Large-scale satellite network service-level computing collaborative resource allocation device and method

By designing a service-level collaborative resource allocation method in a large-scale satellite network, analyzing the service request types and attributes, determining the service order and using a reinforcement learning framework for resource allocation, the complexity and scalability of resource collaborative allocation in the existing technology are solved. The problem has achieved efficient service performance guarantee for diversified services.

CN118524069BActive Publication Date: 2024-11-01XIDIAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410970782.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-01
Estimated Expiration
2044-07-19

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively coordinate communication and computing resources in large-scale satellite networks, and cannot effectively guarantee the service performance of diversified services, especially in terms of massive demands and the order of service requests. The existing method is poorly scalable and difficult to apply. on the ever-expanding network.

Method used

A large-scale satellite network service-level computing collaborative resource allocation method is designed. By dividing the time to be allocated resources into time slots, analyzing the service request types and attributes, generating a service service relationship diagram, and determining using the service request encoder and decoder. Service order, combined with the Actor network computing resource requirements, a reinforcement learning framework is adopted to allocate resources to ensure the service order and strategy of business requests.

Benefits of technology

It effectively improves the performance of large-scale satellite networks to provide services for diversified services, reduces the complexity of the solution of resource collaborative allocation strategies, improves the scalability of the method, and can be applied to networks with increasing scale.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118524069B_ABST
    Figure CN118524069B_ABST
Patent Text Reader

Abstract

The present invention discloses a large-scale satellite network service-level inter-computing collaborative resource allocation device and method, including: a network information management function module, provided with a service request collection unit and a network resource collection unit, the service request collection unit is used to collect service requests to be served that arrive at each satellite node in the network, and the network resource collection unit is used to collect the available resource status information of each satellite in the network; a service pre-processing function module, used to pre-process the service requests to be served; a service feature extraction function module, used to extract features of the pre-processed service requests to be served, and generate service demand features; a service service sequence decision submodule, used to determine the service sequence of the service requests to be served; a service service strategy decision submodule, used to output the service strategy corresponding to the service request. The present invention can effectively ensure that large-scale satellite networks provide high-performance services for various services with inter-computing resource collaborative service requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of satellite communication technology, and in particular relates to a large-scale satellite network service-level computing collaborative resource allocation device and method. Background Art

[0002] As an extension of the ground network, large-scale satellite networks (including hundreds to tens of thousands of satellites) are considered to be important infrastructure for providing on-demand high-performance services for future diversified services with their wide coverage and large capacity, and have become a hot spot in global science and technology and industrial development. However, with the continuous emergence of emerging services and the significant increase in diversified business needs (such as the massive demand for communication resources for ultra-high data density services, and the intensive requests for computing resources for Internet of Things intensive computing offloading services), efficient coordination of computing resources is essential to ensure the service performance of large-scale satellite networks. For example, computing resources are used to compress ultra-high data density services to improve the service performance of network communication resources, and communication resources are used to transmit Internet of Things intensive computing offloading services to other satellite nodes in the network for computing, thereby ensuring the service timeliness of the service and improving the service performance of the computing resources of the entire network. Therefore, it is necessary to effectively coordinate the communication and computing resources in large-scale satellite networks and design large-scale satellite network computing resource collaborative allocation technology.

[0003] In the related technology, Beijing Jiaotong University disclosed a distributed resource collaborative scheduling method and device with general computing integration in its patent document "Distributed resource collaborative scheduling method, device, equipment and medium with general computing integration" (application date: July 12, 2023, application number: 202310851893.X, application publication number: CN116866353A). The method establishes a distributed computing offloading communication model, a computing offloading calculation model, and a distributed computing power load balancing model, and completes the calculation of which computing power node to offload the computing offloading task to by designing a distributed computing power scheduling algorithm based on MADDPG. The decision on row processing is made, thereby realizing network load balancing and full utilization of resources; the designed device includes a communication model and calculation model acquisition module, a load balancing model determination module, a process transformation module, and a solution module, and the proposed method is realized through information interaction between the modules; however, the designed device only includes a general calculation model calculation module after determining the business service strategy, and does not pay attention to the impact of the sequence of business request services when batch businesses arrive on the overall service performance, nor does it have a business analysis module, and can only target one type of business; therefore, it is difficult for the device to effectively guarantee the service performance of diversified businesses with massive demands for general calculation resources.

[0004] Xiangqiang Gao et al. proposed a hierarchical dynamic resource allocation method for computation offloading in low-orbit satellite networks in their paper "Hierarchical Dynamic Resource Allocation for Computation Offloading in LEO Satellite Networks" (IEEE Internet of Things Journal, February 2024). This method focuses on low-orbit satellites as edge computing nodes to provide computing services to ground IoT users within their coverage area. It is divided into two parts: computation offloading and resource allocation. The complexity of the solution is reduced by introducing breadth-first search, which effectively reduces the latency of computation offloading business services and ensures the service performance of the business. However, this method allocates available resources in the network in sequence to a set of arriving business demands, and also does not pay attention to the impact of the order of business demand services. In addition, although this method reduces the complexity of the solution by citing breadth-first search, it is still related to the number of network business demands, the number of nodes in the network, and the number of links between nodes. It is difficult to apply to large-scale satellite networks with expanding constellation sizes in the future. Therefore, the scalability of this method is poor.

[0005] In summary, there is an urgent need to provide a large-scale satellite network service-level computing collaborative resource allocation device and method to improve the above-mentioned problems existing in the prior art. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present invention provides a large-scale satellite network service-level computing collaborative resource allocation device and method. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a large-scale satellite network service-level computing collaborative resource allocation method, comprising:

[0008] The planning time of large-scale satellite networks with resources to be allocated is divided into time slots;

[0009] For the time slot, analyzing the type of each service request to be served arriving at the network to determine the type of the service request to be served; analyzing the attributes of the service request to be served according to the type of the service request to be served to determine the attributes of the service request to be served;

[0010] According to the attributes of the business requests to be served, the relative service priority formula is used to calculate the edge attributes established by connecting the business requests to be served, and a business service relationship graph corresponding to each type of business requests to be served is generated; according to the business service relationship graph, the trained business request representation network is used to extract the service demand characteristics of the business requests to be served;

[0011] A service request encoder is used to encode the service requirement characteristics of the service request to be served, and a network service status of the service request to be served is generated; a service request decoder is used to obtain a decoded hidden state according to the network service status of the service request to be served, and a service request service probability formula is used to calculate the probability of each service request to be served being served, and a service order of the service request to be served is determined according to the probability of being served; both the service request encoder and the service request decoder are constructed by a long short-term memory network;

[0012] According to the type and attributes of the business request to be served, the available resource status information of each satellite node, and the service order of the business request to be served, the business general resource demand formula is used to calculate the demand for communication and computing resources of the business request to be served, and a set of feasible service modes that meet each business request to be served is generated; according to the set of feasible service modes, the probability of all feasible service modes of the business request to be served being selected is calculated using the Actor network, and the corresponding service strategy is selected according to the probability of the feasible service mode being selected;

[0013] Until In a time slot, all service requests to be served select a service strategy, or resources to be allocated are insufficient in large-scale satellite networks;

[0014] Get The service policy for all service requests to be served in a time slot;

[0015] Among them, the business request representation network, business request encoder, business request decoder, and Actor network are trained under the reinforcement learning framework with the goal of maximizing the completion rate of business requests to be served.

[0016] In a second aspect, the present invention further provides a large-scale satellite network service-level computing collaborative resource allocation device, comprising:

[0017] A network information management function module, a service preprocessing function module, a service feature extraction function module and a service-level resource allocation decision function module; the network information management function module includes a service request collection unit and a network resource collection unit, the service request collection unit is used to collect service requests to be served that arrive at each satellite node in the network, and the network resource collection unit is used to collect available resource status information of each satellite node in the network;

[0018] The input end of the service preprocessing function module is communicatively connected to the output end of the service request collection unit, and the service preprocessing function module is used to analyze the type of the service request to be served, determine the type of the service request to be served, and analyze the attributes of the service request to be served according to the type of the service request to be served, and determine the attributes of the service request to be served;

[0019] The input end of the business feature extraction function module is communicatively connected to the output end of the business preprocessing function module. The business feature extraction function module is used to calculate the edge attributes established by connecting the business requests to be served according to the attributes of the business requests to be served, generate a business service relationship graph corresponding to each type of business requests to be served, and extract the service demand characteristics of the business requests to be served according to the business service relationship graph;

[0020] The business-level resource allocation decision function module includes a business service sequence decision submodule and a business service strategy decision submodule; wherein the input end of the business service sequence decision submodule is respectively connected to the output end of the business feature extraction function module and the output end of the network resource collection unit for communication; the business service sequence decision submodule is used to encode the service demand characteristics of the business request to be served, generate the network service status of the business request to be served, calculate the probability of each business request to be served being served according to the network service status of the business request to be served, and determine the service sequence of the business request to be served; the input end of the business service strategy decision submodule is respectively connected to the output end of the business service sequence decision submodule, the output end of the network resource collection unit and the output end of the business preprocessing function module for communication; the business service strategy decision submodule is used to calculate the demand for communication and computing resources of the business request to be served according to the type and attribute of the business request to be served, the available resource status information of each satellite node, and the service sequence of the business request to be served, generate a set of feasible service modes that meet each business request to be served, calculate the probability of all feasible service modes of the business request to be served being selected according to the set of feasible service modes, and output the service strategy of the business request to be served according to the probability of the feasible service mode being selected.

[0021] Beneficial effects of the present invention:

[0022] 1. The present invention provides a large-scale satellite network service-level inter-computing collaborative resource allocation device and method. Compared with the existing resource allocation method that provides services for service requests in sequence and ignores the impact of service sequence on overall service performance, the present invention designs a resource allocation device with a service service sequence decision submodule. By applying the service request encoder in the service coding unit to encode each service request to be served in the network and the available resources in the network, the relationship between each service request of the network and between the service request and the available resources of the network is clarified, and the service sequence of the service requests to be served is obtained using the decoder of the service decoding unit, thereby effectively ensuring the performance of large-scale satellite networks in providing services for various services with inter-computing resource collaborative service requirements.

[0023] 2. The present invention provides a large-scale satellite network business-level general computing collaborative resource allocation device and method. In order to solve the problem that the existing resource allocation methods have high solution complexity and poor scalability, and are difficult to apply to large-scale satellite networks with continuously expanding network scales, the present invention designs a business-level general computing resource collaborative allocation method based on a reinforcement learning network framework. From the business perspective, with the help of the network's ability to process sequence data, reinforcement learning is applied to select service strategies for business requests that have been arranged in a service order, thereby realizing resource allocation that is independent of the network scale. The solution complexity is only related to the number of network business requests, thereby effectively reducing the complexity of solving the large-scale satellite network resource collaborative allocation strategy and improving the scalability of the method.

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of a large-scale satellite network service-level computing collaborative resource allocation device provided by an embodiment of the present invention;

[0026] Figure 2 It is a flow chart of a large-scale satellite network service-level general computing collaborative resource allocation method provided by an embodiment of the present invention;

[0027] Figure 3 is a flow chart for determining a set of feasible service modes provided by an embodiment of the present invention;

[0028] Figure 4 It is a histogram of service request completion rate results under different service request arrival probabilities of three general computing collaborative resource allocation methods in a 66-satellite network scenario provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0030] In view of the deficiencies in the prior art, the present invention proposes a device and method for allocating service-level inter-computing collaborative resources in a large-scale satellite network. By designing a service service sequence decision submodule with a service encoding unit and a service decoding unit, the order of services can be determined according to the service requirements of the service requests to be served, thereby effectively improving the performance of the network in serving various types of services with inter-computing resource collaborative service requirements. By introducing a pointer network reinforcement learning framework, services are provided for the service requests of the entire network from a service-level perspective, thereby avoiding the problem of increased complexity in solving the large-scale satellite network resource collaborative allocation strategy caused by the growth of the network scale. The device has good scalability and can be applied to large-scale satellite networks with expanding constellation sizes in the future.

[0031] See also Figure 1 , Figure 1 The present invention is a schematic diagram of a large-scale satellite network service-level general computing collaborative resource allocation device provided by an embodiment of the present invention. The large-scale satellite network service-level general computing collaborative resource allocation device provided by the present invention includes: a network information management function module, a service preprocessing function module, a service feature extraction function module and a service-level resource allocation decision function module; the network information management function module includes a service request collection unit and a network resource collection unit, the service request collection unit is used to collect service requests to be served that arrive at each satellite node in the network, and the network resource collection unit is used to collect available resource status information of each satellite node in the network;

[0032] The input end of the service preprocessing function module is communicatively connected to the output end of the service request collection unit, and the service preprocessing function module is used to preprocess the service request to be served;

[0033] The input end of the service feature extraction function module is communicatively connected to the output end of the service preprocessing function module, and the service feature extraction function module is used to extract features of the preprocessed service requests to be served, and generate service demand features for each service request to be served;

[0034] The business-level resource allocation decision function module includes a business service sequence decision submodule and a business service policy decision submodule; wherein, the input end of the business service sequence decision submodule is respectively communicated with the output end of the business feature extraction function module and the output end of the network resource acquisition unit, and the business service sequence decision submodule is used to determine the service sequence of the business requests to be served according to the service demand characteristics of each business request to be served and the available resource status information of each satellite node; the input end of the business service policy decision submodule is respectively communicated with the output end of the business service sequence decision submodule, the output end of the network resource acquisition unit and the output end of the business preprocessing function module, and the business service policy decision submodule is used to output the service policy of the business request to be served according to the service sequence of the business request to be served and the available resource status information of each satellite node, as well as the preprocessed business request to be served.

[0035] In an optional embodiment of the present invention, the service pre-processing function module includes a service type analysis submodule and a service attribute analysis submodule;

[0036] The input end of the business type analysis submodule is communicatively connected to the output end of the business request collection unit, the output end of the business type analysis submodule is communicatively connected to the input end of the business attribute analysis submodule, and the output end of the business attribute analysis submodule is communicatively connected to the input end of the business feature extraction function module and the input end of the business service strategy decision submodule respectively; the business type analysis submodule is used to analyze the category of the received business request to be served and obtain the type of the business request to be served, and the business attribute analysis submodule is used to analyze the attributes of the business request to be served according to the type of the business request to be served and obtain the attributes of the business request to be served.

[0037] In an optional embodiment of the present invention, the business feature extraction function module includes a business service relationship construction submodule and a business feature embedding submodule;

[0038] The input end of the business service relationship construction submodule is communicatively connected to the output end of the business attribute analysis submodule, the output end of the business service relationship construction submodule is communicatively connected to the input end of the business feature embedding submodule, and the output end of the business feature embedding submodule is communicatively connected to the input end of the business service sequence decision submodule; the business service relationship construction submodule is used to generate a business service relationship graph according to the attributes of the business request to be serviced, and the business feature embedding submodule is used to extract the service requirement characteristics of the business request to be serviced according to the business service relationship graph.

[0039] In an optional embodiment of the present invention, the business service sequence decision submodule includes a business encoding unit and a business decoding unit;

[0040] The input end of the service encoding unit is communicatively connected to the output end of the service feature embedding submodule, the input end of the service encoding unit is also communicatively connected to the output end of the network resource acquisition unit, the output end of the service encoding unit is communicatively connected to the input end of the service decoding unit, and the output end of the service decoding unit is communicatively connected to the input end of the service service strategy decision submodule; the service encoding unit is used to encode the service request to be served according to the service demand characteristics of each service request to be served and the available resource status information of each satellite node, and generate the network service status of the service request to be served, and the service decoding unit is used to calculate the probability of the service request to be served according to the network service status of the service request to be served, and determine the service order of the service request to be served.

[0041] In an optional embodiment of the present invention, the business service policy decision submodule includes a decision set generation unit and a service policy decision unit;

[0042] The input end of the decision set generation unit is communicatively connected to the output end of the business service relationship construction submodule, the input end of the decision set generation unit is also communicatively connected to the output end of the business decoding unit, and the output end of the decision set generation unit is communicatively connected to the input end of the service policy decision unit; the decision set generation unit is used to generate a set of feasible service modes that meet each business request to be served according to the service order of the business requests to be served, and the service policy decision unit is used to select a corresponding service policy for each business request to be served from the set of feasible service modes according to the service order of the business requests to be served.

[0043] For more details, please see Figure 1 The large-scale satellite network service-level computing collaborative resource allocation device includes a network information management function module, a service preprocessing function module, a service feature extraction function module and a service-level resource allocation decision function module. The network information management function module includes a service request collection unit and a network resource collection unit. The service preprocessing function module includes a service type analysis submodule and a service attribute analysis submodule. The service feature extraction function module includes a service service relationship construction submodule and a service feature embedding submodule. The service-level resource allocation decision function module includes a service service sequence decision submodule and a service service strategy decision submodule. The service service sequence decision submodule includes a service encoding unit and a service decoding unit. The service service strategy decision submodule includes a decision model training unit and a service strategy decision unit.

[0044] The service request collection unit is used to collect the service requests to be served that arrive at each satellite node in the network, construct a sequence of service requests to be served, and output the sequence to the service type analysis submodule through the SS1 interface; wherein the SS1 interface is the information interaction interface between the service request collection unit and the service type analysis submodule.

[0045] The network resource collection unit is used to collect the available resource status information of each satellite node in the network, and output the information to the service coding unit through the RD1 interface and to the decision set generation unit through the RD2 interface; wherein the RD1 interface is the information interaction interface between the network resource collection unit and the service coding unit, and the RD2 interface is the information interaction interface between the network resource collection unit and the decision set generation unit.

[0046] The business type analysis submodule is used to analyze the business type of each arriving business request to be served, and output the analysis results to the business attribute analysis submodule through the SS2 interface; wherein the SS2 interface is the information interaction interface between the business type analysis submodule and the business attribute analysis submodule.

[0047] The business attribute analysis submodule is used to analyze the attributes of each business request to be served in the business request sequence to be served according to the type of the business request to be served, obtain the attributes of each business request to be served, and output the obtained results to the business service relationship construction submodule through the SS3 interface and to the decision set generation unit through the SD1 interface; among which, the SS3 interface is the information interaction interface between the business attribute analysis submodule and the business service relationship construction submodule, and the SD1 interface is the information interaction interface between the business attribute analysis submodule and the decision set generation unit.

[0048] The business service relationship construction submodule is used to construct a business service relationship diagram for each type of business request to be serviced, and output the constructed business service relationship diagram to the business feature embedding submodule through the SS4 interface; wherein the SS4 interface is the information interaction interface between the business service relationship construction submodule and the business feature embedding submodule.

[0049] The business feature embedding submodule is used to extract the service requirement characteristics of each business request to be served based on the constructed business service relationship diagram, and output the business feature information to the business encoding unit through the SD2 interface; wherein the SD2 interface is the information interaction interface between the business feature embedding submodule and the business encoding unit.

[0050] The service encoding unit is used to encode the service request to be served according to the service demand characteristics of each service request and the available resource status information of each satellite node to obtain the network service status of the service request, and output the status to the service decoding unit through the DD1 interface; wherein the DD1 interface is the information interaction interface between the service encoding unit and the service decoding unit.

[0051] The service decoding unit is used to calculate the probability of each service request to be served, determine the service order of the service requests to be served, and output the service order to the service policy decision submodule through the DD2 interface; wherein the DD2 interface is the information interaction interface between the service encoding unit and the service policy decision submodule.

[0052] The decision set generation unit is used to determine the set of feasible service modes that meet each business request to be served according to the service order of the business requests to be served, and output it to the service policy decision unit through the DD3 interface; wherein the DD3 interface is the information interaction interface between the decision set generation unit and the service policy decision unit.

[0053] The service strategy decision unit is used to select a service strategy for each service request to be served from a set of feasible service modes according to the service order of the service requests.

[0054] Based on the same inventive concept, see Figure 2 As shown, Figure 2 1 is a flow chart of a large-scale satellite network service-level general computing collaborative resource allocation method provided by an embodiment of the present invention. The present invention also provides a large-scale satellite network service-level general computing collaborative resource allocation method, which is applied to a large-scale satellite network service-level general computing collaborative resource allocation device provided by the above embodiment. Please refer to the above for the content of the device, which will not be repeated here. The method includes:

[0055] S101, dividing the planning time of the large-scale satellite network to be allocated with resources into time slots;

[0056] S102, for time slot, analyzing the type of each service request to be served arriving at the network to determine the type of the service request to be served; analyzing the attributes of the service request to be served according to the type of the service request to be served to determine the attributes of the service request to be served;

[0057] S103, according to the attributes of the service requests to be served, use the relative service priority formula to calculate the edge attributes established by connecting the service requests to be served, and generate a service service relationship graph corresponding to each type of service requests to be served; according to the service service relationship graph, use the trained service request representation network to extract the service demand characteristics of the service requests to be served;

[0058] S104, using a service request encoder to encode the service requirement characteristics of the service request to be served, and generating a network service status of the service request to be served; according to the network service status of the service request to be served, using a service request decoder to obtain a decoding hidden state, using a service request service probability formula to calculate the probability of each service request to be served being served, and determining the service order of the service request to be served according to the probability of being served; the service request encoder and the service request decoder are both constructed by a long short-term memory network;

[0059] S105. According to the type and attributes of the service request to be served, the available resource status information of each satellite node, and the service order of the service request to be served, the service general resource demand formula is used to calculate the demand for communication and computing resources of the service request to be served, and a set of feasible service modes that meet each service request to be served is generated; according to the set of feasible service modes, the probability of all feasible service modes of the service request to be served being selected is calculated using the Actor network, and the corresponding service strategy is selected according to the probability of the feasible service mode being selected;

[0060] S106, until In a time slot, all service requests to be served select corresponding service strategies, or resources to be allocated are insufficient in large-scale satellite networks;

[0061] S107, Acquisition The service policy for all business requests to be served within a time slot.

[0062] Specifically, in this embodiment, the following process is described in detail, specifically:

[0063] Step S1, divide the resources to be allocated. The planning time of large-scale satellite network is time slots.

[0064] Step S2, load all trained networks.

[0065] Load the trained business request representation network to the business feature embedding submodule, load the trained pointer network composed of the business request encoder and decoder to the business encoding unit and the business decoding unit, and load the trained Actor network to the service strategy decision unit.

[0066] Among them, all the above networks are trained under the reinforcement learning framework with the goal of maximizing the completion rate of business requests.

[0067] The training process of all networks used in the present invention is further described as follows:

[0068] Step S2.1, initialize the training parameters of the business request representation network, pointer network, actor network and critic network.

[0069] Among them, the Critic network is used to evaluate the advantages and disadvantages of the service order and service strategy selected by the pointer network and the Actor network. It is only used in training to enable the network to have higher service performance.

[0070] Step S2.2: Initialize the total number of training cycles The total number of planned time slots in each cycle is .

[0071] Step S2.3: For each training cycle , clear the service request sequence and set the initial resource status of all satellite nodes.

[0072] Step S2.4, in the Each time slot executes the contents described in the following steps S3 to S6, determines the business request of this service and its service strategy, and executes the service strategy to obtain an immediate reward for providing services for the business request of this service, until all business requests arriving at the current time slot are completed, or the resources in the current time slot in the network are insufficient to continue to provide services.

[0073] Step S2.5: The service relationship graph constructed in steps S3 to S6, the available resource status of the acquired satellite nodes, the service strategy selected for each business request and the probability of selecting the service strategy are stored in the experience database as training experience data.

[0074] Step S2.6: Determine whether the amount of experience data stored in the experience database has reached the amount that can be used for network training. If so, apply the loss function of the Actor network to update the training parameters of the business request representation network, the pointer network, and the Actor network, and apply the loss function of the Critic network to update the training parameters of the Critic network. Otherwise, execute step S2.7.

[0075] Among them, the loss function of the Actor network can simultaneously update the training parameters of the business request representation network and the pointer network. The reason is that the output results of the business request representation network and the pointer network are the information required by the Actor network to select service strategies. The performance of the business request representation network and the pointer network will be directly reflected by the performance of the Actor network selected service strategies. Therefore, the loss function of the Actor network can be used to synchronously update the training parameters of the business request representation network and the pointer network.

[0076] Loss function for Actor network It is expressed as:

[0077] ;

[0078] ;

[0079] ;

[0080] in, represents the number of training data sampled from the empirical database, represents the logarithmic operation, Indicates the training parameters of the business request representation network, pointer network, and Actor network. Indicates Business request sequences in training data Adopting business request service strategy sequence The probability of Indicates the selected The service policy for a business request. Indicates that in the business request sequence The Service requests adopt service policies The probability of Indicates Business request sequences in training data Adopting business request service strategy sequence Rewards received, Indicates that in the business request sequence The Service requests adopt service policies Rewards received, represents a baseline function that is independent of the adopted strategy.

[0081] Loss function of the Critic network It is expressed as:

[0082] ;

[0083] in, represents the training parameters of the Critic network, Represents the reward value predicted by the Critic network, Indicates the 2-norm operation.

[0084] Step S2.7: Determination ,if, , return to step S2.4, otherwise execute step S2.8.

[0085] Step S2.8: Determination ,if, , return to step S2.3, otherwise complete the training and obtain the trained business request representation network, pointer network, and Actor network.

[0086] Step S3: Analyze the type and attributes of the service request arriving in the network.

[0087] Step S3.1: The service type analysis submodule analyzes the service type to which each service request belongs based on the service request sequence provided by the service request collection unit through the SS1 interface to obtain the type of each service request.

[0088] The service request sequence is collected by the service request collection unit. It is constructed by the service requests to be served that arrive at each satellite node in the network in each time slot.

[0089] The types of business requests include all the business types that can be provided by large-scale satellite networks at this stage.

[0090] Step S3.2: The service attribute analysis submodule analyzes the attributes of each service request based on the type of service request provided by the service type analysis submodule through the SS2 interface to obtain the attributes of each service request.

[0091] The attributes of a service request include the service request type, service request arrival rate, service request data volume, service request remaining service time, whether the service request needs data processing, the calculation cycle required for each bit of service request data, the compression ratio of service request data, the number of the satellite node where the service request is currently located and its location information, the number of the source node of the service request and its location information, the number of the destination node of the service request and its location information, and the location information of the satellite node adjacent to the satellite node where the service request is currently located. Among them, the two attributes of the calculation cycle required for each bit of service request data and the compression ratio of service request data only exist when the service request needs data processing.

[0092] It should be noted that in actual applications, there is a clear correlation between the service request type and the service request attribute. For example, for large-capacity data transmission services, the service request attribute is that the service request data volume is large and the service request can tolerate a high delay, while for remote machine communication services, the service request data volume is small and the service request can tolerate a low delay. Therefore, after knowing the service request type, the service request attribute can be analyzed according to the service request type.

[0093] Step S4: extracting the service requirement characteristics of the business request.

[0094] Step S4.1, the business service relationship construction submodule determines the connectivity between each business request, and based on the partial business request attributes provided by the business attribute analysis submodule through the SS3 interface, applies the relative service priority formula to calculate the edge attributes established by the connected business requests, and further constructs the authorized service relationship graph.

[0095] Specifically, the specific implementation of constructing the authorized service relationship graph by the business service relationship construction submodule is further described:

[0096] Step 1: Determine whether the service requests in the network are connected based on the satellite node number and service request type where the service request arrives. If the satellite node number and service request type are the same, the two services are considered to be connected and an edge can be constructed. Otherwise, they are considered to be disconnected and an edge cannot be constructed. The constructed edge is a bidirectional edge.

[0097] Step 2: Based on some attributes of the service request, the service priority of each service request is calculated by applying the following service priority formula:

[0098] ;

[0099] in, Indicates The service priority of each business request. Represents the service priority calculation attribute set of the business request, Indicates the operation of finding the number of elements in a collection. Indicates The first business request The weight of the attribute, Indicates The first business request The quantitative value of an attribute.

[0100] The service priority calculation attribute set of the service request includes the data volume of the service request, the remaining service time of the service request, and the distance between the current satellite node where the service request is located and the destination satellite node.

[0101] The distance between the satellite node where the service request is currently located and the destination satellite node may be calculated using the location information of the satellite node where the service request is currently located and the destination satellite node of the service request.

[0102] No. The first business request The weight of the attribute The calculation formula is:

[0103] ;

[0104] in, It indicates the score obtained by comparing the importance of each attribute of the business request using a 0-1 scoring method to effectively evaluate the The weight of each attribute of a business request, The calculation formula is:

[0105] ;

[0106] Indicates The importance of the attribute is higher than The importance of the first attribute is the same as the rest.

[0107] The specific value of is related to the data volume of the service request, the remaining service time of the service request, and the value of the distance between the satellite node where the service request is currently located and the destination satellite node.

[0108] Quantified value of the data volume of the first attribute service request in the attribute set It is expressed as:

[0109] ;

[0110] in, Indicates The amount of data requested by a business, Indicates the maximum amount of business request data.

[0111] The quantitative value of the remaining service time of the second attribute service request in the attribute set It is expressed as:

[0112] ;

[0113] in, Indicates The remaining service time of a business request, Indicates If the remaining service time of a service request is 0, the service request will be deleted, that is, the service fails.

[0114] The third attribute in the attribute set is the quantitative value of the distance between the current satellite node and the destination satellite node. It is expressed as:

[0115] ;

[0116] in, Indicates The satellite node where the service request is currently located and its destination satellite node The distance between Indicates The source satellite node of the service request and its destination satellite node The distance between.

[0117] Step 3: Use the following relative service priority formula to calculate the weight of the bidirectional edge in each direction of any two interconnected service requests. The weight is the edge attribute:

[0118] ;

[0119] in, Indicates The business request is relative to the The service priority of a business request, that is, the starting node is the business request , the termination node is the business request The edge attributes in the direction, Indicates The service priority of the business request. The business request is relative to the The service priority of each business request (The starting node is the business request , the termination node is the business request The edge attributes in the direction can also be calculated using the above relative service priority formula.

[0120] It should be noted that since the service request may go through multiple hops during transmission from the source node to the destination node and each hop of the service request transmission takes time, , and The specific value of may change during the service request transmission process, thereby changing .

[0121] Step 4: Establish an authorized service relationship graph with business requests as nodes, partial attributes of business requests as node attributes, business request connectivity as edges, and edge weights as edge attributes.

[0122] The attributes of the node include the data volume of the service request, the remaining service time of the service request, whether the service request needs data processing, the location information of the satellite node where the service request is currently located, and the distance between the adjacent satellite node of the satellite node where the service request is located and the destination satellite node. The distance is calculated using the location information of the adjacent satellite node of the satellite node where the service request is located and the location information of the destination satellite node.

[0123] Step S4.2, the business feature embedding submodule is based on the service relationship graph provided by the business service relationship construction submodule through the SS4 interface, and applies the business request representation network composed of a graph convolutional neural network to extract the service demand characteristics of the business request.

[0124] Step S5: Select the business request for this service.

[0125] Step S5.1, the service encoding unit constructs a service feature sequence of the service request based on the service requirement features of each service request provided by the service feature embedding submodule through the SD2 interface and the available resource status information of the satellite provided by the network resource acquisition unit through the RD1 interface, and applies the service request encoder constructed by the long short-term memory network to encode the service feature sequence to obtain the network service status of each service request.

[0126] Step S5.2, the service decoding unit calculates the probability of each service request being served by applying the service request service probability formula based on the network service status provided by the service encoding unit through the DD1 interface and the decoded hidden state obtained by the service request decoder constructed by the long short-term memory network, and selects the service request for this service according to the service probability.

[0127] See also Figure 3 , Figure 3 This is a flow chart for determining a set of feasible service modes provided by an embodiment of the present invention, and further describes the specific implementation of the service request selected by the service decoding unit for this service:

[0128] Step 1: Determine whether it is The first service request served by the time slot, if yes, the network service status of the last service request arriving at the current time slot is used as the input status information of the service request decoder, otherwise, the network service status of the last service request served is used as the input status information of the service request decoder.

[0129] In the second step, the input state information is input into the service request decoder constructed by the long short-term memory network to obtain the decoded hidden state.

[0130] Step 3: Based on the network service status and the decoded hidden status, the probability of each service request being served is calculated using the following service request service probability formula:

[0131] ;

[0132] ;

[0133] ;

[0134] in, Indicates Scheduled time slot service request sequence The probability of each service request to be served in being served, represents the normalized exponential function, Indicates Each service request to be served is The input value of Indicates Service time Business requests to be serviced are The input value of represents the hyperbolic tangent function, Indicates The network service status of the business requests to be serviced, Indicates the calculation for The decoded hidden state of the service requests being served, , and Both represent weight parameters, Indicates Service time Whether the business request to be served has been completed, represents the matrix multiplication operation, Represents a matrix transpose operation.

[0135] No. If the first service If a business request is not completed by the service, Set to 0, otherwise Set to a value much larger than the first term in the equation Calculate the value of the result so that Service time The probability of a business request being served approaches 0 and it will not be selected for service.

[0136] Step 4: Select the service request for this service based on the probability of each service request being served.

[0137] Step S6: Select a service policy for the business request of this service.

[0138] Step S6.1, the decision set generation unit applies the business general resource requirement formula to calculate the demand for communication and computing resources of the business request based on the business request information of this service provided by the business decoding unit through the DD2 interface, some business request attributes provided by the business attribute analysis submodule through the SD1 interface, and the satellite available resource status information provided by the network resource acquisition unit through the RD2 interface, and removes the service modes that cannot meet the demand in the service mode set according to the calculation results to obtain a set of feasible service modes for the business request.

[0139] The formula for business general calculation resource requirements is as follows:

[0140] ;

[0141] ;

[0142] in, Indicates The demand for computing resources for the business requests to be served, Indicates the calculation cycle required for unit bit data, Indicates The amount of data for business requests to be served, Indicates Whether the business request to be served is to be processed. Indicates The demand for communication resources for the service requests to be served, Indicates the compression ratio of the business request data to be served.

[0143] It should be noted that, since the service request may go through multiple hops during transmission from the source node to the destination node and data processing may be performed at one of the satellite nodes passed through, and The specific value may change during the service request transmission process.

[0144] The feasible service mode set includes the communication links between adjacent satellite nodes that can be directly used by the business request of this service at the current satellite node and the communication links between adjacent satellite nodes that can be used after compression processing of the computing resources of the current satellite node.

[0145] Step S6.2, the service strategy decision unit calculates the probability of all feasible service modes of the service request being selected based on the set of feasible service modes for the service business request provided by the decision set generation unit through the DD3 interface, and applies the Actor network composed of a fully connected layer, and selects the service strategy for the business request according to the probability, and executes the service strategy to provide service for the selected business request.

[0146] Step S7: loop through steps S5 to S6 until the All service requests arriving in a time slot are served or the resources available in the network are insufficient to continue providing services.

[0147] Step S8: cyclically execute steps S3 to S7 until completion. time slot service.

[0148] In an optional embodiment of the present invention, the effect of the present invention is further described below in conjunction with simulation experiments.

[0149] 1. Simulation experiment conditions:

[0150] The simulation experiment platform of the present invention is: Windows 10 operating system and Python 3.8.

[0151] In the simulation experiment of the present invention, two satellite network scenarios including 66 satellites and 1584 satellites are set respectively. The available communication resources for each satellite to transmit the service request to be served to its adjacent satellite are 200Mbits / s, the available computing resources for each satellite are 2Gcycles / s, each satellite can receive service requests and provide transmission services, the time interval for the arrival of service requests obeys an exponential distribution with a mean of 2s, and the delay demand service of the service requests is uniformly distributed from [5,15]s, and the data size of each service request obeys a uniform distribution of [20,100]Mbits.

[0152] 2. Simulation content and results analysis:

[0153] In the simulation experiment of the present invention, the following three general computing collaborative resource allocation methods are simulated respectively, namely, the service disordered general computing collaborative resource allocation method, the local greedy resource allocation method and the business-level general computing collaborative resource allocation method are simulated respectively, and the service request completion rates of the three general computing collaborative resource allocation methods under different service request arrival probabilities in the 66-satellite network scenario are obtained. Furthermore, the model trained by the business-level general computing collaborative resource allocation method designed by the present invention in the 66-satellite network scenario is applied to the service request completion rate in the 1584-satellite network scenario; see Figure 4 , Figure 4 The histogram of the service request completion rate results of three general-purpose collaborative resource allocation methods under different service request arrival probabilities in a 66-satellite network scenario is described. Figure 4It can be seen that the service-level general computing collaborative resource allocation method designed by the present invention obtains a higher service request completion rate than the other two comparison methods as the probability of service request arrival increases, and when the network has a high service load, the performance gain of the service-level general computing collaborative resource allocation method designed by the present invention is more significant. The reason is that as the service request arrival rate increases, the available general computing resources of the network are limited. It is necessary to plan the service order of each service request according to the data volume, delay requirements and destination node of the service request to reasonably utilize the network general computing resources and ensure that more service requests can be effectively served. Please refer to Table 1, which shows the service request completion rate result table of the model trained in the 66-satellite network scenario and applied to the 1584-satellite network scenario. It can be seen from Table 1 that the model trained in the 66-satellite network scenario can be directly applied to the 1584-satellite network scenario, and good service performance can also be obtained in the 1584-satellite network scenario, which means that the service-level general computing collaborative resource allocation method designed by the present invention has good scalability and can effectively cope with large-scale satellite networks with expanding network scale.

[0154] Table 1 Service request completion rate of the model trained in the 66-satellite network scenario and applied to the 1584-satellite network scenario

[0155]

[0156] It should be noted that the three general computing collaborative resource allocation methods are:

[0157] The first general computing collaborative resource allocation method, under the framework of the design concept of the present invention, removes the business service order decision submodule and does not sort the business requests. The remaining parts are the same as the design of the present invention, and are referred to as the service unordered general computing collaborative resource allocation method.

[0158] The second general collaborative resource allocation method, under the framework of the design concept of the present invention, gives priority to serving business requests that can reach the destination node in one hop, and the rest of the method is the same as the design of the present invention, which is referred to as the local greedy resource allocation method.

[0159] The third general computing collaborative resource allocation method uses step S5 of the present invention to determine the service order of the service requests for the service request sequence arriving at the satellite, and uses step S6 of the present invention to determine the service strategy of the service requests, which is referred to as the service-level general computing collaborative resource allocation method.

[0160] In summary, the present invention provides a large-scale satellite network service-level computing collaborative resource allocation device and method, which has the following beneficial effects:

[0161] 1. Compared with the existing resource allocation method that provides services for business requests in sequence and ignores the impact of service sequence on overall service performance, the present invention designs a resource allocation device with a business service sequence decision submodule, which encodes each business request in the network and the available resources in the network by applying the business request encoder in the business encoding unit, clarifies the relationship between each business request in the network and between the business request and the available resources of the network, and uses the decoder of the business decoding unit to obtain the service sequence of the business request, thereby effectively ensuring the performance of large-scale satellite networks in providing services for various businesses with collaborative service requirements for computing resources.

[0162] 2. In view of the problem that the existing resource allocation methods have high solution complexity and poor scalability, and are difficult to apply to large-scale satellite networks with continuously expanding network scale, the present invention designs a business-level general computing resource collaborative allocation method based on a reinforcement learning network framework. From the business perspective, with the help of the network's ability to process sequence data, reinforcement learning is applied to select service strategies for business requests that have already been arranged in a service order, thereby realizing resource allocation that is independent of the network scale. The solution complexity is only related to the number of network business requests, thereby effectively reducing the complexity of solving the large-scale satellite network resource collaborative allocation strategy and improving the scalability of the method.

[0163] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant is intended to cover non-exclusive inclusion, so that the article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the article or device including the elements. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The orientation or position relationship indicated by "up", "down", "left", "right", etc. is based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0164] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0165] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A large-scale satellite network service-level collaborative resource allocation method, characterized in that: include: The planning time of large-scale satellite networks with resources to be allocated is divided into time slots; For the time slot, analyzing the type of each service request to be served arriving at the network, and determining the type of the service request to be served; Analyzing the attributes of the service request to be serviced according to the type of the service request to be serviced, and determining the attributes of the service request to be serviced; According to the attributes of the service requests to be served, the edge attributes established by connecting the service requests to be served are calculated using the relative service priority formula to generate a service service relationship graph corresponding to each type of service requests to be served; according to the service service relationship graph, the service demand characteristics of the service requests to be served are extracted using the trained service request representation network; Encoding the service requirement characteristics of the service request to be served using a service request encoder to generate a network service status of the service request to be served; According to the network service status of the service request to be served, a service request decoder is used to obtain a decoding hidden state, and a service request service probability formula is used to calculate the probability of each service request to be served being served, and the service order of the service request to be served is determined according to the probability of being served; the service request encoder and the service request decoder are both constructed by a long short-term memory network; According to the type and attribute of the service request to be served, the available resource status information of each satellite node, and the service order of the service request to be served, the service general resource requirement formula is used to calculate the demand of the service request to be served for communication and computing resources, and generate a set of feasible service modes that meet each of the service requests to be served; According to the set of feasible service modes, the probability of all feasible service modes being selected for the business request to be served is calculated using the Actor network, and the corresponding service strategy is selected according to the probability of the feasible service mode being selected; Until In a time slot, all service requests to be served select a service strategy, or resources to be allocated are insufficient in large-scale satellite networks; Get The service policy for all service requests to be served in a time slot; The service request representation network, the service request encoder, the service request decoder, and the Actor network are trained under a reinforcement learning framework with the goal of maximizing the completion rate of service requests to be served, and the service request representation network is composed of a graph convolutional neural network; The service request probability formula is: ; ; ; in, Indicates A sequence of service requests to be served in time slots The probability of each pending service request being served, represents the normalized exponential function, Indicates Each service request to be served is The input value of Indicates Service time Business requests to be serviced are The input value of represents the hyperbolic tangent function, Indicates The network service status of the business requests to be serviced, Indicates the calculation for The decoded hidden state of the service requests being served, , and Both represent weight parameters, Indicates Service time Whether the business request to be served has been completed by the service, represents the matrix multiplication operation, Represents a matrix transpose operation.

2. The large-scale satellite network service-level collaborative resource allocation method according to claim 1, characterized in that: The relative service priority formula is: ; in, Indicates The business request to be serviced is relative to the The service priority of the business request to be served, that is, the starting node is the business request , the termination node is the business request The edge attributes in the direction, Indicates The service priority of the business requests to be serviced, Indicates The service priority of the business requests to be serviced.

3. The large-scale satellite network service-level computing collaborative resource allocation method according to claim 2 is characterized in that: The service priority of the service request to be served The expression is: ; in, A service priority calculation attribute set representing the service request to be serviced, Indicates the operation of finding the number of elements in a collection. Indicates of the business requests to be serviced The weight of the attribute, Indicates of the business requests to be serviced The quantitative value of an attribute.

4. The large-scale satellite network service-level collaborative resource allocation method according to claim 1, characterized in that: The business general calculation resource requirement formula is: ; ; in, Indicates The demand for computing resources for the business requests to be served, Indicates the calculation cycle required for unit bit data, Indicates The amount of data for business requests to be served, Indicates Whether the business request to be served is to be processed. Indicates The demand for communication resources for the service requests to be served, Indicates the compression ratio of the business request data to be served.

5. A large-scale satellite network service-level inter-computing collaborative resource allocation device, used to implement the large-scale satellite network service-level inter-computing collaborative resource allocation method as claimed in any one of claims 1 to 4, characterized in that: include: Network information management function module, service preprocessing function module, service feature extraction function module and service-level resource allocation decision function module; The network information management function module includes a service request collection unit and a network resource collection unit, wherein the service request collection unit is used to collect service requests to be served that arrive at each satellite node in the network, and the network resource collection unit is used to collect available resource status information of each satellite node in the network; The input end of the service preprocessing function module is communicatively connected to the output end of the service request collection unit, and the service preprocessing function module is used to analyze the type of the service request to be served, determine the type of the service request to be served, and analyze the attributes of the service request to be served according to the type of the service request to be served, and determine the attributes of the service request to be served; The input end of the service feature extraction function module is communicatively connected to the output end of the service preprocessing function module, and the service feature extraction function module is used to calculate the edge attributes established by connecting the service requests to be served according to the attributes of the service requests to be served, generate a service service relationship graph corresponding to each type of service requests to be served, and extract the service demand characteristics of the service requests to be served according to the service service relationship graph; The business-level resource allocation decision function module includes a business service sequence decision submodule and a business service strategy decision submodule; wherein the input end of the business service sequence decision submodule is respectively communicated with the output end of the business feature extraction function module and the output end of the network resource acquisition unit, and the business service sequence decision submodule is used to encode the service demand characteristics of the business request to be served, generate the network service status of the business request to be served, calculate the probability of each business request to be served being served according to the network service status of the business request to be served, and determine the service order of the business request to be served; the input end of the business service strategy decision submodule is respectively communicated with ... The output end of the service sequence decision submodule, the output end of the network resource acquisition unit and the output end of the service preprocessing function module are communicatively connected. The service service strategy decision submodule is used to calculate the demand for communication and computing resources of the service request to be served according to the type and attributes of the service request to be served, the available resource status information of each satellite node, and the service sequence of the service request to be served, generate a set of feasible service modes that meet each of the service requests to be served, calculate the probability of all feasible service modes of the service request to be served being selected according to the set of feasible service modes, and output the service policy of the service request to be served according to the probability of the feasible service mode being selected.

6. The large-scale satellite network service-level computing collaborative resource allocation device according to claim 5, characterized in that: The service pre-processing function module includes a service type analysis submodule and a service attribute analysis submodule; The input end of the business type analysis submodule is communicatively connected to the output end of the business request collection unit, the output end of the business type analysis submodule is communicatively connected to the input end of the business attribute analysis submodule, and the output end of the business attribute analysis submodule is communicatively connected to the input end of the business feature extraction function module and the input end of the business service policy decision submodule respectively; the business type analysis submodule is used to analyze the category of the received business request to be served and obtain the type of the business request to be served, and the business attribute analysis submodule is used to analyze the attributes of the business request to be served according to the type of the business request to be served and obtain the attributes of the business request to be served.

7. The large-scale satellite network service-level computing collaborative resource allocation device according to claim 6, characterized in that: The business feature extraction function module includes a business service relationship construction submodule and a business feature embedding submodule; The input end of the business service relationship construction submodule is communicatively connected to the output end of the business attribute analysis submodule, the output end of the business service relationship construction submodule is communicatively connected to the input end of the business feature embedding submodule, and the output end of the business feature embedding submodule is communicatively connected to the input end of the business service sequence decision submodule; the business service relationship construction submodule is used to calculate the edge attributes established to connect the business requests to be served according to the attributes of the business requests to be served, and generate a business service relationship graph corresponding to each type of business requests to be served, and the business feature embedding submodule is used to extract the service requirement characteristics of the business requests to be served according to the business service relationship graph.

8. The large-scale satellite network service-level computing collaborative resource allocation device according to claim 7, characterized in that: The business service sequence decision submodule includes a business encoding unit and a business decoding unit; The input end of the service encoding unit is communicatively connected to the output end of the service feature embedding submodule, the input end of the service encoding unit is also communicatively connected to the output end of the network resource acquisition unit, the output end of the service encoding unit is communicatively connected to the input end of the service decoding unit, and the output end of the service decoding unit is communicatively connected to the input end of the service service policy decision submodule; the service encoding unit is used to encode the service request to be served according to the service demand characteristics of each service request to be served and the available resource status information of each satellite node, and generate the network service status of the service request to be served; the service decoding unit is used to calculate the probability of the service request to be served according to the network service status of the service request to be served, and determine the service order of the service request to be served.

9. The large-scale satellite network service-level computing collaborative resource allocation device according to claim 8, characterized in that: The business service strategy decision submodule includes a decision set generation unit and a service strategy decision unit; The input end of the decision set generation unit is communicatively connected to the output end of the business service relationship construction submodule, the input end of the decision set generation unit is also communicatively connected to the output end of the business decoding unit, and the output end of the decision set generation unit is communicatively connected to the input end of the service policy decision unit; the decision set generation unit is used to generate a set of feasible service modes that meet each of the business requests to be served according to the service order of the business requests to be served, and the service policy decision unit is used to calculate the probability of all feasible service modes of the business request to be served being selected according to the set of feasible service modes, and output the service policy of the business request to be served according to the probability of the feasible service mode being selected.