branches according to the branch inspection order and inspection strategy.

4. The power cable monitoring and analysis method according to claim 3 is characterized in that: The inspection strategy includes: The drone will first inspect the branches according to the branch inspection order set by the local topology network at the node. If there are still branches that have not been inspected after the local topology network corresponding to the wire inspection of the new topology network diagram is completed, the branch will be inspected; Based on image recognition, the drone determines whether the node reached along the branch inspection is the corresponding node in the local topology network. If so, it determines whether all branches corresponding to the node have been traversed. If not, it goes to the next branch in the branch inspection order. If so, it returns to the previous node. If not: calculate the distance between the node and the center point. If the distance is less than or equal to the distance to the edge, the node is determined to be a new node in the partition. If it is greater, the drone is controlled to return to the previous node.

5. The power cable monitoring and analysis method according to claim 4, characterized in that: The S32 includes the following sub-steps: S321. When the UAV reaches the area corresponding to the starting point, it obtains an image of the electric pole at the node; S322. The drone determines the location of the electric pole in the image based on image recognition; S323. The drone selects a position at a preset height above the electric pole as a reference point; S324. The drone flies to the reference point.

6. The power cable monitoring and analysis method according to claim 5, characterized in that: The S35 comprises the following steps: S351. Obtaining pre-recorded node image information; wherein the pre-recorded node image includes a utility pole and a plurality of wires, and the pre-recorded node image is photographed at a reference point corresponding to the utility pole in a preset posture; S352. Acquire a node position image, wherein the node position image is captured at a reference point corresponding to the electric pole in a preset posture; S353. Align the node position image with the pre-recorded node image; S354. Performing wire detection on the node position image to identify thin and long lines or arcs; S355. Perform feature analysis on the detected line segments, and merge the approximately continuous short line segments to obtain the wire contour, wherein the feature analysis includes length analysis, angle analysis, and continuity analysis; S356. In the corresponding registered coordinate system, the starting point, end point, direction and length of the line segments in the two photos are compared, and when the similarity or matching degree meets the set threshold, it is determined to be the same wire; wherein the matching degree includes the endpoint distance, the average offset, and the angle difference; If no matching object can be found or the matching degree is low, it is determined that new wires are added or reduced; the new wires correspond to wires that exist in the node position image but do not exist in the pre-recorded node image; reduced wires correspond to wires that do not exist in the node position image but exist in the pre-recorded node image.

7. The power cable monitoring and analysis method according to claim 6, characterized in that: The S37 comprises the following steps: S371. Control the drone to obtain image information, and obtain wire image information based on instance segmentation; S372. Input to a pre-trained deep learning model to identify the risk type; S373. If a risk is identified, the branch is recorded and the risk type is marked, and a photo of the risk location is taken; Among them, the training steps of the pre-trained deep learning model include: S3721. Determine the types of risks that need to be identified and define the identifiable characteristics of each risk type; S3722. Obtain various types of image samples containing identifiable risk features, and manually mark the identifiable features in the images using annotation tools; S3723. Divide the image samples into a training set and a test set based on a pre-selected partitioning strategy and a partitioning ratio, wherein the training set is used to train the model, and the test set is used to verify and adjust the model, and the pre-selected partitioning strategy is random partitioning, stratified partitioning, or cross-validation; S3724. Train a CNN model using the ResNet architecture based on the training set, using the cross entropy loss function during training; S3725. Input the test set to the trained model, input the risk type identified by the trained model into a preset reward function for calculation and obtain a reward value, wherein the output of the preset reward function is related to the difference between the risk type identified by the CNN model and the actual risk type, and the greater the difference, the lower the reward value, and the smaller the difference, the smaller the reward value; S3726. Adjust the parameters of the trained model based on the reinforcement learning algorithm and the change of the reward value to increase the reward value, thereby obtaining a new risk type, substituting it into the previous step and executing it until the risk type converges to a stable range.

8. A power cable monitoring and analysis system, characterized in that: It comprises a processor, in which a program of the power cable monitoring and analysis method as claimed in any one of claims 1 to 7 is run.

9. A storage medium, characterized in that: A program storing the power cable monitoring and analysis method as described in any one of claims 1 to 7 is provided.

Images