A method for constructing satellite communication networks based on large language model reasoning

Abstract

The present invention relates to a method for constructing a satellite communication network based on large language model reasoning, and belongs to the field of operation, maintenance, and control of satellite communication systems. The method includes: initializing large language model settings; parsing and processing demand input by the large language model; performing multi-step decomposition of the demand input by the large language model; querying and determining whether to call resources, calling resources of a movable point beam based on the query results, generating beam adjustment instruction parameters, and allowing the satellite control system to complete beam pointing adjustment; the large language model calling a resource allocation agent to complete allocation; the large language model calling a resource allocation agent; and the large language model issuing network construction parameters to complete satellite communication network construction. The present invention utilizes the reasoning capability of the LLM model to achieve automatic construction of a satellite communication network, meeting the requirements of intelligent satellite communication.

Description

Technical Field

[0001] The present invention relates to the field of satellite communications, and in particular to a method for constructing a satellite communication network based on large language model reasoning, which can be used for automated and agile network construction in high-orbit satellite communication systems. Background Art

[0002] In recent years, large language model (LLM) technologies, such as those from Deepseek and Qwen, have been widely applied in the communications field. For example, network traffic understanding is achieved by designing traffic representation methods and adjusting the model network output structure; network operations and maintenance management is achieved by optimizing prompt words, network architecture, and assistant design; and communication network security is ensured by identifying network attacks and generating response plans. LLMs have demonstrated a powerful ability to perform complex step-by-step reasoning and problem solving, and have been applied to various issues in terrestrial communication networks.

[0003] Currently, LLM-based communication applications are more focused on network management and control. The benefits they bring include: first, a shift in the way satellite network systems interact, from traditional, full user participation to LLM-initiated understanding and processing of user intent. LLM breaks down user intent and needs and assigns tasks to provide personalized services. This enables system-level AI scheduling capabilities, allowing all services to be invoked using the LLM model as a tool, offering the advantages of precise invocation and rapid response compared to traditional methods. LLM's capabilities enable communication systems to understand intent, change network interaction methods, improve management and control efficiency, and automate network operations, thereby achieving "LLM-enabled networks." Compared to terrestrial communications, satellite communications relies more heavily on operations and maintenance personnel for network management, making it well-suited to implementing related tasks based on large language models.

[0004] Taking satellite communication network construction as an example, when a communication mission arrives in a certain area, a satellite communication network needs to be quickly established for this mission to ensure the communication network in that area. However, current satellite network management still relies on manual entry of mission parameters. This has the following disadvantages: first, it requires high professionalism and reliability of personnel, and cannot ensure the optimal selection of parameters. Second, the network construction process is long, requiring several minutes to complete the entire process. Satellite communication missions can be urgent, and even a few minutes of network construction operations may result in communication delays. Therefore, in order to achieve a closed-loop process from operation and maintenance requirements to the generation and distribution of actual network construction parameters, and to achieve the goal of automated and fast and responsive network construction, a satellite communication network construction method based on LLM model reasoning is needed. Summary of the Invention

[0005] To address these issues, this paper proposes a method for building a satellite communication network based on large language model reasoning. In the field of satellite communication operations and management, this method implements a closed-loop process, from operational requirements to the generation and distribution of actual network construction parameters. This approach achieves automated and responsive network construction, reducing the traditional satellite communication network construction time from minutes to seconds.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A method for constructing a satellite communication network based on large language model reasoning includes the following steps:

[0008] Step 1: Initialize the large language model settings, configure the satellite communication network construction agent based on the large language model, and complete the CoT construction of the network construction scenario;

[0009] Step 2: The user inputs satellite communication requirements through the large language model. The large language model parses and processes the requirements and determines whether to call the satellite communication network construction agent.

[0010] Step 3: After the large language model supplements the required parameters through multiple rounds of inquiries, it decomposes the demand input into multiple steps and calls the resource query agent to query the situation system for beam resources that meet the demand input, and determines whether the queried beam resources meet the demand;

[0011] Step 4: The large language model queries and determines whether to call resources. Based on the query results of the large language model, the resources of the movable spot beam are called. The large language model generates beam adjustment instruction parameters and sends them to the satellite control system. The satellite control system injects the instructions into the satellite to complete the beam pointing adjustment.

[0012] Step 5: The large language model determines the selected beam, completes network selection and user station type selection, and calls the resource allocation agent to complete the allocation;

[0013] Step 6: The large language model calls the resource allocation agent, which calls the small model based on parameters to allocate beam and bandwidth resources.

[0014] Step 7: The large language model sends network construction parameters to satellites, terminals, and operation and maintenance management systems to complete the satellite communication network construction process.

[0015] Furthermore, the specific method of step 1 is:

[0016] Initialize a basic large language model with tool call capabilities and set the parameters of the large language model, including temperature parameters, kernel sampling probability threshold, and maximum generation length;

[0017] The satellite communication network construction agent is configured based on the large language model, and the thinking chain CoT construction of the network construction scenario is completed; the satellite communication network construction agent is a multi-agent collaborative system composed of three agents, including a resource query agent, a resource allocation agent and a network activation agent; the resource query agent completes the query function of satellite, beam and transponder resources, the resource allocation agent completes the allocation function of satellite, beam and transponder resources, and the network activation agent completes the planning of the allocated resources and the network activation function; the thinking chain CoT construction is based on the prompt project to systematically describe the large language model.

[0018] Furthermore, the specific method of step 2 is:

[0019] The user uses natural language to input satellite communication requirements through the large language model. The large language model parses and processes the satellite communication requirement input and determines whether the requirement analysis result is a satellite communication network construction requirement. If so, step 3 is executed; if not, the process ends.

[0020] Furthermore, the specific method of step 3 is:

[0021] The large language model supplements required parameters through multiple rounds of inquiries. The required parameters include bandwidth size, longitude and latitude. Non-required parameters include satellite selection, beam selection, user station type selection, and network selection. If the user input lacks required parameter information, the large language model performs multiple rounds of inquiries through thought chain configuration until the user has supplemented all required parameters.

[0022] The large language model calls the satellite network construction agent and decomposes the satellite communication demand input into multiple steps, converting the satellite network construction demand into a step-by-step calling workflow including a resource query agent, a resource allocation agent, and a network activation agent. The specific method is as follows: the large language model queries the beam resources that meet the satellite communication demand, uses the resource query agent through tools or function calls to query the beam resources, and determines whether the queried beam resources meet the demand. If so, step 5 is executed; if not, step 4 is executed.

[0023] Furthermore, the specific method of step 4 is:

[0024] The large language model asks the user whether to call the resources of the movable spot beam. After the user answers through the large language model, the large language model determines whether to call the resources. If it is not, the process ends; if it is, the large language model parses the context requirement parameters, generates the movable spot beam adjustment instruction parameters, and sends them to the satellite control system. The satellite control system injects the instructions into the satellite to complete the beam pointing adjustment.

[0025] Furthermore, the specific method of step 5 is:

[0026] Resources are selected based on the results obtained from the resource query agent or the results of adjusting the movable spot beams. There are two types of resources: one is the selection of satellites, beams and transponders, and the allocation of transponder frequency resources to the selected satellites and beams; the other is the selection of available earth stations and terminals, and the selection of network systems and user station types based on user preferences and historical data.

[0027] Furthermore, the specific method of step 6 is:

[0028] The large language model calls the resource allocation agent through a function and generates corresponding resource allocation parameters; the resource allocation agent calls the small model of the satellite communication frequency resource allocation method through the parameters to perform beam and bandwidth resource allocation; the small model includes a frequency allocation model and a beam allocation model.

[0029] Furthermore, the specific method of step 7 is:

[0030] The large language model activates the network agent through function calls, sends the network construction parameters to the satellite, terminal and operation and maintenance control system for execution, and completes the construction of the satellite communication network.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. In view of the fact that network management in the current satellite communication field relies more on the demand input of operation and maintenance personnel, the present invention utilizes the reasoning ability of the LLM model and proposes a satellite communication network construction method based on LLM model reasoning, which can meet the needs of intelligent satellite communication.

[0033] 2. The present invention completes the construction of satellite communication networks through a large language model. In the field of satellite communication operation and maintenance management, it can realize a full-process closed loop from operation and maintenance requirements to the generation and issuance of actual parameters for network construction, achieving the goal of automated and fast and agile network construction, and shortening the construction time of traditional artificial satellite communication networks from minutes to seconds. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a flowchart of a method for constructing a satellite communication network based on large language model reasoning in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0036] A method for constructing satellite communication networks based on large language model reasoning, such as Figure 1 As shown, the method includes the following steps:

[0037] Step S1: Initialize the Large Model (LLM) settings. Initialize the basic large model with tool call capabilities and set the parameters of the large model, including the temperature parameter, core sampling probability threshold, and maximum generation length. Specifically, the temperature parameter is set to 0.2. A lower temperature parameter is used to strictly control the randomness of the output and ensure the correctness of parameter generation. The core sampling probability threshold is set to 0.6. This low core sampling probability threshold setting ensures that the LLM model generates stable and reliable answers. The maximum generation length is set to 4k or the maximum number of tokens allowed by the LLM model.

[0038] Subsequently, the satellite communication network construction agent was configured based on the LLM model, and the CoT (CoT) for the network construction scenario was completed. The satellite communication network construction agent is a multi-agent collaborative system composed of three agents: a resource query agent, a resource allocation agent, and a network activation agent. The construction of the CoT uses the prompt project to systematically describe the LLM model. A typical system description example is as follows: "- The network construction agent is used to automate the construction of the satellite communication network. The tools that can be used include the resource query agent, the resource allocation agent, and the network activation agent; - Parameters must be supplemented through multiple rounds of inquiries; function calls cannot be made without required parameters; - If there are no idle resources, inquire whether to call the movable beam and adjust the beam pointing, and call the satellite beam tool appropriately."

[0039] Step S2: The user uses natural language to input satellite communication requirements through the LLM model. The LLM model parses and processes the satellite communication requirements input and determines whether the requirements parsed result is a satellite communication network construction requirement. If so, step 3 is executed; if not, the process ends.

[0040] Step S3: The LLM model supplements the required parameters through multiple rounds of inquiries. The required parameters include bandwidth size, longitude and latitude, and the non-required parameters include satellite selection, beam selection, user station type selection and network system selection. When the required parameter information is missing in the user demand input, the LLM model will perform multiple rounds of inquiry through the thought chain configuration until the user has supplemented all the required parameters. Subsequently, the LLM model calls the satellite network construction agent and decomposes the satellite communication demand input in multiple steps, converting the satellite network construction demand into a step-by-step call workflow including a resource query agent, a resource allocation agent and a network activation agent. First, in order to complete the network construction requirements, the LLM model needs to query the beam resources that meet the requirements, and query the beam resources through the resource query agent using tools or function calls (Function call), and determine whether the queried beam resources meet the requirements. If yes, execute step S5; if not, execute step S4;

[0041] Step S4: The LLM model inquires the user whether to call the resources of the movable spot beam. After the user answers through the LLM model, the LLM model determines whether to call the resources. If the judgment is no, the process ends. If the judgment is yes, the LLM model parses the context requirement parameters, generates the movable spot beam adjustment instruction parameters, sends them to the satellite control system, and injects the instructions into the satellite, completing the beam pointing adjustment and executing step S5.

[0042] Step S5: The LLM model determines the selected beam, completes network selection, selects user station type, and calls the resource allocation agent to complete the allocation. Resource selection is performed based on the results of the resource query agent or the results of adjusting the movable spot beam. This includes two types of resources: the first is the selection of satellites, beams, and transponders. Transponder frequency resources are allocated to the selected satellites and beams. For example, 30 MHz of available resources can be allocated from a 100 MHz transponder resource, ensuring time conflicts. The second is the selection of available earth stations and terminals. Network selection and user station type selection are performed based on user preferences and historical data.

[0043] Step S6: The LLM model calls the resource allocation agent through a function and generates corresponding resource allocation parameters. The resource allocation agent, through the parameters, calls a small model of the satellite communication frequency resource allocation method to perform beam and bandwidth resource allocation. The small model includes a frequency allocation model and a beam allocation model, both of which are prior art and will not be further described here.

[0044] S7: The LLM model calls the network activation agent through a function, sends the network construction parameters to the satellite, terminal, and operation and maintenance control system for execution, and completes the satellite communication network construction process.

[0045] This paper uses a general large-scale model (LLM) and agent-based agents to achieve one-click network construction for satellite communications. In the field of satellite communications operation and maintenance management, the use of large language models to complete satellite communication network construction can achieve a closed loop from operation and maintenance requirements to the generation and distribution of actual network construction parameters, achieving the goal of automated and responsive network construction, and shortening the traditional artificial satellite communication network construction time from minutes to seconds.

[0046] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0047] The above-described embodiments merely represent several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for constructing a satellite communication network based on large language model reasoning, characterized in that: The following steps are involved: Step 1: Initialize the large language model settings, configure the satellite communication network construction agent based on the large language model, and complete the CoT construction of the network construction scenario; Step 2: The user inputs satellite communication requirements through the large language model. The large language model parses and processes the requirements and determines whether to call the satellite communication network construction agent. If the judgment is yes, go to step 3; if the judgment is no, end the process; Step 3: After the large language model supplements the required parameters through multiple rounds of inquiries, it decomposes the demand input into multiple steps and calls the resource query agent to query the situation system for beam resources that meet the demand input, and determines whether the queried beam resources meet the demand; If the judgment is yes, go to step 5; if the judgment is no, go to step 4; Step 4: The large language model queries and determines whether to call resources. Based on the query results of the large language model, the resources of the movable spot beam are called. The large language model generates beam adjustment instruction parameters and sends them to the satellite control system. The satellite control system injects the instructions into the satellite to complete the beam pointing adjustment. Step 5: The large language model determines the selected beam, completes network selection and user station type selection, and calls the resource allocation agent to complete the allocation; Step 6: The large language model calls the resource allocation agent, which calls the small model based on parameters to allocate beam and bandwidth resources. Step 7: The large language model sends network construction parameters to satellites, terminals, and operation and maintenance management systems to complete the satellite communication network construction process.

2. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 1 is: Initialize a basic large language model with tool call capabilities and set the parameters of the large language model, including temperature parameters, kernel sampling probability threshold, and maximum generation length; Based on the large language model, the satellite communication network construction agent is configured and the CoT of thought chain of network construction scenarios is completed. The satellite communication network construction agent is a multi-agent collaborative system composed of three agents, including a resource query agent, a resource allocation agent, and a network activation agent. The resource query agent completes the query function of satellite, beam, and transponder resources, the resource allocation agent completes the allocation function of satellite, beam, and transponder resources, and the network activation agent completes the planning of allocated resources and network activation functions; the thinking chain CoT is built based on the prompt project to systematically describe the large language model.

3. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 2 is: The user uses natural language to input satellite communication requirements through the large language model. The large language model parses and processes the satellite communication requirement input and determines whether the requirement analysis result is a satellite communication network construction requirement. If so, step 3 is executed; if not, the process ends.

4. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 3 is: The large language model supplements required parameters through multiple rounds of inquiries. The required parameters include bandwidth size, longitude and latitude. Non-required parameters include satellite selection, beam selection, user station type selection, and network selection. If the user input lacks required parameter information, the large language model performs multiple rounds of inquiries through thought chain configuration until the user has supplemented all required parameters. The large language model calls the satellite network construction agent and decomposes the satellite communication demand input into multiple steps, converting the satellite network construction demand into a step-by-step calling workflow including a resource query agent, a resource allocation agent, and a network activation agent. The specific method is as follows: the large language model queries the beam resources that meet the satellite communication demand, uses the resource query agent through tools or function calls to query the beam resources, and determines whether the queried beam resources meet the demand. If so, step 5 is executed; if not, step 4 is executed.

5. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 4 is: The large language model asks the user whether to call the resources of the movable spot beam. After the user answers through the large language model, the large language model determines whether to call the resources. If not, the process ends. If the judgment is yes, the context requirement parameters are parsed through the large language model to generate movable point beam adjustment instruction parameters, which are sent to the satellite control system. The satellite control system will inject the instructions into the satellite to complete the beam pointing adjustment.

6. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 5 is: Resources are selected based on the results obtained from the resource query agent or the results of adjusting the movable spot beams. There are two types of resources: one is the selection of satellites, beams and transponders, and the allocation of transponder frequency resources to the selected satellites and beams; the other is the selection of available earth stations and terminals, and the selection of network systems and user station types based on user preferences and historical data.

7. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 6 is: The large language model calls the resource allocation agent through a function and generates corresponding resource allocation parameters; the resource allocation agent calls the small model of the satellite communication frequency resource allocation method through the parameters to perform beam and bandwidth resource allocation; the small model includes a frequency allocation model and a beam allocation model.

8. The method for constructing a satellite communication network based on large language model reasoning according to claim 1, characterized in that: The specific method of step 7 is: The large language model activates the network agent through function calls, sends the network construction parameters to the satellite, terminal and operation and maintenance control system for execution, and completes the construction of the satellite communication network.

Images