Satellite multi-working-condition number-real fusion test method based on transfer learning and digital twinning
By employing a method based on transfer learning and digital twins, the problems of relying on human experience and insufficient integration of digital test models in the formulation of satellite test plans have been solved, enabling efficient, accurate, and adaptive digital testing of satellites under multiple operating conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-03-20
AI Technical Summary
In the current technology, the formulation of satellite test schemes relies on human experience and lacks effective integration of digital test models, making it difficult to cover various working conditions. Furthermore, digital test models lack a mechanism for updating mechanisms and cannot adapt to multiple working conditions.
By employing a method based on transfer learning and digital twins, and through modules for generating and evaluating experimental schemes, evaluating and correcting digital twin models, conducting multi-condition digital experiments, and driving experimental verification and evaluation based on data-real fusion, satellite multi-condition data-real fusion experiments are achieved.
It improves the efficiency and capability of satellite multi-condition testing, reduces the cost of test scheme development, and enhances the accuracy and adaptability of digital testing, enabling effective satellite digital testing under different conditions.
Smart Images
Figure CN121706411A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of aerospace digitalization and computer science, specifically relating to a satellite multi-condition data-real fusion test method based on transfer learning and digital twins. Background Technology
[0002] With the rapid development of satellite projects such as satellite internet constellations, integrated communication, navigation, and remote sensing satellite constellations, and complex mission scientific satellites, improving satellite testing efficiency and facilitating rapid satellite model development has become a key focus of the satellite industry. For test subjects such as low-Earth orbit satellite inter-satellite communication, flexible solar array deployment and power generation, and on-orbit electric propulsion and attitude control, actual on-orbit testing is required to verify performance. However, due to the high cost and limited number of on-orbit tests, it is difficult to achieve comprehensive physical testing covering various operating conditions. Leveraging the data-real-data fusion satellite testing method, fully utilizing the low-cost and high-efficiency advantages of digital testing, and using data from a limited number of physical on-orbit tests, to clearly explore the performance boundaries under various complex coupled operating conditions is the key objective of conducting multi-condition data-real-data fusion satellite tests.
[0003] The current challenges in achieving multi-condition data-physical fusion testing of satellites include: ① The current satellite test scheme development relies heavily on manual experience, resulting in weak batch generation capabilities and difficulty in supporting the development of satellite test schemes covering various operating conditions; ② The current satellite physical and digital tests lack effective integration. Digital test models are mostly static models from the design phase, lacking dynamic update and correction mechanisms and failing to fully utilize physical test data; ③ Conducting satellite digital tests requires building digital models that reflect the corresponding operating conditions. For parts with clear mechanisms such as control, electrical, and mechanical aspects, corresponding mechanism models can be built. For parts with unclear mechanisms such as electromagnetic, radiation, and thermal aspects, corresponding data-driven models can be built. However, there is currently a lack of methods to support the construction of satellite digital test models that cover both mechanism models and data-driven models; ④ Furthermore, how to adapt satellite digital test models to different operating conditions based on the integration of mechanism models and data-driven models, and further conduct multi-condition satellite digital tests, is a problem that needs to be solved. Summary of the Invention
[0004] To address the aforementioned problems, this invention, based on digital twin theory and transfer learning technology, improves the satellite multi-mode data-real fusion test method, thereby enhancing the capability and efficiency of satellite multi-mode testing to a certain extent. The technical problem solved by this invention is achieved through the following technical solution: a satellite multi-mode data-real fusion test method based on transfer learning and digital twins, the specific process of which is as follows:
[0005] Step S101, Test scheme generation and evaluation module, generates and filters specific test schemes for multi-condition tests in batches based on satellite test scheme generative model and satellite digital twin model;
[0006] Step S102, Digital Twin Model Evaluation and Correction Module, uses physical satellite test data as a driving force to determine whether satellite digital twin model correction is needed, and completes the relevant model correction as appropriate;
[0007] Step S103: The multi-condition digital test module based on digital twin and transfer learning integrates the satellite digital twin model and the multi-condition performance prediction model trained by transfer learning to realize satellite digital test under multiple conditions.
[0008] Step S104, the experimental verification and evaluation module driven by data-real fusion, verifies and evaluates the satellite digital test results under multiple operating conditions by establishing an experimental verification and evaluation index system and building a data-real fusion experimental verification scenario.
[0009] More preferably, the test scheme generation and evaluation module in step S101 designs random variables for batch generation of test schemes based on satellite test requirements. It uses a satellite test scheme generative model trained with historical satellite test schemes, historical mission texts, and expert knowledge. Constrained by test mission boundaries and satellite capabilities, the random variables are input into the generative model to batch generate satellite test schemes. The schemes are then evaluated and selected using a satellite digital twin model to obtain a set of satellite test schemes for multi-condition testing. Specifically, this includes:
[0010] ① Based on the requirements of satellite experiments, design random variables for batch generation of satellite experiment schemes, taking into account aspects such as experiment subjects, experiment scenarios, satellite status, experiment process, and experiment parameters;
[0011] ② Train a generative model of satellite test schemes by using historical satellite test schemes, historical mission texts, and satellite expert knowledge. Generate random variables to input into the generative model of satellite test schemes with test mission boundaries and satellite capabilities as constraints, and generate multi-condition satellite test schemes in batches.
[0012] ③ Based on the satellite digital twin model, the generated multi-condition satellite test schemes are evaluated and screened according to the rationality of the test mission objectives, the feasibility of the process, and the representativeness of the results, forming a multi-condition test scheme set that includes physical test schemes and digital test schemes. The physical test schemes guide the execution of physical satellite on-orbit tests or ground tests, while the digital test schemes guide the execution of satellite digital tests.
[0013] More preferably, the digital twin model evaluation and correction module in step S102 identifies the characteristic data of the satellite digital twin model, determines the model evolution and update mechanism, and, based on physical test data obtained from on-orbit or ground tests, judges whether the model needs to be updated and corrected, and corrects the satellite digital twin model that needs correction. Specifically, this includes:
[0014] ① Based on the satellite's operating mechanism, performance evolution law, and expert knowledge, analyze the key characteristic parameters related to satellite performance, and use the multi-factor factorization method to analyze the related characteristic parameters of the satellite in various time-varying processes, preliminarily define the allowable fluctuation range of the characteristic parameters, and establish a model update triggering mechanism;
[0015] ② Input the relevant parameters of the physical experiment into the satellite digital twin model to carry out simulation calculations, and calculate the deviation between the simulation results data and the physical experiment data;
[0016] ③ Determine whether the deviation between the satellite digital twin model and the actual physical satellite exceeds the allowable fluctuation range. If it does, it is determined that it needs to be corrected and jumps to step ④ of S102. If it does not exceed the allowable fluctuation range, jumps to step ① of S103.
[0017] ④ Based on physical test data, the multi-dimensional satellite digital twin model is corrected, and the corrected satellite digital twin model is evaluated and iterated until the model correction is completed.
[0018] More preferably, step S103, based on a multi-condition digital test module using digital twins and transfer learning, firstly designs the digital test architecture and clarifies the required models according to the digital test plan. For parts with clear physical mechanisms, a satellite digital twin model is built, relevant satellite digital twin models are called, and the model assembly, integration, and interface adaptation are completed. For parts with unclear physical mechanisms, a data-driven model is built, historical relevant performance prediction models are called, and performance prediction models under multiple conditions are trained based on transfer learning, combined with physical test data and simulation data. Corresponding interfaces are then developed and adapted. Next, the satellite digital twin model and the multi-condition performance prediction model are integrated using the developed interfaces to build a satellite digital test scenario and conduct multi-condition satellite digital tests. Specifically, this includes:
[0019] ① Based on the digital experiment plan and model, design the digital experiment architecture, including model requirements, interface format, interaction form, calculation process, output format, and call the relevant satellite digital twin model and performance prediction model according to the model requirements;
[0020] ② Based on the digital experiment architecture design, assemble and integrate the assemblable satellite digital twin model;
[0021] ③ Based on the interface format and interaction method, develop the corresponding interface for the satellite digital twin model to adapt the data interaction between the satellite digital twin model and the multi-condition performance prediction model;
[0022] ④ Using the corrected satellite digital twin model, simulate and generate multi-condition satellite performance simulation data. Combine physical test data, multi-condition satellite performance simulation data and related performance prediction models, and train the multi-condition performance prediction model based on transfer learning.
[0023] ⑤ Based on the interface format and interaction form design, develop the corresponding interface for the multi-condition performance prediction model to adapt to the data interaction between the multi-condition performance prediction model and the satellite digital twin model;
[0024] ⑥ By leveraging the developed interface, the satellite digital twin model and multi-condition performance prediction model are integrated to build a satellite digital test scenario for multi-condition digital testing;
[0025] ⑦ Based on the digital test scheme and satellite digital test scenario, conduct multi-condition satellite digital tests and provide digital test results under multiple conditions.
[0026] More preferably, step S104, the data-real fusion driven test verification and evaluation module, constructs an index system for satellite multi-condition test verification and evaluation and designs corresponding evaluation methods. Guided by the index system and evaluation methods, it builds a data-real fusion test verification scenario, injects various operating conditions into the data-real fusion test verification scenario, analyzes and evaluates the results of multi-condition satellite digital tests, and finally provides a comprehensive evaluation result of satellite performance under various operating conditions. Specifically, this includes:
[0027] ① Construct a satellite multi-condition test verification and evaluation index system, including satellite digital twin model test verification and evaluation index, multi-condition performance prediction model test verification and evaluation index, and integrated test verification and evaluation index, and design corresponding analysis and evaluation methods;
[0028] ② Guided by the indicator system and evaluation methods, we build a data-real integrated test and verification scenario. The physical part is a physical equivalent test bench or physical satellite system, and the virtual part is a digital twin model of the environment and satellite. We integrate the two through an interface to build a data-real integrated semi-physical satellite test and verification scenario.
[0029] ③According to the test plan, multi-condition data are injected into the data-real fusion test verification scenario to carry out test verification and obtain data-real fusion test verification data;
[0030] ④ Based on the evaluation method, the results of the satellite digital test under various operating conditions are analyzed and evaluated in combination with the data from the data-real fusion test verification, and a comprehensive evaluation result of the satellite performance under various operating conditions is given.
[0031] Even better, multi-condition data fusion tests include tests on the satellite's electric propulsion performance.
[0032] Even better, the on-orbit electric propulsion performance test of the satellite completes a comprehensive evaluation of the electric propulsion performance by carrying out on-orbit attitude control, orbit maintenance, and orbit change control under multiple operating conditions.
[0033] Better digital twin models related to satellite electric propulsion performance include structural models, control subsystem models, electrical subsystem models, and electric propulsion subsystem models.
[0034] Even better, the physical part includes the onboard computer, electric propulsion test bench, and satellite subsystem equivalent, while the virtual part includes the space environment and digital twin model of the satellite subsystem. The two are integrated through an interface to build a semi-physical satellite electric propulsion performance test and verification scenario that integrates data and reality.
[0035] Even better, the satellite digital twin model is corrected based on physical test data, specifically from the geometric, physical, and behavioral dimensions. The geometric dimension describes the geometry, size, and assembly relationships of the satellite and its subsystems, and its correction is achieved by updating geometric parameters and structural features. The physical dimension describes the physical mechanisms of the satellite and its subsystems in the fields of mechanics, electronics, and thermodynamics, and its correction is achieved by updating parameters and correcting functions based on the model's deviation characteristics. The behavioral dimension describes the behavior of the satellite and its subsystems in space motion, data interaction, and functional operations, and its correction is achieved by updating behavioral descriptions and interaction parameters.
[0036] The advantages of this invention compared to the prior art are:
[0037] (1) Design a test scheme generation and evaluation module. Using historical satellite test schemes, historical mission texts and expert knowledge-trained satellite test scheme generative model, input random variables for batch generation of test schemes into the generative model. With test mission boundaries and satellite capabilities as constraints, batch generate satellite test schemes for multiple operating conditions. Then, use satellite digital twin model to evaluate and screen the schemes to form a test scheme set. This test scheme generation and evaluation method driven by generative model and digital twin model can reduce the cost of manual writing and evaluation of test schemes and improve the efficiency of multi-operating condition satellite test scheme formulation.
[0038] (2) Design a digital twin model evaluation and correction module, set up a model evolution and update mechanism based on feature data identification, and use physical test data obtained from physical satellite on-orbit tests or ground tests as the driving force to update and correct the satellite digital twin model, effectively utilize physical test data and improve the accuracy of the satellite digital twin model, and ensure the credibility of satellite digital tests.
[0039] (3) Design a multi-condition digital test module based on digital twins and transfer learning. Based on the digital test architecture, design and call the satellite digital twin model and the data-driven performance prediction model required by the mechanism. Realize the assembly and integration of the satellite digital twin model and the training of the performance prediction model based on transfer learning respectively. Develop the model interface and use the developed interface to realize the integration of the satellite digital twin model and the performance prediction model, thereby building a satellite digital test scenario that is suitable for multiple conditions and carrying out multi-condition satellite digital tests. This method realizes the effective integration of the satellite digital twin model and the performance prediction model and can adapt to different working conditions, effectively supporting the carrying out of multi-condition satellite digital tests and improving the multi-condition satellite test capability to a certain extent. Attached Figure Description
[0040] Figure 1 This is a flowchart of a satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to the present invention. Detailed Implementation
[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above objectives, this invention adopts the following technical solution.
[0042] According to one embodiment of the present invention, a satellite multi-condition data-real fusion test method based on transfer learning and digital twins is provided. Figure 1 This document presents a flowchart of a satellite multi-condition data-real fusion test method based on transfer learning and digital twins, as described in this invention. Using a typical satellite's electric propulsion performance test as an example, this type of test primarily conducts on-orbit attitude control, orbit maintenance, and orbit change control tests under multiple operating conditions to comprehensively evaluate the electric propulsion performance. By leveraging data-real fusion testing, it achieves coverage of more operating conditions, providing a comprehensive multi-condition satellite electric propulsion performance evaluation. Figure 1 As shown, the specific implementation methods include:
[0043] Step S101, Test scheme generation and evaluation module, generates and selects specific test schemes for multi-condition satellite electric propulsion performance tests in batches based on satellite test scheme generative model and satellite digital twin model;
[0044] Step S102, Digital Twin Model Evaluation and Correction Module, uses physical satellite test data as a driving force to determine whether satellite digital twin model correction is needed, and completes the relevant model correction as appropriate;
[0045] Step S103: The multi-condition digital test module based on digital twin and transfer learning integrates the satellite digital twin model and the multi-condition performance prediction model trained by transfer learning to realize the digital test of satellite electric propulsion performance under multiple conditions.
[0046] Step S104, the experimental verification and evaluation module driven by data-real fusion, verifies and evaluates the digital test results of satellite electric propulsion performance under multiple operating conditions by establishing an experimental verification and evaluation index system and building a data-real fusion experimental verification scenario.
[0047] The test scheme generation and evaluation module in step S101 designs random variables for batch generation of test schemes based on the requirements of satellite electric propulsion performance testing. It utilizes a satellite test scheme generative model trained with expert knowledge from historical satellite test schemes, historical satellite attitude and orbit control mission records, and the satellite as a whole and key subsystems. Constrained by the electric propulsion test mission boundaries, satellite and electric propulsion capabilities, the random variables are input into the generative model to generate satellite test schemes in batches. The schemes are then evaluated and selected using a satellite digital twin model to obtain a set of satellite test schemes for multi-condition testing. Specific implementation includes:
[0048] ① The on-orbit electric propulsion performance test of the satellite mainly completes a comprehensive evaluation of the electric propulsion performance by conducting on-orbit attitude control, orbit maintenance, and orbit change control tests under multiple operating conditions. Based on the requirements of the above-mentioned electric propulsion performance test, random variables for batch generation of satellite test schemes are designed from the aspects of test subjects, test scenarios, satellite status, test process, and test parameters. Among them, the test subjects are divided into attitude control, orbit maintenance, and orbit change control, etc. The test scenarios are designed with corresponding orbit and environmental random variables for the differences in orbit status. The satellite status is designed with relevant random variables from the health status and damage status of the entire satellite to the subsystems. The test process is designed with relevant random test steps and parameters for specific test subjects. The test parameters are designed with corresponding random configuration parameters for the specific status of the satellite.
[0049] ②Use historical satellite test schemes, historical satellite attitude and orbit control mission records, and expert knowledge of the entire satellite and key subsystems to train a generative model of satellite test schemes. Use the boundaries of electric propulsion test missions and the satellite and electric propulsion capabilities as constraints to generate random variables to input into the generative model of satellite test schemes, and generate multi-condition satellite electric propulsion performance test schemes in batches.
[0050] ③ Based on the existing satellite digital twin model, the generated multi-condition satellite electric propulsion performance test schemes are evaluated and screened according to the rationality of the test mission objectives, the feasibility of the process, and the representativeness of the results, forming a multi-condition test scheme set that includes physical test schemes and digital test schemes. The physical test schemes guide the execution of physical satellite on-orbit electric propulsion performance tests and ground electric propulsion performance tests, while the digital test schemes guide the execution of digital tests of satellite electric propulsion performance under various operating conditions.
[0051] The S102 digital twin model evaluation and correction module identifies the characteristic data of the satellite digital twin model related to the satellite electric propulsion performance digital test, determines the model evolution and update mechanism, and judges whether the model needs to be updated and corrected based on the physical test data obtained from the physical satellite on-orbit electric propulsion performance test or ground electric propulsion performance test, and corrects the satellite digital twin model that needs to be corrected. Specific implementation includes:
[0052] ① The digital twin model related to satellite electric propulsion performance mainly includes structural model, control subsystem model, electrical subsystem model, and electric propulsion subsystem model. Based on the satellite's operating mechanism, performance evolution law, and expert knowledge, the key characteristic parameters related to satellite electric propulsion performance are analyzed. The multi-factor factorization method is used to analyze the associated characteristic parameters of satellite electric propulsion in various time-varying processes. The allowable fluctuation range of characteristic parameters is preliminarily defined, and a model update triggering mechanism is established.
[0053] ②Inject relevant parameters from the on-orbit electric propulsion performance test or the ground electric propulsion performance test of the physical satellite into the satellite digital twin model to carry out simulation calculations, and calculate the deviation between the simulation results data and the physical test data;
[0054] ③ Determine whether the deviation between the satellite digital twin model and the actual physical satellite exceeds the allowable fluctuation range. If it does, it is determined that it needs to be corrected and jumps to step ④ of S102. If it does not exceed the allowable fluctuation range, jumps to step ① of S103.
[0055] ④ The satellite digital twin model is corrected based on physical experimental data, focusing on geometric, physical, and behavioral dimensions. The geometric dimension describes the satellite and its subsystems' geometry, size, and assembly relationships; correction primarily involves updating geometric parameters and structural features. The physical dimension describes the physical mechanisms of the satellite and its subsystems in mechanics, electronics, and thermal fields; correction mainly involves updating parameters and correcting functions based on the model's deviation characteristics. The behavioral dimension describes the satellite and its subsystems' behavior in space motion, data interaction, and functional operations; correction primarily involves updating behavioral descriptions and interaction parameters. After corrections are made across these multiple dimensions, the satellite digital twin model undergoes deviation calculation and evaluation again, completing the model correction iteratively.
[0056] Step S103, based on a multi-condition digital test module using digital twins and transfer learning, firstly designs the digital test architecture and clarifies the required models according to the digital test plan. For parts with clear physical mechanisms, a satellite digital twin model is built, relevant satellite digital twin models are called, and the model assembly, integration, and interface adaptation are completed. For parts with unclear physical mechanisms, a data-driven model is built, historical relevant performance prediction models are called, and combined with physical test data and simulation data, a multi-condition performance prediction model is trained based on transfer learning, and corresponding interfaces are adapted and developed. Then, the satellite digital twin model and the multi-condition performance prediction model are integrated using the developed interface to build a satellite digital test scenario and conduct multi-condition digital tests of satellite electric propulsion performance. Specific implementation includes:
[0057] ① Based on the digital test plan and model, design the digital test architecture, including model requirements, interface format, interaction method, calculation process, and output format. According to the characteristics of satellite electric propulsion performance testing, the required satellite digital twin models mainly include structural models, control subsystem models, electrical subsystem models, and electric propulsion subsystem models. The required performance prediction models mainly include battery subsystem charge / discharge performance prediction models, solar panel subsystem power generation performance prediction models, and thermal control subsystem thermal control performance prediction models. Design the interface format and interaction method for the above satellite digital twin models and performance prediction models based on the satellite's structural composition. Design the calculation process for the above satellite digital twin models and performance prediction models based on the satellite's functional logic. Design the test result output format according to the electric propulsion performance evaluation requirements. Call the relevant satellite digital twin models and performance prediction models according to the above model requirements.
[0058] ② Based on the digital test architecture design, the structural model, control subsystem model, electrical subsystem model, and electric propulsion subsystem model are assembled and integrated. The above satellite digital twin models are directly related in terms of structural composition and functional logic, and can be directly assembled and integrated through the satellite digital twin model interface.
[0059] ③ Based on the interface format and interaction method, develop the corresponding interfaces for the aforementioned satellite digital twin model to adapt to the data interaction between the satellite digital twin model and the multi-condition performance prediction model. Specifically, the electrical subsystem model and the electric propulsion subsystem model require the battery subsystem charge / discharge performance prediction model, the solar panel subsystem power generation performance prediction model, and the thermal control subsystem thermal control performance prediction model to provide the overall satellite power supply performance parameters under various conditions, and then calculate the electric propulsion subsystem performance. Therefore, it is necessary to develop the relevant satellite digital twin model data interfaces based on the above functional relationships.
[0060] ④ Using the corrected satellite digital twin model, multi-condition satellite performance simulation data is generated. Combined with physical test data, multi-condition satellite performance simulation data, and relevant performance prediction models, a multi-condition performance prediction model is trained using transfer learning. Digital testing of satellite electric propulsion performance requires training relevant historical performance prediction models through transfer learning to adapt to more operating conditions. The basic idea is... Among them, the required i-performance prediction models are sorted and labeled. This represents the i-th multi-condition performance prediction model after migration. This represents the i-th historical performance prediction model. This represents the i-th source domain data. This represents the data in the i-th target domain, which includes processed physical test data and multi-condition satellite performance simulation data. This relates to transfer learning. The digital performance test of satellite electric propulsion requires the performance prediction model to be transferable to more other operating conditions. Since the feature spaces of the source and target domains are basically the same, feature-based transfer learning is adopted, specifically implemented using a domain adaptation method. Furthermore, a mapping function for the target domain is defined. Domain-adaptive methods need to learn using source domain data. To minimize the expected error in the target domain: Specifically, it is implemented using support vector machines. In order to obtain a multi-condition performance prediction model adapted to the target domain data;
[0061] ⑤ Based on the interface format and interaction design, develop the corresponding interfaces for the above-mentioned multi-condition performance prediction model to adapt to the data interaction between the multi-condition performance prediction model and the satellite digital twin model. Among them, the battery subsystem charge and discharge performance prediction model, the solar panel subsystem power generation performance prediction model, and the thermal control subsystem thermal control performance prediction model need to transmit relevant parameters to each other, and also need to transmit performance parameters to the electrical subsystem model and the electric propulsion subsystem model. Develop relevant data interfaces according to their functional relationships.
[0062] ⑥ Using the developed interface, the structural model, control subsystem model, electrical subsystem model, electric propulsion subsystem model, battery subsystem charge and discharge performance prediction model, solar panel subsystem power generation performance prediction model, and thermal control subsystem thermal control performance prediction model are integrated to build a test scenario for digital testing of satellite electric propulsion performance under multiple operating conditions.
[0063] ⑦ Based on the digital test scheme and satellite digital test scenario, conduct digital tests on satellite electric propulsion performance under multiple operating conditions, and generate digital test results under multiple operating conditions.
[0064] Step S104, the data-real fusion driven test verification and evaluation module, constructs an index system for multi-condition satellite electric propulsion performance test verification and evaluation, and designs corresponding evaluation methods. Guided by the index system and evaluation methods, it builds a data-real fusion test verification scenario, injects various operating conditions into the data-real fusion test verification scenario, analyzes and evaluates the digital test results of multi-condition satellite electric propulsion performance, and finally provides a comprehensive evaluation result of satellite electric propulsion performance under various operating conditions. Specific implementation includes:
[0065] ① Construct a multi-condition satellite electric propulsion performance test verification and evaluation index system, including satellite digital twin model test verification and evaluation indexes, multi-condition performance prediction model test verification and evaluation indexes, and integrated test verification and evaluation indexes. Among them, the satellite digital twin model test verification and evaluation indexes include geometric model accuracy, physical model accuracy, behavioral model accuracy, and condition coverage; the multi-condition performance prediction model test verification and evaluation indexes include accuracy, precision, recall, and generalization of the performance prediction model; and the integrated test verification and evaluation indexes include overall accuracy, overall confidence, and condition coverage. Corresponding analysis and evaluation methods are designed for the above indicators.
[0066] ② Guided by the indicator system and evaluation methods, a data-physical integrated test and verification scenario is built. The physical part includes onboard computers, electric propulsion test benches, satellite subsystem equivalents, etc., while the virtual part includes digital twin models of the space environment and satellite subsystems. The two are integrated through interfaces to build a data-physical integrated semi-physical satellite electric propulsion performance test and verification scenario.
[0067] ③According to the test plan, multi-condition data are injected into the data-real fusion test verification scenario to carry out test verification and obtain data-real fusion test verification data;
[0068] ④ Based on the evaluation method, the digital test results of satellite electric propulsion performance under various operating conditions are analyzed and evaluated in combination with the data from the fusion of data and reality tests, and a comprehensive evaluation result of satellite electric propulsion performance under various operating conditions is given.
[0069] In summary, this invention discloses a satellite multi-condition data-real fusion test method based on transfer learning and digital twins, including a test scheme generation and evaluation module, a digital twin model evaluation and correction module, a multi-condition digital test module based on digital twins and transfer learning, and a data-real fusion-driven test verification and evaluation module. This method can solve the problems existing in current satellite multi-condition data-real fusion tests to a certain extent and improve the capability and efficiency of satellite multi-condition tests.
[0070] The contents not described in detail in this specification are existing technologies known to those skilled in the art.
[0071] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A satellite multi-condition data-real fusion test method based on transfer learning and digital twins, characterized in that, The specific process of the method is as follows: Step S101, Test scheme generation and evaluation module, generates and filters specific test schemes for multi-condition tests in batches based on satellite test scheme generative model and satellite digital twin model; Step S102, Digital Twin Model Evaluation and Correction Module, uses physical satellite test data as a driving force to determine whether satellite digital twin model correction is needed, and completes the relevant model correction as appropriate; Step S103: The multi-condition digital test module based on digital twin and transfer learning integrates the satellite digital twin model and the multi-condition performance prediction model trained by transfer learning to realize satellite digital test under multiple conditions. Step S104, the experimental verification and evaluation module driven by data-real fusion, verifies and evaluates the satellite digital test results under multiple operating conditions by establishing an experimental verification and evaluation index system and building a data-real fusion experimental verification scenario.
2. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 1, characterized in that, Step S101 specifically includes: ① Based on the requirements of satellite experiments, design random variables for batch generation of satellite experiment schemes, taking into account aspects such as experiment subjects, experiment scenarios, satellite status, experiment process, and experiment parameters; ② Train a generative model of satellite test schemes by using historical satellite test schemes, historical mission texts, and satellite expert knowledge. Generate random variables to input into the generative model of satellite test schemes with test mission boundaries and satellite capabilities as constraints, and generate multi-condition satellite test schemes in batches. ③ Based on the satellite digital twin model, the generated multi-condition satellite test schemes are evaluated and screened according to the rationality of the test mission objectives, the feasibility of the process, and the representativeness of the results, forming a multi-condition test scheme set that includes physical test schemes and digital test schemes. The physical test schemes guide the execution of physical satellite on-orbit tests or ground tests, while the digital test schemes guide the execution of satellite digital tests.
3. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 2, characterized in that, Step S102 specifically includes: ①Based on the satellite's operating mechanism, performance evolution law, and expert knowledge, analyze the characteristic parameters related to satellite performance, and use the multi-factor factorization method to analyze the related characteristic parameters of the satellite in various time-varying processes, preliminarily define the allowable fluctuation range of the characteristic parameters, and establish a model update triggering mechanism; ② Input the relevant parameters of the physical experiment into the satellite digital twin model to carry out simulation calculations, and calculate the deviation between the simulation results data and the physical experiment data; ③ Determine whether the deviation between the satellite digital twin model and the actual physical satellite exceeds the allowable fluctuation range. If it does, it is determined that it needs to be corrected and jumps to step ④ of S102. If it does not exceed the allowable fluctuation range, jumps to step ① of S103. ④ Based on physical test data, the multi-dimensional satellite digital twin model is corrected, and the corrected satellite digital twin model is evaluated and iterated until the model correction is completed.
4. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 1, characterized in that, Step S103 specifically includes: ① Based on the digital experiment plan and model, design the digital experiment architecture, including model requirements, interface format, interaction form, calculation process, output format, and call the relevant satellite digital twin model and performance prediction model according to the model requirements; ② Based on the digital experiment architecture design, assemble and integrate the assemblable satellite digital twin model; ③ Based on the interface format and interaction form design, develop the corresponding interface for the satellite digital twin model to adapt the data interaction between the satellite digital twin model and the multi-condition performance prediction model; ④ Using the corrected satellite digital twin model, simulate and generate multi-condition satellite performance simulation data. Combine physical test data, multi-condition satellite performance simulation data and related performance prediction models, and train the multi-condition performance prediction model based on transfer learning. ⑤ Based on the interface format and interaction form design, develop the corresponding interface for the multi-condition performance prediction model to adapt to the data interaction between the multi-condition performance prediction model and the satellite digital twin model; ⑥ By leveraging the developed interface, the satellite digital twin model and multi-condition performance prediction model are integrated to build a satellite digital test scenario for multi-condition digital testing; ⑦ Based on the digital test scheme and satellite digital test scenario, conduct multi-condition satellite digital tests and provide digital test results under multiple conditions.
5. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 1, characterized in that, Step S104 specifically includes: ① Construct a satellite multi-condition test verification and evaluation index system, including satellite digital twin model test verification and evaluation index, multi-condition performance prediction model test verification and evaluation index, and integrated test verification and evaluation index, and design corresponding analysis and evaluation methods; ② Guided by the indicator system and evaluation methods, we build a data-real integrated test and verification scenario. The physical part is a physical equivalent test bench or physical satellite system, and the virtual part is a digital twin model of the environment and satellite. We integrate the two through an interface to build a data-real integrated semi-physical satellite test and verification scenario. ③According to the test plan, multi-condition data are injected into the data-real fusion test verification scenario to carry out test verification and obtain data-real fusion test verification data; ④ Based on the evaluation method, the results of the satellite digital test under various operating conditions are analyzed and evaluated in combination with the data from the data-real fusion test verification, and a comprehensive evaluation result of the satellite performance under various operating conditions is given.
6. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 5, characterized in that, The multi-condition data fusion test includes the satellite's electric propulsion performance test.
7. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 6, characterized in that, The satellite's on-orbit electric propulsion performance test comprehensively evaluates the electric propulsion performance by conducting on-orbit attitude control, orbit maintenance, and orbit change control under multiple operating conditions.
8. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 6, characterized in that, The digital twin models related to satellite electric propulsion performance include structural models, control subsystem models, electrical subsystem models, and electric propulsion subsystem models.
9. A satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 6, characterized in that, The physical component includes an onboard computer, an electric propulsion test bench, and a satellite subsystem equivalent, while the virtual component includes a space environment and a digital twin model of the satellite subsystem. The two components are integrated through an interface to build a semi-physical satellite electric propulsion performance test and verification scenario that combines digital and physical components.
10. The satellite multi-condition data-real fusion test method based on transfer learning and digital twins according to claim 3, characterized in that, Based on physical experimental data, a multi-dimensional satellite digital twin model is corrected, specifically from the geometric, physical, and behavioral dimensions. The geometric dimension describes the geometry, size, and assembly relationships of the satellite and its subsystems, and its correction is achieved by updating geometric parameters and structural features. The physical dimension describes the physical mechanisms of the satellite and its subsystems in the fields of mechanics, electronics, and thermodynamics, and its correction is achieved through parameter updates and function corrections based on the model's deviation characteristics. The behavioral dimension describes the behavior of the satellite and its subsystems in space motion, data interaction, and functional operations, and its correction is achieved by updating behavioral descriptions and interaction parameters.