Large-scale low-orbit satellite network domain division intelligent routing method

Abstract

The invention discloses a large-scale low-orbit satellite network domain-division intelligent routing method, which belongs to the technical field of satellite communication, and comprises the following steps: constructing a network model based on multi-dimensional resource characteristics, and providing resource constraint conditions and space-time correlation characteristics for routing decision by combining with a regional flow prediction algorithm of an improved Transform architecture; virtual nodes and an autonomous domain are divided based on a geographic area, and a low-orbit satellite network routing process is decoupled into an intra-domain part and an inter-domain part; a multi-agent deep Q network is adopted in a domain, a sum tree mechanism-based deep Q network is adopted between domains, a routing algorithm and a resource scheduling algorithm are respectively designed, and a domain intelligent routing system with flexibility and expandability is constructed through hierarchical routing and resource decoupling scheduling. According to the method, the service capability of the low earth orbit satellite network in extreme scenes such as topology high-frequency change, strict resource limitation and traffic space-time mutation is remarkably enhanced, and a high-reliability and low-delay routing solution is provided for space-ground integrated communication, emergency disaster early warning and global real-time data transmission.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and in particular to a domain-specific intelligent routing method for a large-scale low-orbit satellite network. Background Art

[0002] With the rapid development of low-orbit satellite network technology, it has shown great potential in integrated space-ground communications, global coverage, and real-time data transmission. However, low-orbit satellite networks face technical challenges such as highly dynamic topology, strictly limited onboard resources, and uneven multi-dimensional spatiotemporal traffic distribution. Due to the low orbital altitude of low-orbit satellites, their network topology changes frequently and dynamically with satellite operation, making it difficult for traditional routing mechanisms to adapt to the needs of rapid topology reconstruction. At the same time, low-orbit satellite networks face multi-dimensional resource constraints such as limited onboard computing power, limited storage resources, and dynamically changing energy supply, which makes the routing optimization problem more complex. The traffic distribution of low-orbit satellite networks has significant spatiotemporal imbalances. The geographical distribution differences of ground user needs, the dynamic coverage of satellite constellations, and the time-varying characteristics of inter-satellite links result in multi-scale spatiotemporal correlations and nonlinear evolution of network traffic, making it difficult for traditional traffic prediction methods to achieve high-precision predictions.

[0003] In practical applications, routing optimization technology for low-orbit satellite networks faces the following key issues: Traditional routing algorithms are primarily designed based on static network models, making it difficult to effectively cope with the high-frequency topological changes and dynamic resource constraints of low-orbit satellite networks, resulting in slow routing convergence and low resource utilization. Existing traffic prediction methods are mostly based on static spatiotemporal models of independent data sources, which make it difficult to effectively integrate the spatiotemporal correlations of multi-dimensional data, resulting in insufficient filtering of redundant information, deviations in the extraction of key features, and difficulty adapting to the real-time analysis requirements of dynamic constellation reconstruction and environmental mutations. Traditional routing optimization methods ignore the strong coupling characteristics of multi-dimensional resources in satellite networks. Cross-domain coupling effects such as the dynamic correlation between orbital configuration and energy load, and the chain feedback between communication energy consumption and battery life make it difficult to achieve global optimization in routing decisions.

[0004] To address the above issues, the present invention proposes a domain-specific intelligent routing method for large-scale low-orbit satellite networks, which has the following innovations and advantages:

[0005] Based on the extraction and characterization of multi-dimensional resource features, low-orbit satellite network simulation modeling is realized, the influence weight of multi-factor coupling on network performance is quantified, and resource constraints and spatiotemporal correlation characteristics are provided for routing algorithms, including space environment model, energy consumption model, illumination model, on-orbit life model, and multi-dimensional resource coupling characteristics and dynamic interaction model; the space environment model designs constellation configuration, inter-satellite link and communication link model based on orbital parameters such as altitude, inclination, number of orbital planes, number of satellites in each orbit and phase factor of adjacent orbital planes; the energy consumption model quantifies the satellite antenna receiving power, transmitting power and Data transmission energy consumption, establishing the mathematical relationship between satellite communication energy consumption and transmission distance, data volume and channel status; the illumination model predicts the efficiency of solar panel power generation and the impact of the earth's shadow period on satellite power supply through real-time interaction between the satellite's orbital position and the earth's shadow area; the on-orbit life model constructs the relationship between battery reliability and lost life according to the battery cycle life consumption of the lithium-ion battery discharge depth; the dynamic interaction model of multi-dimensional resource coupling characteristics constructs the coupling mechanism and dynamic correlation of multi-dimensional resources by analyzing the model representations of space environment, energy consumption, illumination and on-orbit life, and carries out collaborative modeling and system optimization.

[0006] Based on an improved Transformer architecture, a regional traffic prediction algorithm for low-orbit satellite networks is implemented. A global traffic tidal effect model is constructed, and the prediction results are mapped into a routing decision state space to drive resource allocation and congestion avoidance strategies. Specifically, this includes data service traffic modeling and a low-orbit satellite network traffic prediction algorithm based on the Patch Time Series Transformer (PatchTST). The Earth's surface is divided into 288 traffic blocks at 15° intervals of longitude and latitude. Traffic intensity is converted based on real statistical data to complete the low-orbit satellite network traffic representation for each traffic block. The original traffic sequence is decomposed into multiple single-frequency and independent subsequences using empirical mode decomposition (EMD) technology. The PatchTST model, combined with the Transformer architecture's advantage in efficiently processing long sequences, is used to predict each subsequence, capturing the complex patterns and long-term dependencies in the time series, ultimately improving prediction accuracy and computational efficiency.

[0007] A large-scale low-orbit satellite network model is constructed based on the division of virtual nodes and autonomous domains into geographical areas. Specifically, it includes a hexagonal grid division method for geographical areas and a low-orbit satellite network routing method based on the fusion of virtual nodes and autonomous domains. The earth's surface is divided into equivalent coverage by hexagonal grids, and each grid cell is defined as a virtual node and an autonomous domain. The hexagonal grid is mapped to the constellation through the spatial association between the satellite's ground projection position and the earth's surface. Satellite groups in the same block will belong to the same autonomous domain and be aggregated into an abstract virtual node. As a result, the routing process is decoupled into two parts: intra-domain and inter-domain.

[0008] Based on an improved Multi-Agent Deep Q-Network (MADQN), a routing selection and resource scheduling algorithm within the low-orbit satellite network domain is implemented, specifically including a real-time Markov decision process (RTMDP) and a self-attention mechanism. The routing selection and resource scheduling problem within the autonomous domain is modeled as a real-time Markov decision process. Each satellite acts as an agent, and its input includes multi-dimensional resource feature information of neighboring nodes and the previous action information. The output action is the transmission direction of the next hop. The self-attention mechanism is combined with learning the correlation matrix information of the satellite network state and assigning dynamic weights to the key matrix, thereby enhancing the multi-agent's understanding and response capabilities to the environmental state.

[0009] Based on the Deep Q-Network with SumTree (DQN-ST) mechanism, an inter-domain routing selection and resource scheduling algorithm for low-orbit satellite networks is implemented. Specifically, this algorithm includes a dynamically adaptive domain boundary node mapping mechanism and a DQN-ST algorithm architecture optimized for routing tasks. Domain boundary nodes and topological connection states are updated in real time through weighted scores of node effectiveness and resource availability, establishing a dynamic mapping relationship for cross-domain routing. A SumTree-modified prioritized experience replay mechanism is used to determine the priority of satellite routing task samples based on the absolute value of the temporal difference error. Priority samples are sampled and reinforced learning is performed on them. The optimal next-hop satellite node is selected based on the Q value, and after aggregation, intelligent inter-domain routing decisions are output.

[0010] Through these innovations, the present invention can solve problems such as insufficient multi-dimensional resource coupling modeling, limited spatiotemporal traffic prediction accuracy, and difficulty in global optimization of cross-domain routing in low-orbit satellite networks. It can realize routing decision-making and resource scheduling driven by dynamic resource representation-spatiotemporal collaborative optimization, significantly enhance the service capabilities of low-orbit satellite networks in extreme scenarios such as high-frequency topology changes, strict resource constraints, and spatiotemporal traffic mutations, and provide highly reliable, low-latency constellation-level routing solutions for integrated space-ground communications, emergency disaster warning, and global real-time data transmission. Summary of the Invention

[0011] In view of this, the present invention provides a large-scale low-orbit satellite network domain-specific intelligent routing method to eliminate or improve the defects existing in the prior art, such as insufficient multi-dimensional resource coupling modeling in low-orbit satellite networks, limited spatiotemporal traffic prediction accuracy, and difficulty in global optimization of cross-domain routing. It realizes routing decision-making and resource scheduling driven by dynamic resource representation-spatiotemporal collaborative optimization, thereby enhancing the service capabilities of low-orbit satellite networks in scenarios such as high-frequency topology changes, strict resource constraints, and spatiotemporal traffic mutations.

[0012] This paper provides a low-orbit satellite network simulation modeling method based on multi-dimensional resource feature extraction and characterization. By defining specific parameters for each orbital layer, designing the connection logic between satellites, and considering factors such as the space environment, illumination, and satellite lifespan, a systematic coupling modeling framework is proposed. This provides theoretical and technical support for building efficient and stable space-based Internet services. Specific technical features include the following:

[0013] The space environment model is designed based on orbital parameters such as altitude, inclination, number of orbital planes, number of satellites in each orbit, and phase factor between adjacent orbital planes to design constellation configuration, intersatellite link, and communication link models. The neighbor satellite connection relationship of a satellite logical address (m, n) is defined by the following formula:

[0014]

[0015] The energy consumption model quantifies the satellite antenna receiving power, transmitting power and data transmission energy consumption, establishes the mathematical relationship between satellite communication energy consumption and transmission distance, data volume and channel status; and derives the inter-satellite transmission rate through the Shannon formula. Among them, w (i，j) is the bandwidth of the wireless channel, σ 2 is the noise power; free space propagation loss during transmission The expression is Among them, d (i，j) is the distance between satellites, λ is the wavelength of electromagnetic wave, c is the speed of light, and f is the operating frequency; the antenna receiving power is When the receiving power of a satellite is given, the transmitting power of the satellite can be deduced. The expression is

[0016] The illumination model predicts the efficiency of solar panels and the impact of the earth's shadow period on satellite power supply through real-time interaction between the satellite's orbital position and the earth's shadow area; the cylindrical earth shadow model is used to predict the earth's shadow period, with the following constraints: If this formula is not satisfied, it means that the satellite is in the illuminated area;

[0017] The on-orbit life model is based on the relationship between the battery reliability and the lost life according to the lithium-ion battery discharge depth and the battery cycle life consumption. t1,t2 The functional relationship is The fitting expression of battery reliability R(t) is:

[0018] The dynamic interaction model of multi-dimensional resource coupling characteristics breaks through the limitations of traditional models that independently model single-dimensional resources by revealing the constraints of orbital configuration on communication link distance, the cumulative impact of energy mode switching caused by earth shadow events on battery life, and the interactive mechanism between dynamic tracking strategy and energy consumption distribution. It constructs the coupling mechanism and dynamic association of multi-dimensional resources, and carries out collaborative modeling and system optimization. It quantifies the joint impact of orbital altitude on communication distance and energy load through Monte Carlo simulation, providing data and model basis for subsequent research on low-orbit satellite network traffic prediction and routing technology.

[0019] This paper provides a regional traffic prediction algorithm for low-orbit satellite networks based on an improved Transformer architecture. This algorithm divides the Earth's surface into 288 traffic blocks according to longitude and latitude, and uses EMD technology to decompose the complex original traffic sequence into multiple single-frequency, independent IMF subsequences, effectively removing noise and simplifying the model training process. Subsequently, the PatchTST model is used to predict each IMF subsequence. By combining the advantages of the Transformer architecture, the algorithm captures the complex patterns and long-term dependencies in the time series, ultimately improving prediction accuracy and computational efficiency. The algorithm specifically includes the following technical features:

[0020] The data traffic modeling method divides the earth's surface into regions every 15° according to longitude and latitude, and converts the traffic intensity according to the number of Internet users in each region to obtain the global data traffic intensity L. n , the temporal distribution of traffic is represented by the discount coefficient γ; in each traffic block, the longitude of its center position is selected as a reference to infer the time zone to which the block belongs, assuming that the discount coefficient is 1 when the user is most active and 0 when there is almost no activity; the low-orbit satellite network traffic of any traffic block is represented as Cn =αγL n , n=1,2,…,288, unit is Mbps;

[0021] The PatchTST-based low-orbit satellite network traffic prediction algorithm uses EMD technology to decompose the original traffic sequence into multiple single-frequency, independent IMF subsequences and a residual signal, effectively separating different frequency components, making each subsequence more stable and easier to model. The PatchTST model is introduced for prediction of each IMF subsequence and residual signal. The PatchTST model combines the advantages of the Transformer architecture, can maintain high efficiency when processing long sequences, and can capture complex patterns and dependencies in time series. The output results are summed as the prediction output.

[0022] The implementation method of EMD mentioned above assumes that the original sequence is X(t), which can be decomposed into a finite number of IMF subsequences c j and a remainder term r n , The formula for calculating its standard deviation is: Where h i-1，j 、h i，j represents the intrinsic mode function, i represents the number of iterations, j represents the order of the function, and N represents the total number of iterations;

[0023] The method for constructing the traffic prediction model based on PatchTST is as follows: for a set of multivariate time series samples, assuming that the length of the look-back window, i.e., the time data of the input model, is L, the samples (x1, ..., x t ,...,x L ), where, at time slot t, x t It is a one-dimensional column vector containing M elements, predicting the data value (x L+1 ,...,x L+T ); the model structure includes the forward process, the block process and the Transformer encoder;

[0024] The forward process described above represents the i-th one-dimensional sequence as Each sequence is independently input into the Transformer backbone network, and the prediction result of the sequence is calculated as

[0025] In the block process, each input one-dimensional time series x (i) The block is processed, the block length is recorded as P, and the non-overlapping area between two consecutive blocks is recorded as S. After the block process, a block sequence will be generated Among them, N represents the number of blocks generated, The memory usage and computational complexity of the attention map are expressed in S 2 Therefore, under the conditions of limited training time and GPU memory resources, block operation can allow the model to understand longer historical sequences, thereby significantly improving the model's prediction performance;

[0026] The Transformer encoder is trained through a linear projection W P ∈R D×P , and a learnable additive positional encoding W pos ∈R D×N , mapping the blocks to a potential representation space of dimension D, the data input to the Transformer encoder is represented as in Each attention head h=1,...,H of the multi-head attention transforms it into a query matrix Bond Matrix and the value matrix in Use scaled product to get attention output After the attention output is processed by the normalization layer and the feedforward network layer with residual connection, the generated result is represented as z (i) ∈R D×N , using a fully connected layer with a linear head to obtain the prediction results

[0027] The present invention provides a large-scale low-orbit satellite network model that divides virtual nodes and autonomous domains based on geographic regions. By performing a simple cylindrical projection on the Earth's surface and dividing it into 112 regular hexagonal grid cells, each grid is defined as a virtual node and autonomous domain, thereby shielding the impact of high-speed satellite motion on topological dynamics. Subsequently, the surface grid is associated with the constellation through satellite sub-satellite point position mapping, and satellite clusters within the same block are aggregated into virtual nodes. Global routing is decoupled into real-time intra-domain decision-making and inter-domain scheduling, significantly reducing the size of routing tables and computational complexity. Finally, a hybrid strategy of flooding updates and local updates is combined to improve routing convergence speed and reduce onboard storage resource requirements in scenarios with high-frequency topology reconstruction. Specifically, the model includes the following technical features:

[0028] The hexagonal gridding method for geographic regions uses a simple cylindrical projection of the Earth's surface to transform it into a rectangle with an aspect ratio of 2:1. This rectangle is then divided into 112 regular hexagonal grids, each covering a 15° x 15° latitude and longitude area, representing a virtual node and autonomous system (AS). The grid mapping rules ensure seamless connection between adjacent virtual nodes, reducing routing hops.

[0029] The low-orbit satellite network routing method based on the fusion of virtual nodes and autonomous domains maps the hexagonal grid structure of the earth's surface onto the satellite constellation through the spatial association between the satellite's ground projection position and the surface grid. Satellite clusters in the same block will belong to the same autonomous domain and be aggregated into an abstract virtual node. A routing and resource information update strategy combining flooding updates and local updates is adopted. Inter-domain information is updated through flooding, and the information of satellite nodes at the boundary of the block is flooded to the entire satellite constellation. Intra-domain information is updated through local updates, and information such as the status of each satellite and link is shared and updated within the autonomous domain.

[0030] The present invention provides a routing and resource scheduling algorithm for low-orbit satellite networks based on an improved MADQN. By modeling the problem as a real-time Markov decision process, the algorithm addresses the real-time limitations of traditional MDP frameworks in dynamic environments. Subsequently, the MADQN algorithm is adopted, with each satellite acting as an intelligent agent. The input includes satellite status information and link status information from neighboring nodes, and the output action is one of four possible transmission directions. To more effectively process complex and dynamically changing state information, a self-attention mechanism is introduced to learn the relevant key matrices of satellite network status and dynamically assign weights, thereby enhancing the understanding and response capabilities of the environmental state. The algorithm specifically includes the following technical features:

[0031] The real-time Markov decision process, the state space s t Including satellite energy consumption, link quality, battery reliability and neighbor node status and the satellite node's last action decision result, the action space a t For the transmission direction of “up, down, left, right”, the reward function r(s t , a t ) including comprehensive latency, energy consumption, and link stability; in the RTMDP framework, actions and states change simultaneously. The agent can immediately observe the next state after selecting an action and directly infer the next action. This real-time decision-making mechanism allows the agent to more closely track rapidly changing environmental conditions, ensuring that routing and resource scheduling can respond promptly to changes in the topology and state of the low-orbit satellite network;

[0032] The action value function expression is as follows:

[0033]

[0034] Where, Indicates that in a given state s t and action a t When the transition to a specific state s t+1 The expected probability of . Indicates that in a given state action pair (s t+1 ,a t), execute action a according to the current strategy π t+1 The expected cumulative reward for the state sequence that the agent may encounter after ;

[0035] The state value function expression is as follows:

[0036]

[0037] Where, Indicates that in a given state action pair (s t ,a t ) under the condition of the expected cumulative reward for the state sequence that the agent may encounter;

[0038] The self-attention mechanism maps the input neighbor node state information into query matrices by constructing a multi-head attention structure. Bond Matrix and value matrix The attention weight calculation follows the formula where d k is the dimensionality scaling factor of the key vector, which is used to avoid the gradient saturation problem of the Softmax function. By computing the outputs of multiple attention heads in parallel and concatenating them into a joint representation, the algorithm can dynamically allocate weights, highlight the states of neighboring nodes with high link stability and large resource redundancy, and suppress noise interference. The attention output is input into the fully connected network after residual connection and layer normalization to generate routing decision actions.

[0039] This paper provides a low-orbit satellite network inter-domain routing and resource scheduling algorithm based on DQN-ST. By designing a dynamic and adaptive domain boundary node mapping mechanism, it comprehensively considers node effectiveness and resource availability to solve the problem of boundary node failure caused by high-speed satellite motion. Subsequently, it proposes a deep Q network with priority experience replay based on Sumtree, which realizes global collaborative optimization of cross-domain routing and resource scheduling through task-aware state space modeling and dynamic bandwidth allocation mechanism. The present invention specifically includes the following technical features:

[0040] In the dynamic and adaptive domain boundary node mapping mechanism, when a node needs to communicate across autonomous domains, the node within the domain will forward the request information to the domain boundary node. Due to the continuous high-speed movement of the satellite, the satellite node will continue to move from one area to the next. At this time, the originally selected domain boundary node and the inter-domain communication link will become invalid. Considering the node effectiveness and node available resources and other indicators, the satellite nodes within the domain are evaluated and the mapping satellite of the domain boundary node is selected. The calculation of the node effectiveness index not only considers the geometric distance to the adjacent domain at the current moment, but also needs to be combined with the time sensitivity of the specific routing task. The evaluation of the node resource index needs to be dynamically weighted according to the actual business needs, and the node effectiveness E(s) is defined.i )=dis(s i )s i ∈{s1,s2,...,s n}, node available resources R(s i )=Cal(s i )+Cache(s i )s i ∈{s1,s2,...,s n}, the weight coefficients of the two indicators are set to 0.7 and 0.3 respectively, then the standardized weighted comprehensive score Mapping to actual routing decisions, the system can automatically select edge nodes that can meet the business quality of service (QoS) requirements and have optimal resource utilization;

[0041] The DQN-ST algorithm architecture for routing task optimization applies the priority experience replay mechanism based on Sumtree to the deep Q network algorithm, and the state space s t ={l(t), u(t), τ(t), γ(t)}, where l(t) represents the satellite node where the task of time slot t is located, u(t) represents the link connection status of the satellite node where the task of time slot t is located, τ(t) represents the data volume of the task of time slot t, and γ(t) represents the network status information of the virtual constellation of time slot t; the action space a t ={h(t),B(t)}, where h(t) represents the satellite node where the agent performs the next forwarding in time slot t; B(t) represents the bandwidth resources allocated to the task in time slot t; the priority experience replay mechanism based on Sumtree considers the spatiotemporal correlation of routing tasks in actual deployment, and gives higher TD error weights to task samples with high delay sensitivity or high bandwidth requirements. i =|δ(i)|+∈, where ∈ is a very small constant, and the sampling probability P(i) is calculated according to the formula Calculation; Based on the above calculation and the Q value calculation results, the optimal next-hop satellite node h(t) is selected to ensure the efficiency and stability of cross-domain communication.

[0042] The beneficial effects of the present invention are:

[0043] The present invention provides a domain-specific intelligent routing method for large-scale low-orbit satellite networks, belonging to the field of satellite communication technology. This method is based on low-orbit satellite network simulation modeling with multi-dimensional resource feature extraction and characterization, providing resource constraints and spatiotemporal correlation features for subsequent routing algorithms. A regional traffic prediction algorithm for low-orbit satellite networks based on an improved Transformer architecture constructs a global traffic tidal effect model to drive routing decisions and resource scheduling. A large-scale low-orbit satellite network model is constructed based on the division of virtual nodes and autonomous domains by geographical regions. The autonomous domains are obtained by equivalently covering the earth's surface using a hexagonal network, decoupling the routing process into intra-domain and inter-domain parts. An intra-domain routing selection and resource scheduling algorithm for low-orbit satellite networks based on an improved MADQN implements an efficient real-time decision-making mechanism, ensuring that routing selection and resource scheduling can promptly respond to changes in network topology and status. A DQN-ST-based inter-domain routing selection and resource scheduling algorithm for low-orbit satellite networks addresses the issue of boundary node failure caused by high-speed satellite motion, achieving global collaborative optimization of cross-domain routing and resource scheduling.

[0044] Additional advantages, objects, and features of the present invention will be set forth in part in the following description and will become apparent to those skilled in the art upon examination of the following or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained by the structures particularly pointed out in the description and drawings.

[0045] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the above specific descriptions, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention. In the drawings:

[0047] Figure 1 The figure is a schematic diagram of the steps of a large-scale low-orbit satellite network domain-based intelligent routing method according to one embodiment of the present invention.

[0048] Figure 2 The figure is a flow chart of a large-scale low-orbit satellite network domain-specific intelligent routing method according to one embodiment of the present invention.

[0049] Figure 3 Schematic diagram of a low-orbit satellite network simulation modeling method and a multi-dimensional resource coupling model in one embodiment of the present invention.

[0050] Figure 4 The figure is a flow chart of a method for predicting low-orbit satellite network traffic in one embodiment of the present invention.

[0051] Figure 5 The figure is a flow chart of an empirical mode decomposition method according to an embodiment of the present invention.

[0052] Figure 6 This is a framework diagram of an intra-domain routing and resource scheduling algorithm based on MADQN in one embodiment of the present invention.

[0053] Figure 7 FIG1 is a diagram of an algorithm architecture for inter-domain routing optimization based on DQN-ST in one embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0055] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0056] It should be emphasized that the term "include / comprises" when used herein refers to the existence of features, elements, steps or components, but does not exclude the existence or addition of one or more other features, elements, steps or components.

[0057] It should also be noted that, unless otherwise specified, the term "connection" herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0058] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0059] In order to solve the problems of multi-factor coupling complexity and dynamic characteristics characterization difficulty of large-scale low-orbit satellite networks, the difficulty in predicting the time-space imbalance of low-orbit satellite network traffic, and the difficulty of low-orbit satellite network routing technology to cope with on-orbit changes and environmental impacts, the present invention provides a large-scale low-orbit satellite network domain intelligent routing method. The method is based on a low-orbit satellite network simulation modeling method for multi-dimensional resource feature extraction and characterization and a low-orbit satellite network regional traffic prediction algorithm based on an improved Transformer architecture, providing resource constraints and time-space correlation features for routing decisions, a large-scale low-orbit satellite network model based on geographical area division of virtual nodes and autonomous domains, decoupling the routing process into intra-domain and inter-domain parts, a low-orbit satellite network intra-domain routing selection and resource scheduling algorithm based on an improved MADQN and a low-orbit satellite network inter-domain routing selection and resource scheduling algorithm based on DQN-ST, realizing hierarchical routing and resource decoupling scheduling, breaking through the traditional single-layer routing framework, and building a domain intelligent routing system that takes into account both flexibility and scalability. Figure 1 As shown, the method includes the following steps S101 to S105:

[0060] Step S101: Construct a dynamic representation model of the multi-dimensional resources of the low-orbit satellite network to quantify resource constraints and spatiotemporal correlation characteristics. By receiving satellite constellation configuration parameters and real-time operating status, a space environment model, energy consumption model, illumination model, and on-orbit life model are established to reveal the coupling mechanism and dynamic interaction characteristics of multi-dimensional resources.

[0061] Step S102: Implement low-orbit satellite network traffic forecasting to support efficient and accurate resource management and scheduling decisions, avoiding network congestion and resource waste. The Earth's surface is divided into 288 longitude and latitude blocks. Empirical mode decomposition is used to separate the noise and core components of the traffic series. A block-based time series transformer is used to capture long-range spatiotemporal dependencies, and the forecast results are mapped into spatial states that can be parsed for routing decisions.

[0062] Step S103: Construct a virtual node topology to decouple the global routing optimization problem. A hexagonal grid is used to divide the Earth's surface into equivalent coverage areas. The constellation is mapped into 112 autonomous domains based on the sub-satellite locations of the satellites, forming a satellite cluster topology within the domain. A hybrid update strategy is designed to reduce onboard storage resource requirements.

[0063] Step S104: Execute real-time routing decisions and resource scheduling in a dynamic environment within the autonomous domain to optimize local network transmission efficiency and resource utilization. Through a multi-agent collaborative learning framework, each satellite is modeled as an independent decision-making unit. It inputs the link status, remaining energy, and historical action sequences of neighboring nodes and outputs the next-hop transmission direction. A self-attention mechanism is used to dynamically evaluate the weights of key links, and a multi-dimensional reward function drives policy gradient updates, achieving rapid convergence and load balancing of satellite nodes in scenarios with drastic topological changes.

[0064] Step S105: Global routing optimization and prioritized resource scheduling are implemented in cross-domain scenarios to ensure quality of service for high-value services. Satellites mapping domain boundary nodes are evaluated and selected by comprehensively considering metrics such as node availability and available resources. A prioritized experience replay mechanism based on a Sumtree structure provides global awareness of the network status of the virtual constellation composed of virtual nodes, enabling more efficient and accurate routing and resource scheduling.

[0065] In the present invention, the low-orbit satellite network parameters are composed of satellite constellation configuration parameters and real-time operating status data.

[0066] like Figure 2 The figure shows a flow chart of a domain-based intelligent routing method for a large-scale low-orbit satellite network.

[0067] In step S101, a multi-dimensional resource coupling characterization framework is established to reveal the influence of the interaction between track parameters, energy load, communication requirements and environmental constraints on the dynamic characteristics of the network.

[0068] Specifically, the public datasets for mainstream, large-scale, in-orbit constellations are incomplete, feature complex multi-factor coupling, and are difficult to simulate. This requires the construction of network structure and communication resource models, energy and environmental dynamic models, and multi-dimensional resource coupling and dynamic interaction models. This approach aims to define specific parameters for each orbital layer, design inter-satellite connectivity logic, and consider factors such as the space environment, illumination, and satellite lifespan. This framework, a systematic coupling modeling framework, provides theoretical and technical support for building efficient and stable space-based internet services.

[0069] In step S102, a traffic tidal change model is constructed to capture the complex patterns and dependencies in the time series, improve the accuracy and efficiency of traffic prediction, and support efficient and accurate low-orbit satellite network resource management and scheduling decisions.

[0070] Specifically, the routing algorithms and traffic control strategies of traditional satellite networks have serious lags and other problems. By dividing the Earth's surface into 288 traffic blocks according to longitude and latitude, and using EMD technology to decompose the complex original traffic sequence into multiple single-frequency and independent IMF sub-sequences, the noise can be effectively removed and the model training process is simplified. Subsequently, the PatchTST model is used to predict each IMF sub-sequence. Combined with the advantages of the Transformer architecture, long time series data can be efficiently processed, ultimately improving prediction accuracy and computational efficiency.

[0071] In step S103, autonomous domains and virtual nodes are divided, and the complex global routing problem is decomposed into two easier-to-handle levels: local and global. This makes routing selection and resource scheduling more efficient and flexible, and implements hierarchical routing within and between domains.

[0072] Specifically, the concept of virtual nodes is an effective solution for shielding against the effects of satellite motion. In traditional routing algorithms based on virtual nodes, each virtual node is assigned a fixed geographic location. During constellation operation, only one satellite is actively providing services to ground users at any given time, making it impossible to fully utilize satellite resources. By performing a simple cylindrical projection of the Earth's surface and dividing it into 112 regular hexagonal grid cells, each grid is defined as a virtual node and autonomous domain, thereby shielding against the effects of high-speed satellite motion on topological dynamics. Subsequently, the surface grid is associated with the constellation through satellite sub-satellite point position mapping, and satellite clusters within the same block are aggregated into virtual nodes. Global routing is decoupled into real-time intra-domain decision-making and inter-domain scheduling, significantly reducing the size of routing tables and computational complexity. Finally, a hybrid strategy of flooding updates and local updates is combined to improve routing convergence speed and reduce onboard storage resource requirements in scenarios with high-frequency topology reconstruction.

[0073] In step S104, MADQN is introduced for intra-domain routing and resource scheduling, realizing an efficient real-time decision-making mechanism and ensuring that routing and resource scheduling can respond to changes in network topology and status in a timely manner.

[0074] Specifically, traditional MDP frameworks have limitations for highly dynamic and real-time systems like low-orbit satellite networks. By modeling the problem as a real-time Markov decision process and employing the MADQN algorithm, each satellite acts as an intelligent agent. Its input includes satellite state information and link state information from neighboring nodes, and its output action is one of four possible transmission directions. To more effectively process complex and dynamically changing state information, a self-attention mechanism is introduced to learn key matrices related to the satellite network state and dynamically assign weights, thereby enhancing its understanding of and responsiveness to environmental conditions.

[0075] In step S105, the DQN-ST algorithm is introduced to globally perceive the network status of the virtual constellation composed of virtual nodes, improve the generalization ability of the model, avoid overfitting, and ensure the efficiency and stability of cross-domain communication.

[0076] Specifically, in satellite networks, due to the continuous high-speed motion of satellites, satellite nodes constantly move from one region to the next. At this point, the originally selected domain boundary nodes and inter-domain communication links become invalid, and the satellite network needs to recalculate and select new boundary nodes. By designing a dynamic and adaptive domain boundary node mapping mechanism that comprehensively considers node effectiveness and resource availability, the problem of boundary node failure caused by high-speed satellite motion is addressed. Subsequently, a prioritized experience replay deep Q-network based on an improved Sumtree is proposed. Through task-aware state-space modeling and a dynamic bandwidth allocation mechanism, global coordinated optimization of cross-domain routing and resource scheduling is achieved.

[0077] In some embodiments, based on the low-orbit satellite network simulation modeling method provided by the present invention, multi-dimensional resource feature extraction and characterization are performed to obtain a multi-dimensional resource coupling model, such as Figure 3 As shown, the process specifically includes the following steps S301 to S305:

[0078] Step S301: Construct a space environment model. Based on orbital parameters such as altitude, inclination, number of orbital planes, number of satellites in each orbit, and phase factor between adjacent orbital planes, the constellation configuration, inter-satellite link, and communication link models are designed. The neighbor satellite connection relationship of a satellite logical address (m, n) is defined by the following formula:

[0079]

[0080] Step S302: Construct an energy consumption model to quantify the satellite antenna receiving power, transmitting power and data transmission energy consumption, establish a mathematical relationship between satellite communication energy consumption and transmission distance, data volume and channel status; derive the inter-satellite transmission rate through the Shannon formula Among them, w (i,j) is the bandwidth of the wireless channel, σ 2 is the noise power; free space propagation loss during transmission The expression is Among them, d (i,j) is the distance between satellites, λ is the wavelength of electromagnetic wave, c is the speed of light, and f is the operating frequency; the antenna receiving power is When the receiving power of a satellite is given, the transmitting power of the satellite can be deduced. The expression is

[0081] Step S303: Construct an illumination model and predict the efficiency of solar panels and the impact of the earth's shadow period on satellite power supply through real-time interaction between the satellite's orbital position and the earth's shadow area; use a cylindrical earth shadow model to predict the earth's shadow period, with the following constraints: If this formula is not satisfied, it means that the satellite is in the illuminated area.

[0082] Step S304: Construct an on-orbit life model, and construct a relationship between battery reliability and lost life based on the lithium-ion battery discharge depth D(t) and the battery cycle life consumption L t1,t2 The functional relationship is The fitting expression of battery reliability R(t) is:

[0083] Step S305: Construct a dynamic interaction model of multi-dimensional resource coupling characteristics. By revealing the constraints of orbital configuration on communication link distance, the cumulative impact of energy mode switching caused by earth shadow events on battery life, and the interactive mechanism of dynamic tracking strategy and energy consumption distribution, it breaks through the limitations of traditional models that independently model single-dimensional resources, constructs the coupling mechanism and dynamic association of multi-dimensional resources, and conducts collaborative modeling and system optimization. Through Monte Carlo simulation, it quantifies the joint impact of orbital altitude on communication distance and energy load, providing data and model basis for subsequent research on low-orbit satellite network traffic prediction and routing technology.

[0084] like Figure 4 As shown, the low-orbit satellite network regional traffic prediction algorithm based on the improved Transformer architecture of the present invention specifically includes the following steps S401 to S405:

[0085] Step S401: Data traffic modeling, the earth's surface is divided into 288 traffic blocks at intervals of 15 degrees according to longitude and latitude, and the data traffic intensity is converted according to the number of Internet users in each block. Each person is regarded as a node, and every 10 6 The traffic intensity of each node is equivalent to 1Mps. According to the actual global Internet user statistics, the global data service traffic intensity is obtained, and the traffic intensity of each traffic block is L n . Since most of the business traffic will be distributed through ground communication network facilities, it can be assumed that the proportion of traffic distributed through the low-orbit satellite network α is set to 5%. The temporal distribution of traffic is represented by the discount coefficient γ. In each traffic block, the longitude of its center position is selected as a reference to infer the time zone to which the block belongs. Assuming that the discount coefficient is 1 when the user is most active and 0 when there is almost no activity, the tidal change of traffic can be constructed. The low-orbit satellite network traffic of any traffic block is represented as C n=ɑγL n ,n=1,2,…,288, unit is Mbps.

[0086] Step S402: Empirical Mode Decomposition (EMD) is used to decompose the complex original traffic sequence into multiple single-frequency and mutually independent IMF subsequences. Each IMF subsequence is not only easier to model, but also retains the key features of the original signal, allowing subsequent prediction models to better capture short-term fluctuations and long-term trends. Assuming the original sequence is X(t), the sequence can be decomposed into a finite number of IMF subsequences c j and a remainder term r n ,Right now The formula for calculating its standard deviation is: Where h i-1，j 、h i，j Represents the intrinsic mode function, i represents the number of iterations, j represents the order of the function, and N represents the total number of iterations. The specific EMD algorithm implementation process is as follows Figure 5 shown.

[0087] Step S403: Implement the forward process of the PatchTST model and express the i-th one-dimensional sequence as Each sequence is independently input into the Transformer backbone network, and the prediction result of the sequence is calculated as

[0088] Step S404: Implement the block process of the PatchTST model. Each input one-dimensional time series x (i) The blocks are divided into patches, which can be overlapping or non-overlapping. The patch length is recorded as P, and the non-overlapping area between two consecutive patches is recorded as S. After the patch process, a patch sequence will be generated. Among them, N represents the number of blocks generated, After the block operation, the input token size can be reduced from L to about L / S, and the memory usage and computational complexity of the attention map are reduced to S. 2 Therefore, under the conditions of limited training time and GPU memory resources, the block operation allows the model to understand longer historical sequences, thereby significantly improving the model's prediction performance.

[0089] Step S405: Implement the Transformer encoder, through the trainable linear projection W P ∈R D×P , and a learnable additive positional encoding W pos ∈R D×N, mapping the blocks to a potential representation space of dimension D, the data input to the Transformer encoder is represented as in Each attention head h=1,...,H of the multi-head attention transforms it into a query matrix Bond Matrix and the value matrix in Use scaled product to get attention output After the attention output is processed by the normalization layer and the feedforward network layer with residual connection, the generated result is represented as z (i) ∈R D×N , using a fully connected layer with a linear head to obtain the prediction results

[0090] The large-scale low-orbit satellite network model based on geographical area division of virtual nodes and autonomous domains provided by the present invention specifically includes the following steps S601-S602:

[0091] Step S601: Divide the earth's surface into regular hexagonal grids based on geographical regions. However, since the sphere cannot be strictly divided into regular hexagonal grids, the earth's surface is projected into a simple cylindrical shape and converted into a rectangle with an aspect ratio of 2:1. The rectangle is then divided into 112 regular hexagonal grids, each of which represents a virtual node and autonomous domain.

[0092] Step S602: Based on the decoupled low-orbit satellite network routing method of virtual nodes and autonomous domains, the hexagonal grid structure of the earth's surface is mapped to the satellite constellation through the spatial association between the satellite's ground projection position and the surface grid. Satellite groups in the same block will belong to the same autonomous domain and be aggregated into an abstract virtual node. A routing and resource information update strategy combining flooding updates and local updates is adopted. Inter-domain information is updated through flooding, and the information of satellite nodes at the boundary of the block is flooded to the entire satellite constellation. Intra-domain information is updated through local updates, and information such as the status of each satellite and link is shared and updated within the autonomous domain.

[0093] The present invention also provides a low-orbit satellite network domain routing and resource scheduling algorithm based on the improved MADQN, the framework is as follows Figure 6 As shown, the process specifically includes the following steps S701 to S702:

[0094] Step S701: Modeling a real-time Markov decision process, state space s t Including satellite energy consumption, link quality, battery reliability and neighbor node status and the satellite node's last action decision result, the action space a t For the transmission direction of “up, down, left, right”, the reward function r(st , a t ) includes comprehensive latency, energy consumption and link stability; in the RTMDP framework, actions and states change simultaneously, and the agent can immediately observe the next state after selecting an action and directly infer the next action. This real-time decision-making mechanism allows the agent to more closely track rapidly changing environmental conditions, ensuring that routing selection and resource scheduling can respond promptly to changes in the topology and state of the low-orbit satellite network.

[0095] The action value function expression is as follows:

[0096]

[0097] Where, Indicates that in a given state s t and action a t When the transition to a specific state s t+1 The expected probability of . Indicates that in a given state action pair (s t+1 ,a t ), execute action a according to the current strategy π t+1 The expected cumulative reward for the state sequence that the agent may encounter after ;

[0098] The state value function expression is as follows:

[0099]

[0100] Where, Indicates that in a given state action pair (s t , a t ) is the expected cumulative reward for the state sequence that the agent may encounter under the condition of

[0101] Step S702: Introduce the self-attention mechanism and construct a multi-head attention structure to map the input neighbor node state information into query matrices Bond Matrix and value matrix The attention weight calculation follows the formula where d k is the dimensionality scaling factor of the key vector, which is used to avoid the gradient saturation problem of the Softmax function. By computing the outputs of multiple attention heads in parallel and concatenating them into a joint representation, the algorithm can dynamically allocate weights, highlight the states of neighboring nodes with high link stability and large resource redundancy, and suppress noise interference. The attention output is input into the fully connected network after residual connection and layer normalization to generate routing decision actions.

[0102] The present invention provides a low-orbit satellite network inter-domain routing selection and resource scheduling algorithm based on DQN-ST, the framework is as follows Figure 7 As shown, the process specifically includes the following steps S801 to S802:

[0103] Step S801: Implement dynamic and adaptive domain boundary node mapping, comprehensively consider indicators such as node effectiveness and node available resources, evaluate satellite nodes in the domain, and select satellites that map domain boundary nodes. In actual satellite network cross-domain routing scenarios, the selection of domain boundary nodes needs to be deeply coupled with the characteristics of dynamic routing tasks. When a satellite node detects the need to forward a cross-domain service request, it predicts the candidate node set {s1, s2, ..., s n The orbital position in the future Δt period, where the value of Δt is closely related to the life cycle of a typical routing task. i )=dis(s i ) is calculated not only considering the geometric distance to the adjacent domain at the current moment, but also considering the time sensitivity of the specific routing task. For low-latency services such as real-time video streaming, nodes closer to the next domain are selected to shorten the propagation delay. For batch data transmission tasks, nodes with longer remaining service time are preferred to ensure connection continuity. Node resource index R(s i )=Cal(s i )+CacHe(s i ) evaluation needs to be dynamically weighted based on actual business needs. By mapping the standardized score Score = 0.7E′ + 0.3R′ to the actual routing decision, the system can automatically select the boundary node that can meet the business service quality requirements and has the best resource utilization.

[0104] The node validity is expressed as follows:

[0105] E(s i )=dis(s i )s i ∈{s1,s2,…,s n}

[0106] The available resources of the node are expressed as follows:

[0107] R(s i )=Cal(s i )+Cache(s i )s i ∈{s1,s2,…,s n}

[0108] The weight coefficients of the two indicators are set to 0.7 and 0.3 respectively, and the standardized weighted comprehensive score is shown in the following formula.

[0109]

[0110] Step S802: Implement the DQN-ST algorithm architecture for task optimization, apply the Sumtree-based improved priority experience replay mechanism to the deep Q network algorithm, and the state space s t ={l(t),u(t),τ(t),γ(t)}, where l(t) represents the satellite node where the task of time slot t is located, u(t) represents the link connection status of the satellite node where the task of time slot t is located, τ(t) represents the data volume of the task of time slot t, and γ(t) represents the network status information of the virtual constellation of time slot t; the action space a t ={h(t), B(t)}, where h(t) represents the satellite node where the agent performs the next forwarding in time slot t; B(t) represents the bandwidth resources allocated to the task in time slot t.

[0111] In the state space design described above, l(t) not only contains the location information of the current satellite node, but also integrates the status of the routing task queue undertaken by the node (such as the number of cross-domain connections being processed, the priority of the data packets to be forwarded, etc.), so that the intelligent body can perceive the impact of local load on global routing; the link connection status in u(t) constructs a link stability matrix through the actual measured bit error rate, delay jitter and other parameters to provide physical layer support for action selection. The bandwidth allocation B(t) in the action space needs to be linked with the business type of the specific routing task. For example, in order to meet the sudden large data transmission needs in the low-orbit satellite earth observation mission, the algorithm expands the dynamic adjustment range of B(t) and establishes a nonlinear relationship model between bandwidth allocation and throughput and packet loss rate through online learning. The priority experience replay mechanism based on Sumtree needs to consider the spatiotemporal correlation of routing tasks in actual deployment. By giving higher TD error weights to task samples with high delay sensitivity or high bandwidth requirements, the priority p i =|δ(i)|+∈, where ∈ is a very small constant, and the sampling probability P(i) is calculated according to the formula Calculation; Based on the above calculation and Q value results, the optimal next-hop satellite node h(t) is selected to ensure the efficiency and stability of cross-domain communication.

[0112] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a large-scale low-orbit satellite network domain-specific intelligent routing method.

[0113] Corresponding to the above method, the present invention also provides a device, which includes a computer device, the computer device includes a processor and a memory, the memory stores computer instructions, the processor is used to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0114] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the art.

[0115] In summary, the present invention provides a domain-specific intelligent routing method for large-scale low-orbit satellite networks, belonging to the field of satellite communication technology. This method is based on a low-orbit satellite network simulation modeling method for multi-dimensional resource feature extraction and characterization, and a low-orbit satellite network regional traffic prediction algorithm based on an improved Transformer architecture. It provides resource constraints and spatiotemporal correlation characteristics for routing decisions. It also decouples the routing process into intra-domain and inter-domain parts based on a large-scale low-orbit satellite network model divided into virtual nodes and autonomous domains based on geographical regions. It implements a low-orbit satellite network intra-domain routing and resource scheduling algorithm based on an improved MADQN and a low-orbit satellite network inter-domain routing and resource scheduling algorithm based on a DQN-ST, achieving hierarchical routing and resource decoupling scheduling, breaking through the traditional single-layer routing framework and constructing a domain-specific intelligent routing system that combines flexibility and scalability. This invention significantly enhances the service capabilities of low-orbit satellite networks in extreme scenarios such as high-frequency topology changes, strict resource constraints, and spatiotemporal traffic mutations, providing a highly reliable, low-latency routing solution for integrated space-ground communications, emergency disaster warnings, and global real-time data transmission.

[0116] It should be understood by those skilled in the art that the various exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is specifically performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0117] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0118] In the present 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 replace features of other embodiments.

[0119] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations to the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A domain-specific intelligent routing method for a large-scale low-orbit satellite network, characterized by: Step 1: Implement low-orbit satellite network simulation modeling based on multi-dimensional resource feature extraction and characterization, quantify the impact of multi-factor coupling on network performance, provide resource constraints and spatiotemporal correlation characteristics for routing algorithms, and support the real-time and adaptability of routing decisions; Step 2: Implement a regional traffic prediction algorithm for low-orbit satellite networks based on an improved Transformer architecture, build a global traffic tidal effect model, and map the prediction results into a routing decision state space to drive resource allocation and congestion avoidance strategies. Step 3: Build a large-scale low-orbit satellite network model based on geographically divided virtual nodes and autonomous domains, decoupling the routing process into intra-domain and inter-domain parts. Step 4: Based on the improved multi-agent deep Q network MADQN, the routing selection and resource scheduling algorithm within the low-orbit satellite network domain is implemented. Each satellite is modeled as a multi-agent. Based on the real-time Markov decision process RTMDP, the multi-dimensional resource characteristics of neighboring nodes and the previous action information are input, and the next-hop transmission direction is output. The matrix information of the satellite network state is learned by combining the self-attention mechanism to enhance the multi-agent's understanding and response ability to the environmental state. Step 5: Based on the deep Q network DQN-ST combined with the sum tree mechanism, the inter-domain routing selection and resource scheduling algorithm of the low-orbit satellite network is implemented, a dynamic mapping relationship of cross-domain routing is established, and the priority experience replay mechanism improved by SumTree is used to determine the priority of satellite routing task samples according to the temporal difference error. The optimal next-hop satellite node is selected based on the Q value, and the inter-domain routing intelligent decision is output after aggregation.

2. A large-scale low-orbit satellite network domain-specific intelligent routing method according to claim 1, characterized in that: In the step 1, low-orbit satellite network simulation modeling is performed based on multi-dimensional resource feature extraction and characterization, including a space environment model, an energy consumption model, an illumination model, an on-orbit life model, and a multi-dimensional resource coupling characteristic and dynamic interaction model; The space environment model is designed based on orbital parameters such as altitude, inclination, number of orbital planes, number of satellites in each orbit, and phase factor between adjacent orbital planes to design constellation configuration, intersatellite link, and communication link models. The neighbor satellite connection relationship of a satellite logical address (m, n) is defined by the following formula: The energy consumption model quantifies the satellite antenna receiving power, transmitting power and data transmission energy consumption, and establishes a mathematical relationship between satellite communication energy consumption and transmission distance, data volume and channel status; Derivation of inter-satellite transmission rate using Shannon's formula Among them, w (i，j) is the bandwidth of the wireless channel, σ 2 is the noise power; free space propagation loss during transmission The expression is Among them, d (i，j) is the distance between satellites, λ is the wavelength of electromagnetic wave, c is the speed of light, and f is the operating frequency; the antenna receiving power is When the receiving power of a satellite is given, the transmitting power of the satellite can be deduced. The expression is The illumination model predicts the efficiency of solar panels and the impact of the earth's shadow period on satellite power supply through real-time interaction between the satellite's orbital position and the earth's shadow area; the cylindrical earth shadow model is used to predict the earth's shadow period, with the following constraints: If this formula is not satisfied, it means that the satellite is in the illuminated area; The on-orbit life model is based on the relationship between the battery reliability and the lost life according to the lithium-ion battery discharge depth and the battery cycle life consumption. t1，t2 The functional relationship is The fitting expression of battery reliability R(t) is: The dynamic interaction model of multi-dimensional resource coupling characteristics constructs the coupling mechanism and dynamic association of multi-dimensional resources by analyzing model representations such as space environment, energy consumption, illumination, and on-orbit life, and carries out collaborative modeling and system optimization; and quantifies the joint impact of orbital altitude on communication distance and energy load through Monte Carlo simulation.

3. A large-scale low-orbit satellite network domain-specific intelligent routing method according to claim 1, characterized in that: In the second step, the regional traffic prediction algorithm of the low-orbit satellite network based on the improved Transformer architecture is studied, including data service traffic modeling and traffic prediction algorithm based on PatchTST; The data traffic modeling method divides the earth's surface into regions every 15° according to longitude and latitude, and converts the traffic intensity according to the number of Internet users in each region to obtain the global data traffic intensity L. n , the temporal distribution of traffic is represented by the discount coefficient γ; In each traffic block, the longitude of its center is selected as a reference to calculate the time zone to which the block belongs. It is assumed that the discount coefficient is 1 when the user is most active and 0 when there is almost no activity. The low-orbit satellite network traffic of any traffic block is expressed as C n =αγL n , n=1,2,…,288, unit is Mbps; The PatchTST-based low-orbit satellite network traffic prediction algorithm uses EMD technology to decompose the original traffic sequence into multiple single-frequency and mutually independent intrinsic mode function (IMF) subsequences and a residual signal, effectively separating different frequency components, making each subsequence more stable and easier to model. The PatchTST model is introduced for prediction of each IMF subsequence and residual signal. The PatchTST model combines the advantages of the Transformer architecture, can maintain high efficiency when processing long sequences, and can capture complex patterns and dependencies in time series. The output results are summed as the prediction output. The implementation method of EMD mentioned above assumes that the original sequence is X(t), which can be decomposed into a finite number of IMF subsequences c j and a remainder term r n , The formula for calculating its standard deviation is: Where h i-1，j 、h i，j represents the intrinsic mode function, i represents the number of iterations, j represents the order of the function, and N represents the total number of iterations; The method for constructing the traffic prediction model based on PatchTST is as follows: for a set of multivariate time series samples, assuming that the length of the review window, i.e., the time data of the input model, is L, the samples (x1, ..., x t ,...,x L ), where, at time slot t, x t It is a one-dimensional column vector containing M elements, predicting the data value (x L+1 ,...,x L+T ); the model structure includes the forward process, the block process and the Transformer encoder; The forward process described above represents the i-th one-dimensional sequence as Each sequence is independently input into the Transformer backbone network, and the prediction result of the sequence is calculated as In the block process, each input one-dimensional time series x (i) First, it is divided into blocks, the block length is recorded as P, and the non-overlapping area between two consecutive blocks is recorded as S. After the block process, a block sequence will be generated Among them, N represents the number of blocks generated, The memory usage and computational complexity of the attention map are expressed in S 2 Therefore, under the conditions of limited training time and GPU memory resources, block operation can allow the model to understand longer historical sequences, thereby significantly improving the model's prediction performance; The Transformer encoder is trained through a linear projection W P ∈R D×P , and a learnable additive positional encoding W pos ∈R D×N , mapping the blocks to a potential representation space of dimension D, the data input to the Transformer encoder is represented as in Then each attention head h=1,...,H of the multi-head attention transforms it into a query matrix Bond Matrix and the value matrix in Use scaled product to get attention output After the attention output is processed by the normalization layer and the feedforward network layer with residual connection, the generated result is represented as z (i) ∈R D×N , and finally use a fully connected layer with a linear head to obtain the prediction result 4. A large-scale low-orbit satellite network domain-specific intelligent routing method according to claim 1, characterized in that: In step 3, a large-scale low-orbit satellite network model based on the division of virtual nodes and autonomous domains in geographical areas includes a hexagonal grid division method for geographical areas and a low-orbit satellite network routing method based on the fusion of virtual nodes and autonomous domains; The hexagonal grid division method for geographic regions is to perform a simple cylindrical projection on the Earth's surface and convert it into a rectangle with an aspect ratio of 2:

1. The rectangle is then divided into 112 regular hexagonal grids, each covering a 15°×15° area in longitude and latitude. The grid mapping rules ensure that adjacent virtual nodes are seamlessly connected, reducing the number of routing hops. The low-orbit satellite network routing method based on the fusion of virtual nodes and autonomous domains maps the hexagonal grid structure of the earth's surface onto the satellite constellation through the spatial association between the satellite's ground projection position and the surface grid. Satellite groups in the same block will belong to the same autonomous domain and be aggregated into an abstract virtual node. Adopting a routing and resource information update strategy that combines flooding updates with local updates, the inter-domain information is updated by flooding, and the boundary satellite node information of the block is flooded to the entire satellite constellation; The information within the domain is updated through local updates, and the information such as the status of each satellite and link is shared and updated within the autonomous domain.

5. A large-scale low-orbit satellite network domain-specific intelligent routing method according to claim 1, characterized in that: In step 4, a low-orbit satellite network intra-domain routing selection and resource scheduling algorithm based on an improved MADQN includes a real-time Markov decision process and a self-attention mechanism; The real-time Markov decision process, the state space s t Including satellite energy consumption, link quality, battery reliability and neighbor node status and the satellite node's last action decision result, the action space a t For the transmission direction of "up, down, left, right", the reward function r(s t , a t ) including comprehensive latency, energy consumption, and link stability; in the RTMDP framework, actions and states change simultaneously. The agent can immediately observe the next state after selecting an action and directly infer the next action. This real-time decision-making mechanism allows the agent to more closely track rapidly changing environmental conditions, ensuring that routing and resource scheduling can respond promptly to changes in the topology and state of the low-orbit satellite network; The action value function expression is as follows: Where, Indicates that in a given state s t and action a t When the transition to a specific state s t+1 The expected probability of Indicates that in a given state action pair (s t+1 ,a t ), execute action a according to the current strategy π t+1 The expected cumulative reward for the state sequence that the agent may encounter after ; The state value function expression is as follows: Where, Indicates that in a given state action pair (s t ,a t ) under the condition of the expected cumulative reward for the state sequence that the agent may encounter; The self-attention mechanism maps the input neighbor node state information into query matrices by constructing a multi-head attention structure. Bond Matrix and value matrix The attention weight calculation follows the formula where d k is the dimensionality scaling factor of the key vector, used to avoid gradient saturation in the Softmax function. By computing the outputs of multiple attention heads in parallel and concatenating them into a joint representation, the algorithm can dynamically assign weights, highlighting neighboring nodes with high link stability and high resource redundancy, and suppressing noise interference. The attention output is residually connected and layer normalized, and then input into the fully connected network to generate routing decision actions.

6. A large-scale low-orbit satellite network domain-specific intelligent routing method according to claim 1, characterized in that: In step 5, the DQN-ST-based low-orbit satellite network inter-domain routing selection and resource scheduling algorithm includes a dynamic and adaptive domain boundary node mapping mechanism and a DQN-ST algorithm architecture for routing task optimization; In the dynamic and adaptive domain boundary node mapping mechanism, when a node needs to communicate across autonomous domains, the node within the domain will forward the request information to the domain boundary node. Due to the continuous high-speed movement of the satellite, the satellite node will continue to move from one area to the next. At this time, the originally selected domain boundary node and the inter-domain communication link will become invalid. Considering the node effectiveness and node available resources and other indicators, the satellite nodes within the domain are evaluated and the mapping satellite of the domain boundary node is selected. The calculation of the node effectiveness index not only considers the geometric distance to the adjacent domain at the current moment, but also needs to be combined with the time sensitivity of the specific routing task. The evaluation of the node resource index needs to be dynamically weighted according to the actual business needs, and the node effectiveness E(s) is defined. i )=dis(s i )s i ∈{s1,s2,...,s n }, node available resources R(s i )=Cal(s i )+Cache(s i )s i ∈{s1,s2,...,s n }, the weight coefficients of the two indicators are set to 0.7 and 0.3 respectively, then the standardized weighted comprehensive score Mapping to actual routing decisions, the system can automatically select edge nodes that can meet the business quality of service (QoS) requirements and have optimal resource utilization; The DQN-ST algorithm architecture for routing task optimization applies the priority experience replay mechanism based on Sumtree to the deep Q network algorithm, and the state space s t ={l(t), u(t), τ(t), γ(t)}, where l(t) represents the satellite node where the task of time slot t is located, u(t) represents the link connection status of the satellite node where the task of time slot t is located, τ(t) represents the data volume of the task of time slot t, and γ(t) represents the network status information of the virtual constellation of time slot t; the action space a t ={h(t), B(t)}, where h(t) represents the satellite node where the agent performs the next forwarding in time slot t; B(t) represents the bandwidth resources allocated to the task in time slot t; the priority experience replay mechanism based on Sumtree considers the spatiotemporal correlation of routing tasks in actual deployment, and gives higher TD error weights to task samples with high delay sensitivity or high bandwidth requirements. i =|δ(i)|+∈, where ∈ is a very small constant, and the sampling probability P(i) is calculated according to the formula Calculation; Based on the above calculation and the Q value calculation results, the optimal next-hop satellite node h(t) is selected to ensure the efficiency and stability of cross-domain communication.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images