Wireless broadband ad hoc network communication mode switching method and system based on cluster unmanned scene

By using graph neural network prediction and distributed switching mechanisms, the problem of dynamic adaptation of communication modes in unmanned cluster systems is solved, enabling adaptive switching of wireless self-organizing network modes and improving communication efficiency and security.

CN121442431APending Publication Date: 2026-01-30CHINA SOUTH IND GRP SHANGHAI ELECTRIC CONTROL RES INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511626393.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-07
Publication Date
2026-01-30

AI Technical Summary

Technical Problem

Existing wireless ad hoc network technologies struggle to dynamically adapt communication modes to mission scenarios in unmanned swarm systems, resulting in low energy efficiency, poor communication continuity, and difficulty in ensuring security.

Method used

A pre-trained graph neural network is used to predict communication modes based on multimodal perception data. Combined with a 'request-consensus-double buffer switching-confirmation' mechanism, distributed smooth mode switching is achieved. By integrating task status, motion parameters and environmental interference information, the optimal communication mode is automatically selected.

Benefits of technology

It enables adaptive communication mode switching of unmanned cluster systems at different task stages, ensuring communication continuity and security, reducing the computational overhead and memory usage of mode switching, and supporting efficient and secure wireless ad hoc networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121442431A_ABST
    Figure CN121442431A_ABST
Patent Text Reader

Abstract

The invention relates to a wireless broadband ad hoc network communication mode switching method and system based on a cluster unmanned scene, belongs to the technical field of unmanned aerial vehicle control, and solves the problem that an unmanned cluster system cannot safely and efficiently adapt to a wireless ad hoc network communication mode according to a task scene in the prior art. The method comprises the following specific steps: presetting a plurality of communication modes of an ad hoc network in a task life cycle of the cluster unmanned system; the cluster master control node receives the multi-mode sensing data collected by the slave control nodes respectively, and predicts a target communication mode by using a pre-trained graph neural network based on the multi-mode sensing data to obtain a to-be-switched target communication mode and a corresponding confidence value; whether mode switching is triggered or not is decided based on the confidence value, if the decision is triggered, the ad hoc network executes mode switching, the communication parameters corresponding to the preset target communication mode are configured, and efficient adaptation to the wireless ad hoc network communication mode is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle control, and in particular to a wireless broadband ad hoc network communication mode switching method and system based on a cluster unmanned scenario. BACKGROUND

[0002] With the rapid development of unmanned system technology, clusterization and collaboration have become an important development direction for intelligent agents such as unmanned aerial vehicles, unmanned vehicles, unmanned boats, etc. to perform complex tasks. In application scenarios such as military reconnaissance, emergency rescue, logistics distribution, agricultural plant protection, and environmental monitoring, multiple unmanned devices need to share data and cooperatively control in a high-speed, low-latency, and high-reliability manner in an environment without fixed communication infrastructure support.

[0003] To meet the above needs, wireless broadband ad hoc network (WBAHN) technology is widely used in cluster unmanned systems. This network does not rely on a central node or base station, and each slave node constructs a temporary and dynamic topology communication network through self-organization, supporting real-time video transmission, state synchronization, task scheduling, and other high-bandwidth services. Typically, the wireless ad hoc network protocol based on IEEE 802.11s supports multi-hop routing and distributed topology management, and has been applied in some unmanned systems, but its QoS mechanism is fixed and cannot be dynamically adjusted according to the scene. TDMA / CDMA time slot scheduling mechanism is used to improve the efficiency of concurrent communication of multiple nodes and is commonly used in unmanned aerial vehicle cluster communication, but the scheduling period is fixed and lacks scene awareness capability. Cognitive radio and spectrum sensing technology improves anti-interference capability through dynamic spectrum access, but is mainly used for point-to-point communication and is not linked to the task mode. The software-defined ad hoc network architecture based on SDN / NFV allows network functions to be programmable and has a certain flexibility, but it is complex to deploy on resource-constrained slave nodes and has poor real-time performance.

[0004] It can be seen that the existing wireless ad hoc network technology has the following defects: fixed communication parameters and protocol configurations are mostly used, making it difficult to dynamically adapt network behavior to the task scene, resulting in low energy efficiency; there is a lack of scene awareness and decision-making mechanism, and the task state, environmental information (such as electromagnetic environment, moving speed, and topology density) are not linked to the communication mode selection; mode switching lags or oscillates, and switching decisions rely on manual preset or simple threshold judgment, which can easily lead to switching delay or frequent jitter, affecting communication continuity and increasing the risk of communication interruption; safety and efficiency are difficult to balance, and in the concealment mode, bandwidth is sacrificed to reduce the risk of interception, and in the task mode, high transmission power increases the risk of exposure. In summary, the existing technology cannot achieve the goal of "on-demand networking, context switching, and safe and efficient" cluster communication. SUMMARY

[0005] In view of the above analysis, the embodiments of the present application aim to provide a wireless broadband ad hoc network communication mode switching method and system based on a cluster unmanned scenario, to solve the problem that the existing unmanned cluster system cannot safely and efficiently adapt to the wireless ad hoc network communication mode according to the task scenario.

[0006] The purpose of the present application is mainly realized by the following technical solutions:

[0007] On the one hand, the embodiments of the present application provide a wireless broadband ad hoc network communication mode switching method based on a cluster unmanned scenario, comprising the following steps:

[0008] Pre-set multiple communication modes of the ad hoc network in the task life cycle of the cluster unmanned system; wherein the cluster unmanned system comprises a master node and a plurality of slave nodes;

[0009] The master node receives multi-modal perception data collected by each slave node, and uses a pre-trained graph neural network to predict the target communication mode based on the multi-modal perception data, to obtain the target communication mode to be switched and the corresponding confidence value;

[0010] Based on the confidence value, it is decided whether to trigger mode switching, if it is decided to trigger, the ad hoc network performs mode switching, and the communication parameters corresponding to the pre-set target communication mode are configured.

[0011] Further, the pre-trained graph neural network comprises a spatial convolution module, a time attention and a prediction module; the multi-modal perception data comprises the task phase encoding, speed, relative position and wireless network communication link quality of the slave node itself; obtaining the target communication mode to be switched and the corresponding confidence value comprises:

[0012] Taking each slave node as a graph node, the corresponding speed and relative position as node attributes, and the link quality as edge attributes, a dynamic sparse graph is constructed;

[0013] The spatial convolution module is used to perform spatial coding on the dynamic sparse graph to obtain a plurality of edge-level embedding features;

[0014] The plurality of edge-level embedding features are time-position encoded to obtain time embedding features, and the time attention is used to extract time information from the time embedding features to obtain space-time embedding features;

[0015] The prediction module is used to perform mode prediction based on the space-time embedding features to obtain the target mode to be switched and the corresponding confidence value.

[0016] Further, the mode switching is performed based on the following process:

[0017] Each slave node broadcasts a switching request to the adjacent nodes;

[0018] After reaching a distributed consensus with neighboring nodes, the dual-mode buffering mechanism is activated. After buffering, the current mode communication channel is closed, the new mode configuration is activated, and a confirmation message indicating that the switch is complete is broadcast.

[0019] The post-evaluation module monitors the communication quality under the target mode, and if the communication quality deteriorates, it reverts to the current mode.

[0020] Furthermore, the dual buffering mechanism allows the current mode and the target mode to run in parallel for a preset time.

[0021] Furthermore, if the confidence value of the target mode is greater than a preset threshold and remains so for a preset time, a decision is made to trigger a switch to the target mode; otherwise, the current mode is maintained.

[0022] Furthermore, the communication modes include collaborative mode, task execution mode, anti-interception mode, and emergency withdrawal mode.

[0023] Furthermore, the samples are calibrated based on the triggering conditions of each mode, and the graph neural network is trained using the calibrated samples as input. The triggering conditions include entering the anti-interception mode when strong electromagnetic interference or enemy detection signals are detected; triggering the mission execution mode when the target location is reached and a mission start command is received; and triggering the emergency withdrawal mode when the mission is abnormal or an withdrawal command is received.

[0024] Furthermore, the preset communication parameters include modulation and coding, time slot allocation strategy, transmit power level, and routing protocol.

[0025] On the other hand, embodiments of the present invention provide a wireless broadband self-organizing network communication mode switching system based on cluster unmanned scenarios, including: a mode setting module, used to preset multiple communication modes of the self-organizing network in the task life cycle of the cluster unmanned system;

[0026] The sensing module is used to collect multimodal sensing data;

[0027] The mode decision module is used to predict the target communication mode based on the multimodal perception data using a pre-trained graph neural network, to obtain the target mode to be switched and the corresponding confidence value; and to decide whether to trigger mode switching based on the confidence value.

[0028] The execution module is used to switch execution modes based on the decision that triggers the switching.

[0029] The mode database and configuration management module is used to store the preset communication parameters for each communication mode and configure the communication parameters when switching modes.

[0030] Furthermore, the pre-trained graph neural network includes a spatial convolution module, a temporal attention module, and a prediction module; the multimodal perception data includes the task stage encoding, speed, relative position, and wireless network communication link quality of the control node itself; the target communication mode to be switched and the corresponding confidence value are obtained, including:

[0031] Using each of the slave control nodes as graph nodes, with the corresponding speed and relative position as node attributes, and the link quality as edge attributes, a dynamic sparse graph is constructed.

[0032] The dynamic sparse graph is spatially encoded using a spatial convolution module to obtain multiple edge-level embedding features;

[0033] Temporal position encoding is performed on multiple edge-level embedding features to obtain temporal embedding features. Temporal information is extracted from the temporal embedding features using temporal attention to obtain spatiotemporal embedding features.

[0034] The prediction module uses the spatiotemporal embedding features to perform pattern prediction, thereby obtaining the target pattern to be switched and the corresponding confidence value.

[0035] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0036] 1. This invention proposes a multi-dimensional communication mode system oriented towards the task stage. It pre-defines the communication mode under typical application scenarios, integrates multi-dimensional information such as task status, motion parameters, and environmental interference, constructs a scene feature vector, and achieves adaptive recognition of the mode through a lightweight AI model. This avoids the rigidity problem of traditional threshold methods, so as to realize the adaptive switching of wireless self-organizing network mode between the task stage and the operating environment of the cluster unmanned system.

[0037] 2. A "request-consensus-double-buffered switching-confirmation" mechanism is proposed. Through a smooth switching protocol driven by distributed consensus, communication is not interrupted during mode switching, supporting collaborative switching between cluster nodes and avoiding communication mismatch caused by asynchronous configuration.

[0038] 3. This invention is the first to systematically map the task flow of a cluster unmanned system to the wireless communication mode, and proposes communication mode configuration templates for five typical application scenarios to achieve "task-driven" operation.

[0039] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0040] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0041] Figure 1 This is a flowchart of a wireless broadband self-organizing network communication mode switching method based on a cluster-based unmanned scenario, according to an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram illustrating the communication mode switching state according to an embodiment of the present invention;

[0043] Figure 3 This is a structural block diagram of a wireless broadband self-organizing network communication mode switching system based on a cluster-unmanned scenario, according to an embodiment of the present invention. Detailed Implementation

[0044] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0045] Example 1

[0046] A specific embodiment of the present invention discloses a method for switching communication modes in a wireless broadband ad hoc network based on a cluster-based unmanned scenario, such as... Figure 1 As shown, it includes the following steps:

[0047] Step S1: Pre-set multiple communication modes of the self-organizing network during the task lifecycle of the cluster unmanned system; wherein, the cluster unmanned system includes a master control node and several slave control nodes;

[0048] Step S2: The master control node receives multimodal sensing data collected by each of the slave control nodes, and uses a pre-trained graph neural network to predict the target communication mode based on the multimodal sensing data, thereby obtaining the target communication mode to be switched and the corresponding confidence value.

[0049] Step S3: Based on the confidence value, decide whether to trigger mode switching. If the decision is triggered, the self-organizing network performs mode switching and configures the communication parameters corresponding to the preset target communication mode.

[0050] Using the above methods, a multi-mode wireless broadband self-organizing network architecture and its intelligent switching mechanism for unmanned cluster scenarios were designed. Based on the task stage and operating environment of the unmanned cluster system, a graph neural network was used to predict the communication mode and make switching decisions on the predicted mode to automatically select the optimal communication mode and achieve precise matching between communication performance and task requirements.

[0051] Typically, during actual mission execution, swarm unmanned systems undergo several typical stages, such as: preparation stage (pre-mission configuration, parameter loading, and formation initialization); movement / maneuver stage (path coordination and collision avoidance during swarm movement); stealth penetration stage (anti-interception and anti-jamming communication under low detectability requirements); mission execution stage (target identification, information transmission, and coordinated attack or operation); and withdrawal / recovery stage (status reporting, path convergence, and safe return). The performance requirements for the communication network differ significantly between these stages: the preparation stage emphasizes high-capacity data preloading; the movement stage emphasizes link stability under high mobility; the stealth stage requires low interception probability and strong anti-jamming capabilities; and the mission execution stage requires high throughput and low latency. To automatically select the optimal communication mode based on the mission stage and operating environment of the swarm unmanned system and to achieve smooth and reliable mode switching, this application proposes a wireless broadband ad hoc network communication mode switching method based on swarm unmanned scenarios.

[0052] Specifically, in step S1, the cluster unmanned system includes a master control node and several slave control nodes. In the wireless broadband self-organizing network scenario of the cluster unmanned system, in order to adapt the self-organizing network communication mode according to the actual task of the cluster unmanned system, multiple application modes, i.e. communication modes, of the self-organizing network in the task life cycle of the cluster unmanned system are first preset.

[0053] For example, such as Figure 2 As shown, the diagram divides communication into five core modes: DataLoading Mode, Coordination Mode, Low Probability of Intercept Mode, Mission Execution Mode, and Emergency Retreat Mode. The state transition arrows in the diagram indicate the switching paths between modes. Each path is labeled with its triggering conditions and decision-making basis. The main parameters for each mode are shown in Table 1.

[0054] Table 1

[0055]

[0056] Specifically, in step S2, the cluster of unmanned nodes (including master and slave nodes) is equipped with a wireless broadband communication module, task status sensors, environmental perception units (such as GPS, IMU, RSSI / SNR monitoring, interference source and radiation source detection equipment), and network status monitors for multimodal data acquisition. After the cluster of unmanned nodes recognizes that the data loading is complete, it enters the subsequent automatic identification and switching process of the working modules. This includes:

[0057] S21. Each slave node collects multimodal sensing data in real time, including task stage codes (from the task planning system), node movement speed, relative position, electromagnetic noise intensity, and wireless communication link quality (RSSI, SNR), and uploads it to the master node.

[0058] The system utilizes mission status sensors to receive phase instructions (i.e., mission phase codes) from the mission planning system. The environmental perception unit includes motion sensors and wireless environmental perception units. Motion sensors, such as GPS and IMU, are used to acquire speed, acceleration, and relative position. Wireless environmental perception units include spectrum scanning, received signal strength / signal-to-noise ratio (RSSI / SNR) monitoring, and interference source radiation source reconnaissance equipment (used to detect strong electromagnetic interference or enemy detection signals). The network status monitor is used to monitor link quality, packet loss rate, and latency.

[0059] S22. The master control node uses a pre-trained graph neural network to predict the target communication mode based on the multimodal perception data, obtaining the target communication mode to be switched and the corresponding confidence value; wherein, the pre-trained graph neural network includes a spatial convolution module, a temporal attention module, and a prediction module; the specific process of obtaining the target communication mode is as follows:

[0060] S221. Using each of the slave control nodes as graph nodes, with the corresponding speed and relative position as node attributes, and the link quality as edge attributes, construct a dynamic sparse graph.

[0061] For example, each slave node is used as a graph node, and directed edges are constructed using the K nearest neighbor method (K≤8). The edge set is reconstructed every 5-10 frames to reduce the computational overhead caused by topology jitter. For any node i, when the link packet loss rate is detected to be >30% or RSSI <-90dBm, the corresponding edge is deleted immediately to form a dynamic sparse graph.

[0062] S222. Spatial encoding is performed on the dynamic sparse graph using a spatial convolution module to obtain multiple edge-level embedding features;

[0063] For example, the spatial convolutional module includes a stitching layer, a multilayer perceptron (MLP), and a max-pooling layer. The stitching layer concatenates radio frequency measurements (RSSI, SNR), geometric quantities (relative distance, velocity), and message statistics (packet loss rate), encoding them into an edge feature vector e. ij For each edge (i,j), based on the sending node feature x i (If there is a directed edge from i to j, then the sending node is x) i ), Receiver node feature x j With edge feature vector e ij The edge message m is generated by MLP. ijThe set of all incoming edges at each receiving node is aggregated by max pooling to obtain the node embedding features within time step T, and then multiple edge-level embedding features are obtained in consecutive time steps.

[0064] S223. Perform temporal position encoding on multiple edge-level embedding features of continuous time steps to obtain temporal embedding features. Use temporal attention to extract temporal information from the temporal embedding features to obtain global spatiotemporal embedding features.

[0065] S224. The prediction module performs pattern prediction based on spatiotemporal embedding features to obtain the target pattern to be switched and the corresponding confidence value.

[0066] For example, using a Bayesian head as the prediction module, the spatiotemporal embedding features are input into the Bayesian head, and the predicted probabilities of each target mode are obtained through multiple forward inferences. After comprehensive confidence processing, the target mode to be switched and its confidence value are obtained.

[0067] It should be noted that during the training of the graph neural network, the samples are pattern-labeled based on the triggering conditions of each mode. The pattern-labeled samples are then used as input to train the graph neural network. The triggering conditions include: entering anti-interception mode when strong electromagnetic interference or enemy detection signals are detected; triggering mission execution mode when the target location is reached and a mission start command is received; and triggering emergency withdrawal mode when the mission is abnormal or an withdrawal command is received. Based on these triggering conditions, historical multimodal data is labeled to obtain the original training set.

[0068] Specifically, in step S3, a decision is made based on the predicted target communication mode to be switched and its corresponding confidence value. If the confidence value of the target mode is greater than a preset threshold and remains so for a preset time, then the decision is made to trigger the target mode switch; otherwise, the current mode is maintained. When a decision is made to trigger the switch, the ad hoc network performs mode switching based on the above process:

[0069] S31. Each slave node broadcasts a handover request to its neighboring nodes; the request carries the target mode and the time stamp of the target mode to take effect, which is used to agree that all nodes will switch simultaneously within the time stamp.

[0070] S32. After reaching a distributed consensus with neighboring nodes, start the mode double buffering mechanism, close the current mode communication channel after buffering, activate the new mode configuration, and broadcast a confirmation message of the switch completion. The double buffering mechanism is a preset time for the current mode and the target mode to run in parallel, such as 200ms.

[0071] Furthermore, preset communication parameters for each communication mode are stored in a mode database for later use during mode switching. These communication parameters include modulation and coding schemes (MCS), time slot allocation strategies, transmit power levels, and routing protocols (AODV, OLSR, Fisheye, etc.).

[0072] S33. Monitor the communication quality in the target mode after the switch is executed using the post-evaluation module. If the communication quality deteriorates, revert to the current mode. Deterioration in communication quality refers to RSSI / SNR both being less than a preset threshold for a sustained period.

[0073] Compared with existing technologies, this embodiment provides a wireless broadband self-organizing network communication mode switching method based on cluster unmanned scenarios. It pre-sets communication modes under typical application scenarios, integrates multi-dimensional information such as task status, motion parameters, and environmental interference, constructs scene feature vectors, and achieves adaptive mode recognition through a lightweight AI model. Based on the task stage and operating environment of the cluster unmanned system, it automatically selects the optimal communication mode, and finally achieves smooth and reliable mode switching.

[0074] Example 2

[0075] Another specific embodiment of the present invention discloses a wireless broadband self-organizing network communication mode switching system based on unmanned cluster scenarios, such as... Figure 3 As shown, it includes:

[0076] The mode setting module is used to preset multiple communication modes of the self-organizing network during the mission lifecycle of the cluster unmanned system;

[0077] The sensing module is used to collect multimodal sensing data;

[0078] The mode decision module is used to predict the target communication mode based on the multimodal perception data using a pre-trained graph neural network, to obtain the target mode to be switched and the corresponding confidence value; and to decide whether to trigger mode switching based on the confidence value.

[0079] The execution module is used to switch execution modes based on the decision that triggers the switching.

[0080] The mode database and configuration management module is used to store the preset communication parameters for each communication mode and configure the communication parameters when switching modes.

[0081] The system can switch wireless broadband ad hoc network communication modes according to any of the methods described in Embodiment 1. Related aspects can be referenced from each other, and are not repeated in this embodiment.

[0082] Compared with existing technologies, the wireless broadband self-organizing network communication mode switching system provided in this embodiment, through the cooperation of various modules, enables the wireless self-organizing network mode switching of the unmanned cluster system in an adaptive task phase and operating environment. It features low computational overhead and low memory usage, can run in real time on embedded platforms, and has engineering application value.

[0083] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0084] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A wireless broadband ad hoc network communication mode switching method based on a cluster unmanned scenario, characterized in that, The method comprises the following steps: Pre-set multiple communication modes of the ad hoc network in the task life cycle of the swarm unmanned system; wherein the swarm unmanned system comprises a master node and a plurality of slave nodes; The master node receives multi-modal perception data collected by each slave node respectively, and performs target communication mode prediction based on the multi-modal perception data using a pre-trained graph neural network to obtain a target communication mode to be switched and a corresponding confidence value; Based on the confidence value, it is decided whether to trigger mode switching. If it is decided to trigger, the ad hoc network performs mode switching and configures the communication parameters corresponding to the pre-set target communication mode.

2. The method of claim 1, wherein, The pre-trained graph neural network comprises a spatial convolution module, a time attention module and a prediction module; the multi-modal perception data comprises task phase encoding, speed, relative position and wireless network communication link quality of the slave node itself; To obtain the target communication mode to be switched and the corresponding confidence value, the following steps are included: Taking each slave node as a graph node, the corresponding speed and relative position as node attributes, and the link quality as edge attributes, a dynamic sparse graph is constructed; The spatial convolution module is used to perform spatial encoding on the dynamic sparse graph to obtain a plurality of edge-level embedding features; The time sequence position encoding is performed on the plurality of edge-level embedding features to obtain time sequence embedding features, and the time sequence information is extracted from the time sequence embedding features using the time sequence attention to obtain space-time embedding features; The prediction module is used to perform mode prediction based on the space-time embedding features to obtain the target mode to be switched and the corresponding confidence value.

3. The method of claim 2, wherein, The mode switching is performed based on the following process: Each slave node broadcasts a switching request to the adjacent nodes; After reaching a distributed consensus with the adjacent nodes, a mode double buffering mechanism is started, the current mode communication channel is closed after buffering, the new mode is activated, and a confirmation message of switching completion is broadcasted; The communication quality in the target mode is monitored using a post-evaluation module. If the communication quality deteriorates, the current mode is returned.

4. The method of claim 3, wherein, The double buffering mechanism is that the current mode and the target mode are parallel for a preset time.

5. The method of claim 1, wherein, If the confidence value of the target mode is greater than a preset threshold and the preset time is continued, it is decided to trigger the target mode switching, otherwise the current mode is maintained.

6. The method according to any one of claims 1 to 5, characterized in that, The communication modes include a cooperative mode, a task execution mode, an anti-interception mode and an emergency evacuation mode.

7. The method of claim 6, wherein, Samples are labeled in modes based on the trigger conditions of the modes. The samples labeled in modes are input to train the graph neural network; the trigger conditions include entering the anti-interception mode when strong electromagnetic interference or enemy detection signal is detected; the task execution mode is triggered when the target position is reached and the task start instruction is received; The emergency evacuation mode is triggered when the task is abnormal or the evacuation instruction is received.

8. The method of claim 1, wherein, The pre-set communication parameters include modulation and coding, time slot allocation strategy, transmission power level and routing protocol.

9. A wireless broadband ad hoc network communication mode switching system based on a cluster unmanned scenario, characterized in that, It comprises: A mode setting module for pre-setting multiple communication modes of the ad hoc network in the task life cycle of the swarm unmanned system; An awareness module for collecting multi-modal perception data; A mode decision module is configured to perform target communication mode prediction based on the multi-modal perception data by using a pre-trained graph neural network, to obtain a target mode to be switched and a corresponding confidence value; Decide whether to trigger mode switching based on the confidence value; A performing module is configured to perform mode switching according to the decision of triggering switching; A mode database and configuration management module is configured to store preset communication parameters of each communication mode and to configure communication parameters during switching.

10. The system of claim 9, wherein, The pre-trained graph neural network comprises a spatial convolution module, a time attention and a prediction module; the multi-modal perception data comprises task phase encoding, speed, relative position and wireless network communication link quality of the slave node itself; Obtaining the target communication mode to be switched and the corresponding confidence value comprises: Taking each of the slave nodes as a graph node, corresponding speed and relative position as node attributes, and the link quality as edge attributes, a dynamic sparse graph is constructed; The dynamic sparse graph is spatially encoded by using the spatial convolution module to obtain a plurality of edge-level embedding features; The plurality of edge-level embedding features are time-positionally encoded to obtain time embedding features, time information is extracted from the time embedding features by using time attention, and space-time embedding features are obtained; The prediction module is used to perform mode prediction based on the space-time embedding features, to obtain the target mode to be switched and the corresponding confidence value.