Route planning method for multi-aircraft area coverage search based on improved K-Means algorithm

Abstract

The invention relates to an air route planning method for multi-machine area coverage search based on an improved K-Means algorithm. The method comprises the following steps: using an improved mixed K-Means clustering division strategy; the method comprises the following steps: dividing a target area into a plurality of sub-areas, taking projection points of a starting point and an ending point of multi-unmanned aerial vehicle cooperation at the boundary of the target area as fixed clustering centers, automatically selecting initial values of other clustering centers, and determining a polygonal boundary of the sub-area corresponding to each clustering center based on all the clustering centers; on the basis of preset unmanned aerial vehicle path parameters and polygon boundary information, generating a broken line type coverage path with the minimum total cost for each sub-region, and obtaining a path point sequence of each sub-region; and for the path point sequence of each sub-region, according to a minimum turning radius constraint condition, obtaining a smooth airway path satisfying kinematics constraint. According to the method, the technical problem that the adaptability is insufficient due to the fact that the coverage integrity cannot be guaranteed and path flight cannot be considered in the existing scheme is solved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) search technology, and in particular to a route planning method for multi-aircraft area coverage search based on an improved K-Means algorithm. Background Technology

[0002] In the field of route planning for collaborative multi-UAV regional coverage search, existing solutions mainly include coverage route planning algorithms, path smoothing techniques, region partitioning methods, and modeling related to multi-UAV route planning problems. Among these, coverage route planning algorithms include Boustrophedon path planning based on region segmentation, which uses reciprocating paths to cover sub-regions; the Spanning Tree Coverage (STC) algorithm based on graph theory, which models the environment as a mesh graph and constructs the coverage path using a minimum spanning tree; and the RCPP algorithm based on the "rotating caliper method," suitable for two-dimensional convex regions, which optimizes the overall path length by enumerating parallel directions of the flight path.

[0003] For path smoothing, sharp turning points in the path can be smoothed. Region partitioning methods include equal-area partitioning strategies that cut regions into strips with equal area, and proportional partitioning momentum iteration strategies that combine proportional strip partitioning with momentum gradient descent optimization and iterative adjustment of region boundaries using path length as feedback. Modeling the multi-UAV route planning problem involves simplifying the UAV into a point mass model and constraining the kinematic model to the minimum turning radius; modeling the search region by rasterizing the target area and dividing it into sub-regions; and task allocation and cooperative search strategies that support proportional partitioning momentum iteration segmentation.

[0004] Existing route planning technologies for region coverage search have significant shortcomings in practical applications. In particular, in coverage route planning, the Boustrophedon algorithm is poorly adapted to polygonal and complex-shaped regions, resulting in redundant and uneven paths at corners; the STC algorithm suffers from insufficient flyability of actual paths, and its discretization accuracy is limited by grid resolution; the RCPP algorithm outputs polyline paths with many right angles at turns, which does not meet the minimum turning radius constraints of fixed-wing UAVs.

[0005] In terms of region partitioning, its equal area partitioning strategy only considers the equal area and does not take into account path cost, which can easily lead to load imbalance; the conventional proportional partitioning method lacks a dynamic adjustment mechanism and is difficult to achieve high-precision load balancing.

[0006] Furthermore, the integration of coverage route planning algorithms and path smoothing techniques is not tight enough, making it difficult to ensure both coverage integrity and path flyability. Moreover, existing task allocation and collaborative search strategies are not adaptable enough to complex environments and cannot adequately meet the actual needs of multi-aircraft collaborative coverage search. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] In view of the above-mentioned shortcomings and deficiencies of the existing technology, the present invention provides a route planning method for multi-aircraft area coverage search based on the improved K-Means algorithm, which solves the technical problems of insufficient adaptability caused by the inability to take into account the flightability of the path while ensuring coverage integrity in the existing scheme.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0011] In a first aspect, embodiments of the present invention provide a route planning method for multi-aircraft area coverage search based on an improved K-Means algorithm, comprising:

[0012] S100: Use an improved hybrid K-Means clustering partitioning strategy; divide the target area into multiple sub-regions, each sub-region corresponding to a single drone cooperating with multiple drones;

[0013] Specifically, the starting and ending points of the multi-UAV collaboration are fixed cluster centers at the projection points of the target area boundary. The initial values ​​of other cluster centers are automatically selected using the K-Means++ strategy in hybrid K-Means. Based on all cluster centers, the standard K-Means clustering iteration in hybrid K-Means is used to determine the polygonal boundary of the sub-region corresponding to each cluster center.

[0014] S200. Based on the pre-given UAV path parameters and polygon boundary information, generate a polyline coverage path with the minimum total cost for each sub-region and obtain the path point sequence for each sub-region.

[0015] S300. For each sub-region's path point sequence, obtain a smooth flight path that satisfies kinematic constraints based on the minimum turning radius constraint.

[0016] Optionally, the method further includes:

[0017] Calculate the total cost of the route within each sub-region and evaluate the load balance of all routes;

[0018] Specifically, the total cost C for each route within each sub-region total = The cost C of the route length in this route path len +Turn Cost C turn That is, C total =C len +C turn ;

[0019] Flight path P = {p1, p2, ..., p n The cost of increasing the length of the route is C. len ;

[0020] Where d(p) i ,p i+1 () represents the Euclidean distance between adjacent path points;

[0021]

[0022] (x i y i ), (x i+1 y i+1 () represents adjacent path points;

[0023] Turning Cost C turn for:

[0024] Where, k tum θ is the turning cost coefficient. i Let be the heading angle at the i-th path point.

[0025] Optionally, S100 includes:

[0026] S110. Discretize the target area into a grid point set with a specified precision;

[0027] S120. Project the starting and ending points of all cooperating drones within the target area to the boundary of the target area, and use the projected points of the starting and ending points as the initial values ​​of the fixed cluster centers; and use the projected points of the starting and ending points as key points;

[0028] S130. The K-Means++ strategy in the hybrid K-Means is used to automatically select the initial values ​​of other cluster centers. The automatic selection process of the K-Means++ strategy is performed by sampling weights around the key points and balancing the initial values ​​of the key points and other cluster centers.

[0029] S140. Based on the initial values ​​of all cluster centers, standard K-Means clustering iteration in hybrid K-Means is used to determine the polygon boundaries of the sub-regions corresponding to each cluster center.

[0030] Optionally, S200 includes:

[0031] S210. Initialize the strip spacing dx, start position, end position, cruise speed, rotation speed and polygon boundary information of all cooperative UAV paths during the search; and use the rotating caliper algorithm to determine all heel-to-heel pairs in each sub-region, with each heel-to-heel pair corresponding to a set of scanning directions.

[0032] S220. Traverse each pair of heel points to generate a stripe coverage path in the scanning direction;

[0033] S230. Based on the strip coverage path of each heel point pair, construct the complete auxiliary police in the forward and reverse directions, and obtain the total cost of the path in the construction direction.

[0034] S240. For all strip coverage paths in each sub-region, select the path with the minimum total cost as the polyline coverage path of the sub-region, and obtain the path point sequence of each sub-region.

[0035] Optionally, S300 includes:

[0036] For each sub-region's path point sequence, obtain the heading angle of each path point and refine it through interpolation.

[0037] Select target segments for Dubins curve interpolation from the refined pathpoint sequence; target segments include: the first segment, the last segment, and segments with large turns;

[0038] Based on the minimum turning radius constraint, and using the Dubins curve trajectory smoothing optimization algorithm, the Dubins curve segment that satisfies the minimum turning radius is used to replace the large-angle segment in the target segment;

[0039] By splicing the Dubins curve segment and the other segments in the waypoint sequence except for the large-angle segments, a continuous and smooth route path is output, thus obtaining a smooth route path that satisfies kinematic constraints.

[0040] Optionally, all cooperating drones have the same starting point, the same ending point, and are the same type of drone.

[0041] Secondly, embodiments of the present invention also provide a route planning device for multi-aircraft area coverage search based on an improved K-Means algorithm, comprising:

[0042] Sub-region partitioning unit, used to partition the target region using an improved hybrid K-Means clustering strategy; divide the target region into multiple sub-regions, each sub-region corresponding to a single drone cooperating with multiple drones;

[0043] Specifically, the starting and ending points of the multi-UAV collaboration are fixed cluster centers at the projection points of the target area boundary. The initial values ​​of other cluster centers are automatically selected using the K-Means++ strategy in hybrid K-Means. Based on all cluster centers, the standard K-Means clustering iteration in hybrid K-Means is used to determine the polygonal boundary of the sub-region corresponding to each cluster center.

[0044] The initial path generation unit is used to generate a polyline coverage path with the minimum total cost for each sub-region based on the pre-given UAV path parameters and polygon boundary information, and obtain the path point sequence for each sub-region.

[0045] The route generation unit is used to obtain a smooth route that satisfies kinematic constraints based on the minimum turning radius constraint for the path point sequence of each sub-region.

[0046] Optionally, it also includes:

[0047] The load balancing evaluation unit is used to calculate the total cost of the route within each sub-region and evaluate the load balancing of all route paths.

[0048] Specifically, the total cost C for each route within each sub-region total = The cost C of the route length in this route path len +Turn Cost C turn That is, C total =C len +C tirn ;

[0049] Flight path P = {p1, p2, ..., p n The cost of increasing the length of the route is C. len ;

[0050] Where d(p) i ,p i+1 ) represents the Euclidean distance between adjacent path points.

[0051]

[0052] (x i y i ), (x i+1 y i+1 () represents adjacent path points;

[0053] Turning Cost C turn for:

[0054] Where, k tumθ is the turning cost coefficient. i Let be the heading angle at the i-th path point.

[0055] Thirdly, embodiments of the present invention also provide a ground control device, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program in the memory, specifically executing the steps of the multi-aircraft area coverage search route planning method based on the improved K-Means algorithm described in any of the first aspects above.

[0056] Fourthly, embodiments of the present invention also provide a drone search system, characterized in that it includes the ground control device described in the third aspect and multiple drones, wherein all drones interact with the ground control device to obtain search paths using the multi-drone area coverage search route planning method based on the improved K-Means algorithm described in any of the first aspects.

[0057] (III) Beneficial Effects

[0058] The beneficial effects of this invention are:

[0059] This invention employs an improved hybrid K-Means clustering partitioning algorithm to divide the target area for multi-UAV regional coverage search. Combined with the RCPP path planning algorithm and the Dubins curve trajectory smoothing algorithm, it achieves efficient planning of coverage paths for multi-UAV collaborative operation.

[0060] This invention enables the target area to be rationally divided into sub-regions that match the number of UAVs. The division results fully consider the path characteristics of the UAVs' starting and ending points, improving path smoothness. Clustering ensures that each sub-region is nearly equal in area and number of grid points, laying the foundation for balanced task load distribution. The generated coverage path can completely cover the target area without omission, and after being smoothed by the Dubins curve, it meets the minimum turning radius constraint of fixed-wing UAVs, significantly improving the flightability of the track. It is suitable for complex convex polygonal regions and has good engineering practicality and stability in multi-UAV collaborative coverage search tasks. Attached Figure Description

[0061] Figure 1 A schematic diagram illustrating the RCPP coverage route process provided in this application embodiment;

[0062] Figure 2 A schematic diagram illustrating the process of interpolating and refining RCPP routes provided in this application embodiment;

[0063] Figure 3 A schematic diagram illustrating the comparison between the RCPP route and the Dubins smooth route provided in this application embodiment;

[0064] Figure 4An overall flowchart of a method provided in an embodiment of this application;

[0065] Figure 5 This is a schematic diagram of the improved K-Means clustering partitioning (5UAV) process in an embodiment of this application;

[0066] Figure 6 This is a schematic diagram illustrating the process of improving K-Means clustering to segment covered air routes (5UAVs) in an embodiment of this application;

[0067] Figure 7 A flowchart illustrating a route planning method based on an improved K-Means algorithm for multi-aircraft area coverage search, provided as an embodiment of this application;

[0068] Figure 8 This is a schematic diagram of the path cost distribution when evaluating load, provided in an embodiment of this application. Detailed Implementation

[0069] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0070] Example 1

[0071] like Figure 4 and Figure 7 As shown, this embodiment provides a route planning method for multi-aircraft area coverage search based on an improved K-Means algorithm. The execution entity of this method is a ground control station, which adopts... Figure 7 After dividing the area into sub-regions and determining flight paths, the method shown sends each cooperating UAV its respective sub-region and corresponding flight path, enabling each UAV to perform area coverage search within its sub-region, ensuring no area is missed, and simultaneously achieving load balancing among the UAVs. The method in this embodiment may include the following steps:

[0072] S100: Use an improved hybrid K-Means clustering partitioning strategy; divide the target area into multiple sub-regions, each sub-region corresponding to a single drone cooperating with multiple drones;

[0073] Specifically, the starting and ending points of the multi-UAV collaboration are projected onto the boundary of the target area as fixed cluster centers. The initial values ​​of other cluster centers are automatically selected using the K-Means++ strategy in hybrid K-Means. Based on all cluster centers, the standard K-Means clustering iteration in hybrid K-Means is used to determine the polygonal boundary of the sub-region corresponding to each cluster center.

[0074] Understandably, the traditional K-Means region partitioning method often leads to partitioning results that are out of sync with the actual flight mission of the UAV due to the randomness of its initial cluster centers. For example, a UAV may be assigned to a sub-region far from its takeoff point, resulting in a large number of unnecessary transfer routes, increasing the total range and energy consumption.

[0075] The improved hybrid K-Means clustering strategy in this step strongly couples the region partitioning problem with the start and end points of the task by projecting the common start and end points of all UAVs onto the region boundary and using these two projection points as 'fixed', immovable cluster centers. The remaining cluster centers are intelligently selected around the key points using the K-Means++ strategy.

[0076] S200: Based on the pre-given UAV path parameters and polygon boundary information, generate a polyline coverage path with the minimum total cost for each sub-region, and obtain the path point sequence for each sub-region.

[0077] In this step, when generating the initial coverage path for each sub-region, the shortest route is not simply pursued. This embodiment employs a comprehensive path cost function, which combines the total track length cost and turning cost in a weighted manner. While traversing all possible scan directions using the Rotating Caliper Process (RCPP) algorithm, the comprehensive cost is calculated for both forward and reverse paths in each direction, and the direction with the minimum total cost is selected as the optimal solution. This approach not only considers distance but also pre-considers the path's smoothing potential, selecting an initial path with fewer turns and easier flight for the subsequent Dubins smoothing step, further improving the feasibility of the final route.

[0078] For example, the total cost C of each route within each sub-region total = The cost C of the route length in this route path len +Turn Cost C turn That is, C total =C len +C turn ;

[0079] Flight path P = {p1, p2, ..., p n The cost of increasing the length of the route is C. len ;

[0080] Where d(p) i ,p i+1 ) represents the Euclidean distance between adjacent path points.

[0081]

[0082] (x i y i ), (x i+1 y i+1 () represents adjacent path points;

[0083] Turning Cost C turn for:

[0084] Where, k tum θ is the turning cost coefficient. i Let be the heading angle at the i-th path point.

[0085] Understandably, since changes in the heading angle during actual flight mean additional energy consumption and speed loss, and the turning of the drone is limited by the minimum turning radius of the Dubins model, it is necessary to introduce a turning cost function.

[0086] S300. For each sub-region's path point sequence, obtain a smooth flight path that satisfies kinematic constraints based on the minimum turning radius constraint.

[0087] It should be noted that the 'path point sequence' mentioned in step S200 is generated by the RCPP algorithm. Its data structure is a series of two-dimensional coordinate points (X,Y), which is an N×2 array and stored as a CSV file. At this time, it does not contain heading angle information.

[0088] After proceeding to step S300, the route smoothing algorithm first processes the input sequence of pure coordinate points. For any intermediate point P on the path... i Its heading angle can be determined by its previous point P. i-1 and the next point P i+1 The position is used to calculate the value, for example, using atan2(Y). i+1 -Y i ,X i+1 -X i The heading angle is determined by the first and second points of the path. Therefore, the heading angle is calculated as needed in step S300 and serves as a necessary parameter for subsequent Dubins curve interpolation.

[0089] This embodiment provides a complete technical chain from macro-region allocation to micro-path generation. It first uses an improved K-Means algorithm (S100) to rationally allocate the task area, then generates an efficient initial path using the RCPP algorithm based on comprehensive cost (S200), and finally ensures that the path meets the UAV's kinematic constraints through Dubins curve smoothing optimization (S300). These three steps are interconnected, and the final output is not only a theoretical path point, but also a complete and flyable route that can be directly issued to the UAV for execution, solving the problem of the disconnect between planning and execution in existing technologies.

[0090] In this embodiment, the target area can be reasonably divided into sub-regions that match the number of UAVs. The division results fully consider the path characteristics of the UAVs' starting and ending points, improving path smoothness. Clustering ensures that each sub-region is nearly equal in area and number of grid points, laying the foundation for balanced task load distribution. The generated coverage path can completely cover the target area without omission, and after smoothing, it meets the minimum turning radius constraint of fixed-wing UAVs, significantly improving the flightability of the track. It is suitable for complex convex polygonal regions and has good engineering practicality and stability in multi-UAV collaborative coverage search tasks.

[0091] In this embodiment of the invention, all UAVs performing area coverage searches need to coordinate, starting from the same starting point and ending at the same destination. Each UAV corresponds to a search sub-region, and it performs a comprehensive search within its respective sub-region. In this embodiment, before the UAV search, the ground control station divides the area into sub-regions and determines the flight paths within each sub-region. After determining the flight paths for each sub-region, the ground control station controls each UAV to achieve coverage search within that sub-region. Its application scenarios include information search for a specified target area, personnel search, or target search in a forest.

[0092] Example 2

[0093] This invention proposes a multi-UAV area coverage search route planning method based on an improved K-Means algorithm. By employing an improved hybrid K-Means clustering strategy for area partitioning, combined with the RCPP path planning algorithm and the Dubins curve trajectory smoothing algorithm, it achieves complete coverage of the target area by multiple UAVs. This method can balance the task load of each UAV, making the path cost of each UAV as similar as possible, while satisfying the kinematic constraints of the UAVs (such as minimum turning radius), ensuring the flightability of the route, and overcoming the shortcomings of existing route planning methods in terms of task balance, path smoothness, and adaptability to complex areas.

[0094] The route planning for the area coverage search in this embodiment mainly involves four processes: reasonable area division, initial coverage path generation, trajectory smoothing optimization, and task load balancing, which are explained in detail below:

[0095] 1. Region partitioning: This involves using an improved hybrid K-Means clustering algorithm to allocate sub-regions within the target region.

[0096] Step 1-1: Discretize the target area with a certain precision (e.g., longitude of 5-10 meters) into a grid point set, which will be used as the clustering input data.

[0097] The target region is discretized into a grid set with a certain precision, providing the basic data for clustering. This step ensures that the internal structure of the region is fully sampled, which is especially suitable for target regions with complex and irregular shapes, laying the foundation for the accuracy of subsequent clustering.

[0098] Steps 1-2: Project the start and end points of all collaborative drones onto the boundary of the target area, using the projected points as partial "fixed cluster centers"; the projected points of the start and end points are used as key points.

[0099] The drone's origin and destination points are projected onto the region boundary, with the projected points serving as partial "fixed cluster centers." This improvement makes the partitioning results more closely match the drone's path characteristics, enhances path smoothness, and avoids path redundancy that may result from the random selection of cluster centers in traditional K-Means clustering.

[0100] Steps 1-3: The K-Means++ strategy is used to automatically select the initial values ​​of the remaining cluster centers (as region points). The process of automatically selecting the K-Means++ strategy is to balance the sampling weights of key points and region points; that is, selection is made around key points, so that the sampling weights of key points and region points are basically the same.

[0101] It should be noted that the number of cluster centers is the same as the number of all cooperating drones.

[0102] The remaining cluster centers are automatically selected using the K-Means++ strategy, which balances the sampling weights of key points (start and end point projections) and points within the region. This strategy effectively avoids the problems of local optima and uneven clustering that are common in the traditional K-Means algorithm, thus improving the rationality of sub-region partitioning.

[0103] Steps 1-4: Based on all cluster centers (including the fixed cluster centers and remaining cluster centers mentioned above), perform standard K-Means clustering to divide the target area into K sub-regions matching the number of drones. In this embodiment, each sub-region is considered a cluster.

[0104] That is, the number of drones corresponds to the number of sub-regions, and each drone corresponds to one sub-region in subsequent searches.

[0105] Steps 1-5: Use the convex hull algorithm to generate polygonal boundaries for sub-regions of each cluster.

[0106] Standard K-Means clustering is performed using the determined initial cluster centers as parameters to divide the grid points into K sub-regions matching the number of drones. Then, the polygonal boundaries of the sub-regions are generated using the convex hull algorithm. This process ensures the connectivity and integrity of each sub-region, and each sub-region is nearly equal in area and number of grid points, providing a prerequisite for balanced task load distribution.

[0107] 2. RCPP Initial Coverage Path Generation: This generates a polyline coverage path for each sub-region to ensure that no region is missed in coverage.

[0108] Step 2-1: Initialize parameters;

[0109] The initialization parameters include: sub-region polygon M, strip spacing dx, start point, end point, and velocity parameters (such as translational velocity v_trans, i.e., cruising velocity, and rotational velocity v_rot).

[0110] The cruising speed here refers to the average speed used by all drones during the search process.

[0111] Step 2-2: Based on the initialized parameters, use the Rotating Calipers algorithm to find all antipodal pairs of the polygon. Each antipodal pair corresponds to a set of possible scanning directions.

[0112] Understandably, each subregion can have multiple pairs of heel points.

[0113] Steps 2-3: Traverse the scanning direction of each antipodal point pair and call the bestPathForAntipodalPair function in the rotating caliper algorithm to generate the strip coverage path for each scanning direction.

[0114] Path=bestPathForAntipodalPair(M,current pair ,dx)

[0115] current pair This is the current heel point pair.

[0116] Steps 2-4: Construct complete forward and reverse paths (including start point, covered path, and end point), and calculate the total cost of the two paths (combining translation and rotation costs) using the timeCost2D function in the rotating caliper algorithm;

[0117] FullPath1=[start;Path;end]

[0118] FullPath2=[start; flipud(Path); end]

[0119] cost1=timeCost2D(FullPath1,…)

[0120] cost2=timeCost2D(FullPath2,…)

[0121] Path represents the forward path, flipud(Path) represents the reverse path, and FullPath represents the total path; start is the starting point, and end is the ending point.

[0122] Steps 2-5: Select the path with the lowest cost as the initial coverage path for this sub-region. Return the path point sequence of the optimal path, as shown in the generated result. Figure 1 As shown.

[0123] The pathpoints in the pathpoint sequence are arrays of each NX2, corresponding to latitude and longitude coordinates in reality, and are stored as CSV files.

[0124] The RCPP path planning module calculates all heel-to-heel pairs and evaluates the path costs in different directions, selecting the path with the minimum total cost. This ensures the efficiency of the initial coverage path and achieves complete coverage of the target area.

[0125] 3. Dubins curve trajectory smoothing optimization: This involves smoothing the polyline path generated by RCPP to meet the UAV's turning radius constraints.

[0126] In this embodiment, a minimum turning radius is given in advance. This minimum turning radius is related to the drone's cruising speed, rotation speed, and drone size.

[0127] Step 3-1: For each sub-region, extract the polyline path point sequence from the RCPP output path point sequence and record the heading angle of each point;

[0128] For example, based on the polyline path point sequence (pure coordinate array) output by the aforementioned steps, the initial heading angle of each path point is calculated according to the positional relationship between the preceding and following points.

[0129] Step 3-2: Call the `insert_points` function in the Dubins algorithm to interpolate and refine the extracted polyline path point sequence, focusing on adding points to the start, end, and polyline segments to facilitate smoothing. The result is as follows: Figure 2 As shown.

[0130] The initial and final segments are crucial paths for the drone entering and exiting the area, requiring the insertion of additional waypoints for smoother operation. Figure 1 and Figure 2 This directly displays the section of the drone's path that has bends when it enters and leaves the area. The difference in this embodiment involves inserting more path points onto the original path, which... Figure 2 The interpolated path overlaps with the original path, so it cannot be displayed.

[0131] Specifically, in the interpolation refinement, one point is inserted at the beginning and end of each segment as a smoothing point, marked as the smoothing start point and smoothing end point. The beginning and end segments are then subjected to Dubins smoothing to further enhance the flightability of the path.

[0132] It should be noted that this step does not change the shape of the path, but rather inserts more points on the existing line segment to increase the path density. Its main purposes are twofold: first, to obtain more accurate heading changes in subsequent calculations; and second, to provide more points for the initial segment when the UAV enters the main coverage area and the tail segment when it leaves the coverage area, facilitating smoother curve fitting later.

[0133] Step 3-3: Traverse and calculate each waypoint (including the interpolated waypoint), and select the first segment, the last segment, and the waypoint pairs with large turns as the target segments for Dubins curve interpolation.

[0134] Traverse the interpolated path points to identify the 'target segments' that need smoothing. The target segments mainly include: 1) the 'first segment' when the UAV enters the work area; 2) the 'tail segment' when the UAV leaves the work area; 3) the 'large-angle segment' in the path where a large angle is generated due to a sharp change in heading.

[0135] Steps 3-4: Call the `insertDubinsByDX` function in the Dubins algorithm to insert a Dubins curve segment that satisfies the minimum turning radius `r_min` into the target segment, replacing the original large-angle segment. In other words, call the `insertDubinsByDX` function to generate a corresponding Dubins curve for each identified 'target segment' based on the preset minimum turning radius `r_min`. Then, replace the target segment polyline in the original path with the newly generated smooth Dubins curve segment.

[0136] Steps 3-5: Concatenate the unmarked segments and the segments replaced with Dubins curves to output a smooth, continuous flyable route. The final route, smoothed by the Dubins curve, is then compared with the original RCPP route. Figure 3 As shown.

[0137] That is, all the replaced smooth Dubins curve segments and the original straight segments in the path that were not selected as target segments are seamlessly spliced ​​together. The final output is a complete, flyable route that includes both straight cruise segments and smooth curve segments and satisfies kinematic constraints at every point.

[0138] This implementation uses the Dubins curve model, the purpose of which is to satisfy the shortest path connecting two points under curvature constraints, consisting of a straight line and a circular arc of fixed radius, such as... Figure 3 As shown.

[0139] The Dubins curve trajectory smoothing module refines the flight path through interpolation and inserts curves at key inflection points.

[0140] The Dubins curve segment, at the cost of a slight increase in path length, allows the path to meet the minimum turning radius constraint of fixed-wing UAVs, significantly improving the flightability of the trajectory.

[0141] Based on the above description, combined with Figure 6 The overall flow of the method in this embodiment is described below:

[0142] Step A1: Input initialization, load target area, number of drones, origin and destination information and algorithm parameters (strip spacing dx, minimum turning radius rmin, etc.);

[0143] Step A2: Assign regions to each UAV using an improved hybrid K-Means clustering partitioning algorithm;

[0144] Step A3: Execute the RCPP algorithm for each sub-region to generate an initial coverage path;

[0145] Step A4: Refine the initial path by interpolation and smooth the Dubins curve to obtain a track that satisfies the kinematic constraints;

[0146] Step A5: Calculate the path cost (total length cost + turning cost) for each drone and evaluate load balancing; achieve path balancing, treating the path at that point as the load, such as... Figure 8 As shown.

[0147] The task evaluation uses a comprehensive path cost function consisting of path length and turning cost as the quantitative basis to assess the task load balance, providing data support for the optimization and selection of area division strategies and ensuring that the load of multiple UAVs is as balanced as possible.

[0148] Step A6: Output the final multi-UAV path planning results, including path point coordinates, heading angles, and performance indicators (path length, standard deviation, etc.).

[0149] Simulation Examples

[0150] This invention was simulated in the MATLAB R2023b environment, and the main parameters were set as follows:

[0151] (1) Number of drones: 5;

[0152] (2) Strip spacing (dx): 10 meters;

[0153] (3) Minimum turning radius (r_min): 5 meters;

[0154] (4) Region shape: convex irregular polygon (randomly generated region);

[0155] (5) Start and end points: randomly set (can be specified) to ensure they are outside the area boundary.

[0156] Improved K-Means clustering algorithm partitioning process:

[0157] Step 1: Discretize the target region into a two-dimensional grid point set with a precision of dx to ensure that the internal structure of the region is fully sampled, and generate coordinate point data for clustering.

[0158] Step 2: Project the starting point and ending point onto the region boundary to obtain two projection points as "fixed cluster centers";

[0159] The remaining three cluster centers were selected from the grid point set using the K-Means++ strategy, balancing the sampling weights of key points and region points.

[0160] Step 3: Using the initial cluster center as a parameter, divide all points into 5 clusters by calculating the distance between the grid points and the center. Each cluster corresponds to a sub-region of one UAV.

[0161] Step 4: Apply the convex hull algorithm to the grid points of each cluster to generate closed polygonal sub-region boundaries, ensuring that the sub-regions are connected and do not overlap, such as... Figure 5 As shown.

[0162] Step 5: Perform RCPP route coverage for each sub-area.

[0163] Step 6: Smooth the RCPP route using the Dubins curve.

[0164] Step 7: Perform task load balancing analysis, checking whether the smoothed path completely covers the sub-region, has no overlap, and the turning radius is ≥ r_min; count the path lengths of the 5 drones, calculate the standard deviation of the path lengths of the 5 drones, and evaluate the task balance. The standard deviation in the 5-drone scenario is 132.22 meters. The specific path is as follows: Figure 6 As shown.

[0165] Example 3

[0166] According to the embodiments disclosed in this invention, the present invention also provides a route planning device for multi-aircraft area coverage search based on an improved K-Means algorithm, comprising:

[0167] Sub-region partitioning unit, used to partition the target region using an improved hybrid K-Means clustering strategy; divide the target region into multiple sub-regions, each sub-region corresponding to a single drone cooperating with multiple drones;

[0168] Specifically, the starting and ending points of the multi-UAV collaboration are fixed cluster centers at the projection points of the target area boundary. The initial values ​​of other cluster centers are automatically selected using the K-Means++ strategy in hybrid K-Means. Based on all cluster centers, the standard K-Means clustering iteration in hybrid K-Means is used to determine the polygonal boundary of the sub-region corresponding to each cluster center.

[0169] The initial path generation unit is used to generate a polyline coverage path with the minimum total cost for each sub-region based on the pre-given UAV path parameters and polygon boundary information, and obtain the path point sequence for each sub-region.

[0170] The route generation unit is used to obtain a smooth route that satisfies kinematic constraints based on the minimum turning radius constraint for the path point sequence of each sub-region.

[0171] In practical applications, the above-mentioned device also includes:

[0172] The load balancing evaluation unit is used to calculate the total cost of the route within each sub-region and evaluate the load balancing of all route paths.

[0173] Specifically, the total cost C for each route within each sub-region total = The cost C of the route length in this route path len +Turn Cost C turn That is, C total =C len +C turn

[0174] Flight path P = {p1, p2, ..., p n The cost of increasing the length of the route is C. len ;

[0175] Where d(p) i ,p i+1 ) represents the Euclidean distance between adjacent path points.

[0176]

[0177] (xi y i ), (x i+1 y i+1 () represents adjacent path points;

[0178] Turning Cost C turn for:

[0179] Where, k tum θ is the turning cost coefficient. i Let be the heading angle at the i-th path point.

[0180] The sub-region partitioning unit in this embodiment can be specifically used for: discretizing the target region into a grid point set with a specified precision; projecting the start and end points of all cooperating UAVs within the target region onto the boundary of the target region, and using the projected points of the start and end points as the initial values ​​of fixed cluster centers; and using the projected points of the start and end points as key points; automatically selecting the initial values ​​of other cluster centers using the K-Means++ strategy in hybrid K-Means, wherein the automatic selection process of the K-Means++ strategy is performed around the key points and balancing the sampling weights of the initial values ​​of the key points and other cluster centers; and based on the initial values ​​of all cluster centers, using the standard K-Means clustering iteration process in hybrid K-Means to determine the polygon boundary of the sub-region corresponding to each cluster center.

[0181] For example, the initial path generation unit can be specifically used to initialize the strip spacing dx, start position, end position, cruise speed, rotation speed, and polygon boundary information associated with all cooperative UAV paths during the search; and use a rotating caliper algorithm to determine all heel-to-heel point pairs in each sub-region, with each heel-to-heel point pair corresponding to a set of scanning directions; traverse each heel-to-heel point pair to generate a strip coverage path for that scanning direction; based on the strip coverage path of each heel-to-heel point pair, construct complete paths in both forward and reverse directions, and obtain the total cost of the constructed paths; for all strip coverage paths in each sub-region, select the path with the minimum total cost as the polyline coverage path for the sub-region, and obtain the path point sequence for each sub-region.

[0182] Accordingly, the route generation unit can be specifically used to obtain the heading angle of each path point in the path point sequence of each sub-region, and to refine each path point through interpolation; to select target segments for Dubins curve interpolation processing from the refined path point sequence; the target segments include: the first segment, the last segment, and the large-angle segment with a large turn; based on the minimum turning radius constraint, and using the Dubins curve trajectory smoothing optimization algorithm, the large-angle segment in the target segment is replaced with a Dubins curve segment that meets the minimum turning radius; the Dubins curve segment and other segments in the path point sequence except for the large-angle segment are spliced ​​together to output a continuous and smooth route path, thus obtaining a smooth route path that meets kinematic constraints.

[0183] In this embodiment, all cooperating drones have the same starting point and the same ending point, and they are the same type of drone.

[0184] In addition, embodiments of the present invention also provide a ground control device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program in the memory, specifically performing the steps of the multi-aircraft area coverage search route planning method based on the improved K-Means algorithm described in any of the above embodiments.

[0185] Furthermore, embodiments of the present invention also provide an unmanned aerial vehicle (UAV) search system, including a ground control device and multiple UAVs as described in any of the above embodiments. All UAVs interact with the ground control device to obtain search paths using the multi-aircraft area coverage search route planning method based on the improved K-Means algorithm described in any of the above embodiments.

[0186] The above scheme enables the target area to be reasonably divided into sub-regions that match the number of UAVs. The division results fully consider the path characteristics of the UAVs' starting and ending points, improving path smoothness. Clustering ensures that each sub-region is nearly equal in area and number of grid points, laying the foundation for balanced task load distribution. The generated coverage path can completely cover the target area without omission, and after being smoothed by the Dubins curve, it meets the minimum turning radius constraint of fixed-wing UAVs, significantly improving the flightability of the track. It is suitable for complex convex polygonal regions and has good engineering practicality and stability in multi-UAV collaborative coverage search tasks.

[0187] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0188] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0189] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0190] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions 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 suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0191] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0192] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A route planning method based on multi-aircraft area coverage search using an improved K-Means algorithm, characterized in that, include: S100, Use an improved hybrid K-Means clustering partitioning strategy; The target area is divided into multiple sub-regions, and each sub-region corresponds to a single drone that coordinates multiple drones. Specifically, the starting and ending points of the multi-UAV collaboration are fixed cluster centers at the projection points of the target area boundary. The initial values ​​of other cluster centers are automatically selected using the K-Means++ strategy in hybrid K-Means. Based on all cluster centers, the standard K-Means clustering iteration in hybrid K-Means is used to determine the polygonal boundary of the sub-region corresponding to each cluster center. S200. Based on the pre-given UAV path parameters and polygon boundary information, generate a polyline coverage path with the minimum total cost for each sub-region and obtain the path point sequence for each sub-region. S300. For each sub-region's path point sequence, obtain a smooth flight path that satisfies kinematic constraints based on the minimum turning radius constraint.

2. The method according to claim 1, characterized in that, The method further includes: Calculate the total cost of the route within each sub-region and evaluate the load balance of all routes; Specifically, the total cost C for each route within each sub-region total = The cost C of the route length in this route path len +Turn Cost C turn That is, C total =C len +C turn ; Flight path P = {p1, p2, ..., p n The cost of increasing the length of the route is C. len ; Where d(p) i ,p i+1 () represents the Euclidean distance between adjacent path points; (x i y i ), (x i+1 y i+1 () represents adjacent path points; Turning Cost C turn for: Where, k tum θ is the turning cost coefficient. i Let be the heading angle at the i-th path point.

3. The method according to claim 1, characterized in that, S100 includes: S110. Discretize the target area into a grid point set with a specified precision; S120. Project the starting and ending points of all cooperating drones within the target area to the boundary of the target area, and use the projected points of the starting and ending points as the initial values ​​of the fixed cluster centers; and use the projected points of the starting and ending points as key points; S130. The K-Means++ strategy in the hybrid K-Means is used to automatically select the initial values ​​of other cluster centers. The automatic selection process of the K-Means++ strategy is performed by sampling weights around the key points and balancing the initial values ​​of the key points and other cluster centers. S140. Based on the initial values ​​of all cluster centers, standard K-Means clustering iteration in hybrid K-Means is used to determine the polygon boundaries of the sub-regions corresponding to each cluster center.

4. The method according to claim 1, characterized in that, S200 includes: S210. Initialize the strip spacing dx, start position, end position, cruise speed, rotation speed and polygon boundary information of all cooperative UAV paths during the search; and use the rotating caliper algorithm to determine all heel-to-heel pairs in each sub-region, with each heel-to-heel pair corresponding to a set of scanning directions. S220. Traverse each pair of heel points to generate a stripe coverage path in the scanning direction; S230. Based on the strip coverage path of each pair of heel points, construct the complete paths in the forward and reverse directions, and obtain the total cost of the paths in the constructed directions. S240. For all strip coverage paths in each sub-region, select the path with the minimum total cost as the polyline coverage path of the sub-region, and obtain the path point sequence of each sub-region.

5. The method according to claim 1, characterized in that, The S300 includes: For each sub-region's path point sequence, obtain the heading angle of each path point and refine it through interpolation. Select target segments for Dubins curve interpolation from the refined pathpoint sequence; target segments include: the first segment, the last segment, and segments with large turns; Based on the minimum turning radius constraint, and using the Dubins curve trajectory smoothing optimization algorithm, the Dubins curve segment that satisfies the minimum turning radius is used to replace the large-angle segment in the target segment; By splicing the Dubins curve segment and the other segments in the waypoint sequence except for the large-angle segments, a continuous and smooth route path is output, thus obtaining a smooth route path that satisfies kinematic constraints.

6. The method according to claim 1, characterized in that, All cooperating drones start from the same point, end at the same point, and are the same type of drone.

7. A route planning device for multi-aircraft area coverage search based on an improved K-Means algorithm, characterized in that, include: Sub-region partitioning units are used to partition regions using an improved hybrid K-Means clustering strategy; The target area is divided into multiple sub-regions, and each sub-region corresponds to a single drone that coordinates multiple drones. Specifically, the starting and ending points of the multi-UAV collaboration are fixed cluster centers at the projection points of the target area boundary. The initial values ​​of other cluster centers are automatically selected using the K-Means++ strategy in hybrid K-Means. Based on all cluster centers, the standard K-Means clustering iteration in hybrid K-Means is used to determine the polygonal boundary of the sub-region corresponding to each cluster center. The initial path generation unit is used to generate a polyline coverage path with the minimum total cost for each sub-region based on the pre-given UAV path parameters and polygon boundary information, and obtain the path point sequence for each sub-region. The route generation unit is used to obtain a smooth route that satisfies kinematic constraints based on the minimum turning radius constraint for the path point sequence of each sub-region.

8. The apparatus according to claim 7, characterized in that, Also includes: The load balancing evaluation unit is used to calculate the total cost of the route within each sub-region and evaluate the load balancing of all route paths. Specifically, the total cost C for each route within each sub-region total = The cost C of the route length in this route path len +Turn Cost C turn That is, C total =C len +C turn ; Flight path P = {p1, p2, ..., p n The cost of increasing the length of the route is C. len ; Where d(p) i ,p i+1 () represents the Euclidean distance between adjacent path points; (x i y i ), (x i+1 y i+1 () represents adjacent path points; Turning Cost C turn for: Where, k tum θ is the turning cost coefficient. i Let be the heading angle at the i-th path point.

9. A ground control device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program in the memory, specifically performing the steps of the route planning method for multi-aircraft area coverage search based on the improved K-Means algorithm as described in any one of claims 1 to 6.

10. A drone search system, characterized in that, It includes the ground control device as described in claim 9 and multiple UAVs, and all UAVs interact with the ground control device to obtain search paths using the multi-aircraft area coverage search route planning method based on the improved K-Means algorithm as described in any one of claims 1 to 6.

Images