A generative model driven satellite digital-real fusion test scheme generation method

Abstract

The application discloses a kind of generative model driven satellite digital-real fusion test scheme generation method, belong to spacecraft digitization and computer science technical field.The method includes: based on the test task understanding of knowledge graph and semantic modeling, constructs structured satellite test knowledge system;Utilize generative model to automatically construct the test scheme set in accordance with engineering specification;Through digital twinning technology, the scheme is virtually simulated and feasibility is verified;Intelligent screening and optimization recommendation are carried out to feasible scheme using multi-index decision method;Finally, based on real test data, the digital twinning satellite model and generative model are iterated evolution, realize closed loop self-correction and adaptive optimization.The application effectively solves the problems that traditional test scheme relies on artificial, poor adaptability, verification lag and the like, improves the intelligent level of satellite test, generation efficiency and scheme reliability.

Description

Technical Field

[0001] This invention belongs to the fields of spacecraft digitization and computer science technology, specifically relating to a generative model-driven method for generating satellite data-real fusion test schemes. Background Technology

[0002] As the complexity of space missions continues to increase, satellite systems are evolving towards high performance, multi-satellite collaboration, on-orbit adaptation, and rapid response. This places unprecedented demands on the comprehensiveness, efficiency, and reliability of satellite testing. Traditional satellite test scheme design heavily relies on the experience of domain experts and static procedure templates, which has significant limitations. On the one hand, manual design struggles to systematically cover complex test scenarios involving multi-task coupling and multi-subsystem interaction, easily leading to test blind spots or redundant operations, affecting the integrity and efficiency of testing. On the other hand, existing test schemes lack dynamic adjustment capabilities. When test objectives, platform status, or environmental constraints change, the entire process often needs to be manually redesigned and verified, resulting in slow response and difficulty in effectively identifying potential operational risks and logical conflicts before scheme execution.

[0003] In recent years, the "digital-real fusion" technology system has provided a new approach to addressing the aforementioned challenges. This system constructs a high-fidelity digital twin satellite model, accurately simulating the physical behavior and response characteristics of various satellite subsystems under different operating conditions in virtual space, and dynamically comparing and jointly analyzing this model with real on-orbit or ground-based test data. This method can not only be used for virtual verification and risk prediction of test schemes, but also enables pre-optimization and process control of test strategies, thereby significantly improving the accuracy and safety of testing. Constructing an intelligent and iterative digital-real fusion testing system has significant engineering application value for enhancing the satellite's on-orbit autonomous management capabilities and the rapid testing and verification capabilities for new missions.

[0004] However, in current satellite data-real fusion testing practices, the level of intelligence in the generation of test requirements and final executable solutions remains insufficient. The solution generation process has not yet formed a tight closed loop with subsequent simulation verification, physical testing, and model evolution, resulting in a lack of continuous learning and self-optimization capabilities for the system as a whole. Therefore, there is an urgent need for an intelligent method that can span the entire process of test solution generation, verification, optimization, and evolution to fully leverage the advantages of data-real fusion and achieve a paradigm shift from "manual design" to "intelligent generation and continuous optimization" of test solutions. Summary of the Invention

[0005] To address the aforementioned technical issues, this invention provides a generative model-driven method for generating satellite data-real fusion test schemes. It introduces generative artificial intelligence models and data-real fusion testing technology, and formulates a generative model-assisted satellite test scheme design method. This method can be used for scheme verification and risk prediction, and can also avoid unreasonable test designs and uncontrollable operating conditions, thereby improving the efficiency, reliability, and safety of satellite testing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A generative model-driven method for generating satellite data-real fusion test schemes, the method comprising:

[0008] Step S101: Understanding satellite testing tasks based on knowledge graphs and semantic modeling: Construct a satellite testing knowledge graph and perform semantic structured modeling of satellite testing tasks;

[0009] Step S102: Automatic construction of satellite test scheme set driven by generative model: Based on the test requirements, automatically construct a satellite test scheme set that conforms to engineering specifications using generative model;

[0010] Step S103: Feasibility verification of satellite testing scheme based on digital twin: Use digital twin satellite model to simulate the response of the test scheme set in virtual test scenario, evaluate the scheme effect and identify potential risks;

[0011] Step S104: Scheme selection and intelligent optimization based on multi-index decision-making: Intelligent evaluation of all candidate schemes that have passed simulation verification, and recommendation of the optimal testing strategy;

[0012] Step S105: Evolution of satellite test model based on data-real fusion: Conduct real tests using the optimal testing strategy, correct the digital twin satellite model based on the obtained real test data, and feed the evaluation results back to the generative model to realize the iterative learning and evolution of the satellite test model.

[0013] Furthermore, in step S101, constructing the satellite test knowledge graph includes:

[0014] Key entities are extracted from satellite test procedures, intra-satellite / satellite-ground / inter-satellite system interface documents, ground / on-orbit test record sheets, and fault diagnosis reports of structural / power / thermal control / attitude and orbit control / propulsion / telemetry and remote control / on-board electronics / payload subsystems. These key entities include satellite test items, operating procedures, virtual on-board environment and structural parameter constraints, and satellite subsystem equipment composition. A semantic relationship network between entities is established to construct a hierarchical satellite test knowledge graph. This knowledge graph uses triples to model the semantic relationships of the test process. Each triple includes an initiating test entity, an action relationship, and a target response entity, supporting the structured expression and cross-stage reasoning of satellite test tasks. Graph-text alignment and structural understanding technologies are employed to achieve consistent encoding of information and text semantics and enhance the system's ability to reason structurally about non-textual information.

[0015] Furthermore, step S101 also includes:

[0016] The entity and action relationships in the satellite test knowledge graph are vectorized using a graph embedding model; the structure of the knowledge graph is trained and optimized by minimizing a loss function based on vector representation, which is used to evaluate the rationality of triples and promote the convergence of the knowledge structure.

[0017] Furthermore, step S102 includes:

[0018] A parameterized prompting structure is adopted to construct standardized prompting templates from the task type, satellite platform model, and specific constraints input by the tester, in order to guide the generative model;

[0019] A structure-aware text generation method is adopted to output a hierarchical satellite test plan, with each test phase including the test object, input parameters, execution operation and expected response.

[0020] A logical consistency evaluation mechanism is introduced, which uses a logical consistency loss function to determine whether there are operational logic conflicts that violate physical sequence constraints between each test phase;

[0021] A human-machine collaborative optimization mechanism is introduced, which allows users to modify, block, or suggest replacements of the generated satellite test content through the interface, and adjust the structure of the prompt words to regenerate the content.

[0022] The generative model adopts an overall structure of semantic encoder, graph conditional decoder and constraint decoder, and injects pattern constraints, physical temporal constraints and resource constraints during the generation process. The training objective is mainly instruction learning, combined with structural consistency loss, temporal consistency loss, coverage regularization, risk suppression loss and pattern alignment loss to achieve synergistic optimization of readability, executability and engineering feasibility.

[0023] Furthermore, step S103 includes:

[0024] Virtual simulation models of digital twin satellites at the subsystem level are built, covering parameterized models and interface specifications of structural / power / thermal control / attitude and orbit control / propulsion / telemetry and remote control / onboard electronics / payload subsystems, and historical on-orbit or ground test data are introduced to correct structural parameters and learn residual errors.

[0025] The generated satellite test plan is automatically converted into a simulation input script, which configures test boundary conditions, initializes system state and sets execution timing. The simulation engine executes the operation according to the process and records key parameters.

[0026] A multidimensional data analysis method was used to perform index normalization, error tolerance analysis, and performance distribution comparison on the simulation output data.

[0027] Each subsystem model is parametrically modeled based on the physical characteristics of the real satellite. The thermal control model considers the thermal conductivity of materials, solar irradiance and shadow coverage, while the attitude model considers the inertia matrix and the response time of the actuators. The simulation engine executes each operation according to the set process. The system records key parameters such as temperature curve, attitude deviation, signal strength and stress distribution at each time step, which serve as the basis for evaluating the testing strategy.

[0028] Furthermore, step S104 includes:

[0029] A satellite data-real fusion test scheme evaluation system was constructed, and each candidate scheme was quantitatively evaluated from multiple dimensions such as test coverage, system risk, resource consumption, execution complexity and compatibility.

[0030] A multi-index evaluation algorithm is used to calculate the comprehensive score of each candidate scheme by combining the weights set for each index. The scheme with the highest comprehensive score is selected as the recommended optimal satellite testing strategy. The weights of each index are set according to the mission type. Test coverage reflects the completeness of mission completion, system risk reflects the length of the fault propagation path, resource consumption reflects the test time and power consumption, execution complexity reflects the difficulty of operation, and compatibility reflects the degree of matching with the existing system.

[0031] Furthermore, step S104 also includes:

[0032] For issues related to timing arrangements, resource reuse strategies, or fault-tolerant paths among multiple schemes, we introduce quadratic constraint optimization or reinforcement learning modeling, and use a combination of search strategy and optimization modeling methods to select the globally optimal solution as the recommended optimal satellite testing strategy while satisfying the constraints.

[0033] Furthermore, step S105 includes:

[0034] Collect telemetry data, sensor readings, and environmental records during the actual satellite testing process. Use residual modeling methods to compare the simulation output of the digital twin satellite model with the actual test results, establish an error function or error distribution model, and perform reverse correction on the input parameters, boundary conditions, or internal structure of the digital twin satellite model.

[0035] The feedback data from the actual satellite tests are used to strengthen the generative model, establish a mechanism to punish unreasonable paths and reward highly robust patterns, and realize a dynamic closed-loop evolution from scheme generation, simulation verification, actual testing to model updates.

[0036] Specifically, by comparing simulation outputs with actual observation results using residual modeling methods, a residual model is constructed to fit the error distribution; simulation model parameters are corrected in reverse by minimizing the sum of squared residuals, and key parameters with significant impact on error are adjusted first; a feedback mechanism is established to punish unreasonable generation paths and reward highly robust modes, thereby reducing the output probability of high-risk generation paths and gradually forming an intelligent satellite testing system with self-correction, self-adaptation, and self-evolution characteristics.

[0037] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned generative model-driven satellite data-real fusion test scheme generation method.

[0038] Thirdly, the present invention provides a computer-readable storage medium storing executable instructions thereon, which, when executed by a processor, enable the processor to implement the aforementioned generative model-driven satellite data-real fusion test scheme generation method.

[0039] The beneficial effects of this invention are as follows:

[0040] Improve the efficiency and comprehensiveness of solution generation: Through knowledge graphs and generative models, a large number of structured test solutions that conform to engineering specifications can be automatically and quickly generated based on test requirements. This effectively avoids scenario omissions and logical conflicts that may occur in manual design, and significantly improves design efficiency and solution coverage.

[0041] Enhanced feasibility and reliability: The generated schemes are virtually simulated and verified using high-fidelity digital twin satellite models. This allows for the identification of potential risks and evaluation of scheme effectiveness before actual testing, ensuring that candidate schemes are physically feasible and significantly reducing the probability of failure and mission interruption during physical testing.

[0042] Achieving intelligent decision-making and dynamic optimization: A multi-indicator decision-making method is employed to quantitatively evaluate and intelligently recommend feasible solutions, assisting engineers in selecting the optimal strategy. Simultaneously, continuous feedback and iterative correction are applied to the digital twin satellite model and the generated model using real test data, forming a complete closed loop of "generation-verification-testing-evolution." This enables the system to possess self-learning, self-correction, and adaptive capabilities, continuously optimizing solution quality and model accuracy. Attached Figure Description

[0043] Figure 1 This is a flowchart of a generative model-driven satellite data-real fusion test scheme generation method according to the present invention;

[0044] Figure 2 This is a diagram illustrating the satellite test model evolution method based on data-real fusion of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0046] like Figure 1 The diagram shows a flowchart of a generative model-driven satellite data-real fusion test scheme generation method according to the present invention. The method is illustrated using a typical satellite on-orbit thrust test as an example. Specifically, the method includes:

[0047] Step S101, Satellite test task understanding based on knowledge graph and semantic modeling: Construct a satellite test knowledge graph oriented towards the generation task, and model the semantic structure of the satellite test task;

[0048] Step S102: Automatic construction of satellite test scheme set driven by generative model: Based on the tester's test requirements, automatically construct a satellite test scheme set that conforms to engineering specifications using generative model;

[0049] Step S103: Feasibility verification of satellite testing scheme based on digital twin: Use digital twin satellite model to simulate the response of the test scheme set in virtual test scenario, evaluate the scheme effect and identify potential risks;

[0050] Step S104: Scheme selection and intelligent optimization based on multi-index decision-making: Intelligent evaluation of all feasible candidate schemes that have passed simulation verification, and recommendation of the optimal testing strategy;

[0051] Step S105: Evolution of satellite test model based on data-real fusion: Conduct real tests using the optimal testing strategy, correct the digital twin satellite model based on the obtained real test data, and feed the evaluation results back to the generative model to realize the iterative learning and evolution of the satellite test model.

[0052] Step S101 is a fundamental step in the entire system. It ensures that the generative model possesses a referable engineering semantic context, enhancing the rationality and engineering adaptability of the generated content. This step utilizes natural language processing and knowledge engineering techniques to extract and organize the core semantic units and structural logic in the satellite data-real fusion testing process. Specific implementation includes:

[0053] ① Construct a unified knowledge extraction and organization framework for satellite data-real fusion testing. The tasks, conditions, and parameters involved in satellite data-real fusion testing are widely distributed across various documents, including satellite test procedures, intra-satellite / satellite-ground / inter-satellite system interface documents, ground / on-orbit test record sheets, and fault diagnosis reports for structural / power / thermal control / attitude and orbit control / propulsion / telemetry and remote control / on-board electronics / payload subsystems. Through information extraction technology, key entities are identified from these documents, including satellite test items, operating procedures, virtual on-board environment and structural parameter constraints, and satellite subsystem equipment composition. A semantic relationship network between these entities is established, thereby constructing a hierarchical satellite test knowledge graph. This knowledge graph provides a structural context for generative models, avoiding common-sense errors or engineering taboos in the generated schemes. In constructing a knowledge graph oriented towards test scheme generation, each graph node represents a satellite test object or step with engineering significance, and each edge represents the logical or causal relationship between them. Satellite Test Group Semantic relationships were used to model the testing process, among which As the starting test entity, For the target response entity, This represents the action relationship between the two entities. To achieve semantic reasoning, a graph embedding model is used to vectorize the entities and relationships. Definition To test entity vectors, To test relation vectors, the basic scoring function for the knowledge graph is: ,in, A smaller value indicates a more reasonable test group. During the convergence training of the system's knowledge structure, the following loss function in the overall knowledge base can be minimized:

[0054] ,

[0055] in, It is a constructed negative sample set. It consists of randomly replaced starting test entities, action relationships, and target response entities. These are boundary hyperparameters;

[0056] ② Semantic alignment of multimodal data in satellite data-real fusion testing schemes. The engineering documents generated from actual satellite data-real fusion tests contain highly information-dense data charts, flowcharts, and structural diagrams. Therefore, it is necessary to employ image-text alignment and structural understanding techniques to achieve consistent encoding of information and textual semantics. This step enhances the system's ability to reason about the structure of non-textual information, enabling the generative model to "understand" the engineering meaning behind the graphical content and use this information to guide the reasoning process during generation.

[0057] Step S102 transforms the test objective, platform parameters, and constraints into semantic expressions, and uses generative artificial intelligence technology to output a structured and executable test process. Specific implementation includes:

[0058] ① The generative model adopts an overall structure of semantic encoder, graph conditional decoder, and constraint decoder: The semantic encoder vectorizes the mission type, on-orbit platform model, and operational constraints, and aligns them with the entity / relationship embeddings of the satellite test knowledge graph to form a mission context representation; Under the joint constraints of the mission context and conditional subgraphs, the graph conditional decoder hierarchically generates a structured sequence of stages, steps, parameters, and expected responses; The constraint decoder injects pattern constraints (field order, key-value type), physical timing constraints (pressurization before ignition, attitude control activation before thrust application), and resource constraints (power, propellant, thermal budget) during the generation process; The training objective is mainly instruction learning, combined with structural consistency loss, temporal consistency loss, coverage regularization, risk suppression loss, and pattern alignment loss to achieve synergistic optimization of readability, executability, and engineering feasibility; Incremental learning uses simulation playback and real experimental data as input to continuously fine-tune the model to maintain its adaptability to new missions and platforms.

[0059] ② The input design employs a parametric prompt structure. Testers can input task type, platform model, and specific constraints through the interface. These inputs will be constructed into standardized prompt templates to guide the generative model in understanding the test objectives and constraints. By flexibly constructing the prompt word structure, the model can adapt to various input scenarios, including multi-task requirements, complex boundary conditions, and collaborative testing of multiple subsystems.

[0060] ③ The generation mechanism employs a structure-aware text generation method. Unlike traditional natural language generation, the satellite testing scheme is essentially a structured, multi-stage decision-making problem. Therefore, the generation module is designed to support hierarchical output, with each testing stage corresponding to a sub-module. Each module contains key fields such as the test object, input parameters, execution operations, and expected response. Simultaneously, a structural reasoning mechanism is introduced to ensure that the output step sequence conforms to physical feasibility and engineering process dependencies, avoiding logical conflicts. The output results can be directly used for simulation platform deployment or manually fine-tuned by engineers. The satellite testing scheme can be formally defined as a time-sequential chain of operations: ,in, Indicates the first The testing phase includes the following elements: , The subsystems tested in this phase, For the input parameter vector, such as , Initial temperature (unit: °C). Valve pre-charge pressure (unit: Pa). Expected delay between operation and feedback (in seconds). For the desired response, such as attitude change Thermal equilibrium time Further define a logical consistency loss function to evaluate whether the generated results satisfy the sequential dependency condition: ,in Return whether it violates the physical order constraint. This is an indicator function; its value is 1 if the condition is violated, and 0 otherwise, to determine whether there is a conflict in the operational logic between the two stages.

[0061] ④ To ensure the controllability and reliability of the generated results, a human-machine collaborative optimization mechanism is introduced. After each generation, engineers can modify, disable, or suggest replacements for certain content through the interface, and the system will automatically adjust the structure of the prompts and regenerate that part of the content. This approach balances automation efficiency with the engineer's subjective judgment, making the generation system both intelligent and flexible for practical applications while maintaining engineering security.

[0062] In step S103, the digital twin platform plays a crucial role, undertaking the entire process from solution execution to data collection and performance evaluation, ensuring that candidate solutions entering the evaluation stage have physical feasibility. Specific implementation includes:

[0063] ① In terms of model construction, virtual simulation models at the subsystem level are built for the digital twin satellite. These include subsystems for structure, power supply, thermal control, attitude and orbit control, propulsion, telemetry and remote control, onboard electronics, and payload. Each model is parametrically modeled based on the actual physical characteristics of the satellite. For example, the thermal control model considers material thermal conductivity, solar irradiance, and shadow coverage, while the attitude model considers the inertia matrix and actuator response time. These models will be encapsulated as standard interface components, which can be uniformly scheduled and combined to form mission-level simulation scenarios.

[0064] ② Regarding the execution of the plan, an automated scenario generation and test scheduling approach is adopted. The test plan output by the generated model is automatically converted into simulation input scripts by tools, configuring test boundary conditions, initializing system states, and setting execution timing. The simulation engine will execute each operation according to the set process, and the system records key parameters at each time step, including indicators such as temperature curves, attitude deviations, signal strength, and stress distribution. This data will serve as the basis for evaluating the test strategy and determining whether it meets the requirements for stability, safety, and task completion.

[0065] ③ In terms of simulation evaluation, a multi-dimensional data analysis method is adopted. Through index normalization, error tolerance analysis, and performance distribution comparison methods, unstable operating parameters, abnormal response locations, and potential fault causes are identified. Simultaneously, based on mature mathematical tools, the data results are visualized, using graphs, tables, and curves to facilitate engineers' intuitive understanding of the simulation results. Step S103 achieves a closed-loop process from logical verification of the test plan to physical feasibility verification;

[0066] Step S104 provides engineers with decision-making support, balancing task completion efficiency with test risk management. Specific implementation includes:

[0067] ① Construct a scoring system for satellite data-real fusion testing schemes. For each candidate scheme, quantitative scoring is performed from multiple dimensions, including test coverage, system risk, resource consumption, execution complexity, and compatibility. Weights can be assigned to each indicator based on the task type, and a comprehensive scoring model is constructed using a multi-indicator evaluation algorithm. Each candidate scheme... A comprehensive score will be assigned. It consists of multiple normalized indicators. ,in: For the plan In indicators The scores were normalized to , As an indicator The weights satisfy Commonly used indicators include Task coverage Risk tolerance reflects the length of the fault propagation path. Resource efficiency reflects the test time and power consumption. System stability reflects the response to maximum disturbance and the recovery speed. The scheme with the highest score was ultimately selected. Considering that time-deployment tasks prioritize rapid execution, while security assurance tasks place greater emphasis on the completeness of redundancy strategies, testers can set personalized target weights according to the requirements of the testing tasks;

[0068] ② Recommendation of solution combinations based on policy optimization algorithms. For problems involving the timing arrangement of multiple solutions, resource reuse strategies, and fault-tolerant paths, quadratic constraint optimization or reinforcement learning modeling can be introduced, and a combination selection method based on search strategy and optimization modeling can be adopted. The globally optimal solution is selected under the premise of satisfying constraints, which is particularly suitable for complex tasks such as constellation testing, multi-satellite parallel testing, or limited ground resource allocation. Finally, a list of candidate solutions is output in a ranked manner, with scoring explanations and risk level hints provided to assist testers in making accurate and transparent test solution decisions.

[0069] Step S105 is crucial for achieving "virtual-real consistency," such as... Figure 2 As shown, the specific implementation includes:

[0070] ① After the recommended scheme is executed in actual testing, telemetry data, sensor readings, and environmental records are collected throughout the process and used to correct the data-real fusion error of the digital twin satellite model. Residual modeling methods are used to compare the simulation output with the actual test results, establishing an error function or error distribution model, and making reverse corrections to the simulation input parameters, boundary conditions, or internal structure of the model. This process supports parameter sensitivity analysis, prioritizing adjustments to key parameters with significant error impact, such as heat transfer boundary conditions and system delay terms. This mechanism improves the accuracy of the digital twin satellite in step S103 and reduces the risk of deviation from the generated scheme during actual execution. Let the digital twin output be: The actual test observations are as follows: The residual is then defined as: To correct the simulation model parameters, a residual model can be constructed, assuming that the residuals can be fitted to a function within a certain time interval. , ,in The basis functions are time-varying Gaussian kernels or polynomial bases. For the weight to be determined, Using the index of the basis functions, the sum of squared residuals is used as the objective function. Minimize the objective function to inversely correct the simulation model or to correct the bias distribution of the generated model, thereby continuously improving the accuracy of the twin satellite model;

[0071] ② Use feedback data to strengthen the generative model and achieve a closed-loop evolution of the generative model structure. As real test data is continuously fed back and iterated, establish a mechanism to punish unreasonable paths and reward highly robust patterns, reduce the output probability of "high-risk generation paths", and gradually realize a dynamic closed-loop path from "scheme generation → simulation verification → actual testing → model update", forming a cycle from data to model and then to generation, continuously improving the model's adaptability to the real mission environment, and ultimately forming an intelligent satellite testing system with self-correction, self-adaptation and self-evolution characteristics.

[0072] In summary, this invention discloses a generative model-driven method for generating satellite data-real fusion test schemes, including designing a generative model-driven satellite data-real fusion test scheme generation process and designing a closed-loop self-evolving satellite data-real fusion test scheme verification method. This method can solve the problems existing in the current satellite test scheme formulation to a certain extent and improve the intelligence level and efficiency of satellite mission test scheme generation.

[0073] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned generative model-driven satellite data-real fusion test scheme generation method.

[0074] Thirdly, the present invention provides a computer-readable storage medium storing executable instructions thereon, which, when executed by a processor, enable the processor to implement the aforementioned generative model-driven satellite data-real fusion test scheme generation method.

[0075] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A generative model-driven method for generating satellite data-real fusion test schemes, characterized in that, The method includes: Step S101: Understanding satellite testing tasks based on knowledge graphs and semantic modeling: Construct a satellite testing knowledge graph and perform semantic structured modeling of satellite testing tasks; Step S102: Automatic construction of satellite test scheme set driven by generative model: Based on the test requirements, automatically construct a satellite test scheme set that conforms to engineering specifications using generative model; Step S103: Feasibility verification of satellite testing scheme based on digital twin: Use digital twin satellite model to simulate the response of the test scheme set in virtual test scenario, evaluate the scheme effect and identify potential risks; Step S104: Scheme selection and intelligent optimization based on multi-index decision-making: Intelligent evaluation of all candidate schemes that have passed simulation verification, and recommendation of the optimal testing strategy; Step S105: Evolution of satellite test model based on data-real fusion: Conduct real tests using the optimal testing strategy, correct the digital twin satellite model based on the obtained real test data, and feed the evaluation results back to the generative model to realize the iterative learning and evolution of the satellite test model.

2. The generative model-driven satellite data-real fusion test scheme generation method according to claim 1, characterized in that, In step S101, constructing the satellite test knowledge graph includes: Key entities are extracted from satellite test procedures, intra-satellite / satellite-ground / inter-satellite system interface documents, ground / on-orbit test record sheets, and fault diagnosis reports of structural / power / thermal control / attitude and orbit control / propulsion / telemetry and remote control / on-board electronics / payload subsystems. These key entities include satellite test items, operating procedures, virtual on-board environment and structural parameter constraints, and satellite subsystem equipment composition. A semantic relationship network between entities is established to construct a hierarchical satellite test knowledge graph. The knowledge graph uses triples to model the semantic relationships of the test process. Each triple includes an initiating test entity, an action relationship, and a target response entity, supporting the structured expression and cross-stage reasoning of satellite test tasks.

3. The generative model-driven satellite data-real fusion test scheme generation method according to claim 2, characterized in that, Step S101 further includes: The entity and action relationships in the satellite test knowledge graph are vectorized using a graph embedding model; the structure of the knowledge graph is trained and optimized by minimizing a loss function based on vector representation, which is used to evaluate the rationality of triples and promote the convergence of the knowledge structure.

4. The method for generating a satellite data-real fusion test scheme driven by a generative model according to claim 1, characterized in that, Step S102 includes: A parameterized prompting structure is adopted to construct standardized prompting templates from the task type, satellite platform model, and specific constraints input by the tester, in order to guide the generative model; A structure-aware text generation method is adopted to output a hierarchical satellite test plan, with each test phase including the test object, input parameters, execution operation and expected response. A logical consistency evaluation mechanism is introduced, which uses a logical consistency loss function to determine whether there are operational logic conflicts that violate physical sequence constraints between each test phase; A human-machine collaborative optimization mechanism is introduced, which allows users to modify, block, or suggest replacements of the generated satellite test content through the interface, and adjust the structure of the prompt words to regenerate the content. The generative model adopts an overall structure of semantic encoder, graph conditional decoder and constraint decoder, and injects pattern constraints, physical temporal constraints and resource constraints during the generation process. The training objective is mainly instruction learning, combined with structural consistency loss, temporal consistency loss, coverage regularization, risk suppression loss and pattern alignment loss to achieve synergistic optimization of readability, executability and engineering feasibility.

5. The method for generating a satellite data-real fusion test scheme driven by a generative model according to claim 1, characterized in that, Step S103 includes: Virtual simulation models of digital twin satellites at the subsystem level are built, covering parameterized models and interface specifications of structural / power / thermal control / attitude and orbit control / propulsion / telemetry and remote control / onboard electronics / payload subsystems, and historical on-orbit or ground test data are introduced to correct structural parameters and learn residual errors. The generated satellite test plan is automatically converted into a simulation input script, which configures test boundary conditions, initializes system state and sets execution timing. The simulation engine executes the operation according to the process and records key parameters. A multidimensional data analysis method was used to perform index normalization, error tolerance analysis, and performance distribution comparison on the simulation output data. Each subsystem model is parametrically modeled based on the physical characteristics of the real satellite. The thermal control model considers the thermal conductivity of materials, solar irradiance and shadow coverage, while the attitude model considers the inertia matrix and the response time of the actuators. The simulation engine executes each operation according to the set process. The system records key parameters such as temperature curve, attitude deviation, signal strength and stress distribution at each time step, which serve as the basis for evaluating the testing strategy.

6. The generative model-driven satellite data-real fusion test scheme generation method according to claim 1, characterized in that, Step S104 includes: A satellite data-real fusion test scheme evaluation system was constructed, and each candidate scheme was quantitatively evaluated from multiple dimensions such as test coverage, system risk, resource consumption, execution complexity and compatibility. A multi-index evaluation algorithm is used to calculate the comprehensive score of each candidate scheme by combining the weights set for each index. The scheme with the highest comprehensive score is selected as the recommended optimal satellite testing strategy. The weights of each index are set according to the mission type. Test coverage reflects the completeness of mission completion, system risk reflects the length of the fault propagation path, resource consumption reflects the test time and power consumption, execution complexity reflects the difficulty of operation, and compatibility reflects the degree of matching with the existing system.

7. The generative model-driven satellite data-real fusion test scheme generation method according to claim 6, characterized in that, Step S104 further includes: For issues related to timing arrangements, resource reuse strategies, or fault-tolerant paths among multiple schemes, we introduce quadratic constraint optimization or reinforcement learning modeling, and use a combination of search strategy and optimization modeling methods to select the globally optimal solution as the recommended optimal satellite testing strategy while satisfying the constraints.

8. The generative model-driven satellite data-real fusion test scheme generation method according to claim 1, characterized in that, Step S105 includes: Collect telemetry data, sensor readings, and environmental records during the actual satellite testing process. Use residual modeling methods to compare the simulation output of the digital twin satellite model with the actual test results, establish an error function or error distribution model, and perform reverse correction on the input parameters, boundary conditions, or internal structure of the digital twin satellite model. The feedback data from the actual satellite tests are used to strengthen the generative model, establish a mechanism to punish unreasonable paths and reward highly robust patterns, and realize a dynamic closed-loop evolution from scheme generation, simulation verification, actual testing to model updates. Specifically, by comparing simulation outputs with actual observation results using residual modeling methods, a residual model is constructed to fit the error distribution; simulation model parameters are corrected in reverse by minimizing the sum of squared residuals, and key parameters with significant impact on error are adjusted first; a feedback mechanism is established to punish unreasonable generation paths and reward highly robust modes, thereby reducing the output probability of high-risk generation paths and gradually forming an intelligent satellite testing system with self-correction, self-adaptation, and self-evolution characteristics.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the generative model-driven satellite data-real fusion test scheme generation method according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement a generative model-driven satellite data-real fusion test scheme generation method as described in any one of claims 1-8.

Images