Method for dynamically managing and controlling on-board digital load resources in view of unmanned aerial vehicle mission characteristics

Abstract

This invention discloses a method for dynamic management and control of on-board digital payload resources based on the characteristics of unmanned aerial vehicle (UAV) missions. The method includes: constructing a UAV mission-satellite payload hypergraph model to achieve high-order correlation modeling between UAV mission characteristics and on-board payload resources; extracting dynamic topological features and mission requirement features of the UAV-satellite network based on a graph neural network to generate feature embedding vectors; mining the causal relationship between resource allocation and mission effectiveness based on a causal inference model to determine core management factors; constructing a resource allocation optimization model that integrates hypergraph features and causal factors based on the UAV mission-satellite payload hypergraph model, the feature embedding vectors, and the core management factors, and outputting a dynamic management and control strategy for on-board digital payload resources; and updating model parameters in real time based on UAV flight status feedback to achieve online iterative optimization of the management and control strategy. The corresponding system, electronic equipment, and computer-readable storage medium are also disclosed.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of satellite communication and artificial intelligence, specifically to a method for dynamic management and control of on-board digital payload resources oriented towards the characteristics of UAV missions, and is particularly applicable to an integrated "sky-ground" communication system composed of flexible high-throughput satellites and long-endurance UAVs based on digital transparent processors (DTP). Background Technology

[0002] With the application of digitized flexible payload technology for basic data and the development of space-ground resource management systems, high-throughput satellites are shifting from the traditional high-capacity model to a composite model of "high capacity + high flexibility." Among these technologies, the Digital Transparent Processor (DTP) serves as a core technological support, enabling pooled management and flexible scheduling of satellite application resources, providing feasible conditions for the high-throughput resource needs of mobile terminals and special users. Long-endurance unmanned aerial vehicles (UAVs), as a key platform for airborne information acquisition and transmission, place stringent demands on satellite communication speed, stability, and dynamic resource allocation capabilities in scenarios such as hotspot area intelligence feedback and cross-domain mission execution.

[0003] Existing methods for managing on-board digital payload resources mostly employ static allocation or rule-based dynamic adjustment strategies, which have the following core problems: First, they fail to fully integrate the dynamic characteristics of UAV mission profiles (such as mission priority, data transmission type, and flight trajectory), resulting in low matching between resource allocation and mission requirements, and issues such as bandwidth congestion and delayed intelligence feedback in hotspot areas. Second, traditional management algorithms struggle to rapidly reconfigure resources in the face of dynamic changes in network topology caused by UAV cross-domain flights, failing to meet the low-latency requirements of "one-hop" private network deployments. Third, resource management decisions rely heavily on empirical models, lacking the ability to explore the causal relationships between mission requirements, network status, and payload performance, leading to insufficient robustness in decision-making.

[0004] Graph Neural Networks (GNNs) have a natural advantage in handling network problems with dynamic topological changes, effectively capturing the correlation features between nodes and edges. Causal inference techniques can overcome the correlation limitations of traditional statistical learning, uncovering the causal relationship between resource allocation and task performance, and improving the interpretability of decisions. Hypergraph theory can solve complex many-to-many relationship problems between satellite payloads, UAV nodes, and ground terminals through high-order association modeling. Integrating these artificial intelligence technologies into on-board digital payload resource management is a key path to solving the problem of dynamic resource allocation in UAV communication scenarios. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a dynamic management and control method for on-board digital payload resources tailored to the characteristics of UAV missions. This method, based on graph neural networks, causal inference, and hypergraph theory, constructs a multi-dimensional feature-driven intelligent management and control model to achieve real-time and accurate matching of on-board digital payload resources with UAV mission requirements, thereby improving resource utilization efficiency and mission execution effectiveness.

[0006] The first aspect of this invention is to provide a method for dynamic management and control of on-board digital payload resources oriented towards the characteristics of unmanned aerial vehicle (UAV) missions, applicable to on-board DTP payloads, A unified "sky-ground" communication system comprising one drone node and M ground control nodes includes the following steps:

[0007] S1. Construct a hypergraph model of UAV mission-satellite payload to realize high-order correlation modeling between UAV mission characteristics and on-board payload resources;

[0008] S2, based on graph neural networks, extracts dynamic topological features and mission requirement features of UAV-satellite networks and generates feature embedding vectors;

[0009] S3, based on the causal inference model, explore the causal relationship between resource allocation and task effectiveness, and determine the core control factors;

[0010] S4. Based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factor, construct a resource allocation optimization model that integrates hypergraph features and causal factors, and output a dynamic control strategy for on-board digital payload resources.

[0011] S5, based on drone flight status feedback, updates model parameters in real time to achieve online iterative optimization of control strategies.

[0012] Preferably, S1 includes:

[0013] S11, establish a hypergraph model to describe the many-to-many relationships between UAV missions, on-board payload resources, and ground requirements, defined as... ,in, Represents a vertex set. Represents a superedge set. Represents the weight matrix; where:

[0014] The vertex set It contains three types of vertices, namely, drone mission vertices. Onboard payload resource vertex and ground demand peak ,in, The number of resolvable resource units for the onboard DTP payload;

[0015] The hyperedge set In the diagram, each hyperedge corresponds to a set of "task-resource-requirement" associations, i.e. ,and It contains at least one type of vertex;

[0016] The weight matrix It includes hyperedge weights and vertex weights, which represent the strength of the association and the importance of the vertex, respectively;

[0017] S12, define the UAV mission feature vector as follows: This allows for the quantification of the impact of UAV mission characteristics on the hypergraph structure, where: Task priority is divided into 1-5 levels. This refers to data transmission rate requirements, measured in Mbps. The speed of the drone is expressed in m / s. The importance of the data is indicated by a range of 0-1. The deadline for the task is in seconds.

[0018] S13, define the on-board payload resource feature vector as follows: ,in: This refers to resource bandwidth, measured in Mbps. Resource load factor, %. Resource scheduling delay, in milliseconds. The duration of resource availability, measured in seconds (s).

[0019] S14, calculate the hyperedge weight based on the UAV mission feature vector and the on-board payload resource feature vector. The hyperedge weight is obtained by multiplying the task-resource matching degree and the demand-resource matching degree. The calculation formula is shown in the following formula (1):

[0020] (1);

[0021] in, The cosine similarity function is used. This represents the feature vector of ground demand. ,in This is the normalized form of the onboard payload resource feature vector, satisfying... ;

[0022] S15, determine the vertex weights.

[0023] Preferably, the method for determining vertex weights includes two methods:

[0024] The first method involves directly linking vertex weights to UAV mission characteristics and onboard resource status, determining the vertex weights by matching the core features of the corresponding vertex. For UAV mission vertex weights, the weights are obtained based on mission feature vectors through feature normalization and importance weighting. For onboard payload resource vertex weights, the weights are calculated based on resource feature vectors combined with resource scarcity. For ground demand vertex weights, the weights are quantified by referring to ground demand feature vectors and considering indicators such as urgency and coverage.

[0025] or

[0026] The second method is to determine the vertex weight based on collaboration with the hyperedge weight, including: the vertex weight is determined by reverse iterative optimization through the hyperedge weight, that is, the weight of vertices that frequently participate in high-weight hyperedges is dynamically increased, and the consistency between the association strength and the importance of the vertex is ensured.

[0027] Preferably, S2 includes:

[0028] S21, Construct a graph neural network, wherein the graph neural network consists of a construction layer, a temporal feature fusion layer, and a feature embedding layer;

[0029] S21 includes:

[0030] (1) Constructing a graph construction layer, including: converting the hypergraph model into a dynamic graph sequence. ,in, For the static graph at time t, the edge weights are obtained by mapping the hyperedge weights, i.e. , For containing vertices and The superedge;

[0031] (2) Constructing a temporal feature fusion layer, including: using a structure combining gated recurrent units (GRU) and graph convolutional networks (GCN) to update the graph features at each time step; wherein, the gated recurrent units (GRU) are used to fuse temporal features, and the graph convolutional networks (GCN) are used to extract spatial features. The structure combining gated recurrent units (GRU) and graph convolutional networks (GCN) is used to update the graph features at each time step in the following manner: (2) Dynamic graph convolutional feature update method.

[0032] (2);

[0033] in, As vertex exist Time of the first The feature vector of the layer, The spatial convolution weight matrix, For bias terms, It is the ReLU activation function; As vertex The neighbor set;

[0034] The updated graph features at each time step are then used for temporal feature fusion and updating based on equations (3-1) to (3-4), where equation (3) is:

[0035] (3-1)

[0036] (3-2);

[0037] (3-3);

[0038] (3-4);

[0039] in, To update the gate output value, To reset the gate output value, As vertex exist The final feature embedding vector at time step 1. For element-wise product, The number of convolutional layers in the graph; Use the Sigmoid activation function; To update the weight parameter matrix of the gate; This is a vector concatenation operator; for Peak of Time In the graph convolutional network The output feature vector of the layer; for Peak of Time In the graph convolutional network The output feature vector of the layer; To update the bias parameter vector of the gate; The weight parameter matrix for resetting the gate; This is the offset parameter vector for resetting the gate; It is the hyperbolic tangent activation function; The weight parameter matrix for the candidate features; The bias parameter vector for candidate features;

[0040] (3) Constructing a feature embedding layer, including: concatenating the feature embedding vectors of all vertices to obtain a global feature matrix. ;in, This is the global feature matrix, containing the feature embedding information of all vertices; This represents the total number of vertices in the hypergraph model. For the first The vertex is in Feature embedding vector at time step; superscript This represents the transpose operation of a matrix, ensuring that the rows of the feature matrix correspond to vertices and the columns correspond to feature dimensions, thus adapting to the input requirements of subsequent causal inference and resource optimization modules. This global feature matrix fully integrates... The spatiotemporal correlation characteristics, mission requirement characteristics, and resource status characteristics of the real-time drone-satellite network are the core feature inputs for realizing intelligent control and management decisions, and are used for subsequent causal inference and resource optimization.

[0041] S22, Based on the graph neural network, extract the dynamic topology features and mission requirement features of the UAV-satellite network, and generate feature embedding vectors.

[0042] Preferably, step S3, which involves mining the causal relationship between resource allocation and task performance based on a causal inference model to determine core control factors, includes: employing a dual machine learning framework to mine the causal relationship between resource allocation and task performance to determine core control factors, specifically including:

[0043] S31, Define variables, including: Set the processing variable The allocation ratio of onboard payload resources, with a value range of 0-1, is the result variable. For task performance, a normalized metric is defined, encompassing overall data transmission rate, latency, and success rate, with values ​​ranging from 0 to 1. Covariates... The feature matrix output in step S2 It includes UAV mission characteristics, resource status characteristics, and network topology characteristics, covariates. Dimensions and global feature matrix Consistency is the core input data for causal inference models to capture the correlation between variables;

[0044] S32, construct three sub-models, namely:

[0045] Result prediction model: ;

[0046] Processing variable prediction model: ;

[0047] Residual estimation model ;

[0048] in, The causal effect is the average treatment effect (ATE). and This is the error term; Indicated based on covariates Processing variables The prediction function is represented by the global feature matrix. The processing variables are obtained by fitting a machine learning model as input. The predicted value;

[0049] S33, Calculating the causal effect model Corresponding causal effect This includes: a prediction model based on the training results of the gradient boosting tree XGBoost and a prediction model for the treatment variables, and the causal effect is obtained through residual estimation and average treatment effect ATE, wherein the calculation formula of the causal effect is shown in the following formula (4):

[0050] (4);

[0051] in, and For the trained prediction model, The number of samples;

[0052] S34, Identify core control factors, including: calculating covariates. The impact of various characteristics on causal effects Influence weights, defining feature importance indicators As shown in equation (5):

[0053] (5);

[0054] in, For covariate dimensions, For the first Select one covariate. The characteristics of the UAV mission are used as core control factors. Typical factors include UAV mission priority, data transmission rate requirements, on-board resource load rate, and network topology connectivity.

[0055] Preferably, in step S4, constructing a resource allocation optimization model that integrates hypergraph features and causal factors based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factor, and outputting a dynamic control strategy for on-board digital payload resources, includes: constructing a resource allocation optimization model with the core control factor as a constraint and "maximizing resource utilization efficiency" and "maximizing mission effectiveness" as dual objectives; using a hypergraph attention mechanism to integrate multi-objective weights; and solving for the optimal resource allocation strategy using the Lagrange multiplier method, specifically including:

[0056] S41, Define an objective function, which includes a primary objective function and a secondary objective function. The primary objective function is used to maximize task performance, i.e. The secondary objective function is used to maximize resource utilization efficiency, i.e. ;

[0057] in, For the first The weight of each drone mission is determined by the mission priority. For the first The effectiveness of each task To be assigned to the Resource bandwidth for each task;

[0058] S42, the constraints are determined as follows:

[0059] Total resource constraints: The sum of the resource allocation ratios for all tasks shall not exceed 1;

[0060] Task requirements constraints: Assigned to the The bandwidth of each task meets its rate requirements.

[0061] Delay constraints: Resource scheduling delays shall not exceed the task deadline requirements;

[0062] Causal constraint: , No. Causal effects of individual tasks Not lower than the minimum threshold This ensures the effectiveness of resource allocation;

[0063] S43, Perform hypergraph attention weight fusion, including: to balance the weights of the two objectives, calculate the attention weights of the task and resources based on the hypergraph model. The formula is shown in equation (6) below, which is the formula for calculating the hypergraph attention weight:

[0064] (6);

[0065] in, For attention parameter vectors, This is a vector concatenation operation. For activation functions;

[0066] S44, Solve the model, including: transforming the bi-objective optimization problem into a single-objective problem, constructing a weighted objective function based on formula (7) and formula (8), and using the Lagrange multiplier method to solve the optimal solution of the model;

[0067] (7);

[0068] (8);

[0069] in, The value of the Lagrange function. To weight the overall task effectiveness, The secondary objective weights are determined by the average of the hypergraph attention weights. For resource utilization efficiency, among which, ;, and It is a Lagrange multiplier. For the first The optimal resource allocation ratio for each drone mission; For the first The rate requirement of each task This represents the total bandwidth of the onboard payload. This is the extremum condition for the Lagrange function; The partial derivative of task effectiveness with respect to the resource allocation ratio; To be assigned to the Bandwidth for each task.

[0070] Preferably, step S5, based on UAV flight status feedback, updates model parameters in real time to achieve online iterative optimization of the control strategy. This includes constructing an online model parameter update mechanism to adapt to the dynamic changes in UAV flight status and mission requirements. The specific process includes:

[0071] S51 provides feedback on the flight status of unmanned aerial vehicles (UAVs), including: real-time uploading of flight and communication status by the UAVs, and collection of onboard payload resource status by ground control nodes, forming a feedback data set. ;

[0072] S52, Calculation error, including: the deviation between the calculated actual task performance and the model's predicted performance. ,like This triggers a model update;

[0073] S53, perform parameter updates, including: updating the parameters of the graph neural network and the causal inference model using mini-batch gradient descent, dynamically adjusting the learning rate using a cosine annealing strategy, and calculating the online update of the model parameters based on equations (9) and (10) as follows:

[0074] (9);

[0075] (10);

[0076] in, For the set of model parameters, for Learning rate at any given moment For batch size, Let the mean squared error loss function be . , , This represents the maximum number of iterations.

[0077] S54, perform strategy updates, including recalculating the optimal resource allocation ratio based on the updated model. A new onboard payload resource management strategy is generated and sent to the onboard DTP payload controller via the satellite-to-ground link.

[0078] A second aspect of the present invention is to provide a dynamic management and control system for on-board digital payload resources oriented towards the characteristics of unmanned aerial vehicle (UAV) missions, applicable to on-board DTP payloads, A "sky-ground" integrated communication system comprising one UAV node and M ground control nodes is used to implement the method in the first aspect, including:

[0079] The hypergraph model building module (101) is used to build a hypergraph model of UAV mission-satellite payload to realize high-order correlation modeling between UAV mission characteristics and on-board payload resources.

[0080] The feature embedding vector generation module (102) is used to extract the dynamic topological features and mission requirement features of the UAV-satellite network based on the graph neural network and generate feature embedding vectors.

[0081] The core control factor determination module (103) is used to mine the causal relationship between resource allocation and task effectiveness based on the causal inference model and determine the core control factors.

[0082] The resource allocation optimization model construction module (104) is used to construct a resource allocation optimization model that integrates hypergraph features and causal factors based on the UAV mission-satellite payload hypergraph model, the feature embedding vector and the core control factor, and output the dynamic control strategy of on-board digital payload resources;

[0083] The dynamic control and optimization module (105) is used to update model parameters in real time based on UAV flight status feedback, so as to realize online iterative optimization of control strategies.

[0084] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in the first aspect.

[0085] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions which can be read by a processor and executed as described in the first aspect.

[0086] The beneficial effects of the method and system of the present invention are as follows:

[0087] (1) A high-order association modeling method based on hypergraph theory is proposed to overcome the limitation that traditional graph models cannot describe the many-to-many relationship of "task-resource-requirement", and realize the accurate matching of UAV mission characteristics and on-board payload resources;

[0088] (2) Construct a feature extraction framework that integrates dynamic graph neural networks and causal inference, which not only captures the spatiotemporal dynamic features of network topology, but also explores the causal relationship between resource allocation and task performance, thereby improving the interpretability and robustness of decision-making.

[0089] (3) Design a dual-objective optimization model that integrates the hypergraph attention mechanism, and dynamically balance task performance and resource efficiency through attention weights to avoid the subjective defects of traditional weighted methods;

[0090] (4) Establish an online iteration mechanism based on real-time feedback, combined with a cosine annealing learning rate strategy, to ensure that the model continuously adapts to the dynamically changing task and network state in the cross-domain flight scenario of UAVs. Attached Figure Description

[0091] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0092] Figure 1 This is a flowchart of a method provided according to an embodiment of the present invention;

[0093] Figure 2 A system architecture diagram provided according to an embodiment of the present invention;

[0094] Figure 3 This is a schematic diagram of the UAV mission-satellite payload hypergraph model structure provided in an embodiment of the present invention;

[0095] Figure 4 This is a schematic diagram of the structure of a dynamic graph neural network provided according to an embodiment of the present invention;

[0096] Figure 5 A schematic diagram of a dual machine learning causal inference framework provided according to an embodiment of the present invention;

[0097] Figure 6 This is a hardware architecture diagram of an integrated "sky-ground" communication system provided according to an embodiment of the present invention;

[0098] Figure 7 This is a performance comparison chart of a hotspot area intelligence backhaul scenario provided by an embodiment of the present invention;

[0099] Figure 8 This is a comparison chart of resource reconfiguration time in a cross-domain task execution scenario provided by an embodiment of the present invention;

[0100] Figure 9 This is a structural diagram of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0101] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0102] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0103] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0104] like Figure 1 As shown, the first aspect of the present invention is to provide a method for dynamic management and control of on-board digital payload resources oriented towards the characteristics of unmanned aerial vehicle (UAV) missions, applicable to on-board DTP payloads, A unified "sky-ground" communication system comprising one drone node and M ground control nodes includes the following steps:

[0105] S1. Construct a hypergraph model of UAV mission-satellite payload to realize high-order correlation modeling between UAV mission characteristics and on-board payload resources;

[0106] S2, based on graph neural networks, extracts dynamic topological features and mission requirement features of UAV-satellite networks and generates feature embedding vectors;

[0107] S3, based on the causal inference model, explore the causal relationship between resource allocation and task effectiveness, and determine the core control factors;

[0108] S4. Based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factor, construct a resource allocation optimization model that integrates hypergraph features and causal factors, and output a dynamic control strategy for on-board digital payload resources.

[0109] S5, based on drone flight status feedback, updates model parameters in real time to achieve online iterative optimization of control strategies.

[0110] In a preferred embodiment, S1 includes:

[0111] S11, establish a hypergraph model to describe the many-to-many relationships between UAV missions, on-board payload resources, and ground requirements, defined as... ,in, Represents a vertex set. Represents a superedge set. Represents the weight matrix; where:

[0112] The vertex set It contains three types of vertices, namely, drone mission vertices. Onboard payload resource vertex and ground demand peak ,in, The number of resolvable resource units for the onboard DTP payload;

[0113] The hyperedge set In the diagram, each hyperedge corresponds to a set of "task-resource-requirement" associations, i.e. ,and It contains at least one type of vertex;

[0114] The weight matrix It includes hyperedge weights and vertex weights, which represent the strength of the association and the importance of the vertex, respectively;

[0115] S12, define the UAV mission feature vector as follows: This allows for the quantification of the impact of UAV mission characteristics on the hypergraph structure, where: Task priority is divided into 1-5 levels. This refers to data transmission rate requirements, measured in Mbps. The speed of the drone is expressed in m / s. The importance of the data is indicated by a range of 0-1. The deadline for the task is in seconds.

[0116] S13, define the on-board payload resource feature vector as follows: ,in: This refers to resource bandwidth, measured in Mbps. Resource load factor, %. Resource scheduling delay, in milliseconds. The duration of resource availability, measured in seconds (s).

[0117] S14, calculate the hyperedge weight based on the UAV mission feature vector and the on-board payload resource feature vector. The hyperedge weight is obtained by multiplying the task-resource matching degree and the demand-resource matching degree. The calculation formula is shown in the following formula (1):

[0118] (1);

[0119] in, The cosine similarity function is used. This represents the feature vector of ground demand. ,in This is the normalized form of the onboard payload resource feature vector, satisfying... ;

[0120] S15, determine the vertex weights; there are two methods for determining vertex weights:

[0121] The first method: Vertex weights are directly related to UAV mission characteristics and on-board resource status, and are determined by matching the core features of the corresponding vertices. For UAV mission vertex weights: based on the mission feature vector, they are obtained through feature normalization and importance weighting (e.g., mission priority has a higher weight); for on-board payload resource vertex weights: based on the resource feature vector, combined with resource scarcity (e.g., low load rate vertices have a higher weight); for ground demand vertex weights: referencing the ground demand feature vector, they are quantified according to indicators such as urgency and coverage.

[0122] or

[0123] The second method is to determine the vertex weight based on collaboration with the hyperedge weight, including: the vertex weight is determined by reverse iterative optimization through the hyperedge weight, that is, the weight of vertices that frequently participate in high-weight hyperedges is dynamically increased, and the consistency between the association strength and the importance of the vertex is ensured.

[0124] This forms a complete hypergraph model, which transforms the originally scattered task requirements, resource status, and ground requirements into structured, interconnected data, providing high-level feature support for subsequent intelligent management and control.

[0125] In a preferred embodiment, S2 includes:

[0126] S21, Construct a graph neural network, wherein the graph neural network consists of a construction layer, a temporal feature fusion layer, and a feature embedding layer;

[0127] Considering the dynamic changes in network topology caused by UAV cross-domain flights, a Dynamic Graph Neural Network (DGNN) is used to extract the spatiotemporal correlation features between network topology and task requirements. S21 includes:

[0128] (1) Constructing a graph construction layer, including: converting the hypergraph model into a dynamic graph sequence. ,in, For the static graph at time t, the edge weights are obtained by mapping the hyperedge weights, i.e. , For containing vertices and The superedge;

[0129] (2) Constructing a temporal feature fusion layer, including: using a structure combining gated recurrent units (GRU) and graph convolutional networks (GCN) to update the graph features at each time step; wherein, the gated recurrent units (GRU) are used to fuse temporal features, and the graph convolutional networks (GCN) are used to extract spatial features. The structure combining gated recurrent units (GRU) and graph convolutional networks (GCN) is used to update the graph features at each time step in the following manner: (2) Dynamic graph convolutional feature update method.

[0130] (2);

[0131] in, As vertex exist Time of the first The feature vector of the layer, The spatial convolution weight matrix, For bias terms, It is the ReLU activation function; As vertex The neighbor set;

[0132] The updated graph features at each time step are then used for temporal feature fusion and updating based on equations (3-1) to (3-4), where equation (3) is:

[0133] (3-1)

[0134] (3-2);

[0135] (3-3);

[0136] (3-4);

[0137] in, To update the gate output value, To reset the gate output value, As vertex exist The final feature embedding vector at time step 1. For element-wise product, The number of convolutional layers in the graph; Use the Sigmoid activation function; To update the weight parameter matrix of the gate; This is a vector concatenation operator; for Peak of Time In the graph convolutional network The output feature vector of the layer; for Peak of Time In the graph convolutional network The output feature vector of the layer; To update the bias parameter vector of the gate; The weight parameter matrix for resetting the gate; This is the offset parameter vector for resetting the gate; It is the hyperbolic tangent activation function; The weight parameter matrix for the candidate features; is the bias parameter vector for candidate features.

[0138] (3) Constructing a feature embedding layer, including: concatenating the feature embedding vectors of all vertices to obtain a global feature matrix. ;in, This is the global feature matrix, containing the feature embedding information of all vertices; This represents the total number of vertices in the hypergraph model. For the first The vertex is in Feature embedding vector at time step; superscript This represents the transpose operation of a matrix, ensuring that the rows of the feature matrix correspond to vertices and the columns correspond to feature dimensions, thus adapting to the input requirements of subsequent causal inference and resource optimization modules. This global feature matrix fully integrates... The spatiotemporal correlation characteristics, mission requirement characteristics, and resource status characteristics of the real-time drone-satellite network are the core feature inputs for realizing intelligent control and management decisions, and are used for subsequent causal inference and resource optimization.

[0139] S22, Based on the graph neural network, extract the dynamic topology features and mission requirement features of the UAV-satellite network, and generate feature embedding vectors.

[0140] As a preferred implementation, step S3, which involves mining the causal relationship between resource allocation and task performance based on a causal inference model to determine core control factors, includes: to address the problem of "correlation ≠ causation" in traditional resource management, using a Double Machine Learning (DML) framework to mine the causal relationship between resource allocation (processing variable) and task performance (outcome variable) to determine core control factors;

[0141] In this embodiment, step S3 includes:

[0142] S31, Define variables, including: Set the processing variable The allocation ratio of onboard payload resources, with a value range of 0-1, is the result variable. For task performance, a normalized metric is defined, encompassing overall data transmission rate, latency, and success rate, with values ​​ranging from 0 to 1. Covariates... The feature matrix output in step S2 It includes UAV mission characteristics, resource status characteristics, and network topology characteristics, covariates. Dimensions and global feature matrix Consistency is the core input data for causal inference models to capture the correlation between variables;

[0143] S32, construct three sub-models, namely:

[0144] Result prediction model: ;

[0145] Processing variable prediction model: ;

[0146] Residual estimation model ;

[0147] in, The causal effect is the average treatment effect (ATE). and This is the error term; Indicated based on covariates Processing variables The prediction function is represented by the global feature matrix. (The spatiotemporal characteristics of UAV missions, resource status, and network topology) are used as input, and the processing variables are obtained by fitting a machine learning model. The predicted value.

[0148] S33, Calculating the causal effect model Corresponding causal effect This includes: a prediction model based on the training results of the gradient boosting tree XGBoost and a prediction model for the treatment variables, and the causal effect is obtained through residual estimation and average treatment effect ATE, wherein the calculation formula of the causal effect is shown in the following formula (4):

[0149] (4);

[0150] in, and For the trained prediction model, The sample size is [number of trees]. The XGBoost decision tree has 100 decision trees, a maximum depth of 8, and the core control factors are selected based on [criteria]. .

[0151] S34, Identify core control factors, including: calculating covariates. The impact of various characteristics on causal effects Influence weights, defining feature importance indicators As shown in equation (5):

[0152] (5);

[0153] in, For covariate dimensions, For the first Select one covariate. The characteristics of the UAV mission are used as core control factors. Typical factors include UAV mission priority, data transmission rate requirements, on-board resource load rate, and network topology connectivity.

[0154] As a preferred implementation, step S4, based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factor, constructs a resource allocation optimization model that integrates hypergraph features and causal factors, and outputs a dynamic control strategy for on-board digital payload resources, including: using the core control factor as a constraint, with "maximizing resource utilization efficiency" and "maximizing mission effectiveness" as dual objectives, constructing a resource allocation optimization model, using a hypergraph attention mechanism to integrate multi-objective weights, and solving for the optimal resource allocation strategy using the Lagrange multiplier method.

[0155] In this embodiment, S4 includes:

[0156] S41, Define an objective function, which includes a primary objective function and a secondary objective function. The primary objective function is used to maximize task performance, i.e. The secondary objective function is used to maximize resource utilization efficiency, i.e. ;

[0157] in, For the first The weight of each drone mission is determined by the mission priority. For the first The effectiveness of each task To be assigned to the Resource bandwidth for each task;

[0158] S42, the constraints are determined as follows:

[0159] Total resource constraints: The sum of the resource allocation ratios for all tasks shall not exceed 1;

[0160] Task requirements constraints: Assigned to the The bandwidth of each task meets its rate requirements.

[0161] Delay constraints: Resource scheduling delays shall not exceed the task deadline requirements;

[0162] Causal constraint: , No. Causal effects of individual tasks Not lower than the minimum threshold This ensures the effectiveness of resource allocation.

[0163] S43, Perform hypergraph attention weight fusion, including: to balance the weights of the two objectives, calculate the attention weights of the task and resources based on the hypergraph model. The formula is shown in equation (6) below, which is the formula for calculating the hypergraph attention weight:

[0164] (6);

[0165] in, For attention parameter vectors, This is a vector concatenation operation. The activation function has a slope of 0.2.

[0166] S44, Solve the model, including: transforming the bi-objective optimization problem into a single-objective problem, constructing a weighted objective function based on formula (7) and formula (8), and using the Lagrange multiplier method to solve the optimal solution of the model;

[0167] (7);

[0168] (8);

[0169] in, The value of the Lagrange function. To weight the overall task effectiveness, The secondary objective weights are determined by the average of the hypergraph attention weights. For resource utilization efficiency, among which, ;, and It is a Lagrange multiplier. For the first The optimal resource allocation ratio for each drone mission; For the first The rate requirement of each task This represents the total bandwidth of the onboard payload. This is the extremum condition for the Lagrange function; The partial derivative of task effectiveness with respect to the resource allocation ratio; To be assigned to the Bandwidth for each task.

[0170] In a preferred implementation, step S5, based on UAV flight status feedback, updates model parameters in real time to achieve online iterative optimization of the control strategy. This includes constructing an online model parameter update mechanism to adapt to the dynamic changes in UAV flight status and mission requirements. The specific process includes:

[0171] S51 provides feedback on the UAV's flight status, including: the UAV uploading its flight status (speed, position, remaining battery power) and communication status (transmission rate, latency, bit error rate) in real time; and the ground control node collecting the on-board payload resource status (load rate, available bandwidth) to form a feedback data set. ;

[0172] S52, Calculation error, including: the deviation between the calculated actual task performance and the model's predicted performance. ,like (threshold) This triggers a model update.

[0173] S53, perform parameter updates, including: updating the parameters of the graph neural network and causal inference model using Mini-Batch SGD, dynamically adjusting the learning rate using cosine annealing, and calculating the online update of model parameters based on equations (9) and (10) as follows:

[0174] (9);

[0175] (10);

[0176] in, For the set of model parameters, for Learning rate at any given moment For batch size, Let the mean squared error loss function be . , , This represents the maximum number of iterations.

[0177] S54, perform strategy updates, including recalculating the optimal resource allocation ratio based on the updated model. A new onboard payload resource management strategy is generated and sent to the onboard DTP payload controller via the satellite-to-ground link.

[0178] In this embodiment, the on-board DTP payload includes K divisible resource units, each with a bandwidth of 500Mbps, supports flexible beam reconfiguration and resource pooling scheduling, and has an on-board processor computing power of ≥10TOPS.

[0179] Example 2

[0180] like Figure 2 As shown, this embodiment provides a dynamic management and control system for on-board digital payload resources tailored to the characteristics of UAV missions, applicable to on-board DTP payloads, A "sky-ground" integrated communication system comprising one UAV node and M ground control nodes, used to implement the method of Embodiment 1, includes:

[0181] Hypergraph model building module 101 is used to build a hypergraph model of UAV mission-satellite payload to realize high-order correlation modeling between UAV mission characteristics and on-board payload resources.

[0182] The feature embedding vector generation module 102 is used to extract dynamic topological features and mission requirement features of the UAV-satellite network based on graph neural network and generate feature embedding vectors.

[0183] The core control factor determination module 103 is used to mine the causal relationship between resource allocation and task effectiveness based on the causal inference model, and to determine the core control factors.

[0184] The resource allocation optimization model construction module 104 is used to construct a resource allocation optimization model that integrates hypergraph features and causal factors based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factors, and output a dynamic control strategy for on-board digital payload resources.

[0185] The dynamic control and optimization module 105 is used to update model parameters in real time based on the flight status feedback of the UAV, so as to realize online iterative optimization of the control strategy.

[0186] Application examples: See Figures 3-8 .

[0187] (I) System Hardware Architecture

[0188] The hardware architecture of the "sky-ground" integrated communication system used in this invention consists of three parts:

[0189] (1) Onboard portion: A flexible high-throughput satellite equipped with a digital transparent processor (DTP). The DTP payload contains 16 divisible resource units, each with a bandwidth of 500Mbps, supporting flexible beam reconfiguration and resource pooling scheduling; equipped with an onboard processor (CPU: ARM Cortex-A76, GPU: Mali-G76) to run the resource management algorithm of this invention, with a computing power of 10TOPS.

[0190] (2) Aerospace component: 3 long-endurance UAVs (model: Wing Loong-2), each equipped with a Ka-band satellite communication terminal (transmission rate 0-2Gbps), GPS / BeiDou dual-mode positioning module (positioning accuracy 1m) and status sensors (speed, attitude, battery level), supporting real-time data acquisition and uploading.

[0191] (3) Ground part: The ground control center includes a server cluster (2 CPU servers: Intel Xeon Gold6330, 1 GPU server: NVIDIA A100), monitoring terminals and satellite telemetry and control stations, which are responsible for model training, data storage and control command issuance.

[0192] (er) Software Implementation Process

[0193] The software implementation of this invention is based on Python 3.8 and uses the PyTorch 1.12 deep learning framework. The specific process is as follows:

[0194] (1) Data acquisition module: The UAV terminal collects flight status and communication status data every 100ms, the on-board DTP payload uploads resource status data every 500ms, and the ground control center receives data through the Kafka message queue. The data transmission delay is ≤200ms.

[0195] (2) Hypergraph construction module: Based on the collected task and resource data, calculate the hyperedge weights according to Formula 1, construct the hypergraph model, and use the NetworkX library to realize the storage and visualization of the hypergraph.

[0196] (3) Feature extraction module: A dynamic graph neural network is implemented based on the PyTorch Geometric library, with L=3 graph convolutional layers, GRU hidden layer dimension=256, and feature embedding vector dimension=128; training batch size B=32, and initial learning rate. .

[0197] (4) Causal inference module: The dual machine learning framework is implemented using the XGBoost 1.6.1 library, with 100 decision trees and a maximum depth of 8. The average treatment effect is calculated using Formula 4, and the core control factors are selected using Formula 5.

[0198] (5) Optimization and solution module: The Lagrange multiplier method is implemented based on the optimization toolkit of the SciPy library. The hypergraph attention parameter vector dimension is 64, and the weights are dual-objective. Calculated from the mean of attention weights .

[0199] (6) Online update module: The Flask framework is used to build a RESTful API to realize the real-time update of model parameters and the distribution of control strategies; when the error \( e_t>0.05 \), a mini-batch gradient descent update is triggered with a batch size B=16.

[0200] (III) Test Scenarios and Results

[0201] Two typical scenarios, "hotspot area intelligence feedback" and "cross-domain task execution", were selected for testing to compare the performance differences between the method of this invention and traditional rule-based control methods (control group 1) and static GCN-based control methods (control group 2).

[0202] (1) Test Scenario 1: Intelligence feedback from hotspot areas (3 drones simultaneously perform intelligence collection tasks in a hotspot area with a radius of 5km, and the data transmission requirement is 1Gbps for each drone)

[0203] Test results: The average data transmission rate of the method of this invention reached 980Mbps, the latency was ≤50ms, and the task success rate was 100%; the average transmission rate of control group 1 was 620Mbps, the latency was 120ms, and the success rate was 85%; the average transmission rate of control group 2 was 810Mbps, the latency was 75ms, and the success rate was 92%.

[0204] (2) Test Scenario 2: Cross-domain mission execution (the UAV flies from a plain area (longitude 116°, latitude 39°) to a mountainous area (longitude 117°, latitude 40°), with a flight speed of 150km / h, and the mission priority is increased from level 3 to level 5)

[0205] Test results: The resource reconfiguration time of the method of the present invention is ≤300ms, and the task performance is maintained above 0.9; the resource reconfiguration time of control group 1 is 1.2s, and the task performance drops to 0.65; the resource reconfiguration time of control group 2 is 600ms, and the task performance drops to 0.78.

[0206] (3) Long-term stability test: After running continuously for 72 hours, the model parameters of the method of this invention were updated 48 times, and the average error was... The resource utilization efficiency remained stable at over 85%; the average errors of control group 1 and control group 2 were 0.12 and 0.07, respectively, and the resource utilization efficiency decreased to below 60% over time.

[0207] The application of this invention can promote the deep integration of flexible high-throughput satellites based on DTP and long-endurance UAVs, with specific industrial value reflected in:

[0208] (1) Military field: Enhance the systemic combat capability of UAV equipment, realize real-time transmission of intelligence in hotspot areas and efficient coordination of cross-domain missions, and support the air-based information support strategy;

[0209] (2) Civilian field: Provide highly reliable satellite communication support for emergency rescue, environmental monitoring and other scenarios, and solve the problem of insufficient communication coverage for drones in remote areas;

[0210] (3) Satellite industry: Promote the intelligent upgrading of on-board digital payload technology, improve the resource utilization efficiency and market competitiveness of high-throughput satellites, and promote the construction of an integrated "sky-ground" information network.

[0211] The present invention also provides a memory that stores multiple instructions for implementing the method as described in Embodiment 1.

[0212] like Figure 9 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores a plurality of instructions, which can be loaded and executed by the processor to enable the processor to perform methods as described in Embodiments 2 and 3.

[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic management and control method for on-board digital payload resources tailored to the characteristics of UAV missions, applicable to payloads including on-board DTP payloads, A unified "sky-ground" communication system comprising one UAV node and M ground control nodes, characterized in that... Includes the following steps: S1. Construct a hypergraph model of UAV mission-satellite payload to realize high-order correlation modeling between UAV mission characteristics and on-board payload resources; S2, based on graph neural networks, extracts dynamic topological features and mission requirement features of UAV-satellite networks. Generate feature embedding vectors; S3, based on the causal inference model, explore the causal relationship between resource allocation and task effectiveness, and determine the core control factors; S4. Based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factor, construct a resource allocation optimization model that integrates hypergraph features and causal factors, and output a dynamic control strategy for on-board digital payload resources. S5, based on drone flight status feedback, updates model parameters in real time to achieve online iterative optimization of control strategies.

2. The method for dynamic management and control of on-board digital payload resources based on the characteristics of UAV missions according to claim 1, characterized in that, S1 includes: S11, establish a hypergraph model to describe the many-to-many relationships between UAV missions, on-board payload resources, and ground requirements, defined as... ,in, Represents a vertex set. Represents a superedge set. Represents the weight matrix; where: The vertex set It contains three types of vertices, namely, drone mission vertices. Onboard payload resource vertex and ground demand peak ,in, The number of resolvable resource units for the onboard DTP payload; The hyperedge set In the diagram, each hyperedge corresponds to a set of "task-resource-requirement" associations, i.e. ,and It contains at least one type of vertex; The weight matrix It includes hyperedge weights and vertex weights, which represent the strength of the association and the importance of the vertex, respectively; S12, define the UAV mission feature vector as follows: This allows for the quantification of the impact of UAV mission characteristics on the hypergraph structure, where: Task priority is divided into 1-5 levels. This refers to data transmission rate requirements, measured in Mbps. The speed of the drone is expressed in m / s. The importance of the data is indicated by a range of 0-1. The deadline for the task is in seconds. S13, define the on-board payload resource feature vector as follows: ,in: This refers to resource bandwidth, measured in Mbps. Resource load factor, %. Resource scheduling delay, in milliseconds. The duration of resource availability, measured in seconds (s). S14, calculate the hyperedge weight based on the UAV mission feature vector and the on-board payload resource feature vector. The hyperedge weight is obtained by multiplying the task-resource matching degree and the demand-resource matching degree. The calculation formula is shown in the following formula (1): （1）； in, The cosine similarity function is used. This represents the feature vector of ground demand. ,in This is the normalized form of the onboard payload resource feature vector, satisfying... ; S15, determine the vertex weights.

3. The method for dynamic management and control of on-board digital payload resources based on the characteristics of UAV missions according to claim 2, characterized in that, The methods for determining vertex weights include two types: The first method involves directly linking vertex weights to UAV mission characteristics and onboard resource status, determining the vertex weights by matching the core features of the corresponding vertex. For UAV mission vertex weights, the weights are obtained based on mission feature vectors through feature normalization and importance weighting. For onboard payload resource vertex weights, the weights are calculated based on resource feature vectors combined with resource scarcity. For ground demand vertex weights, the weights are quantified by referring to ground demand feature vectors and considering indicators such as urgency and coverage. or The second method is to determine the vertex weight based on collaboration with the hyperedge weight, including: the vertex weight is determined by reverse iterative optimization through the hyperedge weight, that is, the weight of vertices that frequently participate in high-weight hyperedges is dynamically increased, and the consistency between the association strength and the importance of the vertex is ensured.

4. The method for dynamic management and control of on-board digital payload resources based on the characteristics of UAV missions according to claim 3, characterized in that, S2 includes: S21, Construct a graph neural network, wherein the graph neural network consists of a construction layer, a temporal feature fusion layer, and a feature embedding layer; S21 includes: (1) Constructing a graph construction layer, including: converting the hypergraph model into a dynamic graph sequence. ,in, For the static graph at time t, the edge weights are obtained by mapping the hyperedge weights, i.e. , For containing vertices and The superedge; (2) Constructing a temporal feature fusion layer, including: using a structure combining gated recurrent units (GRU) and graph convolutional networks (GCN) to update the graph features at each time step; wherein, the gated recurrent units (GRU) are used to fuse temporal features, and the graph convolutional networks (GCN) are used to extract spatial features. The structure combining gated recurrent units (GRU) and graph convolutional networks (GCN) is used to update the graph features at each time step in the following manner: (2) Dynamic graph convolutional feature update method. （2）； in, As vertex exist Time of the first The feature vector of the layer, The spatial convolution weight matrix, For bias terms, It is the ReLU activation function; As vertex The neighbor set; The updated graph features at each time step are then used for temporal feature fusion and updating based on equations (3-1) to (3-4), where equation (3) is: （3-1） （3-2）； （3-3）； （3-4）； in, To update the gate output value, To reset the gate output value, As vertex exist The final feature embedding vector at time step 1. For element-wise product, The number of convolutional layers in the graph; Use the Sigmoid activation function; To update the weight parameter matrix of the gate; This is a vector concatenation operator; for Peak of Time In the graph convolutional network The output feature vector of the layer; for Peak of Time In the graph convolutional network The output feature vector of the layer; To update the bias parameter vector of the gate; The weight parameter matrix for resetting the gate; This is the offset parameter vector for resetting the gate; It is the hyperbolic tangent activation function; The weight parameter matrix for the candidate features; The bias parameter vector for candidate features; (3) Constructing a feature embedding layer, including: concatenating the feature embedding vectors of all vertices to obtain a global feature matrix. ;in, This is the global feature matrix, containing the feature embedding information of all vertices; This represents the total number of vertices in the hypergraph model. For the first The vertex is in Feature embedding vector at time step; superscript This represents the transpose operation of a matrix, ensuring that the rows of the feature matrix correspond to vertices and the columns correspond to feature dimensions, thus adapting to the input requirements of subsequent causal inference and resource optimization modules. This global feature matrix fully integrates... The spatiotemporal correlation characteristics, mission requirement characteristics, and resource status characteristics of the real-time drone-satellite network are the core feature inputs for realizing intelligent control and management decisions, and are used for subsequent causal inference and resource optimization. S22, Based on the graph neural network, extract the dynamic topology features and mission requirement features of the UAV-satellite network, and generate feature embedding vectors.

5. The method for dynamic management and control of on-board digital payload resources based on the characteristics of UAV missions according to claim 4, characterized in that, S3, based on a causal inference model, mines the causal relationship between resource allocation and task efficiency to determine core control factors, including: employing a dual machine learning framework to mine the causal relationship between resource allocation and task efficiency to determine core control factors, specifically including: S31, Define variables, including: Set the processing variable The allocation ratio of onboard payload resources, with a value range of 0-1, is the result variable. For task performance, a normalized metric is defined, encompassing overall data transmission rate, latency, and success rate, with values ​​ranging from 0 to 1. Covariates... The feature matrix output in step S2 It includes UAV mission characteristics, resource status characteristics, and network topology characteristics, covariates. Dimensions and global feature matrix Consistency is the core input data for causal inference models to capture the correlation between variables; S32, construct three sub-models, namely: Result prediction model: ; Processing variable prediction model: ; Residual estimation model ; in, The causal effect is the average treatment effect (ATE). and This is the error term; Indicated based on covariates Processing variables The prediction function is represented by the global feature matrix. The processing variables are obtained by fitting a machine learning model as input. The predicted value; S33, Calculating the causal effect model Corresponding causal effect This includes: a prediction model based on the training results of the gradient boosting tree XGBoost and a prediction model for the treatment variables, and the causal effect is obtained through residual estimation and average treatment effect ATE, wherein the calculation formula of the causal effect is shown in the following formula (4): （4）； in, and For the trained prediction model, The number of samples; S34, Identify core control factors, including: calculating covariates. The impact of various characteristics on causal effects Influence weights, defining feature importance indicators As shown in equation (5): （5）； in, For covariate dimensions, For the first Select one covariate. The characteristics of the UAV mission are used as core control factors. Typical factors include UAV mission priority, data transmission rate requirements, on-board resource load rate, and network topology connectivity.

6. The method for dynamic management and control of on-board digital payload resources based on the characteristics of UAV missions according to claim 5, characterized in that, S4, based on the UAV mission-satellite payload hypergraph model, the feature embedding vector, and the core control factor, constructs a resource allocation optimization model that integrates hypergraph features and causal factors, outputting a dynamic control strategy for on-board digital payload resources. This strategy includes: using the core control factor as a constraint, and with "maximizing resource utilization efficiency" and "maximizing mission effectiveness" as dual objectives, constructing a resource allocation optimization model; employing a hypergraph attention mechanism to fuse multi-objective weights; and solving for the optimal resource allocation strategy using the Lagrange multiplier method. Specifically, this includes: S41, Define an objective function, which includes a primary objective function and a secondary objective function. The primary objective function is used to maximize task performance, i.e. The secondary objective function is used to maximize resource utilization efficiency, i.e. ; in, For the first The weight of each drone mission is determined by the mission priority. For the first The effectiveness of each task To be assigned to the Resource bandwidth for each task; S42, the constraints are determined as follows: Total resource constraints: The sum of the resource allocation ratios for all tasks shall not exceed 1; Task requirements constraints: Assigned to the The bandwidth of each task meets its rate requirements. Delay constraints: Resource scheduling delays shall not exceed the task deadline requirements; Causal constraint: , No. Causal effects of individual tasks Not lower than the minimum threshold This ensures the effectiveness of resource allocation; S43, Perform hypergraph attention weight fusion, including: to balance the weights of the two objectives, calculate the attention weights of the task and resources based on the hypergraph model. The formula is shown in equation (6) below, which is the formula for calculating the hypergraph attention weight: （6）； in, For attention parameter vectors, This is a vector concatenation operation. For activation functions; S44, Solve the model, including: transforming the bi-objective optimization problem into a single-objective problem, constructing a weighted objective function based on formula (7) and formula (8), and using the Lagrange multiplier method to solve the optimal solution of the model; （7）； （8）； in, The value of the Lagrange function. To weight the overall task effectiveness, The secondary objective weights are determined by the average of the hypergraph attention weights. For resource utilization efficiency, among which, ;, and It is a Lagrange multiplier. For the first The optimal resource allocation ratio for each drone mission; For the first The rate requirement of each task This represents the total bandwidth of the onboard payload. This is the extremum condition for the Lagrange function; The partial derivative of task effectiveness with respect to the resource allocation ratio; To be assigned to the Bandwidth for each task.

7. The method for dynamic management and control of on-board digital payload resources based on the characteristics of UAV missions according to claim 6, characterized in that, S5, based on UAV flight status feedback, updates model parameters in real time to achieve online iterative optimization of control strategies. This includes constructing an online model parameter update mechanism to adapt to dynamic changes in UAV flight status and mission requirements. The specific process includes: S51 provides feedback on the flight status of unmanned aerial vehicles (UAVs), including: real-time uploading of flight and communication status by the UAVs, and collection of onboard payload resource status by ground control nodes, forming a feedback data set. ; S52, Calculation error, including: the deviation between the calculated actual task performance and the model's predicted performance. ,like This triggers a model update; S53, perform parameter updates, including: updating the parameters of the graph neural network and the causal inference model using mini-batch gradient descent, dynamically adjusting the learning rate using a cosine annealing strategy, and calculating the online update of the model parameters based on equations (9) and (10) as follows: （9）； (10); in, For the set of model parameters, for Learning rate at any given moment For batch size, Let the mean squared error loss function be . , , This represents the maximum number of iterations. S54, perform strategy updates, including recalculating the optimal resource allocation ratio based on the updated model. A new onboard payload resource management strategy is generated and sent to the onboard DTP payload controller via the satellite-to-ground link.

8. A dynamic management and control system for on-board digital payload resources tailored to the characteristics of UAV missions, applicable to on-board DTP payloads, A "sky-ground" integrated communication system comprising one UAV node and M ground control nodes, used to implement the method described in any one of claims 1-7, comprising: The hypergraph model building module (101) is used to build a hypergraph model of UAV mission-satellite payload to realize high-order correlation modeling between UAV mission characteristics and on-board payload resources. The feature embedding vector generation module (102) is used to extract dynamic topological features and mission requirement features of the UAV-satellite network based on graph neural networks. Generate feature embedding vectors; The core control factor determination module (103) is used to mine the causal relationship between resource allocation and task effectiveness based on the causal inference model and determine the core control factors. The resource allocation optimization model construction module (104) is used to construct a resource allocation optimization model that integrates hypergraph features and causal factors based on the UAV mission-satellite payload hypergraph model, the feature embedding vector and the core control factor, and output the dynamic control strategy of on-board digital payload resources; The dynamic control and optimization module (105) is used to update model parameters in real time based on UAV flight status feedback, so as to realize online iterative optimization of control strategies.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing multiple instructions, and the processor being used to read the instructions and execute the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions, which can be read by a processor and executed as described in any one of claims 1-7.

Images