A method, electronic device, and medium for updating inter-satellite laser topology.

Abstract

This invention discloses a method, electronic device, and medium for updating inter-satellite laser topology. The method updates the inter-satellite laser topology at specified time intervals. Each update first acquires the node characteristics of all laser nodes and the link characteristics of all laser links in the laser subnet to determine the set of visible laser links. Then, based on the set of visible laser links, a seed topology is constructed, and candidate laser topologies are generated. Finally, the optimal topology is selected from the candidate topologies for updating. This update method effectively constrains unnecessary laser link establishment while ensuring mission-level performance, thereby significantly reducing network energy consumption.

Description

Technical Field

[0001] This invention relates to the field of aerospace technology, and in particular to a method for updating inter-satellite laser topology, electronic equipment, and media. Background Technology

[0002] Spacecraft swarms typically refer to a distributed system composed of multiple heterogeneous or homogeneous spacecraft, such as satellites, space robots, and probes, which collaborate through perception, intelligent decision-making, autonomous control, and information exchange. With the development of technologies such as microsatellite platforms, attitude and orbit control, inter-satellite microwave communication, and inter-satellite laser communication, swarm systems have become highly engineering feasible. Simultaneously, an increasing number of space missions naturally require multi-spacecraft collaborative execution, such as deep space exploration, space resource surveying, space situational awareness, on-orbit servicing, proactive debris removal, and orbital maneuvering. In these missions, multiple platforms typically share observational information, target status, and relative positions for collaborative cooperation; independent operation by a single platform is insufficient to meet the mission's requirements for timeliness, robustness, and global awareness. Space debris monitoring and removal is one of the most representative and pressing application scenarios. Compared to the Low Earth Orbit (LEO) environment, GEO orbit target cataloging requires a higher detection threshold, but existing ground-based cataloging systems lack sufficient detection capabilities for small-sized debris in GEO regions. Therefore, for debris monitoring and removal scenarios in GEO regions, distributed spacecraft swarms offer significant advantages.

[0003] Because different spacecraft play different roles in missions, their payloads, communication terminals, and computing and energy resources vary, spacecraft swarm networks often contain both high-speed point-to-point laser links and low-speed many-to-many microwave links. The missions of a spacecraft swarm primarily include situational observation, situational understanding, decision generation, and action execution phases. Nodes in each role collaborate through the network to support the mission loop, a mission model consistent with the OODA (Observe–Orient–Decide–Act) loop. Therefore, a spacecraft swarm network can essentially be modeled as a heterogeneous self-organizing network oriented towards the OODA mission loop. Laser links offer advantages such as high bandwidth and low latency, but are limited by satellite visibility, pointing constraints, and the number of laser terminals, requiring dynamic topology planning based on network conditions.

[0004] Existing work on onboard laser topology control primarily focuses on LEO mega-constellations or space backbone networks, typically employing a ground-based centralized planning perspective and modeling network-layer metrics such as connectivity, capacity, latency, energy consumption, and topology stability. In contrast, research on laser topology control for heterogeneous spacecraft clusters is severely lacking, particularly regarding topology control issues driven by differences in node roles, hybrid link configurations, and OODA closed-loop mission requirements. Furthermore, systematic research on how to collaboratively optimize laser topology control and routing decisions under such mission constraints remains insufficient. Summary of the Invention

[0005] To address some or all of the problems of existing technologies, and in order to construct a unified networking framework capable of efficiently completing the OODA task closed loop under conditions of heterogeneous node roles, heterogeneous link types, changing task stages, and time-varying network topology, the first aspect of this invention provides a method for updating inter-satellite laser topology, including updating the inter-satellite laser topology once every specified time interval, wherein each update includes: Obtain the node characteristics of all laser nodes and the link characteristics of all laser links in the laser subnet to determine the set of visible laser links; Based on the aforementioned set of visible laser links, a seed topology is constructed; Candidate laser topologies are generated based on seed topology; The optimal topology is selected from the candidate laser topologies and then updated.

[0006] Furthermore, the inter-satellite laser topology is rooted at a computing satellite and includes several layers of sensing satellites, forming a multi-root layered forest topology (MRLF) structure, wherein the number of sensing satellites in each layer satisfies the maximum degree constraint.

[0007] Furthermore, the seed topology includes Local Sensing-Computation Cycle Topology (LSCC) and Globally Connected Sensing-Computation Cycle Topology (GCSCCC). The Local Sensing-Computation Cycle Topology forms a local small loop for each computational satellite of the multiple hierarchical forest topologies, and the Globally Connected Sensing-Computation Cycle Topology connects the multiple hierarchical forest topologies end to end to form a global large loop.

[0008] Furthermore, candidate laser topologies are generated by randomly deleting edges.

[0009] Furthermore, the optimal topology is selected from the candidate laser topologies using a pre-trained topology selection algorithm.

[0010] Furthermore, the topology selection algorithm uses the window-level evaluation results of the frozen RFD as a supervision signal.

[0011] Furthermore, the window-level evaluation results include: task one-time completion rate and average task latency.

[0012] Based on the update method described above, a second aspect of the present invention provides an electronic device for updating inter-satellite laser topology, comprising a memory and a processor, wherein the memory is configured to store a computer program that executes the update method described above when the processor is running.

[0013] The present invention also provides a computer-readable storage medium for updating inter-satellite laser topology, which stores a computer program that, when run on a processor, executes the update method as described above.

[0014] This invention provides a method for updating inter-satellite laser topology. It constructs and updates the LTC layer using a candidate-based feasibility-value selector (CFVS). Specifically, it generates a candidate topology set based on a seed topology and jointly evaluates the feasibility and performance value of the candidate topologies, achieving supervised topology selection based on the candidate set. This update method effectively constrains unnecessary laser link establishment while ensuring mission-level performance, thereby significantly reducing network energy consumption. Attached Figure Description

[0015] To further illustrate the above and other advantages and features of the various embodiments of the present invention, a more specific description of the various embodiments of the present invention will be presented with reference to the accompanying drawings. It is to be understood that these drawings depict only typical embodiments of the invention and are therefore not intended to limit its scope. In the drawings, identical or corresponding parts will be indicated by identical or similar reference numerals for clarity.

[0016] Figure 1 This diagram illustrates a flowchart of an inter-satellite laser topology update method according to an embodiment of the present invention. Figure 2 This diagram illustrates the structure of a multi-rooted hierarchical forest topology according to an embodiment of the present invention. Figure 3 This diagram illustrates the structure of a local sensing-computation ring topology according to an embodiment of the present invention. Figure 4 A schematic diagram of a global connectivity-aware computational ring topology according to an embodiment of the present invention is shown. Detailed Implementation

[0017] In the following description, the invention is described with reference to various embodiments. However, those skilled in the art will recognize that the embodiments may be practiced without one or more specific details or in conjunction with other alternatives and / or additional methods or components. In other instances, well-known structures or operations are not shown or described in detail so as not to obscure the inventive points of the invention. Similarly, for illustrative purposes, specific numbers and configurations are set forth to provide a comprehensive understanding of embodiments of the invention. However, the invention is not limited to these specific details. Furthermore, it should be understood that the embodiments shown in the drawings are illustrative representations and are not necessarily drawn to scale.

[0018] In this specification, references to "an embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. The phrase "in one embodiment" appearing throughout this specification does not necessarily refer to the same embodiment in all instances.

[0019] It should be noted that the embodiments of the present invention describe the method steps in a specific order; however, this is only for illustrating the specific embodiment and not for limiting the order of the steps. On the contrary, in different embodiments of the present invention, the order of the steps can be adjusted according to actual needs.

[0020] This invention addresses the self-organizing network problem of heterogeneous spacecraft clusters in OODA missions by employing a collaborative optimization framework for laser topology planning and heterogeneous network routing. Considering the differences in time scale between laser link topology reconstruction and service routing decisions, this application adopts a layered approach. The upper layer performs topology control of laser inter-satellite links from a global perspective to construct a laser topology more suitable for the current mission flow and network state. The lower layer, with each node taking a distributed, local perspective, makes service-granular routing decisions based on its own state within the laser-microwave hybrid heterogeneous network, thereby improving the efficiency and reliability of mission data closed-loop transmission. The core problem this application aims to solve is how to construct a unified networking framework capable of efficiently completing the OODA mission closed loop under conditions of heterogeneous node roles, heterogeneous link types, changing mission stages, and time-varying network topology.

[0021] To address the joint optimization problem of heterogeneous spacecraft cluster self-organizing networks in a system-oriented OODA task, this paper first establishes an OODA closed-loop task model to describe the generation and flow relationship of services at different stages. Second, it establishes a laser link visibility model, constructing a set of available laser edges under time-varying geometric constraints. Then, it defines the structure and analyzes the properties of the seed topology upon which the Laser Topology Control (LTC) layer depends, demonstrating its rationality as the basis for candidate topology generation. Finally, it establishes four performance index models: task one-time completion rate, task latency, task energy consumption, and network energy consumption variation coefficient. Based on these models, laser topology control and routing forwarding decisions are formalized into a unified joint optimization problem, providing a theoretical foundation for subsequent algorithm design.

[0022] Based on their onboard payloads and mission roles, cluster nodes are divided into three categories: sensing satellites, computing satellites, and execution satellites, denoted as follows: Then the entire cluster node set can be represented as Let the number of nodes in the three types be respectively... Then there is All satellites are equipped with multiple microwave terminals to achieve near-omnidirectional microwave coverage and ensure basic connectivity. The sensing and computing satellites additionally carry two point-to-point laser terminals for high-capacity image data transmission. Therefore, the network contains two types of links: laser links and microwave links, denoted as […]. and Therefore, the set of link types can be represented as Each OODA mission consists of three phases: collaborative observation, decision computation, and debris removal. First, several sensing satellites collaboratively observe a specific debris target and generate image data. Then, the relevant data is forwarded to one or more computing satellites for mission planning. Finally, the computing satellites generate control commands and send them to the target execution satellite to complete the debris removal operation. Specifically, each mission is denoted as... ,exist Initially, the system randomly selects three sensing satellites to participate in collaborative observation. Each sensing satellite generates one Class A image service, denoted as... ,in This indicates the source sensing satellite for this service. Indicates the size of the business. This indicates the computational load required for this service. For Category A services, the destination node can be any computing satellite; that is, its destination nodes belong to a set. Once a Class A service is processed on a computing satellite, it will be converted into a Class B instruction service, denoted as... ,in This indicates the source computing satellite that generated the instruction service. This indicates the size of the processed instruction. This refers to the set of target execution satellites for the command service. Since a single debris removal operation may require multiple execution satellites to work together, the target node for each Class B service can be 1-2 execution satellites. Therefore, a single OODA task can be abstracted as a two-stage service forwarding process: the first stage is the uploading and processing of Class A image services from the sensing satellite to the computing satellite; the second stage is the distribution and execution of Class B command services from the computing satellite to the execution satellite. Only when... The mission is considered complete only when all corresponding Category B services have been successfully delivered to their respective destinations on the satellite. For the task The entire set of Class A business generated For the reason The entire set of derived B-type business functions, then the task The completion condition can be written as ,in Indicates business Successfully delivered to the execution satellite .

[0023] The visibility of laser inter-satellite links is determined using a distance threshold model based on link budget. The maximum visibility distance is calculated based on the laser terminal parameters. The system then uses the spacecraft's trajectory to determine whether any node pair meets the visibility condition at the current moment. Furthermore, it utilizes future trajectories to estimate the remaining visibility duration of the link, thus characterizing the link's temporal stability.

[0024] Let the total divergence angle of the laser be... The equivalent pointing angle error is Then the loss modeling is as follows: .

[0025] Let the transmission power be The optical efficiencies of the transmitter and receiver are respectively and The receiving aperture is The receiving sensitivity is The link margin is The remaining losses are The minimum received power threshold is: , The linear form of stray loss is: .

[0026] Record the receiving radius Then, the constant term excluding geometric diffusion can be written as: .

[0027] Distance The geometric capture efficiency at that point is expressed as: , Therefore, the received power is: .

[0028] when At that time, the link is considered to meet the visibility requirement. Let , Then the maximum visible distance can be obtained: .

[0029] Set nodes and nodes exist The three-dimensional positions at time t are respectively and Then its interstellar distance is: .

[0030] Therefore, instantaneous laser visibility is defined as: .

[0031] All satisfied The node pairs constitute the set of laser-visible candidate edges at the current moment: .

[0032] To characterize the future sustainability of the link, this invention further utilizes future trajectory samples to calculate the remaining visible duration. Let the trajectory sampling period be... Then the link In the future The visibility of each sampling point is: .

[0033] Since this invention primarily targets GEO spacecraft cluster scenarios, the aforementioned laser link visibility model is based on distance thresholds and does not consider Earth occlusion or satellite attitude. Each spacecraft is equipped with two laser terminals, with an azimuth coverage range of ±180° and an elevation coverage range of -80° to +30°, capable of covering most potential inter-satellite directions within the cluster. Based on this, the distance between nodes is limited to the maximum communication distance obtained from the link budget. As a condition for the visibility of the laser link, no additional angle constraints are introduced. This model reduces computational complexity while effectively supporting subsequent topology selection and routing decisions.

[0034] Based on the aforementioned mathematical modeling foundation Figure 1This diagram illustrates a flowchart of an inter-satellite laser topology update method according to an embodiment of the present invention. Figure 1 As shown, a method for updating inter-satellite laser topology includes updating the inter-satellite laser topology at specified intervals, wherein each update includes: First, in step 101, the set of visible laser links is determined. The node characteristics of all laser nodes and the link characteristics of all laser links in the laser subnet are obtained to determine the set of visible laser links. In each update cycle... Global observation of the laser subnet collected by the network ,in and Let represent the node characteristics of all laser nodes and the link characteristics of all laser links in the network, respectively. Then, based on the method described above, determine the set of candidate laser visible edges at the current time. ; Next, in step 102, a seed topology is constructed. Based on the set of visible laser links, a seed topology is constructed. In subsequent topology selection, instead of performing an unconstrained search on all visible candidate edges, candidate topologies are generated starting from the seed topology to shrink the topology decision space and retain better reachability. Therefore, a reasonable seed topology needs to be constructed first. Let the undirected simple graph be... Due to the large data volume transmission of Category A image services between sensing satellites and computing satellites, in one embodiment of the present invention, reducing the average hop count between sensing satellites and computing satellites is the optimization objective. Meanwhile, since each satellite is equipped with... In one embodiment of the present invention, a uniform maximum constraint is used to characterize the laser connection resource limitations of a laser terminal: .

[0035] For any sensing satellite Define it to the nearest computing satellite The number of hops is: .

[0036] To ensure that all sensing satellites are available for service, the following requirements must be met: .

[0037] Define the total number of hops and the average number of hops as follows: .

[0038] Therefore, the seed topology design objective can be expressed as: , This problem is related to minimizing the average number of hops. equivalence.

[0039] For each sensing satellite We retain the first edge on the shortest path from the sensing satellite to the nearest computing satellite, and repeat this process for subsequent nodes on the path. This subgraph does not increase the shortest distance from any sensing satellite to the computing satellite. Since each retained edge points to a node closer to the computing satellite, when moving along the edge... Strictly decreasing, therefore it is impossible to form a cycle. For forests. Each connected component contains at least one computed satellite as the root, and yes The subgraph is such that the maximum degree does not increase. It can be seen that for any feasible graph satisfying the above constraints... Each has a subgraph , making For a multi-rooted forest with computational satellites as the root, satisfying: , .

[0040] Therefore, when searching for the optimal number of hops, we only need to consider a multi-rooted forest structure with the nearest computational satellite as the root, and denote the nearest computational satellite as... The number of sensing satellites jumping is In Layer 1, each computing satellite can connect to a maximum of [number missing] connections. A sensing satellite or other node, therefore For deeper layers, any position located at the th Each node in the layer already has an edge connecting it to its parent node, therefore, the maximum degree is... Under the constraints, at most [number] connections can be made. Based on this, we can inductively conclude that for any... , .

[0041] Before definition The cumulative capacity of the layer is and order ,in, This means accommodating all under the maximum constraint. The minimum number of layers required for a sensing satellite. As mentioned earlier, any feasible graph can be equivalently reduced to a multi-rooted forest without increasing the hop count, therefore only a lower bound needs to be given for the multi-rooted forest. Meanwhile, the... The maximum number of floors that can accommodate One sensing satellite, before The maximum number of floors that can accommodate One sensing satellite. Therefore, in the front Before the layer is filled, there are still at least Each sensing satellite must be located at a deeper level. Using tomographic summation, we can obtain that the total hop count under any feasible topology is greater than or equal to the lower bound: .

[0042] from Constructing a multi-root forest from a computing satellite: The first layer should place as many [roots] as possible. One sensing satellite, placed in the second layer as much as possible. The first sensing satellite, and so on, until the first... The remaining layers are placed on the floor. A number of sensing satellites. This configuration satisfies the maximum degree constraint and ensures that the number of sensing satellites at each layer is: .

[0043] The total number of jumps can be calculated as follows: , Therefore, the optimal average number of items is: .

[0044] Based on this, in one embodiment of the present invention, when the average shortest hop count from the sensing satellite to the computing satellite is taken as the optimization objective, the optimal topology under the maximum degree constraint presents a multi-root layered forest topology (MRLF) with the computing satellite as the root and shallow-layer priority filling, such as... Figure 2 As shown. Based on this structure, two types of seed topologies can be constructed: Local Sensing-Computation Cycle Topology (LSCC) and Globally Connected Sensing-Computation Cycle Topology (GCSCC). LSCC constructs multiple local small loops on top of MRLF, providing multi-path redundancy while maintaining short-hop reachability, such as... Figure 3 As shown; GCSCC connects the beginning and end of MRLF to enhance global connectivity, as shown. Figure 4 As shown.

[0045] Next, in step 103, candidate laser topologies are generated. Candidate laser topologies are generated based on seed topologies. In one embodiment of the invention, the seed topology set is denoted as... For each seed topology Let its edge set be ,Will The edge itself is added to the candidate set, and then candidate topologies are generated by randomly deleting edges. Specifically, for the number of edges deleted... ,from Randomly delete Strip the edge and repeat this random edge removal sampling. This yields the candidate topology set: , in, Indicates the first The second sampling obtained A set of edges to be deleted. This candidate topology generation method, while preserving the short-hop reachable skeletons of LSCC and GCSCC, generates link pruning to varying degrees, thus providing a search space for the trade-off between energy reduction and performance maintenance. In one embodiment of the invention, candidate laser topologies are generated and subsequently selected using a candidate-based feasibility-value selector (CFVS). The CFVS includes a candidate topology generator, a GCN-based candidate encoder, and a dual-head selector, wherein the candidate topology generator is used to generate a set of candidate laser topologies based on LSCC and GCSCC seed topologies.

[0046] Finally, in step 104, the optimal topology is determined. The optimal topology is selected from the candidate laser topologies and updated. In one embodiment of the invention, the LTC layer is responsible for laser topology control on slower timescales. Since LTC only involves one edge type—the laser link—in one embodiment of the invention, CFVS uses a GCN-based candidate encoder to perform graph encoding on the candidate topologies, extracting the graph-level embedding of the candidate topologies in the current global state. For the candidate topologies... Let its adjacency matrix be . Then the candidate topological embedding can be written as Then, a dual-headed selector is used to output the feasibility score and the value score respectively: The feasibility header is used to determine whether the candidate topology can maintain an acceptable level of performance relative to the seed topology in terms of task performance, and the value header is used to estimate the comprehensive value of the candidate topology between saving links and maintaining task performance, thereby achieving topology selection oriented towards task performance and energy saving.

[0047] In one embodiment of the invention, the CFVS is obtained through training. Specifically, candidate topologies are first generated, and then a frozen (RFD) strategy is used. A window-level rolling evaluation is performed on each candidate topology. The aforementioned frozen (RFD) strategy... This can be obtained through training. For data derived from seed topology... Candidate topology Seed topology and candidate topology were obtained respectively in this One-time completion rate of tasks during the period and average task latency In one embodiment of the invention, a seed topology is used as a reference benchmark for the group of candidates, and a feasibility label is constructed: , in, Let be an indicator function, indicating that a candidate topology is feasible when it does not significantly degrade relative to its source seed topology in terms of task completion rate and task latency. Furthermore, the value label of a candidate topology, combining edge deletion benefits and task performance changes, can be abstractly represented as: ,in This indicates the number of laser links saved by the candidate topology compared to the seed topology. The benefits of edge deletion are rewarded, while penalties are imposed for decreased task completion rate and increased task latency. Therefore, each training sample can be represented as... .

[0048] During the training phase, supervised learning is used to optimize the output of the dual-head selector. The overall loss consists of feasibility classification loss, value regression loss, and candidate topology ranking loss. ,in Used to monitor the feasibility of candidate topologies. Used to fit window-level value labels The LTC policy is obtained after training to enhance the ranking ability of candidate topologies. .

[0049] After training, in practical applications, CFVS will process the candidate set. Each topology in the algorithm is coded and scored by two heads, and the current laser configuration topology is selected according to the rule of prioritizing feasibility and then value. .

[0050] In this invention, LTC training does not independently optimize the topology structure, but uses the actual forwarding effect of frozen RFD on the candidate topology as a supervision signal, so that LTC can select a laser topology with high completion, low latency and low power consumption given a laser subnet state.

[0051] Comparative verification shows that CFVS can maintain high task performance while significantly reducing the number of laser link establishments. Compared with the LSCC and GCSCC seed topologies, at 20 nodes, In typical scenarios, the CFVS group reduced the average number of laser link establishments from 12 to approximately 9, resulting in a 16% and 19.5% reduction in task energy consumption under uniform and hotspot modes, respectively. Furthermore, task completion rate and latency showed almost no deterioration, indicating that there are some redundant links in the seed topology under the current network conditions. CFVS can remove low-yield links based on the current network conditions and task requirements, thereby significantly reducing laser link maintenance energy consumption without sacrificing completion rate or latency. Compared to LTC baselines such as NSGA3, NEAREST, and MTDA, the advantages of CFVS are even more pronounced. NSGA3 achieves a trade-off between latency, hop count, and load balancing by optimizing non-mesh topologies through multi-objective optimization. However, its primary objective is statistical metrics at the topology graph level, lacking direct insight into OODA task flow and routing benefits. Therefore, it tends to retain structurally sound links with limited task benefits. In terms of one-time task completion rate, the CFVS group outperforms the NSGA3 group by 23.5% and 16.1% in uniform and hotspot modes, respectively. Its task latency is 14.3% and 20.1% lower, respectively, and its task energy consumption is 27% and 30.6% lower, respectively. NEAREST tends to select the nearest neighbor links with shorter spatial distances. While this reduces the physical cost of a single link, it easily leads to chain-like structures, requiring more hops for tasks. In terms of one-time task completion rate, the CFVS group outperforms the NEAREST group by 21.7% and 14.6% in uniform and hotspot modes, respectively. Its task latency is 14.1% and 19.3% lower, respectively, and its task energy consumption is 26.5% and 29.1% lower, respectively. MTDA primarily selects links with better physical layer quality based on acquisition probability, received power, and link visibility stability, but does not directly consider the current task load and node congestion status. In uniform and hotspot modes, the CFVS group achieves a 31.6% and 20.8% higher one-time task completion rate than the MTDA group, respectively; task latency is 17.3% and 26.9% lower; and task energy consumption is 32.8% and 30.9% lower. Therefore, while these three methods can construct sparse topologies, the edge deletion locations may not meet task-level transmission requirements. From the network energy consumption CV perspective, CFVS shows an increase compared to other topologies, indicating that some key nodes bear more effective forwarding load after edge deletion. However, based on actual node energy consumption differences, the standard deviation of node energy consumption in the CFVS group does not increase; in fact, it is slightly smaller. Under uniform and hot spot modes, the standard deviations of nodal energy consumption for the CFVS, LSCC, GCSCC, NSGA3, NEAREST, and MTDA groups were 39.5 KJ / 38.7 KJ, 41.8 KJ / 41.2 KJ, 42 KJ / 41.4 KJ, 40.9 KJ / 40.6 KJ, 41.2 KJ / 40.8 KJ, and 41.4 KJ / 41.3 KJ, respectively.Therefore, the increase in CV of CFVS is mainly due to the significant reduction in the average energy consumption of the entire network, which makes the denominator smaller when calculating CV, resulting in a higher CV. In summary, CFVS provides an efficient laser topology for OODA tasks, significantly reducing task energy consumption while ensuring task performance.

[0052] Based on the update method described above, the present invention also provides an electronic device for updating inter-satellite laser topology, comprising a memory and a processor, wherein the memory is configured to store a computer program that executes the update method described above when the processor is running.

[0053] The present invention also provides a computer-readable storage medium for updating inter-satellite laser topology, which stores a computer program that, when run on a processor, executes the update method as described above.

[0054] Furthermore, this invention also provides a self-organizing network structure for heterogeneous spacecraft clusters, comprising a laser topology control layer and a routing and forwarding decision layer. The laser topology control layer selects the laser topology from a global perspective, while the routing and forwarding decision layer executes service-aware routing and forwarding decisions from a local perspective. Both work together to support the information loop of the OODA task. In one embodiment of this invention, a time-scale hierarchical design achieves collaborative optimization between topology and routing. The routing and forwarding decision layer captures heterogeneous link type differences and instantaneous node and link states based on gated graph convolution (Gated-RGCN), and determines the routing and forwarding path through a flow-level policy network. The laser topology control layer includes a feasibility-value selector based on candidate solutions. It generates a candidate topology set based on a seed topology, then selects the optimal topology from the candidate topology set using CFVS, updating the laser topology at specified intervals, thereby reducing redundant laser links while maintaining mission performance. The CFVS is obtained through training. The training process employs a pre-trained multi-agent near-end policy optimization algorithm (GRDC-MAPPO, Gated-RGCNDecision-Critic MAPPO) based on gated graph convolution and decision decoupling evaluation to perform window-level supervised evaluation of candidate topologies. The self-organizing network structure is jointly optimized from the perspectives of topology control and service routing coordination, effectively supporting the communication requirements under the OODA task closed loop. It ensures optimal overall task performance in terms of task completion rate, latency, energy consumption, and network energy balance, while also exhibiting good scalability and robustness across different node scales.

[0055] Furthermore, the present invention also provides a heterogeneous spacecraft cluster, which includes the self-organizing network structure as described above.

[0056] Although various embodiments of the invention have been described above, it should be understood that they are presented by way of example only and not as limitations. It will be apparent to those skilled in the art that various combinations, modifications, and alterations can be made without departing from the spirit and scope of the invention. Therefore, the breadth and scope of the invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should be defined solely by the appended claims and their equivalents.

Claims

1. A method for updating inter-satellite laser topology, characterized in that, The inter-satellite laser topology is updated at specified time intervals, and each update includes: Obtain the node characteristics of all laser nodes and the link characteristics of all laser links in the laser subnet to determine the set of visible laser links; Based on the aforementioned set of visible laser links, a seed topology is constructed; Candidate laser topologies are generated based on seed topology; The optimal topology is selected from the candidate laser topologies and then updated.

2. The update method as described in claim 1, characterized in that, The inter-satellite laser topology is based on a computing satellite and includes several layers of sensing satellites, forming a multi-rooted hierarchical forest topology, wherein the number of sensing satellites in each layer satisfies the maximum degree constraint.

3. The updating method as described in claim 2, characterized in that, The seed topology includes a local sensing-computation loop topology and a globally connected sensing-computation loop topology. The local sensing-computation loop topology forms a local small loop for each computing satellite of the multi-rooted hierarchical forest topology, and the globally connected sensing-computation loop topology connects the multi-rooted hierarchical forest topology end to end to form a global large loop.

4. The update method as described in claim 1, characterized in that, Candidate laser topologies are generated by randomly deleting edges.

5. The update method as described in claim 1, characterized in that, The optimal topology is selected from the candidate laser topologies using a pre-trained topology selection algorithm.

6. The updating method as described in claim 5, characterized in that, The topology selection algorithm uses the window-level evaluation results of the frozen RFD as a supervision signal.

7. The updating method as described in claim 6, characterized in that, The window-level evaluation results include: task one-time completion rate and average task latency.

8. An electronic device for updating inter-satellite laser topology, characterized in that, It includes a memory and a processor, wherein the memory is configured to store a computer program that executes the update method as described in any one of claims 1 to 7 when the processor is running.

9. A computer-readable storage medium for updating inter-satellite laser topology, characterized in that, The system contains a computer program that, when run on a processor, executes the update method as described in any one of claims 1 to 7.

Images