Unmanned aerial vehicle cluster communication resource allocation method and system based on 5g slice technology

Abstract

The application discloses a UAV cluster communication resource allocation method and system based on a 5G slice technology, and belongs to the technical field of UAV communication, and the technical solution points comprise the following steps: obtaining slice configuration requirements according to static basic parameters of each UAV in a UAV cluster and combining preset scene priorities; obtaining operation data of each UAV according to the working scene of the UAV cluster; obtaining resource requirements of each UAV by a fusion feature algorithm according to the operation data; and obtaining a slice allocation sequence of the UAV cluster by a graph interference perception algorithm according to the resource requirements and the slice configuration requirements. The application obtains compliant resource requirements by the fusion feature algorithm comprising spectrum constraints, actively avoids interference between slices by the graph interference perception algorithm, realizes accurate and compliant allocation of low-altitude communication resources of the UAV cluster, improves the feasibility of resource allocation and the spectrum utilization efficiency, and reduces the risk of communication interruption caused by spectrum conflict and interference.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) communication technology, and more specifically to a method and system for allocating UAV swarm communication resources based on 5G slicing technology. Background Technology

[0002] With the rapid development of the low-altitude economy, drone swarms are increasingly being used in scenarios such as intelligent inspection, logistics transportation, emergency rescue, and high-altitude operations. The low-altitude communication quality of drone swarms needs to meet the service quality requirements of different tasks. 5G network slicing technology can allocate dedicated network resources for different drone tasks through multiple slices of a single network, becoming a core technology for solving the competition for communication resources in drone swarms.

[0003] However, existing UAV communication resource allocation schemes have limitations when adapting to complex low-altitude communication scenarios, making it difficult to meet the reliability requirements of practical applications. Existing technologies mostly employ centralized algorithms for resource scheduling, which suffers from delays in obtaining global spectrum occupancy information. Furthermore, given the strict control over low-altitude spectrum, resource allocation results often lead to spectrum usage conflicts. Existing schemes primarily adopt a method of allocating resources first and then checking for resource overruns, making it impossible for the base station to detect and manage inter-slice spatial interference caused by changes in UAV location in advance. Some existing technologies only consider a single dimension of UAV parameters during resource allocation, resulting in allocation results that deviate from the actual needs of various UAV types. Therefore, existing technologies have shortcomings. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for allocating communication resources for UAV swarms based on 5G slicing technology. Based on the static basic parameters and dynamic operational data of each UAV in the UAV swarm, compliant resource requirements are obtained through a joint reinforcement learning decoder including spectrum constraints. Interference between slices is actively avoided through a graph interference perception algorithm, thereby achieving accurate and compliant low-interference allocation of low-altitude communication resources for UAV swarms. This improves the feasibility of resource allocation and spectrum utilization efficiency, and reduces the risk of communication interruption caused by spectrum conflicts and communication interference.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides a method for allocating communication resources in a drone swarm based on 5G slicing technology. The drone swarm includes intelligent inspection drones, logistics transport drones, high-altitude operation drones, and tethered emergency rescue drones. The method includes:

[0007] Based on the static basic parameters of each drone in the drone cluster, and combined with the preset scene priority, the slice configuration requirements of each drone are obtained. The static basic parameters include model, payload type and operation scene.

[0008] Based on the working scenario of the drone swarm, the operational data of each drone in the drone swarm is obtained, and the operational data includes task characteristics, payload status, environmental interference, and topology changes;

[0009] The resource requirements of each UAV are obtained through a fusion feature algorithm based on the operational data. The fusion feature algorithm includes a shared encoder and a multi-task decoder. The resource requirements include bandwidth requirements, connectivity requirements, and bandwidth loss.

[0010] Based on the resource requirements and slice configuration requirements of each UAV, the slice allocation sequence of the UAV cluster is obtained through a graph interference perception algorithm.

[0011] As a further improvement of the present invention, the step of obtaining the slice configuration requirements of each drone based on the static basic parameters of each drone in the drone cluster and in combination with a preset scene priority includes:

[0012] A structured data set is constructed based on the static basic parameters of each drone in the drone cluster;

[0013] Based on the structured data set and the preset operation scenario priority, the service quality index of each UAV is obtained;

[0014] The slice configuration requirements for each UAV are obtained based on the service quality indicators.

[0015] As a further improvement of the present invention, the step of obtaining the resource requirements of each UAV based on the operational data through a fusion feature algorithm includes:

[0016] A standardized feature matrix is ​​obtained by standardizing the aforementioned operational data.

[0017] Based on the standardized feature matrix, a shared feature vector is obtained through a shared encoder;

[0018] The resource requirements of each UAV are obtained through a multi-task decoder based on the shared feature vector.

[0019] As a further improvement of the present invention, the step of obtaining the shared feature vector based on the standardized feature matrix through a shared encoder includes:

[0020] Based on the standardized feature matrix, a temporal feature vector is obtained through a temporal convolutional network;

[0021] Based on the temporal feature vector, a multi-head attention mechanism is used to obtain the scene key feature vector;

[0022] Based on the key feature vectors of the scenario, a shared feature vector is obtained through mapping using a fully connected layer.

[0023] As a further improvement of the present invention, the multi-task decoder includes a bandwidth demand decoder, a connection demand decoder, and a bandwidth loss decoder. The step of obtaining the resource requirements of each UAV through the multi-task decoder based on the shared feature vector includes:

[0024] Based on the shared feature vector, the bandwidth requirement value and the prediction curve for the future preset time period of each UAV are obtained through the bandwidth requirement decoder;

[0025] Based on the shared feature vector and the optimized connection number prediction model, the optimal connection number and spectrum strategy for each UAV are obtained. The optimized connection number prediction model is located in the multi-agent decision network of the connection demand decoder and is obtained by training through historical shared feature vectors.

[0026] The bandwidth loss value is obtained by using the bandwidth loss decoder based on the shared feature vector.

[0027] Based on the bandwidth requirement, optimal number of connections, spectrum strategy, and bandwidth loss value of each UAV, the resource requirements of each UAV are obtained through data integration, parameter normalization, and collaborative verification.

[0028] As a further improvement of the present invention, the connection demand decoder includes a multi-agent decision network and a spectrum compliance environment simulator, and the optimized connection number prediction model is obtained by training through historical shared feature vectors, including:

[0029] The initial connection number prediction model is obtained through a multi-agent decision network based on historical shared feature vectors.

[0030] Spectrum constraints are obtained through a spectrum compliance environment simulator based on the historical operational data of the aforementioned drone cluster.

[0031] Based on the initial connection number prediction model and spectral constraints, and using the shared feature vector as the state input, an optimized connection number prediction model is obtained through training.

[0032] As a further improvement of the present invention, the step of obtaining the slice allocation sequence of the UAV cluster through a graph interference perception algorithm based on the resource requirements and slice configuration requirements of each UAV includes:

[0033] A candidate slice set is obtained based on the slice configuration requirements and spectrum strategy of each UAV;

[0034] Based on the resource requirements of each UAV and the candidate slice set, a suitability score is obtained through weighted calculation, and an initial allocation sequence for each UAV is obtained by sorting in descending order.

[0035] Based on the initial allocation sequence and real-time geographic location information of each UAV, an optimized initial allocation sequence is obtained by constructing an interference relationship graph and using a graph cutting algorithm. The real-time geographic location information is obtained through the UAV payload.

[0036] Based on the optimized initial allocation sequence, the slice allocation sequence of the UAV cluster is obtained through conflict checking and adjustment.

[0037] As a further improvement of the present invention, the step of obtaining the optimized initial allocation sequence by constructing an interference relationship graph and using a graph cutting algorithm includes:

[0038] Based on the initial allocation sequence and real-time geographical location information of each UAV, an interference relationship map is obtained by setting nodes and calculating the interference intensity between nodes;

[0039] Based on the interference relationship graph, a subset of low-interference nodes is obtained using a graph cutting algorithm;

[0040] The optimized initial allocation sequence is obtained by sorting the low-interference node subset and the initial allocation sequence according to their fitness.

[0041] As a further improvement of the present invention, obtaining the operational data of each drone in the drone cluster based on the working scenario of the drone cluster includes:

[0042] Data acquisition rules are derived based on the working scenario of the drone swarm, and the data acquisition rules include acquisition frequency and positioning accuracy.

[0043] The operational data of each UAV is obtained according to the data acquisition rules. The operational data includes mission characteristics, payload status, environmental interference, and topology changes.

[0044] This invention provides a drone swarm communication resource allocation system based on 5G slicing technology, comprising:

[0045] The requirement analysis module obtains the slice configuration requirements of each drone based on the static basic parameters of each drone in the drone cluster and in combination with the preset scenario priority. The static basic parameters include model, payload type and operation scenario.

[0046] The data acquisition module obtains the operational data of each drone in the drone cluster based on the working scenario of the drone cluster. The operational data includes task characteristics, payload status, environmental interference, and topology changes.

[0047] The resource calculation module obtains the resource requirements of each UAV based on the operating data through a fusion feature algorithm. The fusion feature algorithm includes a shared encoder and a multi-task decoder. The resource requirements include bandwidth requirements, connection requirements, and bandwidth loss.

[0048] The slice allocation module obtains the slice allocation sequence of the UAV cluster based on the resource requirements and slice configuration requirements of each UAV using a graph interference perception algorithm.

[0049] This invention obtains slice configuration requirements based on static basic parameters and preset scenario priorities, obtains resource requirements through a fusion feature algorithm based on dynamic operating data, ensures the spectrum compliance of resource requirements through a joint reinforcement learning decoder, and finally obtains a low-interference slice allocation sequence through interference relationship graph and graph cutting algorithm. This invention realizes dynamic adaptation and global optimization of low-altitude communication resources for UAV swarms, improves resource allocation accuracy and spectrum utilization efficiency, ensures the communication quality stability of low-altitude communication for UAV swarms, and reduces spectrum conflicts, interference, and system allocation latency. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the steps of the UAV swarm communication resource allocation method based on 5G slicing technology according to the present invention.

[0051] Figure 2 A flowchart illustrating the steps involved in obtaining resource requirements using a feature fusion algorithm;

[0052] Figure 3 Flowchart for training the joint reinforcement learning decoder;

[0053] Figure 4 This is a schematic diagram illustrating the optimized initial assignment sequence obtained through interference relationship mapping and graph cutting algorithms;

[0054] Figure 5 This is a schematic diagram of the structure of the UAV swarm communication resource allocation system based on 5G slicing technology according to the present invention. Detailed Implementation

[0055] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof.

[0056] Identical parts are indicated by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, while the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific part, respectively.

[0057] The term "and / or" in the following text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0058] like Figure 1 As shown, this embodiment provides a method for allocating communication resources for a drone swarm based on 5G slicing technology. The drone swarm includes intelligent inspection drones, logistics transport drones, high-altitude operation drones, and tethered emergency rescue drones. The method includes:

[0059] Based on the static basic parameters of each drone in the drone cluster, and combined with the preset scenario priority, the slice configuration requirements of each drone are obtained. The static basic parameters include model, payload type and operation scenario.

[0060] Based on the working scenario of the drone swarm, the operational data of each drone in the drone swarm is obtained. The operational data includes task characteristics, payload status, environmental interference, and topology changes.

[0061] Based on the operational data, the resource requirements of each UAV are obtained through a fusion feature algorithm, which includes a shared encoder and a multi-task decoder. The resource requirements include bandwidth requirements, connectivity requirements, and bandwidth loss.

[0062] Based on the resource requirements and slice configuration requirements of each drone, the slice allocation sequence of the drone cluster is obtained through a graph interference perception algorithm.

[0063] Among them, static basic parameters are the basic attribute data of UAV swarm operations, obtained from the UAV product specifications, used to determine the basis for initial resource configuration; model is the hardware specification identifier of the UAV, including hardware parameter information such as fuselage size, endurance, maximum payload weight, and communication module model; payload type is the type of mission execution equipment carried by the UAV, including LiDAR, camera, and GPS module; operation scenario is the type of operation that the UAV can perform, including emergency rescue, logistics delivery, equipment inspection, and high-altitude operations; preset scenario priority is a sorting rule pre-set according to the urgency of each UAV's operation scenario; slice configuration requirements are the 5G slice parameter requirements adapted to the UAV's operation scenario, including slice type and resource reservation ratio; UAV swarm working scenario is the actual scenario and task performed by the UAV swarm; operation data is the real-time operation data of the UAV, obtained through UAV sensors and base station feedback, including task characteristics, payload status, environmental interference, and topology changes; task characteristics are the core data characterizing the UAV task attributes, including task type and swarm size; payload status is a parameter reflecting the impact of the payload on communication requirements, including payload... The data types and resolutions are defined as follows: Environmental interference refers to external factors affecting the quality of low-altitude communication, including terrain obstruction level, channel signal-to-noise ratio, and electromagnetic interference intensity; Topology changes refer to dynamic information related to UAV communication connections, including location trajectory and base station coverage; The fusion feature algorithm is a combined algorithm for extracting data patterns and predicting resource requirements, including a shared encoder and a multi-task decoder; The shared encoder includes a temporal convolutional network, a multi-head attention mechanism, and a fully connected layer; The multi-task decoder includes a bandwidth requirement decoder, a connection requirement decoder, and a bandwidth loss decoder, with the connection requirement decoder being a joint reinforcement learning decoder; Bandwidth requirement refers to the spectrum width required for real-time UAV communication, including the bandwidth requirement value and the prediction curve for a future preset time period; Connection requirement refers to the number of base station connections that the UAV needs to establish and the corresponding spectrum strategy suggestions, including the optimal number of connections and spectrum strategy; Bandwidth loss refers to the effective bandwidth loss caused by environmental interference and channel attenuation, i.e., the bandwidth loss value; The graph interference perception algorithm is an algorithm that obtains the slice allocation sequence by constructing an interference relationship graph and optimizing the graph cutting algorithm; The slice allocation sequence of the UAV swarm is obtained based on the slice allocation sequence of each UAV in the UAV swarm.

[0064] This embodiment, based on the static basic parameters and dynamic operational data of the UAV swarm, achieves precise allocation and optimization of low-altitude communication resources of the UAV swarm by integrating the joint reinforcement learning decoder and the graph interference perception algorithm in the feature algorithm and based on 5G slicing technology. This effectively improves the feasibility of resource allocation and the stability of communication quality, and reduces the risk of communication interruption caused by compliance conflicts and interference.

[0065] Furthermore, this embodiment provides a step for obtaining the slice configuration requirements of each drone based on the static basic parameters of each drone in the drone cluster and combined with preset scene priorities, including:

[0066] A structured data set is constructed based on the static basic parameters of each drone in the drone swarm;

[0067] Based on structured data sets and preset operational scenario priorities, service quality indicators for each drone are obtained.

[0068] The slice configuration requirements for each drone are obtained based on service quality indicators.

[0069] The structured data set is a collection obtained by associating the model, payload type, and operational scenario of each drone. The preset operational scenario priority is a sorting rule pre-defined according to the urgency of each drone's operational scenario. In this embodiment, the preset operational scenario priority is set as follows: emergency scenarios > logistics scenarios > inspection scenarios. Service quality indicators are used to measure the performance of communication services, including latency, bandwidth, and reliability. Latency is the time delay of data transmission, bandwidth is the amount of data transmitted per unit time, and reliability is the stability of data transmission. The slice configuration requirements are the network resource requirements of each drone in the operational scenario, including the required slice type and resource reservation ratio. The 5G slice types include URLLC slices, eMBB slices, etc.

[0070] Specifically, firstly, a structured data group is constructed by associating the model, payload type, and operational scenario information of each drone in the drone cluster. Secondly, based on the operational scenario information in the structured data group and the preset operational scenario priority, initial service quality indicators (SMIs) for each drone are obtained, including the range values ​​of latency, bandwidth, and reliability. Then, a payload requirement threshold is obtained based on the payload type in the structured data group. The initial SMIs are compared with the payload requirement threshold to obtain optimized and precise SMIs, including specific values ​​of latency, bandwidth, and reliability. Finally, based on the optimized SMIs, the slice configuration requirements for each drone are obtained, including the slice type and resource reservation ratio.

[0071] This embodiment obtains structured data groups by the drone model, payload type, and operation scenario. Based on preset operation scenario priorities and payload requirement thresholds, it obtains optimized service quality indicators, and then matches the slice configuration requirements. This achieves precise allocation of drone cluster communication resources, improves the utilization efficiency of 5G network resources, and enhances the stability and reliability of drone operations.

[0072] Furthermore, this embodiment provides a step for obtaining the operational data of each drone within a drone swarm based on the working scenario of the drone swarm, including:

[0073] Data acquisition rules are derived based on the working scenario of the drone swarm, and these rules include acquisition frequency and positioning accuracy.

[0074] The operational data for each UAV is obtained according to the data acquisition rules. The operational data includes mission characteristics, payload status, environmental interference, and topology changes.

[0075] The working scenario of the UAV swarm refers to the actual scenario and tasks performed by the UAV swarm, distinct from the operational scenario of the UAVs. Specifically, it is the combination of the environment and execution status of the actual task. The acquisition frequency is determined based on the real-time requirements of the working scenario of the UAV swarm. Positioning accuracy is used to ensure the consistency between topology data and the actual position of the UAVs. Task characteristics include task type and swarm size. The task type is the specific task actually performed by the UAVs in the working scenario, and the swarm size is the total number of UAVs participating in the current working scenario. Payload status includes payload type and data resolution. The payload type is the task execution device actually carried by the UAV in the current working scenario, and the data resolution is the resolution level of the current payload output data. Environmental interference includes terrain occlusion level, channel signal-to-noise ratio, and electromagnetic interference intensity. The terrain occlusion level is the degree to which the terrain in the working scenario obstructs the communication signal, and it is characterized by hierarchical quantification. The channel signal-to-noise ratio is used to measure the signal quality of the real-time channel environment in the working scenario. The electromagnetic interference intensity is the interference value of the electromagnetic environment in the working scenario on the communication link. Topology changes include position trajectory and base station coverage. The position trajectory is the continuous coordinate sequence of the UAV during flight in the working scenario, and the base station coverage is the coverage area of ​​the base station signal when the UAV is flying in the working scenario.

[0076] Furthermore, this embodiment provides a step-by-step method for obtaining the resource requirements of each UAV based on operational data using a fusion feature algorithm. The fusion feature algorithm includes a shared encoder and a multi-task decoder, and the resource requirements include bandwidth requirements, connectivity requirements, and bandwidth loss. The steps include:

[0077] A standardized feature matrix is ​​obtained by standardizing the operational data.

[0078] A shared feature vector is obtained based on the standardized feature matrix through a shared encoder;

[0079] The resource requirements of each UAV are obtained through a multi-task decoder based on the shared feature vector.

[0080] The standardization process involves converting qualitative parameters in the operational data into quantitative parameters, standardizing the data units, and removing outliers to eliminate data redundancy and format differences, ensuring that the data can be directly input into the algorithm for calculation. The standardized feature matrix is ​​a structured data matrix formed after standardization, containing quantitative representations of four types of data: task characteristics, payload status, environmental interference, and topology changes. The shared feature vector is a core feature vector extracted from the standardized feature matrix that reflects the UAV's mission attributes, environmental interference, and topology changes. The resource requirements are resource parameters that ensure the quality of communication services for the UAV in actual working scenarios.

[0081] Specifically, such as Figure 2 As shown, the collected operational data is first standardized by converting qualitative data such as terrain occlusion levels into level quantification values ​​and removing abnormal data where electromagnetic interference intensity suddenly drops to 0, resulting in a standardized feature matrix. This standardized feature matrix is ​​then input into a shared encoder, which processes the temporal dependencies of the data through a temporal convolutional network and filters key features of the scene using a multi-head attention mechanism, yielding a shared feature vector. Finally, the shared feature vector is input into a multi-task decoder, which obtains the bandwidth requirement value and the prediction curve for a future preset time period through a bandwidth requirement decoder, the optimal number of connections and spectrum strategy through a connection requirement decoder based on joint reinforcement learning, and the bandwidth loss value through a bandwidth loss decoder. Ultimately, the resource requirements for each UAV are obtained collaboratively.

[0082] For example, this embodiment assumes a drone swarm for comprehensive operations in the eastern part of a city, consisting of two tethered emergency rescue drones (numbered...). 、 ), 2 logistics transport drones (number) 、 ), 2 intelligent inspection drones (number) 、 Composed of, among which 、 Carry out the delivery of emergency medical supplies to the community. 、 Delivery of e-commerce parcels to the outlets in this area. 、 The inspection team inspected the power transmission lines within the designated area. First, the operational data was standardized, quantifying terrain obstruction values ​​based on terrain obstruction levels, unifying the electromagnetic interference intensity unit to dBm, and simultaneously eliminating... The normalized feature matrix is ​​obtained from the electromagnetic interference data that abruptly changes to 0. (6 rows × 12 columns), with rows corresponding to each UAV and columns corresponding to the dimensions of mission characteristics and payload status data; the matrix... The input is fed into a shared encoder, where a temporal convolutional network processes the temporal dependencies of the location trajectories, such as... By analyzing the location change patterns over a 10-minute period, a multi-head attention mechanism is used to assign high weights to the electromagnetic interference intensity in emergency scenarios, camera resolution in logistics scenarios, and LiDAR point cloud rate in inspection scenarios, resulting in key feature vectors for each scenario. These vectors are then mapped to a 16-dimensional shared feature vector through a fully connected layer. 、 、 、 、 、 The shared feature vector is input into the multi-task decoder, and the bandwidth requirement value for each UAV is obtained through the bandwidth requirement decoder. 、 、 、 、 、 and its corresponding prediction curve; the optimal number of connections for each drone is obtained by connecting to the demand decoder. 、 、 、 、 、 and its corresponding spectrum strategy; the bandwidth loss value of each UAV is obtained through a bandwidth loss decoder. 、 、 、 、 、 The parameters are mapped to the [0,1] interval using Min-Max normalization, and after collaborative verification, as verified... After matching the resolution with the infrared thermal imager, the complete resource requirements for each UAV are obtained, among which... For drones The corresponding bandwidth requirement value; the meanings of other parameters can be deduced similarly.

[0083] This embodiment eliminates format differences and redundant interference in operational data through standardization processing, accurately extracts feature vectors reflecting the core needs of UAVs through a shared encoder, and achieves parallel calculation and collaborative optimization of bandwidth requirements, connection requirements, and bandwidth loss through a multi-task decoder. In particular, it improves the adaptability of connection requirements and spectrum strategies by relying on joint reinforcement learning, effectively realizing multi-dimensional and high-precision calculation of UAV resource requirements. It takes into account both real-time requirements and future predictions, and ensures the consistency and reliability of parameters. It provides accurate and comprehensive input basis for subsequent slice allocation based on the graph interference perception algorithm, and significantly improves the rationality and efficiency of UAV swarm communication resource allocation.

[0084] Furthermore, this embodiment provides a step for obtaining a shared feature vector based on a normalized feature matrix using a shared encoder, including:

[0085] Temporal feature vectors are obtained through a temporal convolutional network based on the standardized feature matrix;

[0086] The key feature vectors of the scene are obtained through a multi-head attention mechanism based on the temporal feature vectors.

[0087] A shared feature vector is obtained by mapping the key feature vectors of the scene through a fully connected layer.

[0088] Among them, the temporal convolutional network is a deep learning network with causal convolution and residual connection structure, used to extract the temporal dependencies of task features and environmental interference in the standardized feature matrix; the temporal feature vector is a feature vector containing time dimension information obtained by processing the standardized feature matrix through the temporal convolutional network; the multi-head attention mechanism is a mechanism that assigns weights to different dimensions of features of the temporal feature vector through multiple parallel attention heads, used to select core features according to the UAV operation scenario; the scene key feature vector is a feature vector focused on the core requirements of the scene after being weighted and filtered by the multi-head attention mechanism; the fully connected layer is a neural network structure containing an input layer, a hidden layer and an output layer, used to map the high-dimensional scene key feature vector into a low-dimensional feature vector.

[0089] Specifically, the standardized feature matrix is ​​first input into a temporal convolutional network. Causal convolution ensures the temporal rationality of feature extraction, and residual connections prevent gradient vanishing in deep networks, resulting in a temporal feature vector. Then, based on the drone operation scenario type, a multi-head attention mechanism is used to assign high weights to features strongly related to the scenario in the temporal feature vector, integrating them to obtain a key scenario feature vector. Finally, the key scenario feature vector is input into a fully connected layer, and a linear transformation and activation function are used to map high-dimensional features to low-dimensional features, resulting in a shared feature vector.

[0090] Furthermore, this embodiment provides a step for obtaining the resource requirements of each UAV based on a shared feature vector using a multi-task decoder. The multi-task decoder includes a bandwidth requirement decoder, a connection requirement decoder, and a bandwidth loss decoder, including:

[0091] Based on the shared feature vector, the bandwidth requirement value and the prediction curve for the future preset time period are obtained for each UAV through the bandwidth requirement decoder;

[0092] Based on the shared feature vectors and the optimized connection number prediction model, the optimal connection number and spectrum strategy for each UAV are obtained. The optimized connection number prediction model is located in the multi-agent decision network of the connection demand decoder and is trained by historical shared feature vectors.

[0093] The bandwidth loss value is obtained from the shared feature vector using a bandwidth loss decoder.

[0094] Based on the bandwidth requirements, optimal number of connections, spectrum strategy, and bandwidth loss of each drone, the resource requirements of each drone are obtained through data integration, parameter normalization, and collaborative verification.

[0095] The bandwidth requirement decoder includes a fully connected layer and a temporal residual layer, used for temporal prediction of data resolution and task type in the shared feature vector; the bandwidth requirement value is the real-time bandwidth quantification value to meet the data transmission needs of the UAV in the current working scenario; the connection requirement decoder includes a multi-agent decision network and a spectrum compliance environment simulator. The multi-agent decision network includes a feature extraction layer, a decision inference layer, and a policy output layer, used to process the location trajectory and base station coverage data in the shared feature vector; the optimal number of connections is the upper limit of the number of slice connections corresponding to the base station required by the UAV to ensure a balance between communication stability and resource efficiency; the spectrum policy is used to ensure the compliance and efficiency of UAV spectrum use; the bandwidth loss decoder includes a random forest embedding layer and a PID correction layer, which obtains the bandwidth loss value based on environmental interference in the shared feature vector; the bandwidth loss value is the quantification value of the actual bandwidth attenuation caused by environmental factors such as electromagnetic interference and terrain occlusion; parameter normalization is the process of converting parameters with different physical meanings and different numerical ranges to the same scale range, used to eliminate dimensional differences to achieve multi-parameter comparability; collaborative verification is the process of verifying the consistency of the parameter set based on business logic and physical rules, used to identify and correct outliers and logical contradictions.

[0096] Specifically, firstly, the shared feature vector is input into the bandwidth demand decoder. A fully connected layer performs non-linear mapping on the data resolution and task type, and a temporal residual layer is used to obtain the dynamic changes in bandwidth demand, resulting in the real-time bandwidth demand value and the bandwidth prediction curve for a future preset time period. Secondly, the connection demand decoder processes the location trajectory and base station coverage in the shared feature vector. After dimensionality reduction by the feature extraction layer, the decision reasoning layer interacts with the spectrum compliance environment simulator, and the optimal number of connections and the corresponding spectrum strategy are obtained through the policy output layer. Then, the shared feature vector is input into the bandwidth loss decoder. A random forest embedding layer obtains the initial bandwidth loss baseline value based on the electromagnetic interference intensity and terrain occlusion level in the environment. Then, a PID correction layer, combined with historical operating data, corrects the final bandwidth loss value. Finally, the data is integrated based on the bandwidth demand value, the optimal number of connections, the spectrum strategy, and the bandwidth loss value. Minimum-maximum normalization maps the parameters to the [0,1] interval, and then collaborative verification is performed based on preset rules to remove outliers and correct logical errors. By resolving contradictions, the complete resource requirements for each UAV are obtained. The preset rules include preset business logic and physical rules, which are derived from industry technical standards. Based on the preset rules, the normalized bandwidth requirement value, optimal number of connections and spectrum strategy, and bandwidth loss value are subjected to layered verification and contradiction reconciliation. The business logic includes the matching of task type and bandwidth requirement, resource efficiency and system constraints. The physical rules include electromagnetic attenuation rules, spatial connection rules and power bandwidth matching rules. When performing collaborative verification, the parameters are first matched and verified based on the business logic. If the above business logic is not met, it is marked as non-compliant and backtracked to the corresponding decoder for recalculation. Secondly, a second verification is performed based on the physical rules. If there are parameters that violate the physical rules, they are marked as non-compliant and the decoder is triggered to correct them. Then, for scenarios where there is a conflict between business logic and physical rules, the physical support capability is improved by optimizing the spectrum strategy or the business requirement is appropriately reduced to match the physical constraints. Finally, the logically self-consistent resource requirement parameters obtained through collaborative verification are obtained as the complete resource requirements for each UAV.

[0097] For example, this embodiment assumes that the shared feature vector of the aforementioned drone swarm is used. , , , , , First, Input bandwidth requirement decoder: The fully connected layer performs a nonlinear mapping between the infrared thermal imager resolution and the emergency delivery mission, and combines the temporal residual to capture the bandwidth fluctuation pattern within 1 hour, thus obtaining... bandwidth requirement The prediction curve includes a preset sampling time interval and preset prediction sampling points; similarly, the bandwidth requirements and prediction curves for the remaining five drones are obtained; then... The input connection requirement decoder performs dimensionality reduction on location trajectories and base station coverage through a feature extraction layer. Then, through interaction with a spectrum compliance environment simulator via a decision reasoning layer, it obtains... Optimal number of connections And spectrum strategy, including preferred frequency band, alternative frequency band, and avoidance frequency band; similarly, the optimal number of connections and spectrum strategy for the remaining five drones are obtained; then The input bandwidth loss decoder obtains the initial loss value based on terrain occlusion level and electromagnetic interference intensity through a random forest embedding layer, and then obtains the final loss value through a PID correction layer. bandwidth loss value Similarly, the bandwidth loss values ​​for the remaining five drones are obtained; finally, the parameters are processed by Min-Max normalization and after collaborative verification, the complete resource requirements for each drone are obtained.

[0098] This embodiment constructs a multi-task decoder including a bandwidth demand decoder, a connection demand decoder, and a bandwidth loss decoder. Combined with a data integration mechanism of parameter normalization and collaborative verification, it achieves multi-dimensional and high-precision calculation of UAV resource requirements. It considers both real-time requirements and future trends, while ensuring the consistency and reliability of parameters. This provides accurate and comprehensive input basis for the subsequent optimization of UAV swarm resource allocation, effectively improving the rationality and efficiency of resource allocation.

[0099] Furthermore, this embodiment provides a connection demand decoder including a multi-agent decision network and a spectrum compliance environment simulator. The optimized connection number prediction model is obtained through training with historical shared feature vectors, including the following steps:

[0100] The initial connection number prediction model is obtained through a multi-agent decision network based on historical shared feature vectors.

[0101] Spectrum constraints are obtained using a spectrum compliance environment simulator based on historical operational data of drone swarms.

[0102] Based on the initial connection number prediction model and spectral constraints, and using the shared feature vector as the state input, an optimized connection number prediction model is obtained through training.

[0103] Among them, the multi-agent decision network is a reinforcement learning network including a feature extraction layer, a decision reasoning layer, and a policy output layer, used to collaboratively optimize the connection configuration of multiple UAVs; the spectrum compliance environment simulator is a digital model simulating a real channel environment, used to verify the compliance and feasibility of the connection scheme; the initial connection number prediction model is an untrained connection number estimation model based on the preliminary mapping of shared feature vectors; and the spectrum strategy is a combination of preferred frequency bands, alternative frequency bands, and avoidance frequency bands.

[0104] Specifically, such as Figure 3 As shown, a multi-agent decision network is first constructed, comprising a feature extraction layer, a decision inference layer, and a policy output layer. The feature extraction layer includes convolutional layers, batch normalization layers, and a ReLU activation function, used to extract temporal features of location trajectories and spatial features of base station coverage from shared feature vectors. The decision inference layer adopts a deep Q-network architecture, including two hidden layers and a LeakyReLU activation function, and introduces a target network and an experience replay pool. The policy output layer uses a Softmax activation function, and its output dimension corresponds to the maximum number of connections supported by the UAV. Based on historical shared feature vectors, the initial parameters are configured through the multi-agent decision network to obtain the initial connection number prediction model.

[0105] Secondly, a spectrum compliance environment simulator is constructed based on the historical operational data of the UAV swarm. This historical data includes electromagnetic interference intensity and channel signal-to-noise ratio (SNR) over a past period. The simulator includes compliance review, channel simulation, and interference prediction. Hard constraints for civilian UAV communication are derived from the "Radio Frequency Allocation Regulations," including available frequency bands and transmit power limits. An executable compliance review is then generated based on these constraints. This review includes preliminary spectrum constraints and compliance verification logic. Preliminary spectrum constraints include data on available frequency bands and transmit power limits. The compliance verification logic compares the input spectrum strategy with the rule base to obtain compliant or non-compliant labels and records the number of compliant connection attempts, the total number of connection attempts, and the number of non-compliant attempts. Based on the Rayleigh fading model, a channel simulation model is obtained by fitting the historical channel SNR fluctuation pattern, conforming to the actual working scenario. The simulated channel SNR value is obtained through this model. Finally, the interference prediction model is obtained by performing regression analysis on the historical operational data using the least squares method. The final spectrum constraints are derived by combining the preliminary spectrum constraints with the channel simulation model and the interference prediction model.

[0106] Then, based on the real-time shared feature vector, the candidate connection number is obtained through the initial connection number prediction model. The channel environment is simulated through a spectrum compliance environment simulator, with the real-time shared feature vector as the state input, the candidate connection number as the action, and a reward function is set. for:

[0107] ;

[0108] in The spectrum compliance rate is the ratio of the number of successful compliant connection attempts to the total number of connection attempts. For connection stability, the values ​​are obtained based on the simulated channel signal-to-noise ratio, electromagnetic interference intensity, terrain obstruction level, and bandwidth loss. Resource utilization rate is obtained by the ratio of the actual number of slice connections of the drone to the maximum number of supported connections; The penalty for violations is determined by the number of spectrum violations. The corresponding weights are used; in each training round, the actions are input into the spectrum compliance environment simulator to obtain compliance results, interference feedback, and reward values. Samples are randomly sampled through the experience replay pool, and the Adam optimizer is used to update the parameters of the decision inference layer. The target network synchronizes the parameters of the main network once every preset interval, and the training is iterative. The process continues until the reward value converges, resulting in an optimized connection number prediction model.

[0109] Finally, the connection number prediction model optimized by real-time shared feature vector input is processed by the feature extraction layer and then the Q value corresponding to each connection number is obtained by the decision reasoning layer. The value corresponding to the maximum Q value is selected as the optimal connection number for each UAV. Based on the optimal connection number, the spectrum constraints and simulated channel environment obtained by the spectrum compliance environment simulator are used to obtain the preferred frequency band, alternative frequency band, avoidance frequency band, transmit power threshold and access timing. After integration, a complete spectrum strategy is obtained.

[0110] For example, this embodiment assumes that the shared feature vector of the aforementioned drone swarm is used. , , , , , Based on historical operational data, a multi-agent decision-making network is first constructed. The feature extraction layer is set as a convolutional layer with batch normalization and ReLU activation function. The decision inference layer is built using a deep Q-network and an experience replay pool. The policy output layer is a single softmax layer. An initial connection number prediction model is obtained by initialization based on shared feature vectors. A spectrum compliance environment simulator is constructed based on historical operational data. The compliance review module sets the available frequency bands and upper limits of transmit power according to the "Radio Frequency Allocation Regulations". The channel simulation module fits the fluctuation pattern of historical channel signal-to-noise ratio data to obtain a Rayleigh fading model. The interference prediction module obtains the interference model through least squares regression. Finally, the spectrum constraint Y is obtained by integration. Then, real-time data is used to obtain the spectrum constraint Y. Define a reward function for each state and candidate connection number as an action. Where S is the spectrum compliance rate, C is the connection stability, U is the resource utilization rate, and P is the violation penalty, and the Adam optimizer is used for iterative training. The process continues until the reward value converges, resulting in an optimized connection number prediction model. Finally, the shared feature vector is used to obtain the optimal connection number for each drone through the optimized connection number prediction model. 、 、 、 、 、 The spectrum strategy for each UAV is obtained by matching the spectrum constraint Y of the simulator.

[0111] This embodiment deeply integrates reinforcement learning-based multi-agent decision-making with spectrum compliance environment simulation, achieving coordinated optimization of connection count and spectrum strategy. This ensures both the rationality of the spectrum and the communication stability and resource utilization efficiency of the UAV swarm, solving the problem of the disconnect between compliance and optimization objectives in traditional connection configuration methods, and providing accurate connection requirement basis for subsequent slice resource allocation.

[0112] Furthermore, this embodiment provides a step for obtaining the slice allocation sequence of a drone cluster using a graph interference perception algorithm based on the resource requirements and slice configuration requirements of each drone, including:

[0113] A candidate slice set is obtained based on the slice configuration requirements and spectrum strategy of each UAV;

[0114] Based on the resource requirements and candidate slice set of each drone, the suitability score is obtained by weighted calculation, and the initial allocation sequence of each drone is obtained by sorting in descending order.

[0115] Based on the initial allocation sequence and real-time geographic location information of each UAV, an optimized initial allocation sequence is obtained by constructing an interference relationship map and using a graph cutting algorithm. The real-time geographic location information of the UAV is obtained through the UAV payload.

[0116] Based on the optimized initial allocation sequence, the slice allocation sequence of the UAV cluster is obtained through conflict checking and adjustment.

[0117] The candidate slice set is a collection of slices selected from available slices that are compatible with the UAV slice type and meet the spectrum policy requirements; the fit score is an indicator that quantifies the degree of matching between the slice and the UAV's resource requirements; the initial allocation sequence is obtained by sorting the fit scores in descending order; the UAV payload is the mission execution equipment carried by the UAV, in which the GPS module is used to obtain the UAV's real-time geographical location information; the graph cutting algorithm is a spectral clustering algorithm used to find the target with the highest total fit score and the smallest sum of interference weights; the conflict verification and adjustment is a resource coordination mechanism for bandwidth exceeding limits, connection exceeding limits, and inter-slice interference exceeding thresholds, including conflict detection, adjustment logic, and elastic resource pool invocation.

[0118] Specifically, firstly, the slice configuration requirements and avoidance frequency bands in the spectrum strategy of each UAV are extracted. By traversing all slices to exclude slices whose slice type does not match the UAV and slices containing avoidance frequency bands, a candidate slice set is obtained. The candidate slice set consists of one or more candidate slices.

[0119] Secondly, based on the slice type and specifications of each slice in the candidate slice set, the available bandwidth, maximum number of connections, and anti-interference gain of each slice are obtained. Available bandwidth is the amount of bandwidth resources when the slice is unused; maximum number of connections is the maximum number of drone connections a slice can simultaneously support; and anti-interference gain is a quantified value of the slice's anti-interference capability. The bandwidth matching degree is obtained by comparing the available bandwidth with the bandwidth requirements of the drones. The connection matching degree is obtained by comparing the maximum number of connections per slice with the optimal number of connections for the drone. The loss compensation degree is obtained by slicing the anti-interference gain and the bandwidth loss value of the UAV. ; through weighted formula The suitability score is obtained, and the initial allocation sequence for each drone is obtained by sorting the suitability scores in descending order, where the bandwidth matching degree is... The connection matching degree is obtained by comparing the remaining bandwidth of the slice with the bandwidth requirements of the drone. The loss compensation degree is obtained by comparing the maximum number of connections per slice with the optimal number of connections for the drone. The anti-interference gain and bandwidth loss of the UAV are obtained through slicing. Scoring the fit between each drone and the slice. , , These are the weighting coefficients corresponding to bandwidth matching degree, connection matching degree, and loss compensation degree;

[0120] Then, taking each drone and the candidate slice as nodes, the geographical distance between nodes is calculated based on the drone's real-time geographic location information, and combined with the frequency band overlap using the formula... The interference intensity is obtained, where, This represents the interference intensity between the slices corresponding to the two nodes. These are the weighting coefficients. Geographical distance, Frequency band overlap is an indicator that quantifies the degree of frequency band overlap between corresponding slices of two nodes. The larger the value, the higher the interference risk. When the interference intensity exceeds the preset interference intensity threshold, an interference relationship graph is obtained by establishing weighted edges. The interference relationship graph is then cut into low-interference subgraphs by a graph cutting algorithm to obtain a low-interference allocation scheme, i.e., the optimized initial allocation sequence. In the initial allocation sequence, the first slice is the first slice, the second slice is the second slice, the first slice is the best slice allocated first, the second slice is the alternative slice when the first slice is insufficient, and so on.

[0121] Finally, conflict detection is used to verify whether the optimized initial allocation sequence meets the resource requirement constraints. These constraints include: the total allocated bandwidth of each slice does not exceed the available bandwidth of the slice; the total number of connections in each slice does not exceed the maximum number of connections in the slice; and the interference intensity between each node does not exceed a preset interference intensity threshold. If multiple drones share the same first slice, resulting in insufficient available bandwidth for that slice (i.e., the sum of the bandwidth requirements of all drones in that slice exceeds the available bandwidth), then, based on scenario priority, the first slice of the lower-priority drone is adjusted to the second slice in its sequence, and the total bandwidth requirement of that slice is recalculated. If it still exceeds the remaining bandwidth, the next lowest-priority drones are adjusted until the sum of the bandwidth requirements of all drones corresponding to all slices is less than or equal to the available bandwidth of the slice. If there is a connection conflict (i.e., the total number of drone connections in a slice exceeds the maximum number of connections in that slice), then, based on scenario priority... The priority adjustment logic adjusts the second slice of low-priority drones to the first slice until the total number of drone connections in each slice is less than or equal to the maximum number of connections in the slice. If interference exceeds the limit, i.e., the interference intensity between a node corresponding to a certain drone and the nodes corresponding to its neighboring drones is greater than the preset interference intensity threshold, the low-priority drones are adjusted according to the scenario priority. Slices with interference intensity less than the preset interference intensity are obtained sequentially from the allocation sequence of the drone and adjusted to the first slice to ensure that the interference intensity between all nodes is less than or equal to the preset interference intensity after adjustment. Adjustment to non-adjacent frequency band slices; if necessary, the elastic resource pool is called to supplement resources. Through multiple rounds of iterative adjustment, all slice conflicts are resolved. Finally, the slice allocation sequence of the drone cluster is obtained through the slice allocation sequence of each drone, where the slice allocation sequence of the drone cluster only includes the first slice of each drone in the drone cluster.

[0122] For example, this embodiment assumes that the resource requirements and slice configuration requirements of the above-mentioned drone cluster and each drone in the drone cluster are used. Available slices include: URLLC slices. Its available bandwidth is Maximum number of connections is Anti-interference gain is URLLC slices Its available bandwidth is Maximum number of connections is Anti-interference gain is URLLC slices Its available bandwidth is Maximum number of connections is Anti-interference gain is eMBB slices Its available bandwidth is Maximum number of connections is Anti-interference gain is eMBB slices Its available bandwidth is Maximum number of connections is Anti-interference gain is eMBB slices Its available bandwidth is Maximum number of connections is Anti-interference gain is First, a candidate slice set corresponding to each drone is obtained through screening. 、 Excluding the avoidance bands in eMBB and its spectrum strategy, the corresponding candidate slice set is as follows: ; 、 and 、 Excluding URLLC and the avoidance bands in its spectrum strategy, the corresponding candidate slice set is obtained as follows: Next, the fit score between each drone and the candidate slices in its corresponding candidate slice set is calculated. First, the fit score is calculated. and Bandwidth matching Connection matching degree Loss compensation degree The fit score is obtained through the fit score formula. Similarly, calculate and Fit rating , Assuming Then we get the drone initial allocation sequence Similarly, the initial allocation sequence for the remaining drones is obtained; then, based on each drone and its corresponding candidate slice, the node set of the drone cluster is obtained. ,in, For drones With available slices Established nodes For drones With available slices Established nodes For drones With available slices The established nodes are used for calculation. and Geographical distance and interference intensity Assuming Greater than the preset interference intensity threshold Then, an edge is created between these two nodes and weighted. Following this step, all drones in the drone swarm and their corresponding candidate slices are traversed to obtain the edges of all nodes and weight them. The low-interference node corresponding to each drone is obtained through graph cutting algorithm optimization, and a low-interference subset of the drone swarm is constructed. Finally, through conflict checking, it was verified that the total bandwidth and connection requirements of each slice were within their available range, with no interference or over-limit issues, thus obtaining the final allocation sequence for the drone cluster: .

[0123] This embodiment incorporates geographical location and frequency band interference factors into the slice allocation process, thereby achieving the goal of avoiding inter-slice co-channel interference from the source. At the same time, it ensures the stability of resource supply through an elastic resource pool, further improving the rationality and reliability of low-altitude communication resource allocation for UAV clusters.

[0124] Furthermore, this embodiment provides a step of obtaining an optimized initial assignment sequence by constructing an interference relationship graph and using a graph cutting algorithm, including:

[0125] Based on the initial allocation sequence and real-time geographical location information of each UAV, an interference relationship map is obtained by setting nodes and calculating the interference intensity between nodes;

[0126] A subset of low-interference nodes is obtained based on the interference relationship graph using a graph cutting algorithm;

[0127] The optimized initial allocation sequence is obtained by sorting the low-interference node subset and the initial allocation sequence according to their fitness.

[0128] Here, a node is a binary combination formed by combining each UAV with a slice in its candidate slice set. The binary combination includes the UAV number and the candidate slice number. The interference intensity is a numerical value of the channel interference between two nodes, which is calculated by weighting distance and frequency band overlap. The graph cutting algorithm is used to cluster by calculating the eigenvalues ​​and eigenvectors of the Laplacian matrix of the graph. The low-interference node subset is a combination of nodes that have no interference edges between any two nodes or whose total interference weight is less than a threshold. The optimized initial allocation sequence is the allocation sequence of the corresponding slice in the low-interference node subset for each UAV, while maintaining the fitness priority.

[0129] Specifically, firstly, a candidate slice list for each UAV is obtained based on the initial allocation sequence, and then a node set of the UAV and its corresponding candidate slice is constructed. ,in Number the drone. Assign a candidate slice number, corresponding to each drone. 1 node The number of candidate slices is determined; the real-time latitude and longitude coordinates of the UAV are obtained through its onboard payload, such as a GPS module, and the distance between any two nodes is calculated using the semi-versus formula. Simultaneously obtain the center frequencies of the two nodes. , and bandwidth , Through formula Calculate the frequency band overlap The closer the center frequencies of two nodes are, the greater the frequency band overlap; according to The interference intensity between nodes is calculated, where For a set of nodes Middle node With nodes Interference intensity between them The weighting coefficients are determined by fitting historical interference data. The weight representing the impact of geographical distance on interference. The influence weight of frequency band overlap is characterized; the threshold for interference intensity is set as follows. ,like Greater than or equal to Then an edge is established between the two nodes. and will As edge weight Finally, a complete interference relationship map was obtained. ,in For a set of nodes, This is the set of edges between nodes where the interference intensity exceeds a threshold. Let the interference intensity weights be the values ​​corresponding to the edges; secondly, the optimization objective function is determined based on maximizing the total fitness score and minimizing the sum of the interference weights. ,in For nodes The compatibility rating Choose a variable for the node. This indicates that the node is selected. This indicates that it is not selected. To balance the adjustment coefficients for fitness and interference, For nodes With nodes The interference intensity weights corresponding to the edges between them; based on the interference relationship graph. Obtain the normalized Laplace matrix ,in This is the identity matrix, where the elements on the main diagonal are all 1s and all other elements are 0s. It is an adjacency matrix. Given a degree matrix, the first degree is obtained through eigenvalue decomposition. The eigenvectors corresponding to the smallest eigenvalues ​​are used to construct the feature matrix, which is then obtained through k-means clustering. A subset of low-interference nodes, wherein the sum of the interference weights between any two nodes in each subset is less than or equal to 1. , The number of drones in the drone cluster is defined. A subset of low-interference nodes with the highest total fitness score is selected to ensure that each drone corresponds to exactly one node. Then, the mapping relationship between drones and slices in the low-interference node subset is compared with the fitness ranking of the initial allocation sequence. If the selected slice is the first in the candidate slice list for that drone, it is directly retained. If it is not the first in the list, the slice is promoted to the first position in the sequence. The original first in the list and subsequent non-interference slices are adjusted to the candidate sequence in descending order of fitness score. Finally, an optimized initial allocation sequence is generated in which each drone contains only low-interference slices.

[0130] For example, such as Figure 4 As shown, this embodiment assumes that the above-mentioned drone cluster and the initial allocation sequence of each drone within the drone cluster are used. The initial allocation sequence is assumed to be... First, the node set of the drone swarm is obtained based on the initial allocation sequence. The result is obtained from the formula for the semi-versus. and Geographical distance ,according to get and Interference intensity Assuming Greater than the preset interference intensity threshold Then an edge is created between the two nodes and... As edge weights, this step iterates through all drones in the drone swarm and their corresponding candidate slices to obtain the edges of all nodes and weight them to form an interference relationship graph. Secondly, set the optimization objective function. Calculate the normalized Laplace matrix After eigenvalue decomposition, low-disturbance subsets are obtained through k-means clustering. Finally, it is compared with the initial sequence, and its... China The first slice is a low-interference subset, so it is retained and continued as a drone. The first slice, for The second slice, but it is The node with the highest overall fitness score in the objective function response is the low-interference node, therefore... Adjusted to The first slice, and the remaining drones are similarly processed to obtain the optimized initial allocation sequence of the final drone swarm: .

[0131] To address the issue of inter-slice interference, this embodiment constructs a node-based interference relationship graph between UAVs and candidate slices under low-altitude communication. The interference impact of geographical distance and frequency band overlap is quantified into graph weights. Then, a graph segmentation algorithm is used to select low-interference node combinations from a global optimization perspective. This reduces co-frequency interference from the source, shortens the resource allocation process latency, improves spectrum reuse efficiency, and reduces the pressure of subsequent conflict verification and adjustment.

[0132] Furthermore, such as Figure 5 As shown, this application embodiment provides a drone swarm communication resource allocation system based on 5G slicing technology, including:

[0133] The requirement analysis module obtains the slice configuration requirements for each drone based on the static basic parameters of each drone in the drone cluster and the preset scenario priority. The static basic parameters include model, payload type and operation scenario.

[0134] The data acquisition module obtains the operational data of each drone in the drone swarm based on the working scenario of the drone swarm. The operational data includes task characteristics, payload status, environmental interference, and topology changes.

[0135] The resource calculation module obtains the resource requirements of each UAV based on the running data through a fusion feature algorithm. The fusion feature algorithm includes a shared encoder and a multi-task decoder. The resource requirements include bandwidth requirements, connectivity requirements, and bandwidth loss.

[0136] The slice allocation module obtains the slice allocation sequence of the drone cluster based on the resource requirements and slice configuration requirements of each drone using a graph interference perception algorithm.

[0137] The demand analysis module, resource calculation module, and slice allocation module are all located on the server. The server receives data transmitted from the acquisition devices and stores it in the data acquisition module for further analysis. The acquisition devices include different types of UAV payloads.

[0138] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for allocating communication resources for unmanned aerial vehicle (UAV) swarms based on 5G slicing technology, characterized in that, The drone swarm includes intelligent inspection drones, logistics transport drones, high-altitude operation drones, and tethered emergency rescue drones, and the method includes: Based on the static basic parameters of each drone in the drone cluster, and combined with the preset scene priority, the slice configuration requirements of each drone are obtained. The static basic parameters include model, payload type and operation scene. Based on the working scenario of the drone swarm, the operational data of each drone in the drone swarm is obtained, and the operational data includes task characteristics, payload status, environmental interference, and topology changes; The resource requirements of each UAV are obtained through a fusion feature algorithm based on the operational data. The fusion feature algorithm includes a shared encoder and a multi-task decoder. The resource requirements include bandwidth requirements, connectivity requirements, and bandwidth loss. Based on the resource requirements and slice configuration requirements of each UAV, the slice allocation sequence of the UAV cluster is obtained through a graph interference perception algorithm; The step of obtaining the resource requirements of each UAV based on the operational data using a feature fusion algorithm includes: A standardized feature matrix is ​​obtained by standardizing the aforementioned operational data. Based on the standardized feature matrix, a shared feature vector is obtained through a shared encoder; The resource requirements of each UAV are obtained through a multi-task decoder based on the shared feature vector; The step of obtaining the slice allocation sequence of the drone cluster using a graph interference perception algorithm based on the resource requirements and slice configuration requirements of each drone includes: A candidate slice set is obtained based on the slice configuration requirements and spectrum strategy of each UAV; Based on the resource requirements of each UAV and the candidate slice set, a suitability score is obtained through weighted calculation, and an initial allocation sequence for each UAV is obtained by sorting in descending order. Based on the initial allocation sequence and real-time geographic location information of each UAV, an optimized initial allocation sequence is obtained by constructing an interference relationship graph and using a graph cutting algorithm. The real-time geographic location information is obtained through the UAV payload. Based on the optimized initial allocation sequence, the slice allocation sequence of the UAV cluster is obtained through conflict checking and adjustment.

2. The method for allocating unmanned aerial vehicle (UAV) swarm communication resources based on 5G slicing technology according to claim 1, characterized in that, The step of obtaining the slice configuration requirements for each drone based on the static basic parameters of each drone in the drone cluster and in combination with preset scene priorities includes: A structured data set is constructed based on the static basic parameters of each drone in the drone cluster; Based on the structured data set and the preset operation scenario priority, the service quality index of each UAV is obtained; The slice configuration requirements for each UAV are obtained based on the service quality indicators.

3. The method for allocating drone swarm communication resources based on 5G slicing technology according to claim 1, characterized in that, The process of obtaining the shared feature vector based on the standardized feature matrix through a shared encoder includes: Based on the standardized feature matrix, a temporal feature vector is obtained through a temporal convolutional network; Based on the temporal feature vector, a multi-head attention mechanism is used to obtain the scene key feature vector; Based on the key feature vectors of the scenario, a shared feature vector is obtained through mapping using a fully connected layer.

4. The method for allocating unmanned aerial vehicle (UAV) swarm communication resources based on 5G slicing technology according to claim 1, characterized in that, The multi-task decoder includes a bandwidth requirement decoder, a connection requirement decoder, and a bandwidth loss decoder. The step of obtaining the resource requirements of each UAV through the multi-task decoder based on the shared feature vector includes: Based on the shared feature vector, the bandwidth requirement value and the prediction curve for the future preset time period of each UAV are obtained through the bandwidth requirement decoder; Based on the shared feature vector and the optimized connection number prediction model, the optimal connection number and spectrum strategy for each UAV are obtained. The optimized connection number prediction model is located in the multi-agent decision network of the connection demand decoder and is obtained by training through historical shared feature vectors. The bandwidth loss value is obtained by using the bandwidth loss decoder based on the shared feature vector. Based on the bandwidth requirement, optimal number of connections, spectrum strategy, and bandwidth loss value of each UAV, the resource requirements of each UAV are obtained through data integration, parameter normalization, and collaborative verification.

5. The method for allocating unmanned aerial vehicle (UAV) swarm communication resources based on 5G slicing technology according to claim 4, characterized in that, The connection demand decoder includes a multi-agent decision network and a spectrum compliance environment simulator. The optimized connection number prediction model is trained using historical shared feature vectors and includes: The initial connection number prediction model is obtained through a multi-agent decision network based on historical shared feature vectors. The spectrum constraints are obtained through a spectrum compliance environment simulator based on the historical operational data of the drone cluster. Based on the initial connection number prediction model and spectral constraints, and using the shared feature vector as the state input, an optimized connection number prediction model is obtained through training.

6. The method for allocating unmanned aerial vehicle (UAV) swarm communication resources based on 5G slicing technology according to claim 1, characterized in that, The process of constructing an interference relationship graph and obtaining the optimized initial allocation sequence using a graph cutting algorithm includes: Based on the initial allocation sequence and real-time geographical location information of each UAV, an interference relationship map is obtained by setting nodes and calculating the interference intensity between nodes; Based on the interference relationship graph, a subset of low-interference nodes is obtained using a graph cutting algorithm; The optimized initial allocation sequence is obtained by sorting the low-interference node subset and the initial allocation sequence according to their fitness.

7. The method for allocating drone swarm communication resources based on 5G slicing technology according to claim 1, characterized in that, The step of obtaining the operational data of each drone in the drone cluster based on the working scenario of the drone cluster includes: Data acquisition rules are derived based on the working scenario of the drone swarm, and the data acquisition rules include acquisition frequency and positioning accuracy. The operational data of each UAV is obtained according to the data acquisition rules. The operational data includes mission characteristics, payload status, environmental interference, and topology changes.

8. A drone swarm communication resource allocation system based on 5G slicing technology, used to implement the drone swarm communication resource allocation method based on 5G slicing technology as described in any one of claims 1-7, characterized in that, The system includes: The requirement analysis module obtains the slice configuration requirements of each drone based on the static basic parameters of each drone in the drone cluster and in combination with the preset scenario priority. The static basic parameters include model, payload type and operation scenario. The data acquisition module obtains the operational data of each drone in the drone cluster based on the working scenario of the drone cluster. The operational data includes task characteristics, payload status, environmental interference, and topology changes. The resource calculation module obtains the resource requirements of each UAV based on the operating data through a fusion feature algorithm. The fusion feature algorithm includes a shared encoder and a multi-task decoder. The resource requirements include bandwidth requirements, connection requirements, and bandwidth loss. The step of obtaining the resource requirements of each UAV based on the operational data using a feature fusion algorithm includes: A standardized feature matrix is ​​obtained by standardizing the aforementioned operational data. Based on the standardized feature matrix, a shared feature vector is obtained through a shared encoder; The resource requirements of each UAV are obtained through a multi-task decoder based on the shared feature vector; The slice allocation module obtains the slice allocation sequence of the drone cluster based on the resource requirements and slice configuration requirements of each drone using a graph interference perception algorithm. The step of obtaining the slice allocation sequence of the drone cluster using a graph interference perception algorithm based on the resource requirements and slice configuration requirements of each drone includes: A candidate slice set is obtained based on the slice configuration requirements and spectrum strategy of each UAV; Based on the resource requirements of each UAV and the candidate slice set, a suitability score is obtained through weighted calculation, and an initial allocation sequence for each UAV is obtained by sorting in descending order. Based on the initial allocation sequence and real-time geographic location information of each UAV, an optimized initial allocation sequence is obtained by constructing an interference relationship graph and using a graph cutting algorithm. The real-time geographic location information is obtained through the UAV payload. Based on the optimized initial allocation sequence, the slice allocation sequence of the UAV cluster is obtained through conflict checking and adjustment.

Images