A method for merging and splitting a drone cluster

Abstract

The application discloses a kind of unmanned aerial vehicle cluster merging and sub-cluster method, first for multi-cluster merging problem, by introducing cluster label and active times two variables to each unmanned aerial vehicle, unmanned aerial vehicle from different clusters can be ordered according to the requirement of task to complete fusion;For the problem that unmanned aerial vehicle cluster is divided into multiple sub-clusters, introduce genetic algorithm to process the adjacency matrix of unmanned aerial vehicle cluster, find the node division sequence with the least cut edge, disconnect the connection between unmanned aerial vehicle on both ends of cut edge, that is, the cluster can be divided into several sub-clusters required by task, then determine the key node according to the degree centrality index of unmanned aerial vehicle in each cluster, and select it as the new leader unmanned aerial vehicle of each cluster;The application can efficiently realize the merging and sub-cluster of large-scale unmanned aerial vehicle cluster.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) technology, specifically relating to a method for merging and grouping UAV swarms. Background Technology

[0002] With advancements in artificial intelligence, robotics, and data fusion technologies, drone swarms are widely used in both military and civilian fields. Furthermore, with technological development, drone swarms are trending towards miniaturization, intelligence, and large-scale deployment. While fully autonomous large-scale drone swarms represent the future, current limitations in artificial intelligence technology mean that complete replacement of human-machine collaboration by AI is unlikely in the short term. Currently, AI can achieve unmanned operation in some areas through machine learning and deep learning. However, AI is ineffective for creative tasks, tasks lacking clear definitions and boundaries, and tasks requiring digital knowledge and experience. Therefore, for the foreseeable future, intelligent machines will primarily play a supporting role in task execution, with final decision-making still relying on humans. Due to the complexity and variability of drone missions, human involvement in decision-making and external command control are still necessary to ensure successful mission completion. For large-scale drone swarms, a common operating mode is the "leader-follower" model, where a specific drone in the swarm is selected as the leader to directly receive commands from the control center, while the other drones receive commands from the leader, enabling the entire swarm to perform collaborative tasks under external command control.

[0003] For large-scale drone swarms, due to their large size, simultaneous takeoff from the same location is often impossible in practical applications. A common approach is for multiple smaller drone swarms to take off from different locations based on mission requirements, then merge into a larger swarm to execute the mission. When the mission is completed and the drones need to return, or when a small number of drones are required to perform other tasks, the large-scale drone swarm needs to be split into smaller swarms to land or continue performing other tasks. To ensure that the swarm remains under external command control throughout the entire process, and that the merged and split swarms can still quickly achieve state convergence and execute collaborative tasks, the following issues need to be considered:

[0004] When multiple clusters are merged into a large-scale drone swarm, the leader drones are directly dispatched by the control center, allowing for rapid identification of the new leader and completion of the merge. However, follower drones in the swarm lack decision-making capabilities and can only follow the leader drones. Therefore, follower drones are unaware of the specific requirements of the merge task. If follower drones from two clusters interact directly, it inevitably leads to chaotic coordination. Therefore, ensuring the stability of the drone states during the merge of multiple clusters and rapidly completing the merge and state convergence is a crucial issue.

[0005] In the process of dividing a large-scale drone swarm into multiple sub-swarms, it is of great significance to complete the grouping according to the needs of the mission and with the least amount of operation and energy consumption. After the grouping is completed, the multiple different sub-swarms need to quickly select their own new leaders to ensure that the large-scale swarm can still quickly achieve convergence of the cooperative state and continue to execute different tasks. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, this invention provides a method for merging and grouping drone swarms. Firstly, for the problem of merging multiple swarms, two variables—a swarm number and the number of active users—are introduced for each drone, allowing drones from different swarms to be merged in an orderly manner according to task requirements. Secondly, for the problem of dividing a drone swarm into multiple sub-swarms, a genetic algorithm is introduced to process the adjacency matrix of the drone swarm, finding the node partitioning sequence with the fewest cut edges. By disconnecting the drones at the ends of the cut edges, the swarm can be divided into several sub-swarms as required by the task. Subsequently, key nodes are determined based on the degree centrality index of the drones in each swarm, and these nodes are selected as the new leader drone for each swarm. This invention can efficiently achieve the merging and grouping of large-scale drone swarms.

[0007] The technical solution adopted by this invention to solve its technical problem includes the following steps:

[0008] Step 1: Construct a drone swarm model;

[0009] A drone swarm consists of identical drones that can communicate with each other. Therefore, the topology of a drone swarm can be represented by an undirected graph G = (V, E), where V = {v1, v2, ..., v...}. n} represents the set of all drones in the cluster. This represents a set of connections between drones; let Let a represent the adjacency matrix associated with an undirected graph G, where a ij When a is 1, it indicates that drone i and drone j can communicate with each other; when a ijWhen the value is 0, it indicates that drone j is outside the communication range of drone i; each drone swarm consists of a leader and several follower drones;

[0010] The state of each drone in the cluster is represented as follows:

[0011]

[0012] Where x n (t) represents various values ​​related to flight status;

[0013] If the states of all drones in a cluster converge to the state of the leader drone, then the drone cluster is considered to have achieved a cooperative state with external control, i.e., it satisfies the following equation:

[0014]

[0015] Where, x L (t) represents the status information of the leader drone, which is determined by the control center;

[0016] Step 2: Merge the drone swarm;

[0017] Suppose that drone clusters m and n need to be merged, and according to the control center's decision, the leader of the new merged cluster is the leader of cluster n, meaning cluster m needs to be integrated into cluster n. The specific steps are as follows:

[0018] Step 2-1: For drones in cluster m, maintain cluster number g_num as m; for drones in cluster n, maintain cluster number g_num as n; and follow drones only receive information from drones with the same g_num as themselves; set the initial number of active times for drones;

[0019] Step 2-2: When the leaders of clusters m and n receive the merge command, the two clusters begin to approach each other;

[0020] Steps 2-3: When the follower drone in cluster m receives a message from the drone in cluster n, the message is ignored because the cluster numbers are inconsistent.

[0021] Steps 2-4: When the leader in cluster m receives a merge message from a drone in cluster n, it changes its identity to a follower, changes g_num to n, and joins cluster n with the drone from which the message originated as its parent node.

[0022] Steps 2-5: In each iteration, if a follower in cluster m does not receive the leader's status information, the active count is decremented by 1.

[0023] Steps 2-6: When the number of active drones in cluster m drops to 0, change the cluster number g_num to 0 and change itself to leaderless mode;

[0024] Steps 2-7: When a drone in leaderless mode receives a message from cluster n, it changes its identity to follower drone, changes the cluster number g_num to n, and joins cluster n with the drone that sent the message as its parent node.

[0025] Steps 2-8: Once all leaderless drones have joined cluster n, the merging process from cluster m to cluster n is complete.

[0026] Step 3: Drone swarming;

[0027] Step 3-1: Based on the task requirements, determine the number of sub-clusters k needed and the limit on the number of drones in each cluster;

[0028] Step 3-2: Represent the arrangement of nodes corresponding to the adjacency matrix as a vector. The adjacency matrix is ​​transformed by transforming the vector, and the optimal node arrangement is solved using a genetic algorithm.

[0029] Step 3-3: Determine the fitness function that minimizes the number of cut edges. EC is the cut edge set, which is the set of lines connecting nodes in an undirected graph G that are not in the same subset after partitioning. It is defined as follows:

[0030] EC={e s,t |s∈V i ,t∈V j ,1≤i,j≤k,i≠j}

[0031] Among them, e s,t This represents the edge connecting drones located in two different clusters, where s and t represent drones located in different drone clusters, respectively. j A collection of drones representing a drone swarm;

[0032] Steps 3-4: Arrange a vector of nodes Treat it as a single chromosome and initialize an initial population with a fixed number of chromosomes;

[0033] Steps 3-5: In the initial population, record the individual with the highest fitness value;

[0034] Steps 3-6: Randomly select individuals from the initial population to perform crossover and mutation operations to form a new population;

[0035] Step 3-7: Repeat steps 3-4 to 3-6;

[0036] Steps 3-8: Reach the specified number of iterations to obtain the optimal individual, i.e., the optimal node arrangement;

[0037] Steps 3-9: Determine the cut edge set based on the adjacency matrix corresponding to the optimal node arrangement, disconnect the cut edges, and complete the clustering;

[0038] Step 3-10: In each sub-cluster after grouping, sort the nodes according to their degree and select the node with the highest degree in each cluster as the new leader of each cluster.

[0039] Preferably, the flight status includes the speed, position, and acceleration information of the UAV.

[0040] The beneficial effects of this invention are as follows:

[0041] 1. This invention addresses the problem of merging multiple drone clusters by introducing two variables, cluster number and remaining active count, for each drone. This allows drones from different clusters to merge in an orderly manner according to the requirements of the task. It achieves a merging mode where the leader of a cluster merges first, and the followers become leaderless before integrating into the new cluster. This ensures that different drone clusters can quickly reach consistency with the new leader's state after merging, and the collaborative working state of the drone clusters will not be significantly affected.

[0042] 2. To address the problem of dividing a drone swarm into multiple sub-clusters, a genetic algorithm is introduced to process the adjacency matrix of the drone swarm and find the node partitioning sequence with the fewest cut edges. By disconnecting the drones at the two ends of the cut edges, the swarm can be divided into several sub-clusters as required by the task. Then, the key nodes are determined based on the degree centrality index of the drones in each cluster, and they are selected as the new leader drones of each cluster, ensuring that the sub-clusters still have the ability to cooperate in accordance with external instructions after the division. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the drone swarm merging process of the present invention, wherein... Figure 1 (a) indicates the period before cluster merging. Figure 1 (b) indicates that the leader completes the merger first. Figure 1 (c) indicates that the two clusters have been merged.

[0044] Figure 2 This is a schematic diagram of the drone cluster merging method of the present invention.

[0045] Figure 3 This is a schematic diagram illustrating the process of rearranging node sequences to find the optimal UAV swarm grouping in this invention. Figure 3 (a) represents the adjacency matrix corresponding to the initial topology and initial node sequence of the UAV swarm. Figure 3(b) represents the topology of the drone swarm after grouping and the adjacency matrix corresponding to the rearranged node sequence.

[0046] Figure 4 This is a schematic diagram of the drone swarming algorithm of the present invention. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] This invention aims to propose a method for merging and splitting large-scale UAV swarms, enabling the swarm size to be flexibly adjusted according to different mission requirements to meet the needs of takeoff, landing, and other special tasks. Furthermore, it utilizes a suitable method to determine the new leader during the merging and splitting process, allowing the swarm to quickly achieve convergence in a cooperative state after merging and splitting, and continue to perform cooperative tasks under external command control.

[0049] To achieve the above-mentioned objectives, the technical solution proposed by this invention is as follows:

[0050] First, let's introduce the relevant concepts:

[0051] Typically, large-scale drone swarms consist of identical small drones that can communicate with each other. Therefore, the topology of a drone swarm can be represented by an undirected graph G = (V, E), where V = {v1, v2, ..., v...}. n} represents the set of all drones in the cluster. This represents a set of connections between drones. Let a represent the adjacency matrix associated with an undirected graph G, where a ij When a is 1, it indicates that drone i and drone j can communicate with each other; when a ij A value of 0 indicates that drone j is outside the communication range of drone i. The state of each drone in the cluster can be represented as:

[0052]

[0053] Where x n (t) represents various values ​​related to flight state, which can be scalar or vector information such as velocity, position, and acceleration. If the states of all drones in a swarm converge to the state of the leader drone, then the drone swarm can be considered to have achieved a cooperative state with external control, i.e., satisfying the following equation:

[0054]

[0055] Where, x L (t) represents the status information of the leader drone, which is directly determined by the control center.

[0056] 1. Cluster merging methods, such as Figure 2 As shown;

[0057] Each cluster has a leader drone directly controlled by a ground base station. Therefore, the ground base station can select several suitable drone clusters to merge and execute collaborative tasks based on different mission requirements. Since the merged cluster still needs to execute tasks under external command, the cluster merging process can be summarized as follows: The ground base station selects a suitable cluster based on mission requirements and issues a merge command to the leader drone of the cluster. Upon receiving the command, the leaders of different clusters first achieve a collaborative state within their respective clusters and then converge towards a designated area. After the clusters converge, the ground control center selects one of these leader drones as the new leader of the merged cluster based on evaluation criteria such as mission requirements, remaining flight time, and the number of drones in the current cluster. After the new leader is selected, other clusters join the cluster containing the new leader, thus merging multiple clusters.

[0058] However, during the process of two drone swarms executing a merge command, only the two leader drones receive the command message and guide the two swarms closer to each other until they reach a position where they can communicate, and then merge into a new drone swarm. Throughout this process, apart from the leader drone, the other drones in the swarm are unaware of any information about the merge; the drones in the swarm achieve state convergence by maintaining consistency with the leader's state. When the two swarms begin to merge, a drone in one swarm can receive state information from both the other swarm and the leader of its current swarm. The drone itself does not know which side the merged swarm will follow, and therefore does not know which state to choose as its target state. This causes chaos in the swarm merge process and affects the stability of the swarm system.

[0059] To address the aforementioned issues, a variable `g_num` is introduced for each drone to record its cluster number, which is sent to neighboring drones along with its status information during each iteration. Non-leader drones within each cluster can only receive status information from drones with the same `g_num`, ensuring that each drone receives messages only from the same cluster. Simultaneously, a variable `HB` is introduced to record the remaining active count of each drone after leaving its cluster. Each drone's `HB` is typically initialized to 3, indicating that even after losing messages from its original cluster leader, the drone can still maintain its original cluster number `g_num` for three iterations. At the end of each iteration, if the drone receives information from its parent node, `HB` is set to 3; if it loses parent node information, `HB` is decremented by 1, until `HB` reaches 0. This indicates that the drone has exceeded three iterations in its inability to update the original cluster's spanning tree, meaning the original cluster leader has been lost or the drone has left the original cluster. In this state, the drone is in a cluster-free state and can receive messages from other clusters.

[0060] Suppose two clusters, cluster numbers m and n, need to be merged, and the leader of the merged cluster is the original leader of cluster n; that is, the process of cluster m joining cluster n needs to be implemented. Both cluster leaders are directly controlled by a ground center and are therefore aware of the merge information. Therefore, when the leader drone of cluster m... m Once a message is received from cluster n, regardless of whether the message is from the leader of cluster n... n Or other follower drones within cluster n, Leader m All drones execute the merge command, changing their leader status to become followers of cluster n, and adding to the spanning tree maintained by cluster n, using the drone that received the message as their parent node. At this point, drones in cluster m become leaderless. After three iterations, the spanning tree within the cluster still cannot be updated, HB drops to 0, and drones in cluster m set g_num to 0, indicating a no-cluster state, and can receive messages from any cluster. Subsequently, drones originally belonging to cluster m receive messages from cluster n, update and add to the spanning tree of cluster n, completing the cluster merge process. The entire merge process ensures the orderly progress of the cluster merge by having the leader join first, the followers becoming no-cluster, and then joining the new cluster. This does not affect the state of the joined clusters, and the state of the newly added drones can quickly become consistent with the new cluster, achieving stable convergence of the cooperative state throughout the merge process.

[0061] 2. Clustering methods, such as Figure 4 As shown;

[0062] Suppose the task requires the number of sub-clusters to be n. The problem that needs to be solved is how to quickly divide the existing large-scale drone swarm into n sub-clusters based on its topology, quickly determine the new leader in the new cluster, and ensure that the entire process does not have too high overhead, so that the clusters can still maintain a state of collaborative work after the division.

[0063] To address the problem of cluster partitioning, the concept of a cut edge set is introduced, which is the set of lines connecting two nodes that are not in the same subset after the graph has been partitioned. It is defined as follows:

[0064] EC={e s,t |s∈V i ,t∈V j ,1≤i,j≤n,i≠j} (2)

[0065] Since the topology of a drone swarm is an unweighted and undirected graph, partitioning a drone swarm can be seen as dividing a connected undirected graph into multiple disconnected topologies. Therefore, in order to make the partitioning as simple and fast as possible, we need to find a partitioning method that minimizes the number of cut edges in the partitioned topology, i.e., min(EC).

[0066] Genetic Algorithm (GA) is an efficient and practical heuristic algorithm designed based on the evolutionary patterns of organisms in nature. It is a method for searching for optimal solutions by simulating the natural evolutionary process and is suitable for parallel processing of large-scale optimization problems. Solving the optimal sequence problem using a genetic algorithm requires designing four components: chromosome representation, fitness function, crossover operator, and mutation operator.

[0067] (1) Chromosome design: In the sequence optimization problem, the vector of node arrangement. This can be represented as a chromosome; the vector is a permutation of node numbers and also the final solution set. Furthermore, to prevent a sub-cluster from being empty after partitioning, a vector is used... This represents the number of nodes in each of the k sub-clusters.

[0068] (2) Fitness Function: Based on the previous analysis, the adjacency matrix corresponding to the optimal node arrangement has the fewest cut edges. Since the total number of edges in a topological graph is fixed, the fitness function can be written as:

[0069]

[0070] The optimal partitioning sequence is the sequence of nodes that maximizes the fitness function.

[0071] (3) Crossover operator: The crossover operator requires two parent chromosomes to perform structural recombination. It merges the features of the two parent chromosomes to generate a new daughter chromosome. The recombination method is as follows: The main body of one parent chromosome is retained, and the remaining gaps are filled with a non-repeating portion from the other parent chromosome. See the example below:

[0072]

[0073]

[0074]

[0075]

[0076] First, randomly select two intersection points represented by "|" and cut off... The elements between the two intersections and put these elements into The same position. Then, starting from the intersection position, if the child element does not yet have the same element, take the same position. The element value and insert it The position of the first zero element after the intersection point. If the value retrieval process reaches... The tail or interpolation process reaches To find the tail of the vector, we start traversing from the head of the vector.

[0077] (4) Mutation Operator: The mutation operator can increase the diversity of chromosomes and introduce new information into the population. Here, mutation is achieved by introducing a swap operation. For example, two elements 3 and 7 in the parent generation are randomly selected as mutation points, and then these two elements are swapped, as shown below, thus completing the mutation.

[0078]

[0079]

[0080] After completing the above operations, the node sequence is iterated a sufficient number of times to obtain the optimal node partitioning sequence. Using a genetic algorithm, it is no longer necessary to solve the node sorting problem using a brute-force algorithm. Instead, multiple possible sequences are treated as individuals in a generation of the population. The population is then updated by performing crossover and mutation operations, and this process is repeated, while recording the individual with the highest fitness in each generation. When the iterations are repeated a certain number of times, the optimal individual, i.e., the optimal cluster partitioning sequence, can be obtained.

[0081] After finding the optimal cluster partitioning sequence, communication between the drones at both ends of the cut edge is disconnected, thus dividing a large-scale drone swarm into multiple sub-clusters. According to the mission requirements, the sub-clusters still need to possess the ability to perform collaborative tasks; that is, each new cluster still needs a leader to receive external commands and control the new cluster to complete collaborative work. Therefore, it is necessary to find suitable drones within the partitioned clusters to act as new leaders.

[0082] In a cluster network, certain key nodes play a crucial role in the overall network. These key nodes can sometimes determine the network's structure and function, significantly improving its robustness. However, if these nodes are attacked and fail, it will have a huge impact on the entire network. Degree centrality is a classic metric for measuring node importance; it selects key nodes in the network by measuring the node's degree. The higher the degree of a drone, the more drones it interacts with, indicating a more important position. Therefore, degree centrality can be used to find key nodes in the cluster and select them as leader drones.

[0083] After a large-scale drone swarm completes its cluster division and determines new leaders for each sub-cluster, the sub-clusters can quickly achieve a coordinated state and then execute new tasks. The entire process is rapid and unaffected by each other, ensuring that the large-scale drone swarm can adapt to different mission requirements and has the ability to perform collaborative tasks. Specific implementation examples:

[0085] like Figure 1 As shown, suppose cluster m and cluster n need to be merged. And according to the control center's decision, the leader of the new merged cluster is the leader of cluster n, meaning cluster m needs to be integrated into cluster n. The process of merging clusters using a cluster merge algorithm is described below:

[0086] 1) For drones in cluster m, maintain cluster number g_num as m; for drones in cluster n, maintain cluster number g_num as n; and drones only receive information from drones with the same g_num as themselves.

[0087] 2) When the leaders of the two clusters receive a merge command, the two clusters begin to approach each other;

[0088] 3) When a follower drone in cluster m receives a message from a drone in cluster n, the message is ignored because the cluster numbers are inconsistent;

[0089] 4) When the leader drone in cluster m receives a message from a drone in cluster n, it changes its identity to follower, changes g_num to n, and joins cluster n with the drone that sent the message as its parent node.

[0090] 5) In each iteration, if a follower in cluster m does not receive the leader's status information, the remaining active count is reduced by 1.

[0091] 6) When the remaining active count of drones in cluster m drops to 0, change the cluster number g_num to 0 and change itself to leaderless mode;

[0092] 7) When a drone in leaderless mode receives a message from cluster n, it can change its identity to follower, change its cluster number to n, and join cluster n with the drone that sent the message as its parent node.

[0093] 8) Once all leaderless drones have joined cluster n, the merging process from cluster m to cluster n is complete.

[0094] Due to the characteristics of adjacency matrices, the adjacency matrix changes when the node order changes. By adjusting the node sequence of the adjacency matrix, it can be transformed into an approximate diagonal matrix, where each diagonal block represents a newly partitioned sub-cluster. Edges outside the diagonal blocks represent cut edges between different sub-clusters. Disconnecting these cut edges completes the clustering process; therefore, fewer cut edges mean lower energy consumption for partitioning the clusters. Thus, it is necessary to find the optimal node sequence that results in a diagonal adjacency matrix with the fewest cut edges. For example... Figure 3 As shown, when the node sequence is sorted as {1,2,8,4,5,7,9,3,6,10}, the adjacency matrix has the fewest cut edges, and the nodes are divided into three sub-clusters: {1,2,8}, {4,5,7,9}, and {3,6,10}. This partitioning is the optimal solution. The specific description of the cluster partitioning algorithm is as follows:

[0095] 1) Based on the requirements of the task, determine the number of sub-clusters n and the limit on the number of drones in each cluster;

[0096] 2) Represent the arrangement of nodes corresponding to the adjacency matrix as a vector. By transforming the adjacency matrix using this vector, a brute-force search can be used to find a permutation of the adjacency matrix that is closest to the block diagonal matrix. This is then determined by the adjacency matrix and the node arrangement. The cut edge set can be determined. However, brute-force search has too high a time complexity, so we consider using a genetic algorithm to solve for the optimal node arrangement;

[0097] 3) Determine the fitness function that minimizes the number of cut edges.

[0098] 4) A vector that arranges a node Treat it as a single chromosome and initialize an initial population with a fixed number of chromosomes;

[0099] 5) In the initial population, record the individual with the highest fitness value;

[0100] 6) Randomly select individuals from the initial population to perform crossover and mutation operations to form a new population;

[0101] 7) Repeat steps 4) through 6);

[0102] 8) After reaching the specified number of iterations, the optimal individual, i.e., the optimal node arrangement, is obtained;

[0103] 9) Determine the cut edge set based on the adjacency matrix corresponding to the optimal node arrangement, disconnect the cut edges, and complete the clustering;

[0104] 10) In each sub-cluster after the splitting, sort the nodes according to their degree and select the node with the highest degree in each cluster as the new leader of each cluster.

Claims

1. A method for merging and grouping unmanned aerial vehicle (UAV) swarms, characterized in that, Includes the following steps: Step 1: Construct a drone swarm model; A drone swarm consists of identical drones that can communicate with each other. Therefore, the topology of a drone swarm can be represented by an undirected graph G = (V, E), where V = {v1, v2, ..., v...}. n } represents the set of all drones in the cluster. This represents a set of connections between drones; let Let a represent the adjacency matrix associated with an undirected graph G, where a ij When a is 1, it indicates that drone i and drone j can communicate with each other; when a ij When the value is 0, it indicates that drone j is outside the communication range of drone i; each drone swarm consists of a leader and several follower drones; The state of each drone in the cluster is represented as follows: Where x n (t) represents various values ​​related to flight status; If the states of all drones in a cluster converge to the state of the leader drone, then the drone cluster is considered to have achieved a cooperative state with external control, i.e., it satisfies the following equation: Where, x L (t) represents the status information of the leader drone, which is determined by the control center; Step 2: Merge the drone swarm; Suppose that drone clusters m and n need to be merged, and according to the control center's decision, the leader of the new merged cluster is the leader of cluster n, meaning cluster m needs to be integrated into cluster n. The specific steps are as follows: Step 2-1: For drones in cluster m, maintain cluster number g_num as m; for drones in cluster n, maintain cluster number g_num as n; and follow drones only receive information from drones with the same g_num as themselves; set the initial number of active times for drones; Step 2-2: When the leaders of clusters m and n receive the merge command, the two clusters begin to approach each other; Steps 2-3: When the follower drone in cluster m receives a message from the drone in cluster n, the message is ignored because the cluster numbers are inconsistent. Steps 2-4: When the leader in cluster m receives a merge message from a drone in cluster n, it changes its identity to a follower, changes g_num to n, and joins cluster n with the drone from which the message originated as its parent node. Steps 2-5: In each iteration, if a follower in cluster m does not receive the leader's status information, the active count is decremented by 1. Steps 2-6: When the number of active drones in cluster m drops to 0, change the cluster number g_num to 0 and change itself to leaderless mode; Steps 2-7: When a drone in leaderless mode receives a message from cluster n, it changes its identity to follower drone, changes the cluster number g_num to n, and joins cluster n with the drone that sent the message as its parent node. Steps 2-8: Once all leaderless drones have joined cluster n, the merging process from cluster m to cluster n is complete. Step 3: Drone swarming; Step 3-1: Based on the task requirements, determine the number of sub-clusters k needed and the limit on the number of drones in each cluster; Step 3-2: Represent the arrangement of nodes corresponding to the adjacency matrix as a vector. The adjacency matrix is ​​transformed by transforming the vector, and the optimal node arrangement is solved using a genetic algorithm. Step 3-3: Determine the fitness function that minimizes the number of cut edges. EC is the cut edge set, which is the set of lines connecting nodes in an undirected graph G that are not in the same subset after partitioning. It is defined as follows: EC={e s,t |s∈V i ,t∈V j ,1≤i,j≤k,i≠j} Among them, e s,t This represents the edge connecting drones located in two different clusters, where s and t represent drones located in different drone clusters, respectively. j A collection of drones representing a drone swarm; Steps 3-4: Arrange a vector of nodes Treat it as a single chromosome and initialize an initial population with a fixed number of chromosomes; Steps 3-5: In the initial population, record the individual with the highest fitness value; Steps 3-6: Randomly select individuals from the initial population to perform crossover and mutation operations to form a new population; Step 3-7: Repeat steps 3-4 to 3-6; Steps 3-8: Reach the specified number of iterations to obtain the optimal individual, i.e., the optimal node arrangement; Steps 3-9: Determine the cut edge set based on the adjacency matrix corresponding to the optimal node arrangement, disconnect the cut edges, and complete the clustering; Step 3-10: In each sub-cluster after grouping, sort the nodes according to their degree and select the node with the highest degree in each cluster as the new leader of each cluster.

2. The method for merging and grouping unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, The flight status includes the drone's speed, position, and acceleration information.

Images