A virtual resource pool construction method and system based on dynamic satellite clustering, a terminal device, and a storage medium

Abstract

The application discloses a kind of virtual resource pool construction method, system, terminal equipment and storage medium based on dynamic satellite clustering, and is related to satellite network technical field.The method comprises: obtaining the topological connection relationship and resource state of satellite node in time-varying satellite network, and constructing the network topology graph representing the time-varying satellite network;The embedding representation of each satellite node is obtained by a graph neural network model, and the satellite node is divided into multiple star clusters;The evaluation index of each satellite node in the star cluster is calculated and the cluster head node is selected;With the star cluster as the resource management unit, the heterogeneous resources of each satellite node in the star cluster are abstracted and aggregated, and a star cluster level distributed virtual resource pool is constructed.The application solves the problem of single-satellite resource island, realizes efficient pooling and sharing of multi-satellite resources, reduces the management complexity and control overhead of large-scale satellite constellation, and improves system resource utilization, management efficiency and service guarantee capability.

Description

Technical Field

[0001] This invention relates to the field of satellite network technology, and in particular to a method, system, terminal device and storage medium for constructing a virtual resource pool based on dynamic satellite clustering. Background Technology

[0002] In recent years, the construction of global low-Earth orbit satellite constellations has entered a new stage of intensive networking and large-scale development. Currently, satellite network deployment generally adopts a closed model of one satellite for one purpose. With the rapid increase in constellation size, problems such as resource silos, service bloat, and response delays are likely to occur, making it difficult to meet the diverse business needs of sensing, navigation, and intelligence communication.

[0003] Existing methods for constructing virtual resource pools are primarily geared towards relatively stable terrestrial network environments, typically employing a centralized architecture for static pooling and management of resources across multiple nodes. However, these methods are ill-suited for low-Earth orbit (LEO) satellite networks with highly dynamic topologies and frequently changing resource demands. On one hand, constructing a globally centralized unified resource pool within a large-scale LEO satellite constellation incurs extremely high maintenance costs and management complexity. On the other hand, factors such as dynamic topology evolution, intermittent availability of inter-satellite links, and bandwidth and latency constraints make it difficult to achieve timely synchronization of resource awareness, thereby weakening the accuracy and effectiveness of resource scheduling decisions. Furthermore, the control overhead under a centralized architecture grows exponentially with constellation size, severely restricting the system's real-time response capabilities and robustness.

[0004] Therefore, there is an urgent need for a method to construct a cluster-level distributed virtual resource pool driven by dynamic satellite clustering and graph neural networks to fill the gap in existing technologies. Summary of the Invention

[0005] The technical problem this invention aims to solve is that, in the field of satellite network technology, the centralized management of large-scale low-Earth orbit (LEO) satellite constellations leads to an explosive increase in control overhead with the constellation size. Furthermore, the closed, one-satellite-per-use model results in isolated resources for individual satellites and makes resource sharing among multiple satellites difficult. Simultaneously, the highly dynamic nature of LEO satellite network topology makes real-time perception and synchronization of global resource status difficult, severely impacting the accuracy and timeliness of resource scheduling decisions. Therefore, an effective solution is urgently needed to address these technical problems.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for constructing a virtual resource pool based on dynamic satellite clustering, the method comprising: Obtain the topological connections and resource status of satellite nodes in the time-varying satellite network, and construct a network topology map representing the time-varying satellite network; Based on the network topology graph, the topological features and resource status features of each satellite node are aggregated using a graph neural network model to obtain the embedded representation of each satellite node, and the satellite nodes are divided into multiple star clusters based on the embedded representation. Calculate the evaluation index of satellite nodes within each cluster, and select the cluster head node from each cluster based on the evaluation index; Using the star cluster as a resource management unit, the heterogeneous resources of each satellite node within the star cluster are abstracted and aggregated to construct a star cluster-level distributed virtual resource pool.

[0007] In one implementation, acquiring the topological connectivity and resource status of satellite nodes in the time-varying satellite network, and constructing a network topology map representing the time-varying satellite network, includes: The observation period corresponding to the time-varying satellite network is divided into multiple time slots; For each time slot, an undirected network topology graph corresponding to the time slot is constructed based on the inter-satellite link connection relationship between satellite nodes within the time slot; For each satellite node, the available computing resources and available storage resources of the satellite node are obtained through local status monitoring; By broadcasting and interacting with neighboring nodes to synchronize information, the resource status information of each satellite node in each time slot and the available bandwidth resources of each inter-satellite link in each time slot are obtained. The resource status information includes the available computing resources and available storage resources of each satellite node. The resource status information and the available bandwidth resources are used as attribute information of satellite nodes and inter-satellite links in the network topology diagram of the corresponding time slot to obtain the attribute network topology diagram of each time slot. The attribute network topology graphs of each time slot are integrated to obtain the network topology graph, which is a static topology snapshot sequence composed of the attribute network topology graphs of each time slot.

[0008] In one implementation, based on the network topology graph, the topological features and resource status features of each satellite node are aggregated using a graph neural network model to obtain an embedded representation of each satellite node, and the satellite nodes are divided into multiple clusters based on the embedded representation, including: Based on the topological connection relationship and resource status of each satellite node in the network topology diagram, multi-dimensional attribute features of each satellite node are extracted to form a node feature matrix. The multi-dimensional attribute features include satellite location, available computing resources, available storage resources, available bandwidth resources, and node connectivity. Construct an adjacency matrix based on the connection relationships between satellite nodes; The node feature matrix and the adjacency matrix are input into the graph neural network model, and the features of each satellite node are aggregated through the graph neural network model to obtain the embedded representation of each satellite node; Based on the embedded representation of each satellite node, the satellite nodes are clustered.

[0009] In one implementation, the step of inputting the node feature matrix and the adjacency matrix into the graph neural network model, and aggregating the features of each satellite node through the graph neural network model to obtain the embedded representation of each satellite node, includes: Input the node feature matrix and the adjacency matrix into the graph convolutional network; In each layer of the graph convolutional network, the set of neighboring nodes of each node is determined based on the adjacency matrix; For each node, the node's own features are weighted and aggregated with the features of the node's neighboring nodes. After linear transformation through a learnable weight matrix, the node features are processed by a non-linear activation function to obtain an updated node feature representation. Through the layer-by-layer aggregation and propagation of multi-layer graph convolutional networks, node features are gradually integrated with the topological structure information and resource status information of higher-order neighborhoods to obtain the embedded representation of each satellite node.

[0010] In one implementation, the graph convolutional network adopts a two-layer structure, wherein the nonlinear activation function of the first layer is a first nonlinear activation function, and the second layer adopts a classification output function to output the category probability of each satellite node; During the training phase of the graph convolutional network, a classification loss function is constructed based on the label information of satellite nodes with known cluster labels and the predicted class probabilities output by the graph convolutional network. With the goal of minimizing the error between the predicted class probability and the true label, the learnable weight matrix of the graph convolutional network is iteratively optimized.

[0011] In one implementation, the step of calculating the evaluation index of satellite nodes within each cluster and selecting cluster head nodes from each cluster based on the evaluation index includes: For each satellite node, the average node connectivity of the satellite node is obtained based on the network topology diagram; Based on the resource status of the satellite nodes in the network topology diagram, the on-board processing capacity of the satellite nodes is determined. Based on the mission attributes, payload capacity, or orbital position of the satellite node, determine the cluster head priority coefficient of the satellite node; The average connectivity of the nodes, the on-board processing capacity, and the cluster head priority coefficient are weighted and summed to obtain the comprehensive weight value of the satellite nodes as the evaluation index, wherein the sum of the weight factor of the average connectivity of the nodes and the weight factor of the on-board processing capacity is one. The satellite node with the largest comprehensive weight in each star cluster is determined as the cluster head node of that star cluster.

[0012] In one implementation, the step of using the star cluster as a resource management unit to abstract and aggregate the heterogeneous resources of each satellite node within the star cluster to construct a star cluster-level distributed virtual resource pool includes: Obtain the available computing and storage resources of each satellite node, and construct a cluster resource status description to record the resource status of all nodes in the cluster; Virtualization is used to abstract and encapsulate the computing resources of each satellite node within the star cluster, forming a virtual computing resource pool. Virtualization is used to unify and aggregate the storage resources of each satellite node within the cluster, forming a virtual storage pool. By virtualization, the inter-satellite links between satellite nodes within the cluster are abstracted to construct logical virtual links and bandwidth resource pools. By integrating the virtual computing resource pool, virtual storage pool, logical virtual link and bandwidth resource pool, a cluster-level distributed virtual resource pool is obtained, wherein the cluster head node of the cluster performs the intra-cluster autonomous management and inter-cluster collaborative management of the cluster-level distributed virtual resource pool.

[0013] Secondly, embodiments of the present invention also provide a virtual resource pool construction system based on dynamic satellite clustering, the system comprising: The network topology graph construction module is used to obtain the topological connection relationship and resource status of satellite nodes in the time-varying satellite network, and construct a network topology graph representing the time-varying satellite network. The cluster partitioning module is used to aggregate the topological features and resource status features of each satellite node based on the network topology graph using a graph neural network model to obtain the embedded representation of each satellite node, and to partition the satellite node into multiple clusters based on the embedded representation; The cluster head node selection module is used to calculate the evaluation index of satellite nodes in each cluster and select the cluster head node from each cluster based on the evaluation index. The virtual resource pool construction module is used to abstract and aggregate the heterogeneous resources of each satellite node within the star cluster, using the star cluster as the resource management unit, to construct a star cluster-level distributed virtual resource pool.

[0014] Thirdly, embodiments of the present invention also provide a terminal device, the terminal device including a memory, a processor, and a virtual resource pool construction program based on dynamic satellite clustering stored in the memory and executable on the processor. When the processor executes the virtual resource pool construction program based on dynamic satellite clustering, it implements the steps of the virtual resource pool construction method based on dynamic satellite clustering as described in any of the above schemes.

[0015] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a virtual resource pool construction program based on dynamic satellite clustering. When the virtual resource pool construction program based on dynamic satellite clustering is executed by a processor, it implements the steps of the virtual resource pool construction method based on dynamic satellite clustering as described in any of the above schemes.

[0016] Beneficial Effects: This invention discloses a method, system, terminal device, and storage medium for constructing a virtual resource pool based on dynamic satellite clustering, relating to the field of satellite network technology. The method first obtains the topological connections and resource states of satellite nodes in a time-varying satellite network, constructing a network topology graph representing the time-varying satellite network. Then, based on the network topology graph, a graph neural network model is used to aggregate the topological features and resource state features of each satellite node to obtain an embedded representation of each satellite node, and the satellite nodes are divided into multiple clusters based on the embedded representation. Subsequently, evaluation indicators for satellite nodes within each cluster are calculated, and cluster head nodes are selected from each cluster based on the evaluation indicators. Finally, using the cluster as a resource management unit, the heterogeneous resources of each satellite node within the cluster are abstracted and aggregated to construct a cluster-level distributed virtual resource pool.

[0017] This invention utilizes a graph neural network-driven dynamic clustering and multi-attribute weighted cluster leader election mechanism to divide a large-scale low-Earth orbit satellite constellation into several locally autonomous clusters. It then uses these clusters as units to perform unified abstraction and logical aggregation of heterogeneous resources from multiple satellites, constructing a distributed virtual resource pool. This method breaks through the limitations of traditional isolated single-satellite resources, achieving efficient pooling and on-demand scheduling of multi-satellite resources within a cluster. It effectively reduces the control overhead and maintenance complexity of centralized global management of large-scale constellations. Simultaneously, it leverages the topological proximity advantage within clusters to reduce communication overhead for multi-satellite collaboration, improving system resource utilization, scalability, and adaptability to dynamic topological environments. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a specific implementation of the virtual resource pool construction method based on dynamic satellite clustering provided in this embodiment of the invention.

[0019] Figure 2 This is a snapshot model diagram of the time-varying satellite network topology provided in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram illustrating the principle of the graph convolutional network algorithm provided in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram of a dynamic satellite clustering method based on graph convolutional networks provided in an embodiment of the present invention.

[0022] Figure 5 This is a diagram of the star cluster-level distributed virtual resource pool architecture provided in an embodiment of the present invention.

[0023] Figure 6 This is a schematic diagram of the virtual resource pool construction device based on dynamic satellite clustering provided in an embodiment of the present invention.

[0024] Figure 7 This is a block diagram illustrating the internal structure of the terminal device provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0026] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0027] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0028] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, "first control information" and "second control information" are only used to distinguish different control information and do not limit their order.

[0029] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0030] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0031] In recent years, the construction of global low-Earth orbit satellite constellations has entered a new stage of intensive networking and large-scale development. Currently, satellite network deployment generally adopts a vertical siloed approach, presenting a closed "one satellite, one use" model. With the rapid increase in constellation size, problems such as resource silos, service bloat, and response delays are easily encountered, making it difficult to meet the diverse service demands of high-concurrency sensing, navigation, and intelligent communication. By breaking down the physical boundaries of individual satellites and unifying and logically aggregating the heterogeneous resources of multiple satellites to construct a virtual resource pool, efficient pooling and on-demand scheduling of multi-satellite resources can be achieved, alleviating local satellite resource overload bottlenecks and improving the overall system resource utilization and service assurance capabilities.

[0032] Existing methods for constructing virtual resource pools are primarily geared towards relatively stable terrestrial network environments, typically employing a centralized architecture for static pooling and management of resources across multiple nodes. However, these methods are ill-suited for low-Earth orbit (LEO) satellite networks with highly dynamic topologies and frequently changing resource demands. Firstly, constructing a globally centralized unified resource pool within a large-scale LEO satellite constellation incurs extremely high maintenance costs and management complexity. Secondly, factors such as dynamic topology evolution, intermittent availability of inter-satellite links, and bandwidth / latency constraints make it difficult to achieve timely synchronization of resource awareness, thereby weakening the accuracy and effectiveness of resource scheduling decisions. Furthermore, the control overhead under a centralized architecture grows exponentially with constellation size, severely restricting the system's real-time response capabilities and robustness.

[0033] This embodiment provides a method for constructing a virtual resource pool based on dynamic satellite clustering, such as... Figure 1 As shown, the specific steps include the following: Step S100: Obtain the topological connection relationship and resource status of satellite nodes in the time-varying satellite network, and construct a network topology diagram representing the time-varying satellite network.

[0034] In this embodiment, a time-varying satellite network refers to a communication network composed of a large number of low-Earth orbit (LEO) satellites, where the connections between satellite nodes continuously and dynamically change over time. Generally, LEO satellites orbit the Earth at a speed of approximately 7 to 8 km / s. Inter-satellite links are intermittently connected or disconnected due to changes in the relative positions of the satellites, causing the topology of the entire network to change significantly on a timescale of seconds. A satellite node refers to a single LEO satellite in the constellation that possesses onboard computing, storage, and communication capabilities, undertaking the communication functions of data relay and forwarding, and having a certain amount of on-orbit processing and caching capabilities.

[0035] Satellite topology describes whether any two satellites are directly connected via inter-satellite links at the current moment, determining the reachability path and hop count of information within the satellite network. Resource status refers to the amount of available computing resources, available storage capacity, and remaining available bandwidth resources on each inter-satellite link that each satellite node can provide at the current moment. These resource quantities fluctuate dynamically due to the combined influence of the satellite's own payload capacity, on-orbit energy consumption constraints, and real-time service load.

[0036] A network topology graph is a mathematical object that formally represents a satellite network using a graph data structure. The vertices of the graph correspond to individual satellite nodes, the edges correspond to inter-satellite links, and the attributes of the vertices and edges carry resource status information. Since the topology and resource status of real low-Earth orbit satellite networks evolve continuously over time, it is impossible to fully depict all their dynamic states in a static graph model. Therefore, by performing time-slotted discrete modeling on the continuously changing satellite network, it is transformed into a series of time-series arranged static network topology graphs carrying resource attributes. This decouples the highly dynamic and coupled spatiotemporal problem into a structured data processing task that can be analyzed independently graph by graph. Through this processing, the satellite network is transformed from a set of physically high-speed moving and logically constantly reorganizing communication nodes into a structured input that can be directly read and learned by deep models such as graph neural networks.

[0037] In one implementation, the step of acquiring the topological connectivity and resource status of satellite nodes in the time-varying satellite network, and constructing a network topology map representing the time-varying satellite network, specifically includes the following steps: Step S110: Divide the observation period corresponding to the time-varying satellite network into multiple time slots; Step S120: For each time slot, construct the undirected network topology graph corresponding to the time slot based on the inter-satellite link connection relationship between satellite nodes within the time slot; Step S130: For each satellite node, obtain the available computing resources and available storage resources of the satellite node through local status monitoring; Step S140: Synchronize information with neighboring nodes through broadcast interaction to obtain the resource status information of each satellite node in each time slot and the available bandwidth resources of each inter-satellite link in each time slot. The resource status information includes the available computing resources and available storage resources of each satellite node. Step S150: Use the resource status information and the available bandwidth resources as attribute information of satellite nodes and inter-satellite links in the network topology diagram of the corresponding time slot to obtain the attribute network topology diagram of each time slot. Step S160: Integrate the attribute network topology diagrams of each time slot to obtain the network topology diagram, which is a static topology snapshot sequence composed of the attribute network topology diagrams of each time slot.

[0038] In this embodiment, since the relative positions of satellites within the constellation change continuously on a second-level timescale, the observation period is determined based on the orbital period of the low-Earth orbit satellites. This observation period is then divided into multiple equal-length time slots, each with a sufficiently small length to ensure that changes in inter-satellite link connectivity caused by high-speed satellite motion within that time slot are negligible. This allows the network topology within each time slot to be approximated as constant. For each time slot, a corresponding undirected network topology graph is constructed based on the actual inter-satellite links existing within that time slot. The undirected graphs corresponding to each time slot are arranged in chronological order, forming a static topology snapshot sequence. This transforms the dynamic evolution in the time dimension into a serialized graph processing problem in the spatial dimension, such as... Figure 2 As shown.

[0039] Within each time slot, satellite nodes obtain their own available computing and storage resources in real time through local status monitoring. Simultaneously, each satellite node periodically broadcasts Hello packets to interact with neighboring nodes. These Hello packets carry the node's identification information and a summary of its current resource status. Neighboring nodes receive this information and update their locally maintained local network resource views accordingly. Through a distributed information synchronization mechanism, each satellite node obtains its own resource status, as well as the resource status information of its neighboring nodes and the available bandwidth resources of each inter-satellite link. The resource status information specifically includes the available computing and storage resources of each satellite node, while available bandwidth resources refer to the remaining transmission capacity that can be allocated to the inter-satellite links at the current moment.

[0040] Based on this, the obtained resource status information and available bandwidth resources are injected as attribute information for satellite nodes and inter-satellite links in the corresponding time-slot network topology graph, transforming the original topology containing only connectivity relationships into an attribute network topology graph carrying multi-dimensional resource annotations. Finally, the attribute network topology graphs of each time slot are integrated in chronological order to form a complete network topology graph, which is essentially a sequence of static topology snapshots with resource attributes arranged in chronological order. This modeling approach allows each time slot's graph to be processed independently while preserving the temporal continuity of topological evolution at the sequence level, providing a complete input with both structural information and resource semantics for graph neural network models.

[0041] Specifically, considering the high-speed mobility of low-orbit satellites, the total observation time will be... Divided into There are 1 equal-length time slots, each time slot having a length of 1. , The value is chosen to be sufficiently small to ensure that the satellite network topology remains essentially unchanged within each time slot. Therefore, the satellite network topology within each time slot can be modeled as an undirected graph. , ,in Represents a set of satellite nodes. This represents the set of inter-satellite links. The evolution of the entire dynamic satellite network is represented as a sequence of static topology snapshots. This transforms dynamic problems in the time dimension into serialized static graph processing.

[0042] Within each time slot, each satellite node not only performs communication functions but also possesses certain onboard computing and storage capabilities, while inter-node links handle data transmission. Limited by factors such as the physical size and payload capacity of the satellite platform, the computing power, storage capacity, and link bandwidth resources available to each satellite node are all very limited, and the amount of available resources changes dynamically over time. Through local status monitoring, satellite nodes can obtain their own resource status information in real time and interact with neighboring nodes by broadcasting Hello packets. Supported by the inter-satellite synchronization mechanism, this enables dynamic awareness of the local network resource status. (For the satellite network topology diagram...) Define arbitrary satellite nodes In the time slot Available computing resources are Available storage capacity is Link between any two satellite nodes In the time slot Available bandwidth resources are defined as .

[0043] The aforementioned time-varying graphs and node / link attribute representations provide a complete set of input features for graph convolutional networks, thereby supporting dynamic satellite clustering and virtual resource pool construction based on topology and resource status. This transforms complex, dynamic, and coupled satellite networks into analyzable and learnable structured mathematical objects, solving the problem of the difficulty in formally representing large-scale low-Earth orbit satellite networks due to rapid topological changes.

[0044] Step S200: Based on the network topology graph, the topological features and resource status features of each satellite node are aggregated using a graph neural network model to obtain the embedded representation of each satellite node, and the satellite nodes are divided into multiple star clusters based on the embedded representation.

[0045] In this embodiment, after obtaining the network topology map representing the time-varying satellite network, the hundreds or thousands of satellites in the large-scale constellation are reasonably divided into several clusters. The satellites within each cluster are topologically closely connected, complementary and cooperative in terms of resources, and the coupling between clusters is low.

[0046] Graph neural network models are a type of deep neural network specifically designed for processing graph-structured data. They jointly learn from both the intrinsic attributes of nodes and the connectivity relationships between them, making them suitable for structured analysis of satellite nodes in network topology graphs. Topological features refer to the features extracted from the connectivity patterns of satellite nodes in the network, including which neighbors a node is connected to, whether it is located at the center or edge of the network, and the degree of connectivity with its neighbors. These features reflect the structural role of a node in the network topology. Resource status features are extracted from the node's available computing resources, available storage resources, and available bandwidth resources of connected links, reflecting the node's current load level and service capacity.

[0047] Embedded representation refers to the dense vector obtained by a graph neural network model through multiple nonlinear transformations, mapping the original high-dimensional and heterogeneous topological features and resource status features of each satellite node to a low-dimensional continuous vector space. This vector not only lies geometrically in a space where semantically similar nodes are close to each other, but also numerically encodes both the topological location information and resource status information of the nodes. Based on the embedded representation of each satellite node, clustering algorithms can group satellites that are close in distance within the embedded space into the same cluster. A cluster is a logical grouping of several satellite nodes and serves as the basic management unit for constructing a resource pool.

[0048] By using a graph neural network model to simultaneously aggregate both topological and resource features to learn embedded representations, compared to the traditional method of clustering based solely on geographical proximity, the resulting clusters can maintain low latency in intra-cluster communication while ensuring a balanced distribution of resource load. This avoids concentrating a large number of high-load nodes in the same cluster, which can cause local congestion, thereby improving the organizational rationality and resource scheduling efficiency within the cluster.

[0049] In one implementation, based on the network topology graph, the topological features and resource status features of each satellite node are aggregated using a graph neural network model to obtain an embedded representation of each satellite node, and the satellite nodes are divided into multiple clusters based on the embedded representation. Specifically, this includes the following steps: Step S210: Based on the topological connection relationship and resource status of each satellite node in the network topology diagram, extract the multi-dimensional attribute features of each satellite node to form a node feature matrix, wherein the multi-dimensional attribute features include satellite location, available computing resources, available storage resources, available bandwidth resources and node connectivity. Step S220: Construct an adjacency matrix based on the connection relationships between satellite nodes; Step S230: Input the node feature matrix and the adjacency matrix into the graph neural network model, and aggregate the features of each satellite node through the graph neural network model to obtain the embedded representation of each satellite node; Step S240: Based on the embedded representation of each satellite node, the satellite nodes are clustered.

[0050] In this embodiment, as Figure 4 As shown, dynamic satellite clustering based on a network topology graph first requires constructing the input data for a graph neural network model. Specifically, starting from the topological connections and resource status of each satellite node in the network topology graph, multi-dimensional attribute features of each satellite node are extracted to form a node feature matrix. Each row of the node feature matrix corresponds to a satellite node, and each column corresponds to a feature dimension. The multi-dimensional attribute features specifically include satellite position. (Location information), available computing resources Available storage resources (Available storage capacity), available bandwidth resources The model comprises five dimensions: inter-satellite link bandwidth, node connectivity, and more. Satellite location features characterize the spatial proximity of nodes, allowing the model to prioritize grouping nearby satellites into the same cluster, thus ensuring shorter intra-cluster link distances and lower propagation latency. Available computing and storage resources describe the resource occupancy status and service load levels of nodes, ensuring load balancing and preventing the clustering of a large number of high-load nodes into the same cluster, which could cause local congestion. Available bandwidth reflects the transmission capacity margin of the links connected to a node. Node connectivity reflects the number of currently connectable neighbors for a satellite node, measuring network connectivity and topology centrality. This helps the model identify core and edge nodes in the network, promoting the formation of highly cohesive clusters of tightly connected satellites, improving intra-cluster reliability, and reducing cross-cluster communication overhead.

[0051] Simultaneously, an adjacency matrix is ​​constructed based on the connection relationships between satellite nodes. This adjacency matrix is ​​a square matrix with the number of rows and columns equal to the total number of satellite nodes. Elements with a value of 1 represent a direct inter-satellite link between the corresponding two satellites, while a value of 0 indicates no direct connection. The node feature matrix and the adjacency matrix together constitute the input to the graph neural network model.

[0052] After inputting the node feature matrix and adjacency matrix into the graph neural network model, the model aggregates the features of each satellite node and outputs the embedded representation of each satellite node. During the aggregation process, at each layer, the model determines the set of neighboring nodes for each node based on the adjacency matrix, and performs weighted aggregation of each node's own features with the features of its neighboring nodes, thereby gradually fusing the topological structure information and resource status information of the neighborhood. After multiple layers of aggregation and propagation, each node obtains a high-dimensional embedded representation that combines topological correlation and resource similarity.

[0053] Finally, based on the embedded representation of each satellite node, the satellite nodes are clustered. Nodes that are close to each other in the embedding space are assigned to the same cluster, and the resulting clusters are topologically cohesive and resource-complementary.

[0054] It should be noted that the implementation of clustering models is not limited to graph convolutional networks. Other feasible implementations include replacing the graph neural network model with a graph attention network, which adaptively assigns differentiated aggregation weights to different neighbors through an attention mechanism, thereby capturing the topological dependencies between nodes more precisely; alternatively, a distributed weighted clustering algorithm based on region partitioning can be used, employing prior spatial information such as orbital plane and phase for coarse-grained grouping followed by fine-grained adjustments based on resource status; or using classic clustering methods such as K-means to cluster manually constructed node feature vectors, in which case topological structure information needs to be explicitly injected through feature engineering. Furthermore, in addition to the five dimensions mentioned above, the extracted features for satellite nodes can be further extended with features such as persistent link visibility, orbit prediction information, and task type matching degree to enhance the comprehensive perception capabilities of clustering decisions and resource scheduling.

[0055] The clustering method that combines topological features and resource status features effectively improves the topological compactness and resource correlation within a star cluster, resulting in shorter intra-cluster link distances and lower communication latency and interaction overhead.

[0056] In one implementation, the step of inputting the node feature matrix and the adjacency matrix into the graph neural network model, and aggregating the features of each satellite node through the graph neural network model to obtain the embedded representation of each satellite node, specifically includes the following steps: Step S231: Input the node feature matrix and the adjacency matrix into the graph convolutional network; Step S232: In each layer of the graph convolutional network, determine the set of neighboring nodes of each node based on the adjacency matrix; Step S233: For each node, the node's own features are weighted and aggregated with the features of the node's neighboring nodes, and then processed by a non-linear activation function after linear transformation through a learnable weight matrix to obtain the updated node feature representation. Step S234: Through the layer-by-layer aggregation and propagation of the multi-layer graph convolutional network, the node features are gradually integrated with the topological structure information and resource status information of the higher-order neighborhood to obtain the embedded representation of each satellite node.

[0057] In this embodiment, the graph convolutional network is a specific implementation of the graph neural network model, extending the convolution operation of traditional convolutional neural networks on Euclidean space data to graph structure data in non-Euclidean space. For example... Figure 3As shown, graph convolutional networks update the feature representation of each node by aggregating the information of each node's neighbor nodes layer by layer, so that information propagates along the edges of the graph structure.

[0058] In the implementation, the node feature matrix and adjacency matrix are input into the graph convolutional network. In each graph convolutional layer, the set of neighboring nodes for each node is first determined based on the adjacency matrix. Then, for each node, its own features are weighted and aggregated with the features of its neighboring nodes. The aggregation weights are determined by the graph's topology, and a symmetric normalization method is used to scale the contributions of each neighbor, ensuring that nodes with large degree differences are comparable during feature aggregation. The aggregated features undergo a linear transformation through a learnable weight matrix, mapping the features from the feature space of the current layer to the feature space of the next layer. The result of the linear transformation is processed by a non-linear activation function, introducing non-linear expressive power to obtain the updated node feature representation for that layer.

[0059] Specifically, for the first Layer Node Feature Matrix , No. Layer Node Feature Matrix The following propagation rules apply:

[0060] in, It is a symmetric normalized matrix. For adjacency matrices with self-loops, Represents the adjacency matrix. Represents the identity matrix. This is the corresponding degree matrix. It is the first The node feature matrix of the layer, For the first The learnable weight matrix of each layer is updated via backpropagation during model training. Represents a linear feature transformation. It is a non-linear activation function. The initial node feature matrix of the input layer is... By aggregating the adjacency matrix and node information, iterative updates of node features and encoding of higher-order topological information are achieved.

[0061] Through layer-by-layer aggregation and propagation in a multi-layer graph convolutional network, the first layer allows each node to aggregate information from its one-hop neighbors. The second layer further aggregates information from its two-hop neighbors, and so on. As the number of layers increases, node features gradually integrate topological structure information and resource state information from higher-order neighborhoods, resulting in a high-dimensional embedding representation for each node. This approach integrates multi-hop connectivity patterns at the topological level and incorporates the resource state distribution of satellite nodes over a larger range at the resource level, enabling the embedding representation to possess both topological relevance and resource similarity.

[0062] By leveraging the hierarchical propagation mechanism of graph convolutional networks, the clustering model automatically captures the potential topological patterns and resource distribution rules of satellite networks. This eliminates the need for manually designing complex graph topological features or resource distance metrics, thereby improving the automation level and generalization adaptability of the clustering method to different constellation configurations.

[0063] In one implementation, the graph convolutional network adopts a two-layer structure, wherein the nonlinear activation function of the first layer is a first nonlinear activation function, and the second layer adopts a classification output function to output the category probability of each satellite node; The training phase of the graph convolutional network specifically includes the following steps: Step S2301: Construct a classification loss function based on the label information of satellite nodes with known cluster labels and the predicted class probabilities output by the graph convolutional network; Step S2302: With the goal of minimizing the error between the predicted class probability and the true label, iteratively optimize the learnable weight matrix of the graph convolutional network.

[0064] In this embodiment, within the framework of a semi-supervised node classification task, the graph convolutional network employs a two-layer structure. The first layer uses the ReLU function as its nonlinear activation function, and the second layer uses the Softmax function as the classification output function, outputting the probability that each satellite node belongs to its respective cluster category. The complete forward propagation process can be represented as follows:

[0065] in, The input is the node feature matrix. This is the learnable weight matrix for the first layer. This is the learnable weight matrix for the second layer. The first layer uses the ReLU activation function to perform a non-linear mapping on the aggregated features after linear transformation, enhancing the model's ability to fit complex topologies and resource coupling relationships. The second layer uses the Softmax function to normalize the output vector into a class probability distribution, with each satellite node corresponding to a probability vector. The dimension of the probability vector is equal to the total number of star clusters, and the components in the vector represent the probability value of the satellite node being assigned to the corresponding star cluster. The final node embedding captures the node's position in the topology and encodes its resource status and functional characteristics.

[0066] During the training phase of the graph convolutional network, a semi-supervised learning strategy is employed. Specifically, a small subset of satellite nodes with known cluster labels are used as training samples. Their label information is compared with the predicted class probabilities output by the graph convolutional network to construct a classification loss function. The loss function adopts the form of cross-entropy loss, and its mathematical expression is: , in, It is a collection of labeled nodes. For the number of categories, For real labels, To predict probabilities.

[0067] During training, the optimization objective is to minimize the aforementioned loss function. The learnable weight matrices of each layer in the graph convolutional network are then processed using the backpropagation algorithm. and Iterative optimization was performed. Since the loss function is only calculated on labeled nodes, while the model's forward propagation covers all nodes in the graph, including a large number of unlabeled nodes, the topological structure and resource state information of unlabeled nodes still indirectly participate in model training through neighborhood aggregation, having a regularizing effect on feature propagation and gradient backpropagation of labeled nodes. This semi-supervised paradigm, which uses a large number of unlabeled nodes to assist a small number of labeled nodes in training, significantly improves the model's generalization ability. This allows the trained graph convolutional network to adapt to constantly changing satellite network topologies, and can quickly and efficiently complete cluster partitioning when faced with new time-slot topology snapshots without retraining.

[0068] Step S300: Calculate the evaluation index of satellite nodes in each star cluster, and select the cluster head node from each star cluster based on the evaluation index.

[0069] In this embodiment, after the satellite nodes are divided into multiple clusters, a cluster head node is selected in each cluster to be responsible for the control responsibilities such as resource management, status aggregation, task coordination and resource scheduling within the cluster.

[0070] Evaluation metrics are a set of quantifiable values ​​used to measure the suitability of each satellite node within a cluster to serve as the cluster head. Their calculation comprehensively considers factors such as the node's topological centrality, resource carrying capacity, and mission adaptability. The cluster head node is the satellite node with the best evaluation metrics in each cluster, and it assumes the role of cluster control.

[0071] The introduction of evaluation metrics transforms cluster head selection from a traditional method based on a single topology rule or random assignment into a quantitative decision-making process that comprehensively considers multiple dimensions. This ensures that the selected cluster head has good topological accessibility to its members, sufficient onboard resource margins to handle additional management overhead, and also takes into account the preferences required for specific mission scenarios. By determining an optimal cluster head node for each cluster, the collaborative efficiency within the cluster is improved, and the stability and continuous service capability of the cluster organizational structure are enhanced.

[0072] In one implementation, the step of calculating the evaluation index of satellite nodes within each cluster and selecting cluster head nodes from each cluster based on the evaluation index specifically includes the following steps: Step S310: For each satellite node, obtain the average node connectivity of the satellite node based on the network topology diagram; Step S320: Based on the resource status of the satellite nodes in the network topology diagram, determine the on-board processing capacity of the satellite nodes; Step S330: Determine the cluster head priority coefficient of the satellite node based on its mission attributes, payload capacity, or orbital position; Step S340: The average connectivity of the nodes, the on-board processing capability, and the cluster head priority coefficient are weighted and summed to obtain the comprehensive weight of the satellite nodes as the evaluation index, wherein the sum of the weight factor of the average connectivity of the nodes and the weight factor of the on-board processing capability is one. Step S350: Determine the satellite node with the largest comprehensive weight in each star cluster as the cluster head node of the star cluster.

[0073] In this embodiment, the selection of the cluster head node is achieved through multi-attribute weighted evaluation.

[0074] First, for each satellite node, its average node connectivity is obtained based on the network topology graph. Average node connectivity reflects a satellite node's connectivity and topological centrality within the local network. It is obtained by averaging the number of directly connected neighboring nodes within the current cluster. A higher average node connectivity indicates a larger number of directly communicable neighbors, resulting in better network reachability and lower communication forwarding overhead within the cluster, making it more suitable as a control and coordination center within the cluster.

[0075] Secondly, the on-board processing capacity of the satellite node is determined based on the resource status of the network topology diagram. Onboard processing power measures a satellite node's resource support capability for cluster head management tasks under current resource conditions. Specifically, the satellite's current processing power is determined by its own computing power level and real-time load, reflecting the node's available resource margin. The stronger the onboard processing power, the more sufficient the node's computing and storage capacity remains to perform additional cluster management tasks beyond its normal business operations.

[0076] Third, the cluster head priority coefficient is determined based on the satellite node's mission attributes, payload capacity, or orbital position. The cluster head priority coefficient is expressed as... The cluster head priority coefficient is used to reflect the prior priority of a specific satellite node under mission attributes, payload capacity, orbital position, or operational constraints. For example, a scientific observation satellite carrying a specific detection payload may need to assume the role of cluster head in order to perform on-orbit fusion preprocessing of observation data, while a satellite in a special orbital plane may be more suitable to serve as a cross-cluster interaction agent due to its more stable Earth coverage.

[0077] Based on the above three dimensions of indicators, the formula for calculating the comprehensive weight of any satellite node is defined as follows:

[0078] in, This represents the average connectivity of the nodes. Indicates the satellite's processing capacity. This represents the priority coefficient of the node cluster head. and As the attribute weighting factor, satisfying .

[0079] The calculated comprehensive weight is used as the evaluation index for the satellite node. This index unifies considerations from three different dimensions—topological connectivity, resource support capability, and mission adaptation priority—into a comparable value. Finally, among all satellite clusters, the satellite node with the highest comprehensive weight is determined as the cluster head node. Because the weighting factor can be flexibly configured, this election mechanism can adapt to conventional scenarios where topological connectivity is the primary consideration, and also to special scenarios where resource balancing or mission priority is the primary objective.

[0080] Compared to traditional cluster head determination methods based on a single topological metric such as maximum connectivity or random election, the aforementioned multi-attribute weighted mechanism can more comprehensively reflect the topological centrality, resource carrying capacity, and task adaptability of nodes, thereby improving the scientific rigor and robustness of cluster head election results and enhancing the stability and continuous service capability of the star cluster organization structure.

[0081] Step S400: Using the star cluster as a resource management unit, abstract and aggregate the heterogeneous resources of each satellite node within the star cluster to construct a star cluster-level distributed virtual resource pool.

[0082] In this embodiment, after the cluster division and cluster head election are completed, resources are organized using each cluster as the basic resource management unit.

[0083] A resource management unit (RMU) is a logical entity that manages and schedules various resources of multiple satellites within a cluster in a unified manner; its boundaries are defined by the cluster's membership. Heterogeneous resources refer to various types of resources provided by different satellite nodes within a cluster, differing in hardware architecture, performance parameters, and interface protocols. These primarily include computing resources, storage resources, and network bandwidth resources. Abstraction operations involve shielding the underlying physical details of various heterogeneous resources, exposing a unified functional interface and logical view. Aggregation operations merge and integrate similar resources from multiple satellite nodes within a cluster at the logical level, forming a logically unified resource whole.

[0084] Cluster-level distributed virtual resource pools are logical resource sets formed by abstracting and aggregating heterogeneous resources from multiple satellites within a single cluster. They virtualize and integrate previously scattered, heterogeneous, and physically isolated satellite resources into a shared resource pool that upper-layer services can request and schedule on demand. This allows upper-layer applications to request services from the pool without needing to know the specific satellite where a resource is located. Furthermore, because the resource pool is confined to a single cluster, it leverages the topological proximity within the cluster, avoiding the high communication latency and bandwidth overhead associated with long-distance resource scheduling across clusters. This architecture breaks through the limitations of traditional single-satellite resource silos and provides a distributed pooling management solution for the efficient utilization of resources in large-scale low-Earth orbit satellite constellations.

[0085] In one implementation, the step of using the star cluster as a resource management unit to abstract and aggregate the heterogeneous resources of each satellite node within the star cluster to construct a star cluster-level distributed virtual resource pool specifically includes the following steps: Step S410: Obtain the available computing resources and available storage resources of each satellite node, and construct a cluster resource status description to record the resource status of all nodes in the cluster; Step S420: Through virtualization, the computing resources of each satellite node in the star cluster are abstracted and encapsulated to form a virtual computing resource pool; Step S430: Through virtualization, the storage resources of each satellite node in the star cluster are unified and aggregated to form a virtual storage pool; Step S440: Through virtualization, the inter-satellite links between satellite nodes within the star cluster are abstracted to construct logical virtual links and bandwidth resource pools; Step S450: Integrate the virtual computing resource pool, virtual storage pool, logical virtual link and bandwidth resource pool to obtain a cluster-level distributed virtual resource pool, wherein the cluster head node of the cluster performs the intra-cluster autonomous management and inter-cluster collaborative management of the cluster-level distributed virtual resource pool.

[0086] In this embodiment, the construction process and architecture of the cluster-level distributed virtual resource pool are as follows: Figure 5 As shown. First, the available computing and storage resources of each satellite node are obtained, and a cluster resource status description is constructed to record the real-time resource status information of all nodes in the cluster in a structured manner.

[0087] At the computing resource level, virtualization technology is used to abstract and encapsulate the computing resources of each satellite node within the cluster, shielding the hardware differences between different satellite platforms in terms of processor architecture, instruction sets, and operating systems, thus forming a unified virtual computing resource pool. Specifically, this can be achieved by deploying lightweight containers or virtual machine monitors on satellite nodes, mapping the physical computing power of each satellite into several standardized virtual computing units, which can be requested and flexibly scaled by various on-orbit tasks within the cluster as needed.

[0088] At the storage resource level, storage resources of each satellite node within the cluster are uniformly reconstructed and logically aggregated through storage virtualization mechanisms to form a virtual storage pool that can be allocated on demand. Specifically, block-level address mapping is used to map storage blocks on each satellite's physical storage device to a unified logical address space. Logical volume management is used to combine scattered physical volumes into cross-node logical storage volumes. Furthermore, a distributed data sharding mechanism is used to split data according to a strategy and store it across multiple satellites within the cluster. This achieves logical integration of storage capacity while providing data redundancy and parallel access capabilities, and uses a unified namespace management externally.

[0089] At the network resource level, inter-satellite links between satellite nodes within a cluster are abstracted through network virtualization mechanisms, constructing logical virtual links and bandwidth resource pools. Specifically, the control plane and data plane of inter-satellite links within the cluster are separated. The cluster head node centrally maintains the global network view and constructs end-to-end virtual links at the logical level. Simultaneously, combined with bandwidth slicing, the bandwidth resources of physical links are divided into multiple independently schedulable logical slices to meet the differentiated bandwidth and latency requirements of different service flows, achieving elastic allocation and on-demand scheduling of network resources within the cluster.

[0090] Specifically, time slots are defined based on dynamic satellite clustering results. The set of star clusters within is represented as:

[0091] in, Indicates the first A set of satellite nodes within a cluster consists of a cluster head node and multiple member nodes.

[0092] Further define time slots The set of intra-cluster inter-satellite links is represented as:

[0093] in, This refers to the communication links between satellite nodes within the same satellite cluster. Based on the above definition, each satellite cluster is considered a logically autonomous unit with unified resource scheduling and collaborative management capabilities.

[0094] Based on the multidimensional spaceborne resource characterization model, any satellite node in a time slot is defined. The resource state vector is represented as:

[0095] in, and These represent the available computing resources and available storage resources of a node, respectively. This forms the cluster-level resource matrix. It describes the onboard resource status of all nodes within the cluster.

[0096] During the resource pool construction process, different types of heterogeneous resources are virtualized, abstracted, and uniformly aggregated. Within the satellite cluster, virtual machines or containerization technologies are used to abstract and encapsulate the physical computing resources of each satellite node, shielding the underlying hardware heterogeneity and forming a uniformly scheduled virtual computing resource pool. Its logical representation is as follows:

[0097] The virtual computing resource pool aggregates the computing power of multiple nodes within a cluster into a unified set of logical computing resources.

[0098] By employing storage virtualization mechanisms such as block-level address mapping, logical volume management, and distributed data sharding, the dispersed physical storage space within a cluster is uniformly reconstructed and logically aggregated to form a virtual storage pool that can be allocated on demand, as represented by:

[0099] Virtual storage pools enable unified namespace management and logical integration of multi-star distributed storage resources.

[0100] To eliminate physical link heterogeneity and topology differences, software-defined networking (SDN) concepts and bandwidth slicing mechanisms are introduced to abstractly model intra-cluster inter-satellite links, constructing logical virtual links and bandwidth resource pools. The logical link model is represented as follows:

[0101] in, This indicates the available bandwidth resources for inter-satellite links.

[0102] Finally, the virtual computing resource pool, virtual storage pool, and logical virtual link and bandwidth resource pool are integrated to obtain a cluster-level distributed virtual resource pool. In terms of management architecture, the cluster head node performs intra-cluster autonomous management and inter-cluster collaborative management of the cluster-level distributed virtual resource pool. Within a cluster, the cluster head node is responsible for resource status aggregation, task scheduling, member management, and topology updates, uniformly orchestrating and allocating the three types of virtual resource pools within the cluster on demand, achieving local autonomous management within the cluster. Between clusters, the cluster head node acts as an external interaction proxy for its own cluster, responsible for exchanging resource status summaries, discovering cross-cluster resources, collaborative scheduling, and negotiating task migration with the cluster head nodes of other clusters. When the resources within a cluster cannot meet local business needs, the cluster head can query the resource availability of neighboring clusters through the cross-cluster resource discovery mechanism and offload some tasks to neighboring clusters with sufficient resources through task migration negotiation.

[0103] By unifying the abstraction of underlying link resources, elastic allocation and on-demand scheduling of network resources within a cluster can be achieved. Compared to the traditional globally centralized architecture, the distributed resource management system reduces resource management overhead from a hyperlinear growth proportional to the total constellation size to a linear growth proportional to the size of a single cluster, greatly alleviating the pressure on the control plane. Simultaneously, each cluster acts as a relatively independent fault domain; when a node fails or a link is interrupted in an individual cluster, its impact is limited to that cluster and will not affect the resource scheduling of the entire constellation, thus significantly improving the system's robustness. Furthermore, since resource pool construction and scheduling are performed at the cluster level, when the constellation expands with the addition of new satellites, it is only necessary to add the new node to a nearby cluster or build a new cluster in a new orbital plane, demonstrating excellent scalability.

[0104] In summary, this invention aims to provide a method for constructing a virtual resource pool based on dynamic satellite clustering. By introducing a graph convolutional network, a large-scale low-Earth orbit (LEO) satellite constellation is divided into several locally autonomous clusters. A multi-attribute weighted decision mechanism is used to select the cluster head node. Furthermore, heterogeneous resources from multiple satellites within a cluster are uniformly abstracted, virtualized, and logically aggregated to form a cluster-level distributed virtual resource pool. This method effectively reduces the management complexity and control overhead of large-scale LEO satellite networks, breaks the limitation of isolated single-satellite resources, and achieves efficient pooling and sharing of multi-satellite resources.

[0105] like Figure 6As shown in the figure, this embodiment of the invention provides a virtual resource pool construction system based on dynamic satellite clustering. The system includes: a network topology graph construction module 10, a satellite cluster division module 20, a cluster head node selection module 30, and a virtual resource pool construction module 40.

[0106] Specifically, the network topology graph construction module 10 is used to acquire the topological connection relationships and resource status of satellite nodes in the time-varying satellite network, and construct a network topology graph representing the time-varying satellite network; the cluster partitioning module 20 is used to aggregate the topological structure features and resource status features of each satellite node based on the network topology graph using a graph neural network model to obtain the embedded representation of each satellite node, and divide the satellite nodes into multiple clusters based on the embedded representation; the cluster head node selection module 30 is used to calculate the evaluation index of satellite nodes in each cluster, and select the cluster head node from each cluster based on the evaluation index; the virtual resource pool construction module 40 is used to abstract and aggregate the heterogeneous resources of each satellite node in the cluster using the cluster as a resource management unit, and construct a cluster-level distributed virtual resource pool.

[0107] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 7 As shown, the terminal device includes a processor, memory, network interface, display screen, and temperature sensor connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for constructing a virtual resource pool based on dynamic satellite clustering. The display screen can be an LCD screen or an e-ink screen. The temperature sensor is pre-installed inside the terminal device to detect the operating temperature of the internal components.

[0108] Those skilled in the art will understand that Figure 7 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0109] In one embodiment, a terminal device is provided, including a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs including instructions for performing operations as described in the embodiments of the methods above.

[0110] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0111] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for constructing a virtual resource pool based on dynamic satellite clustering, characterized in that, The method includes: Obtain the topological connections and resource status of satellite nodes in the time-varying satellite network, and construct a network topology map representing the time-varying satellite network; Based on the network topology graph, the topological features and resource status features of each satellite node are aggregated using a graph neural network model to obtain the embedded representation of each satellite node, and the satellite nodes are divided into multiple star clusters based on the embedded representation. Calculate the evaluation index of satellite nodes within each cluster, and select the cluster head node from each cluster based on the evaluation index; Using the star cluster as a resource management unit, the heterogeneous resources of each satellite node within the star cluster are abstracted and aggregated to construct a star cluster-level distributed virtual resource pool.

2. The method for constructing a virtual resource pool based on dynamic satellite clustering according to claim 1, characterized in that, The process of acquiring the topological connections and resource status of satellite nodes in the time-varying satellite network, and constructing a network topology map representing the time-varying satellite network, includes: The observation period corresponding to the time-varying satellite network is divided into multiple time slots; For each time slot, an undirected network topology graph corresponding to the time slot is constructed based on the inter-satellite link connection relationship between satellite nodes within the time slot; For each satellite node, the available computing resources and available storage resources of the satellite node are obtained through local status monitoring; By broadcasting and interacting with neighboring nodes to synchronize information, the resource status information of each satellite node in each time slot and the available bandwidth resources of each inter-satellite link in each time slot are obtained. The resource status information includes the available computing resources and available storage resources of each satellite node. The resource status information and the available bandwidth resources are used as attribute information of satellite nodes and inter-satellite links in the network topology diagram of the corresponding time slot to obtain the attribute network topology diagram of each time slot. The attribute network topology graphs of each time slot are integrated to obtain the network topology graph, which is a static topology snapshot sequence composed of the attribute network topology graphs of each time slot.

3. The method for constructing a virtual resource pool based on dynamic satellite clustering according to claim 1, characterized in that, Based on the network topology graph, the topological features and resource status features of each satellite node are aggregated using a graph neural network model to obtain an embedded representation of each satellite node. Based on this embedded representation, the satellite nodes are divided into multiple clusters, including: Based on the topological connection relationship and resource status of each satellite node in the network topology diagram, multi-dimensional attribute features of each satellite node are extracted to form a node feature matrix. The multi-dimensional attribute features include satellite location, available computing resources, available storage resources, available bandwidth resources, and node connectivity. Construct an adjacency matrix based on the connection relationships between satellite nodes; The node feature matrix and the adjacency matrix are input into the graph neural network model, and the features of each satellite node are aggregated through the graph neural network model to obtain the embedded representation of each satellite node; Based on the embedded representation of each satellite node, the satellite nodes are clustered.

4. The method for constructing a virtual resource pool based on dynamic satellite clustering according to claim 3, characterized in that, The step of inputting the node feature matrix and the adjacency matrix into the graph neural network model, and aggregating the features of each satellite node through the graph neural network model to obtain the embedded representation of each satellite node, includes: Input the node feature matrix and the adjacency matrix into the graph convolutional network; In each layer of the graph convolutional network, the set of neighboring nodes of each node is determined based on the adjacency matrix; For each node, the node's own features are weighted and aggregated with the features of the node's neighboring nodes. After linear transformation through a learnable weight matrix, the node features are processed by a non-linear activation function to obtain an updated node feature representation. Through the layer-by-layer aggregation and propagation of multi-layer graph convolutional networks, node features are gradually integrated with the topological structure information and resource status information of higher-order neighborhoods to obtain the embedded representation of each satellite node.

5. The method for constructing a virtual resource pool based on dynamic satellite clustering according to claim 4, characterized in that, The graph convolutional network adopts a two-layer structure, where the first layer uses a first nonlinear activation function and the second layer uses a classification output function to output the class probability of each satellite node. During the training phase of the graph convolutional network, a classification loss function is constructed based on the label information of satellite nodes with known cluster labels and the predicted class probabilities output by the graph convolutional network. With the goal of minimizing the error between the predicted class probability and the true label, the learnable weight matrix of the graph convolutional network is iteratively optimized.

6. The method for constructing a virtual resource pool based on dynamic satellite clustering according to claim 1, characterized in that, The calculation of evaluation metrics for satellite nodes within each cluster, and the selection of cluster head nodes from each cluster based on these evaluation metrics, includes: For each satellite node, the average node connectivity of the satellite node is obtained based on the network topology diagram; Based on the resource status of the satellite nodes in the network topology diagram, the on-board processing capacity of the satellite nodes is determined. Based on the mission attributes, payload capacity, or orbital position of the satellite node, determine the cluster head priority coefficient of the satellite node; The average connectivity of the nodes, the on-board processing capacity, and the cluster head priority coefficient are weighted and summed to obtain the comprehensive weight value of the satellite nodes as the evaluation index, wherein the sum of the weight factor of the average connectivity of the nodes and the weight factor of the on-board processing capacity is one. The satellite node with the largest comprehensive weight in each star cluster is determined as the cluster head node of that star cluster.

7. The method for constructing a virtual resource pool based on dynamic satellite clustering according to claim 1, characterized in that, The method of using the star cluster as a resource management unit to abstract and aggregate the heterogeneous resources of each satellite node within the star cluster to construct a star cluster-level distributed virtual resource pool includes: Obtain the available computing and storage resources of each satellite node, and construct a cluster resource status description to record the resource status of all nodes in the cluster; Virtualization is used to abstract and encapsulate the computing resources of each satellite node within the star cluster, forming a virtual computing resource pool. Virtualization is used to unify and aggregate the storage resources of each satellite node within the cluster, forming a virtual storage pool. By virtualization, the inter-satellite links between satellite nodes within the cluster are abstracted to construct logical virtual links and bandwidth resource pools. By integrating the virtual computing resource pool, virtual storage pool, logical virtual link and bandwidth resource pool, a cluster-level distributed virtual resource pool is obtained, wherein the cluster head node of the cluster performs the intra-cluster autonomous management and inter-cluster collaborative management of the cluster-level distributed virtual resource pool.

8. A virtual resource pool construction system based on dynamic satellite clustering, characterized in that, The system includes: The network topology graph construction module is used to obtain the topological connection relationship and resource status of satellite nodes in the time-varying satellite network, and construct a network topology graph representing the time-varying satellite network. The cluster partitioning module is used to aggregate the topological features and resource status features of each satellite node based on the network topology graph using a graph neural network model to obtain the embedded representation of each satellite node, and to partition the satellite node into multiple clusters based on the embedded representation; The cluster head node selection module is used to calculate the evaluation index of satellite nodes in each cluster and select the cluster head node from each cluster based on the evaluation index. The virtual resource pool construction module is used to abstract and aggregate the heterogeneous resources of each satellite node within the star cluster, using the star cluster as the resource management unit, to construct a star cluster-level distributed virtual resource pool.

9. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a virtual resource pool construction program based on dynamic satellite clustering stored in the memory and executable on the processor. When the processor executes the virtual resource pool construction program based on dynamic satellite clustering, it implements the steps of the virtual resource pool construction method based on dynamic satellite clustering 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 virtual resource pool construction program based on dynamic satellite clustering. When the virtual resource pool construction program based on dynamic satellite clustering is executed by a processor, it implements the steps of the virtual resource pool construction method based on dynamic satellite clustering as described in any one of claims 1-7.

Images