Satellite simulation intelligent modeling system and method based on large model and agent

Abstract

The application provides a satellite simulation intelligent modeling system and method based on a large model and an agent, and the system comprises a requirement understanding agent, a rule engine, a task planning agent, a task execution and interaction agent, a semantic tool set and a knowledge sedimentation unit; the requirement understanding agent analyzes a natural language requirement into a structured task description object with field semantic labels; the rule engine adopts a double-layer rule system composed of surface operation rules and deep meta-rules to complete constraint verification; the task planning agent generates an executable workflow with a dependency relationship graph and a fault-tolerant branch; the task execution and interaction agent schedules the semantic tool set based on a state machine to execute simulation and handle exceptions; the knowledge sedimentation unit collects execution data and updates a rule library, forming a self-enhancing cycle. The application realizes end-to-end automatic conversion from a natural language requirement to a reliable simulation result, reduces professional threshold, improves modeling efficiency and reliability, and supports system continuous evolution.

Description

Technical Field

[0001] This invention relates to the field of satellite simulation technology, and more particularly to an intelligent modeling system and method for satellite simulation based on a large model and an agent. Background Technology

[0002] Satellite simulation technology is a crucial supporting tool in the aerospace engineering field, primarily used to simulate complex scenarios such as satellite on-orbit operation, payload performance, and interactions with the space environment. Traditional satellite simulation relies on mathematical models, physical laws, and expert experience, requiring specialized software (such as STK, MATLAB / Simulink, etc.) for high-precision modeling and verification. However, satellite simulation involves multidisciplinary coupling (such as orbital mechanics, thermal control, and communication links), and its modeling process needs to integrate a large amount of tacit domain knowledge (such as environmental model selection rules and payload-orbit compatibility criteria). This knowledge usually exists in the form of unstructured documents or expert experience, making it difficult to directly translate into machine-executable logic. Furthermore, simulation tasks often require dynamic parameter adjustments and process optimization, while existing technologies lack the semantic understanding of ambiguous user needs, resulting in limited automation.

[0003] Existing satellite simulation technologies are mainly divided into three categories, but all of them have significant shortcomings:

[0004] Traditional professional simulation software (such as STK) can achieve high-fidelity simulations, but it is complex to operate and requires users to have a strong professional background. Its modeling and configuration rely entirely on manual completion, and it cannot automatically convert the intentions described in natural language into simulation parameters, resulting in low efficiency and a high risk of errors.

[0005] Script-based automation frameworks automate certain processes through pre-set scripts, but the script logic is fixed and cannot adapt to dynamic task requirements. For example, changes to the task require rewriting the code, and there is a lack of understanding of domain semantics (such as the inability to automatically derive the correlation rules between orbital altitude and antenna parameters based on "optimizing communication coverage").

[0006] General-purpose AI / RPA tools can simulate user interfaces (such as automatically filling out forms), but they only automate surface processes and lack the understanding of the underlying professional logic (such as link budget constraints and thermal control balance conditions). Their operation is prone to failure due to software interface updates, and they cannot perform domain-based verification and optimization, making it difficult to meet high reliability requirements.

[0007] In summary, existing technologies either rely too heavily on manual labor and lack intelligence, or their automation methods are rigid and unable to handle dynamic demands. Furthermore, they fail to address core issues such as the machine-executable nature of tacit knowledge, dynamic task orchestration, and reliability constraints, thus hindering the improvement of satellite simulation efficiency and intelligence levels. Summary of the Invention

[0008] To address the technical problems in existing technologies, such as the difficulty of machine execution of domain knowledge, the lack of dynamic and intelligent orchestration for complex simulation tasks, the unreliability of large model outputs, and the inability to accumulate and iterate simulation knowledge, this invention provides a satellite simulation intelligent modeling system and method based on a large model and agents. This invention employs a five-layer architecture: large model-driven, rule-based knowledge-constrained, multi-agent collaboration, semantic tool execution, and knowledge self-enhancement. It achieves natural language intent understanding through domain-based fine-tuning of the large model, constrains decision reliability through a two-layer rule engine, completes dynamic planning and execution through multi-agent collaboration, achieves precise scheduling through semantic tool encapsulation, and enables rule base iteration through knowledge accumulation units. This results in low-threshold, high-efficiency, highly reliable, and evolvable intelligent modeling for satellite simulation.

[0009] The technical means employed in this invention are as follows: A satellite simulation intelligent modeling system based on a large model and an agent includes: The demand understanding agent is used to receive natural language demands input by users. It uses a large language model that has been fine-tuned by satellite simulation domain corpus or optimized by prompting engineering to perform semantic parsing on the natural language demands and output a structured task description object containing domain semantic tags. The rule engine is used to receive the structured task description object, perform rule matching and verification, and output the constrained task specification. It includes a two-layer rule system for storing and managing satellite simulation. The two-layer rule system includes surface operation rules and deep meta-rules. The surface operation rules define parameter constraints, process templates and tool calling specifications, while the deep meta-rules define rule conflict resolution strategies, rule priorities and rule activation conditions. The task planning agent receives the constrained task specifications, and based on the process template and constraints provided by the rule engine, performs logical deduction through the built-in inference engine to dynamically generate an executable workflow with dependency graphs and fault-tolerant branches. The task execution and interaction intelligent agent is used to receive the executable workflow, monitor the execution status of each step of the workflow based on the state machine, schedule the semantically encapsulated simulation toolset to perform specific simulation operations according to the exception handling strategy predefined by the rule engine, and trigger retry, tool switching or human-computer interaction when an abnormal state is detected. A semantic toolset is used to encapsulate the callable functions of the satellite simulation system. Each tool is associated with machine-readable metadata, including function description, input parameter definition, preconditions and postconditions, so that the task execution and interaction agent can select and call tools based on the metadata. The knowledge accumulation unit is used to collect configuration data, optimization parameters and verification results generated during the simulation execution process, and feed them back to the rule engine to update the rule base. It transforms the simulation execution data into domain knowledge and optimizes intelligent decision-making capabilities, thus building a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system.

[0010] Furthermore, the demand-understanding agent includes: Domain adaptation module, used to load large language models finely tuned from satellite simulation domain corpus or optimized prompt engineering templates; The semantic parsing module is used to identify key domains in the natural language requirements, including satellite type, payload type, mission objective, regional scope, and performance indicators, and to extract items to be clarified. The structured output module is used to organize the parsed results into a structured task description object containing domain semantic labels, including task type identifier, parameter set, constraints, and a list of parameters to be clarified.

[0011] Furthermore, the demand understanding agent is also used for: When the structured task description object contains parameters to be clarified, the task execution and interaction agent is triggered to generate a clarification question and output it to the user based on the follow-up questioning logic of the corresponding task template in the rule engine. Receive user feedback on the clarification inquiry and update the structured task description object until all required parameters are complete.

[0012] Furthermore, the rule engine includes: A rule repository is used to store the surface operation rules and the deep meta-rules in a hierarchical manner; the surface operation rules include parameter range rules, process template rules, and tool matching rules; the deep meta-rules include conflict resolution rules, priority rules, and condition triggering rules. A rule matcher is used to match the structured task description object with rules in the rule repository to identify applicable process templates and parameter constraints. The rule validator is used to perform consistency checks on the matching results, detect rule conflicts and resolve them according to the deep meta-rules, and output the constrained task specifications.

[0013] Furthermore, the detection rule conflict resolution based on the deep meta-rules includes: When load power consumption rules conflict with platform power supply rules, power supply safety takes priority; when performance indicator rules conflict with cost constraint rules, priority rules are determined based on task type identifiers.

[0014] Furthermore, the task planning agent includes: The reasoning engine module is used to receive the constrained task specifications, perform "if-then-else" logical deduction based on the process template provided by the rule engine, and determine the sequence of task execution steps. The dependency analysis module is used to analyze the data and resource dependencies between each execution step and construct a dependency graph; The fault-tolerant design module is used to pre-set alternative tool paths and exception branches for critical execution steps, generating an executable workflow with fault-tolerant branches; The workflow output module is used to output the executable workflow to the task execution and interaction agent. The executable workflow includes a step sequence, a dependency graph, a tool mapping table, and an exception handling strategy.

[0015] Furthermore, the task execution and interaction agent includes: The status monitoring module is used to monitor the execution status of each step of the workflow in real time, including success, failure, and timeout. The strategy decision module is used to query the exception handling strategy in the rule engine based on the execution status, and decide whether to retry, switch to alternative tools, adjust parameters, or trigger human-computer interaction. The tool scheduling module is used to call the corresponding tool in the semantic toolset based on the decision result, pass parameters and receive execution results; The interaction interface module is used to generate structured query information and receive user feedback when human-computer interaction is required.

[0016] Furthermore, each tool encapsulation in the semantic toolset includes: Function identifiers are used to implement function descriptions, explain the function of a tool, and distinguish different tools; The interface description block is used to define the list of input parameters, parameter types, and parameter constraints for tool calls to the interface; The semantic metadata block is used to describe the functional semantics, applicable scenarios, preconditions and postconditions of the tool. The preconditions define the system state that must be met before the tool is executed, and the postconditions define the changes in the system state after the tool is executed. The execution mapping block is used to establish the mapping relationship between semantic descriptions and the underlying simulation system API.

[0017] Furthermore, the knowledge accumulation module includes: The data acquisition unit is used to collect configuration parameters, execution logs, optimization results, and verification data generated during the simulation process. The knowledge extraction unit is used to extract effective rule fragments and optimization strategies from the collected data. The rule update unit is used to update the extracted valid rule fragments to the rule repository of the rule engine after manual review; The model optimization unit is used to optimize the large language model parameters or prompt engineering template of the requirement understanding agent based on historical task data.

[0018] This invention also provides a satellite simulation intelligent modeling method based on the above-mentioned satellite simulation intelligent modeling system, comprising: S1. Receive the natural language input from the user, perform semantic parsing using a large language model that has been fine-tuned by satellite simulation domain corpus or optimized by prompting engineering, and output a structured task description object containing domain semantic tags. S2. Input the structured task description object into the rule engine. The rule engine performs rule matching and verification based on a two-layer rule system. The two-layer rule system includes surface operation rules that define parameter constraints and process templates, and deep meta-rules that define conflict resolution strategies and priorities. It outputs the constrained task specifications. S3. Based on the task specifications after the constraints, logical deduction is performed through the inference engine to dynamically generate an executable workflow with dependency graphs and fault-tolerant branches; S4. Monitor the execution status of each step of the executable workflow based on the state machine, and schedule the semantically encapsulated simulation toolset to perform specific simulation operations according to the predefined exception handling strategy. Each tool in the toolset is associated with machine-readable metadata including function description, input parameters, preconditions and postconditions. S5. Collect configuration data, optimization parameters and verification results generated during the simulation execution process, and feed them back to the rule engine to update the rule base. Transform the simulation execution data into domain knowledge and optimize intelligent decision-making capabilities to build a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system.

[0019] Compared with the prior art, the present invention has the following advantages: 1. This invention employs a domain-fine-tuned large language model through a demand-understanding intelligent agent, lowering the professional threshold for satellite simulation modeling from mastering professional knowledge and software operation to being able to describe professional tasks. It achieves automatic conversion of natural language requirements into structured task description objects, resolving the contradiction between the implicit nature of domain knowledge and machine executability. Preliminary experimental verification shows that novice users using this system to complete medium-complexity constellation coverage assessment tasks have reduced the average time to complete the task correctly on the first attempt by approximately 85%, and the configuration accuracy can reach over 98%.

[0020] 2. This invention constructs a two-layer rule system through a rule engine, which strictly constrains the generalization and understanding capabilities of large models within a safe boundary composed of domain rules and structured knowledge, ensuring that every automated decision conforms to engineering practice and professional standards, thus bridging the gap between the general capabilities and high reliability requirements of large models.

[0021] 3. This invention solves the problem of the lack of intelligent dynamic orchestration of complex simulation tasks by having a task planning agent perform logical deduction based on templates and constraints provided by a rule engine, dynamically generating an executable workflow with dependency graphs and fault-tolerant branches.

[0022] 4. This invention collects simulation execution data through a knowledge accumulation module and feeds back to update the rule base, constructing a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system, thus solving the problem of the disconnect between knowledge accumulation and system evolution in the simulation closed loop.

[0023] 5. This invention associates machine-readable metadata, including functional identifiers, preconditions, and postconditions, with each tool through a semantic toolset. This enables task execution and interaction agents to select and invoke tools based on semantic understanding, keeping the illusion rate or error rate of large model decision-making within an acceptable engineering range of less than 1%. The reliability of simulation results is comparable to, or even better than, the results manually configured by intermediate engineers, thus improving the accuracy and reliability of simulation execution.

[0024] In summary, the technical solution of this invention can solve the problems of reliance on manual labor, rigid automation, insufficient reliability, and inability to accumulate knowledge in existing satellite simulation technologies. Therefore, the technical solution of this invention solves the technical problems of high threshold, low efficiency, poor reliability, and lack of evolution in existing satellite simulation modeling technologies.

[0025] Based on the above reasons, this invention can be widely applied in fields such as satellite system design, performance evaluation, and mission verification. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a block diagram of the collaborative architecture of the satellite simulation intelligent modeling system of the present invention.

[0028] Figure 2 This is a flowchart of the intelligent modeling method for satellite simulation of the present invention.

[0029] Figure 3 This is a schematic diagram of the collaborative work of intelligent agents and the core data flow of the present invention.

[0030] Figure 4 This is a schematic diagram of the internal structure of the rule engine of this invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0033] like Figure 1 As shown, this invention provides a satellite simulation intelligent modeling system based on a large model and an agent, comprising: The demand understanding agent is used to receive natural language demands input by users. It uses a large language model that has been fine-tuned by satellite simulation domain corpus or optimized by prompting engineering to perform semantic parsing on the natural language demands and output a structured task description object containing domain semantic tags. The rule engine is used to receive the structured task description object, perform rule matching and verification, and output the constrained task specification. It includes a two-layer rule system for storing and managing satellite simulation. The two-layer rule system includes surface operation rules and deep meta-rules. The surface operation rules define parameter constraints, process templates and tool calling specifications, while the deep meta-rules define rule conflict resolution strategies, rule priorities and rule activation conditions. The task planning agent receives the constrained task specifications, and based on the process template and constraints provided by the rule engine, performs logical deduction through the built-in inference engine to dynamically generate an executable workflow with dependency graphs and fault-tolerant branches. The task execution and interaction intelligent agent is used to receive the executable workflow, monitor the execution status of each step of the workflow based on the state machine, schedule the semantically encapsulated simulation toolset to perform specific simulation operations according to the exception handling strategy predefined by the rule engine, and trigger retry, tool switching or human-computer interaction when an abnormal state is detected. A semantic toolset is used to encapsulate the callable functions of the satellite simulation system. Each tool is associated with machine-readable metadata, including function description, input parameter definition, preconditions and postconditions, so that the task execution and interaction agent can select and call tools based on the metadata. The knowledge accumulation unit is used to collect configuration data, optimization parameters and verification results generated during the simulation execution process, and feed them back to the rule engine to update the rule base. It transforms the simulation execution data into domain knowledge and optimizes intelligent decision-making capabilities, thus building a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system.

[0034] In this embodiment, the satellite simulation intelligent modeling system adopts a four-layer collaborative architecture: a large model driving layer, a rule knowledge constraint layer, a multi-agent collaboration layer, and a toolset execution layer. The large model driving layer achieves natural language understanding through the requirement understanding agent; the rule knowledge constraint layer achieves reliability constraints through the rule engine; the multi-agent collaboration layer achieves dynamic planning and execution through the task planning agent and the task execution and interaction agent; and the toolset execution layer achieves simulation operation invocation through the semantic toolset.

[0035] In a specific implementation, as a preferred embodiment of the present invention, the demand understanding agent includes: Domain adaptation module, used to load large language models finely tuned from satellite simulation domain corpus or optimized prompt engineering templates; The semantic parsing module is used to identify key domains in the natural language requirements, including satellite type, payload type, mission objective, regional scope, and performance indicators, and to extract items to be clarified. The structured output module is used to organize the parsed results into a structured task description object containing domain semantic labels, including task type identifier, parameter set, constraints, and a list of parameters to be clarified.

[0036] In a specific implementation, as a preferred embodiment of the present invention, the demand understanding agent is further used for: When the structured task description object contains parameters to be clarified, the task execution and interaction agent is triggered to generate a clarification question and output it to the user based on the follow-up questioning logic of the corresponding task template in the rule engine. Receive user feedback on the clarification inquiry and update the structured task description object until all required parameters are complete.

[0037] In a specific implementation, as a preferred embodiment of the present invention, the rule engine includes: A rule repository is used to store the surface operation rules and the deep meta-rules in a hierarchical manner; the surface operation rules include parameter range rules, process template rules, and tool matching rules; the deep meta-rules include conflict resolution rules, priority rules, and condition triggering rules. A rule matcher is used to match the structured task description object with rules in the rule repository to identify applicable process templates and parameter constraints. The rule validator is used to perform consistency checks on the matching results, detect rule conflicts and resolve them according to the deep meta-rules, and output the constrained task specifications.

[0038] In a specific implementation, as a preferred embodiment of the present invention, the detection of rule conflicts and their resolution based on the deep meta-rules includes: When load power consumption rules conflict with platform power supply rules, power supply safety takes priority; when performance indicator rules conflict with cost constraint rules, priority rules are determined based on task type identifiers.

[0039] In a specific implementation, as a preferred embodiment of the present invention, the task planning agent includes: The reasoning engine module is used to receive the constrained task specifications, perform "if-then-else" logical deduction based on the process template provided by the rule engine, and determine the sequence of task execution steps. The dependency analysis module is used to analyze the data and resource dependencies between each execution step and construct a dependency graph; The fault-tolerant design module is used to pre-set alternative tool paths and exception branches for critical execution steps, generating an executable workflow with fault-tolerant branches; The workflow output module is used to output the executable workflow to the task execution and interaction agent. The executable workflow includes a step sequence, a dependency graph, a tool mapping table, and an exception handling strategy.

[0040] In a specific implementation, as a preferred embodiment of the present invention, the task execution and interaction intelligent agent includes: The status monitoring module is used to monitor the execution status of each step of the workflow in real time, including success, failure, and timeout. The strategy decision module is used to query the exception handling strategy in the rule engine based on the execution status, and decide whether to retry, switch to alternative tools, adjust parameters, or trigger human-computer interaction. The tool scheduling module is used to call the corresponding tool in the semantic toolset based on the decision result, pass parameters and receive execution results; The interaction interface module is used to generate structured query information and receive user feedback when human-computer interaction is required.

[0041] In a specific implementation, as a preferred embodiment of the present invention, each tool encapsulation in the semantic toolset includes: Function identifiers are used to implement function descriptions, explain the function of a tool, and distinguish different tools; The interface description block is used to define the list of input parameters, parameter types, and parameter constraints for tool calls to the interface; The semantic metadata block is used to describe the functional semantics, applicable scenarios, preconditions and postconditions of the tool. The preconditions define the system state that must be met before the tool is executed, and the postconditions define the changes in the system state after the tool is executed. The execution mapping block is used to establish the mapping relationship between semantic descriptions and the underlying simulation system API.

[0042] In a specific implementation, as a preferred embodiment of the present invention, the knowledge accumulation module includes: The data acquisition unit is used to collect configuration parameters, execution logs, optimization results, and verification data generated during the simulation process. The knowledge extraction unit is used to extract effective rule fragments and optimization strategies from the collected data. The rule update unit is used to update the extracted valid rule fragments to the rule repository of the rule engine after manual review; The model optimization unit is used to optimize the large language model parameters or prompt engineering template of the requirement understanding agent based on historical task data.

[0043] like Figure 2 As shown, the present invention also provides a satellite simulation intelligent modeling method based on the above-mentioned satellite simulation intelligent modeling system. include: S1. Receive the natural language input from the user, perform semantic parsing using a large language model that has been fine-tuned by satellite simulation domain corpus or optimized by prompting engineering, and output a structured task description object containing domain semantic tags. S2. Input the structured task description object into the rule engine. The rule engine performs rule matching and verification based on a two-layer rule system. The two-layer rule system includes surface operation rules that define parameter constraints and process templates, and deep meta-rules that define conflict resolution strategies and priorities. It outputs the constrained task specifications. S3. Based on the task specifications after the constraints, logical deduction is performed through the inference engine to dynamically generate an executable workflow with dependency graphs and fault-tolerant branches; S4. Monitor the execution status of each step of the executable workflow based on the state machine, and schedule the semantically encapsulated simulation toolset to perform specific simulation operations according to the predefined exception handling strategy. Each tool in the toolset is associated with machine-readable metadata including function description, input parameters, preconditions and postconditions. S5. Collect configuration data, optimization parameters and verification results generated during the simulation execution process, and feed them back to the rule engine to update the rule base. Transform the simulation execution data into domain knowledge and optimize intelligent decision-making capabilities to build a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system.

[0044] The embodiments of the present invention are described simply because they correspond to those in the embodiments above. For any similarities, please refer to the descriptions in the embodiments above, which will not be elaborated here.

[0045] Example 1 like Figure 3 As shown, the user inputs a natural language request: "I want to evaluate the effectiveness of using small satellites to monitor XX." The semantic parsing module of the demand understanding agent identifies the key domains "small satellite", "surveillance" and "XX". The structured output module generates a structured task description object containing items to be clarified, specifically including: the task type is identified as a maritime surveillance task, the area range is XX, and the list of parameters to be clarified includes specific indicators of the satellite platform, payload type, and surveillance performance indicators.

[0046] The rule matcher of the rule engine matches the structured task description object with the "maritime surveillance" task template in the rule repository. The rule validator detects that the load type and performance indicators are required but missing, and outputs a constraint task specification containing the missing parameter identifier.

[0047] The strategy decision-making module of the task execution and interaction agent triggers human-computer interaction based on the missing parameter identifiers in the constraint task specifications. The interaction interface module generates and outputs clarifying questions based on the questioning logic of the "maritime surveillance" task template: "Please specify the main performance indicators of the surveillance, such as: optical resolution better than meters, or radar synthetic aperture resolution better than 3 meters? The preset platform is a type XX small satellite, and its selectable payloads are A (optical payload) or B (SAR payload). Please select or specify others." After receiving the clarification query, the user adds the following input: "SAR, resolution 5 meters, revisit time within one day." The requirement understanding agent receives the feedback information, updates the structured task description object, determines the payload type as SAR, sets the resolution parameter to 5 meters, and sets the revisit time parameter to 24 hours.

[0048] The rule engine performs rule matching and verification again to confirm that all necessary parameters are complete, and outputs the final constraint task specification. The inference engine module of the task planning agent performs logical deduction based on the constraint task specification to determine the sequence of execution steps: creating a simulation scenario, configuring satellite platform parameters, configuring SAR payload parameters, setting orbit parameters, configuring the ground target area, running coverage analysis simulation, and outputting a results report. The dependency analysis module constructs a dependency graph between steps, and the fault-tolerant design module presets alternative tool paths for the orbit parameter configuration step.

[0049] The task execution and interaction agent monitors each step of the process using a state machine. The tool scheduling module sequentially calls the scene creation tool, platform configuration tool, payload configuration tool, track configuration tool, area configuration tool, simulation run tool, and report generation tool from the semantic toolset. Each tool's function is clearly identified by its functional identifier, and precondition checks in the semantic metadata block ensure the correct tool call order. After execution, the knowledge accumulation module collects the configuration parameters and execution results of the simulation. After analysis by the knowledge extraction unit, the rule update unit updates the valid rule fragments to the rule repository.

[0050] Example 2 like Figure 4 As shown, the user inputs a natural language request: "Compare the communication performance of the Walker and Rose constellations in polar regions, both with 48 satellites." The demand understanding agent parses the comparison objects as the first constellation type Walker constellation and the second constellation type Rose constellation, the evaluation index is communication performance, and the constraints are the number of satellites 48 and the region is the polar region.

[0051] The rules engine matches the "Constellation Communication Performance Evaluation" task template and the "Constellation Comparison Analysis" process template, and outputs a constraint task specification containing two parallel branches.

[0052] The inference engine module of the task planning agent determines the sequence of execution steps, including configuring the Walker constellation in the first branch, configuring the Rose constellation in the second branch, configuring the shared platform payload, parallel simulation execution, and result comparison and analysis. The dependency analysis module constructs a dependency graph, clarifying that the two branches have data dependencies in the platform payload configuration step, and that the same parameters must be used to ensure consistent comparison benchmarks. The fault tolerance design module pre-sets timeout exception handling branches for the simulation execution steps.

[0053] The task execution and interactive agent control branches run in parallel. In the first branch, the tool scheduling module calls the Walker constellation orbit generation tool, inputting the number of satellites (48) and polar coverage constraints to generate the Walker constellation orbit parameters. In the second branch, the tool scheduling module calls the Rose constellation orbit generation tool, inputting the same number of satellites and regional constraints to generate the Rose constellation orbit parameters. The two branches share satellite platform configuration tools and communication payload configuration tools to ensure consistency in all conditions except for the constellation configuration.

[0054] The status monitoring module monitors the simulation execution status of both branches in real time. When a timeout is detected in either branch, the strategy decision module determines whether to trigger a retry or switch to an alternative simulation tool based on the exception handling strategy. After both branches have successfully completed, the tool scheduling module calls the result extraction tool to extract key performance indicators such as coverage and link margin from the simulation output, and calls the comparative analysis tool to generate a comparative analysis report containing parallel graphs and a difference summary table.

[0055] The knowledge accumulation module collects configuration parameters and performance index data of the two constellations. The knowledge extraction unit analyzes the performance difference between the Walker and Rose constellations in polar regions. The rule update unit adds the rule fragment "Rose constellation coverage characteristics are better than Walker constellation in polar communication scenarios" to the rule repository after manual review, for reference in subsequent planning of similar tasks.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A satellite simulation intelligent modeling system based on a large model and an agent, characterized in that, include: The demand understanding agent is used to receive natural language demands input by users. It uses a large language model that has been fine-tuned by satellite simulation domain corpus or optimized by prompting engineering to perform semantic parsing on the natural language demands and output a structured task description object containing domain semantic tags. The rule engine is used to receive the structured task description object, perform rule matching and verification, and output the constrained task specification. It includes a two-layer rule system for storing and managing satellite simulation. The two-layer rule system includes surface operation rules and deep meta-rules. The surface operation rules define parameter constraints, process templates and tool calling specifications, while the deep meta-rules define rule conflict resolution strategies, rule priorities and rule activation conditions. The task planning agent receives the constrained task specifications, and based on the process template and constraints provided by the rule engine, performs logical deduction through the built-in inference engine to dynamically generate an executable workflow with dependency graphs and fault-tolerant branches. The task execution and interaction intelligent agent is used to receive the executable workflow, monitor the execution status of each step of the workflow based on the state machine, schedule the semantically encapsulated simulation toolset to perform specific simulation operations according to the exception handling strategy predefined by the rule engine, and trigger retry, tool switching or human-computer interaction when an abnormal state is detected. A semantic toolset is used to encapsulate the callable functions of the satellite simulation system. Each tool is associated with machine-readable metadata, including function description, input parameter definition, preconditions and postconditions, so that the task execution and interaction agent can select and call tools based on the metadata. The knowledge accumulation unit is used to collect configuration data, optimization parameters and verification results generated during the simulation execution process, and feed them back to the rule engine to update the rule base. It transforms the simulation execution data into domain knowledge and optimizes intelligent decision-making capabilities, thus building a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system.

2. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 1, characterized in that, The demand understanding agent includes: Domain adaptation module, used to load large language models finely tuned from satellite simulation domain corpus or optimized prompt engineering templates; The semantic parsing module is used to identify key domains in the natural language requirements, including satellite type, payload type, mission objective, regional scope, and performance indicators, and to extract items to be clarified. The structured output module is used to organize the parsed results into a structured task description object containing domain semantic labels, including task type identifier, parameter set, constraints, and a list of parameters to be clarified.

3. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 2, characterized in that, The demand understanding agent is also used for: When the structured task description object contains parameters to be clarified, the task execution and interaction agent is triggered to generate a clarification question and output it to the user based on the follow-up questioning logic of the corresponding task template in the rule engine. Receive user feedback on the clarification inquiry and update the structured task description object until all required parameters are complete.

4. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 1, characterized in that, The rule engine includes: A rule repository is used to store the surface operation rules and the deep meta-rules in a hierarchical manner; the surface operation rules include parameter range rules, process template rules, and tool matching rules; the deep meta-rules include conflict resolution rules, priority rules, and condition triggering rules. A rule matcher is used to match the structured task description object with rules in the rule repository to identify applicable process templates and parameter constraints. The rule validator is used to perform consistency checks on the matching results, detect rule conflicts and resolve them according to the deep meta-rules, and output the constrained task specifications.

5. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 4, characterized in that, The detection rule conflict is resolved according to the deep meta-rules, including: When load power consumption rules conflict with platform power supply rules, power supply safety takes priority; when performance indicator rules conflict with cost constraint rules, priority rules are determined based on task type identifiers.

6. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 1, characterized in that, The task planning agent includes: The reasoning engine module is used to receive the constrained task specifications, perform "if-then-else" logical deduction based on the process template provided by the rule engine, and determine the sequence of task execution steps. The dependency analysis module is used to analyze the data and resource dependencies between each execution step and construct a dependency graph; The fault-tolerant design module is used to pre-set alternative tool paths and exception branches for critical execution steps, generating an executable workflow with fault-tolerant branches; The workflow output module is used to output the executable workflow to the task execution and interaction agent. The executable workflow includes a step sequence, a dependency graph, a tool mapping table, and an exception handling strategy.

7. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 1, characterized in that, The task execution and interaction intelligent agent includes: The status monitoring module is used to monitor the execution status of each step of the workflow in real time, including success, failure, and timeout. The strategy decision module is used to query the exception handling strategy in the rule engine based on the execution status, and decide whether to retry, switch to alternative tools, adjust parameters, or trigger human-computer interaction. The tool scheduling module is used to call the corresponding tool in the semantic toolset based on the decision result, pass parameters and receive execution results; The interaction interface module is used to generate structured query information and receive user feedback when human-computer interaction is required.

8. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 1, characterized in that, Each tool encapsulation in the semantic toolset includes: Function identifiers are used to implement function descriptions, explain the function of a tool, and distinguish different tools; The interface description block is used to define the list of input parameters, parameter types, and parameter constraints for tool calls to the interface; The semantic metadata block is used to describe the functional semantics, applicable scenarios, preconditions and postconditions of the tool. The preconditions define the system state that must be met before the tool is executed, and the postconditions define the changes in the system state after the tool is executed. The execution mapping block is used to establish the mapping relationship between semantic descriptions and the underlying simulation system API.

9. The intelligent modeling system for satellite simulation based on a large model and an agent according to claim 1, characterized in that, The knowledge accumulation module includes: The data acquisition unit is used to collect configuration parameters, execution logs, optimization results, and verification data generated during the simulation process. The knowledge extraction unit is used to extract effective rule fragments and optimization strategies from the collected data. The rule update unit is used to update the extracted valid rule fragments to the rule repository of the rule engine after manual review; The model optimization unit is used to optimize the large language model parameters or prompt engineering template of the requirement understanding agent based on historical task data.

10. A satellite simulation intelligent modeling method based on the satellite simulation intelligent modeling system described in any one of claims 1-9, characterized in that, include: S1. Receive the natural language input from the user, perform semantic parsing using a large language model that has been fine-tuned by satellite simulation domain corpus or optimized by prompting engineering, and output a structured task description object containing domain semantic tags. S2. Input the structured task description object into the rule engine. The rule engine performs rule matching and verification based on a two-layer rule system. The two-layer rule system includes surface operation rules that define parameter constraints and process templates, and deep meta-rules that define conflict resolution strategies and priorities. It outputs the constrained task specifications. S3. Based on the task specifications after the constraints, logical deduction is performed through the inference engine to dynamically generate an executable workflow with dependency graphs and fault-tolerant branches; S4. Monitor the execution status of each step of the executable workflow based on the state machine, and schedule the semantically encapsulated simulation toolset to perform specific simulation operations according to the predefined exception handling strategy. Each tool in the toolset is associated with machine-readable metadata including function description, input parameters, preconditions and postconditions. S5. Collect configuration data, optimization parameters and verification results generated during the simulation execution process, and feed them back to the rule engine to update the rule base. Transform the simulation execution data into domain knowledge and optimize intelligent decision-making capabilities to build a data-driven, knowledge-constrained, and intelligently evolving self-reinforcing loop system.

Images