UAV interception and disposal method and system based on frequency hopping tracking

Abstract

The present invention relates to the technical field of unmanned aerial vehicles (UAVs), and discloses a method and system for intercepting and disposing UAVs based on frequency hopping tracking. The method comprises the following steps: dividing a protection area and establishing a basic data set and an area set; constructing a graph sequence hybrid model and generating a node representation set; identifying key nodes, calculating an environmental impact index, and obtaining a spatiotemporal node set; obtaining the frequency hopping frequency of the UAV at the next moment based on frequency hopping characteristics and enhanced feature representations; and generating an interference beam based on the frequency hopping frequency at the next moment to interfere with the key nodes by a defense facility. The present invention analyzes the topological structure and communication relationship of a UAV cluster through a graph sequence hybrid model, accurately identifies key nodes in the cluster in combination with network centrality, predicts future frequency hopping frequencies, and preferentially and directionally interferes with key nodes, thereby effectively disrupting the cluster's collaborative capability, reducing interception resource consumption, and improving the overall interception success rate.

Description

Technical Field

[0001] The present invention relates to the field of drone technology, and more particularly, to a method and system for intercepting and disposing a drone based on frequency hopping tracking. Background Art

[0002] With the rapid development of drone technology, drones are increasingly used in military, civilian and other fields; however, unauthorized drone intrusions into sensitive areas such as important facilities, airports, and borders pose a certain security threat; currently, interception and disposal technologies for drones mainly include physical interception, electronic interference, and signal deception.

[0003] The existing Chinese patent application with publication number CN114003057A discloses a drone prevention and control method based on frequency hopping cracking technology. The method intercepts the remote control signal and image transmission radio frequency signal emitted by the drone to calculate the positioning information of the drone and detects the frequency hopping signal transmitted by the remote control; obtains the frequency hopping characteristic parameters of the frequency hopping signal through the frequency hopping parameter estimation method; compares the frequency hopping characteristic parameters to identify the drone model, and reads the control command information, control logic and signal modulation method of the corresponding drone model; generates a carrier and PN code of the corresponding frequency according to the control command information and modulates them, performs frequency hopping simulation according to the frequency hopping characteristic parameters, generates a drone control command signal and transmits it. This invention solves the problem that existing drone prevention and control means cannot crack the drone remote control signal through frequency hopping cracking technology, making it impossible to remotely control the drone after the drone is suppressed by interference.

[0004] However, in sensitive areas, there may be drone clusters that use frequency hopping communication technology to improve anti-interference capabilities. In this case, existing technologies find it difficult to effectively identify key nodes and communication topology structures in the cluster, resulting in low interception efficiency. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for intercepting and handling drones based on frequency hopping tracking in order to solve the above problems.

[0006] The present invention provides a method for intercepting and handling a UAV based on frequency hopping tracking, comprising the following steps:

[0007] Divide the protection area into a set of areas, obtain multidimensional data within the protection area, and generate a basic data set containing time series information;

[0008] Based on the regional set and basic data set, temporal information is integrated into the graph model to construct a graph sequence hybrid model, which includes a node set and an edge set. A feature vector is defined for each node in the node set, and the feature vector of each node is temporally encoded to obtain a node representation set, where the nodes include drone nodes, defense facility nodes, regional nodes, and environment nodes.

[0009] Based on the graph sequence hybrid model, network centrality analysis is performed to calculate connection centrality and betweenness centrality. Key nodes are identified based on connection centrality and betweenness centrality, and all key nodes form a key node set. Environmental factor analysis is performed based on the graph sequence hybrid model to calculate the environmental impact index. A temporal graph convolutional network is used to update the node representation set through graph convolution operations, marking it as a spatiotemporal node set.

[0010] The feature vector and environmental impact index of each UAV node in the graph sequence hybrid model are combined to obtain the frequency hopping feature, the frequency hopping feature is reconstructed to obtain the latent feature, the features and latent features of the spatiotemporal node set are combined to obtain the enhanced feature representation, the enhanced feature representation and the frequency hopping feature are input into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each UAV at the next moment;

[0011] The defense facility node generates an interference beam based on the frequency hopping frequency at the next moment to interfere with the key node.

[0012] Furthermore, the method of defining a feature vector for each node in the node set includes:

[0013] The raw data of each node is encoded into numerical features, and the attributes of each numerical feature are sequences that change over time, where:

[0014] The regional node feature vector includes: alarm level, defense priority, response time threshold, regional area, regional perimeter, regional center point coordinates, and regional shape features;

[0015] The UAV node feature vector includes: UAV type, UAV three-dimensional spatial coordinates, three-dimensional velocity components, communication frequency, transmission power, signal strength, and protocol characteristics;

[0016] The defense facility node feature vector includes: defense facility three-dimensional spatial coordinates, equipment type, range, power output, equipment status, interference frequency, and frequency tolerance;

[0017] The environmental node feature vector includes: environmental three-dimensional space coordinates, environmental type, environmental intensity, impact direction, and coverage range.

[0018] Furthermore, the edge set includes:

[0019] Inter-UAV communication relationship edge set: connects UAV nodes that have a communication relationship. The communication relationship is determined by analyzing the communication frequency, signal strength, and protocol characteristics of the UAV nodes. When two UAV nodes use the same communication frequency, the signal strength exceeds the preset threshold, and the protocol characteristics match, a communication relationship is determined.

[0020] Defense facility coverage relationship edge set: connects all defense facility nodes and drone nodes;

[0021] Environmental impact relationship edge set: connecting environment nodes and drone nodes;

[0022] Environmental facility relationship edge set: connecting environmental nodes and defense facility nodes;

[0023] Regional node relationship edge set: connects regional nodes with all other types of nodes, where all other types of nodes refer to drone nodes, defense facility nodes, and environment nodes;

[0024] Region-region relationship edge set: connects adjacent region nodes.

[0025] Furthermore, methods for calculating connectivity centrality, betweenness centrality, and environmental impact index include:

[0026] Calculate the edge weights of each drone node in the graph sequence hybrid model and the multiple drone nodes it is connected to, and obtain the connection centrality of the drone node;

[0027] The betweenness centrality of the drone node is obtained by calculating the sum of the proportions of the drone node in the shortest paths of multiple drone nodes connected to it in the graph sequence hybrid model.

[0028] The weighted sum of the impacts of multiple environmental nodes on the UAV node in the graph sequence hybrid model is calculated to obtain the environmental impact index.

[0029] Furthermore, the method of reconstructing the frequency hopping feature to obtain the latent feature is:

[0030] Step 401: Map the frequency hopping feature to a normal distribution in the latent space through an encoder network, and sample the initial latent feature therefrom;

[0031] Step 402: The decoder network reconstructs the original features from the initial latent features, and jointly optimizes the encoder network and the decoder network by minimizing the reconstruction error and the KL divergence.

[0032] Step 403: Repeat steps 401 to 402, iteratively update the encoder network parameters and the decoder network parameters until the reconstruction error converges or reaches a preset number of training rounds, and output the potential features.

[0033] Furthermore, the method further includes inputting the feature vector, enhanced feature representation, and environmental impact index of each drone node into a basic position prediction model, the basic position prediction model outputting the next moment position of each drone, correcting the next moment position by combining the connection centrality, betweenness centrality, environmental impact index, and feature vector of the environmental node to obtain a final position, and assessing the risk level of the drone based on the final position of the drone;

[0034] The defense facility node generates an interference beam based on the frequency hopping frequency at the next moment and points to the final position to interfere with the key node.

[0035] Furthermore, the method for obtaining the final position includes:

[0036] Step 501: Use the connection centrality and betweenness centrality to correct the next moment position of the UAV node to obtain the coordinated position;

[0037] Step 502: Use the influence weight in the environmental influence index and the unit vector of the influence direction in the environmental node feature vector to adjust the collaborative position to obtain the final position.

[0038] Furthermore, the method also includes generating a final objective function by combining the next-moment frequency hopping frequency, final position, eigenvector, key node set, eigenvector of the defense facility node, and environmental impact index of the UAV node, and solving for the optimal interference power allocation and optimal interference frequency allocation under the constraints of limited power and frequency coverage; wherein the optimal interference power allocation is the optimized configuration of the electromagnetic interference power applied by the defense facility node to the target UAV node, and the optimal interference frequency allocation is the interference frequency of the defense facility node that can cover the next-moment frequency hopping frequency of the UAV;

[0039] The defense facility node generates an interference beam based on the optimal interference power allocation and the optimal interference frequency allocation and points it to the final position to perform interference on the key node.

[0040] Furthermore, the method for obtaining the optimal interference power allocation and the optimal interference frequency allocation under the constraints of limited power and frequency coverage includes:

[0041] Step 601: Construct an optimization objective function. The optimization objective function consists of an interference effect term and a power cost term. The interference effect term is the product of the next-moment risk level of all drone nodes and the interference signal-to-noise ratio; the power cost term is the sum of the power outputs in the feature vectors of all defense facility nodes.

[0042] Step 602: Design constraints for the optimization objective function, where the constraints include total power constraints and frequency coverage constraints.

[0043] Step 603: Modify the optimization objective function by modifying the interference effect term of the original optimization objective function to the product of the importance weights of all UAV nodes and the interference signal-to-noise ratio to obtain a modified objective function;

[0044] Step 604: Process the frequency coverage constraint, generate a penalty when the UAV frequency is not covered, and add the generated penalty to the modified objective function to obtain the final objective function;

[0045] Step 605: Process the total power constraint and solve the final objective function to obtain the optimal interference power allocation and the optimal interference frequency allocation.

[0046] The present invention provides a UAV interception and disposal system based on frequency hopping tracking, which is used to store computer instructions. When the computer instructions are read, the above-mentioned UAV interception and disposal method based on frequency hopping tracking is executed. The system includes:

[0047] A partitioning module is used to divide the protection area into a set of areas, obtain multidimensional data within the protection area, and generate a basic data set containing time series information;

[0048] The data processing module integrates time series information into the graph model based on the region set and the basic data set, constructing a graph sequence hybrid model. The graph sequence hybrid model includes a node set and an edge set. It defines a feature vector for each node in the node set and performs time series encoding on the feature vector of each node to obtain a node representation set, where the nodes include drone nodes, defense facility nodes, region nodes, and environment nodes.

[0049] The recognition module performs network centrality analysis based on a graph sequence hybrid model, calculates connection centrality and betweenness centrality, and identifies key nodes based on connection centrality and betweenness centrality. All key nodes form a key node set. It also performs environmental factor analysis based on a graph sequence hybrid model and calculates the environmental impact index. It uses a temporal graph convolutional network to update the node representation set through graph convolution operations, marking it as a spatiotemporal node set.

[0050] The prediction module combines the feature vector of each drone node in the graph sequence hybrid model with the environmental impact index to obtain a frequency hopping feature, reconstructs the frequency hopping feature to obtain a latent feature, combines the features and latent features of the spatiotemporal node set to obtain an enhanced feature representation, inputs the enhanced feature representation and the frequency hopping feature into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each drone at the next moment;

[0051] In the execution module, the defense facility node generates an interference beam based on the frequency hopping frequency at the next moment to interfere with the key node.

[0052] The beneficial effects of the present invention are as follows: the present invention analyzes the topological structure and communication relationship of the drone cluster through a graph sequence hybrid model, accurately identifies key nodes in the cluster in combination with network centrality, predicts future frequency hopping frequencies, and preferentially interferes with key nodes, effectively disintegrating the cluster's collaborative capabilities, reducing interception resource consumption, and improving the overall interception success rate;

[0053] Identify the subsequent location of the drone to ensure the optimal jamming strategy that covers key nodes and meets resource constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the UAV interception and disposal method based on frequency hopping tracking of the present invention;

[0055] Figure 2 This is an example diagram of a graph sequence hybrid model of the UAV interception and disposal method based on frequency hopping tracking of the present invention;

[0056] Figure 3 It is a module diagram of the UAV interception and disposal system based on frequency hopping tracking of the present invention. DETAILED DESCRIPTION

[0057] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described for some examples may be combined in other examples.

[0058] Example 1:

[0059] The embodiment of the present application provides a method for intercepting and handling drones based on frequency hopping tracking, such as Figure 1 As shown, the following steps are included:

[0060] Step 100: Divide the protection area into a set of areas, obtain multidimensional data within the protection area, and generate a basic data set containing time series information;

[0061] Methods for dividing protection areas to obtain area sets include:

[0062] The protection area is divided into multiple irregular areas based on terrain characteristics, security requirements and electromagnetic environment. Each irregular area is composed of multiple three-dimensional spatial coordinates and represented by polygons. An alarm level, response time threshold and defense priority are assigned to each irregular area, and all irregular areas are represented as an area set.

[0063] Divide multiple irregular areas such as:

[0064] The core no-fly zone (3D polygon A) is centered on the runway, with a horizontal radius of 500 meters and a vertical height of 0-1000 meters. The defense priority is 1 (1-10, with lower values ​​indicating higher defense priority), the response time threshold is ≤ 2 seconds, and the alert level is high.

[0065] The outer monitoring area (3D polygon C) covers a 10 km radius around the airport, from ground level to 300 meters above sea level, excluding areas obstructed by mountains. The alert level is low (tracking only, no interception), the response time threshold is ≤ 10 seconds (drone swarm warning), and the defense priority is 5.

[0066] By assigning alarm levels, response time thresholds, and defense priorities to each irregular area, dynamic resource optimization can be achieved. For example, response time can be linked to alarm levels, skipping manual confirmation in high-alarm areas and directly triggering automated interception. Setting up three-dimensional irregular areas can reduce monitoring blind spots (compared to regular grids).

[0067] The method of obtaining multidimensional data within the protection area and generating a basic data set containing time series information includes:

[0068] Obtain multidimensional data on drones, defense facilities, and environmental types within the protection area, and establish a basic data set, which includes a drone data set, a defense facility data set, and an environmental data set.

[0069] In the process of acquiring multidimensional data, timestamp information is recorded to form time series data, including:

[0070] A drone dataset containing time series information: records the 3D spatial coordinates, 3D velocity components, communication frequency, transmission power, signal strength, and protocol feature change sequence of each drone at the past T time points;

[0071] Defense facility dataset containing time series information: records the three-dimensional spatial coordinates, power output, device status, and interference frequency change sequence of each defense facility at the past T time points;

[0072] Environmental datasets containing time series information: Records the 3D spatial coordinates, environmental intensity, impact direction, and coverage change sequence of each environmental type at the past T time points. The 3D spatial coordinates of an environmental type refer to the location information of the area affected by the environmental factor, such as the center coordinates of the electromagnetic interference zone, the center coordinates of the strong wind zone, and the center coordinates of the rainfall zone. These environmental factors have clear spatial distribution characteristics, and their impact ranges need to be located using 3D coordinates to accurately assess their impact on UAV flight and communication.

[0073] These time series data are stored with timestamps as indexes to form a time series database, which provides time series information support for building a graph-sequence hybrid model in subsequent steps.

[0074] Step 200: Based on the region set and the basic data set, the temporal information is integrated into the graph model to construct a graph sequence hybrid model, which includes a node set and an edge set. A feature vector is defined for each node in the node set, and the feature vector of each node is temporally encoded. The static features and temporal dynamic characteristics are integrated to obtain a node representation set, where the nodes include drone nodes, defense facility nodes, region nodes, and environment nodes.

[0075] In one embodiment of the present invention, an example of a graph sequence hybrid model is shown in FIG. Figure 2 shown.

[0076] The node set includes:

[0077] The drone node set, which contains multiple drone nodes, is built based on the drone dataset and includes all detected drones;

[0078] Defense facility node set, which includes multiple defense facility nodes and is built based on the defense facility dataset, and contains all devices that can be used for interception;

[0079] For example, there are full-band drone interception equipment (located in the northwest corner of the airport, with a coverage radius of 3 kilometers and an operating frequency band of 300MHz-6000MHz), and a drone detection and countermeasure system (located at the top of the control tower, with 360-degree coverage, an interception distance of 2 kilometers, and interference frequency bands including 1.2GHz, 1.6GHz, 2.4GHz, 5.2GHz, and 5.8GHz). The purpose of collecting defense facility node sets is to optimize resource allocation, ensure that the most appropriate equipment is used for interception at the right time, improve the interception success rate, and reduce the risk of accidental injuries.

[0080] Irregular regional node set, including multiple regional nodes, is built based on the regional set and includes all divided protection areas;

[0081] The environmental node set, which contains multiple environmental nodes, is constructed based on the environmental dataset and includes all environmental factors that affect interception.

[0082] For example, the electromagnetic interference zone (within 500 meters of the radar station on the east side of the airport, the signal strength is -70dBm, affecting communication quality), the strong wind zone (at the south end of the runway, wind speeds >15m / s, affecting drone stability and flight trajectory), the building obstruction zone (between the control tower and the terminal, forming a radar detection blind spot), and the rainfall zone (south of the airport, rainfall intensity >20mm / h, affecting radar detection effectiveness) are collected to assess the impact of the environment on drones and adjust the interception strategy accordingly to improve interception accuracy and safety.

[0083] Methods for defining a feature vector for each node in a node set include:

[0084] Encode the raw data of each node into machine-understandable numerical features. The attributes of each numerical feature are sequences that change over time, where:

[0085] The regional node feature vector includes: alarm level, defense priority, response time threshold, regional area, regional perimeter, regional center coordinates, and regional shape features. The regional area is calculated using the coordinates of the 3D polygon vertices and is used to assess regional coverage and resource allocation requirements. The regional perimeter is calculated using the polygon boundary length and is used for boundary monitoring and defense deployment planning. The regional center coordinates are obtained by averaging the polygon vertex coordinates and serve as the regional representative point for rapid positioning and calculation. Regional shape features (such as convexity, slenderness, and complexity) are extracted from the polygon shape using a geometric analysis algorithm to identify potential monitoring blind spots and optimize defense strategies. The purpose of collecting these spatial feature data is to achieve precise regional division and differentiated protection, improving the system's adaptability to different regional characteristics.

[0086] The drone node feature vector includes: drone type, drone three-dimensional spatial coordinates, three-dimensional velocity components, communication frequency, transmission power, signal strength, and protocol characteristics. UAV types include quadrotor drones, fixed-wing drones, micro drones, large reconnaissance drones, and relay drones. The three-dimensional spatial coordinates are defined as (x, y, z), which represent the drone's position in three-dimensional space, where x and y represent horizontal coordinates and z represents altitude. The three-dimensional velocity components are defined as (vx, vy, vz), which represent the drone's velocity components in three directions, where vx and vy represent horizontal velocity components and vz represents vertical velocity components. These data are obtained through various detection methods: drone type is obtained through drone image recognition obtained by drone optoelectronic detection equipment (for example, identification is performed using a drone model recognition model trained with a large number of drone images); three-dimensional spatial coordinates are obtained through radar ranging, drone optoelectronic detection equipment, and a multi-station positioning system; three-dimensional velocity components are obtained through Doppler radar velocity measurement and continuous position difference calculation; communication frequency and transmission power are measured using a broadband spectrum analyzer and a directional antenna array; signal strength is measured using dedicated signal receiver evaluation equipment; and protocol characteristics are extracted using deep packet inspection equipment and protocol analysis software. The purpose of collecting this data is to fully understand the physical characteristics and communication characteristics of the UAV, and to provide information support for obtaining the next-moment frequency hopping frequency, optimal interference power allocation, and optimal interference frequency allocation of each UAV.

[0087] The characteristic vectors of defense facility nodes include: the three-dimensional spatial coordinates of the defense facility, equipment type, range, power output, equipment status, interference frequency, and frequency tolerance. These data are obtained through the installation location and introduction manual of the defense facility. The purpose of collecting this data is to accurately evaluate the capability boundaries and optimal working conditions of each defense device, realize the precise scheduling and coordinated operations of defense resources, and improve the interception success rate.

[0088] The environmental node feature vector includes: environmental three-dimensional spatial coordinates, environmental type, environmental intensity, impact direction, and coverage range; the environmental type includes electromagnetic interference areas (such as areas around radar stations and communication base stations), building shielding areas (such as signal blocking areas formed by high-rise buildings, mountains, etc.), strong wind areas (areas where wind speed exceeds a specific threshold), rainfall areas (areas with different rainfall intensities), temperature anomaly areas (areas where temperature is significantly higher or lower than the environmental average), humidity anomaly areas (areas where humidity is significantly higher or lower than the environmental average), etc.; environmental intensity is a quantitative score of the degree of influence of environmental factors, using a standardized scoring system of 0-10, where 0 means no influence, 10 means extremely strong influence, and 10 means extremely strong influence. The calculation method is as follows: for electromagnetic interference areas, scores are calculated based on interference signal strength (dBm) and spectrum width; for building-blocked areas, scores are calculated based on the proportion of blocked area; for strong wind areas, scores are calculated based on wind speed; for rainfall areas, scores are calculated based on rainfall intensity; and for areas with abnormal temperature and humidity, scores are calculated based on the degree of deviation from the normal operating range. The environmental intensity score is standardized to ensure comparability across different environmental factors, providing a quantitative basis for interception decisions. The impact direction refers to the main directional vector of the environmental factor, such as the direction of wind or the direction of electromagnetic interference radiation. This data is collected through meteorological stations, electromagnetic environment monitoring equipment, terrain mapping, and real-time environmental sensor networks. The purpose of collecting this environmental data is to assess the impact of environmental types on the performance of drones and defense facilities, improving the system's adaptability and decision-making accuracy in complex and changing environments.

[0089] The edge set includes:

[0090] Inter-UAV communication relationship edge set: connects UAV nodes that have a communication relationship. The communication relationship is determined by analyzing the communication frequency, signal strength, and protocol characteristics of the UAV nodes. When two UAV nodes use the same communication frequency, the signal strength exceeds the preset threshold, and the protocol characteristics match, it is determined that there is a communication relationship between them. The inter-UAV communication relationship edge set reflects the command and control structure and coordination mode within the UAV cluster, which helps to identify the master UAV and slave UAVs in the UAV cluster.

[0091] Defense facility coverage relationship edge set: connects all defense facility nodes and drone nodes;

[0092] Environmental impact relationship edge set: connecting environment nodes and drone nodes;

[0093] Environmental facility relationship edge set: connecting environmental nodes and defense facility nodes;

[0094] Regional node relationship edge set: connects regional nodes with all other types of nodes, where all other types of nodes refer to drone nodes, defense facility nodes, and environment nodes;

[0095] Region-region relationship edge set: connects adjacent region nodes.

[0096] By constructing these relationship edge sets, the system can fully capture the interactions between all nodes. The drone relationship edge set reflects the collaborative communication patterns within the drone cluster, helping to identify key nodes (key communication nodes are drones that act as signal relays within the drone cluster). The defense facility coverage relationship edge set ensures the precise allocation of defense resources and improves interception efficiency. The environmental impact relationship edge set and the environmental facility relationship edge set enable the system to accurately assess the impact of environmental factors on the performance of drones and defense facilities, enhancing environmental adaptability. The regional node relationship edge set and the region-region relationship edge set optimize resource scheduling.

[0097] In order to effectively process the temporal characteristics of each node, the feature vector of each node is temporally encoded, and the static features and temporal dynamic characteristics are fused to obtain a node representation set. The fusion of static features and temporal dynamic characteristics can simultaneously retain the inherent properties and behavioral change patterns of the nodes, thereby improving the accuracy of the model's prediction of the drone's next-moment frequency hopping frequency.

[0098] Methods for fusing static features and temporal dynamic characteristics to obtain a node representation set include:

[0099] The static feature sequence and dynamic feature sequence of the node are fused into a unified representation, where the static feature sequence is the attributes that do not change with time, such as position and type, and the dynamic feature sequence is the attributes that change with time, such as speed and frequency. For example, for drone nodes, static features include fixed attributes such as drone type, and dynamic feature sequences include time-varying attributes such as three-dimensional spatial coordinates, three-dimensional velocity components, and communication frequency. During the fusion process, the static feature vector is spliced ​​with the dynamic feature vector that has been time-processed to obtain a unified representation. This fused representation not only retains the inherent attribute information of the drone, but also captures the temporal change characteristics of its movement and communication modes.

[0100] In order to make full use of the temporal features of the nodes, we first use the LSTM network to encode the temporal features of the nodes and compress the dynamic feature sequence into a fixed-length vector representation:

[0101]

[0102] in, Representation node The timing characteristics of represents the current time step, For nodes in the past The dynamic feature sequence of time steps, for network.

[0103] Next, a graph attention network is used to integrate the static features and temporal features of the nodes. The static features and temporal features of the nodes are concatenated and converted. After that, the node representation is updated by aggregating the information of neighboring nodes (neighboring nodes are all adjacent nodes that are connected to a node) as the initial representation. The attention mechanism is used for weighting, allowing the node to adaptively obtain information from related neighboring nodes while retaining the temporal dynamic characteristics, capturing the complex dependencies between nodes, and improving the representation ability.

[0104]

[0105]

[0106] in, For nodes The initial feature representation of is the feature conversion function, and The concatenation of is mapped to the latent space, For nodes The static eigenvector of For nodes The temporal characteristics of For nodes The updated feature representation, is the Sigmoid activation function, Representation node The set of all neighbor nodes of is the weight matrix, For nodes The initial feature representation of is the attention coefficient (indicates that the node For Node The attention coefficient of the node For Node The importance of information aggregation is determined by the feature similarity of the node pairs.

[0107] The attention coefficient is determined by calculating the feature similarity between the target node i and the neighbor node j. The specific process is: first, the node features are calculated by the weight matrix Mapping and concatenation, through improved linear rectification function and learnable parameter vector Calculate the unnormalized score, then perform exponential and sum normalization (Softmax) on the neighbor node scores to finally obtain the attention coefficient. This design enables the model to dynamically distinguish the importance of different neighbor nodes and achieve adaptive information aggregation. The specific calculation expression is:

[0108]

[0109] in, represents an exponential function (which converts the attention score into a positive number and amplifies the difference for easy normalization), is an improved linear rectification function (introducing nonlinearity and alleviating the gradient vanishing problem), represents the transpose of the learnable parameter vector (the transpose of the learnable parameter vector is used to calculate the attention score of the relationship between nodes), and represents vector concatenation; for example Representation node and nodes The feature vectors of are spliced ​​to capture the interaction information between the two; Is the weight matrix, used to linearly transform the initial features ; Indicates the node All neighbor nodes of Sum (to achieve local normalization and ensure that the sum of the attention coefficients is 1), Neighbor nodes The initial feature representation of .

[0110] Finally, we get the node representation set , the node representation set includes the updated feature representations of all nodes.

[0111] Step 300: Perform network centrality analysis based on a graph sequence hybrid model, calculate connection centrality and betweenness centrality, identify key nodes based on connection centrality and betweenness centrality, and form a key node set with all key nodes; perform environmental factor analysis based on the graph sequence hybrid model, and calculate the environmental impact index; use a temporal graph convolutional network to update the node representation set through graph convolution operations, and mark it as a spatiotemporal node set; wherein, key nodes refer to nodes that have a core position in the drone cluster. These nodes usually assume the functions of information transfer, command and control, or collaborative decision-making, and have a decisive influence on the overall behavior of the drone cluster; through targeted interference or interception of key nodes, the effect of disintegrating the drone cluster communication network can be maximized, and the defense efficiency can be improved. Key nodes often carry more drone cluster intention information, and focusing on monitoring them helps predict drone cluster behavior. Under limited resources, prioritizing key nodes can achieve optimal allocation of defense resources;

[0112] Based on the graph sequence mixture model, network centrality analysis is performed and the methods for calculating connection centrality include:

[0113] Since the key nodes in the edge set of inter-UAV communication relations usually maintain strong connections with multiple other UAV nodes and play a key role in UAV cluster communication, the sum of the connection strengths (edge ​​weights) of each UAV node with multiple other connected UAV nodes in the graph sequence hybrid model is calculated to obtain the connection centrality of the UAV node, which is used to measure the importance of the UAV node in the communication network. The specific calculation expression of the connection centrality is:

[0114]

[0115] in, For drone nodes The connection centrality of a node reflects its connection strength in the network. represents the set of all drone nodes, For drone nodes and The edge weights between the drone nodes are considered comprehensively. and The exponential term of the spatial distance between them (obtained by calculating the three-dimensional spatial coordinates in the drone node feature vector) and the communication quality are calculated.

[0116] Drone Node and The specific calculation expression of the edge weight between is:

[0117]

[0118] in, and Represents drone nodes respectively and The three-dimensional space position vector of For drone nodes and The Euclidean distance between them is calculated by the three-dimensional spatial coordinates in the drone node feature vector. is the distance attenuation parameter (controls the rate at which the weight decreases with distance, pre-set according to the effective communication range of the drone type of the drone node, and the default value is 1 / 3 of the effective communication range), For drone nodes and The communication quality between the two UAV nodes is determined by the and The transmission power product between the two UAV nodes and The Euclidean distance between them is divided by the square and multiplied by the natural exponential function of the communication frequency similarity. The drone nodes with negative communication frequency similarity are and The absolute value of the difference in communication frequency between The proportional coefficient (adapting to the differences in communication capabilities of different drone types) is calibrated according to the communication capability characteristics of different drone types. The specific setting logic is as follows: for quadrotor drones, since their communication equipment is usually more standardized, the proportional coefficient is set to 1.0 as the benchmark value; for fixed-wing drones, considering that they are usually equipped with higher-power communication equipment, the proportional coefficient is set to 1.2; for micro drones, since their communication capabilities are limited, the proportional coefficient is set to 0.8; for large reconnaissance drones, since they are equipped with professional communication systems, the proportional coefficient is set to 1.5; for relay-type drones, which are specifically used to enhance communication capabilities, the proportional coefficient is set to 2.0; this proportional coefficient setting based on drone type can accurately reflect the actual communication capability differences of different types of drones.

[0119] Based on the node set of the graph sequence mixture model, network centrality analysis is performed. Methods for calculating betweenness centrality include:

[0120] In order to measure the importance of drone nodes in the information transmission path in the edge set of inter-drone communication relations, the sum of the proportions of drone nodes in the shortest paths of multiple drone nodes connected to them in the graph sequence hybrid model is calculated to obtain the betweenness centrality of the drone nodes, which reflects the importance of drone nodes as information transfer stations. UAV nodes with higher importance are located on multiple communication paths. Intercepting them can effectively cut off cluster communication. The specific calculation expression of betweenness centrality is:

[0121]

[0122] in, Represents a drone node The betweenness centrality of Represents a drone node To the drone node The number of shortest paths, Indicates passing through a drone node UAV nodes To the drone node The number of shortest paths is calculated by Dijkstra algorithm, which is a prior art and will not be described in detail here.

[0123] Methods for identifying key nodes based on connectivity centrality and betweenness centrality include:

[0124] When the connection centrality or betweenness centrality of a drone node exceeds a threshold, the drone node is identified as a key node. This is because nodes with high connection centrality usually maintain strong connections with multiple other nodes and play a hub role in the communication network, while nodes with high betweenness centrality are located on multiple communication paths and are highly important as information transfer stations. All key nodes constitute a key node set.

[0125] Methods for analyzing environmental factors based on graph sequence mixture models and calculating environmental impact indexes include:

[0126] In order to quantify the comprehensive impact of complex environments on UAV operations, the importance of different environmental factors is considered. The weighted sum of the influences of multiple environmental nodes on UAV nodes in the graph sequence hybrid model is calculated to obtain the environmental impact index. The weighted sum of the influences is the environmental node Importance scores and environment nodes UAV nodes The specific calculation expression of environmental impact index is:

[0127]

[0128] in, For drone nodes Environmental impact index, Represents the set of all environment nodes, For the environment node The importance score of the environment node The environmental impact index is calculated based on the environmental type, environmental intensity, and coverage. For different types of environmental nodes, such as meteorological conditions (wind, rain, fog), electromagnetic interference sources, and terrain obstacles, basic importance coefficients are set separately. Then, they are weighted according to the environmental intensity. The greater the intensity, the higher the importance score. Finally, the duration of the environmental factor is considered. The longer the duration of the environmental factor, the more significant its impact on the drone, and its importance score increases accordingly. This multi-dimensional assessment ensures that the environmental impact index can accurately reflect the actual impact of various environmental factors on drone operations. For the environment node UAV nodes The influence weight is used to reflect the impact mechanism of different environmental factors on drones. The specific process of calculating the influence weight is: first, the importance of different environmental types is reflected by the environmental type and environmental intensity of the environmental node, and then the environmental node and drone nodes The distance between nodes is reduced to reduce the influence of distant nodes, preventing nodes that are too far away from dominating the influence weight. Then, the absolute value of the product of two cosine functions is introduced, requiring that the influence direction of the environmental node must be directed towards the drone and related to the direction of movement of the drone. This dual constraint improves directional sensitivity. Finally, the scope of action of the environmental node is limited, and the impact outside the coverage range can be ignored.

[0129] The specific calculation expression of the influence weight is:

[0130]

[0131] in, For the environment node UAV nodes The influence weight of For the environment node Environmental type (reflecting the importance of different environmental factors (such as wind and temperature)), For the environment node The environmental strength, Represents an environment node The three-dimensional space position vector of Represents an environment node and drone nodes The distance between nodes, is the cosine function, For the environment node Point to drone node The angle between the direction vector (calculated by the three-dimensional space coordinates between nodes) and the reference coordinate system, For the environment node The direction of influence, For drone nodes The direction of motion (calculated by the three-dimensional velocity components), is the absolute value of the product of two cosine functions, For the environment node coverage.

[0132] Using a temporal graph convolutional network to update a node representation set through graph convolution operations, the methods for labeling a spatiotemporal node set include:

[0133] In order to detect the dynamic behavior pattern of drone clusters, a temporal graph convolutional network is used to update the node representation set through graph convolution operations, which is marked as a spatiotemporal node set. , the spatiotemporal node set includes the updated feature set of all nodes; the temporal graph convolutional network uses the normalized adjacency matrix and degree matrix to transmit information, so that the representation of each node incorporates the local topological structure information, which helps to understand the collaborative behavior of drone clusters.

[0134] Step 400: Combine the feature vector and environmental impact index of each UAV node in the graph sequence hybrid model to obtain a frequency hopping feature, reconstruct the frequency hopping feature to obtain a latent feature, combine the features and latent features of the spatiotemporal node set to obtain an enhanced feature representation, input the enhanced feature representation and the frequency hopping feature into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each UAV at the next moment.

[0135] The method of combining the frequency hopping characteristics is:

[0136] The communication frequency and environmental impact index in the feature vector of each drone node in the graph sequence hybrid model are combined to obtain the frequency hopping feature. For example, the communication frequency of a drone node at a certain moment is 2.4 GHz, and the environmental impact index is 0.75. The combined expression is [2.4, 0.75], which is the frequency hopping feature.

[0137] The specific method of reconstructing the frequency hopping feature to obtain the potential feature is:

[0138] Step 401: Map the frequency hopping feature to a normal distribution (mean is , the standard deviation is ), and sample the initial latent features from it (introducing randomness to enhance generalization).

[0139] Step 402: Then the decoder network The original features are reconstructed from the initial latent features, and the encoder network and decoder network are jointly optimized by minimizing the reconstruction error (the difference between the original features and the frequency-hopping features) and the KL divergence (the difference between the normal distribution of the latent space and the standard normal distribution).

[0140] Step 403: Repeat steps 401 to 402 to iteratively update the encoder network parameters and the decoder network parameters until the reconstruction error converges or reaches a preset number of training rounds, and finally output the potential features.

[0141] The specific calculation expression is:

[0142]

[0143]

[0144] in, Represents a drone node Potential characteristics of (Generated by sampling to introduce randomness and enhance the generalization ability of the encoder network) is sampled from the normal distribution of the encoder output, For drone nodes Frequency hopping characteristics, Indicates the mean , the standard deviation is The normal distribution of and Input for the encoder network The mean and standard deviation of the output (the input frequency hopping feature , which is mapped to the parameters of the normal distribution). For the decoder network (which transforms the latent features Mapping back to the original feature space, the goal is to reconstruct the input features), is the reconstructed original feature (the encoder network is optimized by minimizing the difference between the original feature and the frequency hopping feature).

[0145] In order to ensure that the encoder network retains key information in the process of mapping to the latent space and reconstructing the original features, and that the normal distribution mapped to the latent space is close to the standard normal distribution, the reconstruction loss is optimized based on the square norm and KL divergence of the reconstruction error, where the reconstruction error is the difference between the frequency hopping feature and the original feature.

[0146] The specific calculation expression is:

[0147]

[0148] in, To optimize the reconstruction loss, represents the square norm of the reconstruction error (a measure of how far the decoder output is from the original input), Represents the normal distribution of the latent space and KL divergence between (to prevent overfitting and normalize the latent space), is the Kullback-Leibler divergence, which is used to measure the difference between two probability distributions, prompting the distribution generated by the encoder to be close to the standard normal distribution, thus making the latent space more regular and continuous. represents the standard normal distribution.

[0149] After obtaining the potential features of the drone nodes, the features and potential features of the spatiotemporal node set of each drone node are combined to obtain an enhanced feature representation. The enhanced feature representation and frequency hopping features of each drone node are input into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each drone node at the next moment.

[0150] An example of combining the features and potential features of a spatiotemporal node set to obtain enhanced feature representation is: a drone node in a spatiotemporal node set Features is a 100-dimensional vector, expressed as:

[0151]

[0152] in, 、 and represents the 1st, 2nd, and 100th feature components extracted from the UAV node in the spatiotemporal graph convolutional network, reflecting the spatiotemporal behavior pattern of the UAV;

[0153] Its potential characteristics is a 50-dimensional vector, expressed as:

[0154]

[0155] in, 、 and It represents the 1st, 2nd and 50th latent feature components extracted from the frequency hopping features by the variational autoencoder, which captures the intrinsic structure of the frequency hopping pattern;

[0156] The two are fused through vector concatenation to obtain the drone node Enhanced feature representation of for:

[0157]

[0158] At this time, the enhanced feature representation The dimension of is 150.

[0159] In one embodiment of the present invention, the frequency hopping model adopts a Transformer model;

[0160] The training steps of the frequency hopping model are as follows:

[0161] Prepare training data: Collect the enhanced feature representation and frequency hopping features of drone nodes over a period of time, as well as the corresponding actual frequency hopping frequencies at the next moment as training samples. Divide the training samples into a frequency hopping training set and a frequency hopping test set in a ratio of 7:3.

[0162] Build a frequency hopping model: Use the Transformer model as the basic architecture of the frequency hopping model and set the model parameters according to the input and output dimensions.

[0163] Define the loss function: Use the mean square error (MSE) as the loss function of the frequency hopping model to measure the difference between the frequency hopping frequency predicted by the frequency hopping model and the actual frequency hopping frequency of the drone node at the next moment.

[0164] Training model: The prepared frequency hopping training set is input into the frequency hopping model. The predicted frequency hopping frequency at the next moment is calculated through forward propagation. The error between the predicted value and the true value is then calculated using the loss function. The model parameters are then updated through backpropagation. The process is repeated until the model converges or the preset number of training rounds is reached.

[0165] Evaluate the model: Use an independent frequency hopping test set, input it into the trained frequency hopping model, and calculate the mean square error between the model-predicted frequency hopping frequencies and the actual frequency hopping frequencies to evaluate the model's prediction performance.

[0166] The specific process of using mean square error to evaluate prediction accuracy is as follows: Calculate the prediction value of the frequency hopping model and the true value exist The mean of the squared differences at each time point is used to quantify the prediction bias;

[0167] The specific calculation expression is:

[0168]

[0169] in, Express Take the average of the time points (eliminating the fluctuation of the time dimension and ensuring the stability of the evaluation results), Indicates the time prediction step (indicates the time interval from the current moment to the predicted moment), For drone nodes At the next moment The actual frequency hopping frequency, Predicting drone nodes for frequency hopping models At the next moment The frequency hopping frequency, is the mean square error value.

[0170] Step 500: Input the feature vector, enhanced feature representation, and environmental impact index of each drone node into the basic position prediction model. The basic position prediction model outputs the next moment position of each drone node. The next moment position is corrected by combining the connection centrality, betweenness centrality, environmental impact index, and feature vector of the environmental node to obtain the final position, and the risk level of the drone is evaluated based on the final position of the drone.

[0171] The basic position prediction model's inputs are: the three-dimensional spatial coordinates, enhanced feature representation, and environmental impact index (EI) within each drone node's feature vector. The three-dimensional spatial coordinates provide essential information about the drone's current location, essential for predicting its trajectory. The EI incorporates information from a fusion of spatiotemporal node features and latent features, reflecting the drone's communication status and clustered collaborative behavior. The EI quantifies the impact of the surrounding environment on drone motion, helping the model adapt to complex environmental changes. Together, these three types of data form a complete information chain for position prediction, and the output of the next-moment position more accurately reflects the drone's movement trends in the real world.

[0172] The training method of the basic location prediction model includes:

[0173] Prepare training data: Collect the 3D spatial coordinates, enhanced feature representations, and environmental impact index of drone nodes over a period of time, as well as the corresponding actual position at the next moment (the 3D spatial coordinates of the drone) as training samples. Divide the training samples into a position training set and a position test set in a 7:3 ratio.

[0174] Build a basic location prediction model: Use a feedforward neural network as the architecture of the basic location prediction model and set model parameters based on the input and output dimensions.

[0175] Define the loss function: Use the mean square error (MSE) as the loss function of the basic location prediction model to measure the difference between the model's predicted location and the actual location.

[0176] Training model: Input the prepared position training set into the basic position prediction model, calculate the predicted position at the next moment through forward propagation, then use the loss function to calculate the error between the predicted value and the true value, and then update the model parameters through backpropagation. Continue to iterate until the model converges or reaches the preset number of training rounds, and output the position at the next moment.

[0177] Evaluation model: Use a portion of the independent location test set and input it into the trained basic location prediction model. Calculate the mean square error between the model's predicted next-moment location and the actual location to evaluate the model's prediction performance.

[0178] The method of combining the connection centrality, betweenness centrality, environmental impact index and the characteristic vector of the environmental node to correct the position at the next moment and obtain the final position includes:

[0179] Step 501: Use connection centrality and betweenness centrality to correct the next moment position of the drone node to obtain the coordinated position. This can simulate the impact of the coordinated behavior of the drone cluster (reflected by connection centrality and betweenness centrality) on individual movement.

[0180] Step 502: Use the influence weights in the environmental impact index (i.e., the influence weights of the environmental nodes on the drone nodes described above) and the unit vectors of the influence directions in the environmental node feature vectors to adjust the collaborative position to obtain the final position. This further optimizes the accuracy of position prediction and accounts for the displacement effects caused by various environmental factors.

[0181] The specific process of collaborative position calculation is: based on the next moment position , superimpose the collaborative correction term of the drone node’s neighbor drone node, and the collaborative correction term is determined by the edge weight in the connection centrality and correction vector The product of .

[0182] The specific calculation expression is:

[0183]

[0184] in, For drone nodes The next moment Location, For drone nodes The next moment The collaborative position, Indicates the drone node All neighboring drone nodes Sum, Represents a drone node The set of neighboring drone nodes (i.e., the drone nodes All drone nodes with communication connection) For drone nodes and The edge weights between For drone nodes and The correction vector between them is calculated by the difference in betweenness centrality between the drone node and its neighboring drone nodes, the unit vector of the position difference at the next moment, and the exponential term of the position difference at the next moment. The specific calculation expression of the correction vector is:

[0185]

[0186] in, is the correction coefficient (to control the overall correction amplitude and avoid over-correction), Represents a drone node The betweenness centrality of represents the difference in betweenness centrality (nodes with high betweenness centrality dominate the correction direction, reflecting the concept of "the strong guide the weak"), For drone nodes The next moment position, The unit vector representing the position difference at the next moment (ensuring that the correction direction points to the high betweenness centrality node), is the exponential term of the position difference at the next moment (simulating the physical law that influence decays with distance), is the characteristic distance parameter (determines the attenuation rate, and its default value is the average distance of the drone nodes in the inter-UAV communication relationship edge set, and the average distance is calculated based on the three-dimensional spatial coordinates of the drone nodes).

[0187] The specific process of calculating the final position is: the sum of the coordinated position of the UAV node and the environmental offset terms of multiple environmental nodes. The environmental offset term is the sum of the influence weight of the environmental node on the UAV node and the unit vector of the influence direction in the environmental node feature vector.

[0188] The specific calculation expression of the final position is:

[0189]

[0190] in, For drone nodes The final position at the next moment, Indicates that for all environment nodes Sum (reflecting the superposition effect of multiple environmental factors), For the environment node UAV nodes The influence weight of For the environment node The unit vector of the direction of influence (to ensure that the displacement direction is consistent with the physical effect).

[0191] Methods for assessing the risk level of a drone based on its final location include:

[0192] Use Boolean function to determine whether the final position of the drone at the next moment is within each irregular area, and compare the judgment result with the alarm level of the corresponding area. Multiply them together to get the risk level; if the drone is located in multiple areas at the same time (such as overlapping high-risk and medium-risk areas), its risk level is the sum of the alarm levels of all included areas, reflecting the compound risk effect.

[0193] The specific calculation expression is:

[0194]

[0195] in, For drone nodes The risk level at the next moment, Represents the set of all region nodes, Indicates that for all regional nodes summation (ensuring that all potentially hazardous areas are included in the assessment), Is a Boolean function (determine whether the final position of the drone is in the regional node within), when Located in regional node Returns 1 if it is found, otherwise returns 0. For regional nodes Alert level.

[0196] Step 600: Combine the frequency hopping frequency, final position, eigenvector, key node set, eigenvector of the defense facility node and environmental impact index of the UAV node at the next moment to generate the final objective function, and solve to obtain the optimal interference power allocation and optimal interference frequency allocation under the constraints of limited power and frequency coverage; wherein, the optimal interference power allocation is the optimal configuration of the electromagnetic interference power applied by the defense facility node to the target UAV node, and the optimal interference frequency is the interference frequency of the defense facility node that must cover the frequency hopping frequency of the UAV at the next moment.

[0197] Methods for obtaining optimal interference power allocation and optimal interference frequency allocation under limited power and frequency coverage constraints include:

[0198] Step 601: Construct an optimization objective function. The optimization objective function consists of an interference effect term and a power cost term. The interference effect term is the product of the risk level of all drone nodes at the next moment and the interference signal-to-noise ratio; the power cost term is the sum of the power outputs in the feature vectors of all defense facility nodes; the interference signal-to-noise ratio of the drone node is calculated by accumulating the interference contributions of all defense facility nodes; the interference contribution of each defense facility node is determined by the ratio of the defense facility power output to the transmission power in the drone node feature vector, the relative strength of the interference signal and the drone's own signal path loss, the frequency matching degree between the interference frequency and the frequency hopping frequency (quantified by the Gaussian attenuation model), and the environmental impact index.

[0199] The relative strength of the interference signal and the drone's own signal path loss is the ratio of the square of the distance between the drone node and its communication receiver (the drone node closest to the drone node in the inter-drone communication relationship edge set) (calculated using the three-dimensional spatial coordinates of the two drone nodes) to the square of the distance between the defense facility node and the drone node (calculated based on the drone node's final position at the next moment and the defense facility node's three-dimensional spatial coordinates). When the distance between the defense facility node and the drone node exceeds the defense facility's range, the defense facility's interference contribution to the drone is zero. The interference signal-to-noise ratio (SNR) integrates power, distance, frequency matching, and environmental factors to quantify the intensity of multi-source interference experienced by a drone in a dynamic environment.

[0200] Step 602: To ensure that the jamming strategy is feasible under the actual physical resource constraints and that each target UAV is covered by the frequency of at least one defense facility, constraints are designed for the constructed optimization objective function. The constraints include total power constraints and frequency coverage constraints. The total power constraint is that the total power allocation of all defense facilities must not exceed the total power limit to avoid resource overspending. The frequency coverage constraint is that the frequency hopping frequency of each UAV at the next moment must be covered by the frequency of at least one defense facility node (the difference between the frequency hopping frequency at the next moment and the frequency of the defense facility node is within the frequency tolerance). to ensure comprehensive interception;

[0201] Step 603: To prioritize the interference effect on key nodes and improve the pertinence and effectiveness of interception, the optimization objective function is modified. The interference effect term of the original optimization objective function is modified to the product of the importance weights of all UAV nodes and the interference signal-to-noise ratio. The modified objective function is obtained. The importance weights of UAV nodes are calculated based on the key node set and the risk level at the next moment.

[0202] Step 604: Process the frequency coverage constraint, generate a penalty when the UAV frequency is not covered, and add the generated penalty to the modified objective function to obtain the final objective function;

[0203] Step 605: Process the total power constraint and solve the final objective function to obtain the optimal interference power allocation. and optimal interference frequency allocation ;

[0204] The specific calculation expression of the optimization objective function is:

[0205]

[0206] in, Indicates power allocation and frequency allocation Find the maximum value, To optimize the objective function, Indicates that all The sum of drone nodes, For drone nodes The risk level at the next moment, Indicates that all Defense facilities to sum up, is the trade-off coefficient, For drone nodes Interference signal-to-noise ratio (measures the effectiveness of interference, the larger the value, the stronger the effect), Defense facility node power.

[0207] The specific calculation expression of the interference signal-to-noise ratio of the UAV node is:

[0208] ;

[0209] in, Indicates that all defense facility nodes Sum, Defense facility node The power, For drone nodes The transmission power, For drone nodes The square of the distance between it and its communication receiving end (referring to the drone node closest to the drone node in the inter-drone communication relationship edge set) (reflecting its own signal path loss), Defense facility node With drone nodes The square of the distance between the drone nodes The final position at the next moment and defense facility nodes The three-dimensional spatial coordinates are calculated to reflect the path loss of the interference signal). is the indicator function, when the defense facility node With drone nodes The distance between Less than or equal to defense facility nodes Scope of action The value is 1 when the function is enabled, otherwise it is 0 (to ensure that defense facilities beyond the range do not contribute to interference). For drone nodes At the next moment The frequency hopping frequency, Defense facility node The interference frequency, is the frequency matching parameter (determines the sensitivity of interference to frequency offset), For drone nodes Environmental impact index;

[0210] Modify the specific calculation expression of the objective function as follows:

[0211]

[0212] in, To modify the objective function, For drone nodes Importance weight of

[0213] The calculation method of the importance weight of the drone node is:

[0214] When a drone node belongs to the critical node set, the importance weight of the drone node is the product of the risk level at the next moment and the importance coefficient. The importance coefficient is greater than one and is used to increase the importance weight of the drone node; when the drone node does not belong to the critical node set, the importance weight of the drone node is the risk level at the next moment.

[0215] The specific methods of generating penalties are:

[0216] For each drone node, calculate the absolute value of the difference between its frequency hopping frequency at the next moment and the interference frequencies of multiple defense facility nodes, and take the minimum absolute value as the benchmark. If the minimum absolute value still exceeds the frequency tolerance of the defense facility node, calculate the square value of the excess and add it to the total penalty. The specific calculation expression is:

[0217]

[0218] in, Indicates frequency allocation The penalty value, is the penalty coefficient (a positive number used to control the intensity of the penalty), Defense facility node The interference frequency, Indicates that at all defense facility nodes Take the minimum value, for The absolute value of represents the positive function (when When it is less than 0, it is 0. When greater than 0, take , which ensures that the penalty is incurred only when the frequency difference exceeds the frequency tolerance). Defense facility node Frequency tolerance (reflects the frequency deviation tolerance of the actual communication system. The frequency range that each defense facility node can cover is the interference frequency of the defense facility node minus the frequency tolerance to the interference frequency of the defense facility node plus the frequency tolerance. , rather than a single fixed frequency point).

[0219] The method to obtain the final objective function is:

[0220] The final optimization objective function introduces frequency allocation based on the modified objective function Penalty value , the specific calculation expression is:

[0221]

[0222] in, Indicates the modification of the objective function. is the final objective function;

[0223] The method to handle the total power constraint and solve the final objective function is:

[0224] The alternating gradient ascent method combined with the projected gradient method can not only optimize power and frequency allocation separately, but also handle the total power constraint during the optimization process to avoid violating physical resource constraints. The specific process is as follows:

[0225] Based on the current power allocation and frequency allocation, the product of the final objective function and the power gradient operator is calculated, and the power is updated along the gradient direction for the power learning rate parameter, and then the power is updated through the projection operation. The updated power is mapped to the constraint set to ensure that the sum of the power allocations of all defense facilities does not exceed the total power limit and the power allocation of a single defense facility is non-negative, thereby obtaining the mapped power. The mapped power is then fixed, and the product of the final objective function and the gradient operator of the frequency is calculated. The frequency is updated along the gradient direction as the learning rate parameter of the frequency. The process is repeated until convergence occurs. The converged mapped power is used as the optimal interference power allocation, and the converged updated frequency is used as the optimal interference frequency allocation.

[0226] The specific calculation expression is:

[0227]

[0228]

[0229] in, and Represent the gradient operators for power and frequency (reflecting the optimization direction), and Represents the learning rate parameters of power and frequency respectively (controls the update step size to avoid oscillation or slow convergence), is the number of iterations, and Respectively represent and Secondary power distribution, and Respectively represent and Secondary frequency allocation, Represents a set of directional constraints The projection operation (ensuring that the power allocation always satisfies the constraint set) is performed, where is the total power limit (ensuring that power allocation does not exceed the actual available resources).

[0230] Step 700: Combine the optimal interference power allocation and the optimal interference frequency allocation into an interference strategy, combine the key nodes and the final position, execute the interference strategy and evaluate the interception effect; when executing the interference strategy, accurately point the interference beam of the defense facility node to the final position of the drone node, reducing power waste and increasing interference intensity.

[0231] When implementing the jamming strategy, in order to prioritize the jamming effect on key nodes, concentrate resources on dealing with the targets with the greatest threat, and improve the overall interception efficiency, priority is given to implementing directional jamming on key nodes. The specific methods are as follows:

[0232] By optimizing the power distribution of defense facility nodes, we ensure that the interference energy is concentrated on key nodes. By maximizing the sum of the interference signal-to-noise ratio of all key nodes, we prioritize suppressing the most threatening targets and improve the overall interception efficiency. The specific calculation expression is:

[0233]

[0234] in, Represents a defense facility node The optimal interference power allocation is Indicates the optimal power allocation among all Find the one that maximizes the objective function , represents a set of key nodes, For all key nodes Summation.

[0235] When executing the jamming strategy, in order to optimize the jamming effect on the UAV cluster communication link and reduce its coordination ability by destroying the internal communication of the cluster, hierarchical jamming is implemented for the identified communication relationship edge set between UAVs. The specific method is as follows:

[0236] Prioritize destroying communication links with large edge weights or high frequency matching (the difference between the interference frequency of the defense facility node and the communication frequency of the drone node tends to 0) to maximize the interference effect. The specific calculation expression is:

[0237]

[0238] in, Represents a defense facility node The optimal interference frequency allocation, Indicates that all optimal frequency allocations Find the one that maximizes the objective function , is the communication relationship edge set between UAVs, Indicates that all drone communication edges Sum, Defense facility node The interference frequency, For drone nodes and The communication frequency between is the frequency matching parameter.

[0239] Methods for evaluating interception effectiveness include:

[0240] In order to quantitatively evaluate the effectiveness of the interference strategy, the interference effect is measured by the change rate of the interference signal-to-noise ratio of the key nodes, providing direct feedback for strategy optimization and calculating the interference effect evaluation index. As the interference decreases over time, it indicates that the interference is increasing and the evaluation value approaches 1. Otherwise, the effect is weakened. Therefore, the average interference signal-to-noise ratio of all key nodes is taken to ensure the globality of the evaluation results. The specific calculation expression is:

[0241]

[0242] in, is the interference effect evaluation index, Represents a set of key nodes Take the reciprocal of the size, and UAV nodes The current moment and the previous moment interference signal-to-noise ratio.

[0243] Ultimately, the drone interception goal is achieved, and the interference effect evaluation indicators are fed back in real time for the optimization of subsequent interception missions.

[0244] Example 2:

[0245] See Figure 3 As shown, a UAV interception and disposal system based on frequency hopping tracking is provided, which is used to store computer-readable instructions. When the computer-readable instructions are read, the above-mentioned UAV interception and disposal method based on frequency hopping tracking can be executed. The system includes:

[0246] A division module 101 is used to divide the protection area into a set of areas, obtain multi-dimensional data within the protection area, and generate a basic data set containing time series information;

[0247] Data processing module 102 integrates time series information into a graph model based on the region set and the basic data set, constructing a graph sequence hybrid model, which includes a node set and an edge set. It also defines a feature vector for each node in the node set and performs time series encoding on the feature vector of each node to obtain a node representation set, where the nodes include drone nodes, defense facility nodes, region nodes, and environment nodes.

[0248] Identification module 103 performs network centrality analysis based on a graph sequence hybrid model, calculates connection centrality and betweenness centrality, identifies key nodes based on connection centrality and betweenness centrality, and forms a key node set with all key nodes; performs environmental factor analysis based on a graph sequence hybrid model and calculates an environmental impact index; uses a temporal graph convolutional network to update the node representation set through graph convolution operations, marking it as a spatiotemporal node set;

[0249] Prediction module 104 combines the feature vector of each UAV node in the graph sequence hybrid model with the environmental impact index to obtain a frequency hopping feature, reconstructs the frequency hopping feature to obtain a latent feature, combines the features of the spatiotemporal node set with the latent feature to obtain an enhanced feature representation, inputs the enhanced feature representation and the frequency hopping feature into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each UAV at the next moment;

[0250] In the execution module 105 , the defense facility node generates an interference beam based on the frequency hopping frequency at the next moment to interfere with the key node.

[0251] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A method for intercepting and handling drones based on frequency hopping tracking, characterized in that: The following steps are involved: Divide the protection area into a set of areas, obtain multidimensional data within the protection area, and generate a basic data set containing time series information; Based on the regional set and basic data set, temporal information is integrated into the graph model to construct a graph sequence hybrid model, which includes a node set and an edge set. A feature vector is defined for each node in the node set, and the feature vector of each node is temporally encoded to obtain a node representation set, where the nodes include drone nodes, defense facility nodes, regional nodes, and environment nodes. Based on the graph sequence hybrid model, network centrality analysis is performed to calculate connection centrality and betweenness centrality. Key nodes are identified based on connection centrality and betweenness centrality, and all key nodes form a key node set. Environmental factor analysis is performed based on the graph sequence hybrid model to calculate the environmental impact index. A temporal graph convolutional network is used to update the node representation set through graph convolution operations, marking it as a spatiotemporal node set. The feature vector and environmental impact index of each UAV node in the graph sequence hybrid model are combined to obtain the frequency hopping feature, the frequency hopping feature is reconstructed to obtain the latent feature, the features and latent features of the spatiotemporal node set are combined to obtain the enhanced feature representation, the enhanced feature representation and the frequency hopping feature are input into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each UAV at the next moment; The defense facility node generates an interference beam based on the frequency hopping frequency at the next moment to interfere with the key node.

2. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 1, characterized in that: Methods for defining a feature vector for each node in a node set include: The raw data of each node is encoded into numerical features, and the attributes of each numerical feature are sequences that change over time, where: The regional node feature vector includes: alarm level, defense priority, response time threshold, regional area, regional perimeter, regional center point coordinates, and regional shape features; The UAV node feature vector includes: UAV type, UAV three-dimensional spatial coordinates, three-dimensional velocity components, communication frequency, transmission power, signal strength, and protocol characteristics; The defense facility node feature vector includes: defense facility three-dimensional spatial coordinates, equipment type, range, power output, equipment status, interference frequency, and frequency tolerance; The environmental node feature vector includes: environmental three-dimensional space coordinates, environmental type, environmental intensity, impact direction, and coverage range.

3. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 2, characterized in that: The edge set includes: Inter-UAV communication relationship edge set: connects UAV nodes that have a communication relationship. The communication relationship is determined by analyzing the communication frequency, signal strength, and protocol characteristics of the UAV nodes. When two UAV nodes use the same communication frequency, the signal strength exceeds the preset threshold, and the protocol characteristics match, a communication relationship is determined. Defense facility coverage relationship edge set: connects all defense facility nodes and drone nodes; Environmental impact relationship edge set: connecting environment nodes and drone nodes; Environmental facility relationship edge set: connecting environmental nodes and defense facility nodes; Regional node relationship edge set: connects regional nodes with all other types of nodes, where all other types of nodes refer to drone nodes, defense facility nodes, and environment nodes; Region-region relationship edge set: connects adjacent region nodes.

4. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 3 is characterized in that: Methods for calculating connectivity centrality, betweenness centrality, and environmental impact index include: Calculate the edge weights of each drone node in the graph sequence hybrid model and the multiple drone nodes it is connected to, and obtain the connection centrality of the drone node; The betweenness centrality of the drone node is obtained by calculating the sum of the proportions of the drone node in the shortest paths of multiple drone nodes connected to it in the graph sequence hybrid model. The weighted sum of the impacts of multiple environmental nodes on the UAV node in the graph sequence hybrid model is calculated to obtain the environmental impact index.

5. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 1, characterized in that: The method of reconstructing the frequency hopping feature to obtain the latent feature is: Step 401: Map the frequency hopping feature to a normal distribution in the latent space through an encoder network, and sample the initial latent feature therefrom; Step 402: The decoder network reconstructs the original features from the initial latent features, and jointly optimizes the encoder network and the decoder network by minimizing the reconstruction error and the KL divergence. Step 403: Repeat steps 401 to 402, iteratively update the encoder network parameters and the decoder network parameters until the reconstruction error converges or reaches a preset number of training rounds, and output the potential features.

6. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 1, characterized in that: The method also includes inputting the feature vector, enhanced feature representation, and environmental impact index of each drone node into a basic position prediction model, the basic position prediction model outputting the next moment position of each drone, correcting the next moment position by combining the connection centrality, betweenness centrality, environmental impact index, and feature vector of the environmental node to obtain a final position, and evaluating the risk level of the drone based on the final position of the drone; The defense facility node generates an interference beam based on the frequency hopping frequency at the next moment and points to the final position to interfere with the key node.

7. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 6, characterized in that: Methods for getting the final position include: Step 501: Use the connection centrality and betweenness centrality to correct the next moment position of the UAV node to obtain the coordinated position; Step 502: Use the influence weight in the environmental influence index and the unit vector of the influence direction in the environmental node feature vector to adjust the collaborative position to obtain the final position.

8. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 6, characterized in that: It also includes generating a final objective function by combining the next-moment frequency hopping frequency, final position, eigenvector, key node set, eigenvector of the defense facility node, and environmental impact index of the UAV node, and solving for the optimal interference power allocation and optimal interference frequency allocation under the constraints of limited power and frequency coverage; wherein the optimal interference power allocation is the optimized configuration of the electromagnetic interference power applied by the defense facility node to the target UAV node, and the optimal interference frequency allocation is the interference frequency of the defense facility node, covering the next-moment frequency hopping frequency of the UAV; The defense facility node generates an interference beam based on the optimal interference power allocation and the optimal interference frequency allocation and points it to the final position to perform interference on the key node.

9. The method for intercepting and handling UAVs based on frequency hopping tracking according to claim 8, characterized in that: Methods for obtaining optimal interference power allocation and optimal interference frequency allocation under limited power and frequency coverage constraints include: Step 601: Construct an optimization objective function. The optimization objective function consists of an interference effect term and a power cost term. The interference effect term is the product of the next-moment risk level of all drone nodes and the interference signal-to-noise ratio; the power cost term is the sum of the power outputs in the feature vectors of all defense facility nodes. Step 602: Design constraints for the optimization objective function, where the constraints include total power constraints and frequency coverage constraints. Step 603: Modify the optimization objective function by modifying the interference effect term of the original optimization objective function to the product of the importance weights of all UAV nodes and the interference signal-to-noise ratio to obtain a modified objective function; Step 604: Process the frequency coverage constraint. When the UAV frequency is not covered, a penalty is generated. The generated penalty is added to the modified objective function to obtain the final objective function. Step 605: Process the total power constraint and solve the final objective function to obtain the optimal interference power allocation and the optimal interference frequency allocation.

10. The UAV interception and disposal system based on frequency hopping tracking is characterized by: It is used to store computer instructions. When the computer instructions are read, the method for intercepting and handling a drone based on frequency hopping tracking according to any one of claims 1 to 9 is executed. The system includes: A partitioning module is used to divide the protection area into a set of areas, obtain multidimensional data within the protection area, and generate a basic data set containing time series information; The data processing module integrates time series information into the graph model based on the region set and the basic data set, constructing a graph sequence hybrid model, which includes a node set and an edge set. It also defines a feature vector for each node in the node set and performs time series encoding on the feature vector of each node to obtain a node representation set. The recognition module performs network centrality analysis based on a graph sequence hybrid model, calculates connection centrality and betweenness centrality, identifies key nodes, and forms a key node set with all key nodes. It also performs environmental factor analysis based on a graph sequence hybrid model and calculates the environmental impact index. It uses a temporal graph convolutional network to update the node representation set through graph convolution operations, marking it as a spatiotemporal node set. The prediction module combines the feature vector of each drone node in the graph sequence hybrid model with the environmental impact index to obtain a frequency hopping feature, reconstructs the frequency hopping feature to obtain a latent feature, combines the features and latent features of the spatiotemporal node set to obtain an enhanced feature representation, inputs the enhanced feature representation and the frequency hopping feature into the frequency hopping model, and the frequency hopping model outputs the frequency hopping frequency of each drone at the next moment; In the execution module, the defense facility node generates an interference beam based on the frequency hopping frequency at the next moment to interfere with the key node.

Images