Inter-satellite link dynamic control method and device of multi-layer satellite constellation and storage medium

Abstract

The application provides an inter-satellite link dynamic control method, device and storage medium of a multi-layer satellite constellation, which comprises the following steps: acquiring an inter-satellite distance threshold matrix and an inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation; optimizing the inter-satellite distance threshold matrix through a preset optimization algorithm to generate a preliminary optimized adjacency matrix meeting the inter-satellite link network topology constraint; generating a node feature matrix and inputting the same into a graph convolutional neural network to obtain a network feature vector; generating a link exchange strategy candidate set; inputting the network feature vector into an inter-satellite link control model and taking the link exchange strategy candidate set as an action space to generate a target link exchange strategy; updating the preliminary optimized adjacency matrix according to the target link exchange strategy, performing performance evaluation according to the updated adjacency matrix, and adjusting the inter-satellite links among the satellite nodes in the multi-layer satellite constellation. The application can improve the control performance and reliability of the inter-satellite link control and meet the actual needs of modern satellite network topology optimization.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a method, apparatus and storage medium for dynamic control of inter-satellite links in a multi-layer satellite constellation. Background Technology

[0002] With the rapid development of the information society and next-generation communication technologies, low Earth orbit (LEO) satellite network technology has gradually become a research hotspot. In this field, multi-layered satellite networks composed of satellites at different orbital altitudes have become an important research direction due to their wider coverage and more complex application scenarios. Therefore, conducting research on multi-layered LEO satellite networks is of vital significance to the development of next-generation communication technologies in modern society. In satellite networks, inter-satellite links connect multiple satellites to form the network. From a mathematical modeling perspective, this network can be described as a satellite topology. However, too many inter-satellite links lead to increased network overhead, while insufficient links cannot meet communication needs. Therefore, designing better inter-satellite link construction methods and optimizing satellite topology is a problem worthy of in-depth research.

[0003] Traditional satellite network topology control methods include heuristic search, genetic algorithms, and value ranking. Existing research has proposed various schemes from the perspectives of communication orbit determination, granular computation optimization, and polar-orbiting constellation partitioning, achieving some success in small-scale or single-layer constellations.

[0004] However, existing methods are mostly limited to constellation models with simple structures and limited scale, and are often designed based on idealized assumptions. They do not fully consider actual factors such as dynamic link constraints and changes in the space environment in real multi-layer satellite constellations, resulting in insufficient control performance and reliability in large-scale, highly dynamic scenarios, making it difficult to meet the actual needs of modern satellite network topology optimization. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method, apparatus, and storage medium for dynamic control of inter-satellite links in a multi-layer satellite constellation, to eliminate or improve one or more defects existing in the prior art. It can solve the problem that existing satellite topology control methods suffer from insufficient control performance and reliability in large-scale, highly dynamic scenarios, making it difficult to meet the practical needs of modern satellite network topology optimization.

[0006] One aspect of the present invention provides a dynamic control method for inter-satellite links in a multi-layer satellite constellation, the method comprising the following steps: Obtain the inter-satellite distance threshold matrix and inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation; wherein, the inter-satellite distance threshold matrix is ​​used to characterize whether the distance between each pair of satellite nodes is less than or equal to the inter-satellite distance threshold; the inter-satellite link adjacency matrix is ​​used to characterize the inter-satellite link connection relationship between each satellite node; The inter-satellite distance threshold matrix is ​​optimized by a preset optimization algorithm to generate a preliminary optimized adjacency matrix that satisfies the topological constraints of the inter-satellite link network of a multi-layer satellite constellation. A node feature matrix is ​​generated based on the satellite node network topology characteristics indicated by the inter-satellite link adjacency matrix and the spatial location characteristics of each satellite node. The node feature matrix is ​​input into a pre-trained graph convolutional neural network to obtain network feature vectors that characterize the topological performance of a multi-layer satellite constellation network. The initial optimized adjacency matrix is ​​matched with the inter-satellite distance threshold matrix to filter out inter-satellite link pairs that meet the preset link exchange conditions, and a candidate set of link exchange strategies is generated based on the inter-satellite link pairs. The network feature vector is used as the state parameter input to the pre-trained inter-satellite link control model, and the candidate set of link switching strategies is used as the action space. The inter-satellite link control model outputs a probability distribution based on the state parameter, and samples from the action space based on the probability distribution to generate the target link switching strategy. The inter-satellite link control model is trained based on reinforcement learning and aims to maximize the natural connectivity and constellation network capacity of the multi-layer satellite constellation by constructing a target optimization function. According to the target link switching strategy, the preliminary optimized adjacency matrix is ​​updated, the performance is evaluated according to the updated adjacency matrix, and the inter-satellite links between satellite nodes in the multi-layer satellite constellation are adjusted.

[0007] In some embodiments of the present invention, the inter-satellite link control model is trained through the following steps: The main neural network for obtaining the actor critic network structure; the main neural network includes the initial policy network and the initial value network; Obtain the sample inter-satellite distance threshold matrix and the sample inter-satellite link adjacency matrix; The sample inter-satellite distance threshold matrix is ​​optimized by a preset optimization algorithm to generate an initial sample optimized adjacency matrix that satisfies the link network topology constraints. Input the node feature matrix corresponding to the sample optimized neighbor matrix into the graph convolutional neural network, and output the sample network feature vector as the sample state parameter of the policy network. The sample optimized adjacency matrix is ​​matched with the sample inter-satellite distance threshold matrix to select sample inter-satellite link pairs that meet the preset link exchange conditions, and a sample link exchange policy candidate set is generated based on the sample inter-satellite link pairs as the sample action space of the policy network. The policy network outputs the sample probability distribution based on the sample state parameters, and generates the sample target link exchange policy by sampling from the sample action space based on the sample probability distribution; the value network generates the value corresponding to the sample state parameters based on the sample state parameters. Based on the sample target link exchange strategy and the sample optimized adjacency matrix, a candidate adjacency matrix is ​​generated; If the candidate adjacency matrix satisfies the link network topology constraints, the reward value is calculated by optimizing the adjacency matrix and the candidate adjacency matrix based on the samples and combining them with a preset reward function. The advantage value and target value are calculated using the reward value. The strategy network is updated by limiting the range of change in the probability ratio between the new and old strategies, based on the advantage value. The value network is updated by minimizing the value prediction error, based on the target value. The candidate adjacency matrix is ​​used as the new sample optimized adjacency matrix, and the candidate state parameters are used as the new sample state parameters. The process of matching the sample optimized adjacency matrix with the sample inter-satellite distance threshold matrix is ​​performed. This process is repeated multiple times until a preset termination condition is met. The policy network obtained from the multiple iterations is then used as the inter-satellite link control model.

[0008] In some embodiments of the present invention, a reward value is calculated based on the sample-optimized adjacency matrix and the candidate adjacency matrix, combined with a preset reward function; wherein, the formula for the preset reward function is: In the formula, Indicates training steps The corresponding reward value; Represents the candidate adjacency matrix; This represents the optimized adjacency matrix of the samples. The objective function is the optimization function. The formula for calculation is: In the formula, Indicates in Time-based training steps Optimize the adjacency matrix of the samples below; Indicates the normalization parameter; Indicates the normalization parameter; This indicates that the multi-layered satellite constellation optimizes the adjacency matrix in the sample. The natural connectivity is calculated as follows: In the formula, Represents the natural logarithm function; Represents the inter-satellite link adjacency matrix of the samples The number of rows and columns is also the number of satellite nodes in a multi-layered satellite constellation; This represents the solution formula for the trace of the sample-optimized adjacency matrix, used to optimize the adjacency matrix of the samples, and represents the eigenvalues ​​corresponding to the sample-optimized adjacency matrix. This represents a Taylor expansion of the optimized adjacency matrix of the samples, where, Indicates the index of the series form after expansion; , representing the constellation network capacity of a multi-layered satellite constellation under the sample-optimized adjacency matrix, is calculated as follows: In the formula, Indicates channel condition parameters; Indicates the baseline link capacity; Indicates link distance; This represents the summation function; This represents a distance matrix used to indicate the distances between satellite nodes.

[0009] In some embodiments of the present invention, the advantage value and the target value are calculated using the reward value; wherein, the formula for calculating the advantage value is: In the formula, Indicates training steps The corresponding advantage value; Indicates training steps The corresponding reward value; Indicates the discount factor; Represents a value network; This represents the state parameters generated based on the candidate adjacency matrix; Indicates the sample state parameters; Indicates training steps The corresponding judgment value, after completing all training steps in one iteration of training, The value is 1 if it is 1, otherwise it is 0. This indicates that the dominance value has been normalized, where, This represents the mean of the dominance values; The standard deviation of the dominance value; The formula for calculating the target value is: In the formula, Indicates training steps Corresponding target value; Indicates the discount factor; Represents a value network; This represents the state parameters generated based on the candidate adjacency matrix; Indicates training steps The corresponding judgment value.

[0010] In some embodiments of the present invention, the policy network is updated by combining the advantage value and limiting the range of variation of the probability ratio between the new and old policies. The loss function is calculated as follows: In the formula, Indicates the target link switching strategy in the sample. The loss function corresponding to the policy network; Represents the expected value function; Describes the minimum value function; Indicating in the training steps Below, the target link switching strategy for the sample The corresponding probability ratio; Indicates training steps The corresponding advantage value; This represents a function that restricts the range of values. Limited to to between; Indicates the clipping factor; The value network is updated by minimizing the value prediction error based on the target value. The loss function is calculated as follows: In the formula, In the sample target link switching strategy The loss function corresponding to the value network; Represents the expected value function; Indicates training steps Corresponding target value; Represents a value network; This represents the sample state parameters.

[0011] In some embodiments of the present invention, the graph convolutional neural network includes L graph convolutional layers; wherein L is an integer greater than or equal to 3; the value of L is positively correlated with the number of satellite nodes in the multi-layer satellite constellation; The policy network consists of an input layer, L-2 hidden layers, and an output layer; The value network consists of an input layer, L-2 hidden layers, and an output layer.

[0012] In some embodiments of the present invention, the inter-satellite distance threshold matrix and the inter-satellite link adjacency matrix are symmetric matrices; obtaining the inter-satellite distance threshold matrix and the inter-satellite link adjacency matrix corresponding to a multi-layer satellite constellation includes: Based on the latitude, longitude, and altitude information of each satellite node in the multi-layer satellite constellation, the spatial position coordinates of each satellite node in the geocentric-geofixed coordinate system are obtained. Based on the spatial coordinates of each satellite node, the inter-satellite distance between satellite nodes is calculated and compared with the inter-satellite distance threshold. Based on the comparison results, an inter-satellite distance threshold matrix is ​​constructed. An inter-satellite link adjacency matrix is ​​constructed based on the inter-satellite link connection relationships between each satellite node.

[0013] In some embodiments of the present invention, each inter-satellite link pair includes two independent inter-satellite links; The initial optimized adjacency matrix is ​​matched with the inter-satellite distance threshold matrix to filter out inter-satellite link pairs that meet the preset link switching conditions. Based on the inter-satellite link pairs, a candidate set of link switching strategies is generated, including: Inter-satellite link relationships are extracted based on the preliminary optimized adjacency matrix, resulting in an inter-satellite link set; Based on the inter-satellite distance threshold matrix, link relationships that satisfy the inter-satellite distance threshold are extracted to obtain a set of valid links; The inter-satellite link set is matched with the valid link set, and link pairs that belong to both sets and are independent are selected. For each link pair, if the two new links obtained after link swapping belong to the set of valid links and not to the set of inter-satellite links, it is determined that the preset link swapping conditions are met, and the cosine similarity value between the two new links is calculated. If the cosine similarity value is equal to the preset similarity threshold, the link swapping is used as the link swapping strategy and added to the link swapping strategy candidate set.

[0014] Another aspect of the present invention provides a dynamic control device for inter-satellite links of a multi-layer satellite constellation, including a processor, a memory, and a computer program / instructions stored in the memory. The processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the device implements the steps of the dynamic control method for inter-satellite links of the multi-layer satellite constellation as described above.

[0015] Another aspect of the present invention provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implements the steps of the inter-satellite link dynamic control method for a multi-layer satellite constellation as described above.

[0016] The inter-satellite link dynamic control method and apparatus of this invention can solve the problem that existing satellite topology control methods are insufficient in control performance and reliability in large-scale, highly dynamic scenarios, making it difficult to meet the actual needs of modern satellite network topology optimization. By generating a preliminary adjacency matrix that satisfies basic topological constraints through a preset optimization algorithm, the complexity of problem solving is significantly reduced, laying an efficient starting point for subsequent fine-grained optimization and thus effectively improving control performance. Based on this, a graph convolutional neural network is used to process the node feature matrix containing satellite spatial positions and topological relationships, enabling it to perceive actual physical constraints such as inter-satellite distances and inter-layer geometric relationships, avoiding biases caused by idealized assumptions and enhancing the practicality and reliability of the strategy. The graph convolutional neural network efficiently learns global network features, compressing key topology performance information into low-dimensional feature vectors, avoiding traditional high-dimensional matrix operations, and further improving computational performance in large-scale scenarios. Furthermore, the network feature vectors are used as state inputs to the inter-satellite link control model, and the link exchange candidate set is used as the action space to generate dynamic strategies through probability sampling. This strategy strictly uses a distance threshold matrix for physical feasibility screening and achieves link adaptive adjustment based on real-time topology performance through a "state-action-evaluation" loop, thereby maintaining high system reliability in complex dynamic environments.

[0017] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0018] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings: Figure 1 A flowchart of a dynamic control method for inter-satellite links in a multi-layer satellite constellation provided in an embodiment of the present invention.

[0020] Figure 2 This is a flowchart illustrating the training method of an inter-satellite link control model according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0022] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0023] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0024] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0025] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0026] The following section provides a detailed description of the inter-satellite link dynamic control method for multi-layer satellite constellations provided in this application.

[0027] like Figure 1 As shown, embodiments of this application provide a dynamic control method for inter-satellite links in a multi-layer satellite constellation, comprising at least the following steps S101~S107: Step S101: Obtain the inter-satellite distance threshold matrix and inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation.

[0028] The inter-satellite distance threshold matrix and the inter-satellite link matrix are symmetric matrices. The inter-satellite distance threshold matrix is ​​used to characterize whether the distance between each pair of satellite nodes is less than or equal to the inter-satellite distance threshold. The inter-satellite link adjacency matrix is ​​used to characterize the inter-satellite link connection relationship between each satellite node.

[0029] In some embodiments of the present invention, the inter-satellite distance threshold matrix and inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation are obtained according to a preset discrete time series. The discrete time series can be represented as: in, This indicates the moment when a multi-layered satellite constellation begins normal operation and can establish communication links.

[0030] To discretize the topology network model corresponding to a multi-layer satellite constellation, it is necessary to ensure that the resulting discrete model is an effective approximation of the original model, and the switching time in the topology network model... In discrete time series, it is necessary to select an appropriate time slice length. Discrete time interval ( , ).in, satisfy , The formula for calculation is: In the formula, It represents the average angular velocity of all satellite nodes in a multi-layered satellite constellation; Represents the inverse cosine function; This represents the total number of satellite nodes in a multi-layered satellite constellation. This represents the average orbital radius of all satellite nodes; This represents the speed of light in a vacuum. This indicates the number of orbital planes corresponding to a multi-layered satellite constellation; Represent it as a positive number to ensure Much less than 1; Represents the cosine function; Represents the sine function; This represents the average orbital inclination of all satellite nodes; It represents pi (π).

[0031] For any time period in the discrete time series According to the time period Snapshots of corresponding multi-layered satellite constellations, total number of satellite nodes Number of orbital planes Phase factor Initial value of right ascension of ascending node Data such as these are used to obtain information about multi-layered satellite constellations during specific time periods. Next satellite nodes latitude and longitude coordinates ( , ), orbital altitude Latitude and longitude coordinates, along with orbital altitude, constitute the latitude, longitude, and altitude (LLA) coordinate system of a satellite node. Among these, the time period... , indicating time arrive This continuous period of time.

[0032] In some embodiments of the present invention, each satellite node is in The LLA coordinates for each time period are converted to Earth-Centered, Earth-Fixed (ECEF) coordinates. The distances between satellite nodes are calculated using the corresponding ECEF coordinates and compared with inter-satellite distance thresholds to generate a multi-layered satellite constellation. Inter-satellite distance threshold matrix at any given time; based on the distances of each satellite node at... Inter-satellite link connections over time periods generate multi-layered satellite constellations. Inter-satellite link adjacency matrix under different time periods.

[0033] The inter-satellite distance threshold refers to the maximum visible distance between satellite nodes. The formula for calculating the inter-satellite distance threshold is: In the formula, Indicates the maximum line-of-sight distance between satellite nodes; This represents the Earth's average radius; This indicates the altitude of the satellite above the Earth's surface.

[0034] Specifically, the process of obtaining the inter-satellite distance threshold matrix and inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation includes: obtaining the spatial position coordinates of each satellite node in the geocentric geofixed coordinate system based on the latitude, longitude and altitude information of each satellite node in the multi-layer satellite constellation; calculating the inter-satellite distance between satellite nodes based on the spatial position coordinates of each satellite node, comparing it with the inter-satellite distance threshold, and constructing the inter-satellite distance threshold matrix based on the comparison results; and constructing the inter-satellite link adjacency matrix based on the inter-satellite link connection relationship between each satellite node.

[0035] Let the LLA coordinates of any satellite node be ( , , Then, the coordinates of the satellite node in the ECEF coordinate system can be obtained. , , ).in, axis, axis, The formulas for calculating the axis components are as follows: In the formula, This represents the x-coordinate of the satellite node in the ECEF coordinate system; This represents the ordinate of the satellite node in the ECEF coordinate system; This represents the vertical coordinate of the satellite node in the ECEF coordinate system; Indicates the longitude of the satellite node in the LLA coordinate system; Indicates the latitude of the satellite node in the LLA coordinate system; Indicates the altitude of the satellite node in the LLA coordinate system; Indicates the radius of curvature of the zonal loop; This represents the equatorial radius of the WGS84 ellipsoid, the internationally recognized geometric model of the Earth. This represents the square of the first eccentricity of the WGS84 ellipsoid.

[0036] Based on any two satellites and ECEF coordinate system , Calculate the satellite and satellite The interstellar distances between them. Traversal , , , , ( The inter-satellite distances between all satellites were calculated. Among them, the satellites... and satellite The formula for calculating the interstellar distance is: In the formula, Indicates satellite and satellite Interstellar distances; Indicates satellite The x-coordinate in the ECEF coordinate system; Indicates satellite The ordinate in the ECEF coordinate system; Indicates satellite Vertical coordinate in the ECEF coordinate system; Indicates satellite The x-coordinate in the ECEF coordinate system; Indicates satellite The ordinate in the ECEF coordinate system; Indicates satellite The vertical coordinate in the ECEF coordinate system.

[0037] The inter-satellite distance between satellite nodes and the inter-satellite distance threshold. Compare the results and generate the output. The inter-satellite distance threshold matrix for a multi-layered satellite constellation within a given time period. This inter-satellite distance threshold matrix can be expressed by the following formula: Interstellar distance threshold matrix In, elements Used to represent satellites and Whether it is visible, when hour, ;when hour, Among them, the elements on the diagonal .

[0038] Traversal The inter-satellite distance threshold matrix of the multi-layer satellite constellation at time t is obtained, including .

[0039] Generate based on the inter-satellite link relationships between each satellite node The inter-satellite link adjacency matrix of a multi-layered satellite constellation within a given time period. The inter-satellite link adjacency matrix can be expressed by the following formula: Inter-satellite link adjacency matrix In, elements express Satellite during the time period With satellite The connection relationships between them. Among them, , , , , ;when , indicating satellite and Inter-satellite links are established between satellites, when At that time, it indicates the satellite and No inter-satellite links have been established between the satellites. (Elements on the diagonal) .

[0040] Traversal The inter-satellite link adjacency matrix of the multi-layer satellite constellation under the time period is obtained, including .

[0041] In this embodiment, natural connectivity and constellation network capacity are selected as the two maximization indicators as the overall optimization objectives, and the corresponding results under these two objectives are obtained. The optimal adjacency matrix of the comprehensive solution under different time periods .

[0042] Natural connectivity is a resilience metric used to characterize graph redundancy, which can be represented by the eigenvalue spectral density of the matrix and is also related to the overall average connectivity of the constellation. Constellation network capacity is used to evaluate the overall transmission performance of links in the entire multi-layer satellite constellation network, and is related to the channel distance and channel conditions of the links. Specifically, the objective optimization function can be expressed by the following equation: In the formula, Indicating a multi-layered satellite constellation Inter-satellite link adjacency matrix for a given time period; Indicates the normalization parameter; Indicates the normalization parameter; Indicates multi-layered satellite constellations in the adjacency matrix The natural connectivity of a given condition can be expressed by the following formula: In the formula, Represents the natural logarithm function; Represents the inter-satellite link adjacency matrix The number of rows and columns is also the number of satellite nodes in a multi-layered satellite constellation; Represents the adjacency matrix of inter-satellite links The solution for the trace is used to solve for the inter-satellite link adjacency matrix. , representing the inter-satellite link adjacency matrix The corresponding eigenvalues; Represents the adjacency matrix of inter-satellite links Perform a Taylor expansion, where, Indicates the index of the series form after expansion.

[0043] According to the definition of natural connectivity, its original expression is: In the formula, Representing the adjacency matrix eigenvalues, This represents the number of rows and columns of the matrix.

[0044] To transform the expression for natural connectivity into an adjacency matrix The relevant formulas are expressed in this embodiment, which introduce calculation formulas. Calculation formula Used to solve for the trace of a matrix, thus enabling the use of adjacency matrices. Represents the eigenvalues ​​of the matrix At the same time, the adjacency matrix also needs to be... Perform a Taylor expansion to transform it into an exponential summation, thus obtaining the form in the objective function expression.

[0045] In the objective optimization function, Represents the adjacency matrix of inter-satellite links in a multi-layered satellite constellation. The capacity of the constellation network can be expressed by the following formula: In the formula, Represents random channel condition parameters; Indicates the baseline link capacity. Indicates link distance. and It does not change with the environment; This represents the summation function; This represents a distance matrix used to indicate the distances between satellite nodes.

[0046] According to the definition of network capacity, its original expression is: In order to transform the original expression using an adjacency matrix To represent this, a distance matrix between all satellites needs to be introduced. Multiply it by the adjacency matrix Then use By summing all elements in the matrix, the distance index between all satellites that have established inter-satellite links is obtained, which in turn allows the calculation of the overall constellation network capacity and the form of the objective optimization function expression.

[0047] In this embodiment, the two optimization objectives in the objective optimization function are both maximization objectives, and they are not directly related in computation. Therefore, there are no issues such as mutual calls during the solution process, and they can be regarded as two relatively independent objectives and separated.

[0048] Considering that the calculation of natural connectivity in the optimization objective is relatively simple and is related to the overall average degree of the multi-layer satellite constellation, maximizing natural connectivity can be used as a preliminary solution objective for the optimization problem. In this embodiment, a preset optimization algorithm is used to optimize the inter-satellite distance threshold matrix into a preliminary optimization matrix that satisfies the topological constraints of the inter-satellite link network of the multi-layer satellite constellation.

[0049] Step S102: The inter-satellite distance threshold matrix is ​​optimized using a preset optimization algorithm to generate a preliminary optimized adjacency matrix that satisfies the topological constraints of the inter-satellite link network of the multi-layer satellite constellation.

[0050] Among them, the preset optimization algorithm is a pre-defined optimization method used to optimize the inter-satellite distance threshold matrix. Generate a preliminary optimized adjacency matrix that satisfies the topological constraints of the inter-satellite link network.

[0051] In some embodiments of the present invention, the preset optimization algorithm is a meta-heuristic algorithm, such as simulated annealing, genetic algorithm, particle swarm optimization, or ant colony optimization; or, the preset optimization algorithm is a hyper-heuristic algorithm, such as a hyper-heuristic algorithm based on a greedy strategy or a hyper-heuristic algorithm based on a meta-heuristic algorithm. This embodiment does not limit the implementation method of the preset optimization algorithm.

[0052] Inter-satellite link network topology constraints refer to the constraints that the adjacency matrix generated in this embodiment must satisfy. Specifically, inter-satellite link network topology constraints can be expressed by the following formula: The formula includes four constraints related to the matrix, namely: First constraint This indicates that all elements in the inter-satellite distance threshold matrix and adjacency matrix can only have two values: 0 or 1.

[0053] Second constraint This indicates that the adjacency matrix must be a symmetric matrix.

[0054] The third constraint This indicates that the difference between the elements in the inter-satellite distance threshold matrix and the adjacency matrix is ​​0 or 1, meaning that the adjacency matrix needs to meet the basic condition of visibility.

[0055] Fourth constraint This indicates that the maximum degree constraint for each satellite node represented in the adjacency matrix is ​​. .

[0056] The inter-satellite distance threshold matrix is ​​optimized using a pre-defined algorithm. After optimization, the initial optimized adjacency matrix can be expressed by the following formula: Preliminary optimization of the adjacency matrix In, elements express Satellite during the time period With satellite The connection relationships between them. Among them, , , , , ;when , indicating satellite and Inter-satellite links are established between satellites, when At that time, it indicates the satellite and No inter-satellite links have been established between the satellites. (Elements on the diagonal) .

[0057] A preliminary optimized adjacency matrix is ​​obtained. Subsequently, to further optimize the solution, this embodiment uses an action-compression-based Graph Convolutional Network (GCN) and reinforcement learning to achieve further optimization of the adjacency matrix. Based on this, and taking natural connectivity and constellation network capacity as optimization objectives, the network topology constraints also need to be satisfied. A deep optimization of the satellite topology is then performed to obtain the optimal adjacency matrix. .

[0058] In some embodiments of the present invention, reinforcement learning includes reinforcement learning based on proximal policy optimization (PPO). In actual implementation, reinforcement learning can also be based on temporal networks or multilayer perceptrons. This embodiment does not limit the implementation method of reinforcement learning.

[0059] Step S103: Generate a node feature matrix based on the satellite node network topology features indicated by the inter-satellite link adjacency matrix and the spatial location features of each satellite node.

[0060] The satellite node network topology features include the degree features of the satellite nodes, i.e., the number of inter-satellite links corresponding to each satellite node. The spatial location features of the satellite nodes include their longitude, latitude, and altitude features.

[0061] During the period Any satellite node in a multi-layered satellite constellation eigenvectors It can be expressed by the following formula: In the formula, Indicates the time period Satellite nodes Normalized longitude; Indicates the time period Satellite nodes Latitude after normalization; Indicates the time period Satellite nodes Normalized height; For satellite nodes The degree, i.e., the number of links.

[0062] in, The normalization is calculated from the original range to... The calculation formulas obtained from the intervals are as follows: In the formula, Indicates the time period Satellite nodes Normalized longitude; Indicates the time period Satellite nodes The original longitude; Indicates the maximum longitude; Indicates the minimum longitude; Indicates the time period Satellite nodes Latitude after normalization; Indicates the time period Satellite nodes The original latitude; Indicates the maximum latitude; Indicates the minimum latitude; Indicates the time period Satellite nodes Normalized height; Indicates the time period Satellite nodes Original height; Indicates the maximum height; This indicates the minimum height.

[0063] Time period The feature vectors of all satellite nodes in the lower-level satellite constellation are concatenated to generate a node feature matrix, thus completely containing all the feature parameters of the entire network during that time period. The time period... The corresponding node feature matrix can be represented by the following formula: In the formula, Indicates time period The corresponding node feature matrix; Indicates time period corresponding satellite nodes eigenvectors; Indicates time period corresponding satellite nodes eigenvectors; Indicates time period corresponding satellite nodes eigenvectors; Indicates time period corresponding satellite nodes eigenvectors.

[0064] Step S104: Input the node feature matrix into the pre-trained graph convolutional neural network to obtain the network feature vector used to characterize the topological performance of the multi-layer satellite constellation network.

[0065] In some embodiments of the present invention, before inputting the node feature matrix into the graph convolutional neural network, it is necessary to determine the architectural parameters of the graph convolutional neural network, including the number of layers, the feature dimension of each layer, the activation function, etc. These parameters will determine how the network performs feature aggregation.

[0066] To balance network performance and resource consumption, and to ensure feature extraction capability while avoiding excessive computational complexity, in this embodiment, the number of graph convolutional layers in the graph convolutional neural network is positively correlated with the number of satellite nodes. When the number of satellite nodes is small, the number of graph convolutional layers is 3, and then as the number of satellite nodes increases, the number of layers also increases appropriately.

[0067] In some embodiments of the present invention, the graph convolutional neural network includes L graph convolutional layers.

[0068] Where L is an integer greater than or equal to 3. An L-layer graph convolutional layer consists of an input layer, intermediate layers, and an output layer. The feature dimension of the input layer is represented as... The feature dimensions of the intermediate layer are represented as , No. The weight matrix of the layer is represented as The feature dimension of the output layer is represented as .

[0069] In some embodiments of the present invention, the activation function of the graph convolutional layer is the ReLU function. This is done to address the vanishing gradient problem by introducing a nonlinear mapping. This represents the net input to each neuron in the graph convolutional layer.

[0070] The node feature matrix is ​​input into a graph convolutional neural network, and after... After the operation of the layered graph convolutional layer, feature information is output.

[0071] In some embodiments of the present invention, the feature information output by the graph convolutional neural network includes a node-level feature matrix containing the physical attributes of satellite nodes and global topology association information, and a network feature vector characterizing the overall performance of the current satellite network topology. The node-level feature matrix can be represented by the following formula: In the formula, This represents the node-level feature matrix, in which the first node... row vector For satellite The fusion characteristics.

[0072] The network feature vectors are obtained through mean pooling and can be expressed as follows: In the formula, Indicates time period The corresponding network feature vector; This represents the total number of satellite nodes in a multi-layered satellite constellation; Indicates satellite The fusion characteristics.

[0073] Step S105: The preliminary optimized adjacency matrix is ​​matched with the inter-satellite distance threshold matrix to select inter-satellite link pairs that meet the preset link exchange conditions, and a candidate set of link exchange strategies is generated based on the inter-satellite link pairs.

[0074] In some embodiments of the present invention, each inter-satellite link pair includes two independent inter-satellite links.

[0075] Due to the initial optimization of the adjacency matrix Since the inter-satellite link network topology constraints have already been met, this embodiment will achieve optimization through inter-satellite link switching in subsequent optimizations. That is, in the initial optimization of the adjacency matrix... Based on this, find its relative to the inter-satellite distance threshold matrix. Inter-satellite links that were disconnected during the initial optimization operation will be reconnected to these disconnected links to achieve the desired optimization results. (Time period) The corresponding preliminary optimized adjacency matrix Interstellar distance threshold matrix Both contain a large number of elements equal to 1, representing connections between two satellite nodes. Based on this, valid link sets are defined respectively. Inter-satellite link set These two sets of connections can be represented by the following formula: In the formula, Represents the set of valid links; Represents the set of inter-satellite links; and This represents the total number of links in the corresponding link set; Represents the inter-satellite distance threshold matrix In the middle, satellite nodes With satellite nodes There are links between them; This indicates the initial optimization of the adjacency matrix. In the middle, satellite nodes With satellite nodes There are links between them.

[0076] Effective link set Inter-satellite link set The comparison process selects inter-satellite link pairs that meet the preset link exchange conditions. These preset exchange conditions refer to the conditions that must be met for two independent inter-satellite links to exchange. For any two node pairs, including satellite nodes... Satellite nodes Satellite nodes and satellite nodes There is no intersection between the two pairs of nodes, satisfying... .

[0077] At satellite nodes With satellite nodes Inter-satellite links exist Satellite nodes and satellite nodes Inter-satellite links exist In this case, before the inter-satellite link exchange, the inter-satellite link... Inter-satellite links It belongs to both the set of valid links and the set of inter-satellite links; after inter-satellite link switching, the inter-satellite links Inter-satellite links It belongs to the set of valid links but not to the set of inter-satellite links.

[0078] Specifically, the preset link switching conditions can be expressed by the following formula: In the formula, Represents the set of valid links; Represents the set of inter-satellite links; Indicates satellite node With satellite nodes Inter-satellite links; Indicates satellite node and satellite nodes Inter-satellite links; Indicates satellite node With satellite nodes Inter-satellite links; Indicates satellite node With satellite nodes Inter-satellite links between them.

[0079] This clearly shows that the link switching process involves the satellite nodes... and satellite nodes The link between them is broken, satellite nodes and satellite nodes The link between them was broken; these two links were then adjusted to satellite nodes. and satellite nodes Links between them, and satellite nodes and satellite nodes The link between them. That is, delete the original link. , (make ), and add new links , (make During this process, since both the number of newly disconnected links and the number of newly connected links are 2, the degree of the satellite node remains unchanged.

[0080] Filter and organize to obtain all actions that meet the preset exchange conditions And labeled in a preset order , ... .in, This represents the total number of actions. Based on this, the action space of the near-end policy optimization algorithm can be expressed as: .

[0081] To further reduce the action space of the proximal policy optimization algorithm, achieve action compression in proximal policy optimization, and improve the efficiency of the algorithm during the optimization process, we can build upon the previous steps by using the node-level feature matrix output by the graph convolutional neural network. Action compression is performed. This is done by calculating the cosine similarity value between the two new links after link swapping.

[0082] Specifically, the preliminary optimized adjacency matrix is ​​matched with the inter-satellite distance threshold matrix to filter out inter-satellite link pairs that meet the preset link exchange conditions. A candidate set of link exchange strategies is generated based on the inter-satellite link pairs, including: extracting inter-satellite link relationships based on the preliminary optimized adjacency matrix to obtain an inter-satellite link set; extracting link relationships that meet the inter-satellite distance threshold based on the inter-satellite distance threshold matrix to obtain an effective link set; matching the inter-satellite link set with the effective link set to filter out link pairs that belong to both sets and are independent; for each link pair, if the two new links obtained after link exchange belong to the effective link set but not to the inter-satellite link set, it is determined that the preset link exchange conditions are met, and the cosine similarity value between the two new links is calculated. If the cosine similarity value is equal to the preset similarity threshold, the link exchange is used as a link exchange strategy and added to the candidate set of link exchange strategies.

[0083] In this embodiment, satellite node With satellite nodes Inter-satellite links The eigenvectors of can be represented by the following formula: In the formula, Indicates inter-satellite link The corresponding feature vector; This represents the satellite nodes output by the graph convolutional neural network. The fusion characteristics; This represents the satellite nodes output by the graph convolutional neural network. The fusion characteristics.

[0084] Through two links , The cosine similarity between the feature vectors can be calculated using the following formula: In the formula, Represents the cosine similarity function; Indicates satellite node With satellite nodes Inter-satellite links; Indicates satellite node With satellite nodes Inter-satellite links; This represents the satellite nodes output by the graph convolutional neural network. The fusion characteristics; This represents the satellite nodes output by the graph convolutional neural network. The fusion characteristics; This represents the satellite nodes output by the graph convolutional neural network. The fusion characteristics; This represents the satellite nodes output by the graph convolutional neural network. The fusion characteristics.

[0085] Action groups are divided based on cosine similarity, and a predefined similarity threshold is used. For all cosine similarities, the following condition is met: Furthermore, for links that meet the preset link switching conditions, the corresponding link switching actions are grouped into the same set as link switching strategies, resulting in a candidate set of link switching strategies. Of the multiple link switching strategies in the candidate set, only one needs to be executed in subsequent operations. Therefore, the action space is significantly compressed, and the dimensions of the action space are reduced from... Reduce to .

[0086] Step S106: The network feature vector is input as a state parameter into the pre-trained inter-satellite link control model. The candidate set of link switching strategies is used as the action space. The inter-satellite link control model outputs a probability distribution based on the state parameters and samples the target link switching strategy from the action space based on the probability distribution. The inter-satellite link control model is trained using reinforcement learning and aims to maximize the natural connectivity and network capacity of the multi-layer satellite constellation by constructing a target optimization function.

[0087] Specifically, such as Figure 2 As shown, the inter-satellite link control model is trained through the following steps S10601~S10610: S10601, Obtain the main neural network structure of the Actor-Critic Network. The main neural network includes the initial policy network. (Actor) and the initial value network (Critic).

[0088] In some embodiments of the present invention, the graph convolutional neural network includes L graph convolutional layers. Here, L is an integer greater than or equal to 3; the value of L is positively correlated with the number of satellite nodes in the multi-layer satellite constellation.

[0089] The number of layers in the policy network structure is the same as the number of layers in the graph convolutional layer of the graph convolutional neural network, including the input dimension. Hidden layer dimension and output dimensions The value network has the same number of layers as the policy network, and its input and hidden layer dimensions are the same as the policy network, while its output dimension is 1. Specifically, the policy network consists of one input layer, L-2 hidden layers, and one output layer; the value network consists of one input layer, L-2 hidden layers, and one output layer.

[0090] S10602, obtain the sample inter-satellite distance threshold matrix and the sample inter-satellite link adjacency matrix.

[0091] In some embodiments of the present invention, the sample inter-satellite distance threshold matrix and the sample inter-satellite link adjacency matrix can be matrices generated based on simulated satellite data of a multi-layer satellite constellation (e.g., obtained using STK software or high-precision orbit model simulation), or matrices generated based on real satellite data of a multi-layer satellite constellation. This embodiment does not limit the implementation method of the sample inter-satellite distance threshold matrix and the sample inter-satellite link adjacency matrix.

[0092] In addition, a series of hyperparameters, including the learning rate, were pre-set before performing proximal policy optimization training. Discount Factor Cutting factor Number of training rounds Number of steps per round wait.

[0093] S10603 optimizes the inter-satellite distance threshold matrix of the samples using a preset optimization algorithm to generate an initial sample optimized adjacency matrix that satisfies the link network topology constraints.

[0094] In this embodiment, the sample inter-satellite distance threshold matrix is ​​optimized using a preset optimization algorithm to obtain an initial sample optimized adjacency matrix. Combined with time period The corresponding satellite orbit data is used to initialize the near-end strategy optimization environment and trajectory cache, and an empty list is created. Store transitional samples and initialize the list as follows: in, These are the state parameters for reinforcement learning after initialization; Indicates the action value; This represents the reward value calculated using the reward function; This represents the probability distribution value of the policy for near-end policy optimization; This represents the initial optimized adjacency matrix for the samples; This represents the initial judgment value. For the training step... After completing all training steps in one iteration training round In this case, The value is 1 if it is 1, otherwise it is 0.

[0095] S10604 inputs the node feature matrix corresponding to the sample optimized neighbor matrix into the graph convolutional neural network, and outputs the sample network feature vector as the sample state parameter of the policy network.

[0096] At any training step In the middle, the adjacency matrix is ​​optimized based on the current samples. Generate the corresponding node feature matrix The input is fed into a graph convolutional neural network to obtain sample network feature vectors. and update The state parameters in.

[0097] S10605, the sample optimized adjacency matrix is ​​matched with the sample inter-satellite distance threshold matrix to select sample inter-satellite link pairs that meet the preset link exchange conditions, and a sample link exchange policy candidate set is generated based on the sample inter-satellite link pairs as the sample action space of the policy network.

[0098] S10606 outputs the sample probability distribution based on the sample state parameters through the policy network, and generates the sample target link exchange policy by sampling from the sample action space based on the sample probability distribution; and generates the value corresponding to the sample state parameters through the value network based on the sample state parameters.

[0099] Specifically, the sample state parameters Incoming policy network In this process, through policy networks Output probability distribution This allows for the sampling of target link switching strategies. Through the value network... Generate sample state parameters Corresponding value .

[0100] S10607, Based on the sample target link exchange strategy and the sample optimized adjacency matrix, generate a candidate adjacency matrix.

[0101] Specifically, the target link switching strategy of the sample is parsed into specific link switching operations to generate a candidate adjacency matrix. .

[0102] S10608, assuming the candidate adjacency matrix satisfies the link network topology constraints, the reward value is calculated based on the sample-optimized adjacency matrix and candidate adjacency matrix, combined with a preset reward function. The preset reward function is calculated as follows: In the formula, Indicates training steps The corresponding reward value; Represents the candidate adjacency matrix; This represents the optimized adjacency matrix of the samples. The objective function is the optimization function. The formula for calculation is: In the formula, Indicates in Time-based training steps Optimize the adjacency matrix of the samples below; Indicates the normalization parameter; Indicates the normalization parameter; This indicates that the multi-layered satellite constellation optimizes the adjacency matrix in the sample. The natural connectivity of the following is expressed as: In the formula, Represents the natural logarithm function; Indicates the total number of satellite nodes; This represents the solution formula for the trace of the sample-optimized adjacency matrix, used to optimize the adjacency matrix of the samples, and represents the eigenvalues ​​corresponding to the sample-optimized adjacency matrix. This represents a Taylor expansion of the optimized adjacency matrix of the samples, where, Indicates the index of the series form after expansion; The constellation network capacity of a multi-layer satellite constellation under a sample-optimized adjacency matrix can be expressed by the following formula: In the formula, Indicates channel condition parameters; Indicates the baseline link capacity; Indicates link distance; This represents the summation function; This represents a distance matrix used to indicate the distances between satellite nodes.

[0103] If the candidate adjacency matrix satisfies the link network topology constraints, then set Apply penalties while keeping the sample-optimized adjacency matrix unchanged.

[0104] Update list Among them, when the training step season Stop this round of training, otherwise Continue this round of training.

[0105] S10609 uses the reward value to calculate the advantage value and the target value. The strategy network is updated by limiting the range of change in the probability ratio between the new and old strategies, based on the advantage value. The value network is updated by minimizing the value prediction error, based on the target value.

[0106] The formula for calculating the dominance value is: In the formula, Indicates training steps The corresponding advantage value; Indicates training steps The corresponding reward value; Indicates the discount factor; Represents a value network; This represents the state parameters generated based on the candidate adjacency matrix; Indicates the sample state parameters; Indicates training steps The corresponding judgment value, after completing all training steps in one iteration of training, The value is 1 if it is 1, otherwise it is 0. This indicates that the dominance value has been normalized, where, This represents the mean of the dominance values; The standard deviation of the dominance value.

[0107] The formula for calculating the target value is: In the formula, Indicates training steps Corresponding target value; Indicates the discount factor; Represents a value network; This represents the state parameters generated based on the candidate adjacency matrix; Indicates training steps The corresponding judgment value.

[0108] The policy network is updated by combining the advantage value with limiting the range of variation in the probability ratio between the old and new policies, while the Adam optimizer is used to minimize... To achieve the goal of maximizing optimization, the loss function is calculated as follows: In the formula, Indicates the target link switching strategy in the sample. The loss function corresponding to the policy network; Represents the expected value function; Describes the minimum value function; Indicating in the training steps Below, the target link switching strategy for the sample The corresponding probability ratio; Indicates training steps The corresponding advantage value; This represents a function that restricts the range of values. Limited to to between; This represents the clipping factor.

[0109] The value network is updated by minimizing the value prediction error based on the target value, and the Adam optimizer is used to minimize this error. The loss function is calculated as follows: In the formula, In the sample target link switching strategy The loss function corresponding to the value network; Represents the expected value function; Indicates training steps Corresponding target value; Represents a value network; This represents the sample state parameters.

[0110] S10610, the candidate adjacency matrix is ​​used as the new sample optimized adjacency matrix, the candidate state parameters are used as the new sample state parameters, and the step of matching the sample optimized adjacency matrix with the sample inter-satellite distance threshold matrix is ​​performed. Multiple rounds of iteration are carried out until the preset termination condition is met, and the policy network obtained from multiple rounds of iteration is used as the inter-satellite link control model.

[0111] In some embodiments of the present invention, the preset termination condition includes all training rounds. Or when the average reward change over 10 consecutive rounds is less than When the policy network converges, the policy network training is complete. The objective optimization function is selected during the training process. The sample with the largest value is used to optimize the adjacency matrix as the final optimization result.

[0112] After obtaining the final optimization results, a performance evaluation is performed. Based on the final optimization results, natural connectivity and constellation network capacity are calculated, as well as performance on other performance evaluation metrics, including the shortest link hop count. Overall efficiency Algebraic connectivity Average chain establishment time Average number of handovers These metrics are used to evaluate the performance of the algorithm.

[0113] Among them, the shortest link hop count Indicates from satellite node Reaching the node Minimum number of links required. Global efficiency. The formula for calculation is: In the formula, Indicates overall efficiency; Indicates the total number of satellite nodes; Indicates from satellite node Reaching the node The minimum number of links required.

[0114] Algebraic connectivity The formula for calculation is: In the formula, Represents algebraic connectivity; Represents finding a matrix The second smallest eigenvalue; Indicated by As a diagonal matrix with the main diagonal, where... Representing satellite nodes Satellite nodes and satellite nodes The corresponding degree indicates the number of links established by the satellite node; This represents the final optimization result.

[0115] satellite nodes With satellite nodes Chain establishment time The number of switches is represented by the number of 1s in the adjacency sequence. Represented by the number of times the adjacency sequence changes from 0 to 1. For example, satellite nodes. With satellite nodes The adjacency sequence { If the chain is {0,1,1,0,1,0}, then the chain establishment time is 3. The number of switching times is 2.

[0116] Traversal By obtaining the link establishment time and number of switches between all satellite nodes, the final optimization result can be obtained. Average chain establishment time Average number of handovers .

[0117] Specifically, the average chain establishment time The formula for calculation is: In the formula, Indicates satellite node With satellite nodes Chain establishment time; Indicates satellite node With satellite nodes Number of switching times; This indicates the total number of satellite nodes.

[0118] Average number of handovers The formula for calculation is: In the formula, Indicates satellite node With satellite nodes Number of switching times; This indicates the total number of satellite nodes.

[0119] Step S107: According to the target link switching strategy, update the preliminary optimized adjacency matrix, perform performance evaluation according to the updated adjacency matrix, and adjust the inter-satellite links between satellite nodes in the multi-layer satellite constellation.

[0120] The performance evaluation includes natural connectivity and constellation network capacity, as well as performance on other performance metrics, including the shortest link hop count. Overall efficiency Algebraic connectivity Average chain establishment time Average number of handovers These metrics are used to evaluate performance.

[0121] In summary, the inter-satellite link dynamic control method for multi-layer satellite constellations provided in this embodiment obtains the inter-satellite distance threshold matrix and the inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation. The inter-satellite distance threshold matrix is ​​used to characterize whether the distance between each pair of satellite nodes is less than or equal to the inter-satellite distance threshold; the inter-satellite link adjacency matrix is ​​used to characterize the inter-satellite link connection relationship between each satellite node. The inter-satellite distance threshold matrix is ​​optimized using a preset optimization algorithm to generate a preliminary optimized adjacency matrix that satisfies the topological constraints of the multi-layer satellite constellation's inter-satellite link network. Based on the satellite node network topology features indicated by the inter-satellite link adjacency matrix and the spatial location features of each satellite node, a node feature matrix is ​​generated. The node feature matrix is ​​input into a pre-trained graph convolutional neural network to obtain a network feature vector characterizing the network topology performance of the multi-layer satellite constellation. The process involves matching the initially optimized adjacency matrix with the inter-satellite distance threshold matrix to select inter-satellite link pairs that meet preset link switching conditions, and generating a candidate set of link switching strategies based on these pairs. Network feature vectors are input as state parameters to a pre-trained inter-satellite link control model, and the candidate set of link switching strategies is used as the action space. The inter-satellite link control model outputs a probability distribution based on the state parameters and samples the target link switching strategy from the action space based on this probability distribution. The initially optimized adjacency matrix is ​​updated according to the target link switching strategy, and performance is evaluated based on the updated adjacency matrix. Inter-satellite links between satellite nodes in a multi-layered satellite constellation are then adjusted. This approach addresses the shortcomings of existing satellite topology control methods in terms of control performance and reliability in large-scale, highly dynamic scenarios, making it difficult to meet the practical needs of modern satellite network topology optimization. By generating an initial adjacency matrix that satisfies basic topology constraints through a preset optimization algorithm, the complexity of problem solving is significantly reduced, laying an efficient foundation for subsequent fine-grained optimization and effectively improving control performance. Building upon this foundation, a graph convolutional neural network (GNN) is used to process the node feature matrix containing satellite spatial positions and topological relationships. This enables the GNN to perceive actual physical constraints such as inter-satellite distances and inter-layer geometric relationships, avoiding biases caused by idealized assumptions and enhancing the practicality and reliability of the strategy. The GNN efficiently learns global network features, compressing key topological performance information into low-dimensional feature vectors, avoiding traditional high-dimensional matrix operations and further improving computational performance in large-scale scenarios. Furthermore, the network feature vectors are used as state inputs to the inter-satellite link control model, and the link exchange candidate set is used as the action space to generate dynamic strategies through probabilistic sampling. This strategy rigorously filters physical feasibility based on a distance threshold matrix and achieves adaptive link adjustment based on real-time topological performance through a "state-action-evaluation" loop, thereby maintaining high system reliability in complex dynamic environments.

[0122] Corresponding to the above method, the present invention also provides a dynamic control device for inter-satellite links of a multi-layer satellite constellation. The device includes a computer device, which includes a processor and a memory. The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0123] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned multi-layer satellite constellation inter-satellite link dynamic control method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0124] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0125] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0126] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0127] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic control method for inter-satellite links in a multi-layer satellite constellation, characterized in that, The method includes the following steps: Obtain the inter-satellite distance threshold matrix and inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation; wherein, the inter-satellite distance threshold matrix is ​​used to characterize whether the distance between each pair of satellite nodes is less than or equal to the inter-satellite distance threshold; the inter-satellite link adjacency matrix is ​​used to characterize the inter-satellite link connection relationship between each satellite node; The inter-satellite distance threshold matrix is ​​optimized using a preset optimization algorithm to generate a preliminary optimized adjacency matrix that satisfies the topological constraints of the inter-satellite link network of a multi-layer satellite constellation. Based on the satellite node network topology characteristics indicated by the inter-satellite link adjacency matrix and the spatial location characteristics of each satellite node, a node feature matrix is ​​generated; The node feature matrix is ​​input into a pre-trained graph convolutional neural network to obtain a network feature vector that characterizes the topological performance of the multi-layer satellite constellation network. The preliminary optimized adjacency matrix is ​​matched with the inter-satellite distance threshold matrix to filter out inter-satellite link pairs that meet the preset link switching conditions, and a candidate set of link switching strategies is generated based on the inter-satellite link pairs; The network feature vectors are used as state parameters to input into a pre-trained inter-satellite link control model, and the candidate set of link switching strategies is used as the action space. The inter-satellite link control model outputs a probability distribution based on the state parameters, and samples from the action space based on the probability distribution to generate a target link switching strategy. The inter-satellite link control model is trained based on reinforcement learning and aims to maximize the natural connectivity and constellation network capacity of the multi-layer satellite constellation by constructing a target optimization function. According to the target link switching strategy, the preliminary optimized adjacency matrix is ​​updated, the performance is evaluated according to the updated adjacency matrix, and the inter-satellite links between satellite nodes in the multi-layer satellite constellation are adjusted.

2. The method according to claim 1, characterized in that, The inter-satellite link control model is trained through the following steps: A main neural network for obtaining the actor critic network structure; the main neural network includes an initial policy network and an initial value network; Obtain the sample inter-satellite distance threshold matrix and the sample inter-satellite link adjacency matrix; The sample inter-satellite distance threshold matrix is ​​optimized using the preset optimization algorithm to generate an initial sample optimized adjacency matrix that satisfies the link network topology constraints. The node feature matrix corresponding to the sample optimized neighbor matrix is ​​input into the graph convolutional neural network, and the output sample network feature vector is used as the sample state parameter of the policy network. The sample optimized adjacency matrix is ​​matched with the sample inter-satellite distance threshold matrix to filter out sample inter-satellite link pairs that meet the preset link exchange conditions, and a sample link exchange policy candidate set is generated based on the sample inter-satellite link pairs as the sample action space of the policy network. The policy network outputs a sample probability distribution based on the sample state parameters, and generates a sample target link switching policy by sampling from the sample action space based on the sample probability distribution. The value network generates the value corresponding to the sample state parameters based on the sample state parameters; Based on the target link exchange strategy of the samples and the optimized adjacency matrix of the samples, a candidate adjacency matrix is ​​generated; If the candidate adjacency matrix satisfies the link network topology constraints, the adjacency matrix and the candidate adjacency matrix are optimized based on the samples, and the reward value is calculated by combining the preset reward function. The advantage value and target value are calculated using the reward value. The strategy network is updated by limiting the range of variation in the probability ratio between the new and old strategies, based on the advantage value. The value network is updated by minimizing the value prediction error, based on the target value. The candidate adjacency matrix is ​​used as the new sample optimized adjacency matrix, and the candidate state parameters are used as the new sample state parameters. The step of matching the sample optimized adjacency matrix with the sample inter-satellite distance threshold matrix is ​​performed. Multiple iterations are carried out until a preset termination condition is met. The policy network obtained from multiple iterations is used as the inter-satellite link control model.

3. The method according to claim 2, characterized in that, The reward value is calculated based on the optimized adjacency matrix and the candidate adjacency matrix, combined with a preset reward function; wherein, the preset reward function is calculated as follows: In the formula, Indicates training steps The corresponding reward value; Represents the candidate adjacency matrix; This represents the optimized adjacency matrix of the samples; Let the objective optimization function be the objective optimization function. The formula for calculation is: In the formula, Indicates in Time-based training steps Optimize the adjacency matrix of the samples below; Indicates the normalization parameter; Indicates the normalization parameter; This indicates that the multi-layered satellite constellation optimizes the adjacency matrix in the sample. The natural connectivity is calculated as follows: In the formula, Represents the natural logarithm function; Represents the sample inter-satellite link adjacency matrix The number of rows and columns is also the number of satellite nodes in the multi-layer satellite constellation; The formula for solving the trace of the sample-optimized adjacency matrix is ​​used to represent the eigenvalues ​​corresponding to the sample-optimized adjacency matrix through the sample-optimized adjacency matrix. This indicates that a Taylor expansion is performed on the optimized adjacency matrix of the samples, where, Indicates the index of the series form after expansion; , representing the constellation network capacity of the multi-layer satellite constellation under the sample-optimized adjacency matrix, is calculated as follows: In the formula, Indicates channel condition parameters; Indicates the baseline link capacity; Indicates link distance; This represents the summation function; This represents a distance matrix used to indicate the distances between satellite nodes.

4. The method according to claim 2, characterized in that, The advantage value and target value are calculated using the reward value; wherein, the formula for calculating the advantage value is: In the formula, Indicates training steps The corresponding advantage value; Indicates training steps The corresponding reward value; Indicates the discount factor; This represents the value network; This represents the state parameters generated based on the candidate adjacency matrix; This represents the sample state parameters; Indicates training steps The corresponding judgment value, after completing all training steps in one iteration of training, The value is 1 if it is 1, otherwise it is 0. This indicates that the advantage value has been normalized, where, This represents the mean of the dominance values; The standard deviation of the said dominance value; The formula for calculating the target value is: In the formula, Indicates training steps Corresponding target value; Indicates the discount factor; This represents the value network; This represents the state parameters generated based on the candidate adjacency matrix; Indicates training steps The corresponding judgment value.

5. The method according to claim 4, characterized in that, The policy network is updated by combining the advantage value with a limit on the range of variation in the probability ratio between the new and old policies. The loss function is calculated as follows: In the formula, Indicates the target link switching strategy in the sample The loss function corresponding to the policy network described below; Represents the expected value function; Describes the minimum value function; Indicating in the training steps The sample target link switching strategy The corresponding probability ratio; Indicates training steps The corresponding advantage value; This represents a function that restricts the range of values. Limited to to between; Indicates the clipping factor; The value network is updated by minimizing the value prediction error based on the target value. The loss function is calculated as follows: In the formula, The target link switching strategy in the sample The loss function corresponding to the value network described below; Represents the expected value function; Indicates training steps Corresponding target value; This represents the value network; This represents the sample state parameters.

6. The method according to claim 2, characterized in that, The graph convolutional neural network includes L graph convolutional layers; wherein L is an integer greater than or equal to 3; the value of L is positively correlated with the number of satellite nodes in the multi-layer satellite constellation; The policy network comprises an input layer, L-2 hidden layers, and an output layer; The value network consists of an input layer, L-2 hidden layers, and an output layer.

7. The method according to claim 1, characterized in that, The inter-satellite distance threshold matrix and the inter-satellite link adjacency matrix are symmetric matrices; obtaining the inter-satellite distance threshold matrix and the inter-satellite link adjacency matrix corresponding to the multi-layer satellite constellation includes: Based on the latitude, longitude, and altitude information of each satellite node in the multi-layer satellite constellation, the spatial position coordinates of each satellite node in the geocentric-geofixed coordinate system are obtained. Based on the spatial coordinates of each satellite node, the inter-satellite distance between satellite nodes is calculated and compared with the inter-satellite distance threshold. Based on the comparison results, the inter-satellite distance threshold matrix is ​​constructed. Based on the inter-satellite link connection relationships between each satellite node, the inter-satellite link adjacency matrix is ​​constructed.

8. The method according to claim 1, characterized in that, Each inter-satellite link pair consists of two independent inter-satellite links; The step of matching the preliminary optimized adjacency matrix with the inter-satellite distance threshold matrix to filter out inter-satellite link pairs that meet the preset link switching conditions, and generating a candidate set of link switching strategies based on the inter-satellite link pairs, includes: Based on the preliminary optimized adjacency matrix, inter-satellite link relationships are extracted to obtain an inter-satellite link set; Based on the inter-satellite distance threshold matrix, link relationships that satisfy the inter-satellite distance thresholds are extracted to obtain a set of valid links; The inter-satellite link set is matched with the valid link set to filter out independent link pairs that belong to both sets simultaneously. For each link pair, if the two new links obtained after the link exchange belong to the set of valid links but not to the set of inter-satellite links, it is determined that the preset link exchange condition is met, and the cosine similarity value between the two new links is calculated. If the cosine similarity value is equal to the preset similarity threshold, the link exchange is used as a link exchange strategy and added to the candidate set of link exchange strategies.

9. A dynamic control device for inter-satellite links in a multi-layer satellite constellation, comprising a processor, a memory, and computer programs / instructions stored in the memory, characterized in that, The processor is configured to execute the computer program / instructions, and when the computer program / instructions are executed, the device implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 8.

Images