Unmanned cluster system based on hierarchical description and construction method thereof
Through the three-layer elastic architecture and dynamic update algorithm, the problem of low task execution efficiency and intelligence in complex environments of unmanned cluster systems is solved, and efficient and automatic task completion capabilities are achieved.
Patent Information
- Application Number
- CN202411001712.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-12
AI Technical Summary
The existing unmanned cluster systems cannot meet the needs of diversified tasks in complex environments, and the level of intelligence is low. It is necessary to improve autonomous control and cluster control capabilities to reduce manual intervention.
A three-layer elastic architecture is adopted, including the system layer, the task layer and the cluster layer. Dynamic updates are achieved through sub-graph matching algorithms and node replacement algorithms to ensure that the drone system automatically adjusts in the event of task changes and equipment failures, and improves work efficiency.
The top-down and step-by-step refinement of the unmanned cluster system has been achieved, which improves task execution efficiency and independent decision-making capabilities, reduces manual intervention, and enhances task completion capabilities in complex environments.
Smart Images

Figure CN118938998A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of unmanned cluster systems, and in particular relates to an unmanned cluster system based on hierarchical description and a construction method thereof. Background Art
[0002] In the existing technology, simple unmanned cluster systems using centralized or distributed architectures cannot meet the needs of unmanned clusters in complex environments. Therefore, the current true unmanned cluster system is still imperfect.
[0003] At present, the design of military unmanned swarm systems faces the following requirements:
[0004] (1) Unmanned swarm systems must meet the full spectrum of requirements for different mission scenarios. Faced with complex and harsh environments and diverse mission requirements, unmanned swarm systems need to make reasonable allocations to maximize economic and time benefits while completing missions.
[0005] (2) Unmanned swarm systems need to further improve their autonomous control and intelligence levels, and develop in the direction of low human intervention and high autonomy / intelligence. At present, the intelligence level of unmanned swarm systems is relatively low, mainly based on simple remote control and pre-programmed control, and requires more human intervention. In order to maximize the role of unmanned equipment, it is necessary to improve the control level and intelligence level, so that it has the ability to learn autonomously and adapt to the environment, and can make certain autonomous decisions and avoid dangers.
[0006] (3) Unmanned swarm systems need to continuously improve swarm control. Fully autonomous swarms can effectively enhance their ability to perform tasks in various complex and dangerous environments through close collaboration between individual units, and complete many tasks that are difficult for a single unmanned system to accomplish. This is the highest level of collaborative control of unmanned systems and the ultimate goal of unmanned swarm systems.
[0007] In publication number CN 112488506 A, a scalable distributed architecture and self-organizing method for an intelligent unmanned system cluster is proposed. It only invents a self-organizing method for an unmanned cluster system, does not associate the unmanned cluster system with tasks, does not assign attributes to specific unmanned equipment nodes, and does not explain the logical relationship between unmanned units to form a cluster. Summary of the invention
[0008] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an unmanned cluster system based on hierarchical description and a construction method thereof, which divides the drone environment, describes the unmanned cluster system architecture from the system, cluster, and task perspectives, and constructs the unmanned cluster system architecture.
[0009] In order to achieve the above object, the technical solution adopted by the present invention is:
[0010] An unmanned cluster system based on hierarchical description, the system adopts a three-layer elastic architecture consisting of a system layer, a task layer and a cluster layer;
[0011] The system layer is the bottom layer, which is used to represent the actual equipment nodes, and the corresponding clusters are formed by the nodes;
[0012] The task layer is the top layer, used to represent specific task requirements, and a task layer node represents a specific task;
[0013] The cluster layer is the second layer, which is used to provide the required capabilities for the cluster formed by the system layer. The capabilities include long-range strike, close-range strike, and close-range reconnaissance. A cluster layer node represents the capabilities provided by a group.
[0014] When the task nodes of the task layer in the three-layer elastic architecture change or the equipment nodes of the system layer change, which affects the changes of nodes in other layers, the unmanned cluster is dynamically updated.
[0015] The task layer nodes are supported by cluster layer nodes, and one task layer node can select multiple cluster layer nodes; the cluster layer nodes are supported by system layer nodes, and one cluster layer node can select multiple system layer nodes;
[0016] The three-layer system architecture, from top to bottom, completes the node selection from the upper layer to the lower layer according to the subgraph matching algorithm; from bottom to top, completes the node update from the lower layer to the upper layer according to the node replacement algorithm.
[0017] The system layer, task layer and cluster layer in the three-layer system architecture are all constructed in the form of graphs. The nodes in the system layer graph represent different types of heterogeneous devices; the nodes in the task layer graph represent specific task requirements; the nodes in the cluster layer graph represent the capabilities provided by the system layer nodes combined into clusters, such as long-range strike capabilities, short-range strike capabilities, etc. At the same time, the cluster layer nodes provide support for the task layer nodes through combination to complete the tasks of the corresponding task layer nodes.
[0018] Furthermore, the nodes in the system layer diagram include different types of heterogeneous equipment such as drones, unmanned vehicles, communication stations, satellites, and manned equipment, which are the execution units of the three-layer elastic architecture of the unmanned system cluster; the edges between nodes are determined by the physical locations of the nodes and the communication links between them; each node has static (unchanged) attributes and dynamic (variable attributes) in the attribute description, and the static attributes refer to the type, capability, and equipment of the equipment, which are unchanged attribute information, and the dynamic attributes refer to the attribute information that can be dynamically changed during the process due to the physical changes of the task or the equipment itself; the edge connection between two nodes is determined by the distance of the physical locations and the presence or absence of a communication link;
[0019] The communication link is defined as follows: if in the previous task, a connection relationship is established between two nodes;
[0020] When the distance between two nodes is less than a defined threshold (the distance threshold is determined according to the communication distance of the drone in the specific situation), the two nodes are defined to be connected in the graph description; when there is a communication link connection between the two nodes, the two nodes are also defined to be connected in the graph description; the dynamic properties of the nodes in the graph and the edges between the nodes will be dynamically updated with the migration of time, thereby forming a system-level equipment spatiotemporal graph with equipment as nodes and edges between equipment that evolve dynamically.
[0021] Furthermore, the task layer is the top layer of the three-layer elastic architecture of the unmanned system cluster, which is used to represent dynamic task changes; at the task layer, it is represented in the form of a graph; the unmanned cluster system architecture divides the total task into multiple subtasks based on the task information, each subtask is represented by a separate node, and edge connections are established between multiple subtasks according to the task logic; the task node contains task attributes; the edge connection relationship between task nodes depends on the definition of the task; a task node can correspond to multiple cluster layer nodes, which means that multiple clusters are required to complete the task node, and the task layer nodes can select the same cluster layer node and use them in the order of task time.
[0022] Furthermore, the cluster layer is the second layer of the three-layer elastic architecture of the unmanned system cluster, and is located between the system layer and the task layer; the cluster layer selects capability nodes at the cluster layer according to the requirements of the task layer, and selects equipment at the system layer (a local subgraph of the system layer) according to the capability nodes at the cluster layer;
[0023] Cluster layer nodes represent the capabilities provided by the combination of system layer nodes, such as long-range strikes and close-range reconnaissance; cluster layer capability nodes contain capability attributes; the edges between cluster layer nodes are defined as fully connected, representing that multiple nodes complete a task; a cluster layer node can correspond to multiple system layer nodes, representing that multiple physical equipment are required to achieve a certain capability; in the process of matching capabilities from cluster layer to system layer, the cluster layer capabilities can first be formed into a graph, and then the system layer can be matched with subgraphs according to the attributes to find the maximum matching subgraph to form a cluster.
[0024] Furthermore, when setting the data type of an attribute;
[0025] The task layer node attributes are divided into two types: 1) digital type, describing the task layer node ID; 2) text language, describing the task location of the task layer node;
[0026] Cluster layer node attributes are divided into two types: 1) digital type, describing the cluster layer node ID; 2) text language, describing the component units of the cluster layer node;
[0027] System layer nodes are divided into three types: 1) Boolean type, describing whether the system layer node has a certain function; 2) Digital type, describing the ID of the system layer node; 3) Text language, describing the name, direction, and position of the system layer node.
[0028] Furthermore, different data types are encoded in the experimental verification: 1) Boolean type is encoded using one-hot encoding; 2) digital type data is converted into binary floating point numbers for encoding; 3) text type attributes are encoded using word2vec.
[0029] A method for constructing an unmanned cluster system based on hierarchical description includes the following steps:
[0030] When a task arrives, the task information is extracted through the text key information extraction algorithm and divided into multiple subtasks; each subtask corresponds to a task layer node and is given all its attributes; the cluster layer node is selected according to the capabilities required to complete the task node;
[0031] After the task layer nodes select the cluster layer nodes, the cluster layer nodes first obtain the type and number of system layer nodes required according to the task requirements, and the cluster layer nodes generate the expected subgraph (the same type of devices are fully connected and communicate, and a leader is selected, and different types of devices communicate through the leader), and match the expected subgraph with the generated system layer graph to obtain the optimal subgraph, which is the subgraph corresponding to the cluster layer node; when all cluster layer nodes find the corresponding subgraph, the initial construction of the unmanned cluster system architecture based on the hierarchical description is completed.
[0032] Furthermore, if the mission changes or the drone is damaged, it is necessary to consider the dynamic update strategy of the three-layer architecture of the unmanned cluster, that is, changes in the nodes of one layer will affect the changes in the nodes of other layers; considering the corresponding meanings of the three different layers, the layers that will change due to the actual situation are the system layer and the task layer;
[0033] (1) Task layer node change: Detect whether the original task layer node is affected, and determine whether to build a new task node or update the original node attributes or delete the old node; when the task change is not significant, only update the corresponding task layer node attributes, and determine whether the cluster layer node corresponding to the original task can continue to complete the task. If it can continue to complete the task, no change is required; if change is required, update the system layer node corresponding to the cluster layer node, and make it meet the functions provided by the cluster layer again with the minimum update cost; if the task change is significant or a new task is added, delete the original task node, create a new task layer node and assign the corresponding task attributes, and construct the corresponding graph of the new task node according to the initial situation;
[0034] (2) System layer node changes: When a new device is added to the system architecture, the system layer adds a new node and assigns the corresponding system attributes. The node joins the nearest cluster waiting for tasks based on the distance threshold. When the system layer device fails or cannot meet the capacity requirements of the cluster layer node, it is necessary to search for other backup nodes. Among the replaceable nodes, the optimal node with the lowest cost is selected for replacement, the edges are updated, and the replaced nodes and connecting edges are deleted.
[0035] Beneficial effects of the present invention:
[0036] The three levels of the present invention can be dynamically adjusted according to multi-task requirements or node failures caused by multiple reasons, and the three-layer architecture can update and adjust the nodes by itself. In the three-layer architecture, the task layer is responsible for issuing tasks, the cluster layer provides the required capabilities for the tasks, and the system layer forms a top-down and step-by-step refinement through the corresponding clusters of the respective capabilities between the monomers. The lower layer provides support for the upper layer, which greatly improves work efficiency. In the three-layer architecture, the second layer, the cluster layer, is a virtual node, so considering the corresponding meanings of the three different levels, the levels that will change due to the actual situation are the system layer and the task layer, and the cluster layer will also be affected by the updates of the upper and lower layers. After the system layer and the task layer change, the three-layer system architecture can adjust itself, which will continue to make the task go smoothly.
[0037] The above three-layer architecture based on hierarchical description provides support for dynamic updates from the following two aspects:
[0038] (1) When the task changes, the task layer nodes will change accordingly. The cluster layer nodes will be reorganized according to the task while minimizing the selection of devices for the new capabilities, and will try to use the devices that originally supported the task. According to the changes in the cluster layer, the system layer will regenerate the subgraph based on the attributes of the single device capabilities, location, etc. to support the cluster layer nodes.
[0039] (2) When the number of system-level monomers increases or fails or becomes insufficient, the system-level nodes change accordingly. Failed monomers or monomers with different capabilities will be deleted. The architecture will search for the best device to replace the currently problematic device and regenerate the subgraph so that the task can proceed smoothly. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 It is a system-level dynamic spatiotemporal diagram of the unmanned cluster system diagram based on the hierarchical description of the present invention.
[0041] Figure 2 The present invention is based on a cluster layer diagram of an unmanned cluster system diagram described in a hierarchical manner.
[0042] Figure 3 This is a diagram of an unmanned cluster system based on hierarchical description according to the present invention.
[0043] Figure 4 This is a diagram of the dynamic update of the unmanned swarm system architecture after the mission changes.
[0044] Figure 5 This is a diagram of the dynamic update of the unmanned cluster system after the system layer nodes are changed.
[0045] Figure 6 A subgraph matching algorithm from cluster layer to system layer is constructed for unmanned cluster system.
[0046] Figure 7 It is an algorithm for replacing system-level nodes in an unmanned cluster system after system-level nodes are changed. DETAILED DESCRIPTION
[0047] The present invention will be further described in detail below in conjunction with the accompanying drawings.
[0048] The present invention provides an unmanned cluster system based on hierarchical description and a construction method thereof. In order to meet the needs of military unmanned cluster systems as much as possible, support multi-tasks, and improve operating efficiency, the method is based on hierarchical description and can be dynamically updated; it includes a three-layer elastic architecture;
[0049] The three-layer elastic architecture is divided into a system layer, a task layer, and a cluster layer;
[0050] The system layer is the lowest layer and is used to represent the actual equipment nodes;
[0051] The task layer is the top layer and is used to represent specific task requirements;
[0052] The cluster layer is the second layer, used to represent the functions provided by the cluster formed by the system layer;
[0053] A task layer node represents a specific task, supported by a cluster layer node, and a task layer node can select multiple cluster layer nodes; a cluster layer node represents the capabilities provided by a system layer cluster, supported by a system layer node, and a cluster layer node can select multiple system layer nodes. The three-layer system architecture, from top to bottom, can complete the node selection from the upper layer to the lower layer according to the subgraph matching algorithm; from bottom to top, the node update from the lower layer to the upper layer can be completed according to the node replacement algorithm.
[0054] (1) System layer: The system layer is constructed in the form of a graph. The nodes in the graph include different types of heterogeneous equipment such as drones, unmanned vehicles, communication stations, satellites, and manned equipment. They are the execution units of the three-layer elastic architecture of the unmanned system cluster. The edges between nodes are determined by the physical location of the nodes and the communication links (communication information) between them. Each node has static (unchanging) attributes and dynamic (variable attributes) in the attribute description;
[0055] Static attributes refer to the unchanging attribute information such as the type, capability, and equipment of the equipment. Dynamic attributes refer to the attribute information that can change dynamically during the process due to the physical changes of the task or the equipment itself. UAVs have static attributes and dynamic attributes. The static attributes include model, name, whether it has missiles, whether it has communication radar, whether it has high-definition cameras, whether it has infrared cameras, empty weight, maximum take-off weight, maximum flight altitude, maximum flight speed, range, whether it has reconnaissance function, whether it has strike function, whether it has communication function, etc. The dynamic attributes include system ID, is_using, mission ID, cluster ID, speed, altitude, direction, position, remaining range, link quality, and intelligence information. Unmanned vehicles have static attributes and dynamic attributes. The static attributes include model, name, whether it has missiles, whether it has communication radar, whether it has high-definition cameras, whether it has infrared cameras, weight, maximum speed, range, whether it has reconnaissance function, whether it has strike function, whether it has communication function, etc. The dynamic attributes include system ID, is_using, mission ID, cluster ID, speed, direction, position, remaining range, link quality, and intelligence information. The communication station has static and dynamic attributes. The static attributes include name, location, whether it has optical communication function, whether it has communication command function, whether it has wireless communication function, whether it has military cellular communication function, whether it has data link communication function, whether it has satellite communication function, etc. The dynamic attributes include system ID, is_using, mission ID, cluster ID, intelligence information, command information, and link quality. The satellite has static and dynamic attributes. The static attributes include name and orbital altitude; the dynamic attributes include system ID, is_using, mission ID, cluster ID, intelligence information, command information, and link quality. The edge connection between two nodes is determined by the distance of the physical location and whether there is a communication link. When the distance between two nodes is less than the defined threshold, the two nodes are defined to have a connection relationship in the graph description. When there is a communication link connection between two nodes, the two nodes are also defined to have a connection relationship in the graph description.
[0056] The dynamic properties of the nodes in the graph and the edges between the nodes will be dynamically updated over time, thus forming a system-level equipment spatiotemporal graph with equipment as nodes and edges between equipment dynamically evolving. Figure 1 ,In this figure, V1 to V8 are the equipment nodes that constitute the system layer, which mainly demonstrates the ,dynamic attribute changes of V1, V4, V7, and V8 over time and the changes of the migration edges of the ,entire system layer graph over time.
[0057] The initial devices are V1~V7, of which V1~V4 are in one area, and V5~V7 are in another area. Areas are connected based on physical distance. V3 and V5 have a communication connection before, so there is an edge connection between the two nodes. At time t1, some static and dynamic properties of V1, V4 and V7 are displayed; at time t2, the system layer diagram is updated, V1 node exits the system for some reason, V1 node and edge are deleted, and the dynamic properties of nodes V4 and V7 change; at t n At this moment, according to the attributes of V1, a new node V is selected a8 The V1 node is replaced and connected to other nodes based on physical locations. The dynamic properties of nodes V4 and V7 change.
[0058] (2) Task layer: The task layer is constructed in the form of a graph. The unmanned cluster system architecture divides the total task into multiple subtasks based on the task information. Each subtask is represented by a separate node, and multiple subtasks are connected by edges based on the task logic. The attributes of the task layer node include task ID, task name, equipment type and corresponding quantity required for the task, task target location, task required load, task start time, task end time, task area, task urgency, etc. The edge connection relationship between task nodes depends on the definition of the task. A task node can correspond to multiple cluster layer nodes, which means that multiple clusters are required to complete the task node. The task layer nodes can select the same cluster layer node and use them in the order of task time. For example, if a task is obtained, "Use 10 drones, 10 unmanned vehicles, 2 satellites, and 1 communication station to scout and attack area X", two task nodes "Scout Area X" and "Attack Area X" will be generated in the task layer. The system layer node type and corresponding quantity are assigned to the nodes, and the task attributes are attached to the nodes. The cluster layer node that can complete the task will be selected.
[0059] (3) Cluster layer: The cluster layer is constructed in the form of a graph. The cluster layer nodes represent the capabilities provided by the combination of system layer nodes, such as long-range strikes and close-range reconnaissance. The attributes of cluster layer nodes are cluster ID, task ID, cluster type, cluster corresponding to system layer nodes and quantity, reconnaissance range, reconnaissance accuracy, firepower, communication rate, command and control capabilities, etc. The edges between cluster layer nodes are defined as fully connected, representing that multiple nodes complete a task. A cluster layer node can correspond to multiple system layer nodes, representing that multiple physical equipment are required to achieve a certain capability. In the process of matching capabilities from cluster layer to system layer, the cluster layer capabilities can be first formed into a graph, and then the system layer can be matched with subgraphs based on the attributes to find the maximum matching subgraph to form a cluster. For example, the task of "strike area X" can be completed by cluster layer nodes that provide "long-range strike" capability, "medium-range" strike capability, and "close-range strike" capability. See Figure 2 , where Vb1 ~V b3 There are three nodes in the cluster layer; 3 drones, 2 unmanned vehicles, 1 satellite, and 1 communication station are allocated for long-range strikes; 4 drones, 4 unmanned vehicles, 1 satellite, and 1 communication station are allocated for medium-range strikes; and 2 drones, 1 unmanned vehicle, 1 satellite, and 1 communication station are allocated for close-range strikes. For the "long-range strike" cluster layer node, the expected subgraph can be generated according to its assigned system layer node: nodes of the same type are fully connected, and a leader is generated by calculating the distance. Nodes of different types communicate through the leader, which generates the expected graph of the "long-range strike" cluster layer node. The same is true for the "medium-range strike" cluster layer node and the "close-range strike" cluster layer node, generating the expected graph of the "medium-range strike" cluster layer node and the "close-range strike" cluster layer node. There are two ways to match the subgraph of the system layer for these two nodes: one is that the two nodes match the system layer graph according to the subgraph respectively to find the maximum matching subgraph. The other is to merge the expected subgraphs of the nodes, and then match them to the system layer graph to find the maximum matching subgraph.
[0060] For more information on the matching algorithm, see Figure 6 Algorithm 1.
[0061] (4) When setting the data type of an attribute, the task layer node attributes are divided into two types: 1) Numeric type, describing the task layer node ID, etc.; 2) Text language, describing the task location, etc. of the task layer node. Cluster layer node attributes are divided into two types: 1) Numeric type, describing the cluster layer node ID, etc.; 2) Text language, describing the component units of the cluster layer node, etc. System layer nodes are divided into three types: 1) Boolean type, describing whether the system layer node has a certain function; 2) Numeric type, describing the system layer node ID, etc.; 3) Text language, describing the name, direction, location, etc. of the system layer node.
[0062] (5) Different data types are encoded in the experimental verification: 1) Boolean type is encoded using one-hot encoding; 2) Numerical type data is converted into binary floating point numbers for encoding; 3) Text type attributes are encoded using word2vec.
[0063] In the present invention, by assigning attributes to the equipment nodes at the system layer, unmanned clusters can be established between the equipment nodes through attribute logic, so as to manage the tasks. On the one hand, from top to bottom, the task layer can select unmanned clusters with corresponding capabilities according to the task requirements; on the other hand, from bottom to top, the system layer always supports the completion of the task according to the dynamic changes of the single node.
[0064] A method for constructing an unmanned cluster system based on hierarchical description includes the following steps:
[0065] When a task arrives, it is divided into multiple subtasks by extracting task information. Each subtask corresponds to a task layer node and is given all its attributes. Cluster layer nodes are selected based on the capabilities required to complete the task node. After the cluster layer nodes are selected, the corresponding nodes of the cluster layer first generate the expected subgraph according to the task requirements (equipment of the same type fully connects and communicates, and a leader is selected, and different types of equipment communicate through the leader). The expected subgraph is matched with the generated system layer graph to obtain the optimal subgraph, which is the subgraph corresponding to the cluster layer node. When all cluster layer nodes find the corresponding subgraph, the initial construction of the unmanned cluster system architecture based on the hierarchical description is completed. See Figure 3 Example, where V c1 is the task node, V b1 and V b2 are cluster nodes with different capabilities, V a1 ~V a6 and V a7 ~V a10 It is the node that makes up the corresponding cluster. A task layer node V c1 Two cluster layer nodes V are required b2 and V b3 The capabilities provided are completed at the cluster level node V b2 In the corresponding system layer subgraph, V a1 For type 1, V a2 and V a3 For type 2, V a4 、V a5 and V a6 For type 3, V a7 It is type 4, V a9 It is type 5, full connection communication between the same type, and a leader is selected, V a1 For Type 1 leader, V a3 For type 2 leader, V a4 For type 3 leader, V a7 Is a Type 4 leader, V a9 It is a type 5 leader. Different types communicate with each other through the leader. Cluster layer node V b2 The same is true. a7 and V a9 Yes V b2 and V b3 Shared nodes participate in tasks according to task logic.
[0066] If tasks are added or changed, or drones are added or damaged, it is necessary to consider the dynamic update strategy of the three-layer architecture of the unmanned cluster, that is, changes in nodes in one layer will affect changes in nodes in other layers. Considering the corresponding meanings of the three different layers, the layers that will change due to actual conditions are the system layer and the task layer.
[0067] (1) Task change: Whenever a task changes, the system architecture will be dynamically updated from top to bottom. First, the task layer nodes will be dynamically updated. At the same time, the system architecture will determine the degree of similarity between the changed task and the old task based on the task attributes. If the tasks are similar, first determine whether it is necessary to change the cluster layer nodes. If it is not necessary to change the cluster layer nodes, then continue to analyze the attributes of all the system layer nodes that make up the cluster to determine whether it can continue to support the cluster layer functions. If it cannot support the cluster layer functions, it is necessary to update some of the system layer nodes. The specific update steps are: delete the system layer nodes that can no longer be used, and calculate the attribute capabilities through multiple encoding results within the distance threshold range of the cluster. Under the condition of minimum cost, select unused single or multiple entities so that their attributes can support the cluster layer to complete the corresponding tasks. If a suitable replacement node cannot be found and the task cannot be continued, the output task has encountered a problem. For specific algorithms, please refer to Figure 7 Algorithm 2. If the tasks are not similar, delete the original task node, create a new task layer node and assign the corresponding task attributes, then reselect the cluster layer node, and the cluster layer node also reselects the system layer node. Figure 4 , the figure shows that when task V c1 Change to V c2 When the unmanned cluster architecture is running, it will dynamically update the task layer, cluster layer and system layer from top to bottom. c1 Change to task node V c2 After judging, we know that the two nodes have certain similarities. Cluster layer node V b3 Also has completed task node V c2 Therefore, under the optimal selection condition, the task layer node V c2 Continue to use cluster layer node V b3 But the cluster layer node V b2 Can no longer complete new task node V c2 , so delete the cluster layer node V b2 To complete the task, V c2 Reselect a new cluster layer node V b4 , V b4 Similarly, select the system layer node.
[0068] (2) System-level node changes: When a system-level node is added, the attributes of the added nodes are used to determine whether there are cluster-level nodes that need to be added. If necessary, they are added to the corresponding cluster. If not, they become free nodes. When a system-level node fails or the capabilities provided cannot support the cluster layer to complete the task, the cluster architecture will be dynamically updated from bottom to top. First, the corresponding subgraph of the system layer will delete the system-level nodes that can no longer be used, and within the distance threshold range of the cluster, the attribute capabilities are calculated through multiple encoding results. Under the condition of minimum cost, single or multiple unused entities are selected so that their attributes can support the cluster layer to complete the corresponding task. If a suitable replacement node cannot be found and the task cannot be continued, the output task has encountered a problem. For specific algorithms, please refer to Figure 7 Algorithm 2. When the faulty node is the subgraph leader, the leader responsibility needs to be transferred first, and then the node is dynamically updated. Figure 5 , the figure shows monomer V a10 When changes are made, the unmanned cluster architecture will dynamically update the system layer and cluster layer from bottom to top. a10 Failure occurs and the task cannot be completed. Therefore, under the premise of minimum replacement cost, V a11 and V a12 The sum of the abilities can make up for V a10 The node is vacant, so select V a11 and V a12 Replace V a10 . And because V a10 The node is a leader, so it transfers the leader capability to V a11 . Delete V a10 Nodes and their edges, add new nodes V a11 and V a12 and its edges, and update the attributes of the corresponding cluster layer nodes at the same time.
Claims
1. An unmanned swarm system based on hierarchical description, It is characterized in that The system adopts a three-layer elastic architecture consisting of a system layer, a task layer, and a cluster layer; The system layer is the bottom layer, which is used to represent the actual equipment nodes, and the corresponding clusters are formed by the nodes; The task layer is the top layer, used to represent specific task requirements, and a task layer node represents a specific task; The cluster layer is the second layer, which is used to provide the required capabilities for the cluster formed by the system layer. A cluster layer node represents the capabilities provided by a cluster. When the task nodes of the task layer in the three-layer elastic architecture change or the equipment nodes of the system layer change, which affects the changes of nodes in other layers, the unmanned cluster is dynamically updated.
2. According to claim 1, an unmanned cluster system based on hierarchical description, It is characterized in that The task layer node is supported by the cluster layer node, and one task layer node can select multiple cluster layer nodes; Cluster layer nodes are supported by system layer nodes, and one cluster layer node can select multiple system layer nodes; The three-layer system architecture, from top to bottom, completes the node selection from the upper layer to the lower layer according to the subgraph matching algorithm; from bottom to top, completes the node update from the lower layer to the upper layer according to the node replacement algorithm.
3. The unmanned cluster system based on hierarchical description according to claim 1, It is characterized in that The system layer, task layer, and cluster layer in the three-layer system architecture are all constructed in the form of graphs. The nodes in the system layer graph represent different types of heterogeneous devices; the nodes in the task layer graph represent specific task requirements; The nodes in the cluster layer diagram represent the capabilities provided by the system layer nodes combined into a cluster, which are long-range strike capability and short-range strike capability. At the same time, the cluster layer nodes provide support for the task layer nodes through combination to complete the tasks of the corresponding task layer nodes. The nodes in the system layer diagram include different types of heterogeneous equipment such as drones, unmanned vehicles, communication stations, satellites, and manned equipment, which are the execution units of the three-layer elastic architecture of the unmanned system cluster; the edges between nodes are determined by the physical locations of the nodes and the communication links between them; each node has static attributes and dynamic attributes in the attribute description, and the edge connection between two nodes is determined by the distance between the physical locations and whether there is a communication link; The communication link is defined as follows: if in the previous task, a connection relationship is established between two nodes; When the distance between two nodes is less than the defined threshold, the two nodes are defined to be connected in the graph description; when two nodes are connected by a communication link, the two nodes are also defined to be connected in the graph description; the dynamic properties of the nodes in the graph and the edges between the nodes will be dynamically updated as time passes, thereby forming a system-level equipment spatiotemporal graph with equipment as nodes and edges between equipment that evolve dynamically.
4. The unmanned cluster system based on hierarchical description according to claim 3, It is characterized in that Static attributes refer to the attribute information of the equipment type, capabilities, and equipment that remain unchanged, while dynamic attributes refer to the attribute information that can change dynamically during the process due to physical changes in the task or the equipment itself.
5. The unmanned cluster system based on hierarchical description according to claim 1, It is characterized in that The task layer is the top layer of the three-layer elastic architecture of the unmanned system cluster, and is used to represent dynamic task changes; at the task layer, it is represented in the form of a graph; The unmanned swarm system architecture divides the total task into multiple subtasks based on the task information. Each subtask is represented by a separate node, and edge connections are established between multiple subtasks based on the task logic. Task nodes contain task attributes; the edge connection relationship between task nodes depends on the definition of the task; A task node can correspond to multiple cluster-layer nodes, which means that multiple clusters are needed to complete the task node. The task-layer nodes select the same cluster-layer node and use them in the order of task time.
6. The unmanned cluster system based on hierarchical description according to claim 1, It is characterized in that The cluster layer is the second layer of the three-layer elastic architecture of the unmanned system cluster, and is located between the system layer and the task layer; the cluster layer selects capability nodes at the cluster layer based on the requirements of the task layer, and selects equipment at the system layer based on the capability nodes at the cluster layer; The cluster layer is constructed in the form of a graph. The cluster layer nodes represent the capabilities provided by the combination of system layer nodes. The cluster layer capability nodes contain capability attributes. The edges between nodes at the cluster layer are defined as fully connected, representing multiple nodes completing a task; A cluster layer node can correspond to multiple system layer nodes, which means that multiple physical devices are required to achieve a certain capability; The process of matching capabilities from the cluster layer to the system layer first forms a graph of the cluster layer capabilities, then performs sub-graph matching on the system layer based on the attributes, and finds the maximum matching sub-graph to form a cluster.
7. An unmanned cluster system based on hierarchical description according to any one of claims 3 to 6, It is characterized in that When setting the data type of an attribute; Task layer node attributes are divided into two types: 1) digital type, describing the task layer node ID; 2) text language, describing the task location of the task layer node; Cluster layer node attributes are divided into two types: 1) Numeric type, describing the cluster layer node ID; 2) Text language, describing the components of cluster layer nodes; System layer nodes are divided into three types: 1) Boolean type, describing whether the system layer node has a certain function; 2) Digital type, describing the ID of the system layer node; 3) Text language, describing the name, direction, and position of the system layer node.
8. The unmanned cluster system based on hierarchical description according to claim 7, It is characterized in that Different data types are encoded in the experimental verification: 1) Boolean type is encoded using one-hot encoding; 2) Numerical type data is converted into binary floating point numbers for encoding; 3) Text type attributes are encoded using word2vec.
9. A method for constructing an unmanned cluster system based on hierarchical description according to any one of claims 1 to 8, It is characterized in that The following steps are involved: When a task arrives, the task information is extracted through the key information extraction algorithm and divided into multiple subtasks; each subtask corresponds to a task layer node and is given all its attributes; cluster layer nodes are selected based on the capabilities required to complete the task node; After the task layer nodes select the cluster layer nodes, the cluster layer nodes first obtain the type and number of system layer nodes required according to the task requirements, and the cluster layer nodes generate the expected subgraph (the same type of devices are fully connected and communicate, and a leader is selected, and different types of devices communicate through the leader), and match the expected subgraph with the generated system layer graph to obtain the optimal subgraph, which is the subgraph corresponding to the cluster layer node; when all cluster layer nodes find the corresponding subgraph, the initial construction of the unmanned cluster system architecture based on the hierarchical description is completed.
10. The method for constructing an unmanned cluster system based on hierarchical description according to claim 9, It is characterized in that If the mission changes or the drone is damaged, it is necessary to consider the dynamic update strategy of the three-layer architecture of the unmanned swarm, that is, changes in the nodes of one layer will affect the changes in the nodes of other layers; considering the corresponding meanings of the three different layers, the layers that will change due to the actual situation are the system layer and the task layer; (1) Task layer node change: detect whether the original task layer node is affected, and determine whether to build a new task node, update the original node attributes, or delete the old node; When the task changes slightly, only the corresponding task layer node attributes need to be updated, and it is determined whether the cluster layer node corresponding to the original task can continue to complete the task. If it can continue to complete the task, no change is required; if change is required, the system layer node corresponding to the cluster layer node is updated to meet the functions provided by the cluster layer again at the minimum cost; if the task changes significantly or a new task is added, the original task node is deleted, and a new task layer node is created to assign the corresponding task attributes, and the corresponding graph of the new task node is constructed according to the initial situation; (2) System layer node change: When a new device is added to the system architecture, the system layer adds a new node and assigns the corresponding system attributes. The node joins the nearest cluster based on the distance threshold and waits for tasks; When the system-layer equipment fails or cannot meet the capacity requirements of the cluster-layer nodes, it is necessary to search for other backup nodes, select the optimal node with the lowest cost among the replaceable nodes for replacement, update the edges, and delete the replaced nodes and connecting edges.
