A multi-unmanned aerial vehicle communication anti-interference resource allocation method based on knowledge graph enhancement

By constructing a knowledge graph-based multi-UAV communication anti-interference resource allocation method, and utilizing Stackelberg local altruistic game theory and graph neural networks to optimize the action strategies of UAVs and jammers, the problem of insufficient overall anti-interference effect of multi-UAV swarms is solved, and the feasibility and accuracy of the joint allocation strategy of channel, power and rate are improved.

CN123075708APending Publication Date: 2026-11-10NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611349318.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-02
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing anti-interference methods for UAV communication are difficult to balance the overall anti-interference effect of multiple UAV swarms, and knowledge graphs have limited information fusion and completion capabilities in dynamic interference scenarios, making it difficult to support joint optimization of channel, power, and rate.

Method used

A knowledge graph-based method for allocating anti-jamming resources in multi-UAV communication is constructed. By establishing a Stackelberg local altruistic game model and combining it with a graph neural network, the action strategies of UAVs and jammers are optimized. Information is cross-fused using physical feature encoding and graph feature encoding to predict the optimal actions of the UAV swarm.

Benefits of technology

It improves the anti-interference capability of multi-UAV communication, enhances adaptability to complex interference environments, improves the feasibility and accuracy of joint allocation strategies for channels, power and rate, and reduces the probability of co-channel conflicts and interference channel selection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123075708A_ABST
    Figure CN123075708A_ABST
Patent Text Reader

Abstract

The application relates to the field of unmanned aerial vehicle communication anti-interference technology and discloses a multi-unmanned aerial vehicle communication anti-interference resource allocation method based on knowledge graph enhancement, which comprises the following steps: establishing a multi-unmanned aerial vehicle communication scene in a city environment, constructing an unmanned aerial vehicle air-ground channel model and a jammer ground link model, and extracting communication environment features; constructing a joint action space, constructing a Stackelberg local altruistic game model between the jammer and the unmanned aerial vehicle group based on the joint action space, and extracting the actions of the unmanned aerial vehicle group and the jammer as game features; constructing an unmanned aerial vehicle group communication anti-interference knowledge graph based on the multi-unmanned aerial vehicle communication scene, the communication environment features and the game features; constructing a multi-unmanned aerial vehicle communication anti-interference resource allocation model based on a graph neural network; inputting the communication environment features, the game features and the unmanned aerial vehicle group communication anti-interference knowledge graph into the multi-unmanned aerial vehicle communication anti-interference resource allocation model, and obtaining a multi-unmanned aerial vehicle communication anti-interference resource allocation strategy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of anti-interference technology for unmanned aerial vehicle (UAV) communication, and more specifically, to a resource allocation method for anti-interference of multi-UAV communication based on knowledge graph enhancement. Background Technology

[0002] With the development of low-altitude economy and wireless communication technology, multi-UAV collaborative communication has been widely used in scenarios such as urban inspection, emergency rescue, power line inspection, and data collection. During mission execution, UAVs need to transmit the collected data back to the fusion center in real time; therefore, a stable and reliable communication link is an important foundation for ensuring mission completion.

[0003] Because wireless channels are open, multi-drone communication is susceptible to both internal and external interference. On the one hand, multiple drones share limited spectrum resources, and if they choose the same channel, internal interference is likely to occur. On the other hand, external jammers can maliciously interfere with drone communication links by adjusting interference frequencies, transmission power, or interference modes, thereby reducing link stability and data transmission reliability.

[0004] Existing anti-jamming resource allocation methods mostly employ game theory, power control, or channel selection strategies. However, traditional self-interested game theory usually focuses more on the gains of individual drones and makes it difficult to take into account the overall anti-jamming effect of drone swarms. At the same time, some methods do not adequately characterize the local mutual interference relationship between drones and the hierarchical adversarial relationship between drones and jammers.

[0005] Knowledge graphs can integrate multi-source information such as UAV equipment parameters, channel status, interference modes, and policy sets to provide knowledge support for anti-interference decisions. However, the existing UAV communication knowledge graphs still have limited information fusion and completion capabilities in dynamic interference scenarios, making it difficult to support the joint optimization of channel, power, and rate in a timely manner. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a multi-UAV communication anti-interference resource allocation method based on knowledge graph enhancement, which improves the collaborative decision-making ability and communication reliability of multi-UAVs in complex interference environments.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0008] A knowledge graph-enhanced multi-UAV communication anti-interference resource allocation method includes the following steps: Step S1: Establish a multi-drone communication scenario in an urban environment, including: a drone swarm, a fusion center, and a jammer; Step S2: Based on the multi-UAV communication scenario, construct the air-to-ground channel model from the UAV to the fusion center and the ground link model from the jammer to the fusion center, and extract the communication environment features; Step S3: Discretize the transmission power of the UAV and the jammer, construct a joint action space, construct a Stackelberg local altruistic game model between the jammer and the UAV swarm based on the joint action space, and extract the actions of the UAV swarm and the jammer as game features. Step S4: Construct a knowledge graph of UAV swarm communication anti-interference based on multi-UAV communication scenarios, communication environment characteristics, and game-theoretic characteristics; Step S5: Construct a multi-UAV communication anti-interference resource allocation model based on graph neural networks; Step S6: Input the communication environment characteristics, game characteristics, and UAV swarm communication anti-interference knowledge graph into the multi-UAV communication anti-interference resource allocation model to predict the optimal actions of the UAV swarm as the multi-UAV communication anti-interference resource allocation strategy.

[0009] Furthermore, the communication environment characteristics include: communication link path loss, channel gain, internal mutual interference, external malicious interference, channel received signal-to-interference-plus-noise ratio of the fusion center, channel throughput, UAV communication success indication, and UAV interference success indication.

[0010] Furthermore, the extraction process for the UAV communication success indication and the UAV interference success indication is as follows: Set a threshold for the channel received signal-to-interference-plus-noise ratio (SINR) and determine the upper bound of the channel throughput based on the SINR of the fusion center. If the first The drone in The information transmission rate of each stage is not greater than the corresponding upper bound of the channel throughput, and the corresponding channel received signal-to-interference-plus-noise ratio (SIR) is not less than the corresponding channel received SIR threshold. The drone in The communication success indicator for the first stage is set to 1; otherwise, the communication success indicator for the second stage is set to 1. The drone in The communication success indicator for each stage is set to 0; If the first The drone in The channel in the first stage is the same as the channel being jammed by the jammer, and the corresponding channel's received signal-to-interference-plus-noise ratio (SINR) is less than the corresponding channel's SINR threshold. The drone in The interference success indicator for the first stage is set to 1; otherwise, the first stage is set to 1. The drone in The successful interference indicator for each stage is set to 0.

[0011] Furthermore, the joint action space in step S3 includes: the action space of the UAV and the action space of the jammer; in: No. The drone in Action space at each stage , Indicates the first The drone in The channel selected in each stage Indicates the first The drone in The transmission power at each stage, Indicates the first The drone in Information transmission rate at each stage; jammer In the Action space at each stage , Indicates jammer In the The channel selected in each stage Indicates jammer In the The transmission power at each stage, Indicates jammer In the Interference patterns at each stage.

[0012] Furthermore, the construction process of the UAV swarm communication anti-interference knowledge graph in step S4 is as follows: Entity information is extracted from multi-drone communication scenarios, communication environment characteristics, and game characteristics, including: scenario entities, drone entities, jammer entities, fusion center entities, channel entities, communication link entities, scenario channel entities, scenario channel status entities, drone action entities, jammer action entities, and jamming event entities; Based on the semantic directional relationships between entity information, extended relationships are formed, including: original semantic directional relationships, inverse semantic directional relationships, and self-loop directional relationships; Triples are formed based on entity information and extended relationships to create a knowledge graph for anti-interference in UAV swarm communication.

[0013] Furthermore, the multi-UAV communication anti-interference resource allocation model includes: a physical feature encoding module, a graph feature encoding module, a cross-fusion module, a graph neural network module, and an action prediction output module; The physical feature encoding module is used to encode communication environment features and game features to form spatial mutual interference information between UAV swarms and jammers under the physical view; The graph feature encoding module is used to encode the entity information of the UAV swarm communication anti-interference knowledge graph to form spatial mutual interference information of multiple UAVs under the graph view; The cross-fusion module is used to cross-fuse the spatial mutual interference information of multiple UAVs under the physical view and the spatial mutual interference information of multiple UAVs under the map view to obtain the spatial mutual interference information of multiple UAVs under the physical-map fused view. The graph neural network module is used to transmit information based on the spatial mutual interference information of UAV swarm and jammer under the physical view, the spatial mutual interference information of multiple UAVs under the graph view, and the spatial mutual interference information of multiple UAVs under the physical-graph fusion view, and to capture the complex spatial mutual interference information between multiple UAVs. The motion prediction output module is used to predict the optimal motion of the drone swarm based on the complex spatial mutual interference information between multiple drones.

[0014] Furthermore, it also includes: A joint resource allocation optimization model is constructed based on the UAV communication success indicator and the upper bound of channel throughput. Under the Stackelberg local altruistic game model between the jammer and the drone swarm, the optimal action of the jammer is determined with the goal of maximizing the jammer's utility. Based on the optimal action of the jammer, the reference action of the drone swarm is determined with the goal of maximizing the joint resource allocation optimization model, while maximizing the utility of the drones as much as possible. The reference actions of the UAV swarm are used as training labels for the joint resource allocation optimization model. The goal is to train the multi-UAV communication anti-interference resource allocation model with the convergence of the comprehensive loss function composed of cross-entropy loss from supervised learning and physical constraints.

[0015] Furthermore, the optimal operation of the jammer is expressed as:

[0016] in, Indicates jammer In the The best action at each stage Indicates jammer In the The utility of each stage , Q This represents a positive constant used to ensure the non-negativity of the interference machine's utility. Indicates the number of drones. express index, Indicates the first The drone in Successful interference indication at each stage Indicates the first The drone in Information transmission rate at each stage This represents the power loss coefficient of the drone. Indicates the cost of switching interference modes. This indicates whether the interference patterns of two adjacent stages are consistent. , ,like , .

[0017] Furthermore: The calculation process for the drone's effectiveness is as follows:

[0018] in, Indicates except the first Besides the drones, the other drones were in the first A set of actions for each stage. Indicates the first The drone in The utility of each stage Indicates the first The drone in The benefits at each stage , Indicates the first The drone in Information transmission rate at each stage Indicates the first The drone in Successful communication indicators at each stage Indicates the first The drone in The transmission power at each stage, This represents the power loss coefficient of the drone. This represents the penalty coefficient for breach of contract in the signal transmission rate of the drone. Indicates the first The drone in The upper bound of the channel throughput of each stage. Indicates the first The drone in Neighbor sets at each stage, express index, Indicates the first The drone in Channel in each stage, Indicates the first The drone in Channel in each stage, Denote the Kronecker function, if , ;like , ; Indicates the first The drone in The transmission power at each stage, Indicates the first drones and the first The drone in Local interference coupling strength at each stage; The construction process of the joint resource allocation optimization model is as follows: , in, This represents the objective function of the joint resource allocation optimization model. Indicates the number of drones. express index, This indicates the number of stages in which the drone performs its mission. express index, Indicates the first The drone in The received signal-to-interference-plus-noise ratio of the channel at each stage This represents the channel received signal-to-interference-plus-noise ratio (SIR) threshold. Indicates the first The drone in The transmission power at each stage, This represents the minimum transmit power of the drone. This indicates the maximum transmit power of the drone.

[0019] Furthermore, the comprehensive loss function, which integrates the cross-entropy loss from supervised learning and the physical constraints, is as follows:

[0020] in, Represents the comprehensive loss function. Indicates the first Cross-entropy loss in supervised learning using drones , Indicates the first The channel score selected by the drone Indicates the first Channel training labels corresponding to the drones Represents cross-entropy. This represents the channel loss weighting coefficient. Indicates the first The selected launch power score for the drone Indicates the first Training labels for the transmit power of each drone. This represents the transmit power loss weighting coefficient. Indicates the first The information transmission rate score selected by the drone Indicates the first Training labels for the information transmission rate of each drone. This represents the weighting coefficient for information transmission rate loss. This indicates a penalty for repeated channel occupancy. , This represents the total number of channels. express index, Indicates the first Select channel for drone The probability, express Weighting coefficients; This indicates a channel selection penalty due to interference. , Indicates the machine being interfered with The set of interfering channels express index, Indicates the first Select channel for drone The probability, express Weighting coefficients; This indicates that the coverage is not constrained by interfering channels. , This represents the set of channels that have not been interfered with. express index, This indicates the number of channels participating in the coverage of channels not constrained by interference. This indicates that at least one drone has chosen the first The probability of each channel, express Weighting coefficients; This represents the loss due to information transmission rate requirement constraints. , Indicates the first The information transmission rate requirements of drones Indicates the first The expected information transmission rate of the drone. This represents a normalized scale for information transmission rate. This represents the penalty coefficient for excessive information transmission rate. Indicates taking the positive part. express The weighting coefficients.

[0021] Compared with the prior art, the present invention has the following beneficial effects: (1) Based on the joint action space, this invention constructs a Stackelberg local altruistic game model between the jammer and the UAV swarm. In the sequential confrontation process where the jammer makes decisions first and the UAVs respond later, it can extract the evolution law of the jammer's actions and the characteristics of the cooperative behavior of multiple UAVs. It can also analyze the advantages and disadvantages of each strategy combination based on the changes in the utility of UAVs and the optimization objective value of joint resource allocation, thereby providing game reference for subsequent action label generation and strategy learning. This is conducive to improving the adaptability of the multi-UAV communication anti-jamming resource allocation model to different jamming modes and dynamic scenarios. (2) This invention constructs a knowledge graph of anti-interference for UAV swarm communication by using multiple UAV communication scenarios, communication environment characteristics and game characteristics. It associates and represents information such as scenarios, UAVs, jammers, channels, channel states, and interference events in a structured manner, thereby unifying the originally scattered spatial location relationships, spatial mutual interference relationships and resource allocation relationships into the graph expression. This is conducive to enhancing the modeling ability of complex spatial mutual interference relationships and multi-dimensional resource allocation relationships, improving the feasibility of UAV channel, power and rate joint allocation strategies, and improving the accuracy of conflict avoidance between UAVs and between UAVs and jammers as well as subsequent prediction actions. (3) This invention inputs communication environment features, game features and anti-interference knowledge graph of UAV swarm communication into the multi-UAV communication anti-interference resource allocation model. It uses physical feature encoding, graph feature encoding, cross-fusion and graph neural network to process the spatial mutual interference information between multiple UAVs and predict the optimal action of the UAV swarm accordingly. This achieves the synergistic fusion of communication environment features, game features and anti-interference knowledge graph features, which is conducive to improving the joint prediction accuracy of channel, power and rate, reducing the probability of co-channel conflict, disturbed channel selection and rate default, and improving the global target value of joint resource allocation. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a multi-UAV communication anti-interference scenario according to the present invention; Figure 2 This is a flowchart of the multi-UAV communication anti-interference resource allocation method based on knowledge graph enhancement according to the present invention; Figure 3 This is a schematic diagram of the multi-UAV communication anti-interference resource allocation model in this invention; Figure 4 This is a schematic diagram comparing the optimal action and reference action predicted by the present invention and existing multi-UAV communication anti-interference resource allocation methods. Figure 5This is a diagram showing the channel occupancy at each stage of the Stackelberg game process in this invention. Detailed Implementation

[0023] The technical solution of the present invention will be further explained and described below with reference to the accompanying drawings.

[0024] like Figure 2 This is a flowchart of the multi-UAV communication anti-interference resource allocation method based on knowledge graph enhancement according to the present invention. The multi-UAV communication anti-interference resource allocation method includes the following steps: Step S1: Establish a multi-drone communication scenario in an urban environment, such as... Figure 1 The system comprises a swarm of drones, a data fusion center, and a jamming aircraft. Each drone flies to a specific location in the city to collect data and then transmits the collected data back to the fusion center. During the data collection mission, each drone flies along a pre-set trajectory, and the total mission duration is divided into... K There are 1 stage, and the number of time slots in each stage is 1. T Drones can choose appropriate channels, transmission power, and information transmission rates to establish communication connections with the fusion center, while jammers can also choose channels and power to interfere with drones. Furthermore, it is assumed that the drone swarm and the jammer share the same set of channels, i.e. The drone swarm k The channel selection in each stage is denoted as The jammer will be in the first k The channel selection in each stage is denoted as At the same time, the first n The drone in k The transmission power of each stage is denoted as ,in and These represent the minimum and maximum transmit powers of the UAV, respectively; the transmit power of the jammer is denoted as... ,in and Let be the minimum and maximum transmit power of the jammer, respectively; and let be the set of selectable information transmission rate levels for the UAV. Then the first n The drone in k The information transmission rate for each stage is selected as follows: ;No. n The drone in k The three-dimensional coordinates of each stage can be represented as: The three-dimensional coordinates of the jammer are represented as follows: The three-dimensional coordinates of the fusion center are represented as Obtain a collection of drones. The shared channel set, mission phase, number of time slots, location of each UAV, location of the jammer, location of the fusion center, optional power set of UAVs, optional power set of jammers, and optional rate set of UAVs form the scene characteristics.

[0025] Step S2: Based on the multi-UAV communication scenario, construct the air-to-ground channel model from the UAV to the fusion center and the ground link model from the jammer to the fusion center, and extract communication environment features, including: communication link path loss, channel gain, internal mutual interference, external malicious interference, channel received signal-to-interference-plus-noise ratio of the fusion center, channel throughput, UAV communication success indication, and UAV jamming success indication.

[0026] The line-of-sight link probability is determined based on the pitch angle of the connection between the drone and the fusion center, as well as relevant constants of the urban environment. ,in, a , b , c , d For constants related to the urban environment, Indicates the first n The drone in k The pitch angle formed by connecting each stage with the integration center , , and The first n The drone in k The horizontal, vertical, and axial coordinates of each stage. and These represent the horizontal and vertical coordinates of the jammer.

[0027] The probability of a non-line-of-sight link is expressed as: ; Calculate the channel gain for line-of-sight links and non-line-of-sight links based on the line-of-sight link and non-line-of-sight link probabilities respectively: , in, and These represent the path loss factors for LoS and NLoS links, respectively. and These represent the reference distances for the LoS link and the NLoS link, respectively. Average channel gain at that location, Indicates the first n The drone in k The distance from each stage to the integration center; Based on this, we can then obtain the first... n The drone in k The average path loss of the communication link in each stage is .

[0028] In this invention, internal interference is generated by the transmitted signals of other drones that select the same channel during the communication process between the drone and the fusion center.

[0029] in, Indicates except the first n The remaining drones besides the one mentioned above were gathered together. The Kronecker function is used to determine whether drones are on the same channel. .

[0030] In this invention, external malicious interference includes comb interference, pseudo-random interference, and frequency sweeping interference. Comb interference covers multiple spaced channels within a single phase. Pseudo-random interference determines several affected channels using a pseudo-random sequence. Frequency sweeping interference sweeps across a continuous set of channels. Therefore, the set of affected channels generated by the interference pattern is defined as follows: , in, For the jammer in the first k The interference mode selected at each stage For the jammer in the first k The set of interference channels selected at each stage , and The first k The channel set of comb interference, pseudo-random interference and frequency sweep interference at each stage.

[0031] Based on the set of disturbed channels, the expression for external malicious interference is:

[0032] in, For interference indication function, .

[0033] Based on the aforementioned internal mutual interference and external malicious interference, the channel received signal-to-interference-plus-noise ratio (SINR) of the fusion center is expressed as:

[0034] in, It is additive white Gaussian noise in the channel.

[0035] Based on the channel received signal-to-interference-plus-noise ratio (SINR) of the fusion center, the upper bound of the channel throughput is obtained using Shannon's formula. .

[0036] The extraction process of the UAV communication success indication and the UAV jamming success indication in this invention is as follows: Set the channel receive signal-to-interference-plus-noise ratio threshold. ; If the communication link simultaneously meets the rate feasibility constraint and the signal-to-interference-plus-noise ratio (SINR) threshold requirement, then the UAV communication is considered successful; otherwise, the communication is considered to have failed. Specifically: if the first... The drone in The information transmission rate of each stage is not greater than the corresponding upper bound of the channel throughput, and the corresponding channel received signal-to-interference-plus-noise ratio (SIR) is not less than the corresponding channel received SIR threshold. The drone in The communication success indicator for the first stage is set to 1; otherwise, the communication success indicator for the second stage is set to 1. The drone in The communication success indicator for each stage is set to 0; If the drone and the jammer are on the same channel and the drone fails to communicate, then the jammer is deemed to have successfully interfered with the drone. Specifically: if the first... The drone in The channel in the first stage is the same as the channel being jammed by the jammer, and the corresponding channel's received signal-to-interference-plus-noise ratio (SINR) is less than the corresponding channel's SINR threshold. The drone in The interference success indicator for the first stage is set to 1; otherwise, the first stage is set to 1. The drone in The successful interference indicator for each stage is set to 0.

[0037] Step S3: Discretize the transmission power of the UAV and the jammer to construct a joint action space. Based on the joint action space, construct a Stackelberg local altruistic game model between the jammer and the UAV swarm, and extract the actions of the UAV swarm and the jammer as game features.

[0038] In this invention, the combined action space includes: the action space of the UAV and the action space of the jammer; in: No. The drone in Action space at each stage , Indicates the first The drone in The channel selected in each stage Indicates the first The drone in The transmission power at each stage, , Indicates the first The drone in The number of selectable power levels for each stage Indicates the first The drone in Information transmission rate at each stage; jammer In the Action space at each stage , Indicates jammer In the The channel selected in each stage Indicates jammer In the The transmission power at each stage, Indicates jammer In the Interference patterns at each stage.

[0039] Step S4: Construct a knowledge graph for anti-interference in UAV swarm communication based on multi-UAV communication scenarios, communication environment characteristics, and game theory features. This graph represents information such as scenarios, UAVs, jammers, channels, channel states, and interference events in a structured manner. This unifies the previously dispersed spatial location relationships, spatial mutual interference relationships, and resource allocation relationships into the graph representation, enhancing the modeling ability for complex spatial mutual interference relationships and multi-dimensional resource allocation relationships. It also improves the feasibility of joint UAV channel, power, and rate allocation strategies, and enhances the accuracy of conflict avoidance between UAVs and between UAVs and jammers, as well as subsequent prediction actions. Specifically: Entity information is extracted from multi-UAV communication scenarios, communication environment characteristics, and game-theoretic features, including: scenario entities, UAV entities, jammer entities, fusion center entities, channel entities, communication link entities, scenario channel entities, scenario channel status entities, UAV action entities, jammer action entities, and jamming event entities. Scenario entities represent communication scenarios at specific locations, game stages, and jamming states. UAV entities represent participating UAVs and their associated scenarios, spatial locations, selectable channels, power, speed, and communication links. Jammer entities represent jamming devices and their locations, selectable channels, jamming power, and jamming modes. Fusion center entities represent the target base station connected to the UAV communication. Communication link entities represent the link status between the UAV and the fusion center. Jamming event entities represent the jamming effect of the jammer on the UAV. Scenario channel entities represent the jammed state and potential conflict situations of the channel in a specific scenario. Scenario channel status entities represent the jammed state, conflict pressure, and speed feasibility of the UAV when selecting a candidate channel under current jamming conditions.

[0040] Knowledge graph relationships are used to connect different types of entities and represent the specific connections between entities. These mainly include: the relationship between a scenario containing drones, jammers, and a fusion center; the relationship between the scenario to which a drone or jammer belongs, its location, and the changing stages of the game, including scenario affiliation, location, next stage, and stage update; the link relationship between the drone and the fusion center; the resource relationship between the channels, power, rate, and jamming modes available to the drone or jammer; the jamming strategy relationship between the jammer's currently selected jamming mode, jamming power, center channel, and coverage channel; the relationship between jamming events that affect the drone; the candidate state relationship indicating whether a candidate channel is jammed, its distance from the jamming center, conflict pressure, and rate feasibility; and the co-channel conflict relationship indicating the degree of conflict when different drones choose the same candidate channel.

[0041] Knowledge graphs use triples to store semantic relationships between entities. The original knowledge graph is represented as follows:

[0042] in, This represents the original knowledge graph, which is the set of all valid triples. Represents a set of entities in a knowledge graph. Represents the set of original directional relations. s and o They represent the subject and object entities in the triple, respectively. r Indicates the connection of subject entities s and object entity o Relationship types, Indicates the subject entity s Through relationships r Pointing to the object entity o .

[0043] In this context, the subject and object entities are determined based on the specific relation type. For example, in the triple (Scene, has_uav, UAV), the subject entity is the scene entity, and the object entity is the drone entity, indicating that the scene contains the drone; in the triple (UAV, located_at, Position), the subject entity is the drone entity, and the object entity is the location entity, indicating that the drone is located at the corresponding spatial location; in the triple (JammerStrategy, selects_mode, JammingMode), the subject entity is the jamming strategy entity, and the object entity is the jamming mode entity, indicating that the jamming strategy selects the corresponding jamming mode; in the triple (CandidateChannelState, candidate_channel, Channel), the subject entity is the candidate channel state entity, and the object entity is the channel entity, indicating that the candidate channel state corresponds to the channel; in the triple (CandidateChannelState1, same_channel_conflict, CandidateChannelState2), both the subject and object entities are candidate channel state entities, indicating that there is a same-channel conflict relationship between the two candidate channel states. For UAV action prediction, the knowledge graph also includes action relationships between the UAV and the target channel, target power, and target information transmission rate, represented as uav_channel, uav_power, and uav_rate, respectively. The subject entity in the above action relationship is the UAV entity, and the object entities are the channel entity, the UAV power entity, and the information transmission rate entity, respectively, which are used to represent the channel selection result, power selection result, and information transmission rate selection result corresponding to the UAV.

[0044] Since the original knowledge graph is a directed graph, if only unidirectional relations are retained, messages can only propagate along the original triple relation direction. Entities appearing as objects find it difficult to pass their information back to the subject entity, and the original semantics of the nodes themselves may be weakened during neighbor aggregation. To enable RABGAT to both preserve the original semantic direction of relations and to back-aggregate associated entity information while retaining the information of the central entity, it extends the semantic direction relations between entity information to form extended relations, including: original semantic direction relations, reverse semantic direction relations, and self-loop direction relations.

[0045] For any original triplet ( s , r , o Its reverse semantic representation is derived from the object entity. o Returning to the subject entity via the inverse relation s Its self-loop direction is used to incorporate the entity's own representation into the aggregation process, and the expanded knowledge graph is represented as follows:

[0046] in, This represents the expanded knowledge graph. Representing relations r The corresponding reverse relationship has a message passing direction opposite to the original relationship. Indicates a self-loop relationship, a triple ( s , Θ, s ) represents an entity s Connected to itself, symbol This represents the union of sets. The reverse relation does not simply convert the knowledge graph into an undirected graph, but rather sets an independent relation type for backpropagation, enabling the subsequent multi-UAV communication anti-interference resource allocation model to distinguish between forward and reverse semantics. The extended relation set is represented as:

[0047] in, This represents the expanded set of all relation types. This represents the set of all reverse relationships. This represents a set of self-loop relations. The three types of relations employ independent relation embeddings and direction weight matrices, enabling subsequent RABGAT encoding to assign different attention weights according to the message direction without confusing the semantics of the original direction, reverse direction, and self-loop direction.

[0048] Finally, triples of a knowledge graph for anti-interference in UAV swarm communication are formed based on entity information and extended relationships.

[0049] In this invention, RABGAT encoding takes the central entity, its neighboring entities, and relation sets in a multi-UAV anti-interference knowledge graph as input. First, it interactively models the embeddings of the central entity's neighboring entities and their corresponding relation embeddings, extracting joint semantic information between entities and relations through a cross-interaction method. Different relation parameter matrices are set for original-direction relations, reverse-direction relations, and self-loop relations to enhance the distinguishability of information representation under different relation types. Then, the interactive representation information is input into the relation weight matrix to generate message vectors. An attention network is used to calculate the absolute attention scores of each neighbor message of the central entity, which are then normalized using Softmax to obtain relative attention scores. Weighted aggregation of each neighborhood message is then performed to obtain the updated central entity embedding. Furthermore, a multi-head attention mechanism is introduced, aggregating and averaging the message vectors obtained from multiple attention heads to improve the graph embedding's ability to express complex neighborhood semantics. Simultaneously, a linear transformation is applied to the relation embeddings to maintain dimensionality consistency, ultimately achieving joint updates of entity and relation representations. This allows the graph embedding to more fully represent spatial interference relationships in multi-UAV scenarios, providing high-quality structured semantic features for subsequent resource allocation prediction.

[0050] Step S5: Construct a multi-UAV communication anti-interference resource allocation model based on graph neural networks, such as... Figure 3 It includes a physical feature encoding module, a graph feature encoding module, a cross-fusion module, a graph neural network module, and an action prediction output module. It uses physical feature encoding, graph feature encoding, cross-fusion, and graph neural networks to process spatial mutual interference information between multiple UAVs and predict the optimal actions of the UAV swarm accordingly. This achieves the synergistic fusion of communication environment features, game features, and anti-interference knowledge graph features, which is beneficial to improving the joint prediction accuracy of channel, power, and rate, reducing the probability of co-channel conflict, disturbed channel selection, and rate default, and improving the global target value of joint resource allocation.

[0051] The physical feature encoding module is used to encode communication environment features and game-theoretic features to form spatial mutual interference information between UAV swarms and jammers under the physical view. ,in, Indicates the first A concatenated vector of the communication environment characteristics and game-theoretic characteristics of a drone. This represents a network structure consisting of a linear mapping, BatchNorm, ReLU activation function, and Dropout concatenation. The graph feature encoding module is used to encode the entity information of the UAV swarm communication anti-interference knowledge graph, forming spatial mutual interference information of multiple UAVs under the graph view. ,in, Embedded representation of entity information in a knowledge graph representing anti-interference capabilities for drone swarm communication. This represents a network structure consisting of a linear mapping, BatchNorm, ReLU activation function, and Dropout concatenation. The cross-fusion module is used to cross-fuse the spatial mutual interference information of multiple UAVs under the physical view and the spatial mutual interference information of multiple UAVs under the knowledge graph view, so as to obtain the spatial mutual interference information of multiple UAVs under the physical-knowledge graph fused view; specifically, the physical feature encoding and the knowledge graph feature encoding are concatenated to obtain the initial input of the cross-fusion module. The cross-fusion module includes a Cross branch and a Deep branch. The Cross branch explicitly crosses the initial input and the current layer representation, i.e. ,in, For the first l The weight vector of each Cross layer For bias terms, The number of Cross layers; the Deep branch is used to learn non-linear implicit relationships, i.e. ,in, It consists of linear mapping, BatchNorm, ReLU activation function, and Dropout. The Cross branch output is concatenated with the Deep branch output, and spatial mutual interference information of multiple UAVs under the physical-map fusion view is obtained through linear transformation, LayerNorm, and ReLU.

[0052] in, Represents the ReLU activation function. This indicates the LayerNorm normalization operation. Represents a linear transformation layer. This indicates the bias layer.

[0053] The graph neural network module is used to transmit information based on the spatial mutual interference information between UAV swarms and jammers in the physical view, the spatial mutual interference information between multiple UAVs in the graph view, and the spatial mutual interference information between multiple UAVs in the physical-graph fusion view, capturing complex spatial mutual interference information between multiple UAVs; specifically: To depict the internal interference relationships between multiple drones, using the same scene... N A scene-level graph structure is constructed using drones. Each drone corresponds to a node in the graph, and the initial node for any branch is represented as: , in, Indicates the view index. Represents the physical view. Represents a view of the spectrum. Represents a physics-map fusion view. Indicates the first The first view The encoding characteristics of each node; Considering the potential mutual interference between any two drones sharing channel resources, we construct an approximately fully connected directed graph for drones in the same communication scenario. An approximately fully connected directed graph means that two directed edges with opposite directions are established between any two different drone nodes, allowing each node to receive messages from all other nodes. However, nodes do not establish edges pointing to themselves, thus eliminating the influence of self-loops in message computation. The physical view is represented as:

[0054] in, s An index representing a communication scenario or scenario sample. Indicates the first s Each communication scenario corresponds to a scenario-level directed graph. This represents the set of drone nodes in the graph. Denotes the set of directed edges in the graph. and They represent the first frame and the first The node corresponding to the drone has a directed edge. Indicates from node Pointing to node The direction of message passing, i.e., the node Receiver node Status message sent.

[0055] Let the first The three-dimensional position of the drone is , No. frame and the first The spatial distance between drones is expressed as Using the spatial relationships between UAVs as directed edges, edge features are constructed for each directed edge to describe distance strength, neighbor relationships, and the inverse distance effect:

[0056] in, The threshold for neighbor determination consists of three terms: the first is the normalized distance, used to describe the spatial scale between drones; the second is the neighbor flag, used to emphasize local proximity relationships; and the third is the inverse of the distance, used to represent the physical relationship where closer proximity leads to stronger potential interference. By incorporating these edge features into the message function, the graph neural network no longer relies solely on node semantic similarity but can explicitly perceive the spatial topology of the scene.

[0057] After obtaining the scene graph, a multi-layer message passing network is used to update the node representation of each drone. For the first... t Layered graph neural network, receiving node From the sending node The received message is represented as:

[0058] in, For the first t MLP layer of layered messages, Used to mask loopback messages. Average aggregation of all neighbor messages:

[0059] in, Indicates the first The drone receives scene-level interference information. Subsequently, the current node representation is concatenated with the aggregated message, and the MLP compute node state increment is updated. ,in, For the first t The MLP layer is updated. The updated node representation is:

[0060] The aforementioned message passing process is applied to the physical feature branch, the knowledge graph feature branch, and the DCN cross-fusion branch, respectively, enabling the model to obtain complex spatial mutual interference information between multiple drones in the same scene from different information views.

[0061] The motion prediction output module is used to predict the optimal motion of a drone swarm based on the complex spatial mutual interference information among multiple drones. Specifically: For the physical view branch, to preserve the direct impact of the original physical state on the decision, a physical feature jump connection is added to the output header:

[0062] in, Indicates the first The physical branch of drones in the mission q The classification score vector on the top, Indicates the channel selection task. This indicates the transmit power selection task. Indicates the information transmission rate task. and Representing tasks q The corresponding node represents the mapping matrix and the physical feature jump mapping matrix.

[0063] For knowledge graph view branches, predictions are made using knowledge graph entity representations and candidate action entity similarity:

[0064] in, Indicates the first The knowledge graph branch of the drone in the mission q The classification score vector on the top, Indicates task q The corresponding learnable semantic score scaling factor, This indicates the characteristics and mission of the drone node. q The semantic similarity score vector between each candidate action entity is expressed as follows:

[0065] in, Embedding for candidate action entities, For the task q The set of candidate action entities.

[0066] For the physics-graph fusion view branch, the cross-fused scene representation, physical jump connections, and knowledge graph action semantic scores are used simultaneously:

[0067] in, Indicates the first The mission of the drone in the physics-map fusion view branch q The classification score vector on; This indicates that the branch is in the task. q The semantic similarity score vector between the features of the drone node and the generated candidate action entities.

[0068] After obtaining the outputs of the three branches, learnable branch weights are set for the channel, power, and rate tasks respectively, and normalized using the Softmax function:

[0069] in, Indicates task q The corresponding three-branch weight parameter vector; , and These represent the physical view branch, knowledge graph view branch, and physical-graph fusion view branch in the task, respectively. q The normalized weights in the formula are such that the sum of the three is 1.

[0070] No. n A drone in the mission q The final classification score is obtained by weighted fusion of the three branches:

[0071] in, Indicates the first A drone in the mission q The final classification score vector is obtained by normalizing the vector using the Softmax function to obtain the selection probability of the corresponding candidate action.

[0072] After the channel task output, a scenario-level channel coordination mechanism is introduced. First, the first... Select channel for drone z The probability of occupancy is calculated, and then the score of the current drone on the corresponding channel is penalized based on the probability of other drones occupying the channel in the same scenario. Select channel for drone z The probability is Then, after adding a penalty to the channel z The probability of selection is:

[0073] in, and These represent the numbers before and after coordination, respectively. A drone in the channel z The score on Indicates the first Select channel for drone z The probability, ρ This represents the channel occupancy penalty coefficient, which is taken as [value missing] in the experiment. .

[0074] Step S6: Input the communication environment characteristics, game characteristics, and UAV swarm communication anti-interference knowledge graph into the multi-UAV communication anti-interference resource allocation model to predict the optimal actions of the UAV swarm as the multi-UAV communication anti-interference resource allocation strategy.

[0075] In one technical solution of the present invention, it further includes: A joint resource allocation optimization model is constructed based on the UAV communication success indicator and the upper bound of channel throughput: , in, This represents the objective function of the joint resource allocation optimization model. Indicates the number of drones. express index, Indicates the first The drone in Information transmission rate at each stage Indicates the first The drone in Successful communication indicators at each stage Indicates the first The drone in The transmission power at each stage, This represents the power loss coefficient of the drone. This represents the penalty coefficient for breach of contract in the signal transmission rate of the drone. and Used to balance throughput, energy consumption, and rate constraints. Indicates the first The drone in The upper bound of the channel throughput of each stage. This indicates the number of stages in which the drone performs its mission. express index, Indicates the first The drone in The received signal-to-interference-plus-noise ratio of the channel at each stage This represents the channel received signal-to-interference-plus-noise ratio (SIR) threshold. Indicates the first The drone in The transmission power at each stage, This represents the minimum transmit power of the drone. This indicates the maximum transmit power of the drone.

[0076] In the Stackelberg local altruistic game model between the jammer and the drone swarm, when any drone unilaterally changes its strategy, its change in utility equals the change in potential function, satisfying the definition of an exact potential game. Therefore, the drone subgame has at least one pure strategy Nash equilibrium. The global or local optimal strategy combination of the potential function corresponds to the stable strategy combination in the drone subgame. Thus, the optimal action of the jammer is determined with the goal of maximizing the jammer's utility. Based on the optimal action of the jammer, and under the premise of maximizing the utility of the drones as much as possible, the reference action of the drone swarm is determined with the goal of maximizing the joint resource allocation optimization model. The optimal action of the jammer and the reference action of the drone swarm together constitute the equilibrium solution of the Stackelberg local altruistic game. The aforementioned Stackelberg local altruistic game model can extract the evolutionary patterns of jammer actions and the cooperative behavioral characteristics of multiple UAVs in a sequential adversarial process where the jammer makes decisions first and the UAVs respond later. It also analyzes the advantages and disadvantages of various strategy combinations based on changes in UAV utility and the objective value of joint resource allocation optimization, thus providing game theory references for subsequent action label generation and strategy learning. This is beneficial for improving the adaptability of multi-UAV communication anti-jamming resource allocation models to different jamming modes and dynamic scenarios. The potential function of the UAV subgame in this invention is specifically: ; In this invention, the utility of a drone consists of drone revenue and local benefits. Drone revenue includes: effective transmission revenue, power consumption, and rate violation penalty. Local benefits are determined based on neighboring drones that are close to the current drone and on the same channel. n The neighbor set is determined by both the distance threshold and the co-channel relationship. Specifically: drones n The neighbor set is defined as ,in, For neighbor distance threshold, For the first i The drone to the n The distance between the drones. The local interference cost is defined as:

[0077] in, For the first n The drone in k Action selection at each stage Similarly. Indicates drone n With drones i The local interference coupling strength between them is expressed as follows:

[0078] in, This represents the conflict cost coefficient.

[0079] Therefore, the calculation process for the utility of drones is as follows:

[0080] in, Indicates except the first Besides the drones, the other drones were in the first A set of actions for each stage. Indicates the first The drone in The utility of each stage Indicates the first The drone in The benefits at each stage , Indicates the first The drone in Neighbor sets at each stage, express index, Indicates the first The drone in Channel in each stage, Indicates the first The drone in Channel in each stage, Denote the Kronecker function, if , ;like , ; Indicates the first The drone in The transmission power at each stage, Indicates the first drones and the first The drone in The local interference coupling strength at each stage.

[0081] The goal of the jammer in this invention is to increase the throughput of communication failures caused by external interference while reducing its own transmission power and mode switching costs. The effectiveness of the jammer is as follows:

[0082] in, Indicates jammer In the The utility of each stage Q This represents a positive constant used to ensure the non-negativity of the interference machine's utility. Indicates the number of drones. express index, Indicates the first The drone in Successful interference indication at each stage Indicates the first The drone in Information transmission rate at each stage This represents the power loss coefficient of the drone. Indicates the cost of switching interference modes. This indicates whether the interference patterns of two adjacent stages are consistent. , ,like , .

[0083] The optimal operation of the jammer in this invention is expressed as follows:

[0084] in, Indicates jammer In the The best actions for each stage.

[0085] The reference actions of the UAV swarm are used as training labels for the joint resource allocation optimization model. The goal is to train the multi-UAV communication anti-interference resource allocation model with the convergence of the comprehensive loss function composed of cross-entropy loss from supervised learning and physical constraints.

[0086] The comprehensive loss function in this invention, which integrates the cross-entropy loss from supervised learning and the physical constraints, is specifically as follows:

[0087] in, Represents the comprehensive loss function. Indicates the first Cross-entropy loss in supervised learning using drones , Indicates the first The channel score selected by the drone Indicates the first Channel training labels corresponding to the drones Represents cross-entropy. This represents the channel loss weighting coefficient. Indicates the first The selected launch power score for the drone Indicates the first Training labels for the transmit power of each drone. This represents the transmit power loss weighting coefficient. Indicates the first The information transmission rate score selected by the drone Indicates the first Training labels for the information transmission rate of each drone. This represents the weighting coefficient for information transmission rate loss. This indicates a penalty for repeated channel occupancy. , Indicates the total number of channels. express index, Indicates the first Select channel for drone The probability, express Weighting coefficients; This indicates a channel selection penalty due to interference. , Indicates the machine being interfered with The set of interfering channels express index, Indicates the first Select channel for drone The probability, express Weighting coefficients; This indicates that the coverage is not constrained by interfering channels. , This represents the set of channels that have not been interfered with. express index, This indicates the number of channels participating in the coverage of channels not constrained by interference. This indicates that at least one drone has chosen the first The probability of each channel, express Weighting coefficients; This represents the loss due to information transmission rate requirement constraints. , Indicates the first The information transmission rate requirements of drones Indicates the first The expected information transmission rate of the drone. This represents a normalized scale for information transmission rate. This represents the penalty coefficient for excessive information transmission rate. Indicates taking the positive part. express The weighting coefficients.

[0088] The multi-UAV communication anti-interference resource allocation model of this invention not only learns the optimal resource selection label given in the training data, but is also subject to physical rules such as channel mutual exclusion, interference avoidance, clean channel utilization and rate requirement satisfaction during the training process, and finally outputs the channel, transmission power and information transmission rate selection results of each UAV.

[0089] like Figure 4 This diagram illustrates the comparison between the optimal action and the reference action predicted by the present invention and existing multi-UAV communication anti-interference resource allocation methods. It can be seen that the multi-UAV communication anti-interference resource predicted by the present invention has the highest average global target value, which is better than other comparison algorithms. This shows that the present invention can more effectively balance throughput improvement, power consumption reduction and rate violation penalty suppression.

[0090] like Figure 5 This invention illustrates the channel occupancy status at each stage of the Stackelberg game process. Different colors in the figure represent different drones, jammers, or unoccupied channels, while marker lines indicate conflicts between drones or drones experiencing interference. As the game progresses, multiple drones can achieve a relatively dispersed channel occupancy status and adjust their own channel selection based on the jammer's channel occupancy, thereby reducing co-channel conflicts and interference-related conflicts. This demonstrates the channel coordination and dynamic anti-interference capabilities of the method presented in this invention.

[0091] In one technical solution of the present invention, a computer-readable storage medium is also provided, storing a computer program that enables a computer to execute the knowledge graph-enhanced multi-UAV communication anti-interference resource allocation method of the present invention.

[0092] In one technical solution of the present invention, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the knowledge graph-enhanced multi-UAV communication anti-interference resource allocation method of the present invention.

[0093] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, and portable compact disc read-only memory (CD). ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0094] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0095] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement, characterized in that, Includes the following steps: Step S1: Establish a multi-drone communication scenario in an urban environment, including: a drone swarm, a fusion center, and a jammer; Step S2: Based on the multi-UAV communication scenario, construct the air-to-ground channel model from the UAV to the fusion center and the ground link model from the jammer to the fusion center, and extract the communication environment features; Step S3: Discretize the transmission power of the UAV and the jammer, construct a joint action space, construct a Stackelberg local altruistic game model between the jammer and the UAV swarm based on the joint action space, and extract the actions of the UAV swarm and the jammer as game features. Step S4: Construct a knowledge graph of UAV swarm communication anti-interference based on multi-UAV communication scenarios, communication environment characteristics, and game-theoretic characteristics; Step S5: Construct a multi-UAV communication anti-interference resource allocation model based on graph neural networks; Step S6: Input the communication environment characteristics, game characteristics, and UAV swarm communication anti-interference knowledge graph into the multi-UAV communication anti-interference resource allocation model to predict the optimal actions of the UAV swarm as the multi-UAV communication anti-interference resource allocation strategy.

2. The method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement according to claim 1, characterized in that, The communication environment characteristics include: communication link path loss, channel gain, internal mutual interference, external malicious interference, channel received signal-to-interference-plus-noise ratio of the fusion center, channel throughput, UAV communication success indication, and UAV interference success indication.

3. The method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement according to claim 2, characterized in that, The extraction process for the drone communication success indication and the drone jamming success indication is as follows: Set a threshold for the channel received signal-to-interference-plus-noise ratio (SINR) and determine the upper bound of the channel throughput based on the SINR of the fusion center. If the first The drone in The information transmission rate of each stage is not greater than the corresponding upper bound of the channel throughput, and the corresponding channel received signal-to-interference-plus-noise ratio (SIR) is not less than the corresponding channel received SIR threshold. The drone in The communication success indicator for the first stage is set to 1; otherwise, the communication success indicator for the second stage is set to 1. The drone in The communication success indicator for each stage is set to 0; If the first The drone in The channel in the first stage is the same as the channel being jammed by the jammer, and the corresponding channel's received signal-to-interference-plus-noise ratio (SINR) is less than the corresponding channel's SINR threshold. The drone in The interference success indicator for the first stage is set to 1; otherwise, the first stage is set to 1. The drone in The successful interference indicator for each stage is set to 0.

4. The method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement according to claim 1, characterized in that, The joint action space in step S3 includes: the action space of the UAV and the action space of the jammer; in: No. The drone in Action space at each stage , Indicates the first The drone in The channel selected in each stage Indicates the first The drone in The transmission power at each stage, Indicates the first The drone in Information transmission rate at each stage; jammer In the Action space at each stage , Indicates jammer In the The channel selected in each stage Indicates jammer In the The transmission power at each stage, Indicates jammer In the Interference patterns at each stage.

5. The method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement according to claim 1, characterized in that, The process of constructing the anti-interference knowledge graph for UAV swarm communication in step S4 is as follows: Entity information is extracted from multi-drone communication scenarios, communication environment characteristics, and game characteristics, including: scenario entities, drone entities, jammer entities, fusion center entities, channel entities, communication link entities, scenario channel entities, scenario channel status entities, drone action entities, jammer action entities, and jamming event entities; Based on the semantic directional relationships between entity information, extended relationships are formed, including: original semantic directional relationships, inverse semantic directional relationships, and self-loop directional relationships; Triples are formed based on entity information and extended relationships to create a knowledge graph for anti-interference in UAV swarm communication.

6. The multi-UAV communication anti-interference resource allocation method based on knowledge graph enhancement according to claim 4, characterized in that, The multi-UAV communication anti-interference resource allocation model includes: a physical feature encoding module, a graph feature encoding module, a cross-fusion module, a graph neural network module, and an action prediction output module; The physical feature encoding module is used to encode communication environment features and game features to form spatial mutual interference information between UAV swarms and jammers under the physical view; The graph feature encoding module is used to encode the entity information of the UAV swarm communication anti-interference knowledge graph to form spatial mutual interference information of multiple UAVs under the graph view; The cross-fusion module is used to cross-fuse the spatial mutual interference information of multiple UAVs under the physical view and the spatial mutual interference information of multiple UAVs under the map view to obtain the spatial mutual interference information of multiple UAVs under the physical-map fused view. The graph neural network module is used to transmit information based on the spatial mutual interference information of UAV swarm and jammer under the physical view, the spatial mutual interference information of multiple UAVs under the graph view, and the spatial mutual interference information of multiple UAVs under the physical-graph fusion view, and to capture the complex spatial mutual interference information between multiple UAVs. The motion prediction output module is used to predict the optimal motion of the drone swarm based on the complex spatial mutual interference information between multiple drones.

7. The multi-UAV communication anti-interference resource allocation method based on knowledge graph enhancement according to claim 6, characterized in that, Also includes: A joint resource allocation optimization model is constructed based on the UAV communication success indicator and the upper bound of channel throughput. Under the Stackelberg local altruistic game model between the jammer and the drone swarm, the optimal action of the jammer is determined with the goal of maximizing the jammer's utility. Based on the optimal action of the jammer, the reference action of the drone swarm is determined with the goal of maximizing the joint resource allocation optimization model, while maximizing the utility of the drones as much as possible. The reference actions of the UAV swarm are used as training labels for the joint resource allocation optimization model. The goal is to train the multi-UAV communication anti-interference resource allocation model with the convergence of the comprehensive loss function composed of cross-entropy loss from supervised learning and physical constraints.

8. The multi-UAV communication anti-interference resource allocation method based on knowledge graph enhancement according to claim 7, characterized in that, The optimal operation of the jammer is expressed as follows: in, Indicates jammer In the The best action at each stage Indicates jammer In the The utility of each stage , Q This represents a positive constant used to ensure the non-negativity of the interference machine's utility. Indicates the number of drones. express index, Indicates the first The drone in Successful interference indication at each stage Indicates the first The drone in Information transmission rate at each stage This represents the power loss coefficient of the drone. Indicates the cost of switching interference modes. This indicates whether the interference patterns of two adjacent stages are consistent. , ,like , .

9. A method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement according to claim 7, characterized in that: The calculation process for the drone's effectiveness is as follows: in, Indicates except the first Besides the drones, the other drones were in the first A set of actions for each stage. Indicates the first The drone in The utility of each stage Indicates the first The drone in The benefits at each stage , Indicates the first The drone in Information transmission rate at each stage Indicates the first The drone in Successful communication indicators at each stage Indicates the first The drone in The transmission power at each stage, This represents the power loss coefficient of the drone. This represents the penalty coefficient for breach of contract in the signal transmission rate of the drone. Indicates the first The drone in The upper bound of the channel throughput of each stage. Indicates the first The drone in Neighbor sets at each stage, express index, Indicates the first The drone in Channel in each stage, Indicates the first The drone in Channel in each stage, Denote the Kronecker function, if , ;like , ; Indicates the first The drone in The transmission power at each stage, Indicates the first drones and the first The drone in Local interference coupling strength at each stage; The construction process of the joint resource allocation optimization model is as follows: , in, This represents the objective function of the joint resource allocation optimization model. Indicates the number of drones. express index, This indicates the number of stages in which the drone performs its mission. express index, Indicates the first The drone in The received signal-to-interference-plus-noise ratio of the channel at each stage This represents the channel received signal-to-interference-plus-noise ratio (SIR) threshold. Indicates the first The drone in The transmission power at each stage, This represents the minimum transmit power of the drone. This indicates the maximum transmit power of the drone.

10. A method for allocating anti-interference resources for multi-UAV communication based on knowledge graph enhancement according to claim 8, characterized in that, The comprehensive loss function, which combines the cross-entropy loss from supervised learning with the physical constraints, is as follows: in, Represents the comprehensive loss function. Indicates the first Cross-entropy loss in supervised learning using drones , Indicates the first The channel score selected by the drone Indicates the first Channel training labels corresponding to the drones Represents cross-entropy. This represents the channel loss weighting coefficient. Indicates the first The selected launch power score for the drone Indicates the first Training labels for the transmit power of each drone. This represents the transmit power loss weighting coefficient. Indicates the first The information transmission rate score selected by the drone Indicates the first Training labels for the information transmission rate of each drone. This represents the weighting coefficient for information transmission rate loss. This indicates a penalty for repeated channel occupancy. , This represents the total number of channels. express index, Indicates the first Select channel for drone The probability, express Weighting coefficients; This indicates a channel selection penalty due to interference. , Indicates the machine being interfered with The set of interfering channels express index, Indicates the first Select channel for drone The probability, express Weighting coefficients; This indicates that the coverage is not constrained by interfering channels. , This represents the set of channels that have not been interfered with. express index, This indicates the number of channels participating in the coverage of channels not constrained by interference. This indicates that at least one drone has chosen the first The probability of each channel, express Weighting coefficients; This represents the loss due to information transmission rate requirement constraints. , Indicates the first The information transmission rate requirements of drones Indicates the first The expected information transmission rate of the drone. This represents a normalized scale for information transmission rate. This represents the penalty coefficient for excessive information transmission rate. Indicates taking the positive part. express The weighting coefficients.