A method for integrating low-orbit satellite measurement, operation and control resources

Abstract

The present invention discloses a method for integrating measurement, operation and control resources of low-orbit satellites, which relates to the technical field of measurement, operation and control resource integration. The present invention adopts a spatiotemporal attention module to comprehensively process continuous network status snapshots of low-orbit satellites, aerial platforms and ground stations, integrates data information of multiple time points, and forms a spatiotemporal feature representation of node status, so that mapping decisions can refer to network changes in continuous time slots; through the cooperation of a graph attention network and a self-attention mechanism, the problem of invalid mapping results of a single time snapshot is solved. In virtual network mapping, a two-stage mapping scheme is adopted; in the first stage, a graph convolutional network is used to map nodes in a virtual network request to physical nodes, and the normalized adjacency matrix and weight matrix are fully utilized to fuse neighborhood information; in the second stage, the link path between candidate nodes is calculated by the Floyd algorithm, and the node mapping results and the link mapping are uniformly planned.

Description

Technical Field

[0001] The present invention relates to the technical field of measurement, launch and control resource fusion, and in particular to a method for fusion of measurement, launch and control resources for low-orbit satellites. Background Art

[0002] For the integration of measurement, operation and control resources of low-orbit satellites, static snapshot optimization is currently mostly used, that is, mapping and scheduling are performed according to the current resource status at a certain moment.

[0003] In actual operation, however, the mapping scheme will quickly lose its original optimality due to the ever-changing status and link quality of nodes such as satellites, drones, and ground stations. Most existing systems recalculate the mapping strategy at fixed time intervals to cope with such changes. However, in the continuous evolution of the network, the mapping results frequently fail, and the system has to constantly adjust and migrate, which is rather cumbersome. Such frequent adjustments lead to the continuous migration of resources between nodes, which increases the computational burden and communication overhead. In the case of intensive tasks or drastic environmental changes, reconstructing the mapping scheme not only reduces the overall resource utilization efficiency, but may also cause scheduling delays and affect the task response speed.

[0004] In response to this, some solutions conduct periodic updates or introduce prediction mechanisms. However, in the highly dynamic scenario of low-orbit satellites, such strategies still find it difficult to balance global stability and resource utilization efficiency. Therefore, how to more accurately capture the temporal correlation of network status, predict future status in advance, and reasonably weigh the adjustment costs in mapping decisions have become key challenges that need to be solved urgently. Solving this problem will not only help reduce the migration costs caused by frequent adjustments, but also improve the robustness and response speed of the overall network, providing a solid guarantee for the efficient operation of low-orbit satellite measurement, operation and control systems in complex dynamic environments. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a low-orbit satellite measurement, operation and control resource fusion method to solve the problem that virtual network mapping is often frequently reconstructed in a dynamic network environment due to ignoring temporal associations, resulting in high migration costs and resource waste.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] The embodiment of the present invention provides a method for integrating low-orbit satellite measurement, operation and control resources, which includes:

[0009] Step S1: construct a resource status acquisition unit based on a time-varying graph model to generate a network status snapshot;

[0010] Step S2: Processing the network state snapshots using a spatiotemporal attention module. The spatiotemporal attention module includes a spatial attention layer and a temporal attention layer. The former uses a graph attention network to weight information about node neighborhoods, and the latter uses a self-attention mechanism to perform temporal correlation encoding on states between consecutive snapshots to form a spatiotemporal feature representation of the node state.

[0011] Step S3, performing two-stage mapping on the candidate mapping set according to the spatiotemporal feature representation to obtain mapping parameters;

[0012] Step S4, adjusting the mapping parameters using a trust region policy gradient optimization module, wherein the trust region policy gradient optimization module constrains parameter updates by setting a parameter update trust region, and calculates the migration cost during the mapping process according to a predetermined reward function;

[0013] Step S5: Use the above mapping results to adjust the measurement, operation and control task allocation between the low-orbit satellite and the space-ground platform.

[0014] As a preferred solution of the method for integrating low-orbit satellite measurement, operation and control resources described in the present invention, in step S1, the resource status acquisition unit collects real-time data from low-orbit satellites, aerial platforms and ground station nodes, and generates network status snapshots at predetermined time intervals;

[0015] In step S1, the data collected by the resource status collection unit includes computing resources Cpu, link bandwidth Bdw, remaining computing resources and link transmission delay.

[0016] As a preferred solution of the low-orbit satellite measurement, operation and control resource fusion method described in the present invention, wherein: in step S2, in the spatiotemporal attention module, the spatial attention layer adopts a graph attention network to allocate attention weights within the node neighborhood, and the temporal attention layer adopts a self-attention mechanism to encode the temporal correlation between continuous network state snapshots. After the two are fused, a comprehensive feature representation based on which the mapping decision is based is formed.

[0017] As a preferred solution of the low-orbit satellite measurement, operation and control resource fusion method described in the present invention, in step S2, the step of processing the network status snapshot using the spatiotemporal attention module is as follows:

[0018] To form the spatiotemporal feature representation of node state, a continuous network state snapshot sequence is constructed Each snapshot G t =(V,E,X t ), including the node set V, the edge set E and the node feature matrix X t ;

[0019] The spatial attention layer calculates the attention weight at layer l, and the formula is:

[0020] in,

[0021] Indicates the attention allocation weight of node i to its neighbor node j in the lth layer, SoftMax j Indicates the softmax normalization operation on index j, a (l) Represents the attention weight vector of the lth layer, W (l) is the transformation matrix of the lth layer, represents the feature of node i in layer l, represents the feature of node j at layer l, and the symbol | represents vector concatenation;

[0022] The spatial encoding output is:

[0023]

[0024] in, represents the aggregate representation of node i at layer l+1, W (l) represents the transformation matrix of the lth layer, and σ is the activation function.

[0025] As a preferred solution of the low-orbit satellite measurement, operation and control resource fusion method described in the present invention, in step S2, the step of processing the network status snapshot using the spatiotemporal attention module further includes:

[0026] Representation of node i in consecutive snapshots by the temporal attention layer Calculate the similarity score using the formula:

[0027] in,

[0028] β t,k represents the temporal attention weight between time t and time k, represents the similarity function between node i and k at time t, T represents the total number of time steps, represents the feature representation of node i at time step t;

[0029] The final spatiotemporal feature of node i is represented as h i :

[0030]

[0031] Among them, β t Represents the weighting coefficient corresponding to the time step.

[0032] As a preferred solution of the method for integrating low-orbit satellite measurement, operation and control resources described in the present invention, wherein: in the two-stage mapping, the first stage uses a graph convolutional network to perform node mapping on the virtual network request, and the second stage uses the Floyd algorithm to calculate the link path between candidate nodes to form a virtual link mapping. As a preferred solution of the method for integrating low-orbit satellite measurement, operation and control resources described in the present invention, wherein: in step S3, the step of implementing two-stage mapping on the candidate mapping set is:

[0033] For the mapping between virtual network requests and physical network resources, a two-stage mapping is adopted. In the first stage, the graph convolutional network (GCN) is used to map the nodes in the virtual network request. The node mapping expression is:

[0034]

[0035] Among them, y i represents the mapping output of virtual node i, represents the normalized adjacency matrix, X j is the characteristic of physical node j, W g is the weight matrix of the graph convolution layer;

[0036] In the second stage, the Floyd algorithm is used to calculate the link paths between the candidate nodes in the first stage. The virtual link mapping cost calculation formula is:

[0037]

[0038] Among them, d ij represents the cumulative cost of the mapping link between node i and node j, Indicates taking the minimum value among all paths, ∑ (u,v)∈path l uv represents the cumulative link cost of all edges (u, v) in the path,

[0039] Among them, l uv represents the link cost of edge (u,v).

[0040] As a preferred solution of the method for integrating low-orbit satellite measurement, operation and control resources described in the present invention, in step S4,

[0041] The trust region policy gradient optimization module is used to adjust the mapping parameters.

[0042] The trust region policy gradient optimization module is provided with a trust region for parameter updates.

[0043] The confidence region constrains the parameter update amplitude with a predetermined KL divergence threshold.

[0044] As a preferred solution of the low-orbit satellite measurement, operation and control resource fusion method described in the present invention, in step S4, the step of adjusting the mapping parameters using the trust region policy gradient optimization module is as follows:

[0045] In the mapping parameter adjustment, the mapping migration cost is quantified by constructing a reward function R, which is formulated as follows:

[0046]

[0047] Among them, Rev i represents the benefit brought by mapping, ∑ i Rev i Indicates the sum of the benefits generated by all mappings, Cost i represents the migration cost during the mapping process,

[0048] At the same time, define KL divergence:

[0049]

[0050] Among them, D KL (π old ||π) represents the old policy π old The KL divergence between the new policy π, π old (a|s) and π(a|s) represent the distribution of the old strategy and the new strategy for action a in state s, respectively. a π lod (a|s) represents the sum of the probabilities of all actions a;

[0051] Set the threshold to meet:

[0052] D KL (π old ||π)≤δ,

[0053] Where δ represents the predetermined KL divergence threshold;

[0054] The trust region policy gradient update formula is used to update the mapping parameters. The update formula is:

[0055]

[0056] Among them, W new represents the updated mapping parameters, W old represents the mapping parameters before updating, L(W) represents the objective function based on the reward function R, represents the gradient of the objective function L(W) with respect to the parameter W, and η is the learning rate.

[0057] As a preferred solution of the method for integrating low-orbit satellite measurement, operation and control resources described in the present invention, the scheduling control unit schedules the measurement, operation and control tasks between the low-orbit satellite and the space-ground platform based on the aforementioned mapping results and parameter adjustment results. The scheduling control unit uses continuous network status snapshots as a reference and performs task allocation during the mapping process across time slots.

[0058] In step S5, the aforementioned mapping result refers to the candidate mapping parameter obtained in step S3 and the adjustment result of the parameter in step S4.

[0059] The beneficial effects of the present invention are as follows: the present invention adopts a spatiotemporal attention module to comprehensively process the continuous network status snapshots of low-orbit satellites, aerial platforms and ground stations, integrates the data information of multiple time points, and forms a spatiotemporal feature representation of the node status, so that mapping decisions can refer to network changes in continuous time slots; through the cooperation of the graph attention network and the self-attention mechanism, the problem of invalidation of the mapping results of a single time snapshot is solved, and frequent reconstruction and resource migration caused by state mutations are avoided.

[0060] The present invention adopts a two-stage mapping scheme in virtual network mapping. In the first stage, a graph convolutional network is used to map the nodes in the virtual network request to physical nodes, and the normalized adjacency matrix and weight matrix are fully utilized to fuse the neighborhood information. In the second stage, the link path between candidate nodes is calculated through the Floyd algorithm, and the node mapping results and link mapping are unified for planning.

[0061] The present invention also introduces a trust region policy gradient optimization module to adjust mapping parameters. This module constructs a reward function that combines the benefits of the mapping process with the migration cost, and uses a predetermined KL divergence threshold to limit parameter updates, ensuring that mapping parameters are updated stably in a continuous state. The scheduling control unit relies on continuous network state snapshots and mapping results to execute task allocation between low-orbit satellites and space-ground platforms, achieving cross-time slot scheduling.

[0062] In general, the present invention overcomes the problems of insufficient timing dependence and excessive migration costs caused by frequent mapping reconstruction in traditional low-orbit satellite measurement, operation and control resource fusion methods, and provides a coherent and controlled solution for the dynamic changes of the network environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0064] Figure 1The figure is a flow chart of the low-orbit satellite measurement, operation and control resource integration method of the present invention. DETAILED DESCRIPTION

[0065] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0066] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0067] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0068] Example 1, with reference to Figure 1 This embodiment provides a method for integrating low-orbit satellite measurement, operation and control resources, including:

[0069] Step S1: construct a resource status acquisition unit based on a time-varying graph model to generate a network status snapshot;

[0070] In step S1, the resource status acquisition unit collects real-time data from low-orbit satellites, aerial platforms, and ground station nodes, and generates network status snapshots at predetermined time intervals;

[0071] In step S1, the data collected by the resource status collection unit includes computing resources Cpu, link bandwidth Bdw, remaining computing resources and link transmission delay;

[0072] Step S2: Process the network state snapshots using the spatiotemporal attention module. The spatiotemporal attention module includes a spatial attention layer and a temporal attention layer. The former uses a graph attention network to weight the information of the node neighborhood, and the latter uses a self-attention mechanism to encode the temporal association between the states of consecutive snapshots to form a spatiotemporal feature representation of the node state.

[0073] In step S2, in the spatiotemporal attention module, the spatial attention layer uses a graph attention network to assign attention weights within the node neighborhood, and the temporal attention layer uses a self-attention mechanism to encode the temporal correlation between consecutive network state snapshots. The two are fused to form a comprehensive feature representation based on which the mapping decision is based;

[0074] In step S2, the steps of processing the network state snapshot using the spatiotemporal attention module are as follows:

[0075] To form the spatiotemporal feature representation of node state, a continuous network state snapshot sequence is constructed Each snapshot G t =(V,E,X t ), including the node set V, the edge set E and the node feature matrix X t ;

[0076] The spatial attention layer calculates the attention weight at layer l, and the formula is:

[0077] in,

[0078] Indicates the attention allocation weight of node i to its neighbor node j in the lth layer, SoftMax j Indicates the softmax normalization operation on index j, a (l) Represents the attention weight vector of the lth layer, W (l) is the transformation matrix of the lth layer, represents the feature of node i in layer l, represents the feature of node j at layer l, and the symbol | represents vector concatenation;

[0079] The spatial encoding output is:

[0080]

[0081] in, represents the aggregate representation of node i at layer l+1, W (l) represents the transformation matrix of the lth layer, σ is the activation function;

[0082] In step S2, the step of processing the network state snapshot using the spatiotemporal attention module further includes:

[0083] Representation of node i in consecutive snapshots by the temporal attention layer Calculate the similarity score using the formula:

[0084] in,

[0085] β t,k represents the temporal attention weight between time t and time k, represents the similarity function between node i and k at time t, T represents the total number of time steps, represents the feature representation of node i at time step t;

[0086] The final spatiotemporal feature of node i is represented as h i :

[0087]

[0088] Among them, β t Represents the weighting coefficient corresponding to the time step;

[0089] Specifically, we construct a spatiotemporal graph model for continuous network state snapshots and use a hierarchical attention mechanism to achieve spatiotemporal fusion of node information:

[0090] The spatial attention layer uses neighborhood information to perform weighted aggregation of node features and outputs the encoding through an activation function. Secondly, the temporal attention layer calculates similarities between multiple moments and generates weighted coefficients, thereby integrating information from different moments into a unified representation. This helps to obtain a more stable description of node states in a dynamic environment and form a complete spatiotemporal feature representation.

[0091] Step S3, performing two-stage mapping on the candidate mapping set based on the spatiotemporal feature representation to obtain mapping parameters;

[0092] In the two-stage mapping, the first stage uses a graph convolutional network to perform node mapping on virtual network requests, and the second stage uses the Floyd algorithm to calculate the link paths between candidate nodes to form a virtual link map;

[0093] In step S3, the steps of implementing two-stage mapping on the candidate mapping set are:

[0094] For the mapping between virtual network requests and physical network resources, a two-stage mapping is adopted. In the first stage, the graph convolutional network (GCN) is used to map the nodes in the virtual network request. The node mapping expression is:

[0095]

[0096] Among them, y i represents the mapping output of virtual node i, represents the normalized adjacency matrix, X j is the characteristic of physical node j, W g is the weight matrix of the graph convolution layer;

[0097] In the second stage, the Floyd algorithm is used to calculate the link paths between the candidate nodes in the first stage. The virtual link mapping cost calculation formula is:

[0098]

[0099] Among them, d ij represents the cumulative cost of the mapping link between node i and node j, Indicates taking the minimum value among all paths, ∑ (u,v)∈path l uv represents the cumulative link cost of all edges (u, v) in the path,

[0100] Among them, l uv represents the link cost of edge (u,v);

[0101] Specifically, a two-stage mapping approach is used here to implement resource allocation for virtual network requests in the physical network. A graph convolutional network maps virtual node requests to physical nodes, fusing neighborhood information through an adjacency matrix and a weight matrix to obtain node mapping results. Next, the Floyd algorithm calculates the link costs between candidate nodes and determines the link mapping path. This two-stage mapping combines node mapping with link path evaluation to form an overall plan for nodes and links in the virtual network in the physical network.

[0102] Step S4: Using the trust region policy gradient optimization module to adjust the mapping parameters. The trust region policy gradient optimization module constrains the parameter update by setting the parameter update trust region and calculates the migration cost during the mapping process according to the predetermined reward function.

[0103] In step S4,

[0104] The trust region policy gradient optimization module is used to adjust the mapping parameters.

[0105] The trust region policy gradient optimization module has a trust region for parameter updates.

[0106] And the trust region constrains the parameter update amplitude with a predetermined KL divergence threshold;

[0107] In step S4, the step of adjusting the mapping parameters using the trust region policy gradient optimization module is as follows:

[0108] In the mapping parameter adjustment, the mapping migration cost is quantified by constructing a reward function R, which is formulated as follows:

[0109]

[0110] Among them, Rev i represents the benefit brought by mapping, ∑ i Rev i Indicates the sum of the benefits generated by all mappings, Cost i represents the migration cost during the mapping process,

[0111] At the same time, define KL divergence:

[0112]

[0113] Among them, D KL (π old ||π) represents the old policy π old The KL divergence between the new policy π, πold (a|s) and π(a|s) represent the distribution of the old strategy and the new strategy for action a in state s, respectively. a π old (a|s) represents the sum of the probabilities of all actions a;

[0114] Set the threshold to meet:

[0115] D KL (π old ||π)≤δ,

[0116] Where δ represents the predetermined KL divergence threshold;

[0117] The trust region policy gradient update formula is used to update the mapping parameters. The update formula is:

[0118]

[0119] Among them, W new represents the updated mapping parameters, W old represents the mapping parameters before updating, L(W) represents the objective function based on the reward function R, represents the gradient of the objective function L(W) with respect to the parameter W, η is the learning rate,

[0120] Specifically, a trust region policy gradient optimization module is used here to adjust the mapping parameters. By constructing a reward function that reflects the ratio of the mapping process benefit to the migration cost, the mapping results are quantitatively evaluated. Subsequently, the difference between policies is measured using the KL divergence, and a divergence threshold is set to limit the parameter update amplitude. The update process uses the policy gradient method, so that the parameters are corrected at a predetermined learning rate within the trust region. The dynamic cost of the mapping process is coordinated with the policy update, thereby quantifying and constraining the migration cost in the continuous spatiotemporal mapping process, forming a stable mapping parameter update mechanism.

[0121] Step S5, using the above mapping results to adjust the measurement, operation and control task allocation between the low-orbit satellite and the space-ground platform;

[0122] The scheduling control unit schedules the measurement, operation and control tasks between the low-orbit satellite and the space-ground platform based on the above-mentioned mapping results and parameter adjustment results. The scheduling control unit uses continuous network status snapshots as a reference and allocates tasks during the mapping process across time slots.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for integrating low-orbit satellite measurement, operation and control resources, characterized by: include, Step S1: construct a resource status acquisition unit based on a time-varying graph model to generate a network status snapshot; Step S2: Processing the network state snapshots using a spatiotemporal attention module. The spatiotemporal attention module includes a spatial attention layer and a temporal attention layer. The former uses a graph attention network to weight information about node neighborhoods, and the latter uses a self-attention mechanism to perform temporal correlation encoding on states between consecutive snapshots to form a spatiotemporal feature representation of the node state. Step S3, performing two-stage mapping on the candidate mapping set according to the spatiotemporal feature representation to obtain mapping parameters; Step S4, adjusting the mapping parameters using a trust region policy gradient optimization module, wherein the trust region policy gradient optimization module constrains parameter updates by setting a parameter update trust region, and calculates the migration cost during the mapping process according to a predetermined reward function; Step S5, using the above mapping results to adjust the measurement, operation and control task allocation between the low-orbit satellite and the space-ground platform; In step S2, the step of processing the network state snapshot using the spatiotemporal attention module includes: Representation of node i in consecutive snapshots by the temporal attention layer Calculate the similarity score using the formula: in, β t,k represents the temporal attention weight between time t and time k, represents the similarity function between node i and k at time t, T represents the total number of time steps, represents the feature representation of node i at time step t; The final spatiotemporal feature of node i is represented as h i : Among them, β t Represents the weighting coefficient corresponding to the time step; In step S4, the step of adjusting the mapping parameters using the trust region policy gradient optimization module is as follows: In the mapping parameter adjustment, the mapping migration cost is quantified by constructing a reward function R, which is formulated as follows: Among them, Rev i represents the benefit brought by mapping, ∑ i Rev i Indicates the sum of the benefits generated by all mappings, Cost i represents the migration cost during the mapping process, At the same time, define KL divergence: Among them, D KL (π old ||π) represents the old policy π old The KL divergence between the new policy π, π old (a|s) and π(a|s) represent the distribution of the old strategy and the new strategy for action a in state s, respectively. a π old (a|s) represents the sum of the probabilities of all actions a; Set the threshold to meet: D KL (p old ||π)≤δ, Where δ represents the predetermined KL divergence threshold; The trust region policy gradient update formula is used to update the mapping parameters. The update formula is: Among them, W new represents the updated mapping parameters, W old represents the mapping parameters before updating, L(W) represents the objective function based on the reward function R, represents the gradient of the objective function L(W) with respect to the parameter W, and η is the learning rate.

2. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 1, wherein: In step S1, the resource status acquisition unit collects real-time data from low-orbit satellites, aerial platforms, and ground station nodes, and generates network status snapshots at predetermined time intervals; In step S1, the data collected by the resource status collection unit includes computing resources Cpu, link bandwidth Bdw, remaining computing resources and link transmission delay.

3. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 2, wherein: In step S2, in the spatiotemporal attention module, the spatial attention layer uses a graph attention network to assign attention weights within the node neighborhood, and the temporal attention layer uses a self-attention mechanism to encode the temporal associations between consecutive network state snapshots. The two are fused to form a comprehensive feature representation on which the mapping decision is based.

4. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 3, wherein: In step S2, the step of processing the network state snapshot using the spatiotemporal attention module is as follows: To form the spatiotemporal feature representation of node state, a continuous network state snapshot sequence is constructed Each snapshot G t =(V,E,X t ), including the node set V, the edge set E and the node feature matrix X t ; The spatial attention layer calculates the attention weight at layer l, and the formula is: in, Indicates the attention allocation weight of node i to its neighbor node j in the lth layer, SoftMax j Indicates the softmax normalization operation on index j, a (l) Represents the attention weight vector of the lth layer, W (l) is the transformation matrix of the lth layer, represents the feature of node i in layer l, represents the feature of node j at layer l, and the symbol | represents vector concatenation; The spatial encoding output is: in, represents the aggregate representation of node i at layer l+1, W (l) represents the transformation matrix of the lth layer, and σ is the activation function.

5. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 4, characterized in that: In the two-stage mapping, the first stage uses a graph convolutional network to perform node mapping on virtual network requests, and the second stage uses the Floyd algorithm to calculate the link paths between candidate nodes to form a virtual link mapping.

6. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 5, characterized in that: In step S3, the step of implementing two-stage mapping on the candidate mapping set is: For the mapping between virtual network requests and physical network resources, a two-stage mapping is adopted. In the first stage, the graph convolutional network (GCN) is used to map the nodes in the virtual network request. The node mapping expression is: Among them, y i represents the mapping output of virtual node i, represents the normalized adjacency matrix, X j is the characteristic of physical node j, W g is the weight matrix of the graph convolution layer; In the second stage, the Floyd algorithm is used to calculate the link paths between the candidate nodes in the first stage. The virtual link mapping cost calculation formula is: Among them, d ij represents the cumulative cost of the mapping link between node i and node j, Indicates taking the minimum value among all paths, ∑ (u,v)∈path l uv represents the cumulative link cost of all edges (u, v) in the path, Among them, l uv represents the link cost of edge (u,v).

7. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 6, wherein: In step S4, The trust region policy gradient optimization module is used to adjust the mapping parameters. The trust region policy gradient optimization module is provided with a trust region for parameter updates. The confidence region constrains the parameter update amplitude with a predetermined KL divergence threshold.

8. The method for integrating low-orbit satellite measurement, operation and control resources according to claim 7, characterized in that: The scheduling control unit schedules the measurement, operation and control tasks between the low-orbit satellite and the space-ground platform based on the aforementioned mapping results and parameter adjustment results. The scheduling control unit uses continuous network status snapshots as a reference and allocates tasks during the mapping process across time slots. In step S5, the aforementioned mapping result refers to the candidate mapping parameter obtained in step S3 and the adjustment result of the parameter in step S4.

Images