Task scheduling method and system for low-orbit satellite edge computing network

Abstract

The invention provides a task scheduling method and system for a low earth orbit satellite edge computing network, and the method comprises the steps: enabling an auxiliary control node to recognize a user service intention corresponding to a user service request according to an intention perception mechanism after the low earth orbit satellite edge computing network receives the user service request; the auxiliary control node identifies the current network service preference of each low-orbit satellite node in the corresponding subgroup according to a network service preference sensing mechanism; determining a historical data item which is stored in a weight historical database and has the highest similarity with the current user service request in two aspects of user service intention and network service preference, and generating a task scheduling strategy based on a weight coefficient of the historical data item with the highest similarity; and the auxiliary control node issues the corresponding user service to the corresponding computing node according to the task scheduling strategy to realize task processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of low-orbit satellite edge computing, and particularly relates to a task scheduling method and system for a low-orbit satellite edge computing network. BACKGROUND

[0002] With the continuous development and progress of Internet technology, artificial intelligence, Internet of Things, AR\VR and various emerging multimedia applications are emerging and developing rapidly. Such applications not only require high-speed data transmission capabilities, but also need to ensure low-latency network connections to ensure the smoothness and satisfaction of user experience. The current Internet still has the problem of limited infrastructure deployment range. In remote areas, due to geographical location and deployment of ground infrastructure costs, users have difficulty enjoying high-quality multimedia application services. Therefore, satellite networks are gradually being considered as a key component of future global network architecture, especially low earth orbit (Low Earth Orbit, LEO) satellite networks.

[0003] With the rapid development of low-orbit satellite technology, LEO satellite networks have become an indispensable part of future global communication architecture. LEO satellite networks, with their low latency and high network capacity, are gradually becoming an important part of global communication networks. Compared with traditional ground networks, LEO satellite networks have unique advantages in providing global coverage, especially in remote areas and ocean areas, providing lower latency and higher transmission rates worldwide. In addition, LEO satellite networks can further improve network reliability and flexibility through cluster deployment and distributed resource management. This provides a new solution for efficient content distribution, data transmission and real-time communication, and is conducive to promoting the global digital transformation process. With the continuous progress of technology, LEO satellite networks are expected to become the most promising and promising component of global communication infrastructure.

[0004] Although LEO satellite networks are expected to provide global communication services, they still face many new challenges. First, many latency-sensitive applications such as vehicle networking require lower latency, and many emerging services such as vehicle networking and artificial intelligence require more and more computing and storage resources. Due to physical factors such as volume and mass, LEO satellites cannot carry a large amount of resources. At the same time, there are a large number of differentiated service needs of users in the coverage area of low-orbit satellites, and LEO satellites need to realize differentiated service customization and processing for a large number of user access. However, in most existing research, LEO satellite nodes are generally used as relays for communication between ground cloud servers and users to realize data service transmission. Even in the edge computing scenario, LEO satellites do not classify user services differently when they are used as computing nodes for user services, but uniformly process user services, which cannot realize reasonable allocation of LEO satellite resources for different services. This will make it difficult to effectively utilize various resources of LEO satellites and meet the differentiated needs of users.

[0005] At the same time, LEO satellite networks face huge challenges in computing, storage, bandwidth and other resources in resource-constrained space environments. Although low-orbit satellites can provide extensive coverage, their limited hardware resources make task scheduling and resource allocation more complex. Existing research has focused on collaborative resource optimization between LEO satellites and task scheduling methods under a multi-layer satellite architecture. These methods provide effective solutions to some extent. However, in large-scale LEO satellite networks, how to achieve differentiated service customization according to the different needs and business intentions of users is still a problem that has not been fully solved. In particular, in terms of task processing and resource scheduling, how to perform personalized scheduling according to user service priorities, timeliness requirements, and optimize network performance under limited resources is still a difficult point in current research. How to achieve efficient and customized service delivery under resource constraints is one of the key technical challenges in LEO satellite networks that needs to be overcome. SUMMARY

[0006] In view of this, the embodiments of the present application provide a task scheduling method and system for low-orbit satellite edge computing networks to eliminate or improve one or more defects in the prior art.

[0007] One aspect of the present application provides a task scheduling method for a low-orbit satellite edge computing network, the low-orbit satellite edge computing network is divided into multiple subgroups, each subgroup elects a low-orbit satellite as an auxiliary control node, and the remaining low-orbit satellites in the subgroup are used as computing nodes, the method comprises the following steps: after the low-orbit satellite edge computing network receives a user service request, the auxiliary control node identifies the user service intention corresponding to the user service request according to an intention perception mechanism; wherein the user service intention includes some or all of the dimensions of service name, resource type, data volume and service performance requirement; the auxiliary control node identifies the current network service preference of each low-orbit satellite node in the corresponding subgroup according to a network service preference perception mechanism; wherein the network service preference includes some or all of the dimensions of residual energy, residual CPU resource, residual GPU resource, network load and network reliability; determine the historical data item stored in the weight history database that has the highest similarity to the current user service request in terms of user service intention and network service preference, generate a task scheduling strategy based on the weight coefficient of the historical data item with the highest similarity; wherein each historical data item of the weight history database includes the user service intention, network service preference, user satisfaction and weight coefficient for task scheduling corresponding to a historical user service request; the auxiliary control node distributes the corresponding user service to the corresponding computing node according to the task scheduling strategy to realize task processing.

[0008] In some embodiments of the present application, the step of the auxiliary control node identifying the user service intention corresponding to the user service request according to the intention perception mechanism comprises: the auxiliary control node extracts the user service intention primitive from the user service request by using a pre-trained natural language processing model, performs word analysis and / or semantic mining on the user service intention primitive by using a pre-trained intention translation model, and converts the user service intention primitive into a user service intention conforming to a pre-set RDF framework; wherein the record form of the user service intention is RDF=<service name, resource type, data volume, service performance requirement>, and the record form of the network service preference is NSI=<residual energy, residual CPU resource, residual GPU resource, network load, network reliability>.

[0009] In some embodiments of the present application, the step of determining the historical data item stored in the weight history database that has the highest similarity to the current user service request in terms of user service intention and network service preference comprises: calculating the similarity scores of the current user service request in terms of user service intention and network service preference with each historical data item in the weight history database by using a cosine similarity calculation method; constructing a decision problem by using an analytic hierarchy process to obtain the historical data item with the highest similarity score as the target of the decision problem, and obtaining the historical data item with the highest similarity score.

[0010] In some embodiments of the present application, the analytic hierarchy process constructs a hierarchical decision problem into a target layer, a criterion layer and a scheme layer; the target layer includes a weight coefficient of an objective function for quantitatively evaluating user satisfaction and / or network benefit; wherein the objective function includes some or all of the following indicators: latency, cost, accuracy, energy consumption and load balancing; the criterion layer includes the evaluated user satisfaction and the similarity score; and the scheme layer includes each historical data item in the weight history database.

[0011] In some embodiments of the present application, before starting to process the user service request, the method further includes: constructing the weight history database in advance; and after the secondary control node issues the corresponding user service to the corresponding computing node according to the task scheduling strategy to implement task processing, the method further includes: evaluating the user satisfaction of processing the corresponding user service based on user feedback, and saving the user service request, the corresponding user service intention, network service preference, user satisfaction and weight coefficient for task scheduling as a historical data item of the weight history database.

[0012] In some embodiments of the present application, the method further includes: using a ground cloud center controller located on the ground to perform ground center cloud management and control, receiving historical data items from the low-orbit satellite edge computing network, and updating and maintaining the weight history database.

[0013] In some embodiments of the present application, the weight coefficient of the historical data item includes a user service intention weight and a network service preference weight, and the step of generating a task scheduling strategy based on the weight coefficient of the historical data item with the highest similarity includes: generating a double-weight task scheduling strategy for the user service intention weight and the network service preference weight by using a heuristic algorithm; wherein the heuristic algorithm includes a mayfly algorithm, an ant colony optimization algorithm, a whale optimization algorithm and a firefly algorithm.

[0014] Corresponding to the above method, the present application also provides a task scheduling system for a low-orbit satellite edge computing network, which includes a processor, a memory and a computer program / instruction stored in the memory, and the processor is used to execute the computer program / instruction, and when the computer program / instruction is executed, the system implements the steps of the method according to any one of the above embodiments.

[0015] Corresponding to the above method, the present application also provides a computer readable storage medium having a computer program / instruction stored thereon, and the computer program / instruction is executed by a processor to implement the steps of the method according to any one of the above embodiments.

[0016] Corresponding to the above method, the application also provides a computer program product comprising computer programs / instructions which, when executed by a processor, implement the steps of the method according to any one of the above embodiments.

[0017] The task scheduling method and system for the low-orbit satellite edge computing network can dynamically generate a task scheduling weight and dynamically optimize a task allocation strategy by deeply analyzing a service intention and a network service preference of a user, and preferentially allocate tasks to the most matched satellite node, which is beneficial to improving the accuracy of task scheduling and resource utilization efficiency, reducing task processing delay, optimizing the overall performance of the network, and improving user experience.

[0018] Additional advantages, objects, and features of the application will be set forth in part by the description that follows, and will become apparent to those skilled in the art upon examination of the following detailed description and drawings in which:

[0019] Those skilled in the art will appreciate that the objects and advantages of the application can be implemented without regard to the specific details of the following description and that the application can be implemented in various ways, as will be apparent to persons skilled in the art from the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate the present application and, together with the description, further serve to explain the principles of the application and to enable a person skilled in the relevant art(s) to make and use the application. Figure 1 Flow chart of the task scheduling method for the low-orbit satellite edge computing network in an embodiment of the application.

[0021] Figure 2 Inter-satellite cooperative edge computing system of prior art 1.

[0022] Figure 3 Satellite edge computing system framework of prior art 2.

[0023] Figure 4 Cluster division management system framework for low-orbit satellites in an embodiment of the application.

[0024] Figure 5 Adaptive weight generation mechanism schematic diagram in an embodiment of the application.

[0025] Figure 6 Flow chart of the task scheduling method for the low-orbit satellite edge computing network based on a heuristic algorithm in an embodiment of the application. DETAILED DESCRIPTION

[0026] For the purposes of the present application, the technical solutions and advantages will be made more clear, below, combined with embodiments and drawings, the present application is further described in detail. Here, the illustrative embodiments of the present application and its description are used to explain the present application, but not as a limitation of the present application.

[0027] Here, it also needs to be explained that, in order to avoid unnecessary details, the present application is obscured, only the structure and / or processing steps closely related to the scheme according to the present application are shown in the drawings, and other details not related to the present application are omitted.

[0028] It should be emphasized that the term "comprise / comprising" is used herein to refer to the presence of a feature, element, step or component, but not to exclude the presence or addition of one or more other features, elements, steps or components.

[0029] Here, it also needs to be explained that, if not specially explained, the term "connection" can not only mean direct connection, but also mean indirect connection with the presence of intermediate.

[0030] In the following, the embodiments of the present application will be described with reference to the accompanying drawings. In the drawings, the same reference signs represent the same or similar parts, or the same or similar steps.

[0031] Figure 2 The inter-satellite cooperative edge computing system of the prior art 1. As Figure 2As shown, the related prior art studies the task scheduling and resource allocation problem in the satellite edge computing (SEC) system based on inter-satellite link (ISL). In view of the problem that the uneven distribution of ground users leads to insufficient computing capacity of a single satellite in the high traffic area, an inter-satellite cooperative edge computing scheme is proposed. By establishing an optimization model, the computing resource allocation and user task segmentation are jointly optimized, part of the task is unloaded to the directly connected satellite, and part is transmitted to other satellites through the inter-satellite link for cooperative processing, so as to minimize the system weighted energy consumption. The research adopts an alternating iterative optimization algorithm, decomposes the non-convex problem into multiple sub-problems, and solves it by using the Lagrange dual method and linear programming. The simulation results show that the scheme can effectively reduce the computing burden in the high traffic area, improve the utilization efficiency of satellite computing resources, and significantly reduce the system energy consumption. The satellite edge computing system is composed of LEO satellites and ground user equipment, forming a distributed and cooperative network architecture. The satellites connected by the inter-satellite link not only can provide computing services for the ground user equipments (ground UEs), but also can cooperate with other satellites in tasks, divide the high traffic area and the low traffic area, unload the computing between the high traffic area and the low traffic area, realize the dynamic allocation of computing resources and load balancing. The introduction of this system architecture aims to improve the utilization efficiency of satellite computing resources and the overall performance of the system, and provides a new solution to cope with the computing demand of the high traffic area and reduce energy consumption.

[0032] The disadvantages of Prior Art One are: (1) Prior Art One proposes a satellite edge computing system that collaboratively schedules through inter-satellite links. This system optimizes the allocation of computing resources and task offloading through the cooperation between satellites. Although this solution can effectively reduce system energy consumption and improve resource utilization in theory, it may face many challenges in actual deployment. For example, the high-speed movement and orbital changes of satellites can cause instability of communication links, and the complexity of the system can increase the implementation cost and maintenance difficulty, especially in the scenario of multi-satellite cooperation and dynamic task allocation. These factors may limit the scalability and practical application of the system. (2) Prior Art One mainly focuses on the task cooperation and resource allocation between satellites through inter-satellite links, but in a large-scale satellite network, task scheduling and resource management may face many challenges, such as real-time task allocation, dynamic changes in satellite computing capacity, and adjustment of task priority. The solution does not fully discuss these challenges and corresponding solutions, which are crucial for ensuring the efficient operation of the system in high-traffic areas. The solution also does not discuss in detail how to implement efficient load balancing strategies in a dynamically changing satellite network, which is crucial for optimizing system performance and improving resource utilization. (3) Although Prior Art One optimizes the allocation of computing resources through inter-satellite links, it does not consider the problem of insufficient computing resources that the satellite edge computing system may face when facing sudden high-traffic demand. Therefore, how to maintain the stability and reliability of the system under extreme conditions is a problem that needs further discussion.

[0033] Prior Art Two proposes a Deep Reinforcement Learning (DRL) based satellite edge computing task offloading mechanism, aiming to optimize the allocation of computing resources in LEO satellite networks to achieve low-latency computing services. This mechanism solves the problem of dynamic nature and uneven resource distribution in LEO satellite networks by allowing LEO satellites with limited computing resources to offload part of the tasks to nearby satellites with sufficient resources. Specifically, this mechanism utilizes the dynamic topology and resource variation characteristics of LEO satellite constellations, models the utility function of task offloading, dynamically selects offloading target satellites and allocates computing resources, thereby maximizing the overall utility of task offloading.

[0034] Figure 3 The satellite edge computing system framework of Prior Art Two is shown in FIG. 1. As shown in FIG. 1, the system includes a plurality of LEO satellites 101, a ground control center 102, and a user terminal 103. The LEO satellites 101 are connected to each other through inter-satellite links 104, and the ground control center 102 is connected to the LEO satellites 101 through a ground link 105. The user terminal 103 is connected to the LEO satellites 101 through a user link 106. Figure 3As shown, the prior art two system consists of low earth orbit satellites and ground stations, forming a multi-level, collaborative network structure. LEO satellites not only provide communication services for ground user terminals, but also achieve offloading and collaborative processing of computing tasks through inter-satellite links. The introduction of this system architecture aims to improve the utilization of limited computing resources of LEO satellites, while enhancing the dynamic adaptability and flexibility of the system, providing a new research approach for efficient computing resource allocation and task offloading in satellite networks.

[0035] The disadvantages of prior art one are: (1) Prior art two proposes a task offloading mechanism based on deep reinforcement learning, which is used to optimize the allocation of computing resources in LEO satellite networks. However, this system needs to perform complex task allocation and resource management in dynamic LEO satellite networks, which may face major challenges in actual deployment. Including real-time communication coordination between satellites, rapid adaptability of dynamic network topology, and the complexity of implementing efficient computing offloading on resource-limited satellite platforms. Due to the fast-moving characteristics of LEO satellites, the system needs to quickly adapt to the changing satellite positions and communication link states, which may put pressure on the stability and reliability of the system. (2) Prior art two does not fully discuss how to deal with the challenges of high latency, bandwidth limitations of inter-satellite links, and data transmission during the task offloading process, and does not detail how to implement efficient load balancing and resource allocation strategies in dynamic environments, which is crucial for optimizing task offloading performance and improving resource utilization. (3) Although LEO satellite networks provide a new solution for computing offloading, it still relies on ground infrastructure such as ground stations and data centers. Prior art two may not fully consider the vulnerability of these ground infrastructure, such as natural disasters, technical failures or other unforeseen events that may cause damage or failure of ground stations, affecting the operation of the entire network. How to maintain the continuity and resilience of the network when the ground infrastructure is damaged is a problem that needs further exploration.

[0036] In recent years, the International Telecommunication Union (ITU) has proposed a computing power network (CPN) architecture, a network-centric architecture that connects widely distributed computing resources to provide services to users. With the continuous development of advanced technologies such as satellite Internet and computing power networks, many constructive solutions have been provided for the development of satellite computing power networks.

[0037] This invention proposes a method for cluster partitioning of large-scale satellite networks by effectively sensing resources within the satellite computing network, and for joint optimization scheduling of multi-resources for edge satellite nodes by combining differentiated service intentions of ground users. Drawing on satellite computing networks, this invention proposes a novel task scheduling method and system for low-Earth orbit (LEO) satellite edge computing networks. It implements cluster partitioning for LEO satellite networks, senses different service requirements to implement task scheduling schemes, and optimizes resource allocation while meeting the differentiated needs of users.

[0038] Currently, the application of heuristic algorithms for service intent perception and scheduling in LEO satellite networks is still in its early stages. Further improvements to intent-based scheduling methods in LEO satellite networks to achieve efficient scheduling of diverse services remain unexplored. Therefore, this invention focuses on the scheduling framework in LEO satellite networks, designs an adaptive weight generation method for the service intent perception mechanism, and provides a task scheduling algorithm to address the problem of service intent perception matching based on satellite network status, thus realizing a strategy for service intent perception scheduling in LEO satellite networks.

[0039] Figure 1 This is a flowchart of a task scheduling method for a low-Earth orbit (LEO) satellite edge computing network according to an embodiment of the present invention. The LEO satellite edge computing network is divided into multiple subgroups. Each subgroup selects one LEO satellite as an auxiliary control node, and the remaining LEO satellites within the subgroup serve as computing nodes. The auxiliary control node acts as the execution entity. The method includes the following steps: Step S110: After the low-orbit satellite edge computing network receives a user service request, the auxiliary control node identifies the user service intent corresponding to the user service request according to the intent awareness mechanism; wherein, the user service intent includes some or all of the dimensions of service name, resource type, data volume and service performance requirements.

[0040] The user's business intent can be recorded in the form of RDF=<business name, resource type, data volume, service performance requirements>.

[0041] Low-Earth orbit (LEO) satellite edge computing networks typically receive user service requests and related files from terminal devices via LEO satellites. They then use pre-trained NLP models to extract user service intent primitives from these requests and files. Finally, a pre-trained intent translation model is used to transform these primitives into user service intents that conform to a pre-defined RDF framework.

[0042] Step S120: The secondary control node identifies the current network service preference of each low-orbit satellite node in the corresponding subgroup according to the network service preference perception mechanism; wherein the network service preference includes some or all of the dimensions of residual energy, residual CPU resource, residual GPU resource, network load and network reliability.

[0043] Wherein, the record form of network service preference can be NSI = <residual energy, residual CPU resource, residual GPU resource, network load, network reliability>.

[0044] Step S130: Determine the historical data item stored in the weight history database that has the highest similarity to the current user service request in terms of user service intent and network service preference, and generate a task scheduling strategy based on the weight coefficient of the historical data item with the highest similarity; wherein each historical data item of the weight history database includes the user service intent, network service preference, user satisfaction and weight coefficient for task scheduling corresponding to a historical user service request Wherein, step S130 can be completed by a task scheduling model pre-deployed in the ground cloud center controller or the secondary control node.

[0045] Wherein, in order to make the task scheduling strategy generated by the task scheduling model optimal, a target function for quantitatively evaluating user satisfaction and network benefit is designed, and the optimal task scheduling strategy is found by minimizing the target function, which includes some or all of the following indicators: latency, cost, accuracy, energy consumption and load balancing.

[0046] Step S140: The secondary control node distributes the corresponding user service to the corresponding computing node according to the task scheduling strategy to realize task processing.

[0047] The task scheduling method and system for low-orbit satellite edge computing network proposed in the application can dynamically generate task scheduling weights and dynamically optimize task allocation strategies by deeply analyzing user service intent and network service preference, and preferentially allocate tasks to the most matched satellite nodes. This method is beneficial to improve the accuracy of task scheduling and resource utilization efficiency, reduce task processing delay, optimize the overall performance of the network, and improve user experience.

[0048] In some embodiments of the present application, before the step S110, a process of pre-modeling the low-orbit satellite edge computing network is further included: the low-orbit satellite edge computing network includes three components in architecture, i.e., a terminal device, a LEO computing node (i.e., a low-orbit satellite in a LEO constellation) and a ground cloud center controller; cluster division is performed for a low-earth orbit satellite network (also referred to as a LEO satellite network); the cluster division mechanism is: a cluster division mechanism combining geography and logic is adopted, preliminary division is first performed based on the geographical position (physical distance) and orbit similarity of the satellite, and a fuzzy clustering algorithm is then used to further subdivide into logical subgroups. After the logical subgroup division is completed, a backup control election mechanism based on the Paxos algorithm is adopted, each logical subgroup automatically elects a backup control node according to the computing performance, communication load and task demand of the satellite, which is responsible for coordinating task allocation and resource management, and communicates with the backup control nodes of other subgroups. The backup control node senses the resource status in real time and feeds back to the master control node to support global optimization and scheduling decision. Even if the backup control node fails, the system can automatically trigger the election process to ensure scheduling continuity and cluster stability.

[0049] In some embodiments of the present application, the step of the backup control node identifying the user service intent corresponding to the user service request according to the intent perception mechanism includes: the backup control node extracts user service intent primitives from the user service request by using a pre-trained natural language processing model, performs word analysis and / or semantic mining on the user service intent primitives by using a pre-trained intent translation model, and converts the user service intent primitives into a user service intent conforming to a preset RDF framework.

[0050] The record form of the user service intent is RDF=<service name, resource type, data volume, service performance requirement>, and the record form of the network service preference is NSI=<remaining energy, remaining CPU resource, remaining GPU resource, network load, network reliability>.

[0051] By using the embodiments of the present application, the service intent of the user, such as delay sensitivity, computing demand, cost limit, etc., can be deeply analyzed, the task allocation strategy can be dynamically optimized by continuously sensing and adjusting according to the user intent in real time, and the user experience can be targetedly improved.

[0052] In some embodiments of the present application, the step of determining the historical data item in the weight history database that is most similar to the current user service request in terms of user service intention and network service preference comprises: (1) using a cosine similarity calculation method to calculate the similarity score of the current user service request in terms of user service intention and network service preference with each historical data item in the weight history database; and (2) using an analytic hierarchy process to construct a decision problem to obtain the historical data item with the highest similarity score as the target of the decision problem, thereby obtaining the historical data item with the highest similarity score.

[0053] By using the weight generation mechanism based on historical data and AHP, the weight configuration can be dynamically adjusted under different network states and task demands, which is beneficial to reducing unnecessary resource waste, dynamically adjusting the weight configuration under different network states and task demands, and significantly improving the adaptability and flexibility of task scheduling.

[0054] In some embodiments of the present application, the hierarchy of the decision problem constructed by the analytic hierarchy process comprises a target layer, a criterion layer and a scheme layer. The functions of each layer are as follows: (1) the target layer includes the weight coefficient of the objective function, and the objective function is used to quantitatively evaluate user satisfaction and / or network benefit; wherein the objective function includes some or all of the following indicators: time delay, cost, accuracy, energy consumption and load balancing; (2) the criterion layer includes the evaluated user satisfaction and the similarity score; and (3) the scheme layer includes each historical data item in the weight history database.

[0055] After obtaining the weight, the historical data can be weighted to obtain the historical data item with the highest similarity score, and the historical weight recorded by the historical data item is selected as the adaptively generated weight. If there are historical data items with the same similarity score, the recorded weights can be averaged to obtain the adaptively generated weight.

[0056] By using the dynamic weight generation mechanism based on historical data and the analytic hierarchy process AHP, the weight configuration can be dynamically adjusted under different network states and task demands, which significantly improves the adaptability and flexibility of task scheduling, reduces task processing delay, reduces the pressure on resources, realizes more efficient resource management, optimizes network performance, and improves user experience.

[0057] In some embodiments of the present application, before starting to process the user service request, the method further comprises: constructing the weight history database in advance.

[0058] In some embodiments of the present application, after the secondary control node distributes the corresponding user service to the corresponding computing node according to the task scheduling strategy to realize task processing, the method further comprises: evaluating the user satisfaction of processing the corresponding user service based on user feedback, and saving the user service request, the corresponding user service intention, the network service preference, the user satisfaction and the weight coefficient for task scheduling as a historical data item of the weight history database.

[0059] Exemplarily, a way of constructing a history database is given: the history database stores a tuple data type, which can be represented as , represents the satisfaction evaluation of the user service, represents the network status of the cluster where the user service is located (including the remaining resources , the congestion of the network and the load of the network ), represents the weight coefficient for task scheduling.

[0060] The user satisfaction can be quantitatively evaluated by a target function in the execution of the scheduling decision.

[0061] By using the embodiments of the present application, the current situation can be analyzed based on historical data, the most similar historical data item can be quickly obtained according to the historical data, and the most similar historical data item can be quickly applied, thereby reducing unnecessary waste of computing resources and facilitating to ensure the efficiency of task scheduling and improve user experience.

[0062] In some embodiments of the present application, the method further comprises: using a ground cloud center controller located on the ground to perform ground center cloud management and control, receiving historical data items from a low-orbit satellite edge computing network, and updating and maintaining the weight history database.

[0063] By using the embodiments of the present application, the updating and maintenance of the model and the database can be realized with the help of the ground cloud center controller located on the ground.

[0064] In some embodiments of the present application, the weight coefficient of the historical data item includes a user service intention weight and a network service preference weight, and the step of generating a task scheduling strategy based on the weight coefficient of the historical data item with the highest similarity comprises: generating a double-weight task scheduling strategy for the user service intention weight and the network service preference weight by using a heuristic algorithm. The heuristic algorithm includes mayfly algorithm, ant colony optimization algorithm, whale optimization algorithm and firefly algorithm.

[0065] Further, in some embodiments of the present application, the preference of the network for different demands can be set according to the current network service preference (i.e., network state), for example, if the network load is large, the weight preference for the load can be set to be smaller.

[0066] With the embodiments of the present application, the global optimization capability of the heuristic algorithm can overcome the low efficiency problem of the traditional task scheduling method in a complex network environment, and enhance the adaptability and scalability of the scheme. The method can realize adaptive adjustment in a dynamically changing network environment, and ensure the optimization effect of task scheduling.

[0067] To improve the processing capability of LEO satellite network for diversified user service requests, the present application focuses on solving the following three technical problems: (1) Research on how to realize efficient cluster division and management in LEO satellite network. With the increase of the number of LEO satellites, the traditional centralized management method is difficult to meet the efficient operation demand of large-scale satellite network. Through the cluster division mechanism combining geography and logic, the satellites are first preliminarily divided based on their geographical position and orbital similarity, and then further subdivided into logical subgroups by using fuzzy clustering algorithm, so as to optimize resource allocation and management. This mechanism can not only reduce the management complexity, but also improve the flexibility and efficiency of resource scheduling. (2) Further focus on service intention perception and differentiated service customization. LEO satellite network needs to support diversified services, but existing research lacks customization of scheduling for user differentiated demand. The present application converts the user's service intention into a standardized resource description framework (RDF) through natural language processing and semantic analysis technology, dynamically generates weights by using historical data and analytic hierarchy process (AHP), and optimizes the task scheduling strategy. Combined with the network preference perception mechanism, the task scheduling strategy is dynamically adjusted to adapt to different network environments, so as to realize efficient service customization and resource optimization. (3) A double-weight task scheduling algorithm based on service intention perception and network preference perception is proposed, aiming at solving the dynamic task scheduling problem in LEO satellite network. The algorithm combines user intention weight and network preference weight, and realizes dynamic scheduling of tasks through heuristic algorithm (such as Mayfly algorithm). This method can not only improve the resource utilization efficiency, but also enhance the adaptability and flexibility of the system, so as to realize efficient task processing and resource management in dynamic network environment.

[0068] In some embodiments of the present application, a scheme for chain content pre-placement using heuristic algorithms in low-orbit satellite edge computing networks is designed. The scheme includes: (1) a LEO satellite network cluster division and management scheme; (2) an adaptive weight generation method based on service intention perception and network preference perception; (3) a service scheduling method based on heuristic algorithms.

[0069] (1) The LEO satellite network cluster division and management scheme is as follows: The coverage of ground stations is not global, and in remote areas and marine environments, the communication between LEO satellites and ground stations (Ground Station) may be temporarily interrupted. In the environment of high-speed satellite motion and frequent switching, the system needs to have stable network connection management capability. The star-ground system is composed of different types of satellite nodes and ground equipment resources with different orbital heights to form a heterogeneous computing environment, and the system needs to have the ability of cooperative computing in a heterogeneous environment to realize efficient execution and processing of complex applications. At the same time, the system needs to ensure that it can quickly establish a connection with the terminal device in a high dynamic environment. Based on the above design principles, the architecture of the system is as shown in Figure 4 Figure 4 is the system framework for cluster division and management of low-orbit satellites in an embodiment of the present application. The entire architecture is composed of three parts, i.e., terminal devices, LEO computing nodes, and ground cloud center controllers.

[0070] (1) Terminal device The terminal device (also known as user terminal) is composed of various devices, including Internet of Things sensors, user mobile devices, and ground stations, etc. These devices can collect data in different environments and scenarios and generate computing tasks that need to be processed. Although the terminal user device has certain computing capability, it is limited by physical limitations such as energy consumption, computing resources, and storage capacity, and cannot independently complete all computing tasks. Therefore, these devices often need to schedule and offload part of the computing tasks to LEO satellites for processing. By uploading the computing tasks to the satellite for processing, the terminal user device can efficiently utilize the computing resources of the satellite to quickly complete data processing and analysis tasks, and improve data processing efficiency and response speed.

[0071] (2) LEO computing node The LEO computing node (i.e., LEO satellite for processing computing tasks) has on-orbit computing capability by deploying edge computing servers. Considering the relatively limited computing resources of the satellite, for large-scale data processing and computing tasks, the system needs to perform effective task scheduling to fully utilize the computing resources of each satellite in the constellation. The LEO computing node can perceive the task demand of the user terminal, upload the scheduling result after translating the service intention to the ground cloud center controller, and select the optimal scheduling node for task processing.

[0072] ​(3) Ground cloud center controller The ground cloud center controller assumes the role of ISIP-TS (Inter-Satellite IP Transport Stream) system management control, responsible for comprehensive monitoring and perception of resources. The ground cloud center controller has strong data processing and storage capabilities, capable of supporting large-scale task decision-making and data management. At the same time, through data interaction with the LEO edge computing satellite constellation, the ground cloud center controller can implement resource access management, task scheduling, service orchestration, and other key decision-making of the system, achieving global control and optimized management.

[0073] With the rapid growth of the number of satellites in the LEO edge computing satellite constellation and the diversification of user demand, a single centralized management method has been difficult to meet the efficient and stable network operation demand. In order to cope with the complexity and high dynamics of large-scale low-orbit satellite networks, the present invention proposes an efficient cluster division and auxiliary control selection mechanism, which divides the satellite constellation into multiple dynamically adjusted functional clusters, and selects auxiliary control nodes with optimal computing capability and communication load in each cluster to achieve more efficient task scheduling and resource management within the cluster, significantly improving the efficiency and stability of the system.

[0074] The content of the adaptive weight generation method based on service intention perception and network preference perception is as follows: This scheme first needs to realize the description of the optimization index, and then realize the generation of adaptive weight by adopting the service intention perception mechanism, and realize the business scheduling weight acquisition through the network service preference perception mechanism.

[0075] (1) Set optimization index In the above task scheduling problem based on business intention, the two main bodies of scheduling are business service and LEO satellite computing node. For business service, the satisfaction attribute of service to delay, cost, etc. can be expressed in business intention. For LEO satellite computing node, there are energy consumption of computing node and load balancing performance attribute. Among all the LEO satellite candidate computing nodes, each computing node has different energy consumption, load condition, and each user business has different delay, cost demand. The scheduling of business needs to ensure the matching of user intention and network node, and realize the maximization of user business satisfaction and network benefit index.

[0076] In order to more clearly express the optimization index evaluation in the scheduling process, the target function can be used to quantitatively evaluate user satisfaction and network benefit in the execution of scheduling decision. In order to find the highest satisfaction score, the optimal scheduling decision needs to be found to minimize the target function. The target function is composed of five indexes of delay D, cost C, accuracy A, energy consumption E and load balancing LB.

[0077] (2) Service intention awareness The intention of user service can determine the willingness of users to purchase satellite computing power network computing services. The abstract meaning of intention is that the object wants to do something to achieve a certain effect, which is a high-level abstract form of description of the user's service expression requirements. In the satellite computing power network task scheduling process, different tasks have different service intentions, and user services exist in multiple categories, which can be divided into delay-sensitive, cost-sensitive, and computing-intensive. Service intention types can be classified in multiple ways, classified according to the quality of service that the satellite computing power network can provide, such as delay intention, bandwidth intention, etc., classified according to the subjective requirements of the task, such as reliability intention, cost intention, etc. Business intention usually indicates the subjective expectations that want to be realized in the intention primitive.

[0078] In order to abstract the business intention in the satellite computing power network task scheduling process into the weight of the optimization index, first, the business intention of the user is obtained. After the computing node receives the incoming task and obtains the file and perception data, the original natural language text is converted into standardized intention primitives (words that describe task requirements, such as cost, time delay, accuracy, etc.) based on existing intention dictionaries through natural language processing technology (such as data cleaning, text separation, and feature extraction). After obtaining the standardized intention primitives, the intention requirements of the task are converted into weight configurations in the optimization target, and this process is called intention translation. Using a pre-trained intention translation model, with the help of word analysis and semantic mining, the intention primitive can be converted into a specific RDF, which can be represented as <business name, resource type, data volume, service performance requirement>. Based on the RDF of the service intention, the service optimization index can be determined according to the service performance requirement in the RDF, and the corresponding node selection algorithm can be used to match the service intention with the satellite node, which can complete the conversion of the service intention to the task scheduling strategy.

[0079] However, only according to the service intention perception mechanism, different types of weight framework matching can be generated for business categories, but cannot produce differentiated customized weights according to different network resource environments and task service quantities. The present application can obtain differentiated weight configurations according to matching data by matching RDF and network conditions through a historical database. The present application uses cosine similarity to calculate the similarity of RDF and network service preference (network state B in some embodiments) between data, to reflect the similarity of data in the historical database and the current business and network environment. At the same time, the present application generates user business weights based on historical data through the analytic hierarchy process (AHP), which is a method for decision analysis and can help determine the weights of different indicators. For RDF similarity, network condition similarity and user satisfaction scoring, the generation of weights of the three can be realized through the analytic hierarchy process. After obtaining the weights, the entry with the highest similarity score is obtained by weighting the historical data, and the recorded historical weight is taken as the adaptively generated weight. If there are historical data with the same similarity score, the recorded weights are averaged.

[0080] Figure 5 An adaptive weight generation mechanism for an embodiment of the present application is shown in the figure. The adaptive generation process of the weight coefficient is shown in Figure 5 . , , and represent the corresponding weight coefficients of user business. For each user business satellite computing node, there is a weight history database, and the historical database stores tuple data types, represented as . represents the satisfaction evaluation of user business. After the business scheduling processing result is returned to the user, the user performs satisfaction scoring evaluation and records . represents the network condition of the cluster where the user business is located, including the remaining resources , the congestion of the network , the load of the network , etc. Therefore, the network condition can be represented as .

[0081] (3) Network preference perception The service intent of the user will change with the continuous change of the task, and the network condition is not constant. When the network receives many user services, the service preference of the network will change with the different service intents of the user services. That is, not only the service has a service intent, but also the network has a service intent. The present application has assumed that the service intent RDF of the user can be expressed as RDF = <service name, resource type, data volume, service performance requirement>, and the network service preference (NSI) can be expressed as NSI = <remaining energy, remaining CPU resource, remaining GPU resource, network load, network reliability>. These indicators can represent the current service preference of the low-orbit satellite edge computing network, and can be compared with the service performance requirement in the RDF. The cosine similarity can also be used to represent the adaptability of the current network condition to the scheduled task.

[0082] Among them, the remaining energy refers to the remaining energy of the satellite node in the low-orbit satellite edge computing network. Similarly, the remaining CPU resource, the remaining GPU resource, the network load and the network reliability are all descriptions of the current network service performance or available resources of each satellite node.

[0083] According to all the perceived services in the cluster, the network service preference perception mechanism can calculate the service weight, so as to select the service preferred by the network to achieve better scheduling processing. Assuming that there are N user services in the cluster that need to be scheduled, the cosine similarity can be used to calculate the similarity between the RDF of each service i and the NSI, so that a series of similarity data can be obtained By normalizing the similarity, the normalized weight can be obtained. Thus, the preference of the network for different demands can be set according to the current network state, and different service weights can be set for different services.

[0084] (Three) The content of the service scheduling method based on heuristic algorithm is as follows: Reasonable task scheduling is crucial for the arrangement and management of LEO edge computing satellite constellation and the satisfaction of user task requirements. In the scheduling decision process, different tasks have different optimization objectives. As shown in Figure 3 The task scheduling framework of the system mainly includes intent perception, adaptive weight generation, task scheduling processing, satisfaction feedback link for execution, and the process of the service scheduling based on intent mainly includes the following four steps.

[0085] Step 1: User service request. User services such as vehicle networking service, artificial intelligence service and intelligent survey service in remote areas are sent by the user to the low-orbit satellite for service processing.

[0086] Step two: intention perception. Intention perception includes two parts: business intention perception and network service preference perception. In the business intention perception part, the system obtains the user business intention primitives and converts them into RDF, and then generates adaptive weights for optimization targets such as delay, cost and energy consumption. In the network service preference perception part, the system sets the preference of the network for different demands according to the current network state, for example, if the network load is large, the weight preference for load will be smaller. By comparing the business intention with the network service preference, different business scheduling weights are set for different businesses. Finally, according to the user business intention and network service preference of the current user business request, similar data is searched from the historical database, the similarity score is calculated by cosine similarity, and the optimal weight coefficient is selected. The weight coefficient of the data with the highest similarity score is used as the optimization target weight coefficient.

[0087] Step three: task scheduling. The secondary control node makes scheduling decisions by selecting scheduling algorithms from the algorithm library and issuing them to the task queue of each computing node. The business is forwarded to the corresponding computing node according to the scheduling strategy generated by the algorithm library to implement task processing.

[0088] Step four: satisfaction feedback and storage of business information. After the implementation of task processing, the processing result is returned to the user task, and the final weight is assigned to the user task for intention verification. The satisfaction is evaluated through user feedback, and the weight coefficient and satisfaction are stored in the historical database as historical data for the next scheduling period.

[0089] Figure 6 The flow chart of the task scheduling method for the low-orbit satellite edge computing network based on the heuristic algorithm in an embodiment of the present application. The selection and design of the task scheduling algorithm is a crucial process.

[0090] As Figure 6As shown, the scheduling algorithm aims to optimize resource utilization and improve task execution efficiency by reasonably allocating and scheduling tasks when facing dynamic business demands and complex application scenarios. The scheduling algorithm needs to optimize resource allocation for different satellite nodes based on the characteristics of tasks such as computing requirements, timeliness, etc. The optimization goal of the scheduling algorithm is usually to maximize the utility value of the task. In actual operation, the task scheduling problem often involves a large number of feasible solutions, and the best solution needs to be explored. The common solution idea is to use heuristic algorithms to optimize the scheduling strategy. Heuristic algorithms (such as mayfly algorithm) simulate the interaction between individuals and groups, gradually update the population and find the global optimal solution. Specifically, the algorithm initializes a certain number of individuals to represent task scheduling schemes, evaluates the fitness of each scheme, and simulates the updating process of individual positions. Through competition and cooperation within the group, the algorithm gradually improves the solution, and finally selects the scheduling scheme with the best fitness to achieve the optimal resource allocation and scheduling effect. Through this scheduling algorithm, the system can efficiently handle task scheduling problems, especially suitable for dynamic and complex task scheduling scenarios, meet real-time requirements and business demands, and improve the overall system performance.

[0091] In some embodiments of the present application, the cluster division and auxiliary control selection mechanism proposed by the present application includes the following main steps: (1) The first cluster division is based on the geographical position of the satellite.

[0092] Considering the physical distance between satellites and the similarity of their orbits. By this method, satellites with similar orbits and close in space can be divided into the same cluster. This reduces the communication delay between members within the cluster and enhances the efficiency of data transmission.

[0093] (2) On the basis of geographical position division, use fuzzy clustering algorithm to realize the second division of logical subgroups.

[0094] First, by evaluating the CPU performance, memory, bandwidth and other resource characteristics of the satellite, a comprehensive performance feature vector is constructed. Then, based on these performance data and the preliminary cluster, the fuzzy clustering algorithm is applied to divide the satellites into different subgroups. This helps to ensure the optimal configuration of resources within the cluster and improves the ability of satellites to cooperatively process similar tasks.

[0095] (3) Paxos Algorithm-Based Auxiliary Control Node Election. Each subgroup automatically elects an auxiliary control node based on satellite computing performance, communication load, and mission requirements. This auxiliary control node is responsible for coordinating task allocation and resource management, and communicating with auxiliary control nodes of other subgroups. The auxiliary control node monitors resource status in real time and feeds it back to the master control node to support global optimization and scheduling decisions. Even if an auxiliary control node fails, the system can automatically trigger the election process to ensure scheduling continuity and cluster stability. This election mechanism is beneficial for improving the efficiency of resource management within the cluster and ensuring efficient resource utilization in highly dynamic environments.

[0096] In some embodiments of the present invention, the process of updating the weight coefficients using the analytic hierarchy process is discussed in detail below: (1): Business type classification. Based on the statistical data of business intent, the business intent types, such as latency intent, bandwidth intent, reliability intent and cost intent, are classified and the business RDF is obtained through intent translation.

[0097] (2): Historical database acquisition. Refer to the weighted data set of the historical database. ,in For user business The Historical data for each weighting coefficient, among which This includes user satisfaction, network conditions, and their respective weighting coefficients. If the business is not stored in the historical database... The weighting coefficient is generated by the system based on the user's business intent type.

[0098] (3): Similarity calculation. Based on the obtained current service RDF and the current network status. The cosine similarity is calculated by comparing the data with historical database data to obtain the similarity score between the two.

[0099] (4): The Analytic Hierarchy Process (AHP) is used to obtain the highest similarity score. The hierarchical structure of the decision problem is constructed into three layers, including the objective layer (including the selected weight coefficients). ), criteria layer (including user satisfaction assessment ( Network state similarity score ( ), Business requirement RDF similarity score ( The solution layer (optional data from the historical database) and the solution layer. In the analytic hierarchy process (AHP), to construct the judgment matrix, a framework for user satisfaction assessment can be built. Network state similarity score ( ), Business requirement RDF similarity score ( Find the judgment matrix and calculate the largest eigenvalue of the judgment matrix. and the corresponding feature vector This is to obtain the relative weights of the three. This feature vector This will serve as the basis for similarity-weighted recommendations. The feature vector... Normalization is performed so that the sum of its components is 1 to obtain the weight vector. ,Will Substitute the values ​​into the equations to obtain the similarity score. .

[0100] (5): Obtain the weight of the highest similarity data. Calculate the highest similarity score using the analytic hierarchy process (AHP), and select the weight coefficient of this historical data item. Substitute it into the optimization objective to achieve task scheduling. Evaluate user satisfaction by obtaining user ratings. and network conditions Then, the satisfaction level will be... Network status and the weight coefficients of the objective function The resulting tuple data is stored in the historical database.

[0101] In one specific embodiment of the present invention, when a LEO satellite network operator deploys a satellite constellation and provides global communication services, the corresponding system framework can optimize satellite resource allocation and service response processes by integrating service intent awareness and task scheduling mechanisms. In practice, the satellite network pre-analyzes user tasks based on the service needs and geographical distribution of terminal devices (such as IoT sensors and mobile devices), and allocates tasks to different LEO satellite nodes according to priority and service intent. Each LEO satellite node dynamically adjusts its task scheduling strategy based on the cluster partitioning mechanism and the coordination of auxiliary control nodes, ensuring rapid response to user requests and reducing task processing latency.

[0102] In another specific embodiment of the present invention, the corresponding system framework can improve the resource utilization efficiency of the LEO satellite network through dynamic task scheduling and resource management. When a service request occurs, the system can determine which satellite node will handle the task based on factors such as the resource status of the satellite nodes, the user task type, and the service intent. If the resources of a certain satellite node are insufficient to meet the task requirements, the task will be dynamically allocated to other suitable nodes through coordination by auxiliary control nodes within the cluster, thereby minimizing task processing latency and resource waste. Through this adaptive task scheduling strategy, the framework can effectively improve the quality of user service and optimize the resource utilization of the satellite network.

[0103] In another specific embodiment of the present application, the system framework can convert the service requirements of users (such as delay-sensitive tasks, cost-sensitive tasks, and computing-intensive tasks, etc.) into standardized optimization index weights through a service intention perception mechanism, and dynamically adjust the task scheduling strategy in combination with a network preference perception mechanism. In actual deployment, the ground cloud center controller or the auxiliary control node can generate the optimal scheduling weight for each task according to historical data and the analytic hierarchy process (AHP), to ensure that the task can be efficiently executed on the most suitable satellite node. Through this intelligent task scheduling scheme, the framework can not only improve the resource utilization efficiency of the LEO satellite network, but also enhance the adaptability of the network to diversified service requirements.

[0104] Corresponding to the above method, the present application also provides a task scheduling system for a low-orbit satellite edge computing network, which comprises a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, and the processor being configured to execute the computer instructions stored in the memory, so that the system implements the steps of the above method.

[0105] In the above task scheduling system for a low-orbit satellite edge computing network, the execution subject can be an auxiliary control node, or a ground cloud center controller and an auxiliary control node.

[0106] Corresponding to the above method, the present application also provides a computer readable storage medium storing computer programs / instructions, which, when executed by a processor, implement the steps of the method according to any one of the above embodiments. The computer readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0107] Corresponding to the above method, the present application also provides a computer program product comprising computer programs / instructions, which, when executed by a processor, implement the steps of the method according to any one of the above embodiments.

[0108] In summary, the task scheduling method and system for a low-orbit satellite edge computing network proposed in the present application have the following technical key points: (1) Dual-weight task scheduling strategy based on service intention perception and network preference perception. The purpose is to dynamically generate task scheduling weights by deeply analyzing the user's service intention (such as delay sensitivity, computing demand, cost limit, etc.) and network service preference (such as resource utilization, load balancing, energy consumption, etc.). Through continuous perception and real-time adjustment of user intention and network preference, the system can dynamically optimize the task allocation strategy, and preferentially allocate tasks to the most matched satellite nodes. The advantage of this technology is that it can effectively improve the accuracy and resource utilization efficiency of task scheduling, reduce task processing delay, and optimize the overall performance of the network, and improve user experience.

[0109] (2) LEO satellite cluster division mechanism based on the combination of geography and logic. This mechanism combines the geographical location and orbit characteristics of satellites with their performance characteristics (such as CPU performance, memory, bandwidth, etc.) to build a multi-level cluster architecture, optimizing the management and scheduling of satellite resources. The invention preliminarily organizes satellite clusters through geographical division, and further subdivides them into logical subgroups using fuzzy clustering algorithms, which is conducive to efficient organization and dynamic adjustment of satellite resources. This mechanism not only effectively reduces management complexity, but also improves the flexibility and efficiency of resource scheduling, ensuring efficient resource utilization and task processing in large-scale LEO satellite networks.

[0110] (3) Dynamic weight generation mechanism based on historical data and AHP. This mechanism intelligently generates task scheduling weights, optimizes scheduling strategies, and reduces unnecessary resource waste. To ensure the efficiency of task scheduling, the system uses a weight generation mechanism based on historical data and AHP, which can dynamically adjust weight configuration under different network states and task demands. This mechanism can significantly improve the adaptability and flexibility of task scheduling, reduce task processing delay, and reduce resource pressure, achieving more efficient resource management, especially in resource-constrained LEO satellite networks, which can optimize network performance and improve user experience.

[0111] Compared with the prior art, the beneficial effects of the present solution are: (1) The present invention proposes a framework that integrates service intention perception and task scheduling, aiming to improve the efficiency of task scheduling and the rationality of resource utilization in LEO satellite networks. This technical solution converts the user's service intention into standardized optimization index weights, and dynamically adjusts the task scheduling strategy in combination with network preferences, thereby achieving efficient task allocation and resource management. Through the service intention perception mechanism, the invention can dynamically adjust the task scheduling strategy according to user demand and network state, significantly improving the accuracy of task scheduling and user satisfaction, while reducing task processing delay and resource waste. This solution is particularly suitable for resource-constrained and task-diversified LEO satellite network environments.

[0112] (2) The present application proposes a LEO satellite network management mechanism based on cluster division and auxiliary control node election. By optimizing the management and resource allocation of satellite clusters, the overall performance of the system can be improved. The mechanism first performs preliminary cluster division according to the geographical position and orbital characteristics of the satellites, then further subdivides into logical subgroups using the fuzzy clustering algorithm, and elects auxiliary control nodes. By dynamically adjusting the cluster structure and resource allocation strategy, the present application is beneficial to significantly improve the resource utilization efficiency and management flexibility of LEO satellite networks. In addition, the introduction of auxiliary control nodes makes task scheduling and resource management more efficient, which is beneficial to enhance the stability and reliability of the system.

[0113] (3) The present application proposes a dynamic task scheduling method based on heuristic algorithm, which is used to solve the task scheduling problem in LEO satellite networks. By combining user intention weight and network preference weight, the method optimizes the task allocation strategy, so as to realize efficient utilization of resources and rapid processing of tasks. The present application has strong adaptability and scalability by overcoming the low efficiency of traditional task scheduling methods in complex network environment through the global optimization ability of heuristic algorithm. The method can realize adaptive adjustment in the dynamic network environment, and ensure the optimization effect of task scheduling.

[0114] Those skilled in the art should understand that the exemplary components, systems and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software or a combination thereof. Whether the implementation is in hardware or software depends on the specific application and design constraints imposed on the overall system. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the present application are program or code segments used to perform the required tasks. The program or code segments can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave in a transmission medium or communication link.

[0115] It should be clear that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0116] Features described and / or illustrated with respect to one implementation can be used in the same manner or in a similar manner in one or more other implementations and / or in combination with or in place of features of other implementations.

[0117] The above descriptions are only the preferred embodiment of the application, not intended to limit the application. The application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall fall within the protection scope of the application.

Claims

1. A task scheduling method for low-Earth orbit satellite edge computing networks, characterized in that, The low-Earth orbit (LEO) satellite edge computing network is divided into multiple subgroups. Each subgroup selects one LEO satellite as an auxiliary control node, and the remaining LEO satellites within the subgroup serve as computing nodes. The method includes: After receiving a user service request, the auxiliary control node identifies the user service intent corresponding to the user service request according to the intent awareness mechanism; wherein, the user service intent includes some or all of the dimensions of service name, resource type, data volume and service performance requirements. The auxiliary control node identifies the current network service preferences of each low-Earth orbit satellite node in the corresponding subgroup according to the network service preference perception mechanism; wherein, the network service preferences include some or all of the dimensions of remaining energy, remaining CPU resources, remaining GPU resources, network load and network reliability; Identify the historical data item stored in the weighted historical database that has the highest similarity to the current user's business request in terms of both user business intent and network service preference. Generate a task scheduling strategy based on the weight coefficient of the most similar historical data item. Each historical data item in the weighted historical database includes the user's business intent, network service preference, user satisfaction, and weight coefficient for task scheduling corresponding to a historical user business request. The auxiliary control node distributes the corresponding user services to the corresponding computing nodes according to the task scheduling strategy to achieve task processing.

2. The method according to claim 1, characterized in that, The steps by which the auxiliary control node identifies the user service intent corresponding to the user service request according to the intent-aware mechanism include: The auxiliary control node uses a pre-trained natural language processing model to extract user business intent primitives from the user business request, and uses a pre-trained intent translation model to perform word analysis and / or semantic mining on the user business intent primitives, transforming the user business intent primitives into user business intents that conform to the preset RDF framework. The user's service intent is recorded in the form of RDF=<service name, resource type, data volume, service performance requirements>, and the network service preference is recorded in the form of NSI=<remaining energy, remaining CPU resources, remaining GPU resources, network load, network reliability>.

3. The method according to claim 1, characterized in that, The step of determining the historical data item stored in the weighted historical database that has the highest similarity to the current user's service request in terms of both user service intent and network service preference includes: The cosine similarity calculation method is used to calculate the similarity score between the current user's business request and each historical data item in the weighted historical database in terms of both user business intent and network service preference. The analytic hierarchy process (AHP) is used to construct a decision problem, with the objective of obtaining the historical data item that achieves the highest similarity score.

4. The method according to claim 3, characterized in that, The Analytic Hierarchy Process (AHP) constructs the decision problem into three levels: the objective level, the criterion level, and the alternative level. The target layer includes weight coefficients of the objective function, which is used to quantitatively evaluate user satisfaction and / or network benefits; wherein, the objective function includes some or all of the following indicators: latency, cost, accuracy, energy consumption, and load balancing; The criteria layer includes the assessed user satisfaction and the similarity score; The scheme layer includes each historical data item in the weighted historical database.

5. The method according to claim 3, characterized in that, Before processing user business requests, the method further includes: pre-building the weight history database; After the auxiliary control node distributes the corresponding user services to the corresponding computing nodes according to the task scheduling strategy to realize task processing, the method further includes: evaluating user satisfaction with processing the corresponding user services based on user feedback, and saving the user service request and its corresponding user service intent, network service preference, user satisfaction and weight coefficient used for task scheduling as historical data items in the weight history database.

6. The method according to claim 1, characterized in that, The method further includes: Ground-based cloud control is performed using a ground-based cloud control center, which receives historical data items from the low-orbit satellite edge computing network and updates and maintains the weighted historical database.

7. The method according to claim 1, characterized in that, The weight coefficients of historical data items include user business intent weights and network service preference weights. The step of generating a task scheduling strategy based on the weight coefficients of the historical data items with the highest similarity includes: generating a task scheduling strategy with dual weights for user business intent weights and network service preference weights using a heuristic algorithm. The heuristic algorithms include the mayfly algorithm, the ant colony optimization algorithm, the whale optimization algorithm, and the firefly algorithm.

8. A task scheduling system for low-Earth orbit satellite edge computing networks, comprising a processor, a memory, and computer programs / instructions stored in the memory, characterized in that, The processor is configured to execute the computer program / instructions, and when the computer program / instructions are executed, the system implements the steps of the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images