A delay deterministic space-based routing method and system based on computing power perception space-time expansion graph and resource reservation

Abstract

The application provides a delay deterministic space-based routing method and system based on computing power perception space-time expansion graph and resource reservation, comprising: constructing a hierarchical and domain satellite network architecture, establishing a resource measurement system; generating a binding identifier corresponding to a service identifier for a satellite node, and establishing a service capability mapping relationship between the service identifier and the satellite node; introducing computing edges, storage edges and communication edges according to time copies to construct a computing power perception space-time expansion graph; screening candidate computing power nodes; generating candidate computing transmission paths based on the candidate computing power nodes, and establishing a multi-task computing transmission joint optimization model; solving the multi-task computing transmission joint optimization model, implementing resource reservation according to the time slice of the actual occupied resource of the selected candidate computing transmission path, and maintaining resource state consistency through resource release and rolling update. The application can coordinate multi-task resource competition, provide end-to-end delay deterministic guarantee for the admitted tasks, and improve the utilization efficiency of on-board resources.

Description

Technical Field

[0001] This invention belongs to the fields of satellite communication networks, space-based computing networks, and deterministic routing technology, specifically relating to a time-delay deterministic space-based routing method and system based on computing power-aware spatiotemporal extension maps and resource reservation. Background Technology

[0002] Low Earth Orbit (LEO) satellite networks consist of a large number of high-speed moving satellite nodes, resulting in a highly dynamic network topology. Satellite orbital parameters, ephemeris information, and communication link status are typically available in advance, thus the network topology is predictable within a given planning timeframe. However, with the increasing computing power of satellites, in addition to data forwarding, satellite nodes can perform computational tasks such as image processing, target recognition, and data compression, requiring routing decisions to consider both communication links and computing power distribution.

[0003] In scenarios such as satellite telemetry, tracking, and command (TT&C), disaster monitoring, and emergency rescue, tasks typically have clearly defined deadlines. Their operational effectiveness depends on whether the task data can be transmitted, computed, and returned within the specified timeframe. Since node queuing, bandwidth contention, and computing load can all lead to task timeouts, it is necessary to conduct a feasibility assessment before task execution and provide deterministic latency guarantees for accepted tasks through path planning and time-slice resource reservation.

[0004] Existing satellite routing methods primarily select transmission paths based on link connectivity, transmission delay, or link load, making it difficult to directly address the coupling between task transmission and onboard computing. Snapshot diagrams can discretize dynamic topology into multiple time slices, but they lack sufficient description of cross-time-slice data caching and continuous computation processes. Traditional spatiotemporal extension diagrams, building upon snapshot diagrams, connect communicable nodes within the same time slice via communication edges and the same nodes in adjacent time slices via storage edges to describe data transmission and caching waiting processes. However, this model does not uniformly characterize the heterogeneous computing capabilities, computing service types, computation processing periods, and corresponding resource consumption of satellite nodes. Some studies further introduce computing nodes or computing functions into the time extension diagram, but these are mainly used to describe single task flows or single computation processes, lacking a complete description of computing service types, the duration of computation processes, and the time-slice occupancy of communication, computing, and storage resources by multiple tasks.

[0005] In terms of computational power-aware scheduling, existing methods mainly fall into three categories: First, computation latency is set as node weight, and transmission and propagation latency is set as link weight, searching for computation-propagation paths from "task source node—computation node—destination node" respectively; second, the MEO controller aggregates link and computational power status to uniformly select computation nodes and transmission paths; and third, deep reinforcement learning or graph neural networks are used to generate scheduling strategies based on real-time topology, node load, and task requirements. However, these methods mostly optimize for single-task latency, average latency, or long-term gains, typically making decisions on a task-by-task basis based on the current state. They do not explicitly record the continuous occupation of link bandwidth, computation, and storage resources by tasks in subsequent time slices, nor do they have a mechanism for inheriting and releasing reserved resources for existing tasks. Therefore, resource duplication may occur when multiple tasks run concurrently, causing the planned path to fail to meet the task latency limit during actual execution.

[0006] Different satellite nodes have varying processor types, storage capacities, and communication capabilities, and node resource utilization changes as tasks are executed. Selecting computing nodes solely based on static resource capabilities may assign tasks to nodes with high loads; while making decisions based solely on dynamic states such as current resource utilization and task queue lengths fails to reflect the basic service capabilities of nodes. Furthermore, resources in hierarchical satellite networks are distributed across different nodes and management areas, and existing methods lack a mechanism for mapping computing service needs to node capabilities, resulting in a large range of candidate nodes and low computational power routing efficiency.

[0007] When multiple tasks arrive within the same planning time domain, they will compete for link bandwidth, node computing resources, and storage resources. Routing and selecting computing nodes sequentially based on task arrival order can easily lead to earlier-arriving tasks occupying critical resources, leaving subsequent tasks with no feasible paths. If resource usage within a specific time slice is not recorded, subsequent decisions may duplicate resources already used by active tasks, causing the planned path to fail to meet deterministic latency requirements during actual execution. Even if a path meets current resource constraints at the time of generation, if resources planned for its use in subsequent time slices are not reserved, newly arriving tasks cannot avoid occupying the same link or node resources again, delaying the transmission, waiting, or computation processes of the originally planned path.

[0008] Therefore, this application designs a space-based routing method based on a computing power-aware spatiotemporal extension map and resource reservation. It incorporates task transmission, computation processing, and data queuing into a unified model, and combines satellite node service capabilities, dynamic resource status, and existing resource reservations to conduct multi-task joint decision-making, providing end-to-end latency guarantees for multiple tasks, thereby improving the efficiency of space-based computing power routing and network resource utilization. Summary of the Invention

[0009] The purpose of this invention is to provide a time-deterministic space-based routing method and system based on computing power-aware spatiotemporal extension maps and resource reservation, so as to solve the problems that existing space-based routing methods are difficult to uniformly describe the task transmission, computing processing and data waiting process, and have insufficient executability of planned paths under multi-task competition conditions.

[0010] A time-deterministic space-based routing method based on computing power-aware spatiotemporal extension maps and resource reservation includes: A hierarchical, domain-specific satellite network architecture based on software-defined networking is constructed using GEO, MEO, and LEO satellite layers. Based on this architecture, orbital information, link status, and resource status of satellite nodes within the planning time domain are acquired. The planning time domain is discretized into multiple time slices, and a rolling planning window is established starting from the current decision time slice. A resource measurement system is established for heterogeneous onboard computing, storage, and communication resources to characterize the onboard resource status within each time slice. Service identifiers are defined using computing service types. Based on the static resource capabilities and service matching results of satellite nodes in the satellite network architecture, binding identifiers corresponding to the service identifiers are generated for each satellite node, and a binding relationship is established between the service identifier and the satellite node. The service capability mapping relationship between satellite nodes is established. Within the planning time window, computing edges, storage edges, and communication edges are introduced based on the time replicas of satellite nodes in different time slices to construct a spatiotemporal expansion map for computing power awareness. Based on the service identifier of the arriving task, satellite nodes with corresponding binding identifiers are queried, and candidate computing power nodes are screened by combining remaining available resources and spatiotemporal reachability. Based on the candidate computing power nodes, candidate computing and transmission paths are generated. Based on the candidate computing and transmission paths, a multi-task computing and transmission joint optimization model is established. The multi-task computing and transmission joint optimization model is solved to determine the task acceptance result and the candidate computing and transmission path selection result. Resource reservation is implemented according to the time slice in which the selected candidate computing and transmission paths actually occupy resources, and resource status consistency is maintained through resource release and rolling updates.

[0011] Preferably, in the satellite network architecture: the GEO satellite layer and the MEO satellite layer are the first-level control layer and the second-level control layer, respectively, and the LEO satellite layer is the data layer; the first-level control layer is used to maintain the global network status and implement cross-domain policy coordination; the second-level control layer is used to aggregate the status information of LEO satellite nodes within its region, maintain the regional computing power status table, and perform candidate path generation, multi-task computation and transmission joint optimization, and resource reservation control; the LEO satellites are used to perform task forwarding, computation processing, and status notification.

[0012] Preferably, the method for establishing a resource measurement system includes: dividing onboard resources into node resources and link bandwidth resources and modeling them separately; wherein, the node resources include computing power resources and storage resources, and the computing power resources include four categories: general computing power, parallel computing power, artificial intelligence computing power, and reconfigurable computing power; using a hybrid computing power method combining static and dynamic indicators to characterize the resource status; wherein, the static indicators are used to characterize the basic service capabilities of satellite nodes, including the capacity of each type of computing power, memory capacity, storage capacity, and link bandwidth capacity; the dynamic indicators are used to characterize the resource occupancy status within each time slice, including the utilization rate of each type of computing power, memory occupancy rate, storage occupancy rate, and bandwidth occupancy rate; introducing resource reservation to characterize the amount of resources allocated but not yet executed in multi-task scenarios; and calculating the remaining available resources of satellite nodes and the remaining available bandwidth of links within each time slice based on the static indicators, dynamic indicators, and resource reservation, thus completing the construction of the resource measurement system.

[0013] Preferably, the method for establishing a service capability mapping relationship between service identifiers and satellite nodes includes: classifying service identifiers according to the type of computing service; associating each service identifier with a core resource dimension, an important resource dimension, a resource preference weight, a minimum static resource requirement, and a service capability scoring threshold; for any satellite node, when the core resource capacity meets the minimum static resource requirement, and the service matching score determined based on the static resource capability and the resource preference weight is not lower than the service capability scoring threshold, generating a binding identifier between the satellite node and the corresponding service identifier, and establishing a service capability mapping relationship between the service identifier and the satellite node.

[0014] Preferably, the method for constructing a computing power-aware spatiotemporal expansion graph includes: introducing a rolling planning window, within which time replicas of satellite nodes under different time slices constitute a point set of the computing power-aware spatiotemporal expansion graph; each time replica node in the point set carries the resource status information, service binding identification information, and basic satellite operation status information of the satellite node in the corresponding time slice; based on the point set, introducing communication edges, storage edges, and computing edges to construct an edge set of the computing power-aware spatiotemporal expansion graph; wherein, the communication edge connects time replica nodes with available communication links in the same time slice, used to characterize the data transmission process between nodes; the storage edge connects time replica nodes of the same satellite node in adjacent time slices, used to characterize the cross-time slice caching process of task data; the computing edge is set on satellite nodes with corresponding binding identifications and remaining resources that meet the task processing requirements, used to characterize the process of the satellite node providing corresponding computing services; through the point set and edge set, a computing power-aware spatiotemporal expansion graph that uniformly describes the transmission, storage, and computing processes is constructed.

[0015] Preferably, the method for generating candidate computation-transfer paths includes: obtaining tasks arriving within the current decision time slice, wherein the task attributes include source node, destination node, service type, input data volume, output data volume, maximum tolerable end-to-end latency, task computation workload vector, link bandwidth reservation requirements, and computing node resource requirements; based on the service type of the task, querying the set of satellite nodes with the corresponding binding identifier in the computing power perception spatiotemporal expansion map, and filtering by combining the remaining available resources and link spatiotemporal reachability of each satellite node in the rolling planning window, eliminating nodes that do not meet the task resource requirements or auxiliary constraint index thresholds, to obtain a set of candidate computing power nodes; for each candidate computing power node in the set of candidate computing power nodes, using the Dijkstra algorithm to search for the minimum latency input path from the source node of the task to the candidate computing power node, and the minimum latency output path from the candidate computing power node to the task destination node; sequentially connecting the minimum latency input path, the computation processing process corresponding to the candidate computing power node, and the minimum latency output path to generate a candidate computation-transfer path for the task; traversing all candidate computing power nodes to obtain a set of candidate computation-transfer paths for the task.

[0016] Preferably, the method for establishing a multi-task computation-transfer joint optimization model includes: obtaining a set of tasks arriving at the same routing control node within the current decision time slice and a set of candidate computation-transfer paths corresponding to each task in the task set; setting a path selection variable and a task acceptance variable for each task; the path selection variable is used to indicate whether the task selects the corresponding candidate computation-transfer path; the task acceptance variable is used to indicate whether the task is accepted; calculating the relative delay of each candidate computation-transfer path based on the end-to-end delay and the maximum tolerable end-to-end delay of each task's candidate computation-transfer path, and determining the maximum relative delay among the accepted tasks; and establishing a multi-task computation-transfer joint optimization model with the optimization objective of minimizing the weighted sum of task rejection penalty, maximum relative delay, and resource matching cost.

[0017] Preferably, the multi-task computation-transmission joint optimization model includes the following constraints: task path uniqueness constraint, used to ensure that each accepted task uniquely selects a candidate path, and unaccepted tasks do not select a path; task delay determinism constraint, used to ensure that the end-to-end relative delay of accepted tasks does not exceed 1; maximum relative delay constraint, used to define the maximum relative delay variable of accepted tasks, wherein the maximum relative delay is not less than the relative delay of any candidate path of accepted task; communication link bandwidth constraint, used to ensure that the total bandwidth reserved by all accepted tasks on each link within the same time slice does not exceed the remaining available bandwidth of the corresponding link; computing node resource constraint, used to ensure that the total computing resource occupied by all accepted tasks on each satellite node within the same time slice does not exceed the remaining available computing resources of the corresponding node; and node storage capacity constraint, used to ensure that the total storage resource occupied by all accepted tasks on each satellite node within the same time slice does not exceed the remaining available storage capacity of the corresponding node.

[0018] This invention also provides a time-deterministic space-based routing system based on a computing power-aware spatiotemporal expansion map and resource reservation, used to implement the method, comprising: a satellite network architecture construction module, used to construct a layered and domain-specific satellite network architecture based on software-defined networks using GEO, MEO, and LEO satellite layers; a time-slice partitioning module, used to obtain the orbital information, link status, and resource status of satellite nodes within the planning time domain based on the satellite network architecture, discretely partition the planning time domain into multiple time slices, and establish a rolling planning window starting from the current decision time slice; a resource measurement system construction module, used to establish a resource measurement system for heterogeneous onboard computing power resources, storage resources, and communication resources, characterizing the onboard resource status within each time slice; and a service capability mapping module, used to classify service identifiers based on computing service types, and generate binding identifiers corresponding to the service identifiers for satellite nodes based on the static resource capabilities and service matching results of satellite nodes in the satellite network architecture. The system identifies and establishes a service capability mapping relationship between service identifiers and satellite nodes; a spatiotemporal expansion graph construction module is used to construct a computing power awareness spatiotemporal expansion graph within the planned time window, based on the time replicas of satellite nodes under different time slices, by introducing computing edges, storage edges, and communication edges; a candidate computing power node screening module is used to query satellite nodes with corresponding binding identifiers based on the service identifiers of arriving tasks, and screen candidate computing power nodes in combination with remaining available resources and spatiotemporal reachability; a candidate computing-transfer path generation module is used to generate candidate computing-transfer paths based on the candidate computing power nodes; an optimization model construction module is used to establish a multi-task computing-transfer joint optimization model based on the candidate computing-transfer paths; a model solving module is used to solve the multi-task computing-transfer joint optimization model to determine the task acceptance result and the candidate computing-transfer path selection result; and a state update module is used to implement resource reservation according to the time slices in which the selected candidate computing-transfer paths actually occupy resources, and maintain the consistency of resource state through resource release and rolling updates.

[0019] Preferably, the candidate computation-transmission path generation module, the optimization model construction module, the model solving module, and the state update module are deployed on the secondary control node of the MEO satellite layer; the secondary control node is used to aggregate the state information of LEO satellite nodes in its region, maintain the regional computing power state table, and perform candidate computation-transmission path generation, multi-task computation-transmission joint optimization, and resource reservation control.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention uses a computing power-aware spatiotemporal extension graph to uniformly describe the task transmission, computation processing, and data waiting processes, enabling the end-to-end latency and resource consumption of candidate paths to be predetermined during the path generation stage. This invention establishes a mapping relationship between computing services and satellite static capabilities through service identifiers and binding identifiers, and then combines this with dynamic resource status filtering of candidate nodes, thereby reducing the path search range. This invention performs multi-task joint optimization on a finite set of candidate paths, uniformly determining task acceptance results and path selection results, and coordinating resource competition among different tasks. This invention binds link bandwidth and node resource reservations to specific time slices, and maintains resource status consistency through resource release, reservation inheritance, and local replanning, thereby improving the executability of planned paths and providing end-to-end latency guarantees for accepted tasks. Attached Figure Description

[0021] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating the delay-deterministic space-based routing method based on spatiotemporal extension graphs and resource reservation provided in an embodiment of the present invention; Figure 2 A schematic diagram of a hierarchical and domain-specific satellite network structure based on a software-defined network architecture provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computing power-aware spatiotemporal extension map provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of time-slice resource reservation and rolling update for a candidate computation path provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a time-deterministic space-based routing system based on a computing power-aware spatiotemporal extension map and resource reservation, provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Example 1 like Figure 1 As shown, a time-deterministic space-based routing method based on computing power-aware spatiotemporal extension maps and resource reservation includes: S1: Utilizing the GEO, MEO, and LEO satellite layers, a hierarchical, domain-specific satellite network architecture based on software-defined networks is constructed; a further implementation method is as follows: Figure 2 As shown, in the satellite network architecture: The GEO satellite layer and MEO satellite layer are the first-level control layer and the second-level control layer, respectively, while the LEO satellite layer is the data layer. The primary control layer is used to maintain the global network state and implement cross-domain policy coordination; The secondary control layer is used to aggregate the status information of LEO satellite nodes within its region, maintain the regional computing power status table, and perform candidate path generation, multi-task computation-transmission joint optimization, and resource reservation control. LEO satellites are used for mission relay, computation processing, and status notification.

[0026] See Figure 1 and Figure 2 The method described in this embodiment can be executed by a routing control node in a hierarchical software-defined satellite network. The routing control node obtains the orbit, link, computing, storage, and resource reservation status of satellite nodes within its jurisdiction, generates candidate computation and transmission paths for the task, and issues routing configuration and time-slice resource reservation instructions based on the joint optimization results. To establish a basic graph model that can reflect the dynamic evolution of the satellite network, this embodiment first discretizes the satellite network topology in the planning time domain. In this embodiment, the low-orbit satellite operates along a predetermined orbit, and the node positions and link reachability relationships within the planned time domain can be predicted based on orbital parameters or ephemeris information.

[0027] S2: Based on the satellite network architecture, acquire the orbital information, link status, and resource status of satellite nodes within the planning time domain. Discretely divide the planning time domain into multiple time slices, and establish a rolling planning window starting from the current decision time slice. Specifically, determine the available links for each time slice based on node location, geometric visibility, communication distance, and antenna pointing conditions. Divide the planning time domain into multiple time slices at equal intervals, and establish a rolling planning window using the current decision time slice and a predetermined number of subsequent time slices.

[0028] S3: Establish a resource measurement system for heterogeneous onboard computing, storage, and communication resources to characterize the onboard resource status within each time slice; a further implementation method is that the method for establishing the resource measurement system includes: S31: Onboard resources are divided into node resources and link bandwidth resources, which are modeled separately. Node resources include computing power resources and storage resources. Computing power resources include four categories: general computing power, parallel computing power, artificial intelligence computing power, and reconfigurable computing power. The status of node resources includes static resource capacity, dynamic resource utilization rate, and resource reservation within the time slice. The remaining available resources for each resource dimension within the time slice are obtained by subtracting the resource occupancy determined according to the dynamic resource utilization rate and the existing resource reservation from the corresponding static resource capacity. The status of link resources includes link bandwidth capacity, dynamic bandwidth utilization rate, and bandwidth reservation within the time slice. The remaining available bandwidth of the link within the time slice is obtained by subtracting the bandwidth occupancy determined according to the dynamic bandwidth utilization rate and the existing bandwidth reservation from the link bandwidth capacity.

[0029] Specifically, the computing power in a computing network needs a unified description of heterogeneous computing resources to support subsequent resource mapping, scheduling, and routing decisions. Space-based computing networks have dispersed resources, and heterogeneous computing resources are diverse, which can be categorized by function into computing resources, storage resources, and communication resources. This embodiment models node resources and link resources separately, using resource metrics as shown in Table 1, categorizing them into node resources... Link bandwidth resources Node resources mainly include computing power resources. With storage resources .

[0030] Table 1 The specific representation method is as follows: ; ; ; Among them, computing resources It mainly consists of four parts: general-purpose computing power, parallel computing power, artificial intelligence (AI) computing power, and reconfigurable computing power. General-purpose computing power is primarily handled by CPUs for performing logical operations; parallel computing power is handled by GPUs for satellite image processing; AI computing power is primarily handled by NPUs / TPUs / GPUs, with simple logic but computationally intensive processing; reconfigurable computing power is handled by FPGAs for communication signal processing or AI accelerators; and storage resources... It mainly consists of memory capacity and storage capacity.

[0031] S32: A hybrid computational power method combining static and dynamic indicators is used to characterize resource status. Static indicators are used to describe the basic service capabilities of satellite nodes, including the capacity of various types of computing power (CPU logic computing power, GPU / NPU / TPU intelligent computing power and parallel computing power, FPGA reconfigurable computing power), memory capacity, storage capacity, and link bandwidth capacity. For any satellite node... Its static capacity resource vector is ,and For any satellite node With satellite nodes The link bandwidth capacity between them is .

[0032] Dynamic metrics are used to characterize resource utilization within each time slice, including utilization rates of various computing power types, memory utilization (CPU utilization, GPU utilization, NPU / TPU utilization, FPGA logic resource utilization), storage utilization, and bandwidth utilization; for any time slice Inside, satellite nodes The dynamic resource utilization rate is defined as And satisfy , For any satellite node With satellite nodes The dynamic utilization rate of link bandwidth capacity between them And satisfy .

[0033] S33: Introduce resource reservation to characterize the amount of resources allocated but not yet executed in a multi-task scenario; the dynamic resource occupancy is used to characterize the actual resource occupancy formed within the corresponding time slice, and the resource reservation is used to characterize the resource commitment that has been allocated to the accepted task but has not yet been converted into actual occupancy; the same resource occupancy of the same task is not included in the dynamic resource occupancy and resource reservation at the same time.

[0034] S34: Based on static indicators, dynamic indicators, and resource reservations, calculate the remaining available resources of satellite nodes and the remaining available bandwidth of links within each time slice, and complete the construction of the resource measurement system.

[0035] In addition, two auxiliary constraint indicators are added for screening and penalty, namely satellite power margin. The unit is watts (W), satellite health. ,and Satellite health can comprehensively consider factors such as satellite temperature and failure risk. Unlike allocable resource indicators such as computing, storage, and bandwidth, these indicators characterize the basic operational status of satellite nodes and do not directly participate in resource capacity deduction. The types and number of auxiliary constraint indicators can be increased or decreased according to the satellite platform status and actual business needs, as shown in Table 2.

[0036] Table 2 In scenarios where multi-task computing node resources are reserved, satellite In time slice The resource vectors already reserved in the middle can be represented as ,in Reserve indicator variables for computing nodes. The representative node resource demand vector can be specifically represented as: ; In scenarios where multi-task link bandwidth resources are reserved, satellite nodes With satellite nodes The reserved bandwidth resources of the link between them can be represented as ,in To reserve indicator variables for calculating links, This represents the bandwidth resource demand, which can be specifically expressed as: ; Based on the above description of satellite node resources and link bandwidth resources, the node time slice is calculated. The formula for calculating the remaining available resources is: ; Calculate the remaining available bandwidth of the link ; Considering the different dimensions and value ranges of various resource indicators, to avoid their unbalanced impact on node selection and resource matching results, each resource indicator is normalized and mapped to a unified value range. For any resource quantity to be normalized... Normalization can be performed using the following formula: . This represents a preset, extremely small positive number, used to avoid the normalization formula having a denominator of zero when the resource quantities of each satellite node are the same under the same resource dimension.

[0037] S4: Utilize computing service types to classify service identifiers (SIDs). Based on the matching results of the static resource capabilities and services of satellite nodes in the satellite network architecture, generate binding identifiers (BIDs) corresponding to the service identifiers for each satellite node, and establish a service capability mapping relationship between the service identifiers and satellite nodes. A further implementation method for establishing the service capability mapping relationship between service identifiers and satellite nodes includes: Service identifiers are categorized according to computing service type. Each service identifier is associated with a core resource dimension, an important resource dimension, a resource preference weight, a minimum static resource requirement, and a service capability scoring threshold. For any satellite node, when the core resource capacity meets the minimum static resource requirement, and the service matching score determined based on the static resource capability and resource preference weight is not lower than the service capability scoring threshold, a binding identifier is generated between the satellite node and the corresponding service identifier, and a service capability mapping relationship is established between the service identifier and the satellite node. The service identifier categorization is shown in Table 3. Table 3 Specifically, a Service Identifier (SID) and a Binding Identifier (BID) are introduced. The former describes the type of computing power service required by the task, while the latter describes whether a satellite node has the basic capability to support a certain type of service, enabling the MEO control layer to quickly filter candidate computing power nodes. A set of service identifiers is constructed. For each type of service Its services can be described as ;in, Indicates the first The core resource dimensions of the service class; Indicates the first A collection of key resource dimensions for class services; Indicates the first The resource preference weight vector of the service class satisfies ; This indicates the minimum static resource requirements for this service; This represents the service capability scoring threshold; specifically, the weights of the core resource dimensions also satisfy a lower bound condition: , Indicates service Minimum weighting percentage for core resources.

[0038] Based on the above description of resource quantities, and according to the normalization formula, the static normalized capability vector is calculated. Further calculations of satellites For services The static service matching score is When the satellite simultaneously meets the static hard constraints If the score is higher than the service matching score threshold, then match it with the service. Binding is performed to form a binding identifier. .

[0039] The service can be obtained through the above method. The corresponding statically bound satellite set is single satellite It can be bound to multiple services, and its binding identifier set is: ; To avoid repeatedly reporting node static capabilities, service binding relationships, and dynamic resource status in each time slice, which would incur significant control overhead, this embodiment employs a hierarchical computing power notification mechanism. MEO controller set ,satellite Attribution mapping is Each LEO satellite can only establish a mapping relationship with one MEO controller at any given time slice. The computing power announcement is sent by the LEO satellite node to its associated MEO controller. The MEO controller uses this announcement to maintain a regional computing power status table and provides status information for candidate computing power node selection, updating the computing power awareness spatiotemporal expansion map, and resource reservation.

[0040] The computing power notifications include static registration notifications, periodic status notifications, and event-triggered notifications. Static registration notifications report the node's basic capabilities, service type, and BID binding relationship when a satellite node joins the network, its configuration changes, or its service capabilities change; they are not reported repeatedly when there are no changes. Periodic status notifications report resource utilization, existing reserved resources, and remaining available resources according to a preset period, used to update the regional computing power status table; the notification period can be set to an integer multiple of the time slice length. Event-triggered notifications are reported immediately when there are sudden changes in resources, service capabilities, insufficient remaining power, or abnormal health status; the MEO controller updates the relevant status and performs necessary local replanning accordingly. The dynamic resource status of the current decision time slice is obtained using the latest periodic status notification or event-triggered notification. For subsequent time slices in the rolling planning window, the corresponding remaining available resources are determined based on the current resource status, the planned execution status of existing tasks, and the established future resource reservations; updates can also be made based on load forecasting results if necessary.

[0041] S5: Within the planned time window, based on the time replicas of satellite nodes in different time slices, introduce computing edges, storage edges, and communication edges to construct a computing power-aware spatiotemporal expansion graph; such as... Figure 3As shown, the traditional Time-Expanded Graph (TEG) can be expanded into a Computing-aware Time-Expanded Graph (C-TEG). C-TEG further introduces computation edges on top of the existing transmission and storage edges of the traditional TEG, and adds computing power status, service status, and resource reservation status to each time replica node. A further implementation method for constructing the Computing-aware Time-Expanded Graph includes: S51: Introducing a rolling planning window, based on the node location, available links, node resources, and link resources of each time slice within the rolling planning window, time replicas of satellite nodes under different time slices constitute a point set of a computing power-aware spatiotemporal expansion map; each time replica node in the point set carries the resource status information, service binding identification information, and basic satellite operation status information of the satellite node within the corresponding time slice; specifically, introducing a rolling planning window. ,in The length is the window length; within the scrolling planning window, C-TEG can be represented as... , This represents the set of time replica nodes of C-TEG. Represent the edge set of C-TEG. Note that... For the current decision-making time slice, This is the time slice for the copy in the scrolling window.

[0042] The point set is constructed by defining the time replicas of satellite nodes in different time slices within the rolling window as the point set of the C-TEG. and carries the following status information .

[0043] S52: Based on the point set, a set of edges is constructed by introducing communication edges, storage edges, and computing edges to build a spatiotemporal extension graph of computing power awareness; wherein, the communication edge connects time replica nodes with available communication links within the same time slice, and is used to characterize the data transmission process between nodes; the storage edge connects time replica nodes of the same satellite node in adjacent time slices, and is used to characterize the cross-time slice caching process of task data; the computing edge is set on satellite nodes with corresponding binding identifiers and remaining resources that meet the task processing requirements, and is used to characterize the process by which the satellite node provides the corresponding computing services.

[0044] Specifically, an edge set is constructed, including communication edges, storage edges, and computation edges, represented as follows: The communication edge is defined as follows: The set of communication edges can be represented as ,in, Indicates time slice The set of available links below, each communication edge The link state information that can be carried is , Indicates time slice Next node With nodes The link distance between them; the storage edge is defined as The storage edge set can be represented as For the candidate computing nodes that the task actually passes through, the starting point of the computing edge corresponds to the time replica when the task data arrives at the computing node, and the ending point corresponds to the time replica when the computing process is completed. The computing edge is associated with the computing resource usage in each time slice during the computing period.

[0045] S53: By using a set of points and edges, a unified spatiotemporal extension graph for computing power awareness is constructed to describe the transmission, storage, and computation processes. If satellite nodes... Service bound ,Right now This indicates that the node is capable of providing services. Basic computing power for time slices Time replica node under the scrolling window Corresponding virtual computation edges can be constructed. , indicating satellite node In time slice It has the service Provides computational processing capabilities; the set of computational edges can be represented as .

[0046] The C-TEG can be constructed by following the steps described above.

[0047] S6: Receive tasks arriving within the current decision time slice. Based on the service identifier of the arriving task, query satellite nodes with corresponding binding identifiers, and filter candidate computing power nodes by combining remaining available resources and spatiotemporal reachability; remaining available resources are obtained by subtracting dynamic resource occupancy and existing resource reservation from static resource capacity.

[0048] Specifically, let the current decision-making time slice be... Tasks arriving at the same MEO controller within that time slice constitute a task batch. Its task attributes are .

[0049] in, and These represent the source node and the destination node, respectively; and the relationship between the input and output data volumes. ,in For the task The output data scaling factor of the corresponding computing service; The maximum tolerable end-to-end delay for a task is expressed by the formula: , Indicates the impact coefficient of input data volume on maximum tolerable latency; link bandwidth reservation. ; Calculate the workload vector for the task, where ; To compute the node resource requirement vector, and The memory requirement is Storage requirements , intermediate data volume .

[0050] For any task Its service type can be mapped to After the mission requirement information is sent to the MEO controller, the MEO controller can filter out the satellite nodes that meet the requirements based on the mission's service identifier type. Upon mission arrival, the MEO controller first... Service types, query services available Satellite collection with basic carrying capacity .

[0051] Based on the set of satellite nodes obtained from the previous query that have established BID bindings with the task service identifier and meet the static capability requirements, and further considering the dynamic remaining computing, memory and storage resources, link reachability, and basic status indicators of each node within the current rolling planning window, nodes that do not meet the task resource requirements or auxiliary constraint indicator thresholds are eliminated, resulting in a set of task-level candidate computing power nodes. .

[0052] In the candidate set China needs to further consider... Satellite nodes are sorted using Euclidean distance to avoid wasting high-performance computing resources by assigning tasks to nodes with resources far exceeding their actual needs. Specifically, for each task... Candidate paths Let its corresponding candidate computing power node be First, the task resource requirement vector and the node remaining available resource vector are normalized to obtain the following results: and Through formula The conclusion is Euclidean distance, sorted from low to high, then selected. There are 10 feasible computing nodes, among which... This is a preset upper limit for the number of candidate computing nodes; when the number of feasible computing nodes is less than... At that time, all feasible computation nodes are retained.

[0053] S7: Based on candidate computing power nodes, generate candidate computing transmission paths; a further implementation method includes: The system retrieves tasks arriving within the current decision time slice. Task attributes include source node, destination node, service type, input data volume, output data volume, maximum tolerable end-to-end latency, task computation workload vector, link bandwidth reservation requirements, and computing node resource requirements. Based on the task's service type, it queries the set of satellite nodes with corresponding service type binding identifiers in the computing power awareness spatiotemporal expansion map. This is then filtered based on the remaining available resources and link spatiotemporal reachability of each satellite node within the rolling planning window, eliminating nodes that do not meet the task resource requirements or auxiliary constraint thresholds, resulting in a candidate computing power node set. For each candidate computing power node in the set, the Dijkstra algorithm is used to search for the minimum latency input path from the task's source node to the candidate computing power node, and the minimum latency output path from the candidate computing power node to the task's destination node. The minimum latency input path, the computation process corresponding to the candidate computing power node, and the minimum latency output path are then sequentially connected to generate a candidate computation-transmission path for the task. Finally, all candidate computing power nodes are traversed to obtain the set of candidate computation-transmission paths for the task.

[0054] The resource matching distance is not used to determine whether candidate nodes meet the task resource requirements, but rather to measure the degree of resource suitability among candidate nodes that meet the basic resource constraints, and serves as the resource matching cost in the multi-task computation-transmission joint optimization model.

[0055] Specifically, based on choice For each candidate computing node, the MEO controller uses Dijkstra's algorithm to search for the minimum latency input path from the task source node to the candidate computing node and the minimum latency output path from the candidate computing node to the task destination node, based on the latency of the computation edge, storage edge, and communication edge in the current computing power-aware spatiotemporal expansion graph. The input path, computation process, and output path are then sequentially connected to generate a corresponding set of candidate computation-transmission paths. .

[0056] The task any candidate path Its end-to-end delay consists of transmission delay, propagation delay, computation and processing delay, and storage waiting delay, and can be expressed as: ; , ; ; ; , To store the set of waiting time slots; Among them, the storage waiting latency is distributed in discrete stages on the candidate path, and the duration of each stage is accurately calculated by using the difference between the planning end time and the planning start time.

[0057] In the MEO controller, a candidate path set is generated by individually calculating the path matched to each candidate computing node using Dijkstra's algorithm at the cost of edge latency. .

[0058] S8: Based on candidate computation-transfer paths, establish a multi-task computation-transfer joint optimization model; a further implementation method is that the multi-task computation-transfer joint optimization model is a mixed-integer linear programming model. The method for establishing the multi-task computation-transfer joint optimization model includes: obtaining the set of tasks arriving at the same routing control node within the current decision time slice and the set of candidate computation-transfer paths corresponding to each task in the task set; setting a path selection variable and a task acceptance variable, both of which are binary variables, for each task; the path selection variable is used to indicate whether the task selects the corresponding candidate computation-transfer path; the task acceptance variable is used to indicate whether the task is accepted; calculating the relative delay of each candidate computation-transfer path based on the end-to-end delay and the maximum tolerable end-to-end delay of each task's candidate computation-transfer path, and determining the maximum relative delay among the accepted tasks; establishing the multi-task computation-transfer joint optimization model with the optimization objective of minimizing the weighted sum of task rejection penalty, maximum relative delay, and resource matching cost.

[0059] The resource matching cost is determined based on the resource matching distance between the task's resource requirements and the remaining available resources of the candidate computing nodes.

[0060] A multi-task computation-transmission joint optimization model is established, with the optimization objective being to minimize the weighted sum of task rejection penalty, maximum relative latency, and resource matching cost, i.e., minimizing the comprehensive cost function. The candidate path selection variable is used to describe the final path decision of the task. and its candidate paths It has the following definition: The path selection variable is defined as: ; The task acceptance variable is defined as: ; To eliminate the impact of differences in data volume, computational workload, and maximum tolerable end-to-end latency among different tasks on absolute latency comparisons, the relative latency of candidate computation-transmission paths is defined as follows: The relative latency is the ratio of the end-to-end latency of the candidate computation path to the maximum tolerable end-to-end latency of the corresponding task. The maximum relative latency within a task batch is defined as follows: .

[0061] The multi-task candidate computation path set is the union of the candidate computation path sets for each task. The model uses task acceptance variables. Path selection variables , Indicates task If accepted, it means it was not accepted; Task Select candidate computation path Otherwise, it indicates that no selection was made. All tasks must satisfy... , , , First, configure the non-negative weights, and satisfy the following conditions: .

[0062] A further implementation method is that the multi-task computation-transmission joint optimization model includes the following constraints: Task path uniqueness constraint is used to ensure that each accepted task uniquely selects one candidate path, and unaccepted tasks do not select a path: ; Task delay deterministic constraints are used to ensure that the end-to-end relative delay of accepted tasks does not exceed 1, i.e., the maximum tolerable delay is satisfied. ; The maximum relative delay constraint defines the maximum relative delay variable for accepted tasks. The maximum relative delay is not less than the relative delay of any candidate path for accepted tasks. ; Communication link bandwidth constraints are used to ensure that all accepted tasks within the same time slice are connected to each link. The total bandwidth reserved on the connection shall not exceed the remaining available bandwidth of the corresponding link: Computation node resource constraints are used to ensure that the total computational resource usage of all accepted tasks on each satellite node within the same time slice does not exceed the remaining available computational resources of the corresponding node: Node storage capacity constraints are used to ensure that the total storage resource usage of all accepted tasks on each satellite node within the same time slice does not exceed the remaining available storage capacity of the corresponding node. Among them, the task rejection penalty, maximum relative latency, and resource matching cost characterize the joint optimization objective of multi-task flow computation and transmission from three aspects: task acceptance, latency guarantee, and resource adaptation, respectively. All three are processed in a normalized form to avoid scale bias caused by directly adding indices with different dimensions. Furthermore, since each normalized denominator is a known constant, the objective function still maintains a linear structure. Therefore, this optimization problem can be modeled as a mixed-integer linear programming model and solved using solvers such as Gurobi or CPLEX.

[0063] S9: Solve the multi-task compute-transfer joint optimization model to determine the task acceptance result and candidate compute-transfer path selection result. Specifically, the MEO controller generates a resource reservation instruction based on the execution time slice and resource occupancy associated with each communication edge, computing edge, and storage edge in the selected compute-transfer path, and issues it to the LEO satellite nodes and link scheduling entities involved in the path, reserving link bandwidth, node computing resources, and storage resources respectively within the corresponding time slices. LEO satellite nodes perform data forwarding, computation processing, and buffering according to the resource reservation instruction, and report the execution status back to their respective MEO controllers. Resource reservation is not implemented for candidate compute-transfer paths that are not selected.

[0064] After determining the optimal task acceptance and path selection results, the MEO controller performs resource reservation according to the time slices in which the selected candidate paths actually use resources. The candidate paths not only describe the transmission and computation process of the task in the C-TEG, but also specify the communication, computation, and storage resources occupied by the task in each time slice.

[0065] For the task Candidate paths Its link bandwidth usage Node resource usage All of these can be predetermined after candidate path generation and used as known parameters in the multi-task flow optimization model. Among them, task bandwidth requirements... The optimization model is determined by the amount of task data, the type of business, and the maximum tolerable latency. It only decides whether to accept the task and which candidate path to choose, without continuously optimizing the bandwidth reservation, thus ensuring the linear structure of the model.

[0066] S10: Reserve resources according to the time slice of the actual resources occupied by the selected candidate computing path, and maintain the consistency of resource status through resource release and rolling updates.

[0067] According to the resource allocation plan of the selected candidate compute-transfer paths in each time slice, reserve link bandwidth, node computing resources, and storage resources. When the rolling planning window advances, release the resource reservations for completed time slices, retain the resource reservations for unfinished tasks in subsequent time slices, and update the computing power awareness spatiotemporal expansion map. When a task is completed ahead of schedule or canceled, cancel the unexecuted resource reservations. When a link interruption, node failure, sudden change in resource status, or task execution timeout occurs, perform local replanning for the affected tasks. Repeat steps S6-S10 to continuously update C-TEG.

[0068] Specifically, after the MILP solution is completed, for those that satisfy... and For this task, the MEO controller issues resource reservation instructions within the corresponding time slice according to the selected candidate path. The current task batch is on the link. The newly added bandwidth reservation is .

[0069] The current task batch is at node The newly added number The reserved amount of this type of resource is Storage resources are reserved in the same way, based on the storage edge occupancy in the candidate paths. After reservation, the controller updates the remaining resource status of each time slice in the C-TEG. , .

[0070] Resource reservation is only for the computation-transmission path selected by the joint optimization model, and only within the corresponding time slices where data transmission, computation processing, or caching waiting is planned to be performed on that path, the link bandwidth, node computing resources, and node storage resources are reserved respectively; time slices where the corresponding resources are not used are not reserved, so as to avoid continuously occupying all relevant resources from mission acceptance to mission completion and improve the utilization rate of onboard resources.

[0071] Resource release includes three scenarios: normal release, early release, and abnormal replanning. Under normal circumstances, each resource reservation is bound to a specific time slice. When the corresponding time slice ends, the link bandwidth, computing resources, and storage resources within that time slice are automatically released and do not need to continuously occupy resources in subsequent time slices.

[0072] When the scrolling planning window changes from the current time slice Move to the next time slice When a task is completed, the MEO controller deletes the reserved records for completed time slices, while retaining the resource reservations for accepted but not yet completed tasks in future time slices. The future reservations for existing tasks are used as existing resource occupancy for the next round of optimization. Newly arriving tasks can only use the remaining resources after deducting existing reservations, thus avoiding redundant resource allocation between different decision cycles.

[0073] like Figure 4 As shown, when the scrolling planning window changes from the current time slice... Proceed to the next time segment When the MEO controller is deleted And the resource reservation records of previously executed time slices, retaining unfinished tasks in and resource reservations within subsequent time slices. These reserved resources serve as... For the next round of joint optimization of the current decision time slice, the existing resource occupancy is determined. Newly arriving tasks can only use the remaining available resources after deducting the dynamic resource occupancy and the existing resource reservation, thereby avoiding the duplication of resource allocation.

[0074] When a task is completed ahead of schedule, canceled, or a processing phase ends early, the controller cancels the unexecuted future reservations for that task and adds the corresponding link bandwidth, computing power, and storage resources back to the remaining available resource set for subsequent tasks. When abnormal situations such as link interruption, node failure, or task execution time exceeding the reserved time occur, the controller first updates the resource status in the C-TEG through event-triggered notifications. For affected tasks, while keeping the unaffected accepted tasks and their existing resource reservations unchanged, candidate compute-transfer paths are regenerated, and local joint optimization is performed on the affected tasks to redetermine the path selection results. Unexecuted resource reservations in the original failed paths of the affected tasks are canceled, and new time-slice resource reservations are established according to the newly selected candidate compute-transfer paths.

[0075] Example 2 like Figure 5 As shown, the present invention also provides a time-deterministic space-based routing system based on computing power-aware spatiotemporal extension maps and resource reservation, used to implement the method described in Embodiment 1, including: The satellite network architecture construction module is used to construct a layered and domain-specific satellite network architecture based on software-defined networking, utilizing GEO, MEO, and LEO satellite layers. The time-slice partitioning module, based on the satellite network architecture, acquires the orbital information, link status, and resource status of satellite nodes within the planning time domain, discretely divides the planning time domain into multiple time slices, and establishes a rolling planning window starting from the current decision time slice. The resource measurement system construction module, based on the satellite network architecture, establishes a resource measurement system for heterogeneous onboard computing, storage, and communication resources, characterizing the onboard resource status within each time slice. The service capability mapping module, based on computing service types, classifies service identifiers, generates binding identifiers corresponding to service identifiers for satellite nodes based on the static resource capabilities and service matching results of satellite nodes in the satellite network architecture, and establishes service relationships between service identifiers and satellite nodes. The system includes: a capability mapping relationship; a spatiotemporal expansion graph construction module, used to construct a computing power-aware spatiotemporal expansion graph within the planned time window by introducing computing edges, storage edges, and communication edges based on the time replicas of satellite nodes under different time slices; a candidate computing power node screening module, used to query satellite nodes with corresponding binding identifiers based on the service identifiers of arriving tasks, and to screen candidate computing power nodes by combining remaining available resources and spatiotemporal reachability; a candidate computing-transfer path generation module, used to generate candidate computing-transfer paths based on candidate computing power nodes; an optimization model construction module, used to establish a multi-task computing-transfer joint optimization model based on candidate computing-transfer paths; a model solving module, used to solve the multi-task computing-transfer joint optimization model to determine the task acceptance result and the candidate computing-transfer path selection result; and a state update module, used to implement resource reservation according to the time slices in which the selected candidate computing-transfer paths actually occupy resources, and to maintain resource state consistency through resource release and rolling updates.

[0076] A further implementation method involves deploying the candidate computation-transmission path generation module, the optimization model construction module, the model solving module, and the state update module on the secondary control node of the MEO satellite layer. The secondary control node is used to aggregate the state information of LEO satellite nodes within its region, maintain the regional computing power state table, and perform candidate computation-transmission path generation, multi-task computation-transmission joint optimization, and resource reservation control.

[0077] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A time-deterministic space-based routing method based on computing power-aware spatiotemporal extension maps and resource reservation, characterized in that, include: By utilizing the GEO, MEO, and LEO satellite layers, a hierarchical and domain-specific satellite network architecture based on software-defined networks is constructed. Based on the satellite network architecture, the orbital information, link status and resource status of satellite nodes within the planning time domain are obtained, the planning time domain is discretely divided into multiple time slices, and a rolling planning window is established with the current decision time slice as the starting point. Establish a resource measurement system for heterogeneous onboard computing, storage, and communication resources to characterize the status of onboard resources in each time slice; Service identifiers are categorized by computing service type. Based on the matching results of static resource capabilities and services of satellite nodes in the satellite network architecture, binding identifiers corresponding to the service identifiers are generated for satellite nodes, and a service capability mapping relationship between service identifiers and satellite nodes is established. Within the planned time window, based on the time replicas of satellite nodes under different time slices, computing edges, storage edges, and communication edges are introduced to construct a spatiotemporal expansion graph for computing power perception. Based on the service identifier of the arriving task, query the satellite nodes with the corresponding binding identifier, and filter candidate computing power nodes by combining the remaining available resources and spatiotemporal reachability; Based on the candidate computing power nodes, candidate computing transmission paths are generated; Based on candidate computation-transmission paths, a multi-task computation-transmission joint optimization model is established; Solve the multi-task compute-transfer joint optimization model to determine the task acceptance result and the candidate compute-transfer path selection result; Resource reservations are implemented based on the time slices of resources actually occupied by the selected candidate compute paths, and resource status consistency is maintained through resource release and rolling updates.

2. The method according to claim 1, characterized in that, In the satellite network architecture: The GEO satellite layer and MEO satellite layer are the first-level control layer and the second-level control layer, respectively, while the LEO satellite layer is the data layer. The first-level control layer is used to maintain the global network state and implement cross-domain policy coordination. The secondary control layer is used to aggregate the status information of LEO satellite nodes within its region, maintain the regional computing power status table, and perform candidate path generation, multi-task computing and transmission joint optimization, and resource reservation control. The LEO satellite is used to perform mission relay, computation processing, and status notification.

3. The method according to claim 1, characterized in that, Methods for establishing a resource measurement system include: Onboard resources are divided into node resources and link bandwidth resources, which are modeled separately. The node resources include computing power resources and storage resources. The computing power resources include four categories: general computing power, parallel computing power, artificial intelligence computing power, and reconfigurable computing power. A hybrid computing power method combining static and dynamic indicators is used to characterize resource status. The static indicators are used to describe the basic service capabilities of satellite nodes, including computing power capacity, memory capacity, storage capacity and link bandwidth capacity of various types. The dynamic indicators are used to describe the resource occupancy status within each time slice, including computing power utilization rate, memory occupancy rate, storage occupancy rate and bandwidth occupancy rate of various types. A resource reservation is introduced to represent the amount of resources allocated but not yet executed in a multi-task scenario; Based on the static indicators, dynamic indicators, and resource reservations, the remaining available resources of satellite nodes and the remaining available bandwidth of links within each time slice are calculated, thus completing the construction of the resource measurement system.

4. The method according to claim 1, characterized in that, Methods for establishing a service capability mapping relationship between service identifiers and satellite nodes include: Service identifiers are categorized according to computing service type, and each service identifier is associated with core resource dimensions, important resource dimensions, resource preference weights, minimum static resource requirements, and service capability scoring thresholds. For any satellite node, when the core resource capacity meets the minimum static resource requirements and the service matching score determined based on the static resource capacity and the resource preference weight is not lower than the service capability score threshold, a binding identifier is generated between the satellite node and the corresponding service identifier, and a service capability mapping relationship is established between the service identifier and the satellite node.

5. The method according to claim 1, characterized in that, Methods for constructing a computing power-aware spatiotemporal extension graph include: A rolling planning window is introduced, in which time replicas of satellite nodes under different time slices are used to form a point set of a computing power perception spatiotemporal expansion map; each time replica node in the point set carries the resource status information, service binding identification information and basic satellite operation status information of the satellite node in the corresponding time slice; Based on the aforementioned point set, communication edges, storage edges, and computation edges are introduced to construct an edge set for the computing power-aware spatiotemporal extension graph. The communication edges connect time replica nodes with available communication links within the same time slice, representing the data transmission process between nodes. The storage edges connect time replica nodes of the same satellite node in adjacent time slices, representing the cross-time-slice caching process of task data. The computation edges are set on satellite nodes with corresponding binding identifiers and remaining resources sufficient for task processing, representing the process by which the satellite node provides the corresponding computing services. By using the set of points and the set of edges, a unified spatiotemporal extension graph for computing power awareness that describes the transmission, storage and computation processes is constructed.

6. The method according to claim 1, characterized in that, Methods for generating candidate computational paths include: Obtain the tasks arriving within the current decision time slice. The task attributes include source node, destination node, service type, input data volume, output data volume, maximum tolerable end-to-end latency, task computation workload vector, link bandwidth reservation requirements, and computing node resource requirements. Based on the service type of the task, query the set of satellite nodes with the binding identifier corresponding to the service type in the spatiotemporal expansion map of computing power perception, and filter them by combining the remaining available resources and spatiotemporal reachability of each satellite node in the rolling planning window, and remove nodes that do not meet the task resource requirements or auxiliary constraint index thresholds to obtain a set of candidate computing power nodes. For each candidate computing node in the candidate computing node set, the Dijkstra algorithm is used to search for the minimum latency input path from the source node of the task to the candidate computing node, and the minimum latency output path from the candidate computing node to the task destination node. The minimum latency input path, the computation process corresponding to the candidate computing node, and the minimum latency output path are connected sequentially to generate a candidate computation path for the task. Traverse all candidate computing power nodes to obtain a set of candidate computing transmission paths for the task.

7. The method according to claim 1, characterized in that, Methods for establishing a multi-task computation-transmission joint optimization model include: Obtain the set of tasks arriving at the same routing control node within the current decision time slice, and the set of candidate computation and transmission paths corresponding to each task in the task set; For each task, a path selection variable and a task acceptance variable are set; the path selection variable is used to indicate whether the task selects the corresponding candidate computation path; the task acceptance variable is used to indicate whether the task is accepted. Based on the end-to-end delay and maximum tolerable end-to-end delay of the candidate computation-transmission paths for each task, calculate the relative delay of each candidate computation-transmission path and determine the maximum relative delay among the accepted tasks. A multi-task computation-transmission joint optimization model is established with the optimization objective of minimizing the weighted sum of task rejection penalty, maximum relative latency, and resource matching cost.

8. The method according to claim 7, characterized in that, The multi-task computation-transmission joint optimization model includes the following constraints: The task path uniqueness constraint is used to ensure that each accepted task uniquely selects a candidate path, and unaccepted tasks do not select a path. Task delay deterministic constraints are used to ensure that the end-to-end relative delay of accepted tasks does not exceed 1. The maximum relative delay constraint is used to define the maximum relative delay variable of the accepted task, wherein the maximum relative delay is not less than the relative delay of any candidate path of the accepted task; Communication link bandwidth constraints are used to ensure that the total bandwidth reserved for all accepted tasks on each link within the same time slice does not exceed the remaining available bandwidth of the corresponding link. Computation node resource constraints are used to ensure that the total computing resources occupied by all accepted tasks on each satellite node within the same time slice do not exceed the remaining available computing resources of the corresponding node. The node storage capacity constraint is used to ensure that the total storage resources occupied by all accepted tasks on each satellite node within the same time slice do not exceed the remaining available storage capacity of the corresponding node.

9. A time-deterministic space-based routing system based on computing power-aware spatiotemporal extension maps and resource reservation, used to implement the method described in any one of claims 1-8, characterized in that, include: The satellite network architecture building module is used to construct a layered and domain-specific satellite network architecture based on software-defined networks by utilizing the GEO satellite layer, MEO satellite layer, and LEO satellite layer. The time slice division module is used to obtain the orbit information, link status and resource status of satellite nodes within the planning time domain based on the satellite network architecture, to discretize the planning time domain into multiple time slices, and to establish a rolling planning window starting from the current decision time slice. The resource measurement system construction module is used to establish a resource measurement system for heterogeneous onboard computing resources, storage resources and communication resources, and to characterize the status of onboard resources in each time slice; The service capability mapping module is used to classify service identifiers based on computing service types, generate binding identifiers corresponding to the service identifiers for satellite nodes based on the static resource capabilities and service matching results of satellite nodes in the satellite network architecture, and establish a service capability mapping relationship between service identifiers and satellite nodes. The spatiotemporal extension graph construction module is used to construct a computing power-aware spatiotemporal extension graph by introducing computing edges, storage edges, and communication edges based on the time replicas of satellite nodes under different time slices within the planned time window. The candidate computing power node filtering module is used to query satellite nodes with corresponding binding identifiers based on the service identifier of the arriving task, and filter candidate computing power nodes in combination with the remaining available resources and spatiotemporal reachability. The candidate computing power path generation module is used to generate candidate computing power paths based on the candidate computing power nodes. The optimization model building module is used to establish a multi-task computation-transmission joint optimization model based on candidate computation-transmission paths. The model solving module is used to solve the multi-task computation-transmission joint optimization model and determine the task acceptance result and the candidate computation-transmission path selection result. The status update module is used to reserve resources according to the time slice of the resources actually occupied by the selected candidate computing path, and maintain the consistency of resource status through resource release and rolling updates.

10. The time-deterministic space-based routing system based on computing power-aware spatiotemporal extension map and resource reservation according to claim 9, characterized in that, The candidate computation-transmission path generation module, optimization model construction module, model solving module, and state update module are deployed on the secondary control node of the MEO satellite layer. The secondary control node is used to aggregate the state information of LEO satellite nodes in its region, maintain the regional computing power status table, and perform candidate computation-transmission path generation, multi-task computation-transmission joint optimization, and resource reservation control.

Images