Artificial intelligence-based unmanned aerial vehicle communication protocol feature extraction method and system, electronic device, and storage medium

Abstract

The application provides an unmanned aerial vehicle communication protocol feature extraction method and system based on artificial intelligence, an electronic device and a storage medium, relates to the technical field of unmanned aerial vehicle communication, and comprises the following steps: acquiring the spatial position set of a base station and a group of unmanned aerial vehicles, synchronously collecting real-time speed and acceleration information of each position point to construct a communication environment model; then mapping a virtual space shape, calculating an offset and generating a correction parameter through motion compensation, and adjusting shape parameters in a closed loop; then extracting deep correlation features, generating screening parameters to screen key features, fusing signal attenuation model parameters to generate protocol layer compensation coefficients; finally, extracting space-time change features based on position relationship features, predicting a communication interruption path, combining related parameters to extract protocol features and evaluate risks, and generating a backup node dynamic selection strategy, which can realize unmanned aerial vehicle communication protocol feature extraction and backup communication node dynamic selection based on artificial intelligence.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) communication technology, and in particular to a method, system, electronic device, and storage medium for extracting features of UAV communication protocols based on artificial intelligence. Background Technology

[0002] In complex scenarios where high-speed movement causes sudden changes in three-dimensional topology, there is an urgent technical need for drone ad hoc network relay communication. In emergency rescue scenarios, drones need to navigate through complex terrains such as mountains and ruins, and quickly adjust communication links. This requires communication protocols to be able to perceive topology changes in real time and respond quickly. At the same time, when a large number of drones are operating simultaneously, it is necessary to rationally allocate limited resources such as bandwidth and meet the diverse quality of service requirements of different tasks in terms of data transmission rate, latency, etc.

[0003] Currently, one applicable solution is to use a communication protocol based on Software-Defined Networking (SDN). This protocol, through a centralized controller, can globally control the network topology and flexibly configure network resources. By collecting information such as the location and link quality of each UAV node through the controller, and dynamically adjusting routing strategies based on the real-time network status, it achieves efficient communication link construction and management to adapt to the rapid topology changes caused by the high-speed movement of UAV nodes.

[0004] However, this solution has significant drawbacks. If the centralized controller fails, the entire communication network risks being paralyzed, indicating poor reliability and stability. Furthermore, as the number of drones increases and the network expands, the controller needs to process massive amounts of information, leading to a surge in computational and storage pressure, processing delays, and an inability to respond promptly to sudden changes in 3D topology. This results in degraded communication quality and makes it difficult to meet the stringent requirements for real-time performance and reliability in complex scenarios. Summary of the Invention

[0005] The purpose of this application is to provide an artificial intelligence-based method, system, electronic device, and storage medium for extracting communication protocol features of unmanned aerial vehicles (UAVs), in order to solve the problem in the prior art that it is difficult to extract communication protocol features in real time and accurately and dynamically optimize the selection of backup nodes in the case of UAVs moving at high speed and causing three-dimensional topological changes.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for extracting features of unmanned aerial vehicle (UAV) communication protocols based on artificial intelligence, comprising:

[0007] Obtain the spatial location set of the base station and the drone swarm, synchronously collect the real-time velocity and acceleration information of each location point in the spatial location set, and construct a communication environment model based on the spatial location set and the real-time velocity and acceleration information;

[0008] The set of spatial locations is mapped to generate a virtual space shape. The spatial offset between the geometric center point of the virtual space shape and the center point of the communication range is calculated. The spatial offset is then used to perform motion compensation based on the acceleration information to generate protocol feature correction parameters. The shape parameters of the virtual space shape are then adjusted by feedback through the protocol feature correction parameters to form a closed-loop geometric model.

[0009] Deep correlation features of the communication environment model are extracted, feature filtering parameters are generated based on the dimensional distribution characteristics of the communication environment model, key features of the deep correlation features are filtered based on the feature filtering parameters, and the filtered key features are fused with preset signal attenuation model parameters to generate protocol layer compensation coefficients.

[0010] Based on the positional relationship features output by the closed-loop geometric modeling, the spatiotemporal variation features of the communication network are extracted. Based on the spatiotemporal variation features, the communication interruption path is predicted. Based on the protocol feature correction parameters, protocol layer compensation coefficients, and communication interruption path, the communication protocol features are extracted and the probability of communication interruption risk is evaluated to generate a dynamic selection strategy for backup communication nodes at the communication protocol layer.

[0011] Optionally, the step of extracting communication protocol features based on the protocol feature correction parameters, protocol layer compensation coefficients, and communication interruption paths, and evaluating the probability of communication interruption risk to generate a dynamic selection strategy for backup communication nodes at the communication protocol layer, includes:

[0012] The modulus of the protocol feature correction parameters, the protocol layer compensation coefficient, and the azimuth and elevation angles of the communication interruption path are combined into a communication protocol feature vector. The communication protocol feature vector is then input into a pre-trained risk probability calculation function to output the communication interruption risk probability of each UAV node relative to the base station.

[0013] Sort all drone nodes in ascending order of communication interruption risk probability, and select the N nodes with the lowest communication interruption risk probability to form a backup communication node set;

[0014] The quality parameters of the main communication link are continuously monitored. When the quality parameters are lower than the first quality threshold, the communication routing node is switched to the highest priority node in the backup node set. The quality parameters of the new communication link are detected within a fixed verification time window after the switch. If the quality parameters of the new communication link do not recover to the second quality threshold, the current node is removed from the backup node set and switched back to the highest priority node among the remaining nodes.

[0015] When there are no available nodes in the set of backup nodes, the interruption risk probability calculation is re-executed and a new set of backup nodes is generated.

[0016] Optionally, the step of adjusting the shape parameters of the virtual space shape through the feedback of the protocol feature correction parameters to form a closed-loop geometric model includes:

[0017] The vertex movement direction and movement distance of the virtual space shape are determined based on the protocol feature correction parameters.

[0018] Based on the vertex movement direction and distance, perform synchronous translation operations on the coordinates of all vertices of the virtual space shape, and reconnect the translated vertex coordinates to generate a new virtual space shape.

[0019] Calculate the coordinates of the geometric center point of the new virtual space shape, and measure the difference between the coordinates of the geometric center point and the coordinates of the center point of the base station communication range as the closed-loop verification offset.

[0020] When the closed-loop verification offset is less than a preset threshold, the current virtual space shape is output as the final geometric model. When the closed-loop verification offset is greater than or equal to the preset threshold, the closed-loop verification offset is used as a new spatial offset and input into the motion compensation process to regenerate the protocol feature correction parameters.

[0021] Optionally, the step of extracting the spatiotemporal variation features of the communication network based on the positional relationship features output by the closed-loop geometric modeling, and predicting the communication interruption path based on the spatiotemporal variation features, includes:

[0022] Obtain the coordinates of all boundary vertices in the final geometric model output by the closed-loop geometric modeling, and calculate the distance between adjacent vertices to form a positional relationship feature sequence;

[0023] The positional relationship feature sequence is analyzed, and the distance fluctuation range within each time window is extracted as the amplitude feature. At the same time, the number of times the distance crosses zero within a unit time is calculated as the frequency feature. The amplitude feature and the frequency feature are combined to form the spatiotemporal variation feature.

[0024] In the spatiotemporal variation features, boundary line segments that satisfy the condition of continuously increasing amplitude and frequency exceeding a preset frequency threshold are detected. The extension direction vectors of these line segments are determined as candidate directions for communication interruption paths. Cluster analysis is performed on the candidate directions of the communication interruption paths, and the direction with the largest cluster center is taken as the communication interruption path.

[0025] Optionally, the mapping of the spatial location set generates a virtual spatial shape, the spatial offset between the geometric center point of the virtual spatial shape and the center point of the communication range is calculated, and motion compensation is performed on the spatial offset using the acceleration information to generate protocol feature correction parameters, including:

[0026] The position coordinates of the UAVs in the spatial location set are input into the spatial mapping rules, and a virtual spatial shape is formed by connecting the position coordinates of the outermost UAV.

[0027] Calculate the arithmetic mean of the coordinates of all vertices of the virtual space shape as the coordinates of the geometric center point, and at the same time measure the spatial offset between the coordinates of the geometric center point and the coordinates of the center point of the base station communication range;

[0028] The acceleration information of each UAV node is multiplied by a preset time interval to obtain the instantaneous velocity change. The instantaneous velocity change is multiplied by the same time interval to obtain the displacement compensation component. The displacement compensation components of all UAV nodes are vector-superimposed to generate an overall displacement compensation vector.

[0029] The spatial offset is subtracted from the overall displacement compensation vector to obtain the protocol feature correction parameters.

[0030] Optionally, the step of extracting deep correlation features of the communication environment model, generating feature filtering parameters based on the dimensionality distribution characteristics of the communication environment model, filtering key features of the deep correlation features based on the feature filtering parameters, and fusing the filtered key features with preset signal attenuation model parameters to generate protocol layer compensation coefficients includes:

[0031] The distance change rate and velocity correlation strength between all UAV node pairs are extracted from the spatial topology network of the communication environment model as deep correlation features.

[0032] Calculate the feature dispersion of the deep association features in the three-dimensional coordinate direction, and take the maximum dispersion value in the three directions as the feature selection parameter;

[0033] Deeply correlated features whose distance change rate is greater than the feature filtering parameter and whose velocity correlation strength is greater than the feature filtering parameter are selected as key features;

[0034] The distance compensation component is obtained by multiplying the distance change rate of the key feature by a preset distance attenuation coefficient.

[0035] Multiply the velocity correlation strength of the key feature by a preset motion disturbance coefficient to obtain the motion compensation component;

[0036] The distance compensation component and the motion compensation component are weighted and superimposed to generate the protocol layer compensation coefficient.

[0037] Optionally, the step of obtaining the spatial location set of the base station and the UAV swarm, synchronously collecting real-time velocity and acceleration information of each location point in the spatial location set, and constructing a communication environment model based on the spatial location set and the real-time velocity and acceleration information includes:

[0038] Obtain the spatial location set of all individual drones in the base station and drone swarm, and synchronously record the real-time velocity and acceleration information corresponding to each location point in the spatial location set;

[0039] A spatial topology network containing base station nodes and drone nodes is established based on the spatial location set, and the real-time speeds of the base station nodes and drone nodes are marked at the corresponding positions in the spatial topology network.

[0040] Based on the changes in spatial distance between adjacent nodes and the real-time speed differences, dynamic connection relationships are constructed in the spatial topology network.

[0041] The acceleration information is integrated as a motion state variable into the dynamic connection relationship to form a communication environment model that can be updated in real time.

[0042] Secondly, this application provides an artificial intelligence-based unmanned aerial vehicle (UAV) communication protocol feature extraction system, comprising:

[0043] The acquisition module is used to acquire the spatial location set of the base station and the UAV swarm, synchronously collect the real-time velocity and acceleration information of each location point in the spatial location set, and construct a communication environment model based on the spatial location set and the real-time velocity and acceleration information.

[0044] The compensation module is used to map the set of spatial locations to generate a virtual space shape, calculate the spatial offset between the geometric center point of the virtual space shape and the center point of the communication range, combine the acceleration information to perform motion compensation on the spatial offset to generate protocol feature correction parameters, and use the protocol feature correction parameters to feed back and adjust the shape parameters of the virtual space shape to form a closed-loop geometric model.

[0045] The filtering module is used to extract deep correlation features of the communication environment model, generate feature filtering parameters based on the dimensional distribution characteristics of the communication environment model, filter key features of the deep correlation features based on the feature filtering parameters, and fuse the filtered key features with preset signal attenuation model parameters to generate protocol layer compensation coefficients.

[0046] The generation module is used to extract the spatiotemporal variation features of the communication network based on the positional relationship features output by the closed-loop geometric modeling, predict the communication interruption path based on the spatiotemporal variation features, extract the communication protocol features based on the protocol feature correction parameters, protocol layer compensation coefficients and communication interruption paths, and evaluate the probability of communication interruption risk, so as to generate a dynamic selection strategy for backup communication nodes of the communication protocol layer.

[0047] Thirdly, this application provides an electronic device, comprising:

[0048] Memory, used to store computer programs;

[0049] A processor is configured to implement the steps of the AI-based unmanned aerial vehicle communication protocol feature extraction method as described in the first aspect above when executing the computer program.

[0050] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the artificial intelligence-based UAV communication protocol feature extraction method described in the first aspect above.

[0051] The AI-based UAV communication protocol feature extraction method provided in this application constructs a communication environment model by acquiring the spatial location and motion information of the base station and the UAV swarm, which can accurately characterize the communication scenario. Closed-loop geometric modeling enables dynamic adjustment of the virtual space shape, improving the compensation capability for positional shifts caused by high-speed UAV movement. Extracting deep correlation features and fusing signal attenuation model parameters to generate protocol layer compensation coefficients enhances the targeting and accuracy of feature extraction. Based on positional relationship features, spatiotemporal change features are extracted and communication interruption paths are predicted. Combined with relevant parameters to extract protocol features and assess risks, this method can effectively cope with three-dimensional topological changes. The resulting dynamic selection strategy for backup communication nodes significantly improves the real-time performance, reliability, and anti-interference capability of UAV communication, ensuring stable and efficient communication links in complex dynamic scenarios.

[0052] Furthermore, the modulus of the protocol feature correction parameters, the protocol layer compensation coefficient, and the azimuth and pitch angles of the communication interruption path are combined into a communication protocol feature vector. This vector is then input into a pre-training function to obtain the communication interruption risk probability of each UAV node. The nodes are then sorted in ascending order of probability, and the top N nodes are selected to form a backup node set. Next, the main link quality is monitored. If the quality falls below a first threshold, the node is switched to the highest priority node in the backup set. If the quality of the new link does not reach a second threshold within a fixed time, the node is removed and the process switches to the next node. If no available nodes are found, a new set is recalculated and generated. This method accurately assesses node interruption risk by constructing feature vectors, dynamically selects backup nodes, and combines main link quality monitoring to achieve intelligent node switching and updates. It can quickly respond to changes in link quality, effectively ensure communication continuity, and improve the stability and anti-interruption capability of the UAV communication network in dynamic scenarios. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart illustrating an artificial intelligence-based method for extracting features from unmanned aerial vehicle (UAV) communication protocols, provided as an embodiment of this application;

[0055] Figure 2 A flowchart illustrating a specific implementation of an artificial intelligence-based method for extracting features from unmanned aerial vehicle (UAV) communication protocols, provided in this application embodiment;

[0056] Figure 3 A specific implementation scenario diagram of an artificial intelligence-based UAV communication protocol feature extraction method provided in this application embodiment;

[0057] Figure 4 This is a schematic diagram of the structure of an artificial intelligence-based unmanned aerial vehicle (UAV) communication protocol feature extraction system provided in an embodiment of this application. Detailed Implementation

[0058] In the three-dimensional topology change scenario of UAV ad hoc network relay communication, existing communication protocols based on software-defined networking (SDN) have significant shortcomings. A failure of the centralized controller can lead to network paralysis, resulting in poor reliability. Furthermore, as the number of UAVs increases and the network scales up, the controller is prone to delays when processing massive amounts of information, making it unable to respond to topology changes in a timely manner and failing to meet the requirements for real-time and reliable communication.

[0059] To address the aforementioned issues, this application proposes an AI-based method for extracting features from unmanned aerial vehicle (UAV) communication protocols. This method constructs a communication environment model by acquiring spatial position, velocity, and acceleration information of the base station and the UAV swarm. Through closed-loop geometric modeling, the virtual space shape is dynamically adjusted, deep correlation features are extracted, and signal attenuation model parameters are fused to generate compensation coefficients. Furthermore, spatiotemporal variation features are extracted by combining positional relationship features, and communication interruption paths are predicted. Finally, protocol features are extracted, and risks are assessed to generate a dynamic backup node selection strategy. This solution does not rely on a centralized controller, can accurately extract protocol features in real time, and dynamically optimize backup node selection. It effectively solves the problems of poor reliability, processing latency, and difficulty in handling topology changes in existing solutions, improving the real-time performance and stability of communication.

[0060] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] The core of this application is to provide a method for extracting features of unmanned aerial vehicle (UAV) communication protocols based on artificial intelligence. A flowchart of one specific implementation is shown below. Figure 1As shown, the method includes:

[0062] S101. Obtain the spatial location set of the base station and the UAV swarm, synchronously collect the real-time velocity and acceleration information of each location point in the spatial location set, and construct a communication environment model based on the spatial location set and the real-time velocity and acceleration information.

[0063] Optionally, step S101 may specifically include the following steps:

[0064] S1011. Obtain the spatial location set of all individual drones in the base station and drone swarm, and synchronously record the real-time velocity and acceleration information corresponding to each location point in the spatial location set.

[0065] S1012. Based on the spatial location set, establish a spatial topology network including base station nodes and drone nodes, and mark the real-time speeds of the base station nodes and drone nodes at the corresponding positions in the spatial topology network.

[0066] S1013. Based on the changes in spatial distance between adjacent nodes and the real-time speed differences, construct dynamic connection relationships in the spatial topology network;

[0067] S1014. The acceleration information is integrated as a motion state variable into the dynamic connection relationship to form a communication environment model that can be updated in real time.

[0068] In the above scheme, the spatial location set refers to the information set reflecting the spatial distribution of each UAV in the base station and UAV swarm, including the three-dimensional coordinate data of each location point, which can be used to construct spatial associations between nodes. Real-time velocity refers to the dynamic information reflecting the current movement state of each location point, including the speed and direction parameters, which can be used to analyze the movement trend of nodes. Acceleration information refers to the dynamic signal reflecting the velocity change characteristics of each location point, including the rate of velocity change and direction parameters, which can be used to predict the positional changes of nodes. The communication environment model is a dynamic model constructed by integrating the spatial location set, real-time velocity, and acceleration information, including the connection relationship and motion state association between the base station and UAV nodes, which can be used to characterize the dynamic features of the communication scenario.

[0069] In this embodiment, step S1011 first uses the GPS module on the UAV and the positioning system built into the base station to acquire the spatial position data of the base station and each UAV in the UAV swarm in real time, forming a spatial position set. This data is presented in the form of three-dimensional coordinates (x, y, z). At the same time, using the speed sensors and acceleration sensors on the UAV and the base station, the real-time speed and acceleration information corresponding to each position point is collected synchronously. The real-time speed includes the speed magnitude and direction of motion, and the acceleration also includes the acceleration magnitude and direction of change. This information is temporarily stored in the data acquisition terminal to provide raw data for subsequent steps.

[0070] Secondly, based on the acquired spatial location set in step S1012, a graph theory algorithm is used to treat the base station and each UAV as network nodes. A spatial topology network is constructed in virtual space based on their three-dimensional coordinates, with the node positions in the network corresponding to their actual spatial locations. Then, using data mapping technology, the real-time speed information of the base station nodes and UAV nodes collected in step S1011 is labeled next to the corresponding nodes in the spatial topology network according to the node correspondence. This allows for a direct visual representation of the position and current speed of each node within the topology network.

[0071] Next, based on the spatial topology network constructed in step S1012, step S1013 uses time series analysis to calculate the spatial distance between two adjacent nodes at two consecutive time points, obtaining the distance change. Positive values ​​indicate an increase in distance, while negative values ​​indicate a decrease. Simultaneously, the difference in magnitude and direction of the real-time velocities of these two adjacent nodes is calculated; a greater velocity difference indicates a more asynchronous motion. Then, combining the distance change and velocity difference, a dynamic connection algorithm is used to construct dynamic connections between nodes in the spatial topology network. A strong connection is formed when both the distance change and velocity difference are small, and vice versa.

[0072] Finally, step S1014 further processes the dynamic connection relationship obtained in step S1013. Specifically, the acceleration information collected in step S1011 is used as a motion state variable and integrated into the dynamic connection relationship through data fusion technology. Because acceleration reflects the trend of velocity changes, when the acceleration of a node changes, its future position and velocity will change accordingly, thus affecting the connection relationship between nodes. By introducing the acceleration factor into the dynamic connection relationship and using a real-time update algorithm, the connection relationship can be automatically adjusted with changes in acceleration, ultimately forming a real-time updatable communication environment model. This model can reflect the dynamic changes in the spatial position, motion state, and node connection relationship between the base station and the UAV swarm in real time.

[0073] In a practical application, during a drone inspection mission in a certain area, there is a base station P and a drone swarm consisting of drones Q1, Q2, and Q3. The GPS modules on the drones obtain the three-dimensional coordinates of base station P as (100, 200, 50), and the coordinates of drones Q1 as (150, 220, 60), Q2 as (130, 180, 55), and Q3 as (170, 210, 65). Simultaneously, speed and acceleration sensors collect data showing that base station P has a velocity of 0 and acceleration of 0; Q1 has a velocity of (5, 3, 2) m / s and an acceleration of (1, 0.5, 0.3) m / s²; Q2 has a velocity of (4, -2, 1) m / s and an acceleration of (0.8, -0.4, 0.2) m / s²; and Q3 has a velocity of (6, 2, 3) m / s and an acceleration of (1.2, 0.6, 0.4) m / s². This information... The data is stored in the data acquisition terminal. Then, a graph theory algorithm is used to construct a spatial topology network in virtual space using base station P and nodes Q1, Q2, and Q3 as nodes, based on their three-dimensional coordinates. Data mapping technology is then used to label the velocity information of each node next to its corresponding node. Next, based on this topology network, time series analysis is used to calculate that the distance change between Q1 and Q2 at two consecutive time points is -3 meters, indicating a small velocity difference. A dynamic connection algorithm is used to establish a strong connection between them. Conversely, the distance change between Q1 and Q3 is 5 meters, indicating a larger velocity difference and a weaker connection. Finally, data fusion technology integrates the acceleration information of each node into the dynamic connection relationship. When the acceleration of Q2 changes to (0.7, -0.5, 0.1) m / s², the real-time update algorithm automatically adjusts its connection relationship with surrounding nodes, ultimately forming a real-time updatable communication environment model.

[0074] The overall solution of S101 described above, by acquiring the spatial location set of the base station and the UAV swarm, and synchronously collecting real-time velocity and acceleration information of each location point, constructs a communication environment model that can comprehensively and dynamically present the spatial distribution, motion state, and connection relationships between base station and UAV nodes. This model, through the integration and real-time updating of location, velocity, and acceleration information, provides accurate and realistic basic data for subsequent steps such as closed-loop geometric modeling and feature extraction. This makes the entire communication protocol feature extraction process more adaptable to dynamic changes such as the high-speed movement of UAVs, effectively improving the perception capability of the communication environment and the accuracy of subsequent processing.

[0075] S102. Map the set of spatial locations to generate a virtual space shape, calculate the spatial offset between the geometric center point of the virtual space shape and the center point of the communication range, combine the acceleration information to perform motion compensation on the spatial offset to generate protocol feature correction parameters, and use the protocol feature correction parameters to feed back and adjust the shape parameters of the virtual space shape to form a closed-loop geometric model.

[0076] Optionally, step S102 may specifically include the following steps:

[0077] S1021. Input the position coordinates of the UAVs in the spatial location set into the spatial mapping rules, and form a virtual spatial shape by connecting the position coordinates of the outermost UAVs.

[0078] S1022. Calculate the arithmetic mean of the coordinates of all vertices of the virtual space shape as the coordinates of the geometric center point, and at the same time measure the spatial offset between the coordinates of the geometric center point and the coordinates of the center point of the base station communication range.

[0079] S1023. Multiply the acceleration information of each UAV node by a preset time interval to obtain the instantaneous velocity change, multiply the instantaneous velocity change by the same time interval to obtain the displacement compensation component, and perform vector superposition of the displacement compensation components of all UAV nodes to generate an overall displacement compensation vector.

[0080] S1024. Perform vector subtraction on the spatial offset and the overall displacement compensation vector to obtain the protocol feature correction parameters.

[0081] S1025. Determine the vertex movement direction and movement distance of the virtual space shape based on the protocol feature correction parameters;

[0082] S1026. Perform synchronous translation operation on the coordinates of all vertices of the virtual space shape according to the vertex movement direction and movement distance, and reconnect the translated vertex coordinates to generate a new virtual space shape.

[0083] S1027. Calculate the coordinates of the geometric center point of the new virtual space shape, and measure the difference between the coordinates of the geometric center point and the coordinates of the center point of the base station communication range as the closed-loop verification offset.

[0084] S1028. When the closed-loop verification offset is less than a preset threshold, the current virtual space shape is output as the final geometric model. When the closed-loop verification offset is greater than or equal to the preset threshold, the closed-loop verification offset is used as a new spatial offset and input into the motion compensation process to regenerate the protocol feature correction parameters.

[0085] In the above scheme, the virtual space shape refers to a geometric figure generated based on the spatial position set of UAVs, reflecting the distribution range of the UAV swarm. It includes vertex coordinates and other data to visually represent the spatial distribution of the UAV swarm. The geometric center point is the point corresponding to the arithmetic mean of the coordinates of all vertices of the virtual space shape, representing the center position of the shape. The communication range center point is the center position of the area covered by the base station's communication; it is a fixed or known point. The spatial offset is the positional difference between the geometric center point and the communication range center point, containing distance information in three dimensions, used to measure the positional deviation between the two. The instantaneous velocity change is the change in the speed of the UAV within a preset time interval, reflecting the rate of velocity change; the displacement compensation component is calculated from the instantaneous velocity change and is used to compensate for the positional offset. The overall displacement compensation vector is the vector obtained by superimposing the displacement compensation components of all UAVs, used to comprehensively reflect the impact of the UAV swarm's movement on its position. The protocol feature correction parameter is a parameter obtained by calculating the spatial offset and the overall displacement compensation vector, used to adjust the virtual space shape. Shape parameters describe the shape of the virtual space, such as vertex coordinates; the closed-loop verification offset is the difference between the geometric center point of the new virtual space shape and the center point of the communication range, used to verify the adjustment effect. The preset threshold is the standard value for judging whether the closed-loop verification offset meets the requirements.

[0086] In this embodiment, the position coordinates of all UAVs in the spatial location set are first collected in step S1021. The boundary detection algorithm is used to find the outermost UAVs, which are individuals located at the edge of the UAV swarm and relatively far out. Then, according to the spatial mapping rules, the position coordinates of the outermost UAVs are connected sequentially using geometric connection technology to form a closed virtual space shape. This shape can roughly outline the distribution range of the UAV swarm. For example, if the position coordinates of the outermost UAVs are (2,3,4), (5,6,7), and (8,9,10), the connection will form a triangular virtual space shape.

[0087] Step S1022, based on the virtual space shape obtained in step S1021, first extracts the coordinates of all vertices of the virtual space shape. Then, it uses an arithmetic mean algorithm to calculate the average of these vertex coordinates in the x, y, and z dimensions, obtaining the coordinates of the geometric center point of the virtual space shape. Simultaneously, using a coordinate difference calculation method, it calculates the differences in the x, y, and z directions between the geometric center point coordinates and the coordinates of the pre-set center point of the base station's communication coverage area. These differences are combined to form the spatial offset. For example, if the geometric center point coordinates are (3,4,5) and the base station's communication coverage area center point coordinates are (5,5,6), then the spatial offset is -2 in the x direction, -1 in the y direction, and -1 in the z direction.

[0088] Secondly, step S1023 utilizes the acceleration information of each UAV node. First, a preset time interval is set. Using kinematic formulas, the acceleration information of each UAV is multiplied by the preset time interval to obtain the instantaneous velocity change of each UAV during this time. Then, the instantaneous velocity change is multiplied by the same preset time interval to calculate the displacement compensation component of each UAV. Finally, using a vector superposition method, the displacement compensation components of all UAVs are accumulated in the x, y, and z directions respectively. The sum obtained is the overall displacement compensation vector. For example, if a UAV's acceleration is (1,2,3) m / s², the preset time interval is 2 seconds, its instantaneous velocity change is (2,4,6) m / s, and its displacement compensation component is (4,8,12) m. After superimposing the displacement compensation components of multiple UAVs, the overall displacement compensation vector is output.

[0089] Next, step S1024 processes the obtained spatial offset and overall displacement compensation vector using vector subtraction. The spatial offset vector is subtracted from the overall displacement compensation vector; that is, the components of the two vectors in the x, y, and z directions are subtracted respectively. The result is the protocol feature correction parameter, which reflects the magnitude and direction of the adjustment needed to the virtual space shape. For example, if the spatial offset vector is (-2, -1, -1) and the overall displacement compensation vector is (10, 15, 20), then the protocol feature correction parameter is (-2-10, -1-15, -1-20) = (-12, -16, -21).

[0090] Then, in step S1025, the adjustment direction and movement distance are determined based on the obtained protocol feature correction parameters. The positive and negative values ​​of the x, y, and z components of the protocol feature correction parameters represent the movement direction of the virtual space shape vertices in the corresponding directions. Positive values ​​indicate movement in the positive direction, and negative values ​​indicate movement in the negative direction. The magnitude of the component value represents the distance the vertex needs to move. In step S1026, according to the vertex movement direction and movement distance determined in step S1025, the coordinates of all vertices of the virtual space shape are processed. Using coordinate translation technology, the x, y, and z coordinates of each vertex are adjusted according to the corresponding movement direction and distance. Then, all the translated vertex coordinates are reconnected in the original connection order to form a new virtual space shape. For example, the vertex coordinates (2,3,4) become (2-12,3-16,4-21) = (-10,-13,-17) after adjustment according to the above movement direction and distance. Other vertices are also adjusted in the same way and then reconnected.

[0091] Finally, in step S1027, for the new virtual space shape, the arithmetic mean algorithm is used to calculate the coordinates of the geometric center point of all vertices of the new virtual space shape. Then, the coordinate difference calculation method is used to calculate the difference between the coordinates of this new geometric center point and the coordinates of the center point of the base station communication range in the x, y, and z directions. The above differences are combined as the closed-loop verification offset. Next, in step S1028, it is determined whether the obtained closed-loop verification offset meets the requirements. First, a preset threshold is set, and the components of the closed-loop verification offset in the x, y, and z directions are compared with the preset threshold respectively. If the absolute values ​​of all three components are less than the preset threshold, it means that the current virtual space shape has met the accuracy requirements, and the shape is output as the final geometric model. If the absolute value of any component is greater than or equal to the preset threshold, the closed-loop verification offset is used as a new spatial offset and is resubmitted into the motion compensation process starting in step S1023. The instantaneous velocity change, displacement compensation components, etc. are recalculated to generate new protocol feature correction parameters, and the virtual space shape is adjusted again until the closed-loop verification offset is less than the preset threshold.

[0092] In a practical application scenario involving collaborative drone operations, there is a base station P and a drone swarm, consisting of drones J, K, L, M, and N. First, the position coordinates of these drones are collected. A boundary detection algorithm is used to determine that the outermost drones are J, L, and N, with coordinates (10,10,5), (20,15,5), and (15,20,5) respectively. Based on spatial mapping rules, these three coordinates are sequentially connected using geometric connection techniques to form a virtual space shape. The coordinates of the vertices of this virtual space shape are extracted, and the coordinates of the geometric center point are calculated using an arithmetic mean algorithm. The center point coordinates of the communication range of base station P are (18, 18, 5). The spatial offset is calculated as -3 in the x-direction, -3 in the y-direction, and 0 in the z-direction using the coordinate difference method. Then, using the collected acceleration information of each UAV, a preset time interval of 1 second is set. The acceleration of UAV J is (2, 3, 0) m / s², and its instantaneous velocity change is (2, 3, 0) m / s. The displacement compensation component is (2×1, 3×1, 0×1) = (2, 3, 0) m. Similarly, the displacement compensation components of K, L, M, and N are calculated. All displacement compensation components are accumulated using the vector superposition method to obtain the overall displacement compensation vector as (5, 6, 0) m. The spatial offset vector (-3, -3, 0) and the overall displacement compensation vector (5, 6, 0) are subtracted to obtain the protocol feature correction parameter as (-3-5, -3-6, 0-0) = (-8, -9, 0). Next, based on this parameter, the vertex movement directions are determined to be negative x-direction and negative y-direction, with movement distances of 8 units and 9 units respectively. Following these directions and distances, coordinate translation techniques are used to adjust the coordinates of the three vertices of the virtual space shape: the coordinates of J become (10-8, 10-9, 5-0) = (2, 1, 5), the coordinates of L become (20-8, 15-9, 5-0) = (12, 6, 5), and the coordinates of N become (15-8, 20-9, 5-0) = (7, 11, 5). The adjusted vertices are then reconnected to form a new virtual space shape. Finally, the coordinates of the geometric center point of the new virtual space shape are calculated using an arithmetic mean algorithm. The closed-loop verification offset of the center point (18,18,5) of the communication range of the base station is calculated by the coordinate difference method. The offset is -11 in the x direction, -12 in the y direction, and 0 in the z direction. The preset threshold is set to 10. Since the absolute values ​​of the components of the closed-loop verification offset in the x and y directions are both greater than 10, this offset is used as a new spatial offset and re-substituted into the motion compensation process. After multiple adjustments, the final closed-loop verification offset is less than the preset threshold. The virtual space shape at this time is output as the final geometric model.

[0093] The overall scheme of S102 described above generates a virtual space shape and calculates the spatial offset. It then uses acceleration information for motion compensation to obtain protocol feature correction parameters. These parameters are then used to adjust the virtual space shape, forming a closed-loop geometric model. This process dynamically corrects the deviation between the virtual space shape and the actual communication range caused by the drone's movement, making the constructed geometric model more closely match the real-time spatial distribution of the drone swarm. This provides an accurate spatial geometric basis for subsequent extraction of communication protocol features and enhances adaptability to high-speed drone movement scenarios.

[0094] S103. Extract the deep correlation features of the communication environment model, generate feature filtering parameters according to the dimensional distribution characteristics of the communication environment model, filter the key features of the deep correlation features based on the feature filtering parameters, and fuse the filtered key features with the preset signal attenuation model parameters to generate protocol layer compensation coefficients.

[0095] Optionally, step S103 may specifically include the following steps:

[0096] S1031. Extract the distance change rate and velocity correlation strength between all UAV node pairs from the spatial topology network of the communication environment model as deep correlation features.

[0097] S1032. Calculate the feature dispersion of the deep correlation features in the three-dimensional coordinate direction, and take the maximum dispersion value in the three directions as the feature filtering parameter.

[0098] S1033. Select deep association features that have a distance change rate greater than the feature selection parameter and a velocity association strength greater than the feature selection parameter as key features;

[0099] S1034. Multiply the distance change rate of the key feature by a preset distance attenuation coefficient to obtain the distance compensation component.

[0100] S1035. Multiply the velocity correlation strength of the key feature by a preset motion interference coefficient to obtain the motion compensation component.

[0101] S1036. Weighted superposition of distance compensation components and motion compensation components to generate protocol layer compensation coefficients.

[0102] In the above scheme, deep correlation features are features extracted from the communication environment model that reflect the relationships between UAV nodes. These include the distance change rate and velocity correlation strength between node pairs, used to represent the dynamic connections between nodes. The distance change rate refers to how quickly the distance between two UAV nodes changes over time, including both the numerical value and the trend of the change. The velocity correlation strength is a parameter reflecting the degree of correlation between the velocities of two UAV nodes; a larger value indicates a closer velocity relationship. Feature dispersion refers to the dispersion of deep correlation features along the three-dimensional coordinate directions, used to measure the degree of feature discreteness. Feature filtering parameters are parameters determined from the maximum feature dispersion values ​​in the three coordinate directions, used to filter key features. Key features are the deep correlation features obtained after filtering by the feature filtering parameters; these are features that have a significant impact on communication. Signal attenuation model parameters are pre-set parameters used to calculate compensation components, including the distance attenuation coefficient and the motion interference coefficient. The distance compensation component is the quantity obtained by multiplying the distance change rate of the key feature by the distance attenuation coefficient, used to compensate for the impact of distance changes. The motion compensation component is the quantity obtained by multiplying the velocity correlation strength of the key feature by the motion interference coefficient, used to compensate for interference caused by motion. The protocol layer compensation coefficient is a coefficient obtained by weighted superposition of the distance compensation component and the motion compensation component, and is used to correct relevant features of the communication protocol.

[0103] In this embodiment, step S1031 first starts with the spatial topology network of the communication environment model. Using a feature extraction algorithm based on association rules, all UAV nodes are traversed to find each pair of nodes. Then, by calculating the ratio of the distance difference between any two nodes at consecutive time points to the time interval, the distance change rate between them is obtained. Simultaneously, by analyzing the correlation between the velocity vectors of two nodes, the velocity association strength is obtained. These two data points are used together as deep association features reflecting the dynamic relationship between nodes. For example, in a network containing UAVs A, B, and C, it is calculated that if the distance between A and B decreases by 0.3 meters per second, the velocity association strength is 0.7; if the distance between A and C increases by 0.5 meters per second, the velocity association strength is 0.4, and so on.

[0104] Secondly, based on the deep correlation features in step S1032, the standard deviation algorithm is used to calculate the distribution dispersion of these features in the three-dimensional coordinate directions of x, y, and z, i.e., the feature dispersion degree. First, each deep correlation feature is decomposed into the three coordinate directions, and then the standard deviation of all feature values ​​in each direction is calculated. The larger the standard deviation, the higher the dispersion degree. Then, the feature with the largest dispersion degree in these three directions is selected as the feature selection parameter. For example, if the calculated standard deviation is 2.2 in the x direction, 2.8 in the y direction, and 2.5 in the z direction, then 2.8 is determined as the feature selection parameter.

[0105] Next, in step S1033, deep association features are filtered using feature filtering parameters. The distance change rate and velocity association strength of each deep association feature are examined one by one. Only when the distance change rate and velocity association strength of a deep association feature are both greater than the feature filtering parameters is the feature retained as a key feature. For example, if the feature filtering parameter is 2.8, and a node pair has a distance change rate of 3.0 and a velocity association strength of 3.2, it meets the conditions to become a key feature, while features with a distance change rate of 2.7 or a velocity association strength of 2.6 are excluded.

[0106] Then, the selected key features are processed through steps S1034 and S1035. First, the preset signal attenuation model in the system is called to obtain the distance attenuation coefficient. Then, the distance change rate of the key feature is multiplied by this distance attenuation coefficient, and the result is the distance compensation component. The distance attenuation coefficient is a fixed value preset according to the law of signal attenuation with distance when propagating in space. For example, if the distance change rate of the key feature is 3.0 and the distance attenuation coefficient is 0.5, then the distance compensation component is 3.0 × 0.5 = 1.5. Similarly, the motion interference coefficient is obtained from the preset signal attenuation model. The velocity correlation strength of the key feature is multiplied by the motion interference coefficient to obtain the motion compensation component. The motion interference coefficient is a fixed value preset according to the degree of interference of the motion state on the signal. For example, if the velocity correlation strength is 3.2 and the motion interference coefficient is 0.4, then the motion compensation component is 3.2 × 0.4 = 1.28.

[0107] Finally, in step S1036, the distance compensation component and the motion compensation component are fused. Based on the different weights of the impact of distance changes and motion interference on the communication protocol, the two components are weighted and superimposed. The weight values ​​are determined in advance based on a large amount of experimental data. Specifically, the distance compensation component is multiplied by its corresponding weight, and then the motion compensation component is multiplied by its corresponding weight. The result is the protocol layer compensation coefficient. For example, if the weight of the distance compensation component is 0.6 and the weight of the motion compensation component is 0.4, then the protocol layer compensation coefficient is 1.5 × 0.6 + 1.28 × 0.4 = 0.9 + 0.512 = 1.412.

[0108] In practical applications, in a joint UAV monitoring scenario, the spatial topology network of the communication environment model includes UAVs P, Q, R, and S. First, deep correlation features are extracted from this network. For example, the distance change rate between P and Q is 3.2 m / s, and the velocity correlation strength is 3.0; the distance change rate between Q and R is 2.9 m / s, and the velocity correlation strength is 3.5; the distance change rate between R and S is 3.6 m / s, and the velocity correlation strength is 3.3; and the distance change rate between S and P is 2.7 m / s, and the velocity correlation strength is 2.8. Next, the feature dispersion in the x, y, and z directions is calculated, which are 3.0, 3.3, and 2.9 respectively. 3.3 is selected as the feature filtering parameter. Then, features with both distance change rate and velocity correlation strength greater than 3.3 are selected, namely R and S. The corresponding distance change rate of 3.6 > 3.3, and the velocity correlation strength of 3.3 equals 3.3, which does not meet the requirements. Assuming that the distance change rate between Q and S in the supplementary data is 3.4 and the velocity correlation strength is 3.5, it meets the conditions and becomes a key feature. Then, the distance change rate of this key feature 3.4 is multiplied by the preset distance attenuation coefficient 0.6 to obtain the distance compensation component 3.4 × 0.6 = 2.04. The velocity correlation strength 3.5 is multiplied by the preset motion interference coefficient 0.5 to obtain the motion compensation component 3.5 × 0.5 = 1.75. Finally, these two components are weighted and superimposed, with the distance compensation component weighted at 0.5 and the motion compensation component weighted at 0.5. The protocol layer compensation coefficient is 2.04 × 0.5 + 1.75 × 0.5 = 1.02 + 0.875 = 1.895.

[0109] The overall solution in S103 described above extracts deep-level correlation features from the communication environment model, selects key features based on the dimensional distribution characteristics of these features, and then fuses these key features with signal attenuation model parameters to generate protocol layer compensation coefficients. This process accurately captures features that significantly impact communication. By generating compensation coefficients, it effectively corrects the impact of distance changes and motion interference on the communication protocol, improving the accuracy and reliability of the communication protocol features and making the communication system more adaptable to the dynamically changing operating environment of UAVs.

[0110] S104. Based on the positional relationship features output by the closed-loop geometric modeling, extract the spatiotemporal variation features of the communication network, predict the communication interruption path according to the spatiotemporal variation features, extract the communication protocol features based on the protocol feature correction parameters, protocol layer compensation coefficients and communication interruption path, and evaluate the probability of communication interruption risk, so as to generate a dynamic selection strategy for backup communication nodes of the communication protocol layer.

[0111] Optionally, step S104 may specifically include the following steps:

[0112] S1041. Obtain the coordinates of all boundary vertices in the final geometric model output by the closed-loop geometric modeling, and calculate the distance between adjacent vertices to form a positional relationship feature sequence.

[0113] S1042. Analyze the positional relationship feature sequence, extract the distance fluctuation range within each time window as the amplitude feature, and calculate the number of times the distance crosses zero within a unit time as the frequency feature. Combine the amplitude feature and the frequency feature to form the spatiotemporal change feature.

[0114] S1043. In the spatiotemporal change features, detect boundary line segments that satisfy the condition of continuous amplitude growth and frequency exceeding a preset frequency threshold, determine the extension direction vector of these line segments as candidate directions of communication interruption paths, and perform cluster analysis on the candidate directions of communication interruption paths, taking the direction of the largest cluster center as the communication interruption path.

[0115] S1044. Combine the modulus of the protocol feature correction parameters, the protocol layer compensation coefficient, the azimuth and elevation angles of the communication interruption path into a communication protocol feature vector, and input the communication protocol feature vector into a pre-trained risk probability calculation function to output the communication interruption risk probability of each UAV node relative to the base station.

[0116] S1045. Sort the communication interruption risk probabilities of all UAV nodes in ascending order, and select the N nodes with the lowest communication interruption risk probabilities to form a set of backup communication nodes.

[0117] S1046. Continuously monitor the quality parameters of the main communication link. When the quality parameters are lower than the first quality threshold, switch the communication routing node to the highest priority node in the backup node set, and detect the quality parameters of the new communication link within the fixed verification time window after the switch. If the quality parameters of the new communication link do not recover to the second quality threshold, remove the current node from the backup node set and switch back to the highest priority node among the remaining nodes.

[0118] S1047. When there are no available nodes in the set of backup nodes, re-execute the interruption risk probability calculation and generate a new set of backup nodes.

[0119] In the above scheme, the positional relationship feature sequence is a sequence composed of the distances between adjacent boundary vertices in the final geometric model output by closed-loop geometric modeling, used to reflect changes in the positional relationship between vertices. The spatiotemporal variation feature is a combination of amplitude and frequency features within each time window. The amplitude feature is the range of distance fluctuations, reflecting the magnitude of the change, while the frequency feature is the number of times the distance crosses zero per unit time, reflecting the frequency of change. The communication interruption path refers to the directional path where communication interruption may occur, determined by the extension direction of boundary segments that meet specific conditions. The communication protocol feature vector is a vector composed of the magnitude of the protocol feature correction parameters, the protocol layer compensation coefficient, and the azimuth and elevation angles of the communication interruption path, used to assess the risk of communication interruption. The risk probability calculation function is a pre-trained function used to calculate the probability of communication interruption risk. The communication interruption risk probability is the likelihood of each UAV node experiencing a communication interruption relative to the base station. The backup communication node set is a set of the N nodes with the lowest communication interruption risk probability, used to replace the main link in case of failure. The quality parameters of the main communication link are indicators that measure the communication quality of the main communication link. The first quality threshold and the second quality threshold are standard values ​​for judging whether the communication link quality meets the standards. A fixed verification time window is a fixed period of time used to test the quality of the new link after switching nodes.

[0120] In this embodiment of the application, firstly, the three-dimensional coordinates of all boundary vertices are extracted from the final geometric model output by the closed-loop geometric modeling in step S1041. The Euclidean distance calculation formula is used to calculate the straight-line distance between each pair of adjacent boundary vertices in turn. These distances are arranged in the order of vertex connection to form a positional relationship feature sequence. For example, the boundary vertex coordinates of the final geometric model are (1,1,1), (3,3,3), and (5,5,5), and the distances between adjacent vertices are 3.46 and 3.46, respectively, forming the sequence [3.46,3.46]. Step S1042: Based on the sequence, the sliding window technique is used to divide the time window into multiple time windows. The distance values ​​in each window are analyzed. The amplitude feature is obtained by subtracting the minimum value from the maximum value of the distance in the window. At the same time, the number of times the distance value changes from greater than zero to less than zero or from less than zero to greater than zero per unit time is counted as the frequency feature. The two are combined into a spatiotemporal variation feature. For example, if the maximum distance value in a certain window is 4.0 and the minimum distance value is 2.0, the amplitude feature is 2.0. If the distance changes to zero twice per unit time, the frequency feature is 2. The two are combined into a spatiotemporal variation feature of (2.0,2).

[0121] Next, in step S1043, based on the spatiotemporal variation characteristics, a trend analysis algorithm is used to screen out boundary line segments whose amplitude characteristics continuously increase and whose frequency characteristics exceed a preset frequency threshold. The direction vector calculation method is used to obtain the extension direction vectors of these line segments as candidate directions for communication interruption paths. Then, the K-means clustering algorithm is used to cluster the candidate directions, and the cluster center direction containing the most candidate directions is determined as the communication interruption path. For example, if there are three candidate direction vectors (0.6, 0.5, 0.3), (0.7, 0.4, 0.2), and (0.2, 0.3, 0.8), the largest cluster center direction after clustering is (0.65, 0.45, 0.25), which is the communication interruption path.

[0122] Then, as Figure 2 As shown, step S1044 calculates the magnitude of the protocol feature correction parameter vector and obtains the protocol layer compensation coefficient, as well as the azimuth and elevation angles of the communication interruption path. These four data points are combined sequentially into a communication protocol feature vector, which is then input into a risk probability calculation function trained based on historical data to obtain the communication interruption risk probability of each UAV node relative to the base station. For example, if the vector is (3.0, 2.2, 45°, 30°), the risk probability of a certain node after inputting into the function is 0.15. Next, step S1045 sorts the risk probabilities of all nodes in ascending order from smallest to largest, and selects the top N to form a backup communication node set. For example, if the probabilities after sorting are 0.1, 0.15, 0.2, and 0.3, the first two nodes are selected.

[0123] Finally, the quality parameters of the main communication link, i.e., signal clarity, are continuously tracked through steps S1046 and S1047. When the quality parameter is lower than the first quality threshold, the communication route is switched to the node with the highest priority (i.e., the lowest probability) in the backup set. Subsequently, the quality of the new link is detected within a set fixed verification time window. If it does not reach the second quality threshold, the node is removed from the backup set, and the route is switched to the node with the highest priority among the remaining nodes. If there are no available nodes in the backup set, the risk probability calculation step in step S1044 is re-executed to generate a new set of backup communication nodes. For example, if the quality parameter of the main communication link is 55, which is lower than the first quality threshold of 60, the route is switched to a node with a probability of 0.1. If the new link quality of 65 does not reach the second quality threshold of 70 within 8 seconds, the node is removed, and the route is switched to a node with a probability of 0.15. If this node also fails, a new set is recalculated and generated.

[0124] In practical applications, in a UAV collaborative reconnaissance scenario, the final geometric model boundary vertex coordinates output by closed-loop geometric modeling are (2,2,2), (4,4,4), (6,6,6), and (2,2,2). The distances between adjacent vertices are calculated to be 3.46, 3.46, and 5.66, forming a positional relationship feature sequence [3.46, 3.46, 5.66]. Then, a sliding window is used to divide the time window. Within a certain window, the maximum distance is 6.0, the minimum distance is 3.0, the amplitude feature is 3.0, and the zero-crossing change occurs 3 times per unit time, resulting in a frequency feature of 3, which is combined to form a spatiotemporal variation feature of (3.0, 3). Based on the spatiotemporal variation feature, boundary line segments with continuously increasing amplitude and frequencies exceeding the threshold of 2 are selected, with their extension direction vectors being (0.5, 0.5, 0.5) and (0.6, 0.6, 0.6). This is then processed by K... -means clustering yields the direction of the largest cluster center (0.55, 0.55, 0.55) as the communication interruption path; the protocol feature correction parameter magnitude is calculated to be 4.0, the protocol layer compensation coefficient is 2.5, the interruption path azimuth angle is 60°, and the pitch angle is 30°, which are combined into a feature vector (4.0, 2.5, 60°, 30°). This vector is input into the pre-training function to obtain the interruption risk probabilities of UAVs H, I, J, and K as 0.12, 0.18, 0.25, and 0.3, respectively. After sorting in S1045, the top two are selected to form the backup set {H, I}; finally, if the main link quality parameter 58 is found to be lower than the first threshold 60, the connection is switched to H. If the new link quality 68 fails to reach the second threshold 70 within 8 seconds, H is removed and the connection is switched to I. The new link quality 72 meets the standard. If I also fails, a new backup set is recalculated and generated in S1047.

[0125] The overall solution of S104 described above extracts spatiotemporal variation features based on the positional relationship features of closed-loop geometric modeling, accurately predicts communication interruption paths, assesses risk probabilities by combining relevant parameters, and generates a dynamic selection strategy for backup nodes. By monitoring the quality of the main link in real time and dynamically switching backup nodes, it can promptly respond to communication interruption risks, ensure the continuity and stability of communication, and effectively improve the anti-interference capability and reliability of the UAV communication network in complex dynamic scenarios.

[0126] The following is a complete example for steps 101-104, such as Figure 3As shown, in a drone collaborative operation scenario in a certain area, there exists a base station P and a drone swarm, consisting of drones A, B, C, and D. First, the spatial location set of base station P and drones A, B, C, and D is obtained, acquiring their three-dimensional coordinates. Simultaneously, the real-time velocity of each location is recorded, such as drone A's velocity being 4 m / s southeast, drone B's velocity being 3 m / s northeast, etc. Acceleration information is also recorded, such as drone C's acceleration being 1.5 m / s² southward, etc. Based on these spatial location sets, a spatial topology network including base station P and each drone is established, with real-time velocities labeled. Then, dynamic connection relationships are constructed based on the distance changes and velocity differences between adjacent nodes. Finally, the acceleration information is integrated to form a real-time updatable communication environment model.

[0127] Next, the position coordinates of UAVs A, B, C, and D are input into the spatial mapping rules. Connecting the position coordinates of the outermost A, C, and D forms a triangular virtual space shape. The arithmetic mean of the coordinates of all vertices of this shape is calculated to obtain the geometric center coordinates as (4,5,6), and the communication range center coordinates of base station P are (5,5,6). The spatial offset between the two in the x-direction is -1. Then, at a preset time interval of 2 seconds, the instantaneous velocity change and displacement compensation components of each UAV are calculated. After superposition, the overall displacement compensation vector is obtained as (0.9,0,0). The spatial offset is subtracted from the overall displacement compensation vector to obtain the protocol feature correction parameter as (-1.9,0,0). The vertex positions of the virtual space shape are adjusted according to this parameter. After generating a new virtual space shape, the closed-loop verification offset between its geometric center and the communication range center is calculated as (-0.8,0,0). This value is less than the preset threshold of 1, so the current virtual space shape is output as the final geometric model.

[0128] Then, deep correlation features are extracted from the spatial topology of the communication environment model, such as the distance change rate between A and B being 3.8 m / s and the velocity correlation strength being 3.6, and the distance change rate between B and C being 4.2 m / s and the velocity correlation strength being 3.9. The feature dispersion in the x, y, and z directions is calculated to be 3.1, 3.7, and 3.3, respectively. 3.7 is selected as the feature filtering parameter, and features with both distance change rate and velocity correlation strength greater than 3.7, i.e., the correlation features between B and C, are selected as key features. The distance change rate of this key feature (4.2) is multiplied by a preset distance attenuation coefficient of 0.6 to obtain a distance compensation component of 2.52; the velocity correlation strength (3.9) is multiplied by a preset motion interference coefficient of 0.5 to obtain a motion compensation component of 1.95. These two components are then weighted and superimposed with weights of 0.5 and 0.5, respectively, to generate a protocol layer compensation coefficient of 2.235.

[0129] Finally, the coordinates of the boundary vertices in the final geometric model output from the closed-loop geometric modeling are obtained, and the distances between adjacent vertices are calculated to form a positional relationship feature sequence. This sequence is then analyzed to obtain spatiotemporal variation features. Boundary segments meeting the criteria are selected from these features to determine candidate directions for communication interruption paths and cluster them to obtain the communication interruption paths. The modulus of the protocol feature correction parameters, the protocol layer compensation coefficient of 2.235, the azimuth angle of the communication interruption path of 35°, and the pitch angle of 40° are combined into a communication protocol feature vector. This vector is input into a pre-trained risk probability calculation function, yielding communication interruption risk probabilities of 0.12, 0.18, 0.25, and 0.21 for UAVs A, B, C, and D, respectively. The first two are selected to form a backup communication node set {A, B}. The quality parameters of the main communication link are continuously monitored. When the quality falls below the first quality threshold of 55, the system switches to A. Within a fixed verification time window of 10 seconds, the quality of the new link is checked. If it does not recover to the second quality threshold of 65, A is removed and the system switches to B. If B also fails, a new backup node set is recalculated and generated.

[0130] Figure 4 This is a schematic diagram illustrating a specific implementation of an artificial intelligence-based unmanned aerial vehicle (UAV) communication protocol feature extraction system provided in this application. (Refer to...) Figure 4 The system may include:

[0131] The acquisition module 41 is used to acquire the spatial location set of the base station and the UAV swarm, synchronously collect the real-time velocity and acceleration information of each location point in the spatial location set, and construct a communication environment model based on the spatial location set and the real-time velocity and acceleration information.

[0132] Compensation module 42 is used to map the set of spatial positions to generate a virtual space shape, calculate the spatial offset between the geometric center point of the virtual space shape and the center point of the communication range, combine the acceleration information to perform motion compensation on the spatial offset to generate protocol feature correction parameters, and use the protocol feature correction parameters to feed back and adjust the shape parameters of the virtual space shape to form a closed-loop geometric model.

[0133] The filtering module 43 is used to extract deep correlation features of the communication environment model, generate feature filtering parameters according to the dimensional distribution characteristics of the communication environment model, filter key features of the deep correlation features based on the feature filtering parameters, and fuse the filtered key features with preset signal attenuation model parameters to generate protocol layer compensation coefficients.

[0134] The generation module 44 is used to extract the spatiotemporal variation features of the communication network based on the positional relationship features output by the closed-loop geometric modeling, predict the communication interruption path based on the spatiotemporal variation features, extract the communication protocol features based on the protocol feature correction parameters, protocol layer compensation coefficients and communication interruption path and evaluate the probability of communication interruption risk, so as to generate a dynamic selection strategy for backup communication nodes of the communication protocol layer.

[0135] The AI-based UAV communication protocol feature extraction system of this application embodiment is used to implement the aforementioned AI-based UAV communication protocol feature extraction method. Therefore, the specific implementation of the AI-based UAV communication protocol feature extraction system can be found in the embodiment section of the AI-based UAV communication protocol feature extraction method above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0136] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described artificial intelligence-based drone communication protocol feature extraction method.

[0137] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described artificial intelligence-based UAV communication protocol feature extraction methods.

[0138] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0139] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the artificial intelligence-based UAV communication protocol feature extraction method.

[0140] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0141] The foregoing has provided a detailed description of the method, system, electronic device, and storage medium for extracting features of unmanned aerial vehicle (UAV) communication protocols based on artificial intelligence. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for extracting features of unmanned aerial vehicle (UAV) communication protocols based on artificial intelligence, characterized in that, include: Obtain the spatial location set of the base station and the drone swarm, synchronously collect the real-time velocity and acceleration information of each location point in the spatial location set, and construct a communication environment model based on the spatial location set and the real-time velocity and acceleration information; The set of spatial locations is mapped to generate a virtual space shape. The spatial offset between the geometric center point of the virtual space shape and the center point of the communication range is calculated. The spatial offset is then used to perform motion compensation based on the acceleration information to generate protocol feature correction parameters. The shape parameters of the virtual space shape are then adjusted by feedback through the protocol feature correction parameters to form a closed-loop geometric model. Deep correlation features of the communication environment model are extracted, feature filtering parameters are generated based on the dimensional distribution characteristics of the communication environment model, key features of the deep correlation features are filtered based on the feature filtering parameters, and the filtered key features are fused with preset signal attenuation model parameters to generate protocol layer compensation coefficients. Based on the positional relationship features output by the closed-loop geometric modeling, the spatiotemporal variation features of the communication network are extracted. Based on the spatiotemporal variation features, the communication interruption path is predicted. Based on the protocol feature correction parameters, protocol layer compensation coefficients, and communication interruption path, the communication protocol features are extracted and the probability of communication interruption risk is evaluated to generate a dynamic selection strategy for backup communication nodes at the communication protocol layer.

2. The method according to claim 1, characterized in that, The step of extracting communication protocol features based on the protocol feature correction parameters, protocol layer compensation coefficients, and communication interruption paths, and evaluating the probability of communication interruption risk to generate a dynamic selection strategy for backup communication nodes at the communication protocol layer, includes: The modulus of the protocol feature correction parameters, the protocol layer compensation coefficient, and the azimuth and elevation angles of the communication interruption path are combined into a communication protocol feature vector. The communication protocol feature vector is then input into a pre-trained risk probability calculation function to output the communication interruption risk probability of each UAV node relative to the base station. Sort all drone nodes in ascending order of communication interruption risk probability, and select the N nodes with the lowest communication interruption risk probability to form a backup communication node set; The quality parameters of the main communication link are continuously monitored. When the quality parameters are lower than the first quality threshold, the communication routing node is switched to the highest priority node in the backup node set. The quality parameters of the new communication link are detected within a fixed verification time window after the switch. If the quality parameters of the new communication link do not recover to the second quality threshold, the current node is removed from the backup node set and switched back to the highest priority node among the remaining nodes. When there are no available nodes in the set of backup nodes, the interruption risk probability calculation is re-executed and a new set of backup nodes is generated.

3. The method according to claim 1, characterized in that, The step of adjusting the shape parameters of the virtual space shape through the feedback of the protocol feature correction parameters to form a closed-loop geometric model includes: The vertex movement direction and movement distance of the virtual space shape are determined based on the protocol feature correction parameters. Based on the vertex movement direction and distance, perform synchronous translation operations on the coordinates of all vertices of the virtual space shape, and reconnect the translated vertex coordinates to generate a new virtual space shape. Calculate the coordinates of the geometric center point of the new virtual space shape, and measure the difference between the coordinates of the geometric center point and the coordinates of the center point of the base station communication range as the closed-loop verification offset. When the closed-loop verification offset is less than a preset threshold, the current virtual space shape is output as the final geometric model. When the closed-loop verification offset is greater than or equal to the preset threshold, the closed-loop verification offset is used as a new spatial offset and input into the motion compensation process to regenerate the protocol feature correction parameters.

4. The method according to claim 1, characterized in that, The step of extracting the spatiotemporal variation features of the communication network based on the positional relationship features output by the closed-loop geometric modeling, and predicting the communication interruption path based on the spatiotemporal variation features, includes: Obtain the coordinates of all boundary vertices in the final geometric model output by the closed-loop geometric modeling, and calculate the distance between adjacent vertices to form a positional relationship feature sequence; The positional relationship feature sequence is analyzed, and the distance fluctuation range within each time window is extracted as the amplitude feature. At the same time, the number of times the distance crosses zero within a unit time is calculated as the frequency feature. The amplitude feature and the frequency feature are combined to form the spatiotemporal variation feature. In the spatiotemporal variation features, boundary line segments that satisfy the condition of continuously increasing amplitude and frequency exceeding a preset frequency threshold are detected. The extension direction vectors of these line segments are determined as candidate directions for communication interruption paths. Cluster analysis is performed on the candidate directions of the communication interruption paths, and the direction with the largest cluster center is taken as the communication interruption path.

5. The method according to claim 1, characterized in that, The mapping of the spatial location set generates a virtual spatial shape, the spatial offset between the geometric center point of the virtual spatial shape and the center point of the communication range is calculated, and motion compensation is performed on the spatial offset using the acceleration information to generate protocol feature correction parameters, including: The position coordinates of the UAVs in the spatial location set are input into the spatial mapping rules, and a virtual spatial shape is formed by connecting the position coordinates of the outermost UAV. Calculate the arithmetic mean of the coordinates of all vertices of the virtual space shape as the coordinates of the geometric center point, and at the same time measure the spatial offset between the coordinates of the geometric center point and the coordinates of the center point of the base station communication range; The acceleration information of each UAV node is multiplied by a preset time interval to obtain the instantaneous velocity change. The instantaneous velocity change is multiplied by the same time interval to obtain the displacement compensation component. The displacement compensation components of all UAV nodes are vector-superimposed to generate an overall displacement compensation vector. The spatial offset is subtracted from the overall displacement compensation vector to obtain the protocol feature correction parameters.

6. The method according to claim 1, characterized in that, The process of extracting deep correlation features from the communication environment model, generating feature filtering parameters based on the dimensionality distribution characteristics of the communication environment model, filtering key features of the deep correlation features based on the feature filtering parameters, and fusing the filtered key features with preset signal attenuation model parameters to generate protocol layer compensation coefficients includes: The distance change rate and velocity correlation strength between all UAV node pairs are extracted from the spatial topology network of the communication environment model as deep correlation features. Calculate the feature dispersion of the deep association features in the three-dimensional coordinate direction, and take the maximum dispersion value in the three directions as the feature selection parameter; Deeply correlated features whose distance change rate is greater than the feature filtering parameter and whose velocity correlation strength is greater than the feature filtering parameter are selected as key features; The distance compensation component is obtained by multiplying the distance change rate of the key feature by a preset distance attenuation coefficient. Multiply the velocity correlation strength of the key feature by a preset motion disturbance coefficient to obtain the motion compensation component; The distance compensation component and the motion compensation component are weighted and superimposed to generate the protocol layer compensation coefficient.

7. The method according to claim 1, characterized in that, The process involves acquiring a spatial location set of the base station and the drone swarm, synchronously collecting real-time velocity and acceleration information of each location point in the spatial location set, and constructing a communication environment model based on the spatial location set and the real-time velocity and acceleration information, including: Obtain the spatial location set of all individual drones in the base station and drone swarm, and synchronously record the real-time velocity and acceleration information corresponding to each location point in the spatial location set; A spatial topology network containing base station nodes and drone nodes is established based on the spatial location set, and the real-time speeds of the base station nodes and drone nodes are marked at the corresponding positions in the spatial topology network. Based on the changes in spatial distance between adjacent nodes and the real-time speed differences, dynamic connection relationships are constructed in the spatial topology network. The acceleration information is integrated as a motion state variable into the dynamic connection relationship to form a communication environment model that can be updated in real time.

8. A feature extraction system for unmanned aerial vehicle (UAV) communication protocols based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire the spatial location set of the base station and the UAV swarm, synchronously collect the real-time velocity and acceleration information of each location point in the spatial location set, and construct a communication environment model based on the spatial location set and the real-time velocity and acceleration information. The compensation module is used to map the set of spatial locations to generate a virtual space shape, calculate the spatial offset between the geometric center point of the virtual space shape and the center point of the communication range, combine the acceleration information to perform motion compensation on the spatial offset to generate protocol feature correction parameters, and use the protocol feature correction parameters to feed back and adjust the shape parameters of the virtual space shape to form a closed-loop geometric model. The filtering module is used to extract deep correlation features of the communication environment model, generate feature filtering parameters based on the dimensional distribution characteristics of the communication environment model, filter key features of the deep correlation features based on the feature filtering parameters, and fuse the filtered key features with preset signal attenuation model parameters to generate protocol layer compensation coefficients. The generation module is used to extract the spatiotemporal variation features of the communication network based on the positional relationship features output by the closed-loop geometric modeling, predict the communication interruption path based on the spatiotemporal variation features, extract the communication protocol features based on the protocol feature correction parameters, protocol layer compensation coefficients and communication interruption paths, and evaluate the probability of communication interruption risk, so as to generate a dynamic selection strategy for backup communication nodes of the communication protocol layer.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the AI-based unmanned aerial vehicle communication protocol feature extraction method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the AI-based UAV communication protocol feature extraction method as described in any one of claims 1 to 7.

Images