A method for verifying the fidelity of spacecraft simulation models based on behavior comparison
By designing a sequential experimental strategy using multi-dimensional fidelity indices and Gaussian process regression, the problems of incomplete and inefficient fidelity verification of spacecraft simulation models were solved, achieving efficient consistency verification between spacecraft models and physical objects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PEKING UNIV
- Filing Date
- 2026-02-28
- Publication Date
- 2026-05-26
AI Technical Summary
Existing methods for verifying the fidelity of spacecraft simulation models are incomplete and inefficient, failing to provide a unified answer to the consistency issue between the model and the physical object at the overall level. Traditional methods are not applicable to complex spacecraft systems and require a large amount of experimental data.
We designed a method for verifying the fidelity of spacecraft simulation models based on behavioral comparison. By using a multi-dimensional fidelity index system, combined with Gaussian process regression and sequential experimental strategies, we quantified the model uncertainty and optimized the experimental design to improve efficiency.
This has achieved a comprehensive improvement in the overall fidelity and efficiency of spacecraft simulation models, reduced the amount of testing, and improved the accuracy and reliability of the consistency between the model and the actual object.
Smart Images
Figure CN121744520B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of spacecraft simulation technology, specifically relating to a method for verifying the fidelity of spacecraft simulation models based on behavioral comparison. Background Technology
[0002] The digitalization of spacecraft is developing rapidly, and high-confidence digital modeling and analysis of spacecraft and their subsystems is one of the many challenges facing the field. One of the key issues in digital spacecraft is how to verify the consistency between the established digital model and the actual product. Traditional methods for testing physical spacecraft products compare whether the product output meets design requirements; however, this approach cannot verify whether the digital model matches the physical spacecraft product. Given the increasing number of digital spacecraft models, the need to design a method to verify the fidelity of digital models at a holistic level is extremely urgent.
[0003] Traditional deterministic model validation directly compares the absolute errors of simulation and experimental results. Model validation under uncertainty conditions, however, comprehensively considers various errors that may exist in numerical simulations and physical experiments. The complexity, high coupling, and harsh operating environment of spacecraft systems lead to cognitive errors, such as simplification, in the abstracted mathematical models. Furthermore, the constructed digital models also contain random errors, such as those related to computational parameters. Existing methods such as classical hypothesis testing, Bayesian hypothesis testing, frequency-based metrics, and area-based metrics are mostly suitable for validating digital models of specialized products such as spacecraft components. These methods require extensive experimental data and are biased towards theoretical analysis, failing to provide a unified answer to the question of model fidelity at an overall level.
[0004] In summary, existing methods for verifying the fidelity of digital models lack comprehensive evaluation indicators and are not directly applicable to spacecraft digital models. Furthermore, the numerous experiments required are difficult to conduct at the spacecraft and its subsystem levels, thus significantly limiting the ability to verify the fidelity of spacecraft digital models. Summary of the Invention
[0005] To address the issues of incomplete overall fidelity verification and low implementation efficiency of existing technologies for spacecraft simulation models, this invention provides a fidelity verification method for spacecraft simulation models based on behavior comparison. By comparing spacecraft behavior, fidelity indicators are designed to improve the comprehensiveness and efficiency of the overall fidelity of spacecraft simulation models.
[0006] In this invention, spacecraft behavior refers to the dynamic operational characteristics exhibited by a spacecraft during mission execution, including its orbital evolution, maneuvering intentions, and mission status; the behavior is characterized by telemetry data, including but not limited to orbital root data fed back to the ground during orbital prediction.
[0007] The core of the spacecraft simulation model fidelity verification method based on behavior comparison provided by this invention is to verify and confirm the overall fidelity of the Gaussian process-based digital model by designing multi-dimensional fidelity indices. This includes: designing and determining the overall fidelity indices; designing an overall fidelity index system for the spacecraft simulation model based on spacecraft behavior; designing initial fidelity verification experiments using Latin hypercube sampling to avoid the problem of low efficiency in conducting complete factorial experiments across the entire parameter space due to numerous model input data; quantifying model uncertainty by using Gaussian process regression to establish a proxy model based on the output residuals of the model and the actual spacecraft, considering both model accuracy and variability, and quantifying the uncertainty of the residuals; and designing incremental fidelity verification experiments by using a sequential testing approach based on the results of residual uncertainty quantification, building upon the initial experiments.
[0008] The technical solution of this invention is:
[0009] A method for verifying the fidelity of a spacecraft simulation model based on behavioral comparison is proposed. This method verifies and confirms the overall fidelity of a Gaussian process-based digital spacecraft model by designing multi-dimensional fidelity indices. The method includes the following steps:
[0010] Step 1: Design the overall fidelity index system for the spacecraft simulation model;
[0011] An evaluation system encompassing multiple dimensions, including granularity, repeatability, accuracy, and variability, is constructed, and the fidelity of the comprehensive index is obtained through integrated calculation. , represented as:
[0012]
[0013] in, For accuracy; It is variability; This represents the sample mean output by the spacecraft model. This represents the sample mean of the physical output from the spacecraft. Correcting the standard deviation for the sharpness of actual spacecraft objects; Correct the standard deviation for the sharpness of the spacecraft model;
[0014] Step 2: Design a method for initial fidelity verification testing and conduct initial tests on the spacecraft simulation model;
[0015] Step 3: Establish a proxy model for the output residuals between the spacecraft simulation model and the actual spacecraft, and quantify the uncertainty of the spacecraft simulation model by quantifying the uncertainty of the residuals to obtain the magnitude of the uncertainty of the spacecraft simulation model output;
[0016] The proxy model is represented as:
[0017] in, The surrogate model function for outputting residuals; GP represents a Gaussian process; x and Input the parameter values in the parameter space of the experiment; It is the mean function; It is the covariance function or kernel function;
[0018] By solving the surrogate model, the predicted residuals for the new experiment were obtained, and the covariance of the residuals was also obtained, which is the magnitude of the spacecraft model uncertainty.
[0019] Step 4: Based on the initial experiment, design an incremental fidelity verification experiment;
[0020] Specifically, the method based on maximum uncertainty and the method based on maximum a posteriori probability are used to guide the design of incremental experiments based on the quantitative results of residual uncertainty and experimental resources.
[0021] Based on the quantification results of model uncertainty, the number of test points is gradually increased to achieve accurate calculation of the fidelity index.
[0022] Furthermore, in the overall fidelity index system of this invention:
[0023] The smallest distinguishable unit scale of the output parameters of the granularity characterization model is used to quantify cognitive uncertainty;
[0024] Repeatability is measured by the standard deviation of the output results when the model is run multiple times under the same input, reflecting random uncertainty;
[0025] Accuracy is defined as the dimensionless quantity obtained by normalizing the standard deviation of the model output sample mean and the actual spacecraft output sample mean after granularity correction of the actual object.
[0026] Variability is defined as the normalized value of the product of the granularity correction standard deviations of the model and the actual object, and is used to measure the consistency between the two in terms of dynamic response change characteristics.
[0027] Fidelity is the product of accuracy and variability, and its value ranges from [0, N], where N is the dimension of the model output vector.
[0028] Furthermore, the overall stability of the spacecraft simulation model is adjusted using the sharpness-corrected standard deviation. express:
[0029]
[0030] Where S represents repeatability, i.e., standard deviation; This refers to granularity.
[0031] Furthermore, in step three, the residual data of the spacecraft simulation model and the actual spacecraft output are obtained from the initial experiment. Based on Gaussian process regression (GPR), a GPR surrogate model is established for the output residuals of the spacecraft simulation model and the actual spacecraft. The uncertainty of the residuals is quantified to obtain the predicted variance of each unsampled region. Based on the uncertainty information output by the GPR surrogate model, an incremental test point is designed using a sequential test strategy to increase the number of test points.
[0032] The design of incremental test points is as follows: when test resources are sufficient, the position with the largest prediction variance is selected as the new test point; when test resources are limited, the position with the largest posterior probability is selected as the new test point, until the fidelity index converges or the preset number of tests is reached.
[0033] Furthermore, in step four, the sampling method based on maximum uncertainty specifically involves: selecting the next experimental point. And to maximize the variance of the test point, expressed as:
[0034]
[0035] in, The variance of telemetry data at the next test point of the spacecraft.
[0036] The method based on maximum a posteriori probability is as follows:
[0037] Select the next spacecraft test site And maximize the posterior probability of that test point, expressed as:
[0038] ;
[0039] in, This will serve as the next test site for spacecraft. is the physical simulation model corresponding to the actual behavior of the spacecraft; P is the posterior probability of the surrogate model; This is a spacecraft telemetry dataset; argmaxp() represents the parameter that calculates the maximum probability.
[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0041] This invention provides a method for verifying the fidelity of spacecraft simulation models based on behavior comparison. Based on spacecraft behavior comparison, fidelity indicators are designed, and uncertainty quantification and incremental experiments are conducted to improve the comprehensiveness and efficiency of the overall fidelity of spacecraft simulation models. Attached Figure Description
[0042] Figure 1 This is a schematic diagram showing the relationship between the accuracy index of a spacecraft simulation model and its semi-major axis.
[0043] Figure 2 This is a schematic diagram showing the relationship between the variability index of a spacecraft simulation model and its semi-major axis.
[0044] Figure 3 This is a diagram comparing the fidelity metrics of the TB and SGP4 models.
[0045] Figure 4 This is a schematic diagram of the result obtained by using the method of the present invention to quantify uncertainty.
[0046] Figure 5 This is a flowchart of the method of the present invention. Detailed Implementation
[0047] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.
[0048] This invention proposes a method for verifying the fidelity of spacecraft simulation models based on behavioral comparison. This invention comprehensively considers the behavior and characteristics of spacecraft digital models, designs an overall fidelity index system suitable for spacecraft simulation models, and accurately estimates the overall fidelity of spacecraft simulation models through a sequential experimental strategy based on Gaussian process regression. Figure 5 The diagram illustrates the flow of the method of the present invention. Specifically, the present invention includes the following steps:
[0049] Step 1: Design and determination of overall fidelity indicators;
[0050] The overall fidelity or fidelity of a model is defined as the degree of consistency between the spacecraft simulation model and the actual spacecraft. Specifically, from the perspective of the intended use of the spacecraft simulation model, under the same input conditions, the degree of matching between the remote sensing data output by the spacecraft simulation model and the remote sensing data output by the actual spacecraft. The overall fidelity index system for the spacecraft simulation model designed in this invention includes four dimensions: granularity, repeatability, accuracy, and variability. The overall fidelity is ultimately obtained through comprehensive calculation and serves as the overall fidelity index.
[0051] (1) Particle size
[0052] Granularity refers to the smallest resolvable unit scale of the input data used to characterize the health status monitoring of a spacecraft during its on-orbit operation. Examples include key telemetry parameters such as orbital six-point count, attitude angular velocity, power bus voltage, and thermal control temperature. This metric describes the smallest variation that a model can distinguish or the smallest level of detail it can represent. Granularity can reflect cognitive uncertainty, i.e., parameters that cannot be definitively defined as fixed values within a certain range. Since cognitive uncertainty can be modeled as a uniform distribution (a distribution with equal probability between two boundaries), granularity... It can be defined as the distance between two boundaries.
[0053] For spacecraft, the simplest case can be represented by the granularity of the parameter layer value h.
[0054]
[0055] (2) Repeatability
[0056] Repeatability measures the similarity of the distributions of a model's outputs over multiple iterations, and is a concrete manifestation of random uncertainty. The randomness of a normal distribution can be described by the standard deviation S, while higher-order moments such as skewness and kurtosis can be used to describe the randomness of more complex distributions. Spacecraft typically have limited available data; therefore, it can be assumed that the data follows a normal distribution, and the standard deviation can be used to describe the model's repeatability. Assuming the model outputs a dataset Y, repeatability, i.e., the standard deviation, is expressed as:
[0057]
[0058] in, L is the mean of Y, and L is the total number of elements in Y.
[0059] (3) Accuracy
[0060] Repeatability and granularity together characterize the output stability of the simulation model, and since they originate from independent sources, they are additive. Therefore, the overall stability of the simulation model is calculated using the sharpness-corrected standard deviation. To indicate:
[0061]
[0062] The accuracy metric assesses the degree of consistency between the spacecraft model and the actual spacecraft by measuring the mean difference between the two, and normalizes and makes the standard deviation normalized using the granularity correction of the actual spacecraft. It is defined as follows:
[0063]
[0064] in, This represents the sample mean output by the spacecraft model. This represents the sample mean of the physical output from the spacecraft. Correct the standard deviation for the clarity of the actual object.
[0065] (4) Variability
[0066] Variability refers to the ability of a model to capture significant variations in the output behavior of a real spacecraft if those variations are substantial in reality. Therefore, a high-fidelity model must maintain consistency with the real spacecraft in both accuracy and variability. The similarity in variability between the model and the real spacecraft primarily depends on the difference in granularity correction standard deviations. Variability is defined as the normalized product of the granularity correction standard deviations of the model and the real spacecraft.
[0067]
[0068] in, and The standard deviation for the sharpness correction is applied to both the model and the actual object.
[0069] (5) Fidelity
[0070] The fidelity index of the final spacecraft model should be between 0 and 1, where 0 indicates that the model is completely different from the real object, and 1 indicates that the model is identical to the real object. Based on this concept, fidelity... Defined as the product of accuracy and variability.
[0071]
[0072] in, To ensure fidelity; For accuracy; It is variable.
[0073] The above formula means that the fidelity metric is strictly less than or equal to its accuracy and variability components. If the accuracy or variability is low, the fidelity metric will also be low.
[0074] Step 2: Design an initial fidelity verification test method and conduct initial tests on the spacecraft simulation model;
[0075] In this invention, considering the large number of input parameters for the spacecraft simulation model, conducting a complete factorial experiment across the entire parameter space would be computationally costly and inefficient. Therefore, an initial model validation experiment design based on Latin Hypercube Sampling (LHS) is adopted.
[0076] Suppose the model has p input parameters, and we plan to conduct n initial experiments. The LHS experimental design process is as follows:
[0077] First, generate p independent random permutations from 1 to n, and use them as column vectors to form an n×p index matrix. Then, generate v uniformly distributed random numbers for each dimension in the interval [0,1], and stratify the value intervals of each parameter with equal probability. Finally, by combining the above permutations with random perturbations, construct n test points, such that in each parameter's n sub-intervals, one and only one test point is selected, and any two test points do not fall into the same sub-interval in any dimension.
[0078] The resulting set of test points covers the entire input parameter space, exhibiting good space-filling and representativeness, and is denoted as the LHS test design set.
[0079] Step 3: Quantify the uncertainty of the spacecraft simulation model to obtain the magnitude of the uncertainty output by the spacecraft simulation model;
[0080] Although the initial experiments covered the entire parameter space, the number of experiments was small, posing a risk of inaccurate calculation of the model fidelity index. From the perspective of model accuracy and variability, the risk of inaccurate model fidelity assessment may lie in the fact that the actual object or model may exhibit significant output variations under certain combinations of input parameters, while the other party may not show this characteristic. Therefore, to identify input parameter combinations where the model and the actual object may differ significantly, this invention uses Gaussian process regression (GPR) to establish a surrogate model (GPR model) based on the output residuals of the model and the actual object. The surrogate model is expressed as:
[0081]
[0082] in, The surrogate model function for the output residuals between the spacecraft model and the actual spacecraft; GP represents a Gaussian process; x and Input the parameter values in the parameter space of the experiment; It is the mean function; It is the covariance function or kernel function;
[0083] Furthermore, the uncertainty of the residuals is quantified based on the surrogate model, thereby increasing the number of test points on parameter combinations with greater uncertainty.
[0084] When input parameters Given the physical object, let y be the output, and let the model's output be the output. The residual is denoted as d.
[0085] For the test points composed of input parameters The corresponding residual value is A Gaussian process is completely defined by its mean function and covariance function. It is a mean function, usually set to zero mean, i.e. ; It is the covariance function (or kernel function) used to describe the similarity between input points.
[0086] Kernel function The shape and characteristics of a Gaussian process are determined by its kernel function, and a Gaussian process model with a single kernel function is insufficient to describe the varying characteristics of the input dimensions of a complex system. This invention utilizes the idea of multi-kernel learning support vector machines and selects a composite covariance function. Considering the long and short periodicity characteristics of spacecraft telemetry data, a squared exponential kernel function is chosen:
[0087]
[0088] in, It is the variance of spacecraft telemetry data. It is the length of spacecraft telemetry data.
[0089] Given a training dataset Assume the residuals contain independent and identically distributed Gaussian noise. and When predicting new test points, the prediction distribution can be calculated using the joint Gaussian distribution, expressed as:
[0090]
[0091] in, It is the kernel matrix between the initial test points. It is the kernel vector between the initial test point and the new test point. It is the core value of the new test site itself.
[0092] Within the functional space defined by the prior distribution of a Gaussian process, the predicted output value of the posterior distribution can be computed within a Bayesian framework. That is, the predicted distribution for new experimental points can be obtained through the conditional Gaussian distribution.
[0093]
[0094] Their mean and covariance are as follows:
[0095]
[0096] in, This is the predicted mean of the new test sites. It is the prediction variance, which reflects the predicted uncertainty value for this test point.
[0097] In summary, by solving the GPR model of the residuals, the predicted residuals for the new experiment were obtained. At the same time, the covariance of the residuals is obtained. This refers to the magnitude of uncertainty in the spacecraft model.
[0098] Step 4: Incremental Fidelity Validation Test Design
[0099] Based on the initial experiments, the incremental experimental design is guided by the results of residual uncertainty quantification. In sequential experiments, common methods for selecting new experimental points include uncertainty-based, information gain-based, and maximum a posteriori (MAP)-based methods. Considering that the uncertainty value of residual prediction can be directly obtained from the GPR model, a method based on maximum uncertainty can be used when there are many new experimental resources, while a method based on maximum a posteriori probability can be used when experimental resources are limited. The incremental experimental design is then combined with the uncertainty estimated by the GPR model.
[0100] Sampling method based on maximum uncertainty: For the GPR model, the obtained experimental dataset is as follows: Select the next test point At that time, we want to maximize the variance at that point, that is:
[0101]
[0102] in, The variance of telemetry data at the next test point of the spacecraft;
[0103] Maximum a posteriori probability-based methods: When obtaining spacecraft telemetry datasets... Then, the posterior probability P of the GPR model is:
[0104]
[0105] in, It is the posterior mean. This is the posterior variance. The formulas for calculating the posterior mean and variance are:
[0106]
[0107] Select the next spacecraft test site At that time, we want to maximize the posterior probability at that point, that is:
[0108]
[0109] in, A physical simulation model corresponding to the actual behavior of spacecraft; This is a spacecraft telemetry dataset; argmaxp() represents the parameter that calculates the maximum probability.
[0110] In summary, by gradually increasing the number of experimental points based on the quantification results of model uncertainty, the accuracy of the fidelity index can be finally calculated.
[0111] The following examples use a typical spacecraft orbital model to verify the overall fidelity of the model, including:
[0112] The first step involves designing the indices and calculating the granularity. This embodiment of the invention uses a typical spacecraft orbit model to verify the results of the overall model fidelity confirmation process, illustrating the calculation results of granularity, accuracy, variability, and fidelity indices. This process involves multiple sampling and statistical analysis of experimental data to ensure the model's performance under various conditions. Taking three spacecraft orbit models—TB, SGP4, and SDP4—as examples, their theoretical accuracy is TB < SGP4 < SDP4. Therefore, SDP4 is used as the actual spacecraft, and its output is considered the true value of the corresponding indices. Based on this, the fidelity indices of TB and SGP4 are calculated.
[0113] Then, proceed to step two to generate the initial experimental design. The SGP4 simulation output deviation is much smaller than TB. To visually demonstrate the fidelity index, box plots of accuracy and variability indices as TB changes with the semi-major axis of the SGP4 input are attached. Figure 1 With appendix Figure 2 As shown in the figure. The results show that SGP4's accuracy is significantly higher than TB's, and SGP4's variability is worse than TB's. The accuracy of both models in the low Earth orbit region is lower than in the high Earth orbit region because atmospheric and other factors have a greater impact in the low Earth orbit region, consistent with theoretical understanding. The fidelity index has a value range of [0,1] when the model has only one dimension of output; the upper limit increases by 1 for each additional dimension of output. The orbital model output has 6 dimensions, therefore the fidelity range is [0,6]. To visually demonstrate the overall fidelity index, box plots of the fidelity index for TB and SGP4 with varying inclination angles are attached. Figure 3 As shown in the figure. The results show that the fidelity of SGP4 is significantly higher than that of TB. Both models exhibit poor fidelity at low tilt angles due to the large range of fidelity variations. This result is attributed to the influence of Earth's oblateness, consistent with theoretical understanding.
[0114] Proceed to step three: uncertainty quantification. (See attached document) Figure 4 The solid line represents the quantified uncertainty value. This value identifies locations of high uncertainty and guides the incremental experimental design in step four. After comprehensive calculation, the number of experiments can be reduced by 53.1%, significantly compressing the experimental workload and improving efficiency. Finally, step one is performed again to calculate the fidelity index, which is then used to verify the model's fidelity.
[0115] In summary, the method proposed in this invention accurately, stably, and efficiently confirms the overall fidelity of spacecraft digital models.
[0116] The embodiments of this invention are merely illustrative of the calculation process and are not intended to limit the implementation of the invention. Those skilled in the art will recognize that various modifications, such as parameter adjustments, can be made based on the description herein. This document cannot provide examples of all embodiments; any obvious modifications derived from the technical principles and solutions of this invention are still within the scope of protection of this invention.
Claims
1. A method for verifying the fidelity of a spacecraft simulation model based on behavioral comparison, characterized in that, Includes the following steps: Step 1: Design the overall fidelity index system for spacecraft simulation models: An evaluation system encompassing multiple dimensions, including granularity, repeatability, accuracy, and variability, is constructed, and the fidelity of the comprehensive index is obtained through integrated calculation. , represented as: in, For accuracy; It is variability; This represents the sample mean output by the spacecraft model. This represents the sample mean of the physical output from the spacecraft. Correcting the standard deviation for the sharpness of actual spacecraft objects; Correct the standard deviation for the sharpness of the spacecraft model; In the overall fidelity index system: The smallest resolvable unit scale of the output parameters of the granularity characterization model is used to quantify cognitive uncertainty. Repeatability is measured by the standard deviation of the output results when the model is run multiple times under the same input, reflecting random uncertainty; The accuracy is defined as the dimensionless quantity obtained by normalizing the standard deviation of the actual spacecraft output sample after correcting for the granularity of the actual spacecraft output sample. The variability is defined as the normalized value of the product of the granularity correction standard deviation of the model and the actual object, which is used to measure the consistency of the two in terms of dynamic response change characteristics. The overall stability of the spacecraft simulation model is corrected for by the sharpness standard deviation. express: Where S represents repeatability; Particle size; Step 2: Design a method for initial fidelity verification testing and conduct initial tests on the spacecraft simulation model; Step 3: Establish a proxy model for the output residuals between the spacecraft simulation model and the actual spacecraft, and quantify the uncertainty of the spacecraft simulation model by quantifying the uncertainty of the residuals to obtain the magnitude of the uncertainty of the spacecraft simulation model output; The proxy model is represented as: in, The surrogate model function for outputting residuals; GP represents a Gaussian process; x and Input the parameter values in the parameter space of the experiment; It is the mean function; It is the covariance function or kernel function; By solving the surrogate model, the predicted residuals for the new experiment were obtained, and the covariance of the residuals was also obtained, which is the magnitude of the spacecraft model uncertainty. Step 4: Based on the initial experiment, design an incremental fidelity verification experiment: We employ methods based on maximum uncertainty and maximum a posteriori probability, and use the residual uncertainty quantification results and experimental resources to guide incremental experimental design. Based on the quantification results of model uncertainty, the number of test points is gradually increased to achieve accurate calculation of the fidelity index.
2. The method for verifying the fidelity of spacecraft simulation models based on behavioral comparison as described in claim 1, characterized in that, Fidelity is the product of accuracy and variability, and its value ranges from [0, N], where N is the dimension of the model output vector.
3. The method for verifying the fidelity of spacecraft simulation models based on behavioral comparison as described in claim 1, characterized in that, In step three, the residual data of the spacecraft simulation model and the actual spacecraft output are obtained from the initial experiment. Based on Gaussian process regression (GPR), a GPR surrogate model is established for the output residuals of the spacecraft simulation model and the actual spacecraft. The uncertainty of the residuals is quantified to obtain the predicted variance of each unsampled region. Based on the uncertainty information output by the GPR surrogate model, an incremental test point is designed using a sequential test strategy to increase the number of test points.
4. The method for verifying the fidelity of spacecraft simulation models based on behavioral comparison as described in claim 3, characterized in that, The design of incremental test points is as follows: when test resources are sufficient, the position with the largest prediction variance is selected as the new test point; when test resources are limited, the position with the largest posterior probability is selected as the new test point, until the fidelity index converges or the preset number of tests is reached.
5. The method for verifying the fidelity of spacecraft simulation models based on behavioral comparison as described in claim 1, characterized in that, In step four, the sampling method based on maximum uncertainty specifically involves: selecting the next test point. And to maximize the variance of the test point, expressed as: in, This represents the variance of telemetry data at the next test point of the spacecraft.
6. The method for verifying the fidelity of spacecraft simulation models based on behavioral comparison as described in claim 5, characterized in that, The method based on maximum a posteriori probability is as follows: Select the next spacecraft test site And maximize the posterior probability of that test point, expressed as: ； in, This will serve as the next test site for spacecraft. P represents the physical simulation model corresponding to the actual behavior of the spacecraft; P is the posterior probability of the surrogate model. This is a spacecraft telemetry dataset; argmaxp() represents the parameter that calculates the maximum probability.