An unmanned cluster clustering method based on r-hop control set

Abstract

The present application belongs to the technical field of unmanned cluster cooperative control, and relates to an unmanned cluster clustering method and system based on r-hop control set. In the first stage, local leaders are constructed based on node degree and directed acyclic graph (DAG), and initial cluster heads are identified by combining node influence index and inflection point detection. In the second stage, the boundary is constrained by r-hop communication, the cluster head set is updated, and the next level cluster heads are selected to complement the uncovered area connected components, and the multi-level hierarchical cluster head structure is iteratively generated. The present application is suitable for the engineering application scenarios of actual communication multi-hop forwarding of unmanned cluster and limited effective communication range, avoids the problems of too large single cluster and load imbalance of traditional minimum control set, establishes a cluster head hierarchical mapping mechanism, and realizes that all nodes meet the two-hop reachable cluster head constraint, forms a hierarchical and scalable unmanned cluster backbone topology, and facilitates the implementation of hierarchical task scheduling, distributed negotiation and dynamic replacement of fault nodes of unmanned cluster.

Description

Technical Field

[0001] This invention belongs to the field of unmanned cluster collaborative control and wireless self-organizing network clustering management technology, and relates to an unmanned clustering method and system based on r-hop control set. Background Technology

[0002] Clustered management is a core technology for large-scale networking of unmanned clusters. By dividing a large number of unmanned nodes in the cluster into multiple independent management units (clusters), and electing cluster head nodes to assume the functions of managing members within the cluster, forwarding data between clusters, and coordinating task scheduling, it can effectively reduce network routing overhead, improve topology resilience, optimize node energy allocation, and adapt to the high dynamic mobility characteristics of the cluster. It is the mainstream technology paradigm for distributed management of existing unmanned clusters.

[0003] Early unmanned clustering algorithms commonly used in the industry were represented by the Lowest ID Algorithm (LOWID), the Highest Node Degree Algorithm (HIGHD), and the Weight-Based Clustering Algorithm (WCA). These algorithms have simple models and low computational overhead, and complete cluster head election and cluster partitioning only through a single-dimensional fixed index. They are difficult to adapt to unmanned cluster scenarios with complex working conditions, highly dynamic topologies, and multi-task constraints. They have inherent defects in clustering stability, scenario adaptability, and control capabilities, and are only suitable for the basic networking needs of low-dynamic, small-scale unmanned clusters.

[0004] To address the inherent characteristics of unmanned swarms, such as high mobility, intermittent links, time-varying topology, and limited node energy, existing research has gradually formed a standardized algorithm system and optimization evolution direction. Based on the cluster head election mechanism, current mainstream unmanned swarm clustering algorithms can be divided into two main categories: probabilistic clustering algorithms and deterministic clustering algorithms. Probabilistic clustering algorithm: Based on a stochastic model and probabilistic mechanism, cluster head election is completed. The core optimization objectives are to balance node energy consumption, extend the overall network life cycle, and distribute network load. It is suitable for scenarios with high cluster dynamics and moderate requirements for cluster head stability. Deterministic clustering algorithm: Based on deterministic quantitative indicators such as node remaining energy, communication connectivity, link effective duration, spatial location, and movement trend, it completes cluster head selection and cluster structure division. The core optimization goal is to ensure routing reliability, reduce end-to-end transmission latency, and improve network stability. It is more suitable for application scenarios with high mobility, latency sensitivity, and high task reliability requirements.

[0005] Subsequent existing technologies have further focused on the optimization goal of clustering, forming four major research directions: clustering algorithms based on network energy consumption optimization, clustering algorithms based on cluster mobility adaptation, clustering algorithms based on energy consumption and mobility co-optimization, and clustering algorithms based on spatial positioning assistance. Existing technologies all focus on the above directions, optimizing and iterating on one or more performance indicators such as clustering speed, network lifespan, energy consumption, transmission latency, and clustering stability.

[0006] Based on the clustering implementation logic, existing mature clustering algorithms can be uniformly divided into two basic paradigms: The first is top-down splitting clustering, which takes the entire cluster as the initial single cluster and completes the constraints on the number and size of clusters through hierarchical splitting. Typical examples are K-means clustering, spectral clustering and their extended algorithms. The second is bottom-up merging clustering, which takes a single unmanned node as the initial independent cluster and completes the cluster structure construction by gradually aggregating adjacent nodes. Typical examples are graph clustering algorithms, graph attention-based aggregation clustering algorithms, and multi-hop control set cluster head selection algorithms.

[0007] The core design of existing clustering algorithms revolves around the physical attributes of unmanned nodes (remaining energy, spatial location) and local communication topology characteristics (connectivity, link quality), without incorporating core requirements such as unmanned cluster collaborative operation relationships, actual task types, and task loads. This results in a complete disconnect between clustering logic and actual operational applications, and the clustering results cannot match task-oriented collaborative requirements, leading to low cluster operation efficiency. This is one of the core defects of existing technologies.

[0008] Existing clustering algorithms (split and merge paradigms) can only generate fixed, single-level cluster structures. They lack dynamic adjustment mechanisms and cannot scale, split, or merge cluster structures according to changes in unmanned cluster task types, task load fluctuations, or real-time communication topology changes. They also lack multi-scale community adaptation capabilities with hierarchical nesting and the ability to split and merge, making them unable to cope with complex and ever-changing task scenarios. Their cluster collaboration flexibility and scenario adaptability are extremely poor.

[0009] Existing r-hop multi-hop control set solving algorithms based on communication topology graphs are independent of and not deeply coupled with multi-scale hierarchical clustering logic. Furthermore, the cluster head election and cluster merging processes in the closest solution are disconnected, which cannot provide stable control support for dynamically scalable multi-scale cluster structures. This results in a lack of coordination between cluster head selection, hierarchical control, and clustering strategies, and significant shortcomings in hierarchical control capabilities and collaborative scheduling capabilities, failing to meet the hierarchical and dynamic collaborative operation control requirements of unmanned swarms. Summary of the Invention

[0010] The purpose of this invention is to solve the problems in the prior art and provide an unmanned clustering method and system based on r-hop control sets.

[0011] To achieve the above objectives, the present invention employs the following technical solution: A clustering method for unmanned clusters based on r-hop control sets includes the following steps: Abstract the unmanned cluster into an undirected communication topology graph; Calculate the degree of each node in the undirected communication topology graph, construct a directed acyclic graph based on the degree of each node, and mark the node with an out-degree of zero as a local leader based on the directed acyclic graph. For a node marked as a local leader, perform a breadth-first search in the undirected communication topology graph to determine the upper-level leader of the local leader node. Let the number of hops of the shortest path from the current local leader node to the upper-level leader be the level depth. The comprehensive influence index is calculated based on the degree and level depth of each node. First-level cluster heads are selected based on the comprehensive influence index to generate an initial cluster head set. Based on the initial set of cluster heads, perform a breadth-first search with r hops for each cluster head, and update all covered node cluster heads and the shortest distance from a node to a cluster head according to the search results; The process involves counting the set of nodes not covered by cluster heads. If the set of uncovered nodes is empty, the recursion terminates, and the current set of cluster heads and node affiliation information are output, thus completing the clustering. If the set of uncovered nodes is not empty, the set of cluster heads is expanded, the next level cluster head is added, and the r-hop breadth-first search is re-executed until all nodes are covered. At this point, the current set of cluster heads and node affiliation information are output, thus completing the clustering.

[0012] A further improvement of the present invention is that: The process involves calculating the degree of each node in the undirected communication topology graph, constructing a directed acyclic graph (DAG) based on the degree of each node, and marking nodes with an out-degree of zero as local leaders, including: Calculate the degree of each node based on the undirected communication topology graph; Node The directed edge has the maximum pointing degree and is greater than the node. Neighboring nodes of self-degree Construct a directed acyclic graph (DAG) that satisfies and ,in For nodes The set of neighbors; When node All neighboring nodes have a degree less than or equal to that of node 1. The degree of the node itself, then With no outgoing edges, the current node is identified as a local leader, and a set of local leaders is generated. ; If a node has multiple outgoing edges, one outgoing edge is randomly retained.

[0013] The process of determining the upper-level leader of the local leader node includes: Nodes marked as local leaders Using the source node, perform a breadth-first search in the undirected communication topology graph, where the first encountered node has a degree greater than or equal to the current node. The node is defined as having its own degree and not being identified by other local leaders that have not been searched. The upper-level leader records the current node. The shortest path hop count to the upper-level leader in an undirected communication topology graph is used as the hierarchical position parameter of the source node.

[0014] If there are multiple other local leaders that meet the above conditions in the breadth-first search, the local leader with the highest degree is selected as the superior leader. If there are multiple local leaders with the highest degree, one of them is selected as the superior leader. For the local leader with the highest node degree in the undirected communication topology graph, its hierarchical position parameter is directly set to a preset value or infinity, and breadth-first search is not performed.

[0015] The process involves calculating a comprehensive influence index based on the degree and hierarchical depth of each node, selecting first-level cluster heads based on this comprehensive influence index, and generating an initial cluster head set, including: Degree of each node With path length Multiply to obtain quantitative evaluation indicators. ; Press all nodes Sort the values ​​in descending order to obtain the node sequence. ; Based on node degree With path length Plot a scatter plot, perform inflection point detection on the scatter plot, and select nodes before the inflection point by combining the node sequence. Each node constitutes the initial central set. .

[0016] The process involves performing an r-hop breadth-first search on each cluster head based on the initial cluster head set, and updating all covered node cluster heads and the shortest distance from a node to a cluster head based on the search results, including: Initialize the current initial center set ; For the current set For each cluster head in the undirected communication topology graph, a breadth-first search with r hops is performed to obtain the set of nodes covered by the cluster head, where r represents the preset maximum number of communication hops; If for each node If through a certain cluster head The distance obtained Less than its current shortest path hop count Then update its cluster head. for Shortest path hop count Distance .

[0017] If the set of uncovered nodes is not empty, perform the following steps: Calculate the maximum level in the current cluster head set. The newly extended cluster head level is set to ; In the set of uncovered nodes Induced subgraph Determine each Select the node with the highest degree as the new cluster head. ; New cluster head Add to cluster head set, update cluster head set Set its hierarchy Initialize all newly added cluster heads and re-execute the recursive process until an uncovered set of nodes is reached. If empty, the final hierarchical cluster head structure is obtained.

[0018] An unmanned clustering system based on r-hop control set includes: The initial model building module is used to abstract the unmanned cluster into an undirected communication topology graph; The local leader identification module is used to calculate the degree of each node in the undirected communication topology graph, construct a directed acyclic graph based on the degree of each node, and mark the node with an out-degree of zero as a local leader based on the directed acyclic graph. The upper-level leader identification module is used to perform a breadth-first search in the undirected communication topology graph for nodes marked as local leaders to determine the upper-level leader of the local leader node. Let the number of hops of the shortest path from the current local leader node to the upper-level leader be the level depth. The initial cluster head identification module is used to calculate the comprehensive influence index based on the degree and hierarchical depth of each node, and to select the first-level cluster heads based on the comprehensive influence index to generate the initial cluster head set. The two-hop coverage search module is used to perform an r-hop breadth-first search on each cluster head based on the initial cluster head set, and update all covered node cluster heads and the shortest distance from a node to a cluster head according to the search results; The uncovered node identification module is used to count the set of nodes not covered by the cluster head. If the set of uncovered nodes is empty, the recursion terminates, and the current set of cluster heads and node affiliation information are output, completing the clustering. If the set of uncovered nodes is not empty, the set of cluster heads is expanded, the next level cluster head is added, and the r-hop breadth-first search is re-executed until all nodes are covered. Then, the current set of cluster heads and node affiliation information are output, completing the clustering.

[0019] A computer program product includes a computer program that, when executed by a processor, implements any one of the methods described.

[0020] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of any of the methods described above.

[0021] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described herein.

[0022] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a clustering method for unmanned swarms based on r-hop control sets. It transforms the unmanned swarm clustering problem into a two-hop connected control set problem, aligning with the practical constraints of two-hop communication in engineering. A two-stage hierarchical clustering architecture is proposed. The first stage constructs local leaders based on node degree and a directed acyclic graph (DAG), combining node influence indicators and inflection point detection to identify initial cluster heads. The second stage uses r-hop communication constraints as boundaries, updating the cluster head set and supplementing lower-level cluster heads with connected components in uncovered regions, iteratively generating a multi-level hierarchical cluster head structure, consistent with the realities of unmanned swarms. In engineering applications where communication involves multiple hops and limited effective communication range, this approach avoids the problems of excessively large single clusters and load imbalance in traditional minimum control sets. It also overcomes the current situation where the solution of r-hop control sets and clustering logic are independent of each other. By establishing a cluster head hierarchical mapping mechanism, high-level cluster heads provide overall coverage, and low-level cluster heads fill in edge blind spots, ensuring that all nodes meet the two-hop reachable cluster head constraints. This forms a hierarchical and scalable unmanned cluster backbone topology, which facilitates hierarchical task scheduling, distributed negotiation, and dynamic replacement of faulty nodes in unmanned clusters. Its scalability is far superior to traditional clustering methods without hierarchy. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a modeling diagram of the unmanned cluster communication relationship in this invention; Figure 2 This is a connectivity control set diagram of the present invention; Figure 3 In the first stage of this invention, an initial hierarchical structure diagram of the unmanned cluster is constructed (where a represents the initial communication topology diagram of the unmanned cluster, b represents the calculated node degree; c represents the construction of a directed acyclic graph and identification of local leaders; d represents the determination of upper-level leaders based on local breadth-first search; and e represents the initial cluster head identification). Figure 4 This is the unmanned cluster communication topology under the random unit disk diagram of the present invention; Figure 5 Based on the present invention Cluster head identification decision graph; Figure 6 This is a diagram showing the hierarchical clustering results on the unit disk diagram of the present invention; Figure 7a The clustering results of the present invention are obtained by running three algorithms on a communication topology graph with 32 nodes. Figure 7b The clustering results of the present invention are obtained by running three algorithms on a communication topology graph with 109 nodes. Figure 7c The present invention provides clustering results of three algorithms running on a communication topology graph with 270 nodes. Figure 8 The graph shows the variation of the running time of the three algorithms disclosed in this invention with the node size. Figure 9 This is a graph showing the variation of the runtime of the THLSC disclosed in this invention with node size. Figure 10 The number of cluster heads in the three algorithms of this invention varies with the node size. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0027] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0030] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0031] The present invention will now be described in further detail with reference to the accompanying drawings: Traditional ILP integer programming methods for solving 2-hop control sets are NP-hard problems. As the size of unmanned cluster nodes increases, the model variables and constraints grow exponentially, leading to a sharp increase in solution time. In large-scale scenarios, it is difficult to solve in real time, and only small-scale scenarios can obtain approximate optimal solutions. Traditional SA simulated annealing heuristic algorithms rely on large-scale random iterative search. The number of iterations increases dramatically with the network size, resulting in long running times. Furthermore, there is no hierarchical cluster head design, leading to high cluster head redundancy and no hierarchical collaborative structure, which cannot adapt to the hierarchical management and dynamic reconfiguration requirements of unmanned clusters. Conventional clustering methods only pursue the minimum control set, which can easily lead to excessively large single cluster sizes and concentrated cluster head loads. They also lack the adaptability of hierarchical partitioning and two-hop communication constraints.

[0032] In view of this, addressing the technical problems of existing unmanned swarm clustering algorithms, such as the disconnect between clustering logic and operational requirements, the inability to dynamically adjust cluster structure, and the disconnect between multi-hop control sets and clustering logic, this paper discloses an unmanned swarm clustering method based on r-hop control sets. Combining the practical needs of unmanned swarm collaborative operations, highly dynamic topology adaptation, and multi-scale control, the method first constructs a Directed Acyclic Graph (DAG) based on node degree and selects local leaders, quantifies node influence using topological features, and adaptively determines the initial cluster heads. Then, a strict two-hop communication constraint is introduced, and Breadth-First Search (BFS) is used to complete cluster member affiliation coverage. For uncovered areas, lower-level cluster heads are automatically added according to connected components, forming a multi-level hierarchical structure. The entire process does not require global integer programming solutions or extensive random iterative searches; it only relies on local topology information and a finite number of BFS traversals to complete clustering and networking, achieving hierarchical, dynamic, and task-adaptive clustering of unmanned swarms, providing a reliable communication topology foundation for subsequent distributed collaborative task allocation.

[0033] Therefore, this embodiment of the invention is mainly used to achieve the following objectives: First, it overcomes the shortcomings of existing clustering algorithms that only focus on node physical attributes and local topology and are detached from the needs of collaborative operations. It provides a clustering design scheme that is guided by collaborative operation relationships and actual task requirements, realizes the deep integration of clustering logic and operation application, and improves the adaptability of clustering results to the collaborative needs of task-oriented clusters.

[0034] Second, to address the problem that existing clustering algorithms have fixed cluster structures and cannot be dynamically adjusted, a multi-scale clustering strategy with hierarchical nesting and the ability to split and merge is designed. This enables the cluster structure to dynamically scale according to task type, task load, and communication topology, thereby improving the flexibility of unmanned cluster collaboration and its adaptability to complex scenarios.

[0035] Third, it breaks the current disconnect between r-hop multi-hop control set solution and multi-scale clustering logic, achieves deep coupling between the two, optimizes cluster head selection and hierarchical control mechanism, provides stable control support for dynamic multi-scale cluster structure, meets the hierarchical and dynamic collaborative operation control needs of unmanned clusters, and improves the efficiency and stability of cluster collaborative control.

[0036] Furthermore, key aspects of the embodiments of the present invention include: (1) Key points of topology modeling: The unit disk graph UDG is used to abstractly model the communication topology of the unmanned cluster. The maximum communication distance of the unmanned equipment is used as the edge constraint, and the clustering problem of the unmanned cluster is transformed into a two-hop connected control set solution problem, which fits the actual two-hop communication constraints in the project.

[0037] (2) Key points of the algorithm architecture: A two-stage hierarchical clustering architecture is proposed. In the first stage, a local leader is constructed based on node degree and DAG, and the initial cluster head is identified by combining node influence index and inflection point detection. In the second stage, the lower-level cluster head is selected by BFS traversal and uncovered region connected components, and a multi-level hierarchical cluster head structure is generated iteratively, with 2-hop communication constraints as the boundary.

[0038] (3) Key points for cluster head selection: The node topological influence is evaluated by combining node degree and upper-level path length. The initial number of cluster heads is determined adaptively by detecting inflection points in the scatter plot. In the uncovered subgraph, the node with the highest degree is selected as the new cluster head, taking into account both coverage capability and topological centrality.

[0039] (4) Key points of hierarchical networking: Establish a cluster head hierarchical mapping mechanism, with high-level cluster heads covering the overall area and low-level cluster heads filling the edge blind spots, so that all nodes can meet the two-hop reachable cluster head constraints and form a hierarchical and scalable unmanned cluster backbone topology.

[0040] (5) Key performance advantages: The algorithm's time complexity is close to linear. While ensuring that the cluster head size is close to the global optimal lower bound, it has real-time performance and large-scale scalability that are far superior to integer programming (ILP) and simulated annealing (SA).

[0041] Furthermore, the modeling rules in this embodiment include: Phase 1: Communication Topology Modeling (1) Node abstraction: Each unmanned equipment in the unmanned cluster is abstracted as a vertex in the UDG model. Each vertex uniquely corresponds to one unmanned equipment, and the attributes of the vertex are consistent with the physical attributes and topological attributes (neighboring nodes, communication links) of the corresponding unmanned equipment. Among them, physical attributes include location and communication radius, and topological attributes include neighboring nodes and communication links.

[0042] (2) Construction of edges: The radius of the unit disk corresponding to each vertex is set as the maximum effective communication distance of the unmanned equipment corresponding to that vertex; for any two vertices (corresponding to two unmanned equipment), if the actual physical distance between them does not exceed the maximum effective communication distance, it is considered that the two unmanned equipment can establish a stable bidirectional communication link, which is represented in the UDG model as an undirected edge connecting the two vertices; otherwise, if the actual distance exceeds the maximum effective communication distance, it is considered that a communication link cannot be established directly, and there is an undirected edge between the corresponding vertices in the model.

[0043] like Figure 1 As shown, this modeling method can concisely and accurately reflect the actual communication topology of unmanned swarms. It avoids redundant modeling of complex communication interference and provides a clear and quantifiable topological basis for subsequent clustering algorithm design, cluster head selection, and hierarchical management, thus meeting the actual engineering needs of unmanned swarm collaborative task allocation.

[0044] Phase Two: Transformation and Optimization Objectives of the Clustering Problem Based on the above UDG communication topology model, this invention transforms the clustering problem of unmanned swarms into an optimization problem of solving the Connected Dominating Set (CDS), clarifying the correspondence between clustering and the control set: the nodes in the control set are defined as the cluster heads of each cluster, responsible for the management of members within the cluster, data forwarding between clusters, and task coordination; each "control branch" in the connected control set corresponds to an independent cluster unit, and the member nodes within the cluster are all governed and managed by the corresponding cluster head.

[0045] As shown in Figure 2, the black vertices constitute the minimum control set, which can dominate all nodes but does not guarantee the connectivity of the control set itself. Adding gray vertices to this control set yields a connected control set, which satisfies the need to dominate all nodes while ensuring connectivity between cluster heads, thus supporting inter-cluster collaboration. Based on this connected control set, the unmanned swarm consisting of 16 unmanned equipment units can be divided into 7 clusters (where the bolded nodes are the cluster heads). It is important to note that because the "dominance" relationship of the control set over ordinary nodes may overlap—that is, an ordinary node may be dominated by multiple cluster heads—the actual clustering result is not unique. Multiple equivalent partitioning forms are common, and the appropriate form can be flexibly selected based on specific task requirements.

[0046] However, clustering schemes based on ordinary connectivity control sets have inherent drawbacks such as an excessive number of clusters and a small individual cluster size, making it difficult to balance the requirements of reasonable cluster size and communication efficiency in practical applications: an excessive number of clusters increases inter-cluster coordination overhead, while a small individual cluster size leads to wasted cluster head resources. Considering that communication within unmanned clusters typically allows multi-hop forwarding, and that effective communication range is often limited to two hops in engineering implementations, this invention further introduces an r-hop connectivity control set, where r is the maximum allowed number of communication hops. In this embodiment, r=2 is used to better suit the actual scenario of multi-hop communication in unmanned clusters.

[0047] Under the constraints of a 2-hop connectivity control set, communication links of no more than two hops are allowed between cluster members and the cluster head, without requiring direct connections between the cluster head and all cluster members. This effectively reduces the number of cluster heads, expands the coverage of a single cluster, and optimizes cluster structure rationality and communication efficiency while ensuring network connectivity. Figure 2 For example, when using a 2-hop connected control set, an optimized control set can be obtained. Based on this control set, the 16 unmanned equipment can be divided into 4 clusters (specifically, the clusters are divided as follows). Compared with the ordinary connectivity control set partitioning, the total number of nodes remains unchanged, while the number of clusters is significantly reduced, which better meets the actual needs of "moderate multi-hop, reasonable cluster size, and clear backbone structure" in cluster communication, and provides a good foundation for subsequent hierarchical management and control.

[0048] Further analysis reveals that in practical engineering applications, pursuing only the minimum r-hop control set in a strict sense will result in too few cluster heads and too large a single cluster size, which will significantly increase the communication and computing load of the cluster heads. The cluster heads need to undertake a large number of information forwarding and state management tasks for members within the cluster. The concentrated load will further amplify the information propagation delay and conflict resolution overhead, reduce the overall convergence efficiency of cluster collaboration, and fail to meet the real-time requirements of dynamic tasks.

[0049] Therefore, this invention does not take "solving the minimum r-hop control set" as the sole optimization objective. Instead, it uses the r-hop control set (r=2) as a basis to characterize the hierarchical coverage relationship between cluster heads and cluster members. Furthermore, it introduces a cluster head hierarchy mechanism based on the feasible control set, enabling unmanned swarms to adaptively adjust the number of cluster heads according to the task execution scenario. Specifically, multi-level cluster heads are formed by constructing hierarchical membership relationships between cluster heads.

[0050] Based on the above design concept, the core objective of this invention is to design a clustering method that simultaneously meets two types of requirements: first, it can form a clear cluster head-cluster member organizational structure to ensure the orderliness of intra-cluster control and inter-cluster collaboration; second, it can support hierarchical cluster head management and on-demand reconfiguration to adapt to the dynamic requirements of task type, load fluctuation, and topology changes, thereby providing a communication topology foundation that is closer to engineering practice for the subsequent design of distributed collaborative task allocation algorithms.

[0051] Therefore, this embodiment specifically includes the following steps: Consider an unmanned cluster communication topology, which can be abstracted as an undirected graph. ,in, Represents the set of unmanned equipment nodes. This represents the set of effective communication links. Given that unmanned cluster communication networks typically limit the effective communication range to within two hops in engineering implementations, this invention aims to identify hierarchical cluster head structures that satisfy the two-hop constraint on this topology graph. Specific requirements are as follows: (1) Coverage constraint: For any cluster member node, the shortest path length to its cluster head does not exceed 2; (2) Hierarchical constraints: The cluster head nodes are further organized into a hierarchical leadership backbone structure to characterize the hierarchical collaboration characteristics within the unmanned cluster, thereby realizing hierarchical control and on-demand reconfiguration of the cluster heads. Therefore, this invention, based on the Local Search for Community Detection (LS) algorithm proposed by Shi et al., integrates two-hop communication constraints and cluster head hierarchical modeling requirements to propose a clustering method for constructing a distributed architecture of unmanned clusters—the Two-Hop Hierarchical Local Search Clustering Algorithm (THLSC). This algorithm consists of two stages: "constructing the initial hierarchical structure of the unmanned cluster" and "partitioning clusters based on communication constraints." The pseudocode is shown in Algorithm 1 and Algorithm 2, respectively. The two stages are progressive and closely connected. The specific implementation steps are as follows: Step 1 (Phase 1): Constructing the initial hierarchical structure of the unmanned cluster The core objective of this stage is to identify influential local leaders in the unmanned cluster topology, construct a preliminary hierarchical structure, and provide a basis for subsequent initial cluster head selection. This stage consists of four steps: Step 1.1: Calculate node degree First, calculate the undirected communication topology graph. Each node degree Node degree Defined as the total number of neighboring nodes directly connected to a given node, it is a fundamental topological characteristic reflecting a node's topological influence and communication capabilities. A higher node degree indicates more neighboring nodes, wider communication coverage, and greater suitability as a cluster head for management and control tasks. For ease of explanation and verification in subsequent steps, in... Figure 1 Based on this, the node numbers are rearranged to obtain isomorphic nodes. Figure 3 (a) The degree values ​​of each node calculated accordingly are as follows: Figure 3 As shown in (b), this provides a quantitative basis for the subsequent construction of directed acyclic graphs and identification of local leaders.

[0052] Step 1.2: Construct a directed acyclic graph and identify local leaders Traverse each node in a preset order (in this embodiment of the invention, the node numbers are in ascending order). Based on the degree of nodes, a Directed Acyclic Graph (DAG) is constructed, and local leaders are identified according to the following rules: Node The directed edge has the largest pointing degree and is greater than Neighboring nodes of self-degree That is, satisfying and ,in For nodes The set of neighbors. Since each directed edge points from a node with a lower degree to a neighbor node with a higher degree, there is no directed path from a node with a higher degree back to a node with a lower degree. Therefore, no directed cycles are formed, and a directed acyclic graph (DAG) is finally constructed.

[0053] If no node meets the above conditions , i.e., node All neighboring nodes have a degree no greater than The degree of the node itself, then Nodes with no outgoing edges are identified as "local leaders," and the set of all local leaders is denoted as . .

[0054] After traversal, if a node has multiple outgoing edges (i.e., multiple neighboring nodes with the same degree, all of which are maximum), one outgoing edge is randomly retained to ensure the uniqueness of directed edges. Ultimately, the entire DAG will be constructed as a rooted tree forest, with the root node of each tree being the local leader. Besides the local leader, most nodes have established a local hierarchical structure, clearly demonstrating the asymmetric cluster head-cluster member role relationship (the local leader is a potential cluster head, and nodes pointing to it are potential cluster members). Figure 3 As shown in (c), the gray nodes represent local leaders. Step 1.3: Determine the upper-level leader based on local breadth-first search. To clarify the superior leader of each local leader in the global hierarchical structure and form a complete cluster-head hierarchy, for each local leader... Perform a local breadth-first search (BFS) with the following specific rules: The search process is performed on the original undirected graph. Execute on, from node Start BFS, when the first encounter is a node with a degree not less than 10 ... And local leaders not visited during the current search process Stop the search when the time comes. (Note:) and exist The shortest path length in is and define As a node The upper layer distance, i.e. To its upper-level leaders The number of jumps.

[0055] If multiple degrees not less than 1 exist during the search process If there is a local leader, the node with the highest degree is selected as the upper-level leader; if there are multiple nodes with the same degree, one is randomly selected to ensure the rationality and uniqueness of the selection of the upper-level leader.

[0056] Let the set of local leaders with the largest degree in the topological graph be . For any Instead of performing the aforementioned local breadth-first search, we can directly set its path length to:

[0057] This ensures that this type of local leader is at the top of the global hierarchy, acting as the core cluster head and coordinating the overall cluster structure management. In engineering implementation, it can also be... Set a preset constant threshold that is not less than the above value to simplify the calculation process. For example... Figure 3 In (d), node 14 is the superior leader of node 8, node 8 is the superior leader of node 6, and node 6 is the superior leader of node 5, forming a clear three-level cluster head hierarchy.

[0058] Step 1.4: Initial Cluster Head Identification The initial cluster head is defined as the node degree. With path length Larger local leaders can be identified naturally in rooted forests based on local dominance, but to achieve accurate quantitative identification of cluster leaders, quantitative evaluation indicators need to be introduced. The specific steps are as follows: Step 1.4.1: Calculate quantitative evaluation indicators

[0059] Degree of each node With path length Multiply, we get This indicator comprehensively reflects the topological influence of a node. ) and hierarchical status ( ), The larger the value, the more suitable the node is as the initial cluster head.

[0060] Step 1.4.2: Node sorting and inflection point detection Press all nodes Sort the values ​​in descending order to obtain the node sequence. Simultaneously draw Scatter plots, using inflection point detection methods on scatter plots, determine the initial number of cluster heads. That is, the inflection point corresponds to The mutation point of the value, the node before the mutation point is the potential cluster head with strong influence.

[0061] Initial cluster head set determined: Select the first cluster head after sorting. Each node constitutes the initial central set. .

[0062] It should be noted that the results obtained in this stage Corresponding to the first-level centers with greater influence in the topology, such as Figure 3 As shown in (e), the primary cluster centers can be identified as nodes 14, 8, 6, and 5. However, this initial cluster head set is selected based solely on node topological features and does not guarantee that the distance from all nodes to these centers does not exceed two hops. It needs to be optimized and improved through subsequent cluster partitioning stages. Step 2 (Phase Two): Cluster Partitioning Based on Communication Constraints The core objective of this phase is to establish an initial cluster head set. Based on this, a two-hop coverage constraint is introduced and recursively extended to construct a multi-layer cluster head set that satisfies the communication hop count limit, ensuring that all nodes are within the two-hop coverage range of a certain cluster head. Simultaneously, the cluster head hierarchy structure is improved. The specific recursive iterative process involves the following steps: First, perform initialization settings: Let the current cluster head set be... Cluster head hierarchy is mapped as The initial cluster head The hierarchy is defined as follows (First-level cluster head); simultaneously, for each node Maintain two core parameters—the currently assigned cluster head. and in the original image From arrive Shortest path hop count During initialization , .

[0063] The following recursive process is then executed: Step 2.1: Two-hop cover search In the current cluster head set In a fixed case, for each cluster head In the original image Perform a breadth-first search (BFS) with at most two hops once to obtain all satisfying conditions. The set of nodes, that is, all nodes that can be covered within two hops of the cluster head.

[0064] Step 2.2: Node cluster affiliation update and identification of uncovered nodes For each node If through a certain cluster head The distance obtained Smaller than its current Then update it. for , for This ensures that each node belongs to the nearest cluster head, optimizing intra-cluster communication efficiency. After all nodes are updated, the set of all nodes not covered by any cluster head within a 2-hop range is denoted as:

[0065] Step 2.3: Determine if recursion terminates If the node set is not covered If the value is empty, it means that each node is already within the two-hop range of a certain cluster head, satisfying the two-hop coverage constraint, and the recursive process terminates; at this point, the current cluster head set... Mapping with hierarchy This constitutes the final hierarchical cluster head structure, which can be directly used for subsequent cluster collaborative management and control.

[0066] Step 2.4: Cluster Head Expansion and Hierarchical Update like If the value is not empty, the cluster head set needs to be expanded to include the next-level cluster heads to cover the uncovered nodes. The specific steps are as follows: Calculate the maximum level in the current cluster head set. The newly extended cluster head hierarchy is set to This ensures the continuity and orderliness of the cluster head hierarchy.

[0067] Induced subgraph (i.e., only includes uncovered nodes) Each connected component of the subgraph containing the effective communication links between them In the middle, select the node with the largest degree. Selecting the node with the highest degree as the new cluster head ensures that the new cluster head has strong communication coverage capabilities and can quickly cover the nodes of the connected component.

[0068] New cluster head Add to cluster head set, update cluster head set and set its hierarchy. Simultaneously, initialize all newly added cluster heads. , (The cluster head belongs to itself, and the distance is 0).

[0069] Return to step 2.1 and continue the recursive process until the set of uncovered nodes is reached. If empty, the final hierarchical cluster head structure is obtained.

[0070] Furthermore, this embodiment discloses the algorithm pseudocode specification. Combining the implementation steps of the above two stages, the complete pseudocode of the THLSC algorithm is as follows, see Table 1 and Table 2, which correspond to the two core stages of step 1 and step 2 respectively, to ensure the rigor and reproducibility of the algorithm logic.

[0071] Table 1. Algorithm Implementation Process for Step 1

[0072] Table 2. Implementation process of the algorithm in step 2

[0073] Furthermore, this embodiment discloses simulation experiments and performance evaluations.

[0074] To verify the effectiveness, superiority, and applicability of the THLSC unmanned clustering algorithm proposed in this invention in practical engineering, this embodiment conducts simulation experiments from two aspects: "demonstration of hierarchical clustering effect" and "quantitative evaluation of algorithm performance." Two typical comparison algorithms (Integer Programming Algorithm ILP and Simulated Annealing Algorithm SA) are selected and compared on random topology and public benchmark datasets to comprehensively verify the performance advantages of the THLSC algorithm in terms of clustering effect, computational efficiency, and scalability, providing data support for the engineering application of the algorithm.

[0075] First, experimental design and comparison scheme This experiment employs a dual testing scheme using both random topology and benchmark datasets to ensure the comprehensiveness and reliability of the experimental results. The specific experimental design is as follows: Test topology: First, a randomly generated Unit Disk Graph (UDG) is used to visually demonstrate the operating mechanism and hierarchical cluster head structure characteristics of the THLSC algorithm. Second, we used the publicly available benchmark dataset—the Eternal Dominating Set benchmark dataset provided by Benchmark et al. (https: / / www.ic.unicamp.br / ~cid / Problem-instances / Eternal-Dominating-Set / ), and selected several subgraphs (with 32, 109, and 270 nodes respectively) to verify the clustering effect of the algorithm on real topologies. Thirdly, Erd s–Rényi random graphs are used to test the performance scalability of algorithms in large-scale networks.

[0076] Algorithm Comparison: Two typical clustering algorithms were selected for comparison to ensure the rationality and comprehensiveness of the comparison. Integer Linear Programming (ILP): Based on PuLP modeling and calling the CBC solver, it can obtain an approximate optimal solution for the minimum 2-hop control set, which serves as an optimal lower bound reference for the number of cluster heads. Simulated Annealing (SA): A classic heuristic algorithm that can obtain better clustering results within a finite number of iterations, representing the performance level of existing heuristic clustering algorithms.

[0077] For open-source implementation references: The open-source implementations of ILP and SA can be found in the ds_verifier project (https: / / github.com / MarioGrobler / ds_verifier) ​​to ensure the consistency and reproducibility of the comparison algorithm implementation.

[0078] Performance evaluation metrics: Two core metrics are selected: first, runtime (in seconds), reflecting the algorithm's computational efficiency and real-time performance; second, cluster head set size. This reflects the algorithm's ability to control the cluster head size, among which ILP's As the optimal lower bound, SA and THLSC It reflects the degree of approximation to the optimal level.

[0079] Second, the hierarchical clustering effect is displayed. (1) Clustering effect of random unit disk diagram In a randomly generated unit disk diagram, such as Figure 4 As shown, first construct the local leader's... Decision diagrams, such as Figure 5 As shown, based on this decision graph, a multi-level cluster head set is identified, and the final hierarchical clustering result is as follows. Figure 6 As shown.

[0080] The clustering results clearly show that the cluster heads generated by the THLSC algorithm exhibit a distinct hierarchical structure: when a cluster head cannot cover all nodes within its 2-hop range, a new next-level cluster head will be automatically generated in the remaining uncovered area until all nodes are within the two-hop neighborhood of a certain cluster head, forming a multi-level progressive topology of "core cluster head - secondary cluster head - cluster member". This perfectly matches the "hierarchical and reconfigurable" clustering requirements proposed earlier, and clearly demonstrates the hierarchical membership of cluster heads and the dynamic adaptability of cluster structure.

[0081] (2) Clustering effect of benchmark dataset To further evaluate the performance of the THLSC algorithm on real topologies, several subgraphs from the Eternal Dominating Set benchmark dataset were selected, and the ILP, SA, and THLSC algorithms were run on the same computational example, as shown in Figures 7a to 7b. Figure 7c Clustering visualization results for three algorithms with 32, 109 and 270 nodes are presented (in these benchmark examples, the ILP solution can be regarded as the approximate optimal solution of the minimum 2-hop control set).

[0082] Comparing the clustering structures of the three methods, the following conclusions can be drawn: ILP algorithm: generates the fewest cluster heads, and the cluster heads are mostly concentrated in the degree center region of the topology, which can minimize the cluster head size. However, due to the pursuit of the optimal solution, the computational complexity is extremely high, and the cluster structure has no hierarchy, which cannot adapt to the needs of dynamic reconstruction. SA algorithm: Under the constraints of a limited number of iterations and time, the cluster head positions obtained are basically similar to those of ILP, but there are some redundant cluster heads, the number of cluster heads is slightly more than that of ILP, and there is no clear hierarchical structure, resulting in insufficient clustering flexibility. The THLSC algorithm shows a high degree of agreement between cluster head locations and ILP solutions. Most THLSC cluster heads overlap or are adjacent to ILP cluster heads in the topology, indicating that the cluster head selection method based on local search and hierarchical decision-making has good consistency with the global optimal model. At the same time, since THLSC explicitly introduces a multi-level cluster head mechanism, a small number of lower-level cluster heads are appropriately added in the topology edge region, which effectively improves the intra-cluster path length and communication coverage. This achieves a more reasonable trade-off between controlling the cluster head size and improving communication performance, and is more in line with actual engineering needs.

[0083] Third, quantitative evaluation of algorithm performance To verify the applicability and superiority of the THLSC algorithm in large-scale random communication networks, this invention... On the s–Rényi random graph, the running time and cluster head number of the three algorithms are compared with the node size. The specific experimental setup and results analysis are as follows: (1) Experimental setup Consider a series of random connected graphs with the number of nodes ranging from 10 to 1000, each size Take a sample, denoted as To maintain network sparsity and connectivity, edge generation uses Erd... The s–Rényi model, given an average degree Set connection probability under the target. If the generated graph is not connected, its largest connected subgraph is used as the experimental network to ensure the rationality of the experimental topology. ILP, SA, and THLSC algorithms are run on each random graph, and the running time and cluster head set size of the three algorithms are recorded. Repeat the test three times and take the average value to ensure the stability of the experimental results.

[0084] (2) The variation of running time with node size Figure 8 The running time of three algorithms on a random connected graph is given as a function of the number of nodes. The changing curve, Figure 9 The detailed trend of the running time of the THLSC algorithm is given, and the following analytical conclusions can be drawn from the curves: The THLSC algorithm has a runtime that is almost linearly related to the number of nodes, and its overall magnitude is extremely low. In small-scale networks, the runtime is basically within Order of magnitude (milliseconds); when When the number is increased to 1000, the runtime remains at [previous value]. With a time on the order of magnitude of approximately 0.04 seconds, it possesses near real-time solution capabilities, fully meeting the engineering requirements for real-time task allocation and online reconfiguration in unmanned swarms.

[0085] The SA algorithm's runtime increases significantly with network size. This is mainly because the simulated annealing algorithm requires multiple rounds of random perturbation and acceptance decisions within a large search space, and the number of iterations increases rapidly with network size. At times, the runtime reaches the ten-second level, indicating poor real-time performance.

[0086] ILP algorithms: in small-scale networks ( It runs relatively fast, with a solution time typically less than 1 second, and can find the optimal solution within a limited time; however, as... As the number of model variables and constraints increases, the number of variables and constraints expands exponentially or even exponentially. After that, the solution time for most instances is close to the preset limit (about 60 seconds). The solver often only returns a near-optimal feasible solution, cannot prove the optimality, and has extremely poor real-time performance, making it unsuitable for large-scale cluster scenarios.

[0087] In summary, from the perspective of computational efficiency and scalability, the THLSC hierarchical clustering algorithm... On random networks, cluster head selection and cluster partitioning can be completed in the range of milliseconds to tens of milliseconds, while SA and ILP take tens of seconds or even sixty seconds to run on large-scale networks, which is difficult to meet the requirements of real-time task allocation and online reconstruction. The THLSC algorithm has a significant advantage in computational efficiency.

[0088] (3) The change in the number of cluster heads with node size Figure 10 The number of cluster heads output by the three algorithms is shown. With the number of nodes The changing curves, overall trends, and performance analysis are as follows: ILP algorithm: It produces the fewest cluster heads and can be regarded as the lower bound of the global optimal solution. As the number of nodes increases from 10 to 1000, the number of cluster heads output by ILP increases from 1 to more than 40. The growth rate is significantly lower than linear, which reflects the advantage of the global optimization model in compressing the size of the control set. However, the cost is extremely high computational complexity and no hierarchical structure.

[0089] SA algorithm: The number of cluster heads obtained is generally greater than that of ILP, about 2 to 3 times that of ILP; in small-scale networks ( In large-scale networks, the difference in the number of cluster heads between SA and ILP is relatively small; however, in large-scale networks... On the ), the number of cluster heads output by SA is stable at about 2.5 to 3 times that of ILP, indicating that SA can obtain a high-quality approximate solution in a finite time. However, due to the lack of global optimality guarantee, there are still some redundant cluster heads and no hierarchical structure.

[0090] The THLSC algorithm outputs cluster heads on roughly the same order of magnitude as SA, and is approximately 2.5 to 3 times that of ILP; On random networks, the average number of cluster heads output by THLSC is about 2.8 times that of ILP, and the order of magnitude is similar to that of SA. This shows that the THLSC algorithm can achieve a 2-hop control set size of similar quality to SA with extremely low computational cost, while possessing hierarchical structure and dynamic reconstruction capabilities that the SA algorithm does not have.

[0091] It is worth noting that the THLSC algorithm, in its design, not only aims to minimize the number of cluster heads but also explicitly considers node locality and hierarchical structure: cluster heads tend to be local leaders with higher degrees, while cluster members preferentially belong to cluster heads with lower degrees but closer proximity, thus achieving an optimal trade-off between controlling the cluster head size and maintaining short intra-cluster communication paths. Therefore, compared to the "minimum control set" solution of ILP, the number of cluster heads output by THLSC increases slightly, but in return, the algorithm complexity is reduced from NP-hard integer programming to linear time complexity. Furthermore, it is more suitable for constructing a hierarchical communication structure of multi-level cluster head-cluster members, adapting to the dynamic needs of unmanned cluster collaborative operations.

[0092] Fourth, experimental conclusions Based on simulation results using random topology and benchmark datasets, the following core conclusions can be drawn: The THLSC two-hop hierarchical local search clustering algorithm proposed in this invention, combining a local search algorithm with two-hop communication constraints, has a time complexity of O(n log n). It can guarantee that all nodes can reach the cluster head in two hops and generate a clear cluster head-cluster member hierarchical structure. Compared with the ILP and SA algorithms, the THLSC algorithm has significant advantages in computational efficiency and scalability while maintaining the number of cluster heads close to the optimal level. It can efficiently realize the rapid construction of the distributed organizational structure of unmanned clusters, perfectly solve the three major technical problems of existing clustering algorithms, and adapt to the actual engineering needs of unmanned cluster collaborative operations.

[0093] This embodiment, as the core exposition of the technical solution of the present invention, first completes the unmanned swarm communication topology modeling through the UDG model, transforming the clustering problem into an optimization problem of 2-hop connected control sets, and clarifying the clustering objectives of "hierarchical, reconfigurable, and task-adaptive". Then, the THLSC clustering algorithm is proposed, and the implementation steps of the two core stages of "constructing the initial hierarchical structure" and "cluster partitioning based on communication constraints" are described in detail, and the algorithm pseudocode is standardized. Finally, through simulation experiments of multiple scenarios and multiple indicators, the THLSC algorithm is comprehensively compared with the ILP and SA algorithms, verifying the effectiveness and superiority of the algorithm.

[0094] This embodiment addresses the technical shortcomings of existing unmanned swarm clustering algorithms: the clustering logic is disconnected from job requirements, the cluster structure cannot be dynamically adjusted, and the multi-hop control set is disconnected from the clustering logic. It provides a communication topology foundation that is close to engineering practice for the design of subsequent unmanned swarm distributed collaborative task allocation algorithms, possessing high engineering application value and theoretical innovation significance. Specifically, it includes the following advantages: (1) High computational efficiency and strong real-time performance Since it does not use NP-hard integer programming for modeling and does not require simulated annealing for a large number of random iterations, but is based solely on local search and breadth-first traversal, the algorithm has low complexity. It can maintain millisecond-level solution time in the range of 10 to 1000 nodes, which is far superior to the tens of seconds of ILP and the tens of seconds of SA. It can meet the real-time task requirements of unmanned cluster online networking and dynamic topology reconstruction.

[0095] (2) Cluster head size is close to the optimal size and redundancy is low This invention uses the optimal number of cluster heads in ILP as a lower bound reference and selects the best cluster heads through topological influence. The final cluster head size is on the same order of magnitude as that of the SA algorithm, only 2.5 to 3 times the optimal lower bound. It achieves a near-global optimal clustering effect with extremely low computational cost, while taking into account both cluster head number control and communication coverage performance.

[0096] (3) Naturally adaptable to two-hop communication engineering constraints The algorithm uses two-hop reachability from node to cluster head as a hard constraint throughout, which fits the engineering application scenario of multi-hop forwarding and limited effective communication range in actual unmanned cluster communication, and avoids the problems of excessively large single cluster and load imbalance in traditional minimum control set.

[0097] (4) It has a hierarchical structure and supports hierarchical collaborative management and control. This invention generates a hierarchical cluster head backbone structure with distinct high and low levels. High-level cluster heads are responsible for overall planning, while low-level cluster heads complete edge coverage, forming a tree-like hierarchical topology. This facilitates hierarchical task scheduling, distributed negotiation, and dynamic replacement of faulty nodes in unmanned clusters, and its scalability is far superior to traditional clustering methods without hierarchy.

[0098] (5) Adaptation to large-scale random topologies and real baseline topologies Based on UDG random graphs, public benchmark datasets, and Erd... The s–Rényi large-scale random graph multi-scenario validation shows that the algorithm can stably output reasonable hierarchical and clustered results in artificial topology, real benchmark topology and large-scale random networks. It has a wide range of applicable scenarios and strong robustness.

[0099] This embodiment discloses an unmanned clustering system based on r-hop control sets, including: The initial model building module is used to abstract the unmanned cluster into an undirected communication topology graph; The local leader identification module is used to calculate the degree of each node in the undirected communication topology graph, construct a directed acyclic graph based on the degree of each node, and mark the node with an out-degree of zero as a local leader based on the directed acyclic graph. The upper-level leader identification module is used to perform a breadth-first search in the undirected communication topology graph for nodes marked as local leaders to determine the upper-level leader of the local leader node. Let the number of hops of the shortest path from the current local leader node to the upper-level leader be the level depth. The initial cluster head identification module is used to calculate the comprehensive influence index based on the degree and hierarchical depth of each node, and to select the first-level cluster heads based on the comprehensive influence index to generate the initial cluster head set. The two-hop coverage search module is used to perform an r-hop breadth-first search on each cluster head based on the initial cluster head set, and update all covered node cluster heads and the shortest distance from a node to a cluster head according to the search results; The uncovered node identification module is used to count the set of nodes not covered by the cluster head. If the set of uncovered nodes is empty, the recursion terminates, and the current set of cluster heads and node affiliation information are output, completing the clustering. If the set of uncovered nodes is not empty, the set of cluster heads is expanded, the next level cluster head is added, and the r-hop breadth-first search is re-executed until all nodes are covered. Then, the current set of cluster heads and node affiliation information are output, completing the clustering.

[0100] A schematic diagram of a terminal device according to an embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0101] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0102] The terminal device can be a desktop computer, laptop computer, cloud server, or other device with strong computing power. The terminal device may include, but is not limited to, a processor and memory.

[0103] The optimal choice for the processor is a multi-core high-speed central processing unit (CPU).

[0104] The memory can be used to store the computer program and / or module. The processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0105] If the modules / units integrated into the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0106] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A clustering method for unmanned clusters based on r-hop control sets, characterized in that, Includes the following steps: Abstract the unmanned cluster into an undirected communication topology graph; Calculate the degree of each node in the undirected communication topology graph, construct a directed acyclic graph based on the degree of each node, and mark the node with an out-degree of zero as a local leader based on the directed acyclic graph. For a node marked as a local leader, perform a breadth-first search in the undirected communication topology graph to determine the upper-level leader of the local leader node. Let the number of hops of the shortest path from the current local leader node to the upper-level leader be the level depth. The comprehensive influence index is calculated based on the degree and level depth of each node. First-level cluster heads are selected based on the comprehensive influence index to generate an initial cluster head set. Based on the initial set of cluster heads, perform a breadth-first search with r hops for each cluster head, and update all covered node cluster heads and the shortest distance from a node to a cluster head according to the search results; The process involves counting the set of nodes not covered by cluster heads. If the set of uncovered nodes is empty, the recursion terminates, and the current set of cluster heads and node affiliation information are output, thus completing the clustering. If the set of uncovered nodes is not empty, the set of cluster heads is expanded, the next level cluster head is added, and the r-hop breadth-first search is re-executed until all nodes are covered. At this point, the current set of cluster heads and node affiliation information are output, thus completing the clustering.

2. The unmanned clustering method based on r-hop control set according to claim 1, characterized in that, The process involves calculating the degree of each node in the undirected communication topology graph, constructing a directed acyclic graph (DAG) based on the degree of each node, and marking nodes with an out-degree of zero as local leaders, including: Calculate the degree of each node based on the undirected communication topology graph; Node The directed edge has the maximum pointing degree and is greater than the node. Neighboring nodes of self-degree Construct a directed acyclic graph (DAG) that satisfies and ,in For nodes The set of neighbors; When node All neighboring nodes have a degree less than or equal to that of node 1. The degree of the node itself, then With no outgoing edges, the current node is identified as a local leader, and a set of local leaders is generated. ; If a node has multiple outgoing edges, one outgoing edge is randomly retained.

3. The unmanned clustering method based on r-hop control set according to claim 2, characterized in that, The process of determining the upper-level leader of the local leader node includes: Nodes marked as local leaders Using the source node, perform a breadth-first search in the undirected communication topology graph, where the first encountered node has a degree greater than or equal to the current node. The node is defined as having its own degree and not being identified by other local leaders that have not been searched. The upper-level leader records the current node. In an undirected communication topology graph, the shortest path hop count to the upper-level leader is used as the hierarchical position parameter of the source node. If there are multiple other local leaders that meet the above conditions in the breadth-first search, the local leader with the highest degree is selected as the superior leader. If there are multiple local leaders with the highest degree, one of them is selected as the superior leader. For the local leader with the highest node degree in the undirected communication topology graph, its hierarchical position parameter is directly set to a preset value or infinity, and breadth-first search is not performed.

4. The unmanned clustering method based on r-hop control set according to claim 1, characterized in that, The process involves calculating a comprehensive influence index based on the degree and hierarchical depth of each node, selecting first-level cluster heads based on this comprehensive influence index, and generating an initial cluster head set, including: Degree of each node With path length Multiply to obtain quantitative evaluation indicators. ; Press all nodes Sort the values ​​in descending order to obtain the node sequence. ; Based on node degree With path length Plot a scatter plot, perform inflection point detection on the scatter plot, and select nodes before the inflection point by combining the node sequence. Each node constitutes the initial central set. .

5. The unmanned clustering method based on r-hop control set according to claim 1, characterized in that, The process involves performing an r-hop breadth-first search on each cluster head based on the initial cluster head set, and updating all covered node cluster heads and the shortest distance from a node to a cluster head based on the search results, including: Initialize the current initial center set ; For the current set For each cluster head in the undirected communication topology graph, a breadth-first search with r hops is performed to obtain the set of nodes covered by the cluster head, where r represents the preset maximum number of communication hops; If for each node If through a certain cluster head The distance obtained Less than its current shortest path hop count Then update its cluster head. for Shortest path hop count Distance .

6. The unmanned clustering method based on r-hop control set according to claim 5, characterized in that, If the set of uncovered nodes is not empty, perform the following steps: Calculate the maximum level in the current cluster head set. The newly expanded cluster head hierarchy is set to ; In the set of uncovered nodes Induced subgraph Determine each Select the node with the highest degree as the new cluster head. ; New cluster head Add to cluster head set, update cluster head set Set its hierarchy Initialize all newly added cluster heads and re-execute the recursive process until an uncovered set of nodes is reached. If empty, the final hierarchical cluster head structure is obtained.

7. An unmanned clustering system based on r-hop control sets, characterized in that, include: The initial model building module is used to abstract the unmanned cluster into an undirected communication topology graph; The local leader identification module is used to calculate the degree of each node in the undirected communication topology graph, construct a directed acyclic graph based on the degree of each node, and mark the node with an out-degree of zero as a local leader based on the directed acyclic graph. The upper-level leader identification module is used to perform a breadth-first search in the undirected communication topology graph for nodes marked as local leaders to determine the upper-level leader of the local leader node. Let the number of hops of the shortest path from the current local leader node to the upper-level leader be the level depth. The initial cluster head identification module is used to calculate the comprehensive influence index based on the degree and hierarchical depth of each node, and to select the first-level cluster heads based on the comprehensive influence index to generate the initial cluster head set. The two-hop coverage search module is used to perform an r-hop breadth-first search on each cluster head based on the initial cluster head set, and update all covered node cluster heads and the shortest distance from a node to a cluster head according to the search results; The uncovered node identification module is used to count the set of nodes not covered by the cluster head. If the set of uncovered nodes is empty, the recursion terminates, and the current set of cluster heads and node affiliation information are output, completing the clustering. If the set of uncovered nodes is not empty, the set of cluster heads is expanded, the next level cluster head is added, and the r-hop breadth-first search is re-executed until all nodes are covered. Then, the current set of cluster heads and node affiliation information are output, completing the clustering.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images