Unmanned aerial vehicle power distribution network autonomous inspection path planning method and system adaptive to unmanned aerial vehicle nest

Abstract

The invention discloses an unmanned aerial vehicle power distribution network autonomous routing inspection path planning method and system adaptive to an unmanned aerial vehicle nest, and relates to the technical field of unmanned aerial vehicles. Firstly, in a data preparation and cache stage, routing inspection coordinate point information and an initial position of a nest vehicle are extracted from various power grid diagrams of a power distribution network information system and are accessed to a traffic data interface; constructing a quadtree data structure, initializing a linked list, calculating and caching distance information and the like; constructing an initial loop path, and combining multiple factors to select a minimum weight scheme insertion point to form a complete loop path; segmenting a ring path, randomly segmenting for multiple times by using a Monte Carlo algorithm, and evaluating through a loss function; a path is optimized through a simulated annealing algorithm, path adjustment is accepted or rejected according to a criterion, and temperature convergence is updated to a better solution; and finally, performing final path cutting and outputting, and uniformly distributing to the multiple unmanned aerial vehicles. The system comprises a server and a processor, executes related steps and outputs a complete inspection path plan for a ground control center to monitor and dispatch.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular to a method and system for autonomous inspection path planning of an unmanned aerial vehicle distribution network adapted to an unmanned aerial vehicle nest. Background Art

[0002] With the continuous development of the distribution network, the operation and maintenance of the distribution network faces complex situations such as dense lines, numerous branches, various types of poles and towers, and small connection fittings. These factors make the inspection task arduous, and the traditional manual inspection methods, such as naked eye observation, telescope inspection, and handheld infrared instrument detection, not only have blind spots in vision, but also rely on personal experience to identify line defects, resulting in low inspection efficiency and quality. Against the background of the growing development of Internet of Things technology, the role of drones in power system inspections has received more and more attention. Drone inspections can replace traditional manual inspections, thereby reducing labor input, reducing the cost of line operation and maintenance, and improving the precise control of line operation status, enhancing the system's ability to resist risks. However, despite the obvious advantages of drones in power inspections, their application still faces some challenges.

[0003] In terms of data collection, in the early days, it often only relied on simple tower coordinate information or some key equipment location data, lacking comprehensive and detailed information integration of the distribution network. At the same time, the application of drone parameters such as flight speed and node inspection time is not accurate enough, and the relevant information of drone nest vehicles is rarely considered, especially in terms of the dynamic position changes of nest vehicles and the impact of traffic conditions on their movement. There is almost no consideration of path planning, resulting in poor coordination between nest vehicles and drones during actual inspections, which increases the overall inspection time and cost.

[0004] In the path construction stage, traditional methods mostly use relatively simple rules, such as randomly selecting the starting point and then connecting the inspection points in sequence, or building the path based on a fixed priority order. This method does not fully consider the relationship between the inspection point and the initial position of the vehicle in the machine nest, which can easily cause the path to be circuitous, which not only wastes the energy of the drone, but also prolongs the inspection cycle. Moreover, when faced with large-scale and complex distribution network inspection tasks, there is a lack of effective data structures to quickly locate and process the relationship between inspection points, making the path construction inefficient and difficult to meet the growing demand for power inspections.

[0005] In the path optimization link, although some existing technologies attempt to optimize, the optimization strategy is relatively simple. Only the path length is used as the only optimization target, ignoring important factors such as the difference in inspection time of drones at different nodes, the scheduling cost of the vehicle nest, and the load balance of the path. This may cause the drone tasks on some paths to be too heavy, while other path resources are idle, reducing the reliability and stability of the entire inspection system. At the same time, due to the lack of an effective global optimization algorithm, it is easy to fall into a local optimal solution and cannot find the truly optimal path combination.

[0006] To sum up, the existing UAV distribution network inspection path planning technology has shortcomings and is difficult to adapt to the high precision, high efficiency and high reliability requirements of modern complex distribution network inspections. Therefore, there is an urgent need for a more advanced and comprehensive UAV distribution network autonomous inspection path planning method and system that is adapted to UAV nests to solve these problems. Summary of the invention

[0007] The purpose of the present invention is to collect multi-source data and construct data structure and cache distance, then construct weighted initial loop path, and then use Monte Carlo algorithm to split the loop path and select the best one, and then optimize the path with simulated annealing algorithm, and finally check and cut the overloaded path and distribute it evenly to multiple UAVs. The supporting system stores the program, executes the steps, outputs the plan and supports monitoring scheduling and real-time adjustment and optimization to solve the above problems.

[0008] To achieve the above object, the present invention provides the following technical solutions: A method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest, characterized by comprising the following steps: S1: Data preparation and caching: Collect and organize the required data for the entire inspection planning algorithm, build a suitable data structure, and pre-calculate and cache the distance to provide basic information support for subsequent path planning.

[0009] S2: Build an initial loop path: Based on the previously prepared data, start building a preliminary inspection path framework, which covers all points that need to be inspected in the form of a ring, and then gradually optimize it later.

[0010] S3: Split the loop path; split the constructed initial loop path into multiple sub-paths according to a certain strategy to better allocate tasks and perform subsequent optimization.

[0011] S4: Simulated annealing path optimization: Use the simulated annealing algorithm to optimize the segmented path, and iteratively search for a better path combination to balance various costs and constraints.

[0012] S5: Final path cutting and output. Check whether the optimized path meets the working capacity limit of the drone. If not, make cutting adjustments and assign the finalized path to each drone, and output the relevant inspection plan information.

[0013] Step S1 data preparation and caching specifically includes the following steps: S11: Data collection, extracting the coordinate point information of all power grid equipment that need to be inspected from the power grid map of the distribution network information system, the power grid map includes a single-line topology map, a geographic GIS map, and a tower 3D map. The coordinate points cover the tower location, key equipment installation points, and preset inspection points. At the same time, determine the number of drones used in the inspection task, obtain the accurate flight speed parameters of each drone and the inspection time of each node, collect the initial position information of the drone nest vehicle, and access the real-time traffic status data interface to obtain the estimated travel time of different sections. The drone nest vehicle may be affected by traffic conditions during movement, so as to more accurately estimate its round-trip cost and time schedule; S12: Data structure construction, based on the collected two-dimensional coordinate point information, a quadtree data structure is constructed, and the initial position of the movable drone nest vehicle is added as a special node; for each node i in the quadtree, the coordinate point information, child node pointer and distance to the initial position of the drone nest vehicle are recorded; at the same time, space is reserved for each node to record the real-time distance information from the dynamic position of the drone nest vehicle, and a bidirectional circular linked list is initialized to store path information; S13: Distance cache calculation, using the Euclidean distance formula to calculate the distance between all coordinate points: , in, and is the value of any two coordinate points; The calculation results are stored in a hash table, with the coordinate point pair as the index and the distance value as the storage content. In the subsequent path construction and optimization algorithm, when the distance between two points needs to be obtained, it is queried from the cache to avoid repeated distance calculations; at the same time, the distance d from each coordinate point to the initial position of the drone nest vehicle is calculated and stored. i,n0 , initialize an array for storing the dynamic position information of the drone nest vehicle, record the position coordinates of the drone nest vehicle at different time points and the corresponding timestamps, and initially only contain the initial position of the drone nest vehicle (x n0 ,y n0 ) and initial time t0.

[0014] Step S2 constructs the initial loop path and specifically includes the following steps: S21: Initialize the path segment and calculate the distance weight from each inspection point to the initial position of the drone nest vehicle: , in, is the distance weight, d i,n0 is the distance from the coordinate point to the initial position of the drone nest vehicle; In the set of all coordinate points that need to be inspected, two different coordinate points p1 and p2 are randomly selected by manual designation or random number generator, and these two points are used as the starting point and end point of the initial path segment. A new path node sequence is created, and the nodes corresponding to p1 and p2 are inserted into the sequence in sequence. At the same time, the predecessor pointer of the p1 node is set to point to the p2 node, and the successor pointer of the p2 node is set to point to the p1 node, forming a simple closed-loop path segment; S22: Loop to build the path and enter the main loop. The loop condition is that there are still coordinate points that have not been included in the current path. In each loop iteration, a coordinate point p is randomly selected from the set of coordinate points that have not been selected. i , using the constructed quadtree data structure, quickly find p i The K nearest neighbors in the current path.

[0015] The specific operation of step S22 loop construction path is as follows: S221: Starting from the root node of the quadtree, according to p i The coordinate position of p is recursively searched for i The smallest sub-region of the node with a distance p is selected from the path nodes in this sub-region and the adjacent sub-regions. i The nearest K points; S222: For each p of the K neighbor points found n Try to p i Insert into p n The previous and next positions of the node are used to form new candidate paths, and the total path weight of each candidate path is calculated. The calculation of the total path weight is based on the unit path weight of each point. : , Among them, v d is the flight speed of the drone, t d It takes time to inspect each node; S223: Based on the distance relationship between the path and the initial position of the drone nest vehicle and the estimated round trip cost W after considering the dynamic movement of the drone nest vehicle, the specific calculation formula is: , in, Unit path weight for each point , m is the number of path nodes after insertion, is a parameter that adjusts the initial distance weight, d i,n0 is the distance from the i-th point in the path to the initial position of the drone nest vehicle, is the parameter that adjusts the dynamic round trip cost of the drone nest vehicle, C i,n (t) is the estimated round-trip cost from the i-th point in the path to the drone nest vehicle at time t; S224: Compare the total weights of all candidate solutions, select the insertion solution with the smallest total weight, and i Insert into the corresponding position in the current path and update the predecessor and successor pointer relationship between the path nodes; S225: Repeat the above cycle steps until all coordinate points are successfully inserted into the path, and a closed loop path covering all points to be inspected is formed.

[0016] Step S3 of segmenting the ring path specifically includes the following steps: S31: random segmentation attempt, setting a larger number of iterations N to control the exploration depth of the Monte Carlo algorithm, performing a random segmentation operation on the currently constructed loop path in each iteration, specifically randomly selecting M-1 segmentation points on the loop path, where M is the number of segmentation paths set in advance, and segmenting the loop path into M sub-paths; S32: Loss function evaluation. For each of the M paths obtained by segmentation, the load of each path is calculated. The calculation of the path load takes into account the number of points in the path and the weight of each point. The formula is: , in, is the number of points in the jth path, The weight of the i-th point in the path; Calculate the sum of all path loads L sum and the standard deviation of the path load ; Calculate the average distance D of the drone from the last inspection point to the nearest drone nest vehicle position in each segmented path avg (t), for the last point of each segmented path, according to the dynamic location information data structure of the drone nest vehicle based on the real-time traffic status data, the location of the drone nest vehicle when the drone is expected to arrive at the point is obtained, the distance to the location is calculated, and then the average value is obtained; at the same time, according to the energy consumption model of the drone nest vehicle and the driving path and speed, the driving cost E of the drone nest vehicle corresponding to each segmented path is calculated nest (t); According to the set loss function: , in, To adjust the parameters of the standard deviation, To adjust the parameters affecting the distance of the vehicle returning to the drone nest, To adjust the parameters affecting the driving cost of the drone nest vehicle, the loss value of the segmentation is calculated; S33: Select the optimal segmentation. After completing N segmentation attempts, compare the loss function value of each segmentation, and select the segmentation method with the smallest loss value as the optimal segmentation result of this Monte Carlo algorithm. The path set corresponding to the optimal segmentation result is used as the starting path for subsequent simulated annealing algorithm iterations, and the quadtree TE is adjusted according to these paths. Specifically, the points belonging to each path are stored in the corresponding TE tree and the points that do not belong to each path are stored in the TE tree, so as to prepare data for subsequent path optimization calculations.

[0017] Step S4 simulated annealing optimization path specifically includes the following steps: S41: Initialize parameters, set the initial temperature T0 of the simulated annealing algorithm, and set the annealing rate , used to control the temperature drop rate during the iteration process, and a smaller probability value p is used to determine whether the temperature rise strategy is hit, setting the parameters that adjust the distance change , adjust the parameters affecting the dynamic round trip cost of the drone nest vehicle ; S42: Main loop iteration, entering the main loop of the simulated annealing algorithm, the loop condition is that the temperature drops below a preset threshold.

[0018] Step S42 main loop iteration specifically includes the following steps: In each loop iteration: S421: First, determine whether the temperature increase strategy in the annealing algorithm is hit with probability p. If the temperature increase strategy is hit, perform the following operations: S422: Randomly select a path p from the current path set r , for path p r Every point p in k , use the quadtree to find its nearest K neighbor points; S423: For each neighbor point p n , try inserting it into p k The previous and next positions are used to form new candidate paths, and the growth value of the path load of each candidate is calculated. , the change of the distance from the path to the drone nest vehicle And the change of dynamic round-trip cost of drone nest vehicle ; S424: Define a comprehensive evaluation metric Select the candidate with the smallest E. If the smallest E is less than 0, that is, the comprehensive cost of the path is reduced after insertion, then directly insert the neighbor point p n Delete it from its original path and add it to path p r If the minimum E is greater than or equal to zero, the insertion operation is accepted with a certain probability according to the simulated annealing acceptance criterion; the simulated annealing acceptance criterion formula is: , in is the change value of the comprehensive cost of the path, T is the current temperature, that is, a random number is generated ,if , then accept the insertion operation, otherwise give up; S425: If the warm-up strategy is not hit, a path with a 50% probability is selected according to the principle of path with larger load: First, calculate the load L of each path j , and then softmax the load to get the selection probability: , Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the calculated probability s ; S426: For path P s Every point P in m , use the quadtree to find the nearest neighbor point that does not belong to the path, calculate the distance to these neighbor points, and select the neighbor point p with the shortest distance n , p n Remove from its original path and add to path P s At the same time, the change in the distance from the inserted path to the drone nest vehicle is calculated , and the dynamic round-trip cost change of the drone nest vehicle If the comprehensive cost increases too much, a path with a smaller path load is selected with a probability of 50% according to the simulated annealing acceptance criterion: first negate the load of each path and then perform softmax to obtain the selection probability: , Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the probability t ; S427: For path Pt Every point P in q , find the nearest neighbor point that does not belong to the path and perform the above removal and addition operations, taking into account the distance change , Dynamic round-trip cost changes of drone nest vehicles and acceptance criteria; S428: recalculate the overall loss function value Loss according to the new path load condition, distance change and dynamic round-trip cost change of the drone nest vehicle; Regardless of which random selection operation is used, after completing the transfer of neighbor points, recalculate the load of the path that accepts the neighbor point transfer and the path that has common points with the path, as well as the change in the distance from the overall path to the drone nest vehicle and the change in the dynamic round-trip cost of the drone nest vehicle, in the same way as the loss function Loss calculation set above; S429: according to the changes of the current temperature T and the loss function value, decide whether to accept the new path adjustment according to the simulated annealing acceptance criteria; if accepted, update the path set; if not accepted, keep the original path set unchanged; S4210: Update temperature That is, the temperature is lowered according to the annealing rate, the exploration scope of the algorithm in subsequent iterations is reduced, and it gradually converges to a better solution. Check whether the loop end condition is met. If not, continue to the next loop iteration; if satisfied, the iterative process of the simulated annealing algorithm ends.

[0019] Step S5: Final path cutting and output specifically includes the following steps: S51: Path cutting check, traverse all paths after simulated annealing optimization, for each path p i Check whether its total load exceeds the maximum daily workload W of the drone max ,If the load of a path exceeds the maximum workload, the path needs to be cut; S52: Cutting operation, using a greedy strategy to cut the path, starting from the starting point of the path, gradually accumulating the weights of the points on the path until the accumulated sum is close to but does not exceed the maximum daily workload of the drone W max When the point is taken as a candidate cutting point, for each candidate cutting point, the distance d from the end point of the path after cutting to the dynamic moving estimated position of the UAV nest vehicle is calculated. end,n (t) and the cost E of the drone nest vehicle to travel to that location nest (t), define a cutting point evaluation function: , Where L sub is the load of the subpath after cutting, is the parameter that adjusts the distance effect, dend,n (t) is the distance from the end point of the path after cutting to the estimated position of the drone nest’s dynamic movement, It is a parameter that adjusts the impact of the UAV's nest vehicle driving cost. The candidate cutting point with the smallest F value is selected for cutting, and the path is divided into two sub-paths. This process is repeated until all paths meet the UAV's maximum daily workload limit; S53: Multi-UAV task allocation and output, the cut path is divided into the number of UAVs n d The task load of each UAV is balanced. The task volume is measured according to factors such as the length, number of points or load of the path. Paths with smaller task volumes are preferentially allocated to UAVs with larger task volumes, so as to gradually balance the workload of each UAV. Specifically, the task volume index, path length and load of each sub-path after cutting are calculated, and the paths are sorted from small to large according to the task volume. Then, the paths are allocated to UAVs in turn. Each time the UAV with the smallest total task volume is selected for allocation, the total task volume of the UAV is updated.

[0020] A drone distribution network autonomous inspection path planning system adapted to drone nests, characterized in that it includes a server and a processor, wherein the server stores a system program, and the processor implements the steps of any one of the methods described in claims 1 to 8 when executing the computer program.

[0021] The system outputs a complete inspection path plan for each UAV, including a sequence of path coordinate points, estimated flight time, and estimated arrival time at each inspection point based on real-time traffic status data, estimated return time to the UAV nest vehicle, and location information of the UAV nest vehicle at each key time point. This information is stored in a database for the ground control center to monitor and schedule tasks. During the inspection process, the UAV can communicate with the ground control center based on the actual flight conditions, and the ground control center can adjust and optimize the path plan based on real-time information to ensure the smooth completion of the inspection task.

[0022] The working mechanism of the autonomous inspection planning method and system of the drone distribution network adapted to the drone nest is: First, in the data preparation and caching stage, the inspection coordinate point information covering poles, key equipment and preset inspection points is extracted from various power grid diagrams of the distribution network information system, the number of drones is determined, and their flight speed and node inspection time are obtained. At the same time, the initial position of the machine nest vehicle is collected and connected to the traffic data interface. Based on these data, a quadtree data structure is constructed, the initial position of the machine nest vehicle is set as a special node, the detailed information of each node is recorded and the dynamic distance information space is reserved, the bidirectional circular linked list is initialized, and the Euclidean distance formula is used to calculate and cache the distance between all coordinate points and the distance from each point to the initial position of the machine nest vehicle, and the dynamic position information array of the machine nest vehicle is recorded.

[0023] When constructing the initial loop path, first calculate the distance weight from each inspection point to the initial position of the machine nest vehicle, and randomly select two coordinate points to construct a simple closed loop path segment as the starting framework. Then enter the loop construction, randomly select one coordinate point that has not yet been included in the path, and use the quadtree to quickly find its nearest neighbor points in the current path. Consider the unit path weight, the distance relationship with the initial position of the machine nest vehicle, and the estimated round-trip cost after the dynamic movement of the machine nest vehicle. Calculate the total path weight of each candidate solution, select the insertion solution with the smallest total weight to insert the point into the path, and repeat until all coordinate points are successfully included to form a complete initial loop path.

[0024] Then, the ring path segmentation operation is performed, and a larger number of iterations is set. The Monte Carlo algorithm randomly selects several segmentation points for each iteration of the ring path to segment it into a specified number of sub-paths. For each sub-path segmented, the path load, load standard deviation, average distance from the last inspection point to the nearest vehicle location, and vehicle travel cost are calculated in depth. The loss value of the segmentation is calculated based on a carefully set loss function. After completing multiple segmentation attempts, the loss function value is compared, and the segmentation method with the smallest loss value is selected as the optimal segmentation result. The quadtree data structure is adjusted accordingly to provide a good data foundation for the subsequent simulated annealing algorithm.

[0025] In the simulated annealing path optimization link, the key parameters of the simulated annealing algorithm are initialized first, including the initial temperature, annealing rate, probability value of the heating strategy, and parameters affecting the distance and dynamic round-trip cost of the machine nest vehicle. Entering the main loop iteration, each loop first determines whether the heating strategy is hit based on probability. If it is hit, a path is randomly selected, and the nearest neighbor points of each point in the path are found by using the quadtree. Candidate solutions are formed for the front and back positions of the neighbor points, and the path load growth value, the change in the distance to the machine nest vehicle, and the change in the dynamic round-trip cost of the machine nest vehicle are calculated. Through comprehensive evaluation index screening and based on the simulated annealing acceptance criteria, it is decided whether to accept the insertion operation; if the heating strategy is not hit, the path is selected according to the principle of path load size with a certain probability, and the neighbor points are also found for removal and addition operations, and the relevant changes are calculated and the acceptance criteria are considered. Regardless of the operation, after completion, the load of the path that accepts the transfer of neighbor points and its related paths, the distance to the machine nest vehicle, and the dynamic round-trip cost changes of the machine nest vehicle are recalculated, and the overall loss function value is recalculated based on the new situation. Then, according to the current temperature and the change in the loss function value, the simulated annealing acceptance criterion is used to decide whether to accept the new path adjustment. If accepted, the path set is updated, and then the temperature is reduced according to the annealing rate. The iteration is continued until the temperature drops below the preset threshold, thereby gradually converging to a better solution.

[0026] Finally, the final path cutting and output step is to carefully traverse all the paths after simulated annealing optimization, check whether the total load of each path exceeds the maximum workload of the drone in a single day, and use a greedy strategy to cut once the load of a path exceeds the limit. The weights of the points are gradually accumulated from the starting point of the path. When it is close to but not exceeding the maximum workload of the drone in a single day, the candidate cutting point is determined. The distance from the end point of the path to the estimated position of the dynamic movement of the machine nest vehicle after cutting each candidate cutting point and the cost of the machine nest vehicle to travel to that position are calculated. The optimal candidate cutting point is selected according to the cutting point evaluation function to split the path into two sub-paths. This process is repeated until all paths meet the workload limit. After that, the cut paths are allocated according to the number of drones, and the task volume is measured by comprehensively considering factors such as the length, number of points or load of the path. Paths with smaller task volumes are preferentially allocated to drones with larger task volumes to achieve balanced task loads for each drone.

[0027] The supporting inspection planning system includes a server and a processor. The server stores the system program, and the processor implements the various steps of the above method when executing the program. The system can output a detailed inspection path plan for each drone, covering the path coordinate point sequence, estimated flight time, estimated arrival time at each inspection point based on real-time traffic status data, estimated return time to the machine nest vehicle, and location information of the machine nest vehicle at various key time points. This information is stored in the database for the ground control center to monitor and schedule tasks at any time. During the entire inspection process, the drone can communicate with the ground control center in real time according to the actual flight situation, and the ground control center adjusts and optimizes the path plan based on real-time information, ensuring that the inspection task can be completed smoothly and efficiently in all aspects, and effectively ensuring the stable operation and safe maintenance of the distribution network.

[0028] Compared with the prior art, the present invention has the following beneficial effects: In terms of data utilization, by comprehensively collecting coordinate point information in various types of distribution network maps and integrating multiple data such as drones, machine nest vehicles and traffic conditions, a more accurate planning model that fits the actual scenario can be constructed, taking into account the dynamic traffic impact of machine nest vehicles.

[0029] In the early stage of path construction, the distance weight from the inspection point to the initial position of the vehicle in the machine nest is comprehensively considered when constructing the initial ring path, and the quadtree is used to efficiently find neighbor points to build the path, which can form the initial path framework faster and more reasonably, reduce unnecessary path detours, and improve overall efficiency.

[0030] The segmentation loop path link uses the Monte Carlo algorithm to perform multiple random segmentation attempts and loss function evaluations. This approach based on a large number of attempts and comprehensive evaluation indicators provides a better starting point for subsequent path optimization and better balances factors such as the load of each sub-path and the interaction cost with the machine nest vehicle.

[0031] In the simulated annealing path optimization stage, by setting parameters and combining the heating strategy with multiple path selection rules, a balance is achieved between exploring new paths and accepting better solutions, effectively avoiding falling into the local optimal solution while taking into account various complex constraints and cost factors.

[0032] In the final path cutting and output part, a greedy strategy is used in combination with a cutting point evaluation function that takes into account the dynamic position of the vehicle in the machine nest to reasonably cut the overloaded path. In the multi-UAV task allocation, tasks are evenly allocated based on a variety of factors to make the workload of each UAV more balanced, extend the overall service life of the UAV, and improve the reliability and sustainability of the inspection tasks. At the same time, the system also has complete monitoring and scheduling and real-time adjustment and optimization functions to ensure that the inspection tasks can flexibly respond to various emergencies in actual flight. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The present invention is a flow chart of a method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention will be fully described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0035] like Figure 1 As shown, A method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest, characterized by comprising the following steps: S1: data preparation and caching; S2: construct the initial loop path; S3: segmentation loop path; S4: simulated annealing optimization path; S5: Final path cutting and output.

[0036] Step S1 data preparation and caching specifically includes the following steps: S11: Data collection, extracting the coordinate point information of all power grid equipment that need to be inspected from the power grid map of the distribution network information system, the power grid map includes a single-line topology map, a geographic GIS map, and a tower 3D map. The coordinate points cover the tower location, key equipment installation points, and preset detection points. At the same time, determine the number of drones used in the inspection task, obtain the accurate flight speed parameters of each drone and the inspection time of each node, collect the initial position information of the drone nest vehicle, and access the real-time traffic status data interface to obtain the estimated travel time of different sections; S12: Data structure construction, based on the collected two-dimensional coordinate point information, a quadtree data structure is constructed, and the initial position of the movable drone nest vehicle is added as a special node; for each node i in the quadtree, the coordinate point information, child node pointer and distance to the initial position of the drone nest vehicle are recorded; at the same time, space is reserved for each node to record the real-time distance information from the dynamic position of the drone nest vehicle, and a bidirectional circular linked list is initialized to store path information; S13: Distance cache calculation, using the Euclidean distance formula to calculate the distance between all coordinate points: , in, and is the value of any two coordinate points; The calculation results are stored in a hash table, with the coordinate point pair as the index and the distance value as the storage content. In the subsequent path construction and optimization algorithm, when the distance between two points needs to be obtained, it is queried from the cache to avoid repeated distance calculations; at the same time, the distance d from each coordinate point to the initial position of the drone nest vehicle is calculated and stored. i,n0 , initialize an array for storing the dynamic position information of the drone nest vehicle, record the position coordinates of the drone nest vehicle at different time points and the corresponding timestamps, and initially only contain the initial position of the drone nest vehicle (x n0 ,y n0 ) and initial time t0.

[0037] Step S2 constructs the initial loop path and specifically includes the following steps: S21: Initialize the path segment and calculate the distance weight from each inspection point to the initial position of the drone nest vehicle: , in, is the distance weight, d i,n0 is the distance from the coordinate point to the initial position of the drone nest vehicle; In the set of all coordinate points that need to be inspected, two different coordinate points p1 and p2 are randomly selected by manual designation or random number generator, and these two points are used as the starting point and end point of the initial path segment. A new path node sequence is created, and the nodes corresponding to p1 and p2 are inserted into the sequence in sequence. At the same time, the predecessor pointer of the p1 node is set to point to the p2 node, and the successor pointer of the p2 node is set to point to the p1 node, forming a simple closed-loop path segment; S22: Loop to build the path and enter the main loop. The loop condition is that there are still coordinate points that have not been included in the current path. In each loop iteration, a coordinate point p is randomly selected from the set of coordinate points that have not been selected. i , using the constructed quadtree data structure, quickly find p i The K nearest neighbors in the current path.

[0038] The specific operation of step S22 loop construction path is as follows: S221: Starting from the root node of the quadtree, according to p i The coordinate position of p is recursively searched for i The smallest sub-region of the node with a distance p is selected from the path nodes in this sub-region and the adjacent sub-regions. i The nearest K points; S222: For each p of the K neighbor points found n Try to p i Insert into p nThe previous and next positions of the node are used to form new candidate paths, and the total path weight of each candidate path is calculated. The calculation of the total path weight is based on the unit path weight of each point. : , Among them, v d is the flight speed of the drone, t d It takes time to inspect each node; S223: Based on the distance relationship between the path and the initial position of the drone nest vehicle and the estimated round trip cost W after considering the dynamic movement of the drone nest vehicle, the specific calculation formula is: , in, Unit path weight for each point , m is the number of path nodes after insertion, is a parameter that adjusts the initial distance weight, d i,n0 is the distance from the i-th point in the path to the initial position of the drone nest vehicle, is the parameter that adjusts the dynamic round trip cost of the drone nest vehicle, C i,n (t) is the estimated round-trip cost from the i-th point in the path to the drone nest vehicle at time t; S224: Compare the total weights of all candidate solutions, select the insertion solution with the smallest total weight, and i Insert into the corresponding position in the current path and update the predecessor and successor pointer relationship between the path nodes; S225: Repeat the above cycle steps until all coordinate points are successfully inserted into the path, and a closed loop path covering all points to be inspected is formed.

[0039] Step S3 of segmenting the ring path specifically includes the following steps: S31: random segmentation attempt, setting a larger number of iterations N to control the exploration depth of the Monte Carlo algorithm, performing a random segmentation operation on the currently constructed loop path in each iteration, specifically randomly selecting M-1 segmentation points on the loop path, where M is the number of segmentation paths set in advance, and segmenting the loop path into M sub-paths; S32: Loss function evaluation. For each of the M paths obtained by segmentation, the load of each path is calculated. The calculation of the path load takes into account the number of points in the path and the weight of each point. The formula is: , in, is the number of points in the jth path, The weight of the i-th point in the path; Calculate the sum of all path loads Lsum and the standard deviation of the path load ; Calculate the average distance D of the drone from the last inspection point to the nearest drone nest vehicle position in each segmented path avg (t), for the last point of each segmented path, according to the dynamic location information data structure of the drone nest vehicle based on the real-time traffic status data, the location of the drone nest vehicle when the drone is expected to arrive at the point is obtained, the distance to the location is calculated, and then the average value is obtained; at the same time, according to the energy consumption model of the drone nest vehicle and the driving path and speed, the driving cost E of the drone nest vehicle corresponding to each segmented path is calculated nest (t); According to the set loss function: , in, To adjust the parameters of the standard deviation, To adjust the parameters affecting the distance of the vehicle returning to the drone nest, To adjust the parameters affecting the driving cost of the drone nest vehicle, the loss value of the segmentation is calculated; S33: Select the optimal segmentation. After completing N segmentation attempts, compare the loss function value of each segmentation, and select the segmentation method with the smallest loss value as the optimal segmentation result of this Monte Carlo algorithm. The path set corresponding to the optimal segmentation result is used as the starting path for subsequent simulated annealing algorithm iterations, and the quadtree TE is adjusted according to these paths. Specifically, the points belonging to each path are stored in the corresponding TE tree and the points that do not belong to each path are stored in the TE tree, so as to prepare data for subsequent path optimization calculations.

[0040] Step S4 simulated annealing optimization path specifically includes the following steps: S41: Initialize parameters, set the initial temperature T0 of the simulated annealing algorithm, and set the annealing rate , used to control the temperature drop rate during the iteration process, and a smaller probability value p is used to determine whether the temperature rise strategy is hit, setting the parameters that adjust the distance change , adjust the parameters affecting the dynamic round trip cost of the drone nest vehicle ; S42: Main loop iteration, entering the main loop of the simulated annealing algorithm, the loop condition is that the temperature drops below a preset threshold.

[0041] Step S42 main loop iteration specifically includes the following steps: In each loop iteration: S421: First, determine whether the temperature increase strategy in the annealing algorithm is hit with probability p. If the temperature increase strategy is hit, perform the following operations: S422: Randomly select a path p from the current path set r , for path p r Every point p in k , use the quadtree to find its nearest K neighbor points; S423: For each neighbor point p n , try inserting it into p k The previous and next positions are used to form new candidate paths, and the growth value of the path load of each candidate is calculated. , the change of the distance from the path to the drone nest vehicle And the change of dynamic round-trip cost of drone nest vehicle ; S424: Define a comprehensive evaluation metric Select the candidate with the smallest E. If the smallest E is less than 0, that is, the comprehensive cost of the path is reduced after insertion, then directly insert the neighbor point p n Delete it from its original path and add it to path p r If the minimum E is greater than or equal to zero, the insertion operation is accepted with a certain probability according to the simulated annealing acceptance criterion; the simulated annealing acceptance criterion formula is: , in is the change value of the comprehensive cost of the path, T is the current temperature, that is, a random number is generated ,if , then accept the insertion operation, otherwise give up; S425: If the warm-up strategy is not hit, a path with a 50% probability is selected according to the principle of path with larger load: First, calculate the load L of each path j , and then softmax the load to get the selection probability: , Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the calculated probability s ; S426: For path P s Every point P in m , use the quadtree to find the nearest neighbor point that does not belong to the path, calculate the distance to these neighbor points, and select the neighbor point p with the shortest distance n , pn Remove from its original path and add to path P s At the same time, the change in the distance from the inserted path to the drone nest vehicle is calculated , and the dynamic round-trip cost change of the drone nest vehicle If the comprehensive cost increases too much, a path with a smaller path load is selected with a probability of 50% according to the simulated annealing acceptance criterion: first negate the load of each path and then perform softmax to obtain the selection probability: , Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the probability t ; S427: For path P t Every point P in q , find the nearest neighbor point that does not belong to the path and perform the above removal and addition operations, taking into account the distance change , Dynamic round-trip cost changes of drone nest vehicles and acceptance criteria; S428: recalculate the overall loss function value Loss according to the new path load condition, distance change and dynamic round-trip cost change of the drone nest vehicle; Regardless of which random selection operation is used, after completing the transfer of neighbor points, recalculate the load of the path that accepts the neighbor point transfer and the path that has common points with the path, as well as the change in the distance from the overall path to the drone nest vehicle and the change in the dynamic round-trip cost of the drone nest vehicle, in the same way as the loss function Loss calculation set above; S429: according to the changes of the current temperature T and the loss function value, decide whether to accept the new path adjustment according to the simulated annealing acceptance criteria; if accepted, update the path set; if not accepted, keep the original path set unchanged; S4210: Update temperature That is, the temperature is lowered according to the annealing rate, the exploration scope of the algorithm in subsequent iterations is reduced, and it gradually converges to a better solution. Check whether the loop end condition is met. If not, continue to the next loop iteration; if satisfied, the iterative process of the simulated annealing algorithm ends.

[0042] Step S5: Final path cutting and output specifically includes the following steps: S51: Path cutting check, traverse all paths after simulated annealing optimization, for each path p iCheck whether its total load exceeds the maximum daily workload W of the drone max ,If the load of a path exceeds the maximum workload, the path needs to be cut; S52: Cutting operation, using a greedy strategy to cut the path, starting from the starting point of the path, gradually accumulating the weights of the points on the path until the accumulated sum is close to but does not exceed the maximum daily workload of the drone W max When the point is taken as a candidate cutting point, for each candidate cutting point, the distance d from the end point of the path after cutting to the dynamic moving estimated position of the UAV nest vehicle is calculated. end,n (t) and the cost E of the drone nest vehicle to travel to that location nest (t), define a cutting point evaluation function: , Where L sub is the load of the subpath after cutting, is the parameter that adjusts the distance effect, d end,n (t) is the distance from the end point of the path after cutting to the estimated position of the drone nest’s dynamic movement, It is a parameter that adjusts the impact of the UAV's nest vehicle driving cost. The candidate cutting point with the smallest F value is selected for cutting, and the path is divided into two sub-paths. This process is repeated until all paths meet the UAV's maximum daily workload limit; S53: Multi-UAV task allocation and output, the cut path is divided into the number of UAVs n d The task load of each UAV is balanced. The task volume is measured according to factors such as the length, number of points or load of the path. Paths with smaller task volumes are preferentially allocated to UAVs with larger task volumes, so as to gradually balance the workload of each UAV. Specifically, the task volume index, path length and load of each sub-path after cutting are calculated, and the paths are sorted from small to large according to the task volume. Then, the paths are allocated to UAVs in turn. Each time the UAV with the smallest total task volume is selected for allocation, the total task volume of the UAV is updated.

[0043] A drone distribution network autonomous inspection path planning system adapted to drone nests, characterized in that it includes a server and a processor, wherein the server stores a system program, and the processor implements the steps of any one of the methods described in claims 1 to 8 when executing the computer program.

[0044] The system outputs a complete inspection path plan for each UAV, including a sequence of path coordinate points, estimated flight time, and estimated arrival time at each inspection point based on real-time traffic status data, estimated return time to the UAV nest vehicle, and location information of the UAV nest vehicle at each key time point. This information is stored in a database for the ground control center to monitor and schedule tasks. During the inspection process, the UAV can communicate with the ground control center based on the actual flight conditions, and the ground control center can adjust and optimize the path plan based on real-time information to ensure the smooth completion of the inspection task.

[0045] The specific implementation method is: S1: Data preparation and caching; specifically includes the following steps: S11: Data collection, extracting the coordinate point information of all power grid equipment that need to be inspected from the power grid map of the distribution network information system, the power grid map includes a single-line topology map, a geographic GIS map, and a tower 3D map. The coordinate points cover the tower location, key equipment installation points, and preset detection points. At the same time, determine the number of drones used in the inspection task, obtain the accurate flight speed parameters of each drone and the inspection time of each node, collect the initial position information of the drone nest vehicle, and access the real-time traffic status data interface to obtain the estimated travel time of different sections; S12: Data structure construction, based on the collected two-dimensional coordinate point information, a quadtree data structure is constructed, and the initial position of the movable drone nest vehicle is added as a special node; for each node i in the quadtree, the coordinate point information, child node pointer and distance to the initial position of the drone nest vehicle are recorded; at the same time, space is reserved for each node to record the real-time distance information from the dynamic position of the drone nest vehicle, and a bidirectional circular linked list is initialized to store path information; S13: Distance cache calculation, using the Euclidean distance formula to calculate the distance between all coordinate points: , in, and is the value of any two coordinate points; The calculation results are stored in a hash table, with the coordinate point pair as the index and the distance value as the storage content. In the subsequent path construction and optimization algorithm, when the distance between two points needs to be obtained, it is queried from the cache to avoid repeated distance calculations; at the same time, the distance d from each coordinate point to the initial position of the drone nest vehicle is calculated and stored. i,n0 , initialize an array for storing the dynamic position information of the drone nest vehicle, record the position coordinates of the drone nest vehicle at different time points and the corresponding timestamps, and initially only contain the initial position of the drone nest vehicle (x n0 ,yn0 ) and initial time t0.

[0046] S2: Constructing the initial loop path; specifically including the following steps: S21: Initialize the path segment and calculate the distance weight from each inspection point to the initial position of the drone nest vehicle: , in, is the distance weight, d i,n0 is the distance from the coordinate point to the initial position of the drone nest vehicle; In the set of all coordinate points that need to be inspected, two different coordinate points p1 and p2 are randomly selected by manual designation or random number generator, and these two points are used as the starting point and end point of the initial path segment. A new path node sequence is created, and the nodes corresponding to p1 and p2 are inserted into the sequence in sequence. At the same time, the predecessor pointer of the p1 node is set to point to the p2 node, and the successor pointer of the p2 node is set to point to the p1 node, forming a simple closed-loop path segment; S22: Loop to build the path and enter the main loop. The loop condition is that there are still coordinate points that have not been included in the current path. In each loop iteration, a coordinate point p is randomly selected from the set of coordinate points that have not been selected. i , using the constructed quadtree data structure, quickly find p i The K nearest neighbors in the current path.

[0047] The specific operations are: S221: Starting from the root node of the quadtree, according to p i The coordinate position of p is recursively searched for i The smallest sub-region of the node with a distance p is selected from the path nodes in this sub-region and the adjacent sub-regions. i The nearest K points; S222: For each p of the K neighbor points found n Try to p i Insert into p n The previous and next positions of the node are used to form new candidate paths, and the total path weight of each candidate path is calculated. The calculation of the total path weight is based on the unit path weight of each point. : , Among them, v d is the flight speed of the drone, t d It takes time to inspect each node; S223: Based on the distance relationship between the path and the initial position of the drone nest vehicle and the estimated round trip cost W after considering the dynamic movement of the drone nest vehicle, the specific calculation formula is: , in, Unit path weight for each point , m is the number of path nodes after insertion, is a parameter that adjusts the initial distance weight, d i,n0 is the distance from the i-th point in the path to the initial position of the drone nest vehicle, is the parameter that adjusts the dynamic round trip cost of the drone nest vehicle, C i,n (t) is the estimated round-trip cost from the i-th point in the path to the drone nest vehicle at time t; S224: Compare the total weights of all candidate solutions, select the insertion solution with the smallest total weight, and i Insert into the corresponding position in the current path and update the predecessor and successor pointer relationship between the path nodes; S225: Repeat the above cycle steps until all coordinate points are successfully inserted into the path, and a closed loop path covering all points to be inspected is formed.

[0048] S3: Split the ring path; specifically includes the following steps: S31: random segmentation attempt, setting a larger number of iterations N to control the exploration depth of the Monte Carlo algorithm, performing a random segmentation operation on the currently constructed loop path in each iteration, specifically randomly selecting M-1 segmentation points on the loop path, where M is the number of segmentation paths set in advance, and segmenting the loop path into M sub-paths; S32: Loss function evaluation. For each of the M paths obtained by segmentation, the load of each path is calculated. The calculation of the path load takes into account the number of points in the path and the weight of each point. The formula is: , in, is the number of points in the jth path, The weight of the i-th point in the path; Calculate the sum of all path loads L sum and the standard deviation of the path load ; Calculate the average distance D of the drone from the last inspection point to the nearest drone nest vehicle position in each segmented path avg (t), for the last point of each segmented path, according to the dynamic location information data structure of the drone nest vehicle based on the real-time traffic status data, the location of the drone nest vehicle when the drone is expected to arrive at the point is obtained, the distance to the location is calculated, and then the average value is obtained; at the same time, according to the energy consumption model of the drone nest vehicle and the driving path and speed, the driving cost E of the drone nest vehicle corresponding to each segmented path is calculatednest (t); According to the set loss function: , in, To adjust the parameters of the standard deviation, To adjust the parameters affecting the distance of the vehicle returning to the drone nest, To adjust the parameters affecting the driving cost of the drone nest vehicle, the loss value of the segmentation is calculated; S33: Select the optimal segmentation. After completing N segmentation attempts, compare the loss function value of each segmentation, and select the segmentation method with the smallest loss value as the optimal segmentation result of this Monte Carlo algorithm. The path set corresponding to the optimal segmentation result is used as the starting path for subsequent simulated annealing algorithm iterations, and the quadtree TE is adjusted according to these paths. Specifically, the points belonging to each path are stored in the corresponding TE tree and the points that do not belong to each path are stored in the TE tree, so as to prepare data for subsequent path optimization calculations.

[0049] S4: simulated annealing optimization path; specifically includes the following steps: S41: Initialize parameters, set the initial temperature T0 of the simulated annealing algorithm, and set the annealing rate , used to control the temperature drop rate during the iteration process, and a smaller probability value p is used to determine whether the temperature rise strategy is hit, setting the parameters that adjust the distance change , adjust the parameters affecting the dynamic round trip cost of the drone nest vehicle ; S42: Main loop iteration, entering the main loop of the simulated annealing algorithm, the loop condition is that the temperature drops below a preset threshold.

[0050] The specific steps include: In each loop iteration: S421: First, determine whether the temperature increase strategy in the annealing algorithm is hit with probability p. If the temperature increase strategy is hit, perform the following operations: S422: Randomly select a path p from the current path set r , for path p r Every point p in k , use the quadtree to find its nearest K neighbor points; S423: For each neighbor point p n , try inserting it into p k The previous and next positions are used to form new candidate paths, and the growth value of the path load of each candidate is calculated. , the change of the distance from the path to the drone nest vehicle And the change of dynamic round-trip cost of drone nest vehicle ; S424: Define a comprehensive evaluation metric Select the candidate with the smallest E. If the smallest E is less than 0, that is, the comprehensive cost of the path is reduced after insertion, then directly insert the neighbor point p n Delete it from its original path and add it to path p r If the minimum E is greater than or equal to zero, the insertion operation is accepted with a certain probability according to the simulated annealing acceptance criterion; the simulated annealing acceptance criterion formula is: , in is the change value of the comprehensive cost of the path, T is the current temperature, that is, a random number is generated ,if , then accept the insertion operation, otherwise give up; S425: If the warm-up strategy is not hit, a path with a 50% probability is selected according to the principle of path with larger load: First, calculate the load L of each path j , and then softmax the load to get the selection probability: , Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the calculated probability s ; S426: For path P s Every point P in m , use the quadtree to find the nearest neighbor point that does not belong to the path, calculate the distance to these neighbor points, and select the neighbor point p with the shortest distance n , p n Remove from its original path and add to path P s At the same time, the change in the distance from the inserted path to the drone nest vehicle is calculated , and the dynamic round-trip cost change of the drone nest vehicle If the comprehensive cost increases too much, a path with a smaller path load is selected with a probability of 50% according to the simulated annealing acceptance criterion: first negate the load of each path and then perform softmax to obtain the selection probability: , Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L jis the path load, and a path P is selected according to the probability t ; S427: For path P t Every point P in q , find the nearest neighbor point that does not belong to the path and perform the above removal and addition operations, taking into account the distance change , Dynamic round-trip cost changes of drone nest vehicles and acceptance criteria; S428: recalculate the overall loss function value Loss according to the new path load condition, distance change and dynamic round-trip cost change of the drone nest vehicle; Regardless of which random selection operation is used, after completing the transfer of neighbor points, recalculate the load of the path that accepts the neighbor point transfer and the path that has common points with the path, as well as the change in the distance from the overall path to the drone nest vehicle and the change in the dynamic round-trip cost of the drone nest vehicle, in the same way as the loss function Loss calculation set above; S429: according to the changes of the current temperature T and the loss function value, decide whether to accept the new path adjustment according to the simulated annealing acceptance criteria; if accepted, update the path set; if not accepted, keep the original path set unchanged; S4210: Update temperature That is, the temperature is lowered according to the annealing rate, the exploration scope of the algorithm in subsequent iterations is reduced, and it gradually converges to a better solution. Check whether the loop end condition is met. If not, continue to the next loop iteration; if satisfied, the iterative process of the simulated annealing algorithm ends.

[0051] S5: Final path cutting and output, specifically including the following steps: S51: Path cutting check, traverse all paths after simulated annealing optimization, for each path p i Check whether its total load exceeds the maximum daily workload W of the drone max ,If the load of a path exceeds the maximum workload, the path needs to be cut; S52: Cutting operation, using a greedy strategy to cut the path, starting from the starting point of the path, gradually accumulating the weights of the points on the path until the accumulated sum is close to but does not exceed the maximum daily workload of the drone W max When the point is taken as a candidate cutting point, for each candidate cutting point, the distance d from the end point of the path after cutting to the dynamic moving estimated position of the UAV nest vehicle is calculated. end,n (t) and the cost E of the drone nest vehicle to travel to that location nest (t), define a cutting point evaluation function: , Where L sub is the load of the subpath after cutting, is the parameter that adjusts the distance effect, d end,n (t) is the distance from the end point of the path after cutting to the estimated position of the drone nest’s dynamic movement, It is a parameter that adjusts the impact of the UAV's nest vehicle driving cost. The candidate cutting point with the smallest F value is selected for cutting, and the path is divided into two sub-paths. This process is repeated until all paths meet the UAV's maximum daily workload limit; S53: Multi-UAV task allocation and output, the cut path is divided into the number of UAVs n d The task load of each UAV is balanced. The task volume is measured according to factors such as the length, number of points or load of the path. Paths with smaller task volumes are preferentially allocated to UAVs with larger task volumes, so as to gradually balance the workload of each UAV. Specifically, the task volume index, path length and load of each sub-path after cutting are calculated, and the paths are sorted from small to large according to the task volume. Then, the paths are allocated to UAVs in turn. Each time the UAV with the smallest total task volume is selected for allocation, the total task volume of the UAV is updated.

[0052] A drone distribution network autonomous inspection path planning system adapted to drone nests, characterized in that it includes a server and a processor, wherein the server stores a system program, and the processor implements the steps of any one of the methods described in claims 1 to 8 when executing the computer program.

[0053] The system outputs a complete inspection path plan for each UAV, including a sequence of path coordinate points, estimated flight time, and estimated arrival time at each inspection point based on real-time traffic status data, estimated return time to the UAV nest vehicle, and location information of the UAV nest vehicle at each key time point. This information is stored in a database for the ground control center to monitor and schedule tasks. During the inspection process, the UAV can communicate with the ground control center based on the actual flight conditions, and the ground control center can adjust and optimize the path plan based on real-time information to ensure the smooth completion of the inspection task.

Claims

1. A method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest, characterized in that: The following steps are involved: S1: data preparation and caching; S2: construct the initial loop path; S3: segmentation loop path; S4: simulated annealing optimization path; S5: Final path cutting and output.

2. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: Step S1 data preparation and caching specifically includes the following steps: S11: Data collection, extracting the coordinate point information of all power grid equipment that need to be inspected from the power grid map of the distribution network information system, the power grid map includes a single-line topology map, a geographic GIS map, and a tower 3D map. The coordinate points cover the tower location, key equipment installation points, and preset detection points. At the same time, determine the number of drones used in the inspection task, obtain the accurate flight speed parameters of each drone and the inspection time of each node, collect the initial position information of the drone nest vehicle, and access the real-time traffic status data interface to obtain the estimated travel time of different sections; S12: Data structure construction, based on the collected two-dimensional coordinate point information, a quadtree data structure is constructed, and the initial position of the movable drone nest vehicle is added as a special node; for each node i in the quadtree, the coordinate point information, child node pointer and distance to the initial position of the drone nest vehicle are recorded; at the same time, space is reserved for each node to record the real-time distance information from the dynamic position of the drone nest vehicle, and a bidirectional circular linked list is initialized to store path information; S13: Distance cache calculation, using the Euclidean distance formula to calculate the distance between all coordinate points: ， in, and is the value of any two coordinate points; The calculation results are stored in a hash table, with the coordinate point pair as the index and the distance value as the storage content. In the subsequent path construction and optimization algorithm, when the distance between two points needs to be obtained, it is queried from the cache to avoid repeated distance calculations; at the same time, the distance d from each coordinate point to the initial position of the drone nest vehicle is calculated and stored. i,n0 , initialize an array for storing the dynamic position information of the drone nest vehicle, record the position coordinates of the drone nest vehicle at different time points and the corresponding timestamps, and initially only contain the initial position of the drone nest vehicle (x n0 ,y n0 ) and initial time t0.

3. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: Step S2 constructs the initial loop path and specifically includes the following steps: S21: Initialize the path segment and calculate the distance weight from each inspection point to the initial position of the drone nest vehicle: ， in, is the distance weight, d i,n0 is the distance from the coordinate point to the initial position of the drone nest vehicle; In the set of all coordinate points that need to be inspected, two different coordinate points p1 and p2 are randomly selected by manual designation or random number generator, and these two points are used as the starting point and end point of the initial path segment. A new path node sequence is created, and the nodes corresponding to p1 and p2 are inserted into the sequence in sequence. At the same time, the predecessor pointer of the p1 node is set to point to the p2 node, and the successor pointer of the p2 node is set to point to the p1 node, forming a simple closed-loop path segment; S22: Loop to build the path and enter the main loop. The loop condition is that there are still coordinate points that have not been included in the current path. In each loop iteration, a coordinate point p is randomly selected from the set of coordinate points that have not been selected. i , using the constructed quadtree data structure, quickly find p i The K nearest neighbors in the current path.

4. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: The specific operation of step S22 loop construction path is as follows: S221: Starting from the root node of the quadtree, according to p i The coordinate position of p is recursively searched for i The smallest sub-region of the node with a distance p is selected from the path nodes in this sub-region and the adjacent sub-regions. i The nearest K points; S222: For each p of the K neighbor points found n Try to p i Insert into p n The previous and next positions of the node are used to form new candidate paths, and the total path weight of each candidate path is calculated. The calculation of the total path weight is based on the unit path weight of each point. : ， Among them, v d is the flight speed of the drone, t d It takes time to inspect each node; S223: Based on the distance relationship between the path and the initial position of the drone nest vehicle and the estimated round trip cost W after considering the dynamic movement of the drone nest vehicle, the specific calculation formula is: ， in, Unit path weight for each point , m is the number of path nodes after insertion, is a parameter that adjusts the initial distance weight, d i,n0 is the distance from the i-th point in the path to the initial position of the drone nest vehicle, is the parameter that adjusts the dynamic round trip cost of the drone nest vehicle, C i,n (t) is the estimated round-trip cost from the i-th point in the path to the drone nest vehicle at time t; S224: Compare the total weights of all candidate solutions, select the insertion solution with the smallest total weight, and i Insert into the corresponding position in the current path and update the predecessor and successor pointer relationship between the path nodes; S225: Repeat the above cycle steps until all coordinate points are successfully inserted into the path, and a closed loop path covering all points to be inspected is formed.

5. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: Step S3 of segmenting the ring path specifically includes the following steps: S31: random segmentation attempt, setting a larger number of iterations N to control the exploration depth of the Monte Carlo algorithm, performing a random segmentation operation on the currently constructed loop path in each iteration, specifically randomly selecting M-1 segmentation points on the loop path, where M is the number of segmentation paths set in advance, and segmenting the loop path into M sub-paths; S32: Loss function evaluation. For each of the M paths obtained by segmentation, the load of each path is calculated. The calculation of the path load takes into account the number of points in the path and the weight of each point. The formula is: ， in, is the number of points in the jth path, The weight of the i-th point in the path; Calculate the sum of all path loads L sum and the standard deviation of the path load ; Calculate the average distance D of the drone from the last inspection point to the nearest drone nest vehicle position in each segmented path avg (t), for the last point of each segmented path, according to the dynamic position information data structure of the drone nest vehicle based on the real-time traffic status data, the position of the drone nest vehicle when the drone is expected to arrive at the point is obtained, the distance to the position is calculated, and then the average value is obtained; at the same time, according to the energy consumption model of the drone nest vehicle and the driving path and speed, the driving cost E of the drone nest vehicle corresponding to each segmented path is calculated nest (t); According to the set loss function: ， in, To adjust the parameters of the standard deviation, To adjust the parameters affecting the distance of the vehicle returning to the drone nest, To adjust the parameters affecting the driving cost of the drone nest vehicle, the loss value of the segmentation is calculated; S33: Select the optimal segmentation. After completing N segmentation attempts, compare the loss function value of each segmentation, and select the segmentation method with the smallest loss value as the optimal segmentation result of this Monte Carlo algorithm. The path set corresponding to the optimal segmentation result is used as the starting path for subsequent simulated annealing algorithm iterations, and the quadtree TE is adjusted according to these paths. Specifically, the points belonging to each path are stored in the corresponding TE tree and the points that do not belong to each path are stored in the TE tree, so as to prepare data for subsequent path optimization calculations.

6. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: Step S4 simulated annealing optimization path specifically includes the following steps: S41: Initialize parameters, set the initial temperature T0 of the simulated annealing algorithm, and set the annealing rate , used to control the temperature drop rate during the iteration process, and a smaller probability value p is used to determine whether the temperature rise strategy is hit, setting the parameters that adjust the distance change , adjust the parameters affecting the dynamic round trip cost of the drone nest vehicle ; S42: Main loop iteration, entering the main loop of the simulated annealing algorithm, the loop condition is that the temperature drops below a preset threshold.

7. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: Step S42 main loop iteration specifically includes the following steps: In each loop iteration: S421: First, determine whether the temperature increase strategy in the annealing algorithm is hit with probability p. If the temperature increase strategy is hit, perform the following operations: S422: Randomly select a path p from the current path set r , for path p r Every point p in k , use the quadtree to find its nearest K neighbor points; S423: For each neighbor point p n , try inserting it into p k The previous and next positions are used to form new candidate paths, and the growth value of the path load of each candidate is calculated. , the change of the distance from the path to the drone nest vehicle And the change of dynamic round-trip cost of drone nest vehicle ; S424: Define a comprehensive evaluation metric Select the candidate with the smallest E. If the smallest E is less than 0, that is, the comprehensive cost of the path is reduced after insertion, then directly insert the neighbor point p n Delete it from its original path and add it to path p r If the minimum E is greater than or equal to zero, the insertion operation is accepted with a certain probability according to the simulated annealing acceptance criterion; the simulated annealing acceptance criterion formula is: ， in is the change value of the comprehensive cost of the path, T is the current temperature, that is, a random number is generated ,if , then accept the insertion operation, otherwise give up; S425: If the warm-up strategy is not hit, a path with a 50% probability is selected according to the principle of path with larger load: First, calculate the load L of each path j , and then softmax the load to get the selection probability: ， Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the calculated probability s ; S426: For path P s Every point P in m , use the quadtree to find the nearest neighbor point that does not belong to the path, calculate the distance to these neighbor points, and select the neighbor point p with the shortest distance n , p n Remove from its original path and add to path P s At the same time, the change in the distance from the inserted path to the drone nest vehicle is calculated , and the dynamic round-trip cost change of the drone nest vehicle If the comprehensive cost increases too much, a path with a smaller load is selected with a probability of 50% according to the simulated annealing acceptance criterion: the load of each path is first inverted and then softmaxed to obtain the selection probability: ， Among them, P j is the probability that the jth path is selected, m is the number of paths, is the adjustment factor, L j is the path load, and a path P is selected according to the probability t ; S427: For path P t Every point P in q , find the nearest neighbor point that does not belong to the path and perform the above removal and addition operations, taking into account the distance change , Dynamic round-trip cost changes of drone nest vehicles and acceptance criteria; S428: recalculate the overall loss function value Loss according to the new path load condition, distance change and dynamic round-trip cost change of the drone nest vehicle; Regardless of which random selection operation is used, after completing the transfer of neighbor points, recalculate the load of the path that accepts the neighbor point transfer and the path that has common points with the path, as well as the change in the distance from the overall path to the drone nest vehicle and the change in the dynamic round-trip cost of the drone nest vehicle, in the same way as the loss function Loss calculation set above; S429: according to the changes of the current temperature T and the loss function value, decide whether to accept the new path adjustment according to the simulated annealing acceptance criteria; if accepted, update the path set; if not accepted, keep the original path set unchanged; S4210: Update temperature That is, the temperature is lowered according to the annealing rate, the exploration scope of the algorithm in subsequent iterations is reduced, and it gradually converges to a better solution. Check whether the loop end condition is met. If not, continue to the next loop iteration; if satisfied, the iterative process of the simulated annealing algorithm ends.

8. The method for autonomous inspection path planning of a UAV distribution network adapted to a UAV nest according to claim 1 is characterized in that: Step S5: Final path cutting and output specifically includes the following steps: S51: Path cutting check, traverse all paths after simulated annealing optimization, for each path p i Check whether its total load exceeds the maximum daily workload W of the drone max ,If the load of a path exceeds the maximum workload, the path needs to be cut; S52: Cutting operation, using a greedy strategy to cut the path, starting from the starting point of the path, gradually accumulating the weights of the points on the path until the accumulated sum is close to but does not exceed the maximum daily workload of the drone W max When the point is taken as a candidate cutting point, for each candidate cutting point, the distance d from the end point of the path after cutting to the dynamic moving estimated position of the UAV nest vehicle is calculated. end,n (t) and the cost E of the drone nest vehicle to travel to that location nest (t), define a cutting point evaluation function: ， Where L sub is the load of the subpath after cutting, is the parameter that adjusts the distance effect, d end,n (t) is the distance from the end point of the path after cutting to the estimated position of the drone nest’s dynamic movement, It is a parameter that adjusts the impact of the UAV's nest vehicle driving cost. The candidate cutting point with the smallest F value is selected for cutting, and the path is divided into two sub-paths. This process is repeated until all paths meet the UAV's maximum daily workload limit; S53: Multi-UAV task allocation and output, the cut path is divided into the number of UAVs n d The task load of each UAV is balanced. The task volume is measured according to factors such as the length, number of points or load of the path. Paths with smaller task volumes are preferentially allocated to UAVs with larger task volumes, so as to gradually balance the workload of each UAV. Specifically, the task volume index, path length and load of each sub-path after cutting are calculated, and the paths are sorted from small to large according to the task volume. Then, the paths are allocated to UAVs in turn. Each time the UAV with the smallest total task volume is selected for allocation, the total task volume of the UAV is updated.

9. A drone distribution network autonomous inspection path planning system adapted to drone nests, characterized in that: The method comprises a server and a processor, wherein the server stores a system program, and the processor implements the steps of the method according to any one of claims 1 to 8 when executing the computer program.

10. The autonomous inspection path planning system for UAV distribution network adapted to UAV nest according to claim 9 is characterized in that: The system outputs a complete inspection path plan for each UAV, including a sequence of path coordinate points, estimated flight time, and estimated arrival time at each inspection point based on real-time traffic status data, estimated return time to the UAV nest vehicle, and location information of the UAV nest vehicle at each key time point. This information is stored in a database for the ground control center to monitor and schedule tasks. During the inspection process, the UAV can communicate with the ground control center based on the actual flight conditions, and the ground control center can adjust and optimize the path plan based on real-time information to ensure the smooth completion of the inspection task.

Images