Heterogeneous task low-orbit satellite task unloading method based on space-time diagram attention network

Abstract

The invention discloses a heterogeneous task low orbit satellite task unloading method based on a space-time diagram attention network. The method comprises the steps of obtaining space-time diagram data related to a to-be-unloaded task; the time-space diagram data are input into a time-space conversion model, a prediction result of a subsequent network state is obtained, the time-space conversion model integrates a graph convolutional network module, a long-short-term memory network module and a multi-head attention module, the graph convolutional network module extracts spatial features of satellite network topology under each time step, and the multi-head attention module extracts the spatial features of the satellite network topology under each time step; the long-short-term memory network module captures satellite network states and time sequence dependence of task loads, and the multi-head self-attention module learns association weights among different spatial-temporal characteristics and outputs prediction results of future satellite network states; and based on the prediction result and the attribute of the to-be-unloaded task, setting an optimization objective function of an opportunity task unloading algorithm problem to determine a decision result of task unloading. According to the method, the task unloading efficiency and the resource utilization rate in the low-orbit satellite network can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary field of LEO satellite network technology, edge computing technology, artificial intelligence, and machine learning applications, and in particular to a low-orbit satellite task offloading method based on a spatiotemporal graph attention prediction network, which is particularly suitable for opportunistic task offloading. Background Art

[0002] As the low-Earth orbit (LEO) satellite market continues to expand, the portfolio of services it offers is becoming increasingly diverse, extending from traditional internet access services to emerging applications such as precision navigation, remote sensing, and direct device connectivity. However, the inherent spatiotemporal uncertainty of LEO networks—manifested by dynamic fluctuations in link quality, real-time changes in the availability of resources such as computing, bandwidth, and energy, and the continuous evolution of network topology—poses a fundamental challenge to the effective provision of these diverse services. In this context, leveraging LEO satellite communication signals as signals of opportunity has become a highly promising technological approach, and "task offloading" technology is emerging as a key means to efficiently capture and exploit these fleeting communication and computing opportunities in such opportunistic networks.

[0003] In the early stages of research on task offloading in dynamic environments, academia and industry have explored various technical approaches. Traditional solutions often rely on graph-based models, such as using time-varying graphs (TEGs) to characterize network dynamics to support the efficient deployment of virtual network functions (VNFs), or deriving steady-state probability matrices to achieve optimal task allocation. While these approaches account for network dynamics to some extent, they often lack the flexibility to capture and quickly respond to short-lived offloading opportunities. As the highly dynamic nature of LEO systems has become a focus of industry attention, artificial intelligence methods such as deep learning (DL) and reinforcement learning (RL) have gained widespread attention and application. For example, innovative solutions such as the LSTM network model for network traffic forecasting, the multi-agent deep reinforcement learning (MADRL) algorithm for collaborative offloading in satellite mobile edge computing (SMEC) scenarios, and offloading strategies designed to achieve latency optimization have been proposed. Furthermore, the use of dynamic graph neural networks (DGNNs) or GCN-LSTM models to explicitly model and learn spatiotemporal dependencies also represents an important research direction in this field.

[0004] However, it is worth noting that many current task offloading strategies are primarily designed for terrestrial network environments (such as vehicular ad hoc networks or mobile edge computing scenarios). These strategies are difficult to directly and efficiently apply to the network scale, node characteristics, and dynamic behavior unique to LEO constellations. Currently, task offloading solutions specifically designed for LEO constellation environments often lack sufficiently sophisticated prediction and evaluation capabilities to leverage the spatiotemporal correlations of multi-dimensional information such as link quality, node computing and storage capabilities, and specific mission requirements. This makes it difficult to accurately judge and grasp the true quality of instantaneous offloading opportunities. This limitation significantly restricts their effectiveness in making forward-looking and truly opportunistic decisions, and it is often difficult to fully and accurately capture and address the complex interactions and influences between various uncertainties. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art and provide a low-orbit satellite task offloading method based on a spatiotemporal graph attention network. The method comprises the following steps:

[0006] Obtaining spatiotemporal graph data related to the task to be offloaded, where the spatiotemporal graph data is based on a spatiotemporal graph representation constructed for a satellite network model, where nodes represent satellites or ground stations, edges represent connections between them, and the graph dynamically evolves over time;

[0007] Inputting the spatiotemporal graph data into a spatiotemporal conversion model to obtain a prediction result of the satellite network state at a subsequent time, wherein the spatiotemporal conversion model integrates a graph convolutional network module, a long short-term memory network module, and a multi-head attention module. The graph convolutional network module is used to extract the spatial features of the satellite network topology at each time step, the long short-term memory network module is used to capture the time series dependency of the satellite network state and the mission load, and the multi-head self-attention module is used to learn the association weights between different spatiotemporal features and output a prediction result of the future satellite network state.

[0008] Based on the prediction results and the attributes of the task to be offloaded, an optimization objective function of the opportunistic task offloading algorithm problem is set to determine a decision result of task offloading;

[0009] Based on the decision result, the task data to be offloaded and the execution instruction are sent to the determined offloading target and the task execution process is monitored.

[0010] Compared with existing technologies, the present invention's LEO satellite task offloading method, based on a spatiotemporal graph attention network, overcomes the limitations of traditional methods through more intelligent data analysis and decision-making capabilities, providing a more efficient and adaptable solution for task offloading in LEO satellite networks. This invention is of great significance for fully leveraging the potential of LEO constellations and supporting various emerging applications in future integrated space-ground networks.

[0011] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0013] Figure 1 is a flowchart of a low-orbit satellite task offloading method based on a spatiotemporal graph attention network according to an embodiment of the present invention;

[0014] Figure 2 is a structural block diagram of a spatiotemporal transformation model ST-Trans according to an embodiment of the present invention;

[0015] Figure 3 is a performance comparison curve of the spatiotemporal transformation model ST-Trans according to one embodiment of the present invention and a benchmark method in global attribute prediction;

[0016] Figure 4 This is a performance comparison curve of the average task completion delay of a heterogeneous task low-orbit satellite task offloading method based on a spatiotemporal graph attention network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention.

[0018] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0019] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0020] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0021] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0022] In response to the shortcomings of existing low-orbit satellite network task offloading methods in dynamic spatiotemporal state perception, complex data dependency modeling, and opportunistic offloading decisions, which lead to problems such as low resource utilization and high task processing delay, the present invention provides a low-orbit satellite opportunistic task offloading method based on spatiotemporal graph attention network. It introduces a novel ST-Trans model (or spatiotemporal transformation model), which integrates GCN (graph convolutional network), LSTM (long short-term memory network) and multi-head self-attention mechanism to achieve accurate spatiotemporal prediction of satellite network topology evolution, link quality changes and node load status, and combines an opportunistic task offloading algorithm based on the prediction results to achieve near-optimal task allocation and routing selection while meeting task QoS (quality of service) constraints, thereby minimizing average task completion delay and system energy consumption.

[0023] In general, the low-orbit satellite opportunistic task offloading method based on spatiotemporal graph attention network provided by the present invention has the following features and advantages:

[0024] 1) Build a dynamic satellite network spatiotemporal graph representation method that effectively characterizes satellite nodes, ground stations, their connections over time, and node / link attributes. This representation method is the foundation for subsequent accurate predictions and intelligent decision-making.

[0025] 2) Design an ST-Trans prediction model, which includes a graph convolutional network (GCN) layer to extract network topological spatial features, a long short-term memory (LSTM) layer to capture the time series dependencies between network status and task load, and a multi-head self-attention layer to learn the deep correlations and importance weights between different spatiotemporal features. The model inputs can include historical satellite orbit data, link status, node load, and task information, and outputs predicted values ​​for key network parameters for a period of time in the future.

[0026] 3) Develop an opportunistic task offloading mechanism that uses the future network state prediction results output by the ST-Trans model, combined with a heuristic task offloading algorithm, to dynamically determine the optimal offloading target (specific satellite node or ground station) and corresponding execution strategy for each pending task based on real-time task requirements and dynamically changing decision metrics (such as predicted node computing load, link transmission delay, task priority, satellite remaining energy, etc.).

[0027] 4) Define the system model, including the network model (describing the satellites, ground stations, and their connectivity), the satellite model (describing the computational, storage, and energy properties of satellite nodes), the mission model (describing the computational effort, data size, and deadlines of the mission), the communication model (describing the data transmission characteristics of the inter-satellite and satellite-to-ground links), and the computational model (describing the execution process of the mission on the nodes). Related mathematical representations or key equations, such as the formula for calculating mission completion delay and the system energy consumption model, are also defined to provide a basis for quantifying optimization objectives and designing algorithms.

[0028] 5) Simulation experiments demonstrate that compared to existing benchmark offloading algorithms (such as nearest visible satellite offloading, random offloading, and some published heuristic algorithms), the present invention demonstrates significant advantages in key performance indicators, such as average task completion latency, total system energy consumption, and balanced network resource utilization. For example, experimental results show that the present invention can reduce average task completion latency by 8.4% and more effectively balance network load.

[0029] Specifically, see Figure 1 As shown, the provided low-orbit satellite task offloading method based on spatiotemporal graph attention network includes the following steps:

[0030] Step S1: Initialize the LEO network model.

[0031] For example, a low-orbit satellite network model is initialized, which includes multiple low-orbit satellite nodes, ground station nodes, and the dynamic visibility, communication link characteristics, and three heterogeneous task load characteristics of computing, communication, and remote sensing between them.

[0032] Step S2: construct a space-time graph.

[0033] Based on the satellite network model, a spatio-temporal graph (STG) is constructed, where the nodes of the graph represent satellites or ground stations, and the edges represent the connection relationships between them, and it evolves dynamically over time.

[0034] Step S3: Input the spatiotemporal graph STG into the ST-Trans model.

[0035] The ST-Trans model comprehensively analyzes multi-source heterogeneous feature information to capture deep correlations between different feature dimensions, achieving global and accurate perception and prediction of node and link resource availability. For example, input spatiotemporal graph data into the ST-Trans model, which integrates a GCN module, an LSTM module, and a Multi-head Self-Attention module.

[0036] Step S4: Use the GCN module to extract spatial features.

[0037] The GCN module is used to extract the spatial features of the satellite network topology at each time step.

[0038] Step S5: Use the LSTM module to capture time dependency.

[0039] The LSTM module is used to capture the time series dependencies of satellite network status and heterogeneous mission loads.

[0040] Step S6: Use the multi-head self-attention mechanism to learn associations and predict future states.

[0041] The multi-head self-attention module is used to learn the complex association weights between different spatiotemporal features and output prediction results for future network status or available resources.

[0042] Step S7: Execute opportunistic task offloading decision.

[0043] In one embodiment, an opportunistic task offloading algorithm is used to make a decision based on the prediction result and the attributes of the task to be offloaded.

[0044] Step S8: Determine the best uninstallation solution.

[0045] The opportunistic task offloading algorithm determines the optimal offloading target satellite or ground station and data transmission path to optimize the individual completion delay of heterogeneous tasks, the total completion delay and system energy consumption.

[0046] For example, optimizing the total task completion latency and resource utilization is based on a preset utility function that performs a weighted combination of latency and resource utilization.

[0047] Step S9: Execute task unloading and monitor

[0048] Send task data and execution instructions to the selected offload target and monitor the heterogeneous task execution process.

[0049] It should be noted that accurately capturing the brief but valuable high-quality unloading window when a LEO satellite passes by can be achieved by predicting the future network status and available resources in step S6, combined with the decision-making algorithm in step S7, to predict and plan the unloading timing and path before the satellite enters the communication coverage of the mission source node or ground station.

[0050] The core function of the ST-Trans model is to predict the future state of network nodes (for example, node resource availability, link quality between nodes, etc.) and the dynamic evolution properties of multi-dimensional heterogeneous tasks, so as to accurately predict potential offloading opportunities. Figure 2 As shown in the figure, when implementing the ST-Trans model proposed in this invention, GCN is first used to extract spatial features; then, an LSTM network is used to preliminarily model the time series data; finally, a multi-head self-attention mechanism is used to effectively capture the complex and long-range temporal dependencies in the data and learn the importance weights between different features.

[0051] In one embodiment, a layer of GCN updates the representation of a node by aggregating information from its neighboring nodes. The update rule is as follows:

[0052]

[0053] Among them, H (l) represents the hidden representation of the lth layer; represents the normalized adjacency matrix; Represents the corresponding diagonal matrix; W (l) represents a trainable weight matrix; σ represents the activation function. The LSTM network processes sequence data through its internal input gate, forget gate, and output gate, updating its cell state and hidden state to capture time series patterns in the data. The multi-head self-attention mechanism calculates weighted input features using the following formula:

[0054]

[0055] Where K is the key matrix, V is the value matrix, Q is the query matrix, and d k is the scaling factor.

[0056] This enables the model to dynamically assess the relative importance of spatial and temporal features and weight them accordingly.

[0057] The ST-Trans model processes the input sequence X={X1,X2,…,X T},in The F-dimensional features of N network nodes and their associated tasks at time step t are encoded. The initial spatial features are extracted by the GCN layer. These spatially aware features are then fed into the LSTM layer to model sequence characteristics. The output of the LSTM is further processed by a multi-head attention mechanism to capture long-range dependencies and effectively integrate spatiotemporal information. This results in a comprehensive spatiotemporal embedding that represents the predicted service status. Based on these prediction properties, the future resource availability prediction for node i at time t+1 is as follows:

[0058]

[0059] in, is the total capacity, and Based on used resources As predicted.

[0060] Using the predicted service status from the ST-Trans model This module is designed to select the optimal LEO satellite node by considering the predicted future state (opportunity) and mission-specific requirements to perform proactive and opportunistic offloading of heterogeneous missions.

[0061] The opportunistic task offloading problem proposed in this paper aims to capture the optimal offloading opportunities to minimize the Total weighted completion time

[0062]

[0063] Among them, when task k is assigned to satellite s, y k,s =1;T k (Y,f) is the completion time of task k; w k is the predefined priority weight of task k. f k,s represents the computing resource allocation decision variable, Represents a set of tasks, represents the set of satellite nodes, v k represents the task delay constraint, F s Indicates the satellite computing resource capacity. Subject to indicates that it is subject to (i.e., a constraint condition).

[0064] The suitability of opportunistically offloading task j to candidate LEO node i is evaluated using a decision metric. The decision metric D(i, j) used to evaluate the quality of the prediction opportunity is expressed as follows:

[0065]

[0066] This metric uses an adjustable weight w k To balance: predicted transmission cost C trans(i,j;α j )(It depends on the task data size α j and predicted link conditions), and the relative prediction capability of task j on node i under the predicted conditions The computational load (W j is the task workload, is the predicted processing capability of the node). Represents the energy consumption of predicted task execution.

[0067] In one embodiment, the opportunistic task offloading algorithm adopts a collaborative decision-making method based on multi-agent reinforcement learning (MARL). Its state space includes the current satellite network topology, the predicted value of the resource status of each node, the predicted value of the link quality, and the queue information of heterogeneous tasks to be offloaded; the action space is the decision of offloading the task to any available node in the network or local processing; the reward function is set comprehensively based on the task completion delay, completion rate and QoS (quality of service).

[0068] The offloading component iteratively processes prioritized, heterogeneous tasks. This selection is based on an evaluation of the decision metric D(i, j) of potential target nodes, effectively selecting the best predicted opportunity. Dynamic resource allocation and availability updates are performed on the nodes during each task assignment to ensure the consumption of the identified opportunities.

[0069] To further verify the effect of the present invention, the ST-Trans model, opportunistic task offloading algorithm, etc. involved can be implemented and simulated based on the PyTorch deep learning framework using Python language. Figure 3 2 is a performance comparison curve of the ST-Trans model according to the present invention and the benchmark method in global attribute prediction. Figure 4 This is a performance comparison curve of the low-orbit satellite task offloading method of heterogeneous tasks based on the spatiotemporal graph attention network of the present invention in terms of the average task completion delay.

[0070] In summary, to address the technical issues of uneven computing resource allocation, high offloading decision latency in dynamic network environments, and insufficient capture of multi-dimensional spatiotemporal data dependencies in traditional heterogeneous task offloading in low-orbit satellite networks, this paper proposes an intelligent heterogeneous task offloading method for low-orbit satellite networks. By constructing a dynamically evolving spatiotemporal graph model of satellite networks, using a graph convolutional network to extract spatial dependencies, and combining long-short-term memory networks and a multi-head attention mechanism to capture temporal dynamics and complex feature associations, this method achieves accurate prediction of heterogeneous task characteristics in satellite network states and dynamic evaluation of available computing resources. Furthermore, with the help of an opportunistic offloading decision algorithm that minimizes task completion time and energy consumption, this method achieves optimal task allocation and routing while meeting Quality of Service (QoS) constraints.

[0071] In summary, the core idea of ​​the opportunistic task offloading method for low-orbit satellites based on a spatiotemporal graph attention network provided by the present invention is to optimize the execution efficiency and resource utilization of tasks in the LEO satellite network through accurate dynamic network state prediction and intelligent offloading decisions. This can significantly improve the offloading efficiency and resource utilization of computing, communication, and remote sensing tasks in the low-orbit satellite network, and reduce task processing delays. The present invention can be applied to earth observation data processing and space Internet of Things data analysis scenarios. By optimizing the allocation and execution of onboard edge computing tasks, the overall data processing efficiency and response speed can be improved.

[0072] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0073] Computer-readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device.Computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device or any suitable combination thereof.More specific examples (non-exhaustive list) of computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove having instructions stored thereon, and any suitable combination thereof.Computer-readable storage medium used herein is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagated by waveguides or other transmission media (for example, light pulses by fiber optic cables), or electrical signals transmitted by wires.

[0074] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0075] The computer program instructions for performing the operation of the present invention can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, Python, and conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer readable program instructions can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), is personalized by utilizing the state information of the computer readable program instructions, and the electronic circuit can execute the computer readable program instructions, thereby realizing various aspects of the present invention.

[0076] Various aspects of the present invention are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0077] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0078] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0079] The flowcharts and block diagrams in the accompanying drawings show the possible implementation architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of an instruction, and the module, program segment or part of the instruction contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0080] While various embodiments of the present invention have been described above, the foregoing description is intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A method for offloading heterogeneous low-orbit satellite tasks based on a spatiotemporal graph attention network, comprising the following steps: Obtaining spatiotemporal graph data related to the task to be offloaded, where the spatiotemporal graph data is based on a spatiotemporal graph representation constructed for a satellite network model, where nodes represent satellites or ground stations, edges represent connections between them, and the graph dynamically evolves over time; Inputting the spatiotemporal graph data into a spatiotemporal conversion model to obtain a prediction result of the satellite network state at a subsequent time, wherein the spatiotemporal conversion model integrates a graph convolutional network module, a long short-term memory network module, and a multi-head attention module. The graph convolutional network module is used to extract the spatial features of the satellite network topology at each time step, the long short-term memory network module is used to capture the time series dependency of the satellite network state and the mission load, and the multi-head self-attention module is used to learn the association weights between different spatiotemporal features and output a prediction result of the future satellite network state. Based on the prediction results and the attributes of the task to be offloaded, an optimization objective function of the opportunistic task offloading algorithm problem is set to determine a decision result of task offloading; Based on the decision result, the task data to be offloaded and the execution instruction are sent to the determined offloading target and the task execution process is monitored.

2. The method according to claim 1, characterized in that The objective function is set as: Among them, y k,s is the task assignment instruction. When task k is assigned to satellite s, y k,s =1, T k (Y,f) is the completion time of task k, w k is the predefined priority weight of task k, f k,s represents the computing resource allocation decision variable, Represents a set of tasks, represents the set of satellite nodes, ν k represents the task delay constraint, F s Indicates the satellite computing resource capacity, and subject to indicates the limitation.

3. The method according to claim 1, characterized in that The input of the spatiotemporal conversion model is multi-source heterogeneous feature information, including the orbital parameters of the low-orbit satellite node, the remaining resources on the satellite, the current utilization rate of the computing unit, the current occupancy rate of the storage unit, the bandwidth, signal-to-noise ratio, transmission delay of the communication link characteristics, and the current task queue information.

4. The method according to claim 1, wherein For the spatiotemporal graph, node features include the instantaneous computing capacity, remaining storage space, available resources, and current task load of the satellite node, and edge features include the predicted bandwidth, delay, and stability of the link.

5. The method according to claim 1, wherein The opportunistic task offloading algorithm is implemented by collaborative decision-making based on multi-agent reinforcement learning. Its state space includes the current satellite network topology, the predicted value of each node's resource status, the predicted value of the link quality, and the queue information of the tasks to be offloaded. The action space is the decision of offloading the task to available nodes in the satellite network or processing it locally; The reward function is set comprehensively based on task completion delay, completion rate and service quality QoS.

6. The method according to claim 5, characterized in that The monitoring task execution process includes collecting task completion status and actual resource consumption information for feedback adjustment of the decision result.

7. The method according to claim 1, characterized in that The tasks to be offloaded are heterogeneous tasks, including computationally intensive data processing tasks, communication relay tasks, and remote sensing data acquisition and preliminary processing tasks.

8. The method according to claim 2, characterized in that For the opportunistic task offloading algorithm, the suitability of opportunistically offloading task j to candidate satellite node i is evaluated by the following decision metric: Among them, w1, w2 and w3 are the weights of the corresponding items, C trans (i,j;α j ) is the predicted transmission cost, α j is the task data size, W j is the task workload, is the predicted processing capacity of node i, i is the node index, j is the task index, Represents the energy consumption of predicted task execution.

9. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of being run on the processor is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images