An unmanned aerial vehicle cluster distributed intelligent cooperative decision-making method based on a virtual phased array

Abstract

The application belongs to the technical field of unmanned aerial vehicle cluster control, and discloses a kind of unmanned aerial vehicle cluster distributed intelligent cooperative decision-making method based on virtual phased array, the method comprises: deploying unmanned aerial vehicle cluster carrying millimeter wave subarray, real-time acquisition unmanned aerial vehicle individual local state information and generation formation global state information;By reasoning global state, obtain master-slave election probability distribution and beam configuration parameter;Combined probability distribution and local state comprehensive score, select temporary master node according to score sorting;According to beam configuration parameter, the beam pointing of each millimeter wave subarray is adjusted in cooperation, and the virtual phased array is formed by coherent synthesis signal;Collect virtual phased array radiation capacity feedback, real-time detection master node heartbeat signal;When master node does not respond, re-election new master node and synchronizes beam parameter, completes virtual phased array reorganization.The application realizes unmanned aerial vehicle node autonomous decision and dynamic master-slave election, and improves the advantage of cluster invulnerability.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control technology, and in particular to a distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array. Background Technology

[0002] Phased array networking of UAV swarms will become the underlying core infrastructure for low-altitude economy, UAV swarms, and security and emergency response. Currently, there are two main types of phased array networking schemes for UAV swarms: one is to equip a single UAV with a large-scale phased array antenna, achieving long-distance, high-bandwidth communication through the large array element antenna of a single UAV; the other is to adopt a centralized control architecture, setting up a central node UAV, which uniformly schedules the phased array beams and communication links of all subordinate nodes to achieve swarm collaboration.

[0003] However, the two traditional solutions mentioned above have obvious technical flaws and are difficult to meet the practical application requirements of large-scale drone swarms: 1. The solution of using a single drone equipped with a large-scale phased array antenna has the problems of high cost and difficult deployment. Large-scale phased array antennas (usually containing dozens or even hundreds of array elements) are large in size, heavy in weight, and consume a lot of power, making them difficult to adapt to small drones. Moreover, the cost of a single array element is high, and the overall cost increases exponentially when deployed on a large scale. 2. Centralized control architecture suffers from poor resilience and insufficient scalability. The central node is the core of the entire cluster network; if the central node fails, the entire cluster network system will be completely paralyzed, unable to continue collaborative operations. In addition, centralized control requires the central node to process the scheduling information of all nodes. When the number of drones reaches more than 100, the computing and communication bandwidth pressure on the central node will increase sharply, easily leading to scheduling delays, link congestion, and other problems, making it impossible to achieve efficient collaboration in large-scale formations.

[0004] From the perspective of technological development trends, the distributed intelligence paradigm is gradually replacing the traditional centralized control paradigm. Combining distributed intelligence with phased array technology to achieve autonomous and collaborative networking of UAV swarms has become a key direction for breaking through traditional technological bottlenecks. Summary of the Invention

[0005] This invention provides a distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array, which enables autonomous decision-making and dynamic master-slave election of UAV nodes, improves the swarm's resilience, and allows for automatic re-election when the master node fails, ensuring continuous and uninterrupted decision-making and supporting low-latency and high-reliability collaborative effects for large-scale UAV swarms.

[0006] To achieve the above objectives, this invention provides a distributed intelligent cooperative decision-making method for unmanned aerial vehicle (UAV) swarms based on a virtual phased array, comprising: S1: Deploy a drone swarm, collect local status information of individual drones in the drone swarm in real time, generate global status information of the drone swarm based on the local status information, wherein each drone is equipped with a millimeter-wave subarray, and the millimeter-wave subarrays equipped by the drone swarm form an initial virtual phased array. S2: By reasoning the global state information through the distributed negotiation module built into the individual UAV, the master-slave election probability distribution and beam configuration parameters of the individual UAV are obtained; S3: Combining the master-slave election probability distribution and the local state information, a comprehensive score is given to each individual drone to obtain the election score result of the individual drone. According to the order of the election score results and the preset fixed term, the individual drones are selected as the temporary master nodes of the drone cluster in sequence. S4: Based on the beam configuration parameters, coordinately adjust the beam direction of the millimeter-wave subarray carried by the individual UAV to obtain the cooperative beam of the individual UAV. Based on the cooperative beam, coherently synthesize the signals of all the millimeter-wave subarrays in the UAV cluster to form an equivalent virtual phased array. S5: Collect the real-time radiation capability of the virtual phased array, obtain the radiation capability feedback of the virtual phased array, and detect the heartbeat signal of the individual UAV to which the temporary master node belongs in real time through the radiation capability feedback; S6: When the heartbeat signal is unresponsive, a new master node of the UAV cluster is selected and the beam configuration parameters are sent out synchronously to obtain a recombined virtual phased array.

[0007] Preferably, the real-time acquisition of local state information of individual drones in the drone cluster, and the generation of global state information of the drone cluster based on the local state information, includes: The system collects the three-dimensional position coordinates and three-dimensional velocity vector of the individual UAV in real time, as well as the transmit power, signal phase, beam weight, subarray power consumption, and channel signal-to-noise ratio of the millimeter-wave subarray, and encapsulates them into local state information. The local state information is normalized to obtain the normalized local state vector of the individual UAV. The normalized local state vector is broadcast in real time to all other drones in the drone cluster via the distributed communication link onboard the individual drone. Each of the aforementioned drones receives and aggregates the normalized local state vectors of all other drones, and concatenates them to form the global state information representing the state of all individuals in the drone cluster.

[0008] Preferably, the step of inferring the global state information through the distributed negotiation module built into the individual UAV to obtain the master-slave election probability distribution and beam configuration parameters of the individual UAV includes: The global state information is inferred based on the lightweight dual-branch deep neural network model of the distributed negotiation module in the individual UAV. The first output branch of the lightweight dual-branch deep neural network model outputs the probability distribution of master and slave nodes. The second output branch of the lightweight dual-branch deep neural network model outputs a beam weight vector and a signal phase vector, which together constitute the beam configuration parameters.

[0009] Preferably, the step of combining the master-slave election probability distribution and the local state information to comprehensively score each individual drone and obtain the election score result of the individual drone includes: Based on the master-slave election probability distribution, the channel signal-to-noise ratio in the local state information, the power consumption of the subarray, and the equivalent omnidirectional radiation power of the initial virtual phased array, a multi-objective joint optimization reward function is constructed. The scalar reward value for each individual UAV is calculated using the multi-objective joint optimization reward function, and the scalar reward value is the election score result.

[0010] Preferably, the step of sequentially selecting individual drones as temporary master nodes of the drone swarm according to the order of the election scores and a preset fixed term includes: The election score sequence of the drone cluster is obtained by sorting all the individual drones in the drone cluster from high to low according to the election score results. The drone individual ranked first in the election scoring sequence is selected as the temporary master node; A fixed term is set for the temporary master node. After the fixed term expires, a new round of election is automatically triggered according to the election scoring sequence.

[0011] Preferably, the step of coordinating the beam pointing of the millimeter-wave subarray carried by the individual UAV according to the beam configuration parameters to obtain the cooperative beam of the individual UAV includes: Based on the beam weight vector in the beam configuration parameters, the beam weight of the millimeter-wave subarray carried by the UAV is adjusted to obtain the beam adjustment weight of the UAV. Based on the signal phase vector in the beam configuration parameters, the signal phase of the millimeter-wave subarray carried by the UAV is adjusted to obtain the signal adjustment phase of the UAV. The beam adjustment weights and the signal adjustment phases are packaged into the cooperative beam of the individual UAV.

[0012] Preferably, the step of coherently combining the signals of all the millimeter-wave subarrays within the UAV cluster based on the cooperative beam to form an equivalent virtual phased array includes: Based on the beam adjustment weight and signal adjustment phase of the cooperative beam, the millimeter-wave subarrays carried by all the individual UAVs are directionally radiated to obtain electromagnetic waves from the millimeter-wave subarrays pointing towards the target airspace. Based on the superposition characteristics of the electromagnetic waves when propagating in space, all the electromagnetic waves in the target spatial direction are spatially coherently superimposed to obtain the enhanced composite signal of the millimeter wave subarray in the target spatial direction. Based on the spatial convergence effect of the synthesized signal, the original dispersed transmission power of all the millimeter-wave subarrays in the UAV cluster is spatially coherently synthesized to obtain the equivalent large-aperture radiation source of the UAV cluster in the direction of the target airspace, and the large-aperture radiation source is the virtual phased array.

[0013] Preferably, the step of acquiring the real-time radiation capability of the virtual phased array and obtaining the radiation capability feedback of the virtual phased array includes: The power detection unit on each of the aforementioned UAVs is used to collect the radiated power density of the virtual phased array in the target direction in real time. Based on the radiation power density and the spatial location information of the UAV cluster, the actual equivalent isotropic radiation power of the virtual phased array is calculated, and the actual equivalent isotropic radiation power is the radiation capability feedback.

[0014] Preferably, the step of detecting the heartbeat signal of the individual UAV to which the temporary master node belongs in real time through the radiation capability feedback includes: The radiation capability feedback is compared with the radiation capability threshold of the individual drone under normal operation to obtain the radiation capability comparison result of the individual drone. Based on the radiation capability comparison results, the heartbeat signal of the individual UAV corresponding to the temporary master node is determined.

[0015] Preferably, when the heartbeat signal is unresponsive, a new master node is selected for the UAV cluster, and the beam configuration parameters are simultaneously sent out to obtain a reassembled virtual phased array, including: When the radiation capability feedback remains below the normal operating threshold for an extended period exceeding the preset timeout duration, it is determined that the heartbeat signal of the individual drone corresponding to the temporary master node is unresponsive. The individual drone re-collects the local state information, infers the new master-slave election probability distribution through the distributed negotiation module, and selects a new master node based on the new master-slave election probability distribution; Based on the new master node, the latest beam configuration parameters are synchronously sent to all other drones in the drone cluster; All the individual drones, according to the issued beam configuration parameters, readjust the cooperative beams of the individual drones and perform coherent synthesis to form the recombined virtual phased array.

[0016] Beneficial effects This solution utilizes distributed collaborative decision-making and virtual large-scale phased array synthesis technology, offering advantages such as low cost, lightweight design, and ease of deployment. It is adaptable to small UAVs while achieving a communication EIRP comparable to that of a large physical phased array. The system supports large-scale UAV swarms, exhibiting low decision latency and high communication speeds, thus meeting the information exchange and data transmission needs of large-scale swarm collaborative operations.

[0017] The system in this solution adopts a distributed architecture without a central node. Each UAV node collaborates on an equal footing, with no single point of failure. Even if some UAVs malfunction (crash or be interfered with), the remaining UAVs can still reassemble the virtual phased array through distributed negotiation and continue to complete collaborative operations.

[0018] This solution optimizes the master-slave node and scheduling strategy in real time based on channel signal-to-noise ratio, power consumption, and EIRP virtual phased array communication, achieving adaptive optimal networking. Compared with fixed master-slave, fixed scheduling, and fixed array configurations, it has higher throughput and lower power consumption. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a distributed intelligent collaborative decision-making method for unmanned aerial vehicle (UAV) swarms based on a virtual phased array, provided as an embodiment of the present invention.

[0020] Figure 2 This invention provides a distributed reinforcement learning dual-branch network model structure in a distributed intelligent collaborative decision-making method for unmanned aerial vehicle (UAV) swarms based on a virtual phased array, as an embodiment of the present invention.

[0021] Figure 3 This is a hardware block diagram of a single UAV in a distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array, provided as an embodiment of the present invention.

[0022] Figure 4 This is a system framework diagram of a distributed intelligent collaborative decision-making method for unmanned aerial vehicle (UAV) swarms based on a virtual phased array, provided in an embodiment of the present invention.

[0023] Figure 5This invention provides a method for distributed intelligent collaborative decision-making of unmanned aerial vehicle (UAV) swarms based on a virtual phased array, which includes the hardware connection relationship of the UAV swarm.

[0024] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0025] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0026] This application provides a distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0027] Reference Figure 1 The diagram shown is a flowchart illustrating a distributed intelligent cooperative decision-making method for UAV swarms based on a virtual phased array, according to an embodiment of the present invention. In this embodiment, the distributed intelligent cooperative decision-making method for UAV swarms based on a virtual phased array includes: S1: Deploy a drone swarm, collect local status information of individual drones in the drone swarm in real time, generate global status information of the drone swarm based on the local status information, wherein each drone is equipped with a millimeter-wave subarray, and the millimeter-wave subarrays equipped by the drone swarm form an initial virtual phased array. In this embodiment, the real-time acquisition of local state information of individual drones in the drone cluster, and the generation of global state information of the drone cluster based on the local state information, includes: The system collects the three-dimensional position coordinates and three-dimensional velocity vector of the individual UAV in real time, as well as the transmit power, signal phase, beam weight, subarray power consumption, and channel signal-to-noise ratio of the millimeter-wave subarray, and encapsulates them into local state information. The local state information is normalized to obtain the normalized local state vector of the individual UAV. The normalized local state vector is broadcast in real time to all other drones in the drone cluster via the distributed communication link onboard the individual drone. Each of the aforementioned drones receives and aggregates the normalized local state vectors of all other drones, and concatenates them to form the global state information representing the state of all individuals in the drone cluster.

[0028] Specifically, the three-dimensional position coordinates are the position parameters of an individual UAV in a three-dimensional spatial coordinate system. They are obtained in real time by the airborne positioning module and are used to characterize the spatial position of a single UAV in the formation airspace. They include coordinate values ​​in three orthogonal directions.

[0029] The three-dimensional velocity vector is the velocity parameter of an individual UAV in a three-dimensional spatial coordinate system. It is calculated by the onboard velocity measurement module in combination with positioning data and is used to characterize the motion direction and speed of a single UAV. It includes velocity components in three orthogonal directions.

[0030] Millimeter-wave subarrays are small millimeter-wave antenna arrays mounted on each individual UAV, possessing independent signal transmission and reception capabilities and beam control capabilities. They are the basic radiating units that constitute virtual phased arrays.

[0031] Transmit power is the power value of electromagnetic wave signals radiated outward by the millimeter-wave subarray. It is acquired in real time by the power detection unit built into the subarray and is used to characterize the signal radiation intensity of a single subarray.

[0032] Signal phase is the phase value of the electromagnetic wave signal radiated by the millimeter-wave subarray. It is acquired in real time by the phase detection unit built into the subarray and is used to characterize the phase state of the subarray output signal. It is the core parameter for realizing coherent synthesis of spatial signals.

[0033] Beam weight is the amplitude weighting coefficient of the radiation signal of each element of the millimeter-wave subarray. It is read in real time by the subarray control unit and is used to adjust the direction and shape of the subarray beam.

[0034] Subarray power consumption is the total electrical power consumed by the millimeter-wave subarray when it is in operation. It is collected in real time by the detection unit of the subarray power supply circuit and is used to characterize the energy consumption level of a single subarray.

[0035] The channel signal-to-noise ratio (SNR) is the ratio of useful signal power to noise power in an individual UAV communication link. It is obtained in real time by the onboard communication module and is used to characterize the quality of communication links between UAVs.

[0036] Local state information is a dataset obtained by packaging all state parameters collected by a single drone in a fixed format. It only contains the motion state of the corresponding single drone and the working parameters of the subarray.

[0037] Normalization is a process of linearly transforming state parameters with different dimensions and numerical ranges according to the upper and lower limits of the physical range of the corresponding parameters. It is used to eliminate the dimensional differences between different parameters and make all parameters fall within a unified numerical range.

[0038] Normalized local state vectors are one-dimensional vector data formed by normalizing the local state information of a single UAV and arranging it in a fixed order. They have a unified numerical range and fixed dimensions and can be directly used for cross-node data interaction.

[0039] Distributed communication links are peer-to-peer communication network links deployed on each individual drone. Each drone node is equal in status and can send and receive data without relying on a central node, supporting the exchange of status information within the cluster.

[0040] Broadcasting is a communication operation in which a single UAV synchronously sends its normalized local state vector to all other UAVs within the formation communication range via a distributed communication link.

[0041] Aggregation is an operation in which a single drone summarizes, verifies, and caches the normalized local state vectors received from all other drones, and is used to gather the state data of all individuals in the formation.

[0042] Global state information is a complete dataset formed by concatenating the normalized local state vector of a single UAV with the normalized local state vectors of all other UAVs in a fixed order. It can represent the operational status of all individuals in the entire UAV swarm at the current moment.

[0043] Furthermore, the individual drone activates its onboard positioning module, speed measurement module, and various detection units of the millimeter-wave subarray to synchronously collect its own motion state parameters and the working parameters of the millimeter-wave subarray at a unified sampling frequency.

[0044] The motion state parameters are continuously output by the positioning and velocity measurement module through spatial positioning calculations, while the operating parameters of the millimeter-wave subarray are directly read by each detection unit within the subarray. After all parameters are collected, they are packaged and encapsulated according to a pre-agreed data format to form local state information specific to the drone.

[0045] During the data acquisition process, the sampling times of all parameters are kept aligned to ensure that the data within the same set of local status information has temporal consistency.

[0046] Each drone will linearly transform the parameters in its encapsulated local state information according to its own upper and lower limits of physical range.

[0047] Position and velocity spatial motion parameters are mapped to their upper and lower limits according to their own range. Within a given range, parameters such as power, phase, weight, power consumption, and signal-to-noise ratio are mapped to the upper limit of their respective ranges. Interval.

[0048] After all parameters have been processed, they are arranged and combined according to a pre-agreed parameter order to form a normalized local state vector with fixed dimensions and a uniform numerical range.

[0049] During the processing, the mapping logic and arrangement order of various parameters are kept globally consistent within the formation to ensure that the vectors output by different UAVs have a consistent data format, enabling direct cross-node interaction.

[0050] A single drone invokes its onboard distributed communication link to send its generated normalized local state vector, along with the drone's identity identifier and the timestamp of the current sampling time, omnidirectionally to all other drones within the formation's communication coverage area.

[0051] The broadcasting process is continuously executed according to a fixed transmission cycle. Each broadcast carries the latest state vector data to ensure that other drones in the formation can obtain the latest operating status of the drone in real time.

[0052] While broadcasting its own state data, each drone continuously listens for and receives the normalized local state vectors broadcast by all other drones in the formation through a distributed communication link.

[0053] The drone first verifies the identity and time sequence validity of each set of vector data it receives, and then summarizes and stores all valid vector data after removing invalid data.

[0054] Subsequently, the UAV concatenates and combines its own normalized local state vector with the state vectors of the other UAVs obtained from the aggregation, according to a unified identity number order, to finally form global state information that can cover the operating status of all individuals in the entire formation.

[0055] For example, after all drones are started, they enter autonomous flight mode, maintaining a preset initial formation spacing (5-10m). Real-time collection of local status information from drone nodes is also performed. Indicates the location of the drone node itself, Indicates the speed of the drone node itself, This indicates the transmit power of the millimeter-wave subarray of the UAV node. Indicates the phase of the subarray. Indicates the subarray beam weights. Indicates the power consumption of the subarray. This indicates the channel signal-to-noise ratio.

[0056] The drone collects its own 11-dimensional state information in real time through a local perception module, performs normalization preprocessing on the local state vector, and maps physical quantities such as position and velocity to... The interval maps physical quantities such as power, power consumption, phase, and signal-to-noise ratio to... The interval is used to eliminate dimensional differences and meet the model input requirements.

[0057] The drone broadcasts its preprocessed local state vector to all drones in the cluster in real time via an onboard distributed communication link. Each drone receives and aggregates the local state vectors of the other 99 drones, forming a 1100-dimensional global state vector corresponding to 100 drones. This achieves real-time synchronization of global state information, providing a unified input basis for distributed collaborative decision-making.

[0058] S2: By reasoning the global state information through the distributed negotiation module built into the individual UAV, the master-slave election probability distribution and beam configuration parameters of the individual UAV are obtained; In this embodiment, the step of inferring the global state information through the distributed negotiation module built into the individual UAV to obtain the master-slave election probability distribution and beam configuration parameters of the individual UAV includes: The global state information is inferred based on the lightweight dual-branch deep neural network model of the distributed negotiation module in the individual UAV. The first output branch of the lightweight dual-branch deep neural network model outputs the probability distribution of master and slave nodes. The second output branch of the lightweight dual-branch deep neural network model outputs a beam weight vector and a signal phase vector, which together constitute the beam configuration parameters.

[0059] Specifically, the distributed negotiation module is an edge computing processing unit deployed on each individual UAV. It is interconnected with the UAV flight control system, millimeter-wave subarray control unit, and airborne distributed communication unit through an onboard high-speed interface. It has functions such as state data preprocessing, decision-making, parameter distribution, and fault self-healing detection. It can complete collaborative decision-making within the cluster without a central node. The distributed negotiation module adopts a decentralized edge-side online distributed reinforcement learning training mechanism to construct a joint loss function composed of master-slave election cross-entropy loss, beam weight mean square error loss, and signal phase mean square error loss. Each UAV independently performs forward inference and backward propagation, and asynchronously updates the model parameters.

[0060] The lightweight dual-branch deep neural network model is a decision-making operation structure embedded in the storage unit of the distributed negotiation module. It contains two parallel output paths, and its overall computational and parameter quantities are adapted to the computing power, power consumption, and latency constraints of airborne edge computing chips. It is used to convert global state information into cluster collaborative decision parameters.

[0061] The first output branch is the output path responsible for master-slave node election decision in the lightweight dual-branch deep neural network model. It corresponds to the computational output path for cluster node role allocation and is used to output the master node adaptation parameters of each drone.

[0062] The master-slave node probability distribution is a numerical sequence that characterizes the suitability of each drone in the cluster to be selected as the master node. Each value in the sequence corresponds one-to-one with a drone. The higher the value, the more suitable the drone is to assume the scheduling responsibility of a temporary master node.

[0063] The second output branch is the output path responsible for configuring the virtual phased array beam parameters in the lightweight dual-branch deep neural network model. It corresponds to the output path for calculating the radiation parameters of the millimeter-wave subarray and is used to output the control parameters of each UAV subarray.

[0064] The beam weight vector is a numerical sequence that represents the weighting coefficients of the radiation signal amplitude of each UAV millimeter-wave subarray within the cluster. Each value in the sequence corresponds one-to-one with a UAV millimeter-wave subarray and is used to adjust the radiation intensity and beam shape of the subarray, supporting the beam pointing control of the virtual phased array.

[0065] The signal phase vector is a numerical sequence that characterizes the phase value of the radiated signal of each UAV millimeter-wave subarray within the cluster. Each value in the sequence corresponds one-to-one with a UAV millimeter-wave subarray and is used to adjust the phase state of the subarray radiated signal to ensure that all subarray signals can be coherently superimposed in the target airspace.

[0066] Beam configuration parameters are a set of parameters used to control the radiation state of millimeter-wave subarrays. They include two types of parameters: beam weight and signal phase. They can be directly sent to the millimeter-wave subarray control unit to drive the subarray to complete the adjustment of beam pointing and radiation phase.

[0067] Furthermore, after individual drones complete the splicing and aggregation of global state information during swarm flight, the onboard distributed negotiation module calls the internally fixed dual-path decision processing unit to process the global state information layer by layer.

[0068] First, perform a linear dimension mapping on the global state data to transform the high-dimensional cluster state data into intermediate-dimensional feature data.

[0069] Then, perform time-series correlation feature extraction on the feature data of the intermediate dimension to capture the correlation features of changes in the cluster state at different times.

[0070] The extracted features are then compressed to further reduce feature dimensionality while retaining core decision information.

[0071] Finally, an optimization mapping is performed on the compressed features to obtain feature data that meets the requirements of the decision output.

[0072] After feature processing is completed, the first output path of the dual-path decision processing unit generates a master-slave node probability distribution covering all UAVs in the cluster. Each value in the master-slave node probability distribution corresponds one-to-one with a UAV, representing the degree of suitability of the corresponding UAV to assume the role of temporary master node, and providing a quantitative basis for subsequent cluster master-slave node election.

[0073] The second output path of the dual-path decision processing unit synchronously generates beam weight vectors and signal phase vectors covering all UAVs in the cluster. Each value in the beam weight vector corresponds to the radiation amplitude weighting coefficient of a UAV millimeter-wave subarray, and each value in the signal phase vector corresponds to the radiation signal phase value of a UAV millimeter-wave subarray. The two types of vectors together form beam configuration parameters that can be directly sent to the millimeter-wave subarray control unit, providing a control basis for the subsequent spatial coherent synthesis of the virtual phased array.

[0074] The model described in this invention employs an online training mechanism, which does not rely on a pre-set offline dataset. The training samples are generated in real time by a cluster of drones, forming an online training set.

[0075] The training set consists of the local state information of all nodes within the drone swarm: In the formula, Indicates the first Normalized local state information set of each UAV node subarray .

[0076] The distributed negotiation module incorporates a lightweight dual-branch deep neural network model, which is a fully connected temporal feature fusion structure, consisting of an input layer, a linear mapping layer, a long short-term memory (LSTM) temporal feature extraction layer, a feature compression layer, a feature mapping layer, and dual output branches.

[0077] The global state vector is input into the model. A linear mapping layer maps the 1100-dimensional features to 128 dimensions. An LSTM layer extracts temporal correlation features, a feature compression layer compresses them to 64 dimensions, and a feature mapping layer optimizes the feature representation. The output is then processed by two parallel branches: the first branch, activated by Softmax, outputs a 100-dimensional master-slave node probability distribution vector; the second branch, activated by Sigmoid, outputs a 100-dimensional beam weight vector and a 100-dimensional signal phase vector. This invention employs a distributed reinforcement learning dual-branch network model to achieve joint optimization of master-slave election, beam weighting, and signal phase allocation, such as... Figure 2 As shown.

[0078] The model input dimension of 1100 corresponds to 11 states of 100 drones, and uses a lightweight hidden layer of 128 / 64 to balance expressive power, inference latency and onboard computing power limitations.

[0079] Input layer: Drone swarm state sequence .

[0080] Linear mapping layer: Linear (1100→128), activation function ReLU.

[0081] Temporal feature extraction layer: LSTM (128→128), capturing temporal correlations of states. Feature compression layer: Linear (128→64), activation function ReLU.

[0082] Feature mapping layer: Linear (64→64), activation function Tanh.

[0083] Branch 1 (Master-Slave Election): After Linear mapping and softmax normalization, output the master-slave node probability distribution vector (dimensions). ), to complete dynamic master-slave election; Branch 2 (Virtual Phased Array Beam Weights and Signal Phase Allocation): After Linear mapping and Sigmoid normalization, the output beam weights and signal phase distribution vectors are obtained to maximize the EIRP of the virtual phased array.

[0084] Model outputs: master-slave election results, optimal beam weights, and optimal signal phase.

[0085] The ReLU mentioned above is the Modified Linear Unit Activation Function. Its operation rule is to directly output the input value when the input value is greater than 0, and to output 0 when the input value is less than or equal to 0. This function has low computational complexity, which can improve network computation speed and alleviate the gradient vanishing problem in deep networks. It is typically deployed in the feature transformation layer in the middle of the network to perform non-linear filtering on features, retaining positive and effective features. In this scheme, ReLU is deployed in the linear mapping layer and the feature compression layer, performing non-linear activation on the features after the cluster state mapping, strengthening the effective feature representation related to collaborative decision-making.

[0086] Tanh is the hyperbolic tangent activation function, which nonlinearly maps any input value to... The output values ​​are centered around 0, ensuring a balanced discriminative power between positive and negative features. This function is typically deployed in the feature mapping layer, performing a normalized nonlinear transformation on the compressed features to ensure the feature values ​​lie within a symmetrical, regular interval, thus improving the stability of subsequent decision outputs. In this scheme, Tanh is deployed in the feature mapping layer, mapping the compressed cluster decision features to... The interval optimizes the balance of feature representation, providing regular feature inputs for the decision output of the two branches.

[0087] Sigmoid is a sigmoid nonlinear activation function that can nonlinearly map any input value to... The output value represents a continuous physical quantity such as proportion or intensity within a given range. This function is typically deployed in the continuous value output branch, normalizing the raw values ​​obtained from network computation to a range that conforms to actual physical constraints. In this scheme, the Sigmoid function is deployed in the second output branch, mapping the raw values ​​of beam weights and signal phase obtained from network computation to... The range is matched with the beam weight value range of 0 to 1 and the normalized signal phase value range, and the output is a continuous control parameter that can be directly sent to the millimeter-wave subarray control unit.

[0088] Softmax is a normalized exponential activation function. During its operation, it takes the natural exponent of each element in the input vector, and then normalizes the output vector by using the sum of all exponents as the denominator, ensuring that all elements in the output vector take values ​​within the normal range. Furthermore, the sum of all element values ​​is 1, which can represent the probability distribution in multi-category scenarios. This function is deployed in the classification decision output branch, converting the network computation results into the probability distribution of each option, adapting to multiple-choice decision scenarios. In this scheme, Softmax is deployed in the first output branch, converting the raw master-slave election values ​​obtained from the network computation into the master node probability distribution of all drones in the cluster. The sum of the probability values ​​of all drones is 1. The higher the probability, the more suitable the corresponding drone is for the temporary master node role, providing a quantitative basis for dynamic master-slave election in the cluster.

[0089] S3: Combining the master-slave election probability distribution and the local state information, a comprehensive score is given to each individual drone to obtain the election score result of the individual drone. According to the order of the election score results and the preset fixed term, the individual drones are selected as the temporary master nodes of the drone cluster in sequence. In this embodiment, the step of combining the master-slave election probability distribution and the local state information to comprehensively score each individual drone and obtain the election score result of the individual drone includes: Based on the master-slave election probability distribution, the channel signal-to-noise ratio in the local state information, the power consumption of the subarray, and the equivalent omnidirectional radiation power of the initial virtual phased array, a multi-objective joint optimization reward function is constructed. The scalar reward value for each individual UAV is calculated using the multi-objective joint optimization reward function, and the scalar reward value is the election score result.

[0090] Specifically, the channel signal-to-noise ratio (SNR) is the ratio of useful signal power to noise power in the airborne communication link of an individual UAV. It is obtained in real time by the airborne communication module and is used to reflect the quality of the communication link between an individual UAV and other UAVs in the formation.

[0091] Subarray power consumption is the total electrical power consumed by the millimeter-wave subarray carried by an individual UAV when it is in operation. It is collected in real time by the detection unit of the subarray power supply circuit and is used to reflect the energy consumption level of a single UAV millimeter-wave subarray.

[0092] The equivalent omnidirectional radiation power of a virtual phased array is the equivalent omnidirectional radiation power in the target direction after the signals of all millimeter-wave subarrays in the UAV swarm are spatially coherently combined. The value is used to reflect the signal radiation intensity and communication coverage capability of the virtual phased array in the target direction.

[0093] The multi-objective joint optimization reward function is a quantitative calculation rule that integrates multi-dimensional evaluation indicators such as communication quality, energy consumption level, and cluster radiation performance. It is used to convert state parameters of different dimensions into a unified and comparable score value and provides a quantitative judgment basis for the election of UAV master nodes.

[0094] The scalar reward value is a single numerical result obtained by multi-objective joint optimization of the reward function, which corresponds to the overall adaptation degree of the master node of a single UAV.

[0095] The election score is the final scoring basis for the drone master node ranking election. It corresponds one-to-one with the scalar reward value of a single drone. The higher the score, the higher the priority of the drone in the master node election.

[0096] Furthermore, the input parameters for four evaluation dimensions are first determined: master-slave election fit, channel signal-to-noise ratio, subarray power consumption, and equivalent isotropic radiated power.

[0097] Then, a corresponding weight coefficient is set for each evaluation dimension. The sum of the weight coefficients for all dimensions is 1. The weight coefficients are used to adjust the proportion of influence of different evaluation dimensions in the overall score.

[0098] Next, the input parameters are normalized. The channel signal-to-noise ratio is based on the maximum channel signal-to-noise ratio in the current drone cluster. The channel signal-to-noise ratio of a single drone is converted into a relative value in the range of 0 to 1. The higher the value, the better the communication quality.

[0099] The power consumption of the subarray is based on the maximum allowable power consumption of the millimeter-wave subarray. The power consumption of a single UAV's subarray is converted into an inverse relative value in the range of 0 to 1. The lower the power consumption, the higher the corresponding value.

[0100] The equivalent isotropic radiation power is based on the theoretical maximum equivalent isotropic radiation power of the virtual phased array. The actual equivalent isotropic radiation power is converted into a relative value in the range of 0 to 1. The higher the value, the stronger the cluster radiation performance.

[0101] Finally, the normalized values ​​of each item are multiplied by their corresponding weight coefficients and summed to form a complete multi-objective joint optimization reward function calculation rule.

[0102] For each individual drone in the drone swarm, extract the master-slave election adaptation value, the normalized channel signal-to-noise ratio value, the normalized reverse power consumption value, and the normalized equivalent isotropic radiated power value corresponding to that drone, and substitute these values ​​into the calculation rules of the multi-objective joint optimization reward function.

[0103] The numerical calculation is performed according to the weighted summation rules set in the reward function to obtain the scalar reward value for the corresponding individual drone. This scalar reward value is directly used as the election score result for the corresponding individual drone, and the election scores of all drones can be directly used for subsequent sorting comparisons and selection of temporary master nodes.

[0104] In the decision-making process, to achieve decentralized distributed autonomous collaboration, this invention designs a reward function based on multi-objective joint optimization. The core evaluation dimensions are the communication quality, power consumption level, and overall cluster communication EIRP of the UAV nodes. This quantifies the suitability of each node as a master node, forming a target guidance signal for the distributed collaboration strategy. The multi-objective joint optimization reward function is as follows:

[0105] in, Scoring for the master node election of the first drone. The channel signal-to-noise ratio of the first drone is... The maximum signal-to-noise ratio of the drones in the current drone swarm. For the subarray power consumption of the first drone, This represents the maximum allowable power consumption of the subarray. The equivalent isotropic radiated power for actual communication using a virtual phased array. This represents the theoretical maximum communication equivalent isotropic radiation power of a virtual phased array. For adaptive scoring weights ( ).

[0106] The reward function guides the collaborative strategy through goal quantification, action feedback, and dynamic weight adaptation. (1) Target Quantification Guidance: The reward function transforms the three cluster-level targets of high communication quality, low power consumption, and high cluster radiation power into scalar reward values ​​that can be calculated and compared for individual nodes. The signal-to-noise ratio term prompts the strategy to prioritize nodes with reliable communication links; the power consumption term prompts the strategy to select low-power nodes to extend the cluster's endurance; the EIRP term prompts the strategy to be guided by the optimal global network performance, avoiding individual optimization leading to global imbalance.

[0107] (2) Action feedback guidance: After the UAV outputs master-slave decision, beam weight and phase configuration actions based on local state, the environment returns reward values ​​in real time. Actions with excellent performance (such as high signal-to-noise ratio nodes being selected as master nodes, beam configuration improving EIRP) receive high rewards; actions with poor performance receive low rewards, forming an immediate evaluation of the strategy and guiding behavioral preferences.

[0108] (3) The formula for calculating the equivalent isotropic radiated power (EIRP) is as follows:

[0109] The meanings of each parameter are explained in detail below: Number of drone clusters (N=100), i.e., the total number of sub-arrays; : No. The transmit power (in W) of each miniaturized subarray is collected in real time by the subarray module. ); : No. Beam weights of individual subarrays (0≤ ≤1), dynamically allocated by the distributed negotiation module; : No. The signal phase (unit: rad) of each subarray is dynamically allocated by the distributed negotiation module; : No. The gain of a single element in a miniaturized subarray (unit: linear value, dimensionless) is an intrinsic parameter of the subarray (microstrip patch antenna). ); Coherent synthesis efficiency (considering losses such as phase synchronization error and multipath interference), ).

[0110] (4) Adaptive weight dynamic adaptation: The weights are dynamically updated iteratively using an online learning method. The update formula is:

[0111] In the formula, For the first Adaptive weights for each scoring metric For the first The incremental synergistic effect of each scoring indicator is calculated in real time by the model based on the master node scheduling efficiency corresponding to multiple weight matching schemes stored in history.

[0112] In this embodiment, the step of sequentially selecting individual drones as temporary master nodes of the drone cluster according to the order of the election scores and a preset fixed term includes: The election score sequence of the drone cluster is obtained by sorting all the individual drones in the drone cluster from high to low according to the election score results. The drone individual ranked first in the election scoring sequence is selected as the temporary master node; A fixed term is set for the temporary master node. After the fixed term expires, a new round of election is automatically triggered according to the election scoring sequence.

[0113] Specifically, the election scoring sequence is an ordered set of data formed by arranging the election scores of all drones in the formation from high to low. Each set of data in the set corresponds one-to-one with the identity identifier of a drone.

[0114] Temporary master nodes are individual drones elected from the drone formation and responsible for cluster communication scheduling and collaborative command forwarding for a fixed period of time.

[0115] The preset fixed term is a threshold for the effective duration of a single performance of duties for a temporary master node. Once this threshold is reached, the scheduling authority of the temporary master node will automatically expire.

[0116] The new round of elections is the process of all drones in the formation re-selecting a temporary master node after the fixed term expires.

[0117] Furthermore, the election score results for each individual drone in the drone cluster are obtained, and all drones are arranged in descending order of score based on the size of the score.

[0118] During the sorting process, the binding relationship between the identity of each drone and its corresponding score value is preserved simultaneously to ensure that there is no mismatch between identity and score in the sorted data, and finally an election score sequence covering all drones in the formation is formed.

[0119] Then, the generated election scoring sequence is read, the first scoring item in the sequence is located, the individual drone identity information corresponding to the item is extracted, the individual drone is identified as the temporary master node of the current formation, and the identity information of the temporary master node is broadcast to all individual drones in the formation through the airborne distributed communication link to clarify its scheduling authority during its term of office.

[0120] Finally, a fixed term of service is set for the determined temporary master node. During the term of service of the temporary master node, the onboard timing unit keeps a continuous timer. When the timer reaches the set fixed term threshold, the next temporary master node selection process is started directly based on the generated election scoring sequence without the need for additional external triggering instructions, thus completing the orderly rotation of master nodes.

[0121] In detail, the distributed negotiation module sorts all drones in the cluster in descending order based on the master-slave node probability distribution output by the model, selects the drone with the highest master node probability as the temporary master node, and the remaining drones as slave nodes; the master node is set to a fixed term of 2 seconds, and a new round of election is automatically triggered after the term expires, to avoid a single node occupying scheduling authority for a long time; the master node is only responsible for communication scheduling and collaborative command forwarding within the cluster, and does not interfere with the independent flight decisions of slave nodes, so that slave nodes maintain autonomous decision-making capabilities and realize decentralized distributed collaboration.

[0122] S4: Based on the beam configuration parameters, coordinately adjust the beam direction of the millimeter-wave subarray carried by the individual UAV to obtain the cooperative beam of the individual UAV. Based on the cooperative beam, coherently synthesize the signals of all the millimeter-wave subarrays in the UAV cluster to form an equivalent virtual phased array. In this embodiment, the step of coordinating the beam pointing of the millimeter-wave subarray carried by the UAV according to the beam configuration parameters to obtain the cooperative beam of the UAV includes: Based on the beam weight vector in the beam configuration parameters, the beam weight of the millimeter-wave subarray carried by the UAV is adjusted to obtain the beam adjustment weight of the UAV. Based on the signal phase vector in the beam configuration parameters, the signal phase of the millimeter-wave subarray carried by the UAV is adjusted to obtain the signal adjustment phase of the UAV. The beam adjustment weights and the signal adjustment phases are packaged into the cooperative beam of the individual UAV.

[0123] Specifically, the beam weight vector is a numerical sequence that represents the weighting coefficient of the radiation signal amplitude of each UAV millimeter-wave subarray within the cluster. Each value in the sequence corresponds one-to-one with a UAV millimeter-wave subarray and is used to adjust the radiation intensity and beam shape of the subarray, supporting the beam pointing control of the virtual phased array.

[0124] Individual drones are independent flight units that make up a drone swarm. They are equipped with millimeter-wave subarrays and airborne control units, and have independent flight control, signal transmission and reception, and status awareness capabilities.

[0125] Beam weighting is an amplitude weighting coefficient of the radiated signal of each element in a millimeter-wave subarray, used to control the direction and shape of the subarray beam.

[0126] The beam adjustment weight is the final beam weight value after the millimeter-wave subarray receives the control command and completes the amplitude weighting coefficient adjustment. It can be directly used to drive the subarray elements to perform signal radiation.

[0127] The signal phase vector is a numerical sequence that characterizes the phase value of the radiated signal of each UAV millimeter-wave subarray within the cluster. Each value in the sequence corresponds one-to-one with a UAV millimeter-wave subarray and is used to adjust the phase state of the subarray radiated signal to ensure that all subarray signals can be coherently superimposed in the target airspace.

[0128] The signal phase adjustment is the final signal phase value after the millimeter-wave subarray receives the control command and completes phase adjustment. It can be directly used to drive the subarray elements to perform signal radiation of the corresponding phase.

[0129] Cooperative beamforming is a radiation beam with unified control parameters formed by a single UAV millimeter-wave subarray after beam weighting and signal phase adjustment. It can work together with the beams of other subarrays in the formation to achieve spatial coherent synthesis.

[0130] Furthermore, the weight values ​​corresponding to the local millimeter-wave subarray are extracted from the beam weight vector in the beam configuration parameters. The weight values ​​are then distributed to each element according to the arrangement rules of the elements within the subarray, and converted into amplitude modulation electrical signals corresponding to each element. By adjusting the signal output amplitude of the radio frequency link of each element, the beam weight of the millimeter-wave subarray is adjusted, and finally, the beam adjustment weights adapted to the formation coordination requirements are obtained.

[0131] The phase values ​​corresponding to the local millimeter-wave subarray are extracted from the signal phase vector in the beam configuration parameters. The phase values ​​are then distributed to each element according to the arrangement rules of the elements within the subarray. The phase values ​​are converted into phase modulation electrical signals corresponding to each element. By adjusting the signal phase offset of the radio frequency link of each element, the phase of the millimeter-wave subarray signal is adjusted, and finally, the signal adjustment phase adapted to the formation coordination requirements is obtained.

[0132] The adjusted beam weights and signal phases are encapsulated and integrated according to the agreed data format to form a set of control parameters that can directly drive the synchronous radiation of all elements in the subarray, i.e., the cooperative beam of a single UAV, providing a unified radiation parameter benchmark for the spatial coherent synthesis of all subarrays in the subsequent formation.

[0133] In this embodiment, the step of coherently synthesizing the signals of all the millimeter-wave subarrays within the UAV cluster based on the cooperative beam to form an equivalent virtual phased array includes: Based on the beam adjustment weight and signal adjustment phase of the cooperative beam, the millimeter-wave subarrays carried by all the individual UAVs are directionally radiated to obtain electromagnetic waves from the millimeter-wave subarrays pointing towards the target airspace. Based on the superposition characteristics of the electromagnetic waves when propagating in space, all the electromagnetic waves in the target spatial direction are spatially coherently superimposed to obtain the enhanced composite signal of the millimeter wave subarray in the target spatial direction. Based on the spatial convergence effect of the synthesized signal, the original dispersed transmission power of all the millimeter-wave subarrays in the UAV cluster is spatially coherently synthesized to obtain the equivalent large-aperture radiation source of the UAV cluster in the direction of the target airspace, and the large-aperture radiation source is the virtual phased array.

[0134] Specifically, directional radiation is a radiation method in which a millimeter-wave subarray concentrates electromagnetic wave energy and emits it in a specified spatial direction according to its configuration parameters.

[0135] The target airspace direction is the preset spatial orientation corresponding to the mission executed by the UAV swarm, and is the target direction in which the energy of the virtual phased array beam converges.

[0136] Electromagnetic waves are the millimeter-wave frequency electromagnetic energy radiated outward by millimeter-wave subarrays, and are the physical carrier for realizing signal transmission and spatial coherent synthesis.

[0137] The superposition property is a physical characteristic in which the amplitudes of multiple electromagnetic waves of the same frequency band superimpose each other according to their phase relationship when they meet during the propagation of electromagnetic waves in space. It is the physical basis for achieving coherent synthesis.

[0138] Spatial coherent superposition is a physical process in which electric and magnetic field components superimpose each other according to phase relationship when multiple electromagnetic waves of the same frequency and phase match propagate in the target's spatial domain, thereby enhancing the signal amplitude in the target's direction.

[0139] The enhanced synthetic signal is a synthesized electromagnetic signal with significantly enhanced amplitude obtained by coherently superimposing the electromagnetic waves radiated by all millimeter-wave subarrays in the target spatial direction.

[0140] The spatial convergence effect is a phenomenon in which multiple electromagnetic waves, after coherent superposition, concentrate their energy in the target spatial direction, forming a spatial energy distribution with an energy density much higher than that of a single subarray radiation.

[0141] Spatial coherent superposition is a technique that utilizes the coherent superposition characteristics of electromagnetic waves to concentrate and superimpose the emission power of multiple dispersed radiation sources in the target direction, thereby effectively enhancing the overall radiation capability.

[0142] Transmit power is the power value of electromagnetic wave signals radiated outward by a millimeter-wave subarray carried by a single UAV, characterizing the signal radiation intensity of a single subarray.

[0143] A large-aperture radiation source is an equivalent radiation device with a large radiation aperture formed by spatially coherently combining all dispersed millimeter-wave subarrays in the target direction.

[0144] An equivalent virtual phased array is an equivalent large-scale phased array radiation system formed by multiple distributed millimeter-wave subarrays carried by UAVs through beam coordinated control and spatial coherent synthesis.

[0145] Furthermore, the airborne control unit of each UAV transmits the beam weights and signal phase parameters that have been configured in collaboration to its own millimeter-wave subarray control unit. Each millimeter-wave subarray synchronously adjusts the radiation amplitude weighting coefficients of each internal array element and the phase value of the output signal according to the received parameters, so that the main lobe of the subarray is aligned with the preset target airspace direction.

[0146] After the parameters are adjusted, all millimeter-wave subarrays in the formation start signal transmission while maintaining time synchronization, radiating millimeter-wave electromagnetic waves of the corresponding parameters towards the target airspace. Each subarray independently completes its own directional radiation output, and the radiation beams of all subarrays point to the same target airspace direction.

[0147] The electromagnetic waves radiated by each millimeter-wave subarray propagate toward the target airspace in a free-space environment. All electromagnetic waves are in the same operating frequency band, and the signal phases are pre-matched.

[0148] When each electromagnetic wave propagates to the corresponding spatial position in the target airspace, its electric field component and magnetic field component are superimposed according to their own phase relationship. The position signal amplitudes with matching phases are added to each other, while the position signal amplitudes with opposite phases cancel each other out. Finally, in the specified direction of the target airspace, an enhanced composite signal with a signal amplitude much higher than that of the single subarray radiation signal is formed, thus completing the concentrated superposition of signal energy in the target direction.

[0149] The electromagnetic waves of all millimeter-wave subarrays in the formation continuously and coherently superimpose in the direction of the target airspace. The subarray transmission power, which was originally dispersed on each UAV, is concentrated in the direction of the target through spatial coherent synthesis, forming a spatial radiation effect with highly concentrated energy.

[0150] This convergence effect is equivalent to the radiation effect produced by a radiation source whose radiation aperture covers the entire distribution range of the UAV swarm. This equivalent large-aperture radiation source does not require a physical large-size antenna array structure. It can achieve the same radiation effect simply through the spatial beam coordination of the distributed subarrays, ultimately forming a virtual phased array of the UAV swarm.

[0151] In this process, each UAV controls the onboard millimeter-wave subarray to adjust the beam direction in real time based on the beam weight and signal phase output by the model, so that the signals of all subarrays in the cluster are coherently combined to form an equivalent large-scale virtual phased array; the actual communication EIRP of the virtual phased array is calculated in real time and fed back to the distributed negotiation module to form a closed-loop optimization mechanism to continuously improve the cluster communication EIRP.

[0152] S5: Collect the real-time radiation capability of the virtual phased array, obtain the radiation capability feedback of the virtual phased array, and detect the heartbeat signal of the individual UAV to which the temporary master node belongs in real time through the radiation capability feedback; In this embodiment, the step of acquiring the real-time radiation capability of the virtual phased array and obtaining the radiation capability feedback of the virtual phased array includes: The power detection unit on each of the aforementioned UAVs is used to collect the radiated power density of the virtual phased array in the target direction in real time. Based on the radiation power density and the spatial location information of the UAV cluster, the actual equivalent isotropic radiation power of the virtual phased array is calculated, and the actual equivalent isotropic radiation power is the radiation capability feedback.

[0153] Specifically, the power detection unit is an airborne hardware detection device installed on each individual UAV, which is matched with the millimeter-wave subarray. It can receive millimeter-wave electromagnetic waves propagating in space and complete the detection and numerical conversion of signal power amplitude.

[0154] The target direction is the pre-defined beam pointing airspace when the UAV swarm performs collaborative tasks, and it is the designated spatial direction where the radiation energy of the virtual phased array is concentrated.

[0155] Radiation power density is the numerical value of millimeter-wave radiation power passing through a unit area in the space in the target direction. It is used to characterize the energy intensity of electromagnetic waves at that spatial location and is obtained by sampling and conversion by the power detection unit.

[0156] The spatial location information of a drone swarm is the set of position coordinates of all individual drones in the formation in a three-dimensional spatial coordinate system, which can reflect the distribution points and relative spacing of each millimeter-wave subarray in space.

[0157] The actual equivalent isotropic radiation power is the equivalent isotropic radiation power value of the virtual phased array in the target direction, which is obtained by converting actual sampled data and is used to quantify the actual radiation performance of the virtual phased array.

[0158] Radiation capability feedback is a quantitative parameter characterizing the real-time radiation capability of a virtual phased array, which can be used for subsequent master node heartbeat status detection and closed-loop adjustment of beam parameters.

[0159] Further, during the formation flight of each drone, the power detection unit matched with the airborne millimeter-wave subarray is started, and the signal receiving surface of the power detection unit is oriented to a preset target direction.

[0160] The power detection unit continuously receives and samples millimeter-wave electromagnetic waves propagating in the space in the target direction at a fixed sampling frequency, converts the received electromagnetic wave signals into electrical signals, obtains the power amplitude of the signals through the amplitude detection circuit, and then calculates the radiant power density in the target direction at the current sampling moment in combination with the effective receiving area of the detection unit.

[0161] The power detection units of all drones in the formation are time-synchronized uniformly, and perform collection operations at the same sampling moment to ensure that the radiant power density data output by each node has time consistency.

[0162] The local processing unit of the drone summarizes the radiant power density data collected by itself, obtains the radiant power density data collected by all other drones in the formation through the distributed communication link at the same time, and then retrieves the three-dimensional position coordinates of all drones in the formation at the current moment.

[0163] Combined with the attenuation characteristic of electromagnetic waves propagating in free space and the spatial position distribution of each drone relative to the target direction, the total combined radiation power of all subarrays in the target direction is converted, and the actual equivalent isotropically radiated power of the entire virtual phased array in the target direction is obtained.

[0164] The actual equivalent isotropically radiated power is locally stored as radiation capability feedback data, and synchronized to other drones in the formation at the same time, so as to provide data support for subsequent master node state detection and beam parameter cooperative adjustment.

[0165] In this embodiment, the step of detecting the heartbeat signal of the individual drone to which the temporary master node belongs in real time through the radiation capability feedback includes: Comparing the radiation capability feedback with the radiation capability threshold of the individual drone under normal operation to obtain a radiation capability comparison result of the individual drone; Determining the heartbeat signal of the individual drone corresponding to the temporary master node based on the radiation capability comparison result.

[0166] Specifically, the radiation capability comparison result is an intermediate determination result obtained by comparing the value of the radiation capability feedback acquired in real time with the value of the radiation capability threshold, and is a state identifier used to reflect whether the radiation capability of the current virtual phased array is in the normal working range.

[0167] The heartbeat signal of the individual drone corresponding to the temporary master node is a status indicator that represents whether the temporary master node is working properly. It indirectly reflects the online and working status of the temporary master node through the change of the overall radiation capability of the virtual phased array, and is used to determine whether the master node can continue to assume the responsibilities of cluster communication scheduling and collaborative command forwarding.

[0168] The heartbeat signal is a status indicator that represents whether the individual drones belonging to the temporary master node in the drone swarm are operating normally. It indirectly reflects the online performance status of the temporary master node through the actual radiation capability changes of the virtual phased array, and serves as the core basis for determining whether the temporary master node can continue to assume the responsibilities of swarm communication scheduling and collaborative command forwarding.

[0169] In a physical scenario where a drone swarm maintains formation flight, and each drone carries a millimeter-wave subarray that radiates millimeter waves directionally and forms a virtual phased array through spatial coherent synthesis, the heartbeat signal is categorized into two states: normal response and no response. When the radiation capability feedback is not lower than the normal operating threshold, the heartbeat signal is in a normal response state, and the millimeter-wave subarray of the corresponding temporary master node can participate in beam coordination normally, possessing the ability to continuously perform its functions. When the radiation capability feedback is consistently lower than the normal operating threshold and exceeds the preset timeout period, the heartbeat signal is in a no-response state, and the corresponding temporary master node cannot output a millimeter-wave signal that meets the requirements, losing the conditions for performing swarm scheduling functions, which can trigger the swarm master node re-election and virtual phased array reconfiguration process.

[0170] Furthermore, as the drone swarm maintains formation flight and all millimeter-wave subarrays coordinate to radiate and form a virtual phased array, each drone continuously collects millimeter-wave radiation power density data propagating in space by using its onboard power detection unit, pointing towards the target airspace. Combined with the three-dimensional position coordinates of all drones in the formation at the current moment, the radiation capability feedback value of the entire virtual phased array is calculated according to the attenuation law of electromagnetic wave propagation in free space.

[0171] At the same time, each drone retrieves the pre-set radiation capability threshold value corresponding to the normal operating state of the drone from the local storage unit, and compares the radiation capability feedback value with the radiation capability threshold value according to the same power dimension.

[0172] When the radiation capability feedback value is greater than or equal to the radiation capability threshold, a comparison result with normal radiation capability is generated; when the radiation capability feedback value is less than the radiation capability threshold, a comparison result with abnormal radiation capability is generated, and finally a radiation capability comparison result that can be used for subsequent status determination is obtained.

[0173] After obtaining the radiation capability comparison results, the drones will determine the working status of the individual drones currently undertaking the temporary master node responsibilities.

[0174] When the radiation capability comparison result is normal, it indicates that the millimeter-wave subarray carried by the temporary master node is in normal working condition, and can output millimeter-wave signals that meet the requirements and participate in the spatial coherent synthesis of the virtual phased array. At this time, it is determined that the individual heartbeat signal of the UAV corresponding to the temporary master node is in normal response condition, and the temporary master node can continue to assume the responsibility of communication scheduling and collaborative command forwarding within the cluster.

[0175] When the radiation capability comparison result is abnormal, it indicates that the millimeter-wave subarray carried by the temporary master node cannot output signals normally or the signal parameters deviate from the preset range, resulting in a decrease in the overall radiation capability of the virtual phased array. At this time, it is determined that the individual heartbeat signal of the UAV corresponding to the temporary master node is in an unresponsive state. This determination result will serve as the starting condition for triggering the subsequent master node re-election and virtual phased array reorganization.

[0176] Among them, the status of the master node is monitored by heartbeat signal. When the master node experiences a heartbeat timeout of more than 200 milliseconds, or no response for three consecutive heartbeat cycles (or the channel signal-to-noise ratio is lower than 5dB or the power consumption of the subarray exceeds 4.5W), the master node is determined to be faulty.

[0177] S6: When the heartbeat signal is unresponsive, a new master node of the UAV cluster is selected and the beam configuration parameters are sent out synchronously to obtain a recombined virtual phased array.

[0178] In this embodiment, when the heartbeat signal is unresponsive, selecting a new master node for the UAV cluster and simultaneously sending out the beam configuration parameters to reassemble the virtual phased array includes: When the radiation capability feedback remains below the normal operating threshold for an extended period exceeding the preset timeout duration, it is determined that the heartbeat signal of the individual drone corresponding to the temporary master node is unresponsive. The individual drone re-collects the local state information, infers the new master-slave election probability distribution through the distributed negotiation module, and selects a new master node based on the new master-slave election probability distribution; Based on the new master node, the latest beam configuration parameters are synchronously sent to all other drones in the drone cluster; All the individual drones, according to the issued beam configuration parameters, readjust the cooperative beams of the individual drones and perform coherent synthesis to form the recombined virtual phased array.

[0179] Specifically, the normal operating threshold is a pre-set minimum equivalent omnidirectional radiated power value corresponding to the normal cooperative working state of the virtual phased array. It is a benchmark value for comparing with real-time radiated capability feedback and judging the cooperative state of the cluster.

[0180] The timeout duration is a pre-set threshold for the duration of continuous abnormal states used to determine master node failure. It is the time-based condition for triggering the master node re-election and phased array reassembly process.

[0181] A lack of response to the heartbeat signal is a status indicator that the temporary master node has lost its ability to perform its duties. This means that the temporary master node is unable to output millimeter-wave signals that meet the coordination requirements and can no longer assume the responsibilities of cluster communication scheduling and command forwarding.

[0182] The probability distribution of the new master-slave election is a numerical sequence generated by recalculation after the original master node fails. Each value in the sequence corresponds to a drone. The higher the value, the more suitable the corresponding drone is to undertake the scheduling responsibility of the master node, which is the direct basis for selecting the new master node.

[0183] The new master node is an individual drone that is elected through a new round of elections after the original temporary master node fails, and is responsible for cluster communication scheduling and forwarding of collaborative commands.

[0184] The recombined virtual phased array is an equivalent large-scale phased array radiation system formed by the remaining UAVs after the failure of the original temporary master node and the completion of spatial coherent synthesis.

[0185] Furthermore, the drone swarm maintains a preset formation and flies stably in the target airspace. Each drone carries a millimeter-wave subarray that radiates millimeter-wave signals toward the preset target direction. The signals from all subarrays coherently superimpose after propagating in free space, forming a virtual phased array covering the target direction.

[0186] During this process, each drone continuously receives millimeter-wave signals from the target direction through its own power detection unit, and calculates the current radiation capability feedback value of the virtual phased array in real time by combining it with its own spatial position. This value is then compared with the pre-set normal operating threshold every moment.

[0187] When the radiation capability feedback value is continuously below the normal working threshold, and the uninterrupted duration of this low value state exceeds the preset timeout period, it is confirmed that the UAV belonging to the temporary master node currently responsible for cluster scheduling is unable to output millimeter wave signals that meet the coordination requirements, and does not have the conditions to continue to perform its duties of cluster communication scheduling and coordination command forwarding, and finally completes the state determination of no heartbeat signal response.

[0188] After determining that the temporary master node is not responding to heartbeats, all drones in the formation that are in normal working condition simultaneously initiate the master node re-election process.

[0189] Each drone re-collects its own operating parameters such as three-dimensional position coordinates, three-dimensional velocity vector, millimeter-wave subarray transmit power, signal phase, beam weight, subarray power consumption, and channel signal-to-noise ratio at a unified sampling frequency through its own mounted positioning module, speed measurement module, and various detection units matching the millimeter-wave subarray, and encapsulates all parameters into the latest local state information according to a pre-agreed data format.

[0190] Each drone broadcasts its own latest local state information to other drones in the formation through the airborne distributed communication link, while continuously receiving and aggregating the local state information broadcast by all other drones, and forming a complete state data set covering all normal drones after verification and validation.

[0191] The distributed negotiation module on each drone invokes the internal processing unit to perform arithmetic processing on the aggregated full-formation state data set, obtains the adaptation degree sorting result of each drone assuming the responsibility of the master node, and selects the drone with the highest adaptation degree as the new master node for the new round.

[0192] After the identity of the new master node is determined, the drone to which the new master node belongs first broadcasts the master node identity confirmation information to all other normal drones in the formation through the airborne distributed communication link to clarify its scheduling authority.

[0193] Subsequently, the new master node synchronously transmits the latest beam configuration parameters adapted to the current formation state to each slave node drone in the formation in a broadcast manner through the airborne distributed communication link. The transmission process is accompanied by a unified timestamp and a data verification identifier, ensuring that all slave nodes can receive complete, time-sequence consistent beam configuration parameters, so as to provide a unified parameter reference for the subsequent reorganization of the virtual phased array.

[0194] After all drones in the formation receive the latest beam configuration parameters, they first verify and confirm the parameters, and then parse the parameters into the beam weight value and signal phase value corresponding to their own millimeter-wave subarray.

[0195] Each drone adjusts the signal output amplitude and phase offset of the radio frequency link of each array element inside the subarray through the control unit of the millimeter-wave subarray, completes the adjustment of the beam weight and signal phase of its own millimeter-wave subarray, and forms a cooperative beam adapted to the current formation state.

[0196] Triggered by a unified time synchronization signal, all the millimeter-wave subarrays of the UAVs synchronously radiate adjusted millimeter-wave signals toward the preset target direction. When the electromagnetic waves radiated by each subarray propagate in the target airspace, they undergo spatial coherent superposition according to their own phase relationship. The signal amplitudes at phase-matching positions are added to each other, while the signal amplitudes at phase-opposite positions cancel each other out. Ultimately, the subarray transmission power dispersed on each UAV is concentrated and converged in the target direction, re-forming a recombined virtual phased array with an equivalent large-aperture radiation effect.

[0197] In this embodiment, in order to achieve the high-precision phase synchronization required for spatial coherent synthesis, the GNSS timing module carried by each UAV receives satellite common-view signals to achieve absolute time synchronization of each node in the cluster.

[0198] Simultaneously, carrier phase differential positioning technology is used to obtain its own centimeter-level three-dimensional position in the global coordinate system in real time. At the beginning of each communication time slot, the UAV calculates the initial phase difference introduced by the path difference based on its current position coordinates, the target airspace unit direction vector broadcast by the temporary master node, and the master node's position coordinates.

[0199] Subsequently, the subarray control unit subtracts the initial phase difference from the target signal phase output by the distributed negotiation module to obtain the compensated phase shifter phase value and performs open-loop phase pre-compensation, so that the signals emitted by each UAV achieve phase alignment when they propagate to the target airspace.

[0200] To further eliminate residual phase errors introduced by platform micro-vibrations and atmospheric disturbances, the temporary master node periodically broadcasts a phase correction command with extremely low bandwidth by monitoring the quality of the synthesized signal. Each slave node receives the command and makes fine adjustments to form a closed-loop correction, ultimately ensuring that all subarray signals achieve highly consistent coherent superposition in the target airspace direction.

[0201] This invention employs a decentralized, edge-based, online distributed reinforcement learning training method, combined with adaptive weight updates, to achieve real-time model optimization and continuous decision-making. Due to the lightweight training model, online training consumes less energy, making it suitable for real-time, low-latency requirements.

[0202] (1) Each UAV collects local status in real time, forms training input vectors, and completes the online construction of training set.

[0203] (2) The model performs forward reasoning on the state vector within the cell and outputs the master-slave election results, beam weights, and signal phase. The node master-slave election score is calculated according to the formula of the reward function, and this score is used as the reward function. The total loss consists of the master-slave election cross-entropy loss, the beam weight mean square error, and the signal phase mean square error (MSE) loss. Among them, the master-slave election is a classification task, and the cross-entropy is used to ensure the election probability distribution; the beam weights and phase are continuously optimized, and the MSE is used to ensure the coherent synthesis accuracy.

[0204] (3) Use the gradient descent algorithm with a small learning rate to perform backpropagation and update the model parameters asynchronously to ensure stable convergence of distributed training.

[0205] (4) Update the scoring weights according to the adaptive weight dynamic adaptation formula.

[0206] (5) The failure of the master node is determined by heartbeat timeout (200ms) and communication interruption (three consecutive heartbeats without response); after the trigger, the UAV swarm will re-elect within ≤50ms to ensure continuous decision-making, uninterrupted training process, and continuous online optimization.

[0207] In this embodiment, the hardware connection relationship of the drone cluster is as follows: Figure 5 As shown, the hardware block diagram of a single drone is as follows: Figure 3 As shown, the drone swarm consists of 100 small multi-rotor drones, each equipped with a miniaturized millimeter-wave subarray. The subarray contains 6 array elements made of GaN material, featuring lightweight and low power consumption. The subarray measures 8cm × 8cm × 1.5cm, weighs 150g, consumes 3W, operates in the 24GHz millimeter-wave band, has a beam scanning range of 0-360°, a scanning speed of 150° / s, and an element spacing of 6.25mm based on the characteristics of the 24GHz band, ensuring a beam pointing accuracy of ≤0.5°.

[0208] Each drone is equipped with a distributed negotiation module, which uses the NVIDIA Jetson Nano edge computing chip. It has powerful local computing capabilities, can process local status information and collaborative decision-making tasks in real time, and has low decision latency; at the same time, it has low power consumption.

[0209] The distributed negotiation module is interconnected with the UAV flight control system and millimeter-wave subarray control unit through the onboard high-speed serial port, and forms a peer-to-peer network with other UAVs in the cluster through the onboard distributed communication unit, thus building a decentralized, distributed, real-time collaborative airborne intelligent decision-making hardware environment.

[0210] The distributed negotiation module receives real-time UAV status data and millimeter-wave subarray operation data output by the local sensing module through a multi-channel parallel acquisition interface. The sampling frequency is no less than 100Hz, and a single frame of data contains 11-dimensional status information. Simultaneously, through ROS2 distributed communication, it receives local status data broadcast by other drones in the cluster in real time, completing the parallel reception and caching of global status information to form a complete cluster status dataset.

[0211] The distributed negotiation module has a built-in data preprocessing unit that performs standardization processing on the collected local and global state data, eliminates differences in different physical dimensions, and finally outputs a 1100-dimensional global state feature vector with regular dimensions and aligned features, which serves as the standard input for the deep neural network model.

[0212] The distributed negotiation module pre-installs and embeds a lightweight dual-branch deep neural network model. After training and convergence, the model parameters are stored locally in the eMMC, avoiding the computational overhead of onboard online training. The module calls JetsonNano's TensorRT acceleration engine to compress the model inference latency to ≤50ms, meeting the requirements of onboard real-time decision-making. During inference, the preprocessed global state feature vector is input into the model and calculated layer by layer through the linear mapping layer, LSTM temporal feature extraction layer, feature compression layer, and feature mapping layer. The dual branches output two types of decision results in parallel: one is a 100-dimensional master-slave node probability distribution vector, and the other is a 100-dimensional beam weight vector and a 100-dimensional signal phase vector.

[0213] The distributed negotiation module has a built-in dynamic power management unit that monitors the chip load in real time: during high-load inference periods, it automatically increases the core frequency to ensure real-time performance; during low-load idle periods, it automatically reduces the frequency, shuts down idle cores, and reduces GPU power consumption, so that the overall power consumption of the module is always ≤5W, adapting to the power supply limitations of drones and taking into account both high performance and low power consumption requirements.

[0214] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0215] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0216] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0217] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0218] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0219] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A distributed intelligent collaborative decision-making method for unmanned aerial vehicle (UAV) swarms based on a virtual phased array, characterized in that, The method includes: S1: Deploy a drone swarm, collect local status information of individual drones in the drone swarm in real time, generate global status information of the drone swarm based on the local status information, wherein each drone is equipped with a millimeter-wave subarray, and the millimeter-wave subarrays equipped by the drone swarm form an initial virtual phased array. S2: By reasoning the global state information through the distributed negotiation module built into the individual UAV, the master-slave election probability distribution and beam configuration parameters of the individual UAV are obtained; S3: Combining the master-slave election probability distribution and the local state information, a comprehensive score is given to each individual drone to obtain the election score result of the individual drone. According to the order of the election score results and the preset fixed term, the individual drones are selected as the temporary master nodes of the drone cluster in sequence. S4: Based on the beam configuration parameters, coordinately adjust the beam direction of the millimeter-wave subarray carried by the individual UAV to obtain the cooperative beam of the individual UAV. Based on the cooperative beam, coherently synthesize the signals of all the millimeter-wave subarrays in the UAV cluster to form an equivalent virtual phased array. S5: Collect the real-time radiation capability of the virtual phased array, obtain the radiation capability feedback of the virtual phased array, and detect the heartbeat signal of the individual UAV to which the temporary master node belongs in real time through the radiation capability feedback; S6: When the heartbeat signal is unresponsive, a new master node of the UAV cluster is selected and the beam configuration parameters are sent out synchronously to obtain a recombined virtual phased array.

2. The distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array as described in claim 1, characterized in that, The real-time acquisition of local state information of individual drones in the drone cluster, and the generation of global state information of the drone cluster based on the local state information, includes: The system collects the three-dimensional position coordinates and three-dimensional velocity vector of the individual UAV in real time, as well as the transmit power, signal phase, beam weight, subarray power consumption, and channel signal-to-noise ratio of the millimeter-wave subarray, and encapsulates them into local state information. The local state information is normalized to obtain the normalized local state vector of the individual UAV. The normalized local state vector is broadcast in real time to all other drones in the drone cluster via the distributed communication link onboard the individual drone. Each of the aforementioned drones receives and aggregates the normalized local state vectors of all other drones, and concatenates them to form the global state information representing the state of all individuals in the drone cluster.

3. The distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array as described in claim 1, characterized in that, The process of inferring the global state information through the distributed negotiation module built into the individual UAV to obtain the master-slave election probability distribution and beam configuration parameters of the individual UAV includes: The global state information is inferred based on the lightweight dual-branch deep neural network model of the distributed negotiation module in the individual UAV. The first output branch of the lightweight dual-branch deep neural network model outputs the probability distribution of master and slave nodes. The second output branch of the lightweight dual-branch deep neural network model outputs a beam weight vector and a signal phase vector, which together constitute the beam configuration parameters.

4. The distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array as described in claim 2, characterized in that, The process of combining the master-slave election probability distribution and the local state information to comprehensively score each individual drone and obtain the election score result of the individual drone includes: Based on the master-slave election probability distribution, the channel signal-to-noise ratio in the local state information, the power consumption of the subarray, and the equivalent omnidirectional radiation power of the initial virtual phased array, a multi-objective joint optimization reward function is constructed. The scalar reward value for each individual UAV is calculated using the multi-objective joint optimization reward function, and the scalar reward value is the election score result.

5. The distributed intelligent cooperative decision-making method for UAV swarms based on a virtual phased array as described in claim 1, characterized in that, The step of selecting individual drones as temporary master nodes of the drone cluster in sequence according to the order of the election scores and a preset fixed term includes: The election score sequence of the drone cluster is obtained by sorting all the individual drones in the drone cluster from high to low according to the election score results. The drone individual ranked first in the election scoring sequence is selected as the temporary master node; A fixed term is set for the temporary master node. After the fixed term expires, a new round of election is automatically triggered according to the election scoring sequence.

6. The distributed intelligent cooperative decision-making method for UAV swarms based on a virtual phased array as described in claim 1, characterized in that, The step of coordinating the beam pointing of the millimeter-wave subarray carried by the UAV according to the beam configuration parameters to obtain the cooperative beam of the UAV includes: Based on the beam weight vector in the beam configuration parameters, the beam weight of the millimeter-wave subarray carried by the UAV is adjusted to obtain the beam adjustment weight of the UAV. Based on the signal phase vector in the beam configuration parameters, the signal phase of the millimeter-wave subarray carried by the UAV is adjusted to obtain the signal adjustment phase of the UAV. The beam adjustment weights and the signal adjustment phases are packaged into the cooperative beam of the individual UAV.

7. The distributed intelligent cooperative decision-making method for UAV swarms based on a virtual phased array as described in claim 6, characterized in that, The step of coherently combining the signals of all the millimeter-wave subarrays within the UAV cluster based on the cooperative beam to form an equivalent virtual phased array includes: Based on the beam adjustment weight and signal adjustment phase of the cooperative beam, the millimeter-wave subarrays carried by all the individual UAVs are directionally radiated to obtain electromagnetic waves from the millimeter-wave subarrays pointing towards the target airspace. Based on the superposition characteristics of the electromagnetic waves when propagating in space, all the electromagnetic waves in the target spatial direction are spatially coherently superimposed to obtain the enhanced composite signal of the millimeter wave subarray in the target spatial direction. Based on the spatial convergence effect of the synthesized signal, the original dispersed transmission power of all the millimeter-wave subarrays in the UAV cluster is spatially coherently synthesized to obtain the equivalent large-aperture radiation source of the UAV cluster in the direction of the target airspace, and the large-aperture radiation source is the virtual phased array.

8. The distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array as described in claim 1, characterized in that, The process of acquiring the real-time radiation capability of the virtual phased array and obtaining feedback on the radiation capability of the virtual phased array includes: The power detection unit on each of the aforementioned UAVs is used to collect the radiated power density of the virtual phased array in the target direction in real time. Based on the radiation power density and the spatial location information of the UAV cluster, the actual equivalent isotropic radiation power of the virtual phased array is calculated, and the actual equivalent isotropic radiation power is the radiation capability feedback.

9. The distributed intelligent collaborative decision-making method for UAV swarms based on a virtual phased array as described in claim 1, characterized in that, The step of real-time detection of the heartbeat signal of the individual UAV to which the temporary master node belongs through the radiation capability feedback includes: The radiation capability feedback is compared with the radiation capability threshold of the individual drone under normal operation to obtain the radiation capability comparison result of the individual drone. Based on the radiation capability comparison results, the heartbeat signal of the individual UAV corresponding to the temporary master node is determined.

10. The distributed intelligent cooperative decision-making method for UAV swarms based on a virtual phased array as described in claim 9, characterized in that, When the heartbeat signal is unresponsive, a new master node is selected for the UAV cluster, and the beam configuration parameters are simultaneously sent out to reassemble the virtual phased array, including: When the radiation capability feedback remains below the normal operating threshold for an extended period exceeding the preset timeout duration, it is determined that the heartbeat signal of the individual drone corresponding to the temporary master node is unresponsive. The individual drone re-collects the local state information, infers the new master-slave election probability distribution through the distributed negotiation module, and selects a new master node based on the new master-slave election probability distribution; Based on the new master node, the latest beam configuration parameters are synchronously sent to all other drones in the drone cluster; All the individual drones readjust their cooperative beams according to the issued beam configuration parameters and perform coherent synthesis to form the recombined virtual phased array.

Images