An On-Demand Routing Agent System and Generation Method for Low-Earth Orbit Satellite Networks

Abstract

This application provides an on-demand routing intelligent agent system and generation method for low-Earth orbit (LEO) satellite networks, relating to the field of satellite communication technology. The intelligent agent system includes a planning and decision-making module, a network perception module, and an execution module. The planning and decision-making module includes a large model integrating carefully designed system prompts and parses user request text to obtain communication request parameters. The network perception module is used to perceive the time-varying network state of the LEO satellite network and output a network topology map. The execution module includes a satellite access tool and a routing selection tool. The satellite access tool is used to determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the originating and ending users based on the network topology map and the uplink and downlink access satellites, using QoS parameter values ​​as constraints and a routing calculation strategy corresponding to the QoS parameter type, to determine the routing path between the uplink and downlink access satellites, thereby improving the service quality and intelligence level of the LEO satellite network.

Description

Technical Field

[0001] This application relates to the field of satellite communication technology, and in particular to an on-demand routing intelligent agent system and generation method for low-Earth orbit satellite networks. Background Technology

[0002] Low-Earth orbit (LEO) satellite networks, with their advantages of wide coverage and low transmission latency, have become a core support solution for global communications. However, the high-speed movement of satellite nodes leads to dynamic changes in network topology, posing a severe challenge to routing planning. Some existing routing technologies rely on pre-defined algorithms to statically calculate paths, while others are based on reinforcement learning or deep neural network algorithms. These typically require a large amount of historical data and simulation samples for training, resulting in limited model generalization ability, rigid model structure, and a single training objective. Consequently, they lack adaptability to dynamically changing network topologies and diverse service demands, thus restricting the service quality and intelligence level of LEO satellite networks. Summary of the Invention

[0003] This application provides an on-demand routing agent system and generation method for low-Earth orbit satellite networks to alleviate or solve one or more technical problems existing in the prior art.

[0004] In a first aspect, embodiments of this application provide an on-demand routing agent for low-Earth orbit satellite networks, comprising:

[0005] The planning and decision-making module includes: a pre-trained large model and a tool invocation module; the large model is used to parse user requirement text to obtain communication requirement parameters. The user requirement text is the user's communication requirements described in natural language. The communication requirement parameters include: the latitude and longitude coordinates of the origin and destination users, the QoS parameter type, and the corresponding QoS parameter value; the tool invocation module is used to invoke satellite access tools and routing tools.

[0006] The network sensing module is used to dynamically sense the time-varying network status of the low-Earth orbit satellite network and output a network topology map.

[0007] The execution module includes a satellite access tool and a routing tool. The satellite access tool is used to determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the originating and ending users by combining the network topology map. The routing tool is used to determine the routing path between the uplink and downlink access satellites based on the network topology map and the uplink and downlink access satellites, with QoS parameter values ​​as constraints, and based on the routing calculation strategy corresponding to the QoS parameter type.

[0008] In some embodiments of this application, the satellite access tool is further specifically used to generate multiple sets of uplink and downlink access satellites that correspond one-to-one according to multiple access strategies, and the routing tool is further specifically used to calculate the routing path between each set of uplink and downlink access satellites based on the routing calculation strategy, so as to obtain multiple routing paths that correspond one-to-one with the multiple sets of uplink and downlink access satellites, as well as the communication performance parameter value of each routing path.

[0009] Accordingly, the large model is also used to select the optimal routing path among multiple routing paths that matches the communication performance parameter value with the QoS parameter value.

[0010] In some embodiments of this application, multiple access strategies include: a multi-parameter access strategy; the satellite access tool generates multiple sets of uplink and downlink access satellites corresponding to each other according to the multi-parameter access strategy, specifically including: obtaining multi-dimensional evaluation parameters of multiple sets of candidate satellite combinations; quantifying the information contribution rate of each evaluation parameter according to the information entropy model, and dynamically assigning corresponding weights to each evaluation parameter based on the information contribution rate; calculating the multi-parameter weighted evaluation value of each set of candidate satellite combinations according to the weights assigned to each evaluation parameter; and selecting the candidate satellite combination with the largest multi-parameter weighted evaluation value as a set of uplink and downlink access satellites corresponding to the multi-parameter access strategy.

[0011] In some embodiments of this application, the planning decision module further includes:

[0012] The prompting module is used to generate prompt text based on user-defined text and preset prompt words, which is then input into the large model. The prompt text guides the large model to perform the following operations:

[0013] The system parses the latitude and longitude coordinates of the start and end users based on the user's request text and context information, and then inputs these coordinates into the satellite access tool.

[0014] The system identifies the communication service type based on the user's text requirements, matches the corresponding QoS parameter type and QoS parameter value in the preset service database according to the communication service type, and inputs the QoS parameter type and QoS parameter value into the routing tool.

[0015] In some embodiments of this application, the large model is also used to output structured thought analysis text before calling the satellite access tool or routing tool. The structure of the thought analysis text is set by the prompting engineering module. The thought analysis text is used to describe the parameter basis, constraints and reasoning process of the tool call.

[0016] In some embodiments of this application, the planning decision module further includes a session memory unit, which is used to store user requirement text, parsing results of large models, call records of tool calling modules, and routing paths; the prompt engineering module is also used to generate context information based on the stored information of the session memory unit.

[0017] In some embodiments of this application, the QoS parameter types correspond one-to-one with the routing calculation strategies:

[0018] If the QoS parameter type is delay, the routing calculation strategy is the weighted Dijkstra algorithm based on the shortest delay, and the edge weight of the algorithm is the sum of the inter-satellite propagation delay and the transmission delay.

[0019] If the QoS parameter type is throughput, the routing calculation strategy is the weighted Dijkstra algorithm based on minimum load, and the edge weight of the algorithm is the standard deviation of the satellite node neighbor queue length.

[0020] If the QoS parameter type is inter-satellite hop count, the routing calculation strategy is a weighted Dijkstra algorithm based on the minimum hop count, and the edge weight of the algorithm is fixed at 1.

[0021] Secondly, embodiments of this application provide an on-demand route generation method for low-Earth orbit satellite networks. This method is applied to the system provided in any one of the embodiments of this application, and includes:

[0022] Obtain the user request text; the user request text is the user's communication requirements described in natural language.

[0023] The pre-trained large model is used to parse user request text to obtain communication request parameters. The communication request parameters include: latitude and longitude coordinates of the start and end users, QoS parameter type and corresponding QoS parameter value.

[0024] Based on the network topology diagram of the low-Earth orbit satellite network, determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the originating and ending users;

[0025] Based on the network topology and uplink / downlink access satellites, and using QoS parameter values ​​as constraints, the routing path between uplink / downlink access satellites is determined according to the routing calculation strategy corresponding to the QoS parameter type.

[0026] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor implements the method provided in the second aspect of embodiments of this application when executing the computer program.

[0027] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method provided in the second aspect of embodiments of this application.

[0028] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method provided in the second aspect of embodiments of this application.

[0029] Based on any of the above technical solutions, this application has at least the following beneficial effects or advantages:

[0030] After obtaining the user's request text, a pre-trained large model is used to parse the user's request text to obtain communication request parameters. The communication request parameters include: the latitude and longitude coordinates of the origin and destination users, the QoS parameter type and the corresponding QoS parameter value. Based on the network topology map of the low-Earth orbit satellite network, the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the origin and destination users are determined. According to the network topology map and the uplink and downlink access satellites, with the QoS parameter value as a constraint, the routing path between the uplink and downlink access satellites is determined based on the routing calculation strategy corresponding to the QoS parameter type, thereby improving the service quality and intelligence level of the low-Earth orbit satellite network.

[0031] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0032] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments according to this application and should not be construed as limiting the scope of this application.

[0033] Figure 1 This illustration shows an on-demand routing intelligent agent system for low-Earth orbit satellite networks provided in an embodiment of this application. Figure 1 ;

[0034] Figure 2 The flowchart illustrates an on-demand route generation method for low-Earth orbit satellite networks provided in an embodiment of this application.

[0035] Figure 3 This illustration shows an on-demand routing intelligent agent system for low-Earth orbit satellite networks provided in an embodiment of this application. Figure 2 ;

[0036] Figure 4A block diagram of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0037] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the concept or scope of this application. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0038] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following related technologies are optional solutions and can be arbitrarily combined with the technical solutions of the embodiments of this application, all of which fall within the protection scope of the embodiments of this application. It should be noted that the application scenarios or application examples provided in this application are for ease of understanding, and the embodiments of this application do not specifically limit the application of the technical solutions.

[0039] Furthermore, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0040] The technological evolution and large-scale deployment of Low Earth Orbit (LEO) satellite communication networks are accelerating. Compared with Medium Earth Orbit (MEO) and Geostationary Earth Orbit (GEO) satellites, LEO communication networks have advantages such as flexible deployment, high spectrum utilization, and low latency. They have shown significant application potential, especially in remote areas, oceans, and airspace where terrestrial networks are difficult to cover, thus supporting important applications such as global wide-area access, broadband communication, emergency response, and remote sensing.

[0041] In complex network environments with multiple nodes and links, routing mechanisms directly impact the stability of constellation communication. They play a decisive role in load balancing and real-time calculation of the least-latency path, serving as a core control element in low-Earth orbit satellite network architecture and playing a crucial role in selecting data transmission paths. Their performance directly determines the network's quality of service (including latency, throughput, and reliability). Routing algorithms dynamically select appropriate transmission paths based on multi-dimensional parameters such as inter-satellite link status, node locations, and network load, achieving efficient coordination in inter-satellite communication and optimal decision-making for data forwarding.

[0042] Against this backdrop, access control and routing planning for satellite networks have become core technical challenges for LEO communication networks. Various routing algorithms for LEO communication networks have been proposed in existing technologies. In one example, a multi-objective optimization model is trained to generate differentiated routing strategies; in another, reinforcement learning is used to optimize node connectivity and load distribution. While these solutions can achieve feasible path planning under known network topologies and achieve certain optimization effects on performance metrics such as latency and throughput, they still face the following key problems: they primarily focus on path optimization and topology learning at the network layer, lacking the ability to understand and map upper-layer service requirements, and failing to form a closed-loop mechanism from user requirement analysis to satellite access selection, path planning, and execution feedback; algorithms based on reinforcement learning or deep neural networks typically require a large amount of historical data and simulation samples for training, resulting in limited model generalization ability, rigid model structures, and singular training objectives. They are ill-suited to adapting to dynamically changing network topologies and diverse service requirements, making it difficult to balance multi-dimensional performance metrics such as latency, throughput, and reliability.

[0043] Therefore, this application aims to provide an on-demand routing intelligent agent system and generation method for low-Earth orbit (LEO) satellite networks, solving the problem that existing LEO communication network satellite routing algorithms rely on preset strategies or fixed algorithms and cannot generate routing schemes on demand according to users' real-time service needs. This intelligent agent achieves closed-loop management of the entire process from user demand analysis and network status awareness to satellite access and route selection. It can autonomously select the optimal satellite communication path in dynamic topology, high concurrency, and multi-service scenarios, and supports the automatic generation and real-time adjustment of service-oriented differentiated routing strategies, effectively enhancing the system's adaptability to complex environments and unexpected tasks.

[0044] The on-demand routing agent constructed in this application is based on a Large Language Model (LLM) and is implemented through the following technical solutions: First, it intelligently identifies the starting position of the task from the user's input natural language requirements and parses it into corresponding latitude and longitude information, while simultaneously identifying the service type and determining the corresponding routing algorithm; then, it obtains real-time network topology and link status information through network state awareness and selects access nodes in conjunction with satellite access tools; subsequently, it uses the selected access nodes as input and calls the service-driven routing algorithm to generate the corresponding path node sequence, completing end-to-end intelligent access and routing planning for user service transmission. The agent constructed in this application is a novel intelligent routing management architecture with high-level semantic understanding, dynamic environment awareness, autonomous planning, and closed-loop execution capabilities, enabling adaptive optimal route selection in multi-service, multi-target, and multi-node environments. This overcomes the bottleneck of existing routing management technologies in on-demand intelligent selection and solves the problem that existing LEO communication network satellite routing algorithms typically rely on preset strategies or fixed algorithms, failing to generate routing schemes on demand based on users' real-time service requirements.

[0045] The core inventive points of the intelligent agent constructed in this application mainly include three aspects: First, it constructs a closed-loop intelligent agent architecture for low-Earth orbit satellite network management, which integrates planning and decision-making, perception, and execution. The planning and decision-making module analyzes user service requirements and coordinates the operation of the perception and execution modules. The perception module obtains the current network topology status, and the execution module calls satellite access tools and service-driven routing algorithms. Second, it improves the model's ability to analyze the requirements of low-Earth orbit satellite network management based on a large language model and combined with prompting engineering access to a domain database containing satellite service data requirements and technical specifications, thereby achieving location identification, latitude and longitude analysis, and service requirement understanding and routing algorithm matching. Third, it adopts an on-demand intelligent routing selection mechanism, models the low-Earth orbit satellite network as a graph structure, where nodes represent satellites and edges represent inter-satellite links, and dynamically updates link status information. Different edge weights are calculated for latency-sensitive services, bandwidth-sensitive services, and best-effort services. The optimal transmission path is generated through a weighted Dijkstra algorithm, realizing a differentiated routing strategy based on service type.

[0046] Compared with existing technologies, the advantages and positive effects of this application are reflected in the following aspects: It achieves autonomous operation and intelligent collaboration of the system under unmanned control through a closed-loop intelligent agent architecture, significantly improving the automation and real-time performance of network routing; it realizes intelligent mapping of user semantic needs to network tasks based on a large language model, accurately extracting start and end positions and identifying service types, automatically matching the optimal routing algorithm, and reducing the complexity of manual configuration; and it achieves differentiated service guarantees by dynamically selecting the optimal access node and routing path based on real-time network topology and service requirements through on-demand intelligent routing selection.

[0047] The technical solution of this application and how it solves the aforementioned technical problems are described in detail below with specific embodiments. The listed specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0048] See Figure 1 The diagram shows the architecture of an on-demand routing agent system for low-Earth orbit satellite networks. This agent system includes:

[0049] The planning and decision-making module 11 includes a pre-trained large model 111 and a tool invocation module 112. The large model is used to parse user request text to obtain communication request parameters. The user request text is the user's communication request described in natural language. The communication request parameters include: the latitude and longitude coordinates of the origin and destination users, the QoS parameter type, and the corresponding QoS parameter value. The tool invocation module is used to invoke satellite access tools and routing tools.

[0050] The network sensing module 12 is used to dynamically sense the time-varying network status of the low-orbit satellite network and output a network topology map.

[0051] The execution module 13 includes a satellite access tool 131 and a routing tool 132. The satellite access tool is used to determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the originating and ending users by combining the network topology map. The routing tool is used to determine the routing path between the uplink and downlink access satellites based on the network topology map and the uplink and downlink access satellites, with QoS parameter values ​​as constraints, and based on the routing calculation strategy corresponding to the QoS parameter type.

[0052] The core architecture of this system consists of a planning and decision-making module, a network perception module, and an execution module. These modules work together to achieve the entire process of transforming natural language requirements into routing paths.

[0053] The pre-trained large-scale model in the planning and decision-making module possesses semantic understanding capabilities in the field of low-Earth orbit satellite communication. It can receive communication request texts described by users in natural language. For example, if a user inputs "high-definition video transmission from Shanghai to London," the large-scale model will accurately parse the communication request parameters. When parsing the latitude and longitude coordinates of the origin and destination users, the large-scale model can combine a pre-set geographic information database to convert locations such as "Shanghai" and "London" into corresponding latitude and longitude values, forming the origin and destination user latitude and longitude coordinates. When identifying the QoS parameter type and its corresponding QoS parameter value, the large-scale model first determines that the communication service type is bandwidth-sensitive, and then matches the corresponding QoS parameter type as throughput, with its parameter value set to no less than 200Mbps. The tool invocation module then, based on the parsing results of the large-scale model, initiates invocation commands for satellite access tools and routing selection tools according to pre-set tool invocation rules.

[0054] The network awareness module is responsible for dynamically capturing the time-varying network state of the low-Earth orbit satellite network. For time slot t, it first acquires the ephemeris data of the current mission moment, parses it to obtain the latitude and longitude coordinates of each satellite node, and converts them into three-dimensional rectangular coordinates in the Earth-Centered Earth-Fixed (ECEF) system. Subsequently, according to the Walker-Delta constellation topology rules, it constructs the adjacency relationships between satellite nodes, calculates the inter-satellite link distance and link rate, and collects the load status and link duration of each satellite node. Finally, it generates and outputs a network topology map containing satellite node status and inter-satellite link information. The network awareness module can update independently according to a preset update cycle, for example, updating the network topology map at a cycle of 100ms.

[0055] After receiving the network topology map and the latitude and longitude coordinates of the originating and ending users, the satellite access tool in the execution module first performs spatial geometric modeling of the user's geographical location and the instantaneous position of the satellites. It calculates basic parameters such as the center angle, elevation angle, and satellite-to-ground distance between the user and each satellite, filters out the set of visible satellites, and then determines the suitable uplink and downlink access satellites from the set. The routing tool, based on the network topology map and the uplink and downlink access satellites, uses QoS parameter values ​​as hard constraints. For example, when the QoS parameter type is latency and the parameter value is ≤150ms, the corresponding routing calculation strategy is used to search for transmission paths that meet the constraints in the network topology map, and finally determines the routing path between the uplink and downlink access satellites.

[0056] The intelligent agent system provided in this application constructs a complete technical link from user natural language requirements to route path generation through the collaborative work of three major modules. The large model of the planning and decision-making module achieves accurate conversion of natural language into structured communication requirement parameters, eliminating the need for manual intervention and standardized parameter input, thus significantly reducing operational complexity. The network perception module dynamically generates a network topology map, providing real-time and accurate network environment data support for subsequent satellite selection and route calculation, ensuring the timeliness and reliability of decision-making. Based on the parsing results and network status, the execution module accurately matches the access satellite with the route path, achieving deep adaptation between route planning and service requirements. This effectively improves the intelligence and adaptability of low-Earth orbit satellite network routing selection, meeting the needs of efficient response to diverse communication demands in dynamic network environments.

[0057] In some embodiments of this application, the satellite access tool is further specifically used to generate multiple sets of uplink and downlink access satellites that correspond one-to-one according to various access strategies, and the routing selection tool is further specifically used to calculate the routing path between each set of uplink and downlink access satellites based on the routing calculation strategy, so as to obtain multiple routing paths that correspond one-to-one with the multiple sets of uplink and downlink access satellites, as well as the communication performance parameter value of each routing path; accordingly, the large model is also used to select the optimal routing path that matches the communication performance parameter value and the QoS parameter value among the multiple routing paths.

[0058] When determining uplink and downlink access satellites, the satellite access tool can employ multiple access strategies in parallel calculations to generate multiple sets of uplink and downlink access satellites. Access strategies include, but are not limited to: maximum elevation angle access strategy, minimum load access strategy, random access strategy, and multi-parameter access strategy. Under the maximum elevation angle access strategy, the satellite with the highest elevation angle is selected from the set of visible satellites as the access satellite to improve link gain; the minimum load access strategy prioritizes the satellite with the lowest current data packet load to avoid overloading hotspot satellites; the random access strategy randomly selects from the visible satellites for performance comparison benchmarks; and the multi-parameter access strategy selects based on a comprehensive consideration of multiple dimensions. Each strategy is calculated independently, outputting a separate set of uplink and downlink access satellites, ultimately forming multiple independent sets of access satellite combinations.

[0059] The routing tool calculates paths one by one for each uplink / downlink access satellite combination using a routing calculation strategy corresponding to the QoS parameter type. For example, when the QoS parameter type is throughput, the routing calculation strategy is a weighted Dijkstra algorithm based on minimum load. The tool combines information such as the load status and link rate of satellite nodes in the network topology diagram to calculate a unique routing path for each access satellite combination. During the calculation process, the tool simultaneously derives the communication performance parameter values ​​for each path, including latency, throughput, and inter-satellite hop count, ensuring that each routing path has corresponding quantified performance data. Ultimately, this forms multiple sets of correspondence data between "uplink / downlink access satellite combination - routing path - communication performance parameter value".

[0060] After receiving multiple sets of routing paths and their corresponding communication performance parameter values, the large model first extracts the parsed QoS parameter types and values, and then uses these as a benchmark to filter all routing paths. For example, if the QoS parameter type is inter-satellite hop count and the parameter value is the lowest, the large model compares the inter-satellite hop count of all paths and selects the path with the fewest hops. When the QoS parameter type is latency and the parameter value is ≤150ms, paths with latency exceeding 150ms are first eliminated, and then the path with the optimal latency is selected from the remaining paths. Finally, the optimal routing path that matches the communication performance parameter value with the QoS parameter value is determined.

[0061] Employing multiple access strategies allows for a more comprehensive consideration of satellite access combinations, avoiding the limitations of a single strategy and providing diverse starting points for subsequent routing calculations. The routing tool independently calculates paths and outputs communication performance parameters for each satellite access combination, forming a rich library of alternative routing schemes. The large-scale model filters for optimal paths based on QoS parameter values, ensuring that the selected paths accurately match core service requirements, satisfying performance constraints while achieving optimized path selection. This multi-strategy, multi-path parallel computation and precise filtering mechanism enhances the diversity and adaptability of routing schemes, effectively addressing the dynamic characteristics of low-Earth orbit satellite networks and guaranteeing communication service quality in different service scenarios.

[0062] In some embodiments of this application, multiple access strategies include: a multi-parameter access strategy; the satellite access tool generates multiple sets of uplink and downlink access satellites corresponding to each other according to the multi-parameter access strategy, specifically including: obtaining multi-dimensional evaluation parameters of multiple sets of candidate satellite combinations; quantifying the information contribution rate of each evaluation parameter according to the information entropy model, and dynamically assigning corresponding weights to each evaluation parameter based on the information contribution rate; calculating the multi-parameter weighted evaluation value of each set of candidate satellite combinations according to the weights assigned to each evaluation parameter; and selecting the candidate satellite combination with the largest multi-parameter weighted evaluation value as a set of uplink and downlink access satellites corresponding to the multi-parameter access strategy.

[0063] When a satellite access tool uses a multi-parameter access strategy to generate uplink and downlink access satellites, it first acquires multi-dimensional evaluation parameters for multiple candidate satellite combinations. These candidate satellite combinations can be any two satellites from a low-Earth orbit (LEO) satellite network, or any two satellites from a selected set within the LEO satellite network. For example, the selection criteria could include whether the satellite is currently available or whether its load is excessive. The multi-dimensional evaluation parameters include, but are not limited to, key indicators such as satellite-to-ground link duration, satellite node load, and total inter-satellite hop count. For each candidate satellite combination, specific values ​​for each dimension of the evaluation parameters are collected. For example, a candidate satellite combination might have a satellite-to-ground link duration of 120 seconds, a satellite node load of 80, and a total inter-satellite hop count of 3, forming a complete dataset of evaluation parameters.

[0064] Subsequently, the information contribution rate of each evaluation parameter is quantified using the information entropy model. First, a probability matrix of the indicators is constructed to measure the relative contribution of each indicator in the same dimension. Then, the information utility value of each evaluation parameter is calculated using the information entropy formula. A higher information utility value indicates a more significant role for the indicator in distinguishing different satellite combinations. Based on the calculated information contribution rate, weights are dynamically assigned to each evaluation parameter. For example, if the information contribution rate of the satellite-to-ground link duration is 0.4, the satellite node load is 0.3, and the total inter-satellite hops are 0.3, then the corresponding weights are set to 0.4, 0.3, and 0.3 respectively, ensuring that the weight allocation matches the actual importance of the indicators.

[0065] Next, the evaluation parameters are normalized to unify the numerical ranges of indicators with different dimensions and directions. Then, the multi-parameter weighted evaluation value for each candidate satellite combination is calculated based on the assigned weights. The weighted evaluation value can be calculated by multiplying the normalized indicator values ​​by their corresponding weights and summing the results to obtain the comprehensive score for each combination. Finally, the weighted evaluation values ​​of all candidate satellite combinations are sorted, and the candidate satellite combination with the highest score is selected as the uplink and downlink access satellite combination corresponding to the multi-parameter access strategy.

[0066] The multi-parameter access strategy provided in this application avoids the one-sidedness of single-indicator decision-making by comprehensively evaluating multiple dimensions such as satellite-to-ground link duration, satellite load, and inter-satellite hop count, making the selection of access satellites more comprehensive and scientific. The dynamic weight allocation mechanism based on the information entropy model can adjust weights according to the actual information contribution rate of each indicator, adapting to the needs of different network states and service scenarios, offering greater flexibility compared to fixed weight allocation methods. By calculating weighted evaluation values ​​to select the optimal satellite combination, the comprehensive advantages of access satellites in multiple dimensions are ensured, providing a solid foundation for subsequent routing path optimization, effectively improving the rationality and reliability of access satellite selection, and thus guaranteeing the transmission performance of the entire communication link.

[0067] In some embodiments of this application, the planning decision module further includes: a prompting engineering module 113, used to generate prompt text and input it into the large model based on the user's request text and preset prompt words; the prompt text is used to guide the large model to perform the following operations:

[0068] The system parses the latitude and longitude coordinates of the origin and destination users based on the user request text and context information, and inputs these coordinates into the satellite access tool. It also identifies the communication service type based on the user request text, matches the corresponding QoS parameter type and QoS parameter value in the preset service database based on the communication service type, and inputs the QoS parameter type and QoS parameter value into the routing tool.

[0069] In the optional implementation described above, the planning and decision-making module further includes a prompting engineering module. This module generates more refined prompt text to guide the large model's parsing process. The prompting engineering module can pre-store preset prompt words, which clarify the large model's task flow, parsing rules, and output format. Upon receiving the user's request text, the prompting engineering module can concatenate and integrate the user's request text with the preset prompt words to generate a complete prompt text. For example, if the user's request text is "emergency communication from Beijing to Moscow," the preset prompt words could include location parsing rules, service type identification standards, etc. After concatenation, a complete prompt text guiding the large model's parsing is formed and input into the large model.

[0070] Specifically, the prompt text first guides the large model to parse the latitude and longitude coordinates of the originating and ending users. Combining this with contextual information, the large model identifies the communication origin and ending locations "Beijing" and "Moscow" in the user's request text, retrieves geographic information data from the preset service database, and converts the locations into corresponding latitude and longitude coordinates. Subsequently, it inputs these coordinates into the satellite access tool according to a preset data format, providing data support for satellite access selection. Simultaneously (which can be relatively independent of parsing the originating and ending user latitude and longitude coordinates), the prompt text can also guide the large model to identify the communication service type. By analyzing the keyword "emergency communication" in the user's request text, the large model determines the service type to be "latency-sensitive." Next, it searches the preset service database for the corresponding QoS parameter type and value, determining the QoS parameter type to be latency, with a corresponding value of ≤100ms. Finally, the large model inputs the QoS parameter type and value into the routing tool according to the tool's required format, providing the basis for selecting the routing calculation strategy and the data required for calculation.

[0071] The prompting module provides a clear framework for the large model's parsing process, indicating the parsing flow, invocation process, and the format of the output data. This ensures the accuracy and consistency of the large model's parsing results and avoids parameter extraction errors caused by ambiguous parsing rules. Guided by the prompting text, the large model can accurately complete the parsing of the start and end user's latitude and longitude coordinates and the determination of QoS parameters, achieving efficient conversion of user natural language requirements into structured tool input parameters. The application of the preset service database provides a standardized basis for matching service types and QoS parameters, ensuring accurate adaptation between routing calculation strategies and service requirements. This further enhances the intelligence and standardization of the system's routing planning and reduces the impact of human factors on the parsing results.

[0072] In some embodiments of this application, the large model is also used to output structured thought analysis text before calling the satellite access tool or routing tool. The structure of the thought analysis text is set by the prompting engineering module. The thought analysis text is used to describe the parameter basis, constraints and reasoning process of the tool call.

[0073] After receiving the prompt text and parsing the communication requirement parameters, the large model does not directly invoke the tool, but first outputs a structured thought analysis text. The structure of the thought analysis text is pre-set by the prompt engineering module, including three core parts: parameter basis, constraints, and reasoning process, ensuring that the text is logically clear and information is complete.

[0074] Regarding the basis of the parameters, the analysis text of the large model can explain in detail the source of each communication requirement parameter. For example, the latitude and longitude coordinates of the start and end users are obtained by querying the geographic information in the preset service database based on the two locations "Guangzhou" and "Dubai" in the user requirement text; the QoS parameter type and parameter value are determined by matching the corresponding standard parameters in the database based on the identified "remote sensing data transmission" service type.

[0075] Regarding constraints, the analysis text of the large model can list the limiting factors that need to be followed during the tool call process. For example, hard constraints on QoS parameter values, such as throughput ≥ 300Mbps, and implicit constraints such as satellite node status and link rate in the network topology diagram, while explaining the specific impact of these constraints on the tool execution.

[0076] Regarding the reasoning process, the analysis text of the large model can explain the logical deduction chain from demand analysis to tool invocation. For example, based on the service characteristics of "remote sensing data transmission", it is determined that it has high bandwidth requirements. Therefore, the QoS parameter type is determined to be throughput, and then the corresponding routing calculation strategy is selected. At the same time, it explains why this group of uplink and downlink access satellites was selected, as well as the correlation between the parameters, to ensure that the entire thinking process is traceable and verifiable.

[0077] The structured analysis of the output text makes the tool-calling decision-making process of large models more transparent and traceable, avoiding blind decision-making. The parameter basis demonstrates the rationality and accuracy of the tool's input parameters, and the constraints provide clear boundary limits for tool execution, reducing invalid calculations. The explanation of the reasoning process clarifies the logical connection between needs and decisions, facilitating subsequent solution optimization and problem identification. This transparent decision-making mechanism improves the system's reliability and maintainability, enhances the credibility of routing planning results, provides additional assurance for the accuracy of low-Earth orbit satellite network routing selection, and increases user trust in the system.

[0078] In some embodiments of this application, the planning decision module further includes a session memory unit 114, which is used to store user requirement text, parsing results of large models, call records of tool calling modules and routing paths; the prompt engineering module is also used to generate context information based on the stored information of the session memory unit.

[0079] The planning and decision-making module may also include a session memory unit, which has data storage and retrieval functions. The stored content includes user request text, large model parsing results, tool call records, and routing paths. User request text includes, but is not limited to: natural language communication requests entered by the user each time; parsing results include, but are not limited to: structured data such as the start and end user latitude and longitude coordinates, QoS parameter types, and parameter values ​​generated each time; tool call records include, but are not limited to: the type of tool called, the call time, and input and output parameters; and routing paths include the optimal path calculated each time and the corresponding communication performance parameter values.

[0080] The session memory unit can employ a structured storage method to classify and manage different types of data, facilitating rapid querying and retrieval. When a new user request text is entered, the prompting engineering module can retrieve relevant historical data from the session memory unit, such as previous communication requests from the same user, parsing results for similar service types, and routing schemes, integrating this historical data into contextual information. The prompting engineering module generates more targeted prompt text by combining the new user request text with contextual information and preset prompt words. For example, when a user makes a similar video conferencing communication request again, the prompting engineering module retrieves information from the session memory unit, such as the user's previous latency parameter preferences and historical satellite access selections, and incorporates it into the prompt text, guiding the large model to refer to historical data during the parsing process, improving parsing efficiency and accuracy. It can be understood that the session memory unit may include the storage module itself, or it may not include a storage module but instead communicate with an external storage module for read and write operations to achieve its access functionality.

[0081] The session memory unit effectively accumulates and reuses historical service information by storing various types of data. The contextual information generated by the prompting engineering module based on historical data makes the parsing process of large models more targeted, quickly adapting to user habits and service preferences, improving parsing efficiency and parameter matching accuracy. Simultaneously, the reuse of historical data provides a reference for routing planning in similar service scenarios, reducing redundant calculations and improving system response speed. This historical data-driven optimization mechanism further enhances the system's intelligence and user adaptability, ensuring continuous performance optimization during long-term use.

[0082] In some embodiments of this application, the QoS parameter types correspond one-to-one with the routing calculation strategies. If the QoS parameter type is latency, the routing calculation strategy is a weighted Dijkstra algorithm based on the shortest latency, and the edge weight of the algorithm is the sum of the inter-satellite propagation latency and the transmission latency; if the QoS parameter type is throughput, the routing calculation strategy is a weighted Dijkstra algorithm based on the minimum load, and the edge weight of the algorithm is the standard deviation of the satellite node's neighbor queue length; if the QoS parameter type is inter-satellite hop count, the routing calculation strategy is a weighted Dijkstra algorithm based on the minimum hop count, and the edge weight of the algorithm is fixed at 1.

[0083] There is a strict one-to-one correspondence between QoS parameter types and routing calculation strategies; each QoS parameter type is matched with a specific routing calculation strategy and algorithm rules. When the QoS parameter type is latency, the corresponding routing calculation strategy is the weighted Dijkstra algorithm based on the shortest latency. The edge weights of this algorithm consist of two parts: inter-satellite propagation delay and transmission delay. The inter-satellite propagation delay is calculated by dividing the inter-satellite distance by the speed of light, while the transmission delay is determined by dividing the data packet size by the link rate. The algorithm searches for the transmission path with the minimum total latency by accumulating the weights of each edge on the path.

[0084] When the QoS parameter type is throughput, the routing calculation strategy is a weighted Dijkstra algorithm based on minimum load. The edge weights of the algorithm are set to the standard deviation of the satellite node's neighbor queue length. By calculating the dispersion of the satellite node's neighbor queue length, the load balancing status of the satellite is quantified. The smaller the weight, the more balanced the load. The algorithm prioritizes links with smaller weight values ​​to build paths, so as to ensure high throughput of the path.

[0085] When the QoS parameter type is inter-satellite hop count, the corresponding routing calculation strategy is the weighted Dijkstra algorithm based on the minimum hop count. This algorithm fixes the weight of each edge to 1, and the total weight of the path is equal to the inter-satellite hop count in the path. The algorithm searches for the path with the minimum total weight to achieve the routing plan with the fewest inter-satellite hops, simplifying the transmission process and reducing the risk of packet loss.

[0086] The one-to-one correspondence between QoS parameter types and routing calculation strategies enables precise binding between routing calculations and service requirements. Differentiated algorithms and weight designs are employed for different QoS parameter types, ensuring that routing paths accurately match the core performance requirements of services. Algorithms based on minimum latency guarantee the real-time performance of latency-sensitive services, algorithms based on minimum load meet the high throughput requirements of bandwidth-sensitive services, and algorithms based on minimum hop count improve the transmission reliability of best-effort services. This precise matching mechanism effectively improves the performance adaptability of routing paths, ensures the communication quality of various services in low-Earth orbit satellite networks, and enhances the system's responsiveness to diverse service demands.

[0087] Corresponding to the on-demand routing intelligent agent system for low-Earth orbit satellite networks provided in the embodiments of this application, the embodiments of this application also provide an on-demand route generation method for low-Earth orbit satellite networks. For example, this method can be... Figure 1 The planning and decision-making module of the system shown is executed.

[0088] See Figure 2 The flowchart shown illustrates an on-demand route generation method for low-Earth orbit satellite networks. This method specifically includes the following steps:

[0089] Step 201, obtain the user request text; the user request text is the user's communication request described in natural language;

[0090] Step 202: Use the pre-trained large model to parse the user request text and obtain the communication request parameters. The communication request parameters include: the latitude and longitude coordinates of the start and end users, the QoS parameter type and the corresponding QoS parameter value. Before parsing the user request text, the user request text can be integrated with system prompt words, and the integrated prompt text can be input into the large model.

[0091] Step 203: Based on the network topology map of the low-Earth orbit satellite network, determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the originating and ending users;

[0092] Step 204: Based on the network topology and uplink / downlink access satellites, and using QoS parameter values ​​as constraints, determine the routing path between uplink / downlink access satellites according to the routing calculation strategy corresponding to the QoS parameter type.

[0093] The method provided in this embodiment can be applied to the aforementioned routing agent system.

[0094] First, the system performs the user request text acquisition step. It receives the user's communication request described in natural language through the interactive interface. These requests can be for various scenarios such as video conferencing, data transmission, and emergency communication. For example, the user can input "high-definition video conferencing from Chengdu to Sydney" or "remote sensing data transmission from Urumqi to Frankfurt". The system stores and processes these inputs as raw request text.

[0095] Next, the communication requirement parameter parsing step begins. The pre-trained large model performs semantic analysis on the acquired user requirement text. When parsing the latitude and longitude coordinates of the start and end users, the large model identifies the communication start and end locations in the text, calls a pre-set geographic information database, and converts the locations into corresponding latitude and longitude values. For example, "Chengdu" is parsed as 30.67°N, 104.06°E, and "Sydney" as 33.87°S, 151.21°E. When identifying QoS parameter types and values, the large model first determines the service type, such as high-definition video conferencing being latency-sensitive and remote sensing data transmission being bandwidth-sensitive. Then, it matches the corresponding QoS parameter type and value. For example, the QoS parameter type for latency-sensitive services is latency, with a value of ≤150ms, while the QoS parameter type for bandwidth-sensitive services is throughput, with a value of ≥250Mbps.

[0096] Subsequently, the uplink and downlink access satellite determination step is performed. The system calls the network topology map generated by the network perception module, and combines the latitude and longitude coordinates of the origin and destination users obtained by parsing to perform spatial geometric calculations on the user's geographical location and satellite position, filter out the set of visible satellites, and then select the appropriate uplink and downlink access satellites through the access strategy. For example, the satellite with the highest elevation angle is selected as the uplink and downlink access satellite based on the maximum elevation angle strategy.

[0097] Finally, the routing path determination step is executed. Based on the network topology map and uplink / downlink access satellites, the system uses QoS parameter values ​​as constraints and calls the routing calculation strategy corresponding to the QoS parameter type. For example, when the QoS parameter type is delay, the weighted Dijkstra algorithm based on the shortest delay is used to search for transmission paths that meet the delay constraints in the network topology map, and finally determine the routing path between uplink / downlink access satellites.

[0098] The method provided in this application follows a logical flow of "demand acquisition - parameter parsing - satellite access determination - route path generation," achieving fully automated processing from user natural language requirements to route paths. The demand parsing step accurately extracts communication requirement parameters using a large model, eliminating the need for manual intervention and improving the efficiency and accuracy of parameter acquisition. The determination of uplink and downlink access satellites is based on a real-time network topology map, ensuring the adaptability and availability of access satellites. Route path generation ensures that the path meets the core performance requirements of the service through precise matching of QoS parameters and routing calculation strategies. The entire method is concise and efficient, adapting to the dynamic characteristics and diverse service requirements of low-Earth orbit satellite networks, effectively improving the intelligence and adaptability of route generation, and providing reliable methodological support for efficient communication in low-Earth orbit satellite networks.

[0099] It is understood that the specific implementation methods and corresponding technical effects of the methods provided in the embodiments of this application can be referred to the relevant descriptions of the systems provided in the embodiments of this application, and will not be repeated here.

[0100] refer to Figure 3 As an example of a routing agent system for low-Earth orbit satellite networks in a specific application scenario, the following section combines... Figure 3 This example explains the intelligent agent system. Using the emergency communication requirement of "a fire occurs in location A, and remote sensing images of the disaster area need to be transmitted back to the disaster relief command center in location B in real time" as an example, this embodiment fully presents the system's workflow and technical details.

[0101] (1) Initialization of the intelligent agent system

[0102] After the agent system starts, it first loads the environment variable file, obtains the path to the MCP (Model Context Protocol) tool server script, and establishes a standard input / output channel through the Stdio Server Parameters configuration class (the parameter configuration class for the standard input / output communication service, used to define the core configuration items when stdio_client establishes a connection with the satellite network tool library, and is the basic set of parameters supporting the initialization of MCP protocol communication). Subsequently, stdio_client and ClientSession complete the initial connection, realize the registration and handshake with the satellite network tool library, and the agent system enters the ready state.

[0103] In this embodiment, the low-Earth orbit satellite communication intelligent agent system uses the Walker-Delta constellation model to model the satellite distribution, assuming the constellation consists of... It consists of several orbital planes, each evenly distributed. The difference in right ascension between the ascending nodes of two satellites in adjacent orbital planes is 1. The orbital phase difference is ,in This is the phase factor. The satellite's orbital altitude is H=550km, and the orbital inclination is... In the intelligent agent system requesting a time slot Internally, the network sensing module models the network topology of the low-Earth orbit satellite communication agent system as an undirected graph. , Indicates in time slot The set of satellite nodes within, Indicates in time slot The set of effective links within the network. Satellite nodes establish communication connections through inter-satellite links (ISLs). Each satellite node can establish 4 ISLs with its 4 surrounding satellites (including 2 satellites in the same orbit and 2 satellites in adjacent orbits). The edge parameters include inter-satellite distances. and link rate .

[0104] (2) Planning and Decision Module

[0105] When a user's natural language communication request is received, the planning and decision-making module integrates the request with historical dialogues, execution results, and thought processes to form complete contextual information. This contextual information is then concatenated with the agent system's prompts and input into the LLM. The agent system's prompts clearly define the agent system's identity, task flow, thought analysis specifications, tool invocation rules, and operation output format. The intelligent agent system's prompts guide the LLM to first identify the origin of the communication request and parse its latitude and longitude coordinates. These coordinates are then input into the satellite node access tool (Satellite_access) in a standardized format. Satellite_access outputs the uplink and downlink access nodes. Next, the system analyzes the communication service type in the user request, such as video conferencing or remote sensing data transmission, and maps corresponding key QoS parameters, including latency and throughput, based on the service type. Following the prompts, the system identifies the routing algorithm type required for the service, including latency-sensitive, bandwidth-sensitive, and best-effort types. The LLM then uses the QoS parameters and routing algorithm type as input to call the routing node selection tool (Route_selection), generating multiple compliant communication paths and their performance metrics. After the tool completes execution, the intelligent agent system matches the access policy returned by Satellite_access with the path results from Route_selection, integrating them into a complete end-to-end communication scheme. Each scheme includes the ingress satellite, egress satellite, path satellite sequence, and corresponding latency and throughput metrics. Before each tool call, the LLM must output... <think>The analysis paragraphs explain the basis for the selected parameters, possible constraints, and the reasoning process. The intelligent agent system strictly adheres to a structured format in its output and stores operation records and dialogue history for reuse and reference in subsequent user requests, enabling continuous parsing and dynamic optimization of user needs. Based on a large language model and combined with access to a domain database containing satellite service data requirements and technical specifications, the intelligent agent system can parse user service requirements and select routing algorithm types consistent with those requirements, further enhancing the model's ability to parse requirements within the low-Earth orbit satellite network management domain.

[0106] In this embodiment, when a user inputs the natural language request "A fire has occurred in location A, and it is necessary to communicate with the disaster relief command center in location B in real time to request rescue teams", the prompting engineering module in the planning and decision-making module integrates the request with preset prompt words and historical information stored in the conversation memory unit to generate a complete prompt text that is input into the pre-trained large model.

[0107] Next, the preset prompts will guide the large model to perform the following operations: First, identify the communication start and end points "location A" and "location B", call the geographic information in the preset service database, and parse it into the corresponding latitude and longitude coordinates, i.e., Input str:(lat1,lon1);(lat2,lon2), where lat1 and lon1 are the latitude and longitude of location A, and lat2 and lon2 are the latitude and longitude of location B; then analyze the service type. "Real-time call" is a latency-sensitive service, so the corresponding QoS parameter type is mapped to latency, and the QoS parameter value is set to Input str: latency=X (X is the latency threshold to meet real-time transmission, X=150ms in this embodiment), and the corresponding routing algorithm type is matched as the weighted Dijkstra algorithm based on the shortest latency.

[0108] Before calling the satellite access tool and routing tool, the large model outputs a structured thinking analysis text, clarifying that the basis for the analysis of latitude and longitude parameters is the matching result of the geographic information database, the constraint condition of QoS parameters is the latency requirement of real-time calls, and the reasoning process is "real-time calls need to prioritize low latency, so a latency-sensitive routing algorithm is selected".

[0109] The planning and decision-making module sends perception commands to the perception module, which then provides the perception results, including a network topology diagram.

[0110] (3) Perception module

[0111] The sensing module acquires the ephemeris data for the current mission time t, and reads the latitude and longitude coordinates of all satellites at that time based on the pre-calculated satellite orbit data in the ephemeris file. And it is mapped to three-dimensional rectangular coordinates in the Earth-centered Earth-fixed system (ECEF) using coordinate transformation functions. The specific calculation method is as follows:

[0112] (1)

[0113] (2)

[0114] (3)

[0115] (4)

[0116] in For the Earth's radius, This represents the orbital altitude. After coordinate transformation, the intelligent agent system obtains the set consisting of all satellite nodes. .

[0117] The sensing module constructs the adjacency relationships between satellites according to the Walker-Delta constellation topology. For any satellite Its neighbor set consists of its preceding and following orbital neighbors, adjacent orbital neighbors, and cross-orbit offset neighbors. For any adjacent satellite... and The intelligent agent system calculates inter-satellite distances based on its three-dimensional coordinates:

[0118] (5)

[0119] Based on physical distance, the sensing module further calculates the link signal-to-noise ratio (SNR) using the free-space path loss model and the Shannon formula to calculate the actual achievable link rate. The calculation method is as follows:

[0120] (6)

[0121] (7)

[0122] in For carrier frequency, At the speed of light, For channel bandwidth, For transmission power, , These are the transmit and receive antenna gains, respectively. For noise power spectral density, Here, represents the Boltzmann constant. The link rate varies periodically over time; as the satellite orbits move, the distance between satellites and the line-of-sight relationship change dynamically, resulting in a distinct time-varying characteristic for the link status. Ultimately, the network state awareness module generates a set of satellite links in the following structured manner:

[0123] (8)

[0124] Based on this, the sensing module constructs a time-varying satellite network topology:

[0125] (9)

[0126] Next, the perception module can feed the perception results back to the planning and decision-making module, which will then perform standardization and other processes to adapt to the input format of the satellite access tool. Alternatively, the perception module can directly generate a network topology diagram that conforms to the input format of the satellite access tool; this embodiment does not specifically limit this approach.

[0127] (4) Satellite access tools Figure 3 (Satellite node access in the execution module)

[0128] The satellite access tool in the execution module receives latitude and longitude parameters (lat1, lon1); (lat2, lon2) and network topology diagrams provided by the planning and decision module. First, spatial geometric modeling is performed on the user's geographical location and the instantaneous satellite position.

[0129] Known satellites exist The latitude and longitude coordinates at that time are The user's latitude and longitude coordinates are The calculated geocentric angle between the user and the satellite is as follows:

[0130] (10)

[0131] This reflects the angle between the user and the satellite in the geocentric coordinate system, and the satellite-to-ground geometric elevation angle is further obtained as follows:

[0132] (11)

[0133] In satisfying ( Under the condition that the satellite is visible, the satellite-to-ground link visibility time can be obtained from the ephemeris data. For the current moment and The time difference is significant for users. , The respective candidate access satellite sets are as follows:

[0134] (12)

[0135] In addition, for subsequent link budgeting, this embodiment provides a formula for calculating the satellite-to-ground slant range (distance from the user to the satellite):

[0136] (13)

[0137] To characterize the inter-satellite routing cost, this embodiment further defines a minimum inter-satellite hop count model based on a Walker-Delta constellation structure. The number of constellation orbital planes is... The number of satellites in each orbit is Phase factor is The longitude interval between the ascending nodes of adjacent orbits is:

[0138] (14)

[0139] The phase interval between adjacent satellites in the same orbit is:

[0140] (15)

[0141] The relative phase offset between adjacent satellites in adjacent orbits is:

[0142] (16)

[0143] By performing spherical geometric derivation of the instantaneous latitude, longitude, and ascent / descent orbit states of the satellites, the longitude of the ascending node of the orbits of the two satellites can be obtained. , and orbital phase , If it is a satellite in ascending orbit:

[0144] (17)

[0145] (18)

[0146] If it is a de-orbiting satellite:

[0147] (19)

[0148] (20)

[0149] in For the satellite's latitude and longitude, Let be the orbital inclination angle, then the longitude difference between the ascending intersection points of the two orbits is:

[0150] (twenty one)

[0151] in This indicates that the angle is normalized to... Inter-satellite horizontal jump count. With the number of track jumps It can be written as:

[0152] (twenty two)

[0153] (twenty three)

[0154] in This represents the rounding operation to the nearest integer. Therefore, the minimum total inter-satellite hop count between two access satellites can be defined as follows: Based on this, this embodiment introduces three types of satellite node access strategies and reconstructs a multi-parameter access strategy.

[0155] ① Maximum elevation angle access strategy

[0156] The maximum elevation angle strategy aims to prioritize access satellites with the highest elevation angle to improve link gain and reduce obstruction risk. For users... In its candidate set of visible satellites Internally, the following optimization criteria are used to select access satellites:

[0157] (twenty four)

[0158] This allows us to obtain user information. Maximum elevation angle access satellite index And calculate the corresponding satellite-to-ground distance. And the number of horizontal hops between two access satellites With the number of track jumps .

[0159] ② Minimum load access strategy

[0160] To balance inter-satellite load and avoid overloading of hotspot satellites, this embodiment performs the following during constellation initialization: Each satellite randomly generates its data packet payload. The arrival process of service flows on satellites is described using the Poisson model:

[0161] (25)

[0162] in For the first Average arrival rate of satellites The maximum average arrival rate is given to the agent system.

[0163] Based on this, the minimum load access strategy, while ensuring elevation angle constraints, prioritizes selecting the satellite with the lowest current data packet load from the candidate visible set. For users... Its access satellite index is:

[0164] (26)

[0165] Selected Further used to calculate the corresponding link duration Distance between stars and Earth Inter-star horizontal jump count Number of jumps along the track and total number of jumps This ensures coverage while effectively reducing the risk of further congestion from heavily loaded satellites.

[0166] ③ Random access strategy

[0167] To provide a benchmark performance comparison, this embodiment introduces a random access strategy. In this strategy, the user randomly selects an access satellite with equal probability from only a candidate set that satisfies the elevation angle constraint, regardless of elevation angle and load differences. For the user... Its satellite access follows the following:

[0168] (27)

[0169] in For users Number of candidate satellites. Satellite-to-ground distance output by the random policy. Link duration And interstellar jumps , , As a benchmark, it is used to evaluate the benefits brought by other intelligent access strategies.

[0170] ④ Multi-parameter access strategy

[0171] To achieve a comprehensive trade-off among multiple metrics such as elevation angle, link duration, satellite load, and inter-satellite hop count, this embodiment constructs a multi-parameter access strategy. For a given time... ,user and The candidate sets are respectively and Choose any pair of candidate access satellite combinations. Its evaluation vector is defined as:

[0172] (28)

[0173] Among them, subscript For a specific combination of satellites, For the duration of the satellite-to-ground link, To access the current data packet payload of the satellite, This represents the total number of inter-satellite hops between the two access satellites.

[0174] To unify indicators with different dimensions and directions (benefits / costs), a normalization process based on extreme values ​​is introduced. Let the... The maximum and minimum values ​​of each indicator across all satellite combinations are respectively , For each combination With indicators Normalized value Defined as:

[0175] (29)

[0176] in, correspond , correspond , 4 and 4 respectively correspond to , , correspond .

[0177] Then construct the probability matrix of the index. This is used to measure the relative contribution of each indicator on the same indicator dimension:

[0178] (30)

[0179] in This represents the number of candidate satellite combinations.

[0180] Using the concept of information entropy, we define the first... Information utility value of each indicator for:

[0181] (31)

[0182] And further, the normalized weights of each indicator are obtained:

[0183] (32)

[0184] in, The larger the value, the higher the value. The higher the information contribution rate of each indicator in distinguishing different satellite combinations, the higher its corresponding weight. Also bigger.

[0185] Ultimately, the comprehensive evaluation value of each pair of candidate satellite combinations can be obtained. Defined as a weighted sum:

[0186] (33)

[0187] The satellite combination with the highest comprehensive evaluation value is selected as the access result of the multi-parameter access strategy.

[0188] (34)

[0189] Using the four access strategies described above, this embodiment outputs the user's latitude and longitude, the selected access satellite number, and the satellite-to-ground link duration for different service requirements. Distance between stars and Earth Angle of elevation Inter-star horizontal jump count Number of track jumps and access satellite instantaneous load These parameters provide a unified and scalable description of satellite node access for subsequent routing and transmission performance evaluation.

[0190] refer to Figure 3 In this embodiment, the satellite access tool ultimately outputs the uplink and downlink access satellites as access satellite 310 (access from location A) and access satellite 948 (access from location B), and simultaneously outputs parameters such as satellite-to-ground link duration, satellite-to-ground distance, elevation angle, inter-satellite hop count, and satellite load.

[0191] (5) Routing tools Figure 3 "On-demand intelligent routing selection" in the execution module

[0192] The routing tool in the execution module is responsible for generating the optimal transmission path from the source node to the target node for the task after determining the four sets of access satellite nodes provided by the satellite access tool. The inputs of the routing tool include the access node number, the target node location, the network topology status, and service type constraints. The final output is the path satellite nodes.

[0193] In this embodiment, the routing tool receives uplink and downlink access satellites (310, 948), network topology graph G(t), and QoS parameters (latency ≤ 150ms).

[0194] Based on the parsed communication service type, the routing tool can automatically select the corresponding routing algorithm: for latency-sensitive services with latency as the sensitive QoS parameter type, the Dijkstra algorithm based on the shortest latency is used; for bandwidth-sensitive services with bandwidth as the sensitive QoS parameter type, the Dijkstra algorithm based on the minimum load is used; and for best-effort services with hop count as the sensitive QoS parameter type, the Dijkstra algorithm based on the minimum hop count is used.

[0195] Intelligent agent system intelligently constructs satellite node diagram Let any edge The weight is The routing problem then becomes the problem in the graph. China seeks access to satellite To the target satellite The minimum cost path.

[0196] Let the path be The total path cost is:

[0197] (35)

[0198] in For the edge The weights vary depending on the routing strategy.

[0199] ① For best-effort routing algorithms, the edge weight is 1:

[0200] (36)

[0201] ② For delay-sensitive routes, the weight of an edge is the sum of its propagation delay and its transmission delay:

[0202] (37)

[0203] (38)

[0204] (39)

[0205] ③ For bandwidth-sensitive routes, obtain the set of neighbor queue lengths for each node as follows: Then the weight of the edge is the standard deviation of the queue length of that node:

[0206] (40)

[0207] (41)

[0208] in For satellite Neighboring satellites.

[0209] For the QoS parameter types that are sensitive to the service type after the user demand is parsed, the corresponding Dijkstra algorithm is selected and calculated with the weight of the corresponding algorithm to obtain the four path node sets calculated under the four access strategies, as well as the communication performance parameter values ​​(latency, throughput, inter-satellite hops).

[0210] The intelligent agent system ultimately integrates the four results and the analyzed service type, first outputting four routing selection results and returning them to the planning and decision module, which then outputs the optimal result to the user. Specifically, the planning and decision module's large model receives multiple routing paths and communication performance parameter values, filters out the path with latency ≤150ms and optimal performance as the final routing path, forming a complete end-to-end communication scheme that includes ingress satellite 310, egress satellite 948, path satellite sequence, latency, and throughput indicators, and outputs the routing selection results.

[0211] The routing agent system provided in this embodiment uses a planning and decision-making module as its core. Its main function is to transform user-initiated natural language service requests into executable satellite access and route planning operations. The main processes include: ① Service requirement analysis: When a user inputs, such as "I need to travel from Beijing to New York for a video conference," the module first generates agent-level prompts and combines the user input with the prompts to form a standardized LLM message sequence. The LLM performs semantic understanding on this sequence, extracting communication start and end points, service types, and key QoS parameters; ② Tool invocation and task decomposition: Based on the MCP protocol, the module establishes asynchronous communication with external tool services and loads available tools (Satellite_access, Route_selection). Subsequently, a directed task graph is constructed using LangGraph, which decomposes the parsed tasks into a series of ToolNode call sequences, clarifies the input-output dependencies of each tool node, and realizes the mapping from task to operation. In the LangGraph task graph, LLM acts as a policy node, dynamically selecting the order of subsequent tool calls based on the context state and execution feedback; ③ Result integration and memory update: The module integrates the satellite access scheme, path satellite sequence, latency and other information output by each tool node into a structured decision result, and writes the user input, thought process and execution result into conversation_history. In subsequent calls, historical information can be used as experience guidance.

[0212] The perception module is responsible for providing real-time network situational awareness support for planning and decision-making, and undertakes the task of dynamically monitoring the topology and link status of the low-Earth orbit satellite network. The module perceives two types of information: ① network topology and inter-satellite link information: link reachability, link bandwidth, and link distance; ② node status information: satellite node load and node location. Based on the perceived information, the network status is constructed into a graph. The perception results are fed back to the planning and decision-making module, providing accurate network environment input for strategy generation and feasible operational basis for the execution module.

[0213] The execution module maps the satellite access and routing policies generated by the planning and decision-making module into specific network operations, enabling the tasks to be implemented. The execution module receives structured data output from the planning and decision-making module, and based on the MCP protocol, sequentially inputs the data into the Satellite_access and Route_selection tools. Finally, it integrates the operation execution results into standardized feedback information for subsequent user requirement analysis and policy optimization reference, thereby improving the service quality and intelligence level of the low-Earth orbit satellite network.

[0214] Figure 4 This is a block diagram of an electronic device used to implement embodiments of this application. For example... Figure 4 As shown, the electronic device includes a memory 401 and a processor 402. The memory 401 stores a computer program that can run on the processor 402. When the processor 402 executes the computer program, it implements the method described in the above embodiments. The number of memories 401 and processors 402 can be one or more. In a specific implementation, the electronic device may also include a communication interface 403 for communicating with external devices and performing data exchange and transmission.

[0215] In practical implementation, if the memory 401, processor 402, and communication interface 403 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0216] Optionally, in a specific implementation, if the memory 401, processor 402 and communication interface 403 are integrated on a single chip, the memory 401, processor 402 and communication interface 403 can communicate with each other through an internal interface.

[0217] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method provided in this application.

[0218] This application provides a computer program product, including a computer program that, when executed by a processor, implements the method provided in this application.

[0219] This application also provides a chip including a processor for calling and executing instructions stored in a memory, causing a communication device with the chip installed to perform the method provided in this application.

[0220] This application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in the application embodiment.

[0221] It should be understood that the aforementioned processor can be a CPU (Central Processing Unit), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting Advanced Reduced Instruction Set Machines (ARM) architecture.

[0222] Further, optionally, the aforementioned memory may include read-only memory and random access memory. The memory may be volatile memory or non-volatile memory, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available. Examples include Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Sync Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0223] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0224] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0225] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0226] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0227] The logic and / or steps described in the flowchart or otherwise herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0228] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0229] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0230] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope described in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.< / think>

Claims

1. An on-demand routing intelligent agent system for low-Earth orbit satellite networks, characterized in that, include: The planning and decision-making module includes: a pre-trained large model and a tool invocation module; the large model is used to parse user request text to obtain communication request parameters, the user request text is the user's communication request described in natural language, and the communication request parameters include: the latitude and longitude coordinates of the origin and destination users, the QoS parameter type and the corresponding QoS parameter value; the tool invocation module is used to invoke satellite access tools and routing tools. The network sensing module is used to dynamically sense the time-varying network status of the low-Earth orbit satellite network and output a network topology map. The execution module includes: the satellite access tool and the routing tool. The satellite access tool is used to determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the originating and ending users by combining the network topology map. The routing tool is used to determine the routing path between the uplink and downlink access satellites based on the network topology map and the uplink and downlink access satellites, with the QoS parameter value as a constraint, and based on the routing calculation strategy corresponding to the QoS parameter type. The planning and decision-making module also includes: The prompting module is used to generate prompt text based on the user's request text and preset prompt words, and input the prompt text into the large model; the prompt text is used to guide the large model to perform the following operations: The origin and destination user latitude and longitude coordinates are parsed based on the user request text and context information, and then input into the satellite access tool. The communication service type is identified based on the user's text requirements, and the corresponding QoS parameter type and QoS parameter value are matched in the preset service database according to the communication service type. The QoS parameter type and QoS parameter value are then input into the routing tool. The planning and decision-making module further includes a session memory unit, which stores the user request text, the parsing results of the large model, the call records of the tool calling module, and the routing path. The prompting engineering module is also used to generate the context information based on the stored information in the session memory unit. The context information is used to adapt the parsing process of the large model to the user's usage habits and service preferences. The prompting engineering module is also used to generate the prompt text by combining the new user request text with the context information and the preset prompt words.

2. The intelligent agent system according to claim 1, characterized in that, The satellite access tool is also specifically used to generate multiple sets of uplink and downlink access satellites that correspond one-to-one according to multiple access strategies. The routing tool is also specifically used to calculate the routing path between each set of uplink and downlink access satellites based on the routing calculation strategy, so as to obtain multiple routing paths that correspond one-to-one with the multiple sets of uplink and downlink access satellites, as well as the communication performance parameter value of each routing path. Accordingly, the large model is also used to select the optimal routing path among multiple routing paths whose communication performance parameter value matches the QoS parameter value.

3. The intelligent agent system according to claim 2, characterized in that, The multiple access strategies include a multi-parameter access strategy. The satellite access tool generates multiple sets of uplink and downlink access satellites corresponding to each other according to the multi-parameter access strategy, specifically including: obtaining multi-dimensional evaluation parameters of multiple sets of candidate satellite combinations; quantifying the information contribution rate of each evaluation parameter according to the information entropy model, and dynamically assigning corresponding weights to each evaluation parameter based on the information contribution rate; calculating the multi-parameter weighted evaluation value of each set of candidate satellite combinations according to the weights assigned to each evaluation parameter; and selecting the candidate satellite combination with the largest multi-parameter weighted evaluation value as a set of uplink and downlink access satellites corresponding to the multi-parameter access strategy.

4. The intelligent agent system according to claim 1, characterized in that, The large model is also used to output structured thought analysis text before invoking the satellite access tool or the routing tool. The structure of the thought analysis text is set by the prompting engineering module. The thought analysis text is used to describe the parameter basis, constraints and reasoning process of the tool invocation.

5. The intelligent agent system according to claim 1, characterized in that, The QoS parameter types correspond one-to-one with the routing calculation strategies: If the QoS parameter type is latency, the routing calculation strategy is a weighted Dijkstra algorithm based on the shortest latency, and the edge weight of the algorithm is the sum of inter-satellite propagation latency and transmission latency; If the QoS parameter type is throughput, the routing calculation strategy is a weighted Dijkstra algorithm based on minimum load, and the edge weight of the algorithm is the standard deviation of the satellite node neighbor queue length; If the QoS parameter type is inter-satellite hop count, the routing calculation strategy is a weighted Dijkstra algorithm based on minimum hop count, and the edge weight of the algorithm is fixed at 1.

6. A method for on-demand route generation for low-Earth orbit satellite networks, characterized in that, The method is applied to the intelligent agent system as described in any one of claims 1-5, and the method includes: Obtain the user request text; the user request text is the user's communication request described in natural language. The user request text is parsed using a pre-trained large model to obtain communication request parameters; the communication request parameters include: the latitude and longitude coordinates of the start and end users, the QoS parameter type and the corresponding QoS parameter value. Based on the network topology diagram of the low-Earth orbit satellite network, determine the uplink and downlink access satellites corresponding to the latitude and longitude coordinates of the starting and ending users; Based on the network topology and the uplink / downlink access satellites, and using the QoS parameter values ​​as constraints, the routing path between the uplink / downlink access satellites is determined according to the routing calculation strategy corresponding to the QoS parameter type.

7. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory, wherein the processor, when executing the computer program, implements the method of claim 6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of claim 6.

Images