Unmanned aerial vehicle cluster collaborative planning method based on dynamic Voronoi diagram and improved RRT

Abstract

The invention discloses an unmanned aerial vehicle cluster collaborative planning method based on a dynamic Voronoi diagram and an improved RRT, and the method comprises the steps: taking an unmanned aerial vehicle track data set stored in a sub-airspace as training data, constructing a dynamic Voronoi diagram model based on a Bowyer-Watson algorithm, and carrying out the path planning through combining a plurality of strategies, thereby obtaining a cluster collaborative model; utilizing received unmanned aerial vehicle tracks, resources and mobile equipment data to perform edge computing collaborative optimization training, and updating to obtain a cluster collaborative prediction model; and successively predicting the to-be-predicted track, predicting track data of two broadcast moments in the future at each broadcast moment, and repeating the prediction until the track ends. The total energy consumption of the unmanned aerial vehicle cluster can be reduced, the average energy consumption of ground users is reduced, the average transmission delay of the users is shortened, and efficient data collection and collaborative operation of the unmanned aerial vehicle cluster in a complex environment are realized.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) cluster control and path planning, and in particular to a UAV cluster collaborative planning method based on a dynamic Voronoi diagram and an improved RRT. Background Art

[0002] Drone swarm deployment and trajectory planning are key research areas in drone technology. In recent years, with the rapid development of wireless communication and production technologies, small drone technology has experienced rapid growth. However, traditional methods struggle to achieve uniform coverage in complex environments and are prone to detection blind spots. Static path planning also struggles to adapt to real-time obstacle changes and struggles to balance energy consumption and latency. Currently, drone swarm deployment and trajectory planning face significant challenges.

[0003] To address these challenges, collaborative deployment and trajectory planning for drone swarms, a technology widely used in smart city construction and characterized by high efficiency, adaptability, low energy consumption, and robustness, is an attractive option. Efficient collaborative deployment and precise trajectory planning ensure stable drone operation in complex environments, improve mission success rates, and meet the growing and diverse needs of various industries. Therefore, overcoming the challenges of existing drone swarms in complex environments, such as uneven deployment, path conflicts, and energy bottlenecks, and achieving efficient collaborative deployment and dynamic trajectory planning, has become a critical issue that needs to be addressed.

[0004] The invention with application number 202110953148.7 discloses a method for autonomous exploration planning of unknown spaces, which uses Voronoi i-diagrams and RRT trees to control robots to achieve rapid, safe, and efficient autonomous exploration of unknown underground spaces. However, the flight environment of drone swarms is more complex and involves airspace, making the aforementioned method unsuitable for drone swarms. Summary of the Invention

[0005] To address the shortcomings of existing drone swarm deployment and trajectory planning technologies in complex environments, such as uneven regional coverage, lack of dynamic adaptability in path planning, and difficulty balancing energy consumption and delay, the present invention provides a drone swarm collaborative planning method based on a dynamic Voronoi diagram and improved RRT. This system integrates multiple technologies, divides the mission area into sub-airspaces, utilizes the ADS-B system to achieve drone positioning and information exchange, constructs a dynamic Voronoi diagram model based on the Bowyer-Watson algorithm for initial deployment, uses an improved RRT algorithm for path planning, and combines it with a mobile edge computing model for energy consumption optimization, thereby jointly optimizing the deployment location, flight trajectory, and energy distribution of drone swarms. The present invention aims to simultaneously reduce the total energy consumption of drone swarms, reduce the average energy consumption of ground users, and shorten the average transmission delay of users, thereby achieving efficient data collection and collaborative operations for drone swarms in complex environments and improving the overall performance and application value of drone swarms in complex scenarios.

[0006] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0007] A collaborative planning method for a UAV cluster based on a dynamic Voronoi diagram and an improved RRT, the method comprising the following steps:

[0008] S1: Divide the mission area into several sub-airspaces and build a drone positioning and information exchange system. Each sub-airspace is monitored by a central control station and equipped with an ADS-B ground base station. Drones in the sub-airspace obtain their current location information while exchanging resource information and mobile device information with the base station.

[0009] S2, based on the information exchange results of step S1, the UAV uses the airborne ADS-BOUT to broadcast its own position, resources and mobile equipment information to the sub-airspace where it is located at the broadcast time;

[0010] In step S3, based on the broadcast operation in step S2, the ADS-B ground base station responsible for receiving information in the sub-airspace receives the corresponding information and sends it to the control center. The control center stores the drone track data in the order of drone number and ADS-B broadcast time;

[0011] S4, based on the track data stored in step S3, the stored UAV track dataset in the sub-airspace is used as training data, a dynamic Voronoi diagram model is constructed based on the Bowyer-Watson algorithm, and path planning is performed in combination with multiple strategies to obtain a cluster collaboration model;

[0012] S5, based on the cluster collaboration model obtained in step S4, using the received drone track, resources and mobile device data, through edge computing collaborative optimization training, to update the cluster collaboration prediction model;

[0013] S6, using the cluster collaborative prediction model obtained in step S5, predict the track to be predicted one by one. At each broadcast time, the track data for the next two broadcast times is predicted, and the prediction is repeated until the track ends.

[0014] Step S1 further comprises:

[0015] S11, divide the mission area map into multiple sub-airspaces, set up a master control station in each sub-airspace for drone monitoring; equip each sub-airspace with an ADS-B ground base station to build an information exchange infrastructure between drones and base stations; determine that the drones in the sub-airspace are positioned using four Beidou satellites to obtain the current location information of the drones; and clarify the resource information of the base station, including wireless access points, micro-clouds, and charging piles. s , and the micro-cloud deployment logo S g (a i ), Charging pile deployment sign B g (a i ), where a i is the identifier of the i-th device or access point; the device information of the drone covers the device set D g ={d0,d1,...,d N-1}, and each device d i Maximum speed Maximum power Discretized position number L g and device status R g , where S g (a i ), B g (a i ), R g (a i )∈{0,1},d0,d1,...,d N-1 is the number of the device or access point; when S g (a i )=1, it means that the i-th access point deploys a micro cloud; when B g (a i )=1, it means that the charging pile is deployed at the i-th access point; R g (t) = 1 represents the transmission state, R g (t) = 0 represents the detection state; L g (t)∈{0,1,...,J} represents the location of the device at time t, where J represents the upper limit of the device location number;

[0016] S12. After completing the setting of the sub-airspace and related equipment information, the drone uses the onboard ADS-BOUT equipment to broadcast its received location information and mobile device information to the sub-airspace where it is located at the time of broadcasting, so that information can be shared between drones in the sub-airspace and between drones and base stations.

[0017] Furthermore, in step S2, the ground base station responsible for receiving ADS-B information in the sub-airspace receives the ADS-B information broadcast by the drone in step S1 according to the established communication protocol and frequency, and sends the received information to the control center through a wired or wireless communication link.

[0018] Step S4 further comprises:

[0019] S41, using the drone track data in the sub-airspace g stored by the control center as the initial training set, the ADS-B data is pre-processed by mean normalization;

[0020] S42, based on the mission area map, the number of drones N and the device set D = {d0, d1, ..., d N-1}, perform Delaunay triangulation to form any tetrahedron T(d j ,d k ,d l ,d m ), calculate its circumscribed sphere center C(x c ,y c ,z c ), radius r, generate Voronoi diagram to divide N sub-regions; where d j ,d k ,d l ,d m The j, k, l, and m in the Delaunay triangulation are vertex indices. j ,d k ,d l ,d m They do not directly represent elements in the device set, but rather points in the Delaunay triangulation.

[0021] S43, constructs an independent RRT tree for each UAV, with the starting point being the centroid of the Voronoi diagram and the end point being the mission target point. It detects environmental obstacles in real time and dynamically expands RRT nodes to ensure that the path is collision-free, thus obtaining a cluster collaboration model;

[0022] S44, the cluster coordination model is trained using the pre-processed initial training set. During each training process, a broadcast time is randomly selected as the starting point of the model training data. The n consecutive track points including the starting point are used as the input data of the model. The track points at the two broadcast times after the input data are used as the cluster coordination data. After the training is completed, the initial low-altitude airspace UAV cluster coordination model is obtained.

[0023] Step S42 further includes:

[0024] Generate a weighted Voronoi diagram based on a 3D terrain model that includes elevation and obstacle data:

[0025] Weight ω(p) = α·elevation(p) + β·obstacle density(p);

[0026] Among them, p is a symbol related to weight, which represents the weight related to the attributes of the points in the task area (such as height, obstacle density, etc.).

[0027] Initialize the construction of the hypertetrahedron T containing all points init , calculate the center of the circumscribed sphere based on the vertices of the tetrahedron:

[0028]

[0029] Among them, j, k, l, m are the vertex indices of Delaunay triangulation, x j ,y j ,z j is the three-dimensional position coordinate of the vertex with index j in the Delaunay triangulation, x c ,y c ,z c is the calculated three-dimensional position coordinate of the center of the circumscribed sphere.

[0030] Insertion point d new If d≤r, delete the conflicting tetrahedron to form a cavity Ψ, and connect the boundary surface of the cavity with d new Connect, generate new tetrahedrons, and then verify the Delaunay conditions:

[0031]

[0032] Where C is the center of the circumscribed sphere, which is used to calculate the circumscribed sphere of the tetrahedron in the Delaunay triangulation.

[0033] Generate Voronoi diagram by optimizing region partitioning through convex hull monitoring:

[0034]

[0035] Among them, α and β are adjustment coefficients, and d is the insertion point d new The distance to the circumcenter of tetrahedron T, ||xd i || is the Euclidean distance, and i and j are the indices of the device or task point.

[0036] Step S43 further includes:

[0037] The target bias strategy and obstacle density adaptive step size strategy are introduced to directly expand to the target point with probability ρ to accelerate convergence. Multi-path generation, efficiency optimization and path smoothing are introduced to enhance planning capabilities. The target bias strategy is:

[0038]

[0039] The obstacle density adaptive step size strategy is:

[0040]

[0041] The efficiency optimization formula is:

[0042] ||q new -q goal ||2≤Thr;

[0043] The path smoothing formula is:

[0044]

[0045] Among them, Δq is the adaptive expansion step size, Δ base is the basic expansion step, q rand is a random sampling point, q near For q rand The node with the closest Euclidean distance; if q near to q goal If the Euclidean distance of the original path P = {p1, p2, ..., p n} to simplify, starting from the starting point p j Search backward for the farthest reachable point p k , satisfying the line segment If there is no collision, then p k* Join P smooth , update j*=k*, repeat until reaching the end point, and achieve path smoothing; where j* and k* are the indices of each node in the path.

[0046] Step S5 further comprises:

[0047] S51, obtain the stored UAV track data DU g(i) Received resource data and mobile device data, providing the acquired data as input to the cluster collaboration model

[0048] S52, using edge computing technology, takes the total mobile energy consumption minimization objective function as the objective function to analyze the cluster collaboration model Perform collaborative optimization training to obtain a cluster collaborative prediction model The object of edge computing is the drones in the entire airspace, and the objective function is:

[0049]

[0050] The constraints are:

[0051] Coverage constraints:

[0052] Service delay constraints:

[0053] Power Constraint: P i (t) = P i (0)-α2·D(L i (t),L i (t+1))-β2·R(t,i)+γ·B(L i (t))≥0 and P i (t)≤p i max

[0054] Speed constraint: D(L i (t),L i (t+1))≤v i max ·Δt

[0055] Among them, D(A,B) represents the distance between positions A and B, α2 is the energy consumption coefficient of moving unit distance, x(t,j,i)∈{0,1}, when device d i When the coverage area j' is set to 1 at time t, β2 is the energy consumption coefficient of the transmission state, and γ is the charging efficiency coefficient of the charging pile; P i (t) represents the power of device i at time t, R(t,i) represents the energy consumption of the transmission state, L i (t) represents the location of device i at time t in the entire airspace, and L i (t)∈{0,1,...,N-1},B(L i(t)) indicates whether a charging pile is deployed at the current location of the device, and Δt is the time interval; the coverage constraint requires that each detection area j'∈{0,1,...,J} be visited at least once within time T, the service delay constraint requires that the data of each device be uploaded to the cloud at least once within time T, the power constraint requires that the device power meets the lower limit and does not exceed the maximum capacity, and the speed constraint requires that the device movement speed does not exceed its maximum capacity.

[0056] Step S6 further comprises:

[0057] S61, using cluster collaborative prediction model The predicted track is calculated to obtain the predicted data for the next two broadcast times t+1 and t+2.

[0058] S62, at broadcast time t+1, the cluster collaborative prediction model is used again The predicted track is calculated to obtain the predicted data for the next two broadcast times t+2 and t+3, and this operation is repeated until the track ends.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] First, the UAV cluster collaborative planning method based on dynamic Voronoi diagram and improved RRT of the present invention deeply integrates the unique advantages of the ADS-B system in UAV cluster positioning. ADS-B can achieve millisecond-level information updates, support high-frequency position refreshes, and improve timeliness by 10 times compared to GPS. Its three-dimensional positioning accuracy is high, and it can simultaneously obtain heading angle and airspeed data to achieve dynamic track prediction. The system adopts a distributed architecture, with wide coverage of a single base station and low construction cost, and is particularly suitable for complex terrain. Actual measurements show that ADS-B reduces cluster trajectory collaborative errors by 62%, supports high-density real-time path planning for 200+ drones, and significantly improves airspace utilization efficiency and emergency response capabilities.

[0061] Second, the present invention's collaborative planning method for drone swarms, based on a dynamic Voronoi diagram and improved RRT, leverages multi-algorithm collaborative optimization to address complex scenarios. This method uses a dynamic Voronoi diagram to implement regional responsibility division, combined with an improved RRT algorithm for local obstacle avoidance. This addresses the contradiction between uneven global coverage and local path conflicts in traditional methods. A weighted Voronoi diagram model is introduced for complex terrain, dynamically adjusting regional divisions based on elevation and obstacle density to ensure reasonable deployment in three-dimensional environments such as mountainous and urban areas.

[0062] Third, the present invention's collaborative planning method for drone swarms, based on a dynamic Voronoi diagram and improved RRT, leverages the advantages of dynamic route planning and edge computing for real-time obstacle avoidance and energy optimization. The improved RRT algorithm directly expands toward the target point with probability ρ to accelerate convergence. A greedy algorithm simplifies the path and eliminates redundant nodes. Based on a mobile edge computing model, it integrates power constraints, coverage requirements, and low-latency transmission requirements to dynamically coordinate charging paths and task priorities. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A model of a UAV collaborative positioning system based on the ADS-B system and auxiliary edge computing servers in low-altitude airspace;

[0064] Figure 2 This is a flow chart of the algorithm proposed in the present invention;

[0065] Figure 3 The process and results of drawing Voronoi diagrams for the case of random distribution of drone points;

[0066] Figure 4 Comparison of the time required to deploy the Voronoi diagram under different distribution conditions;

[0067] Figure 5 A comparison chart of single path and multi-path for RRT path planning;

[0068] Figure 6 A preview of the algorithm effect to show whether the path planning takes efficiency into consideration;

[0069] Figure 7 Schematic diagram of the effect of using a greedy algorithm for smoothing problems;

[0070] Figure 8 It is the result of auxiliary edge computing;

[0071] Figure 9 A comparison chart of energy optimization results under different algorithms. DETAILED DESCRIPTION

[0072] The embodiments of the present invention are described in further detail below with reference to the accompanying drawings.

[0073] The embodiment of the present invention provides a UAV cluster collaborative planning method based on a dynamic Voronoi diagram and an improved RRT, and the specific steps are as follows:

[0074] Step 1: Divide the mission area into several sub-airspaces and build a drone positioning and information exchange system. Each sub-airspace is monitored by a central control station and equipped with an ADS-B ground base station. Drones within the sub-airspace obtain their current location information using four Beidou satellites and exchange resource and mobile device information with the base station.

[0075] Step 1 specifically includes:

[0076] Step 1-1: Divide the mission area map into multiple sub-airspaces, and set up a master control station in each sub-airspace for drone monitoring. Equip each sub-airspace with an ADS-B ground base station to build an information exchange infrastructure between drones and base stations. Determine that the drones in the sub-airspace are positioned using four BeiDou satellites to obtain the current location of the drone. At the same time, clarify the resource information of the base station, including wireless access points, micro-clouds, and charging piles. s , and the micro-cloud deployment logo S g (a i ), Charging pile deployment sign B g (a i ), where a i is the identifier of the i-th device; the device information of the drone covers the device set D g ={d0,d1,...,d N-1}, and each device d i Maximum speed Maximum power Discretized position number L g and device status R g , where S g (a i ), B g (a i ), R g (a i )∈{0,1}, i represents the number of the access point, d0,d1,...,d N-1 is the number of the device or access point; when S g (a i )=1, it means that the i-th access point deploys a micro cloud; when B g (a i )=1, it means that the charging pile is deployed at the i-th access point; R g (t) = 1 represents the transmission state, R g (t) = 0 represents the detection state; L g (t)∈{0,1,...,J} represents the location of the device at time t, where J represents the upper limit of the device location number;

[0077] Step 1-2: After completing the sub-airspace and related equipment information settings, the drone uses the onboard ADS-BOUT equipment to broadcast its received position information U g (i) (A g 、S g (a i ), B g (a i)) and mobile device information (D g , L g 、R g ), broadcast to the sub-airspace where it is located, and realize information sharing between drones in the sub-airspace and between drones and base stations. g is the collection of charging piles in the sub-airspace g.

[0078] Step 2: Based on the information obtained in step 1, the drone uses the onboard ADS-BOUT to broadcast its own location, resources, and mobile device information to the sub-airspace where it is located at the broadcast time.

[0079] Step 2 specifically includes:

[0080] Step 2-1: The ground base station responsible for ADS-B information reception in the sub-airspace receives the ADS-B information broadcast by the drone in step 1-2 according to the established communication protocol and frequency.

[0081] Step 2-2: After receiving the ADS-B information, the ground base station sends this information to the control center through a wired or wireless communication link.

[0082] Step 3: Based on the broadcast operation in step 2, the ADS-B ground base station responsible for receiving information in the sub-airspace receives the corresponding information and sends it to the control center;

[0083] Step 3 specifically includes:

[0084] Step 3-1: The control center receives the ADS-B information sent by the ground base station in step 2-2, and sorts the information according to the ADS-B broadcast time according to the drone number.

[0085] Step 3-2: The control center stores the sorted information as the drone track DU g (i).

[0086] Step 4: Based on the information collected in step 3, the control center stores the drone track data in the order of drone number and ADS-B broadcast time, in preparation for subsequent cluster collaboration;

[0087] Step 4 specifically includes:

[0088] Step 4-1: Using the drone track data in the sub-airspace g stored by the control center as the initial training set, the ADS-B data is preprocessed by mean normalization before inputting the data into the drone cluster collaborative deployment and trajectory planning model training.

[0089] Step 4-2: Based on the mission area map, the number of drones N and the equipment set D = {d0, d1, ..., d N-1}, perform Delaunay triangulation to form any tetrahedron T(d j ,d k ,d l ,d m ), calculate its circumscribed sphere center C(x c ,y c ,z c ), radius r, generate a Voronoi diagram to divide N sub-areas.

[0090] Weight ω(p) = α·elevation(p) + β·obstacle density(p) (1)

[0091]

[0092] Among them, p is a symbol related to weight, which represents the weight related to the attributes of the points in the task area (such as height, obstacle density, etc.); j, k, l, m are the vertex indices of Delaunay triangulation, x j ,y j ,z j is the three-dimensional position coordinate of the vertex with index j in the Delaunay triangulation, x c ,y c ,z c is the calculated three-dimensional position coordinate of the center of the circumscribed sphere; C represents the center of the circumscribed sphere, which is used to calculate the circumscribed sphere of the tetrahedron in the Delaunay triangulation; α and β are adjustment coefficients, and d is the insertion point d new The distance to the circumcenter of tetrahedron T, ||xd i || is the Euclidean distance, i and j are the indices of the device or task point.

[0093] Based on the Bowyer-Watson algorithm, a dynamic Voronoi diagram model is constructed. Generally speaking, a weighted Voronoi diagram (Formula 1) is generated based on the three-dimensional terrain model (including elevation and obstacle data). First, the hypertetrahedron T containing all points is initialized. init , calculate the center of the circumscribed sphere according to the vertices of the tetrahedron (Formula 2), and then insert point d new If d≤r, delete the conflicting tetrahedron (Formula 3) to form a cavity Ψ, and connect the boundary surface of the cavity with d new Connect and generate new tetrahedrons, then verify the Delaunay condition, namely the empty sphere criterion (Formula 4), and generate the Voronoi diagram by optimizing the region partition through convex hull monitoring to achieve efficient initial deployment of the UAV cluster.

[0094] Next, a separate RRT tree is constructed for each UAV, starting at the Voronoi diagram centroid and ending at the mission target. Environmental obstacles are detected in real time, and RRT nodes are dynamically expanded to ensure a collision-free path. A target bias strategy (Equation 6) and an obstacle density adaptive step size strategy (Equation 7) are introduced to directly expand toward the target point with probability ρ, accelerating convergence.

[0095]

[0096] ||q new -q goal ||2≤Thr (8)

[0097]

[0098] Among them, Δq is the adaptive expansion step size, Δ base is the basic expansion step, q rand is a random sampling point, q near For q rand The node with the closest Euclidean distance.

[0099] At the same time, we introduce multi-path generation, efficiency optimization (Formula 8) and path smoothing (Formula 9) to enhance planning capabilities. goal If the Euclidean distance of the original path P = {p1, p2, ..., p n} to simplify, starting from the starting point p i Search backward for the farthest reachable point p k* , satisfying the line segment If there is no collision, then p k* Join P smooth , update j*=k*, repeat until reaching the end point, and achieve path smoothing. Where j* and k* are the indices of each node in the path.

[0100] During each training process, a broadcast time is randomly selected as the starting point of the model training data, and then n consecutive track points including the starting point are used as the input data of the model. The track points of the two broadcast times after the input data are used as cluster coordination data. After the training is completed, the initial low-altitude airspace UAV cluster coordination model is obtained.

[0101] Step 5: Based on the track data stored in step 4, the stored UAV track dataset in the sub-airspace is used as training data, and a dynamic Voronoi diagram model is constructed based on the Bowyer-Watson algorithm. Path planning is performed in combination with multiple strategies to obtain a cluster collaboration model.

[0102] Step 5 specifically includes:

[0103] Step 5-1: Get the drone track data DU stored in step 4 g (i) Received resource data and mobile device data, which are provided as input to the cluster collaboration model

[0104] Step 5-2: Using edge computing technology, the target is the drones in the entire airspace, and the objective function is to minimize the total mobile energy consumption (Formula 10):

[0105]

[0106] At the same time, the cluster collaboration model can meet the coverage constraint (Formula 11), service delay constraint (Formula 12), power constraint (Formula 13), speed constraint (Formula 14) and other constraints. Perform collaborative optimization training to obtain a cluster collaborative prediction model

[0107]

[0108] P i (t) = P i (0)-α2·D(L i (t),L i (t+1))-β2·R(t,i)+γ·B(L i (t))≥0 and P i (t)≤p i max (13)

[0109] D(L i (t),L i (t+1))≤v i max ·Δt (14)

[0110] Among them, D(A,B) represents the distance between positions A and B, α2 is the energy consumption coefficient of moving unit distance, x(t,j,i)∈{0,1}, when device d i When the coverage area j' is set to 1 at time t, β2 is the energy consumption coefficient of the transmission state, and γ is the charging efficiency coefficient of the charging pile. i (t) represents the power of device i at time t, R(t,i) represents the energy consumption of the transmission state, L i (t) represents the location of device i at time t in the entire airspace, and L i (t)∈{0,1,...,N-1},B(L i(t)) indicates whether a charging station is deployed at the device's current location, and Δt is the time interval. The coverage constraint requires that each detection area j'∈{0,1,...,J} be visited at least once within time T. The service delay constraint requires that each device's data be uploaded to the cloud at least once within time T. The power constraint requires that the device's power meet the lower limit and do not exceed the maximum capacity. The speed constraint requires that the device's movement speed does not exceed its maximum capacity.

[0111] Step 6: Based on the cluster collaboration model obtained in step 5, the received drone track, resource and mobile device data are used to perform edge computing collaborative optimization training to update the cluster collaboration prediction model.

[0112] Step 6 specifically includes:

[0113] Step 6-1: Use cluster collaborative prediction model The predicted track is calculated to obtain the predicted data for the next two broadcast times t+1 and t+2.

[0114] Step 6-2: At broadcast time t+1, use the cluster collaborative prediction model again The predicted track is calculated to obtain the predicted data for the next two broadcast times t+2 and t+3, and this operation is repeated until the track ends.

[0115] Step 7: Use the cluster collaborative prediction model obtained in step 6 to predict the trajectory one by one. At each broadcast time, predict the trajectory data for the next two broadcast times. Repeat this prediction until the trajectory is complete.

[0116] Step 7 specifically includes:

[0117] Step 7-1: Set the starting broadcast time for the first prediction and specify the task of using the cluster collaborative prediction model Pre2 to predict the data for the next two broadcast times.

[0118] Step 7-2: At each broadcast moment, the latest drone location, resource, and mobile device information is input into the cluster collaborative prediction model Calculate and output the track prediction data for the next two broadcast times, and continue this operation until the track ends.

[0119] Figure 1The figure shows a collaborative drone positioning system in low-altitude airspace based on ADS-B and auxiliary edge computing servers. This system uses Voronoi diagrams to dynamically divide airspaces into sub-areas. MEC nodes are deployed in each sub-area to receive drone ADS-B data in real time. This system combines distributed preprocessing with the interaction of spatiotemporal parameters of neighboring nodes to build a collaborative solution architecture with multiple base station virtual anchors. This system achieves high-precision correction of drone positions through a dynamic load balancing strategy across MEC domains. The fused positioning results are fed back to ATC for trajectory prediction and conflict warning, significantly improving collaborative obstacle avoidance response capabilities in low-altitude, densely populated scenarios.

[0120] Figure 3 The process and results of drawing Voronoi diagrams when the UAV points are randomly distributed are given. The Delaunay triangulation construction process based on the Bowyer-Watson algorithm optimizes the generation accuracy of the three-dimensional Voronoi diagram and provides a geometric basis for the initial deployment of UAVs. The figure shows the process of drawing Voronoi diagrams from the initial Delaunay triangulation ( Figure 3 (a) in the figure), after convex hull monitoring optimization ( Figure 3 (b)) to the final Voronoi diagram ( Figure 3 The complete process of (c)).

[0121] Figure 4 The generation performance of Voronoi diagrams under different point distribution patterns was compared. The time consumption for Voronoi diagram generation was calculated for five point distribution patterns (random, circular, double-row, random triangle with jitter, and center-centered spiral). The horizontal axis represents the distribution type, and the vertical axis represents the deployment time (in seconds). The results show that the algorithm is most efficient (<3.5 seconds) under uniform distribution (circular, double-row). Complex distribution (random triangle with jitter) requires multiple cavity reconstructions, which takes longer, but still outperforms traditional methods. Through convex hull monitoring and an adaptive step size strategy, the overall deployment time is reduced by 30%.

[0122] Figure 5 The improved RRT algorithm demonstrates its multi-path generation capabilities. The black areas in the figure represent obstacles, the gray lines represent the decision-making process for path planning, and the colored curves represent the final planned path. Unlike the single-tree growth model of the traditional RRT algorithm (left), the improved RRT algorithm (right) uses a parallel tree expansion mechanism to form a multi-branch collaborative exploration topology network, continuously retaining the growth potential of alternative paths and reducing the risk of path interruption in narrow passages or dynamic obstacle scenarios. The results show that the system can quickly generate multiple collision-free paths in complex environments, demonstrating a high degree of robustness.

[0123] Figure 6This comparison shows the path planning performance of an improved RRT algorithm (considering efficiency optimization) and a traditional RRT algorithm. The black areas in the figure represent obstacles, the gray segments represent the path planning decision-making process, and the colored curves represent the final planned path. The left figure, which ignores efficiency, exhibits many redundant nodes, slow convergence, and unnecessary detours. The right figure, which considers efficiency by introducing a target bias strategy and adaptive step size, reduces the path length by 20% and planning time by 40%. The obstacles in the background demonstrate the algorithm's dynamic obstacle avoidance capabilities.

[0124] Figure 7 The results before and after path smoothing are compared. The left figure shows a single path plan, and the right figure shows a multi-path plan. The colored curve in the figure represents the original path generated by RRT, which exhibits jagged fluctuations and is detrimental to stable flight. The black curve represents the smoothed path generated using a greedy algorithm. This algorithm gradually connects the farthest reachable points from the starting point, eliminating redundant nodes. This improves path smoothness by 60% and reduces energy consumption by 15%. The smoothed path better conforms to the drone's kinematic constraints.

[0125] Figure 8 This paper demonstrates the collaborative optimization architecture of a mobile edge computing model. A swarm of drones uploads ADS-B data to the edge cloud (auxiliary edge computing nodes) in real time. Mobile edge computing uses a dynamic Voronoi diagram to divide areas of responsibility and generates routing instructions based on an improved RRT. The power monitoring module dynamically dispatches drones to charging stations (marked by squares), while micro-cloud nodes (marked by contour lines) prioritize high-priority task data. The overall architecture achieves joint optimization of energy consumption, coverage, and latency, reducing total system energy consumption by 25%.

[0126] Figure 9 The results of energy optimization using different algorithms in a mobile edge technology model are presented. Experiments show that within 1-30 iterations, the total energy consumption of the cluster system using the dynamic power optimization solution (HUECM) is reduced from the initial 850J to 358J, a 56.9% reduction compared to the traditional fixed power deployment solution (FFP), and the coverage is increased by 2.32 times. It can be seen that dynamic power optimization achieves performance improvements through the following mechanisms: 1) Balance energy consumption and load, through L E The system optimizes data throughput per unit energy; 2) real-time power regulation dynamically adjusts transmit power based on channel conditions (with a threshold of ±15dBm); and 3) applies a convergence acceleration algorithm with an adaptive iterative step size, reducing stabilization time to one-third that of traditional solutions. This approach demonstrates the significant advantages of dynamic optimization strategies in high-density IoT scenarios, making it suitable for large-scale mobile node deployments such as emergency communications and wide-area IoT.

[0127] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0128] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT, characterized by: The method comprises the following steps: S1: Divide the mission area into several sub-airspaces and build a drone positioning and information exchange system. Each sub-airspace is monitored by a central control station and equipped with an ADS-B ground base station. Drones in the sub-airspace obtain their current location information while exchanging resource information and mobile device information with the base station. S2, based on the information exchange results of step S1, the UAV uses the airborne ADS-BOUT to broadcast its own position, resources and mobile equipment information to the sub-airspace where it is located at the broadcast time; In step S3, based on the broadcast operation in step S2, the ADS-B ground base station responsible for receiving information in the sub-airspace receives the corresponding information and sends it to the control center. The control center stores the drone track data in the order of drone number and ADS-B broadcast time; S4, based on the track data stored in step S3, the stored UAV track dataset in the sub-airspace is used as training data, a dynamic Voronoi diagram model is constructed based on the Bowyer-Watson algorithm, and path planning is performed in combination with multiple strategies to obtain a cluster collaboration model; S5, based on the cluster collaboration model obtained in step S4, using the received drone track, resources and mobile device data, through edge computing collaborative optimization training, to update the cluster collaboration prediction model; S6, using the cluster collaborative prediction model obtained in step S5, predict the track to be predicted one by one. At each broadcast time, the track data for the next two broadcast times is predicted, and the prediction is repeated until the track ends.

2. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 1 is characterized in that: Step S1 further comprises: S11, divide the mission area map into multiple sub-airspaces, set up a master control station in each sub-airspace for drone monitoring; equip each sub-airspace with an ADS-B ground base station to build an information exchange infrastructure between drones and base stations; determine that the drones in the sub-airspace are positioned using four Beidou satellites to obtain the current location information of the drones; and clarify the resource information of the base station, including wireless access points, micro-clouds, and charging piles. s , and the micro-cloud deployment logo S g (a i ), Charging pile deployment sign B g (a i ), where a i is the identifier of the i-th device or access point; the device information of the drone covers the device set D g ={d0,d1,...,d N-1 }, and each device d i Maximum speed Maximum power Discretized position number L g and device status R g , where S g (a i ), B g (a i ), R g (a i )∈{0,1},d0,d1,...,d N-1 is the number of the device or access point; when S g (a i )=1, it means that the i-th access point deploys a micro cloud; when B g (a i )=1, it means that the charging pile is deployed at the i-th access point; R g (t) = 1 represents the transmission state, R g (t) = 0 represents the detection state; L g (t)∈{0,1,...,J} represents the location of the device at time t, where J represents the upper limit of the device location number; S12. After completing the setting of the sub-airspace and related equipment information, the drone uses the onboard ADS-BOUT equipment to broadcast its received location information and mobile device information to the sub-airspace where it is located at the time of broadcasting, so that information can be shared between drones in the sub-airspace and between drones and base stations.

3. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 1 is characterized in that: In step S2, the ground base station responsible for receiving ADS-B information in the sub-airspace receives the ADS-B information broadcast by the drone in step S1 according to the established communication protocol and frequency, and sends the received information to the control center through a wired or wireless communication link.

4. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 1 is characterized in that: Step S4 further comprises: S41, using the drone track data in the sub-airspace g stored by the control center as the initial training set, the ADS-B data is pre-processed by mean normalization; S42, based on the mission area map, the number of drones N and the device set D = {d0, d1, ..., d N-1 }, perform Delaunay triangulation to form any tetrahedron T(d j ,d k ,d l ,d m ), calculate its circumscribed sphere center C(x c ,y c ,z c ), radius r, generate Voronoi diagram to divide N sub-regions; where d j ,d k ,d l ,d m j, k, l, m are the vertex indices in the Delaunay triangulation; S43, constructs an independent RRT tree for each UAV, with the starting point being the centroid of the Voronoi diagram and the end point being the mission target point. It detects environmental obstacles in real time and dynamically expands RRT nodes to ensure that the path is collision-free, thus obtaining a cluster collaboration model; S44, the cluster coordination model is trained using the pre-processed initial training set. During each training process, a broadcast time is randomly selected as the starting point of the model training data. The n consecutive track points including the starting point are used as the input data of the model. The track points at the two broadcast times after the input data are used as the cluster coordination data. After the training is completed, the initial low-altitude airspace UAV cluster coordination model is obtained. .

5. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 4 is characterized in that: Step S42 further includes: Generate a weighted Voronoi diagram based on a 3D terrain model that includes elevation and obstacle data: Weight ω(p) = α·elevation(p) + β·obstacle density(p); Among them, p is a symbol related to weight, which represents the weight related to the attribute of the point in the task area; Initialize the construction of the hypertetrahedron T containing all points init , calculate the center of the circumscribed sphere based on the vertices of the tetrahedron: Among them, j, k, l, m are the vertex indices of Delaunay triangulation, (x j ,y j ,z j )、(x k ,y k ,z k )、(x l ,y l ,z l ) and (x m ,y m ,z m ) are the three-dimensional position coordinates of the vertices indexed as j, k, l and m in the Delaunay triangulation, x c ,y c ,z c is the calculated three-dimensional position coordinate of the center of the circumscribed sphere; Insertion point d new If d≤r, delete the conflicting tetrahedron to form a cavity Ψ, and connect the boundary surface of the cavity with d new Connect, generate new tetrahedrons, and then verify the Delaunay conditions: Where C represents the center of the circumscribed sphere, which is used to calculate the circumscribed sphere of the tetrahedron in the Delaunay triangulation; Generate Voronoi diagram by optimizing region partitioning through convex hull monitoring: Among them, α and β are adjustment coefficients, and d is the insertion point d new The distance to the circumcenter of tetrahedron T, ||xd i || is the Euclidean distance, and i and j are the indices of the device or task point.

6. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 4 is characterized in that: Step S43 further includes: The target bias strategy and obstacle density adaptive step size strategy are introduced to directly expand to the target point with probability ρ to accelerate convergence. Multi-path generation, efficiency optimization and path smoothing are introduced to enhance planning capabilities. The target bias strategy is: The obstacle density adaptive step size strategy is: The efficiency optimization formula is: ||q new -q goal ||2≤Thr; The path smoothing formula is: Among them, Δq is the adaptive expansion step size, Δ base is the basic expansion step, q rand is a random sampling point, q near For q rand The node with the closest Euclidean distance; if q near to q goal If the Euclidean distance of the original path P = {p1, p2, ..., p n } to simplify, starting from the starting point p j Search backward for the farthest reachable point p k , satisfying the line segment If there is no collision, then p k* Join P smooth , update j*=k*, repeat until reaching the end point, and achieve path smoothing; where j* and k* are the indexes of each node in the path.

7. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 1, characterized in that: Step S5 further comprises: S51, obtain the stored UAV track data DU g (i) Received resource data and mobile device data, providing the acquired data as input to the cluster collaboration model S52, using edge computing technology, takes the total mobile energy consumption minimization objective function as the objective function to analyze the cluster collaboration model Perform collaborative optimization training to obtain a cluster collaborative prediction model The object of edge computing is the drones in the entire airspace, and the objective function is: The constraints are: Coverage constraints: Service delay constraints: Power Constraint: P i (t) = P i (0)-α2·D(L i (t),L i (t+1))-β2·R(t,i)+γ·B(L i (t))≥0 and P i (t)≤p i max Speed constraint: D(L i (t),L i (t+1))≤v i max ·Δt Among them, D(A,B) represents the distance between positions A and B, α2 is the energy consumption coefficient of moving unit distance, x(t,j,i)∈{0,1}, when device d i When the coverage area j' is set to 1 at time t, β2 is the energy consumption coefficient of the transmission state, and γ is the charging efficiency coefficient of the charging pile; P i (t) represents the power of device i at time t, R(t,i) represents the energy consumption of the transmission state, L i (t) represents the location of device i at time t in the entire airspace, and L i (t)∈{0,1,...,N-1},B(L i (t)) indicates whether a charging pile is deployed at the current location of the device, and Δt is the time interval; the coverage constraint requires that each detection area j'∈{0,1,...,J} be visited at least once within time T, the service delay constraint requires that the data of each device be uploaded to the cloud at least once within time T, the power constraint requires that the device power meets the lower limit and does not exceed the maximum capacity, and the speed constraint requires that the device movement speed does not exceed its maximum capacity.

8. The UAV swarm collaborative planning method based on dynamic Voronoi diagram and improved RRT according to claim 1 is characterized in that: Step S6 further comprises: S61, using cluster collaborative prediction model The predicted track is calculated to obtain the predicted data for the next two broadcast times t+1 and t+2. S62, at broadcast time t+1, the cluster collaborative prediction model is used again The predicted track is calculated to obtain the predicted data for the next two broadcast times t+2 and t+3, and this operation is repeated until the track ends.

Images