A spacecraft intelligent software virtual simulation test scene adaptive optimization method

Abstract

This invention relates to an adaptive optimization method for virtual simulation test scenarios in spacecraft intelligent software. The method involves acquiring a set of scenario parameters and defining a feasible region, generating N initial scenario parameter samples within the feasible region. Each scenario parameter sample is then converted into a simulation scenario script and loaded into a simulation engine to run the spacecraft intelligent software and collect operational status data. Based on a predefined index system, a comprehensive risk score for the scenario is calculated. Using this as training data, a Gaussian process regression model is fitted, and a data acquisition function is constructed to find the next set of most worthy-of-evaluation scenario parameter samples. Simulation evaluation is then performed to obtain a new score. This score is added to the historical dataset, and the surrogate model is updated. This process is iterated until the maximum number of iterations is reached or the comprehensive risk score converges. This method solves the problems of low testing efficiency and insufficient coverage of key scenarios caused by the lack of intelligent guidance in test scenario generation.

Description

Technical Field

[0001] This invention belongs to the field of spacecraft software testing and simulation technology, and relates to an adaptive optimization method for virtual simulation test scenarios of spacecraft intelligent software. It is particularly suitable for testing, verification and performance evaluation of key intelligent software such as spacecraft autonomous navigation systems, environmental perception algorithms and intelligent decision-making modules. Background Technology

[0002] With the increasing demands for autonomy and intelligence in space missions such as deep space exploration and on-orbit servicing, spacecraft intelligent software, represented by environmental perception, autonomous navigation, and intelligent decision-making, has become the core support for mission execution. This type of software is typically based on artificial intelligence technologies such as deep learning and reinforcement learning. Its decision-making process has "black box" characteristics, exhibiting high complexity and uncertainty in complex space environments, posing a severe challenge to traditional testing methods.

[0003] High-fidelity virtual simulation testing is a key technical means to verify the reliability of aerospace intelligent software, and its testing efficiency is highly dependent on the quality of the test scenario. However, current methods for generating test scenarios for aerospace intelligent software have significant limitations: While existing methods such as MarsSim and ISMRS simulation platforms can achieve high-fidelity physical and visual environment simulations and support the procedural generation of complex terrains, their test scenario generation strategies still primarily rely on manual experience design or random generation. These methods lack specificity for the failure modes of aerospace intelligent software and struggle to systematically generate test scenarios that effectively trigger unique defects in intelligent software, such as perception algorithm misjudgments, navigation decision errors, and abnormal control commands. For example, critical test scenarios such as the performance degradation of visual recognition algorithms under extreme lighting conditions, the failure of multi-sensor fusion systems under specific interference modes, and path planning errors of autonomous decision-making modules in complex terrains often cannot be effectively covered by random generation or manual experience design.

[0004] While random generation methods can produce a large number of scenario variations, the lack of guidance mechanisms targeting the vulnerabilities of intelligent software means that most of the generated scenarios remain within the "comfort zone" of the intelligent software, making it difficult to reach the system's stability boundaries. Manual design methods, on the other hand, are inefficient and heavily rely on the accumulated experience of testing experts, making it difficult to systematically uncover the potential risks of intelligent software under complex operating conditions.

[0005] Therefore, there is an urgent need for an adaptive optimization method that can automatically discover and focus on high-risk testing scenarios based on the characteristics of aerospace intelligent software, and efficiently expose deep defects in spacecraft intelligent software by intelligently guiding the testing direction, thereby comprehensively improving the relevance and effectiveness of testing. Summary of the Invention

[0006] The technical problem addressed by this invention is to overcome the shortcomings of existing technologies and propose an adaptive optimization method for virtual simulation test scenarios in spacecraft intelligent software. This method aims to solve the core problems of low testing efficiency and insufficient coverage of key scenarios in current spacecraft intelligent software simulation testing, caused by the lack of intelligent guidance in test scenario generation. Specifically, this invention addresses the following three key technical problems: (1) Existing testing methods lack guidance and are difficult to automatically discover key high-risk scenarios: Current testing mainly relies on manual experience to design scenarios or adopts random generation strategies. Manual design is difficult to fully foresee all complex boundary conditions that can trigger system failure; while random generation is a "blind" search, which will generate a large number of irrelevant normal operating conditions, making it extremely inefficient for discovering deep defects. Neither of these methods can systematically and automatically guide the testing process to the vulnerable links and stability boundaries of the spacecraft's intelligent software, causing many potential defects to be exposed only in real missions, bringing huge risks.

[0007] (2) Insufficient automation and intelligence in the testing process, failing to form a self-optimizing closed loop: Traditional testing processes are typically "open-loop": scenario design -> test execution -> manual result analysis. This process heavily relies on the intervention of testers to analyze the causes of failures and attempt to design new, potentially more effective test scenarios. This method cannot achieve autonomous iterative optimization of the testing process, is time-consuming and labor-intensive, and highly subjective, becoming a major bottleneck in improving the depth and breadth of testing.

[0008] (3) Insufficient utilization of the massive amount of data generated by simulation and lack of quantitative evaluation and mining methods: Although simulation testing can generate a large amount of operational data, existing methods lack a set of quantitative evaluation index systems that can comprehensively measure explicit failures and implicit risks for the failure modes of spacecraft intelligent software. This makes it impossible to automatically and quickly identify potential high-risk scenarios from the data, and even more impossible to use this data to drive the optimization of subsequent tests, resulting in a waste of data value and keeping test analysis at the qualitative level.

[0009] The solution to the technical problem of this invention is: an adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios, comprising the following steps: We analyze the key scenario factors affecting the performance of spacecraft intelligent software and abstract them into a set of optimizable scenario parameters. And define the feasible domain of scene parameters. The Latin hypercube sampling method is used in the feasible region. Generate N initial scene parameter samples. ; Sample each scene parameter , i =1~N The simulation scenario script is converted into a simulation scenario script and loaded into the simulation engine to run the intelligent software of the spacecraft under test, and to monitor and collect the software operation status data in real time. The defined indicator system includes a basic fault score and a continuous risk score. After the simulation, the comprehensive risk score for the scenario is calculated using the collected state data based on the predefined indicator system. The basic faults are divided into collisions, boundary crossings, mission failures, and loss of contact events, which are assigned high fixed penalty points. The continuous risk score is normalized and weighted based on the vehicle's maximum tilt angle, minimum obstacle distance, average slip rate, and path detour index. Using all evaluated parameters-rating pairs As training data, a Gaussian process regression model is fitted to construct a sampling function to find the next set of scene parameter samples most worthy of evaluation. ,right A simulation evaluation was conducted to obtain a new score. and will apply the new data to Add historical datasets, update the agent model, and iterate repeatedly until the maximum number of iterations is reached or the comprehensive risk score converges. During the iteration process, sample all scenario parameters that trigger failures or whose comprehensive risk scores exceed the threshold. The corresponding scripts and simulation process data are automatically saved to the high-risk scenario library.

[0010] Furthermore, the scene parameter classification includes: Obstacle layout parameters This includes the number, location, size, shape, and distribution of obstacles; Environmental disturbance parameters This includes sensor noise intensity, lighting conditions, ground friction coefficient, and communication delay. Task start condition parameters This includes the spacecraft's initial attitude, velocity, and target location.

[0011] Furthermore, the software runtime status data includes position, speed, attitude, sensor data, and task status logs.

[0012] Furthermore, let the basic faults of collision, boundary crossing, mission failure, and loss of connection be categorized as follows: , , , ,but .

[0013] Furthermore, the maximum tilt angle, minimum obstacle distance, and path detour indexes are applicable to all spacecraft types, while the average slip rate index is only processed when the spacecraft is a vehicle type.

[0014] Furthermore, the calculation method for the continuous risk score is as follows: Define the normalization function:

[0015] Where value, min, and max are the calculated value, lower limit, and upper limit of a certain indicator, respectively; Maximum body camber ,in, Indicates the angle of the vehicle's left and right tilt. Indicates the angle at which the front of the vehicle is tilted up or down; the fractional angle is:

[0016] in, For a safe tilt angle, This is the maximum permissible tilt angle; Minimum obstacle distance ,in, The distance from the spacecraft to the obstacle; the obstacle distance fraction is:

[0017] in, To maintain a safe distance; Average slip rate ,in, To the expected speed of the probe, The actual speed of the probe; the slippage rate fraction is:

[0018] in, This indicates the upper limit of the slippage rate, which defaults to 1.0. Path detour ,in, This represents the actual distance traveled from the starting point to the ending point. The detour score represents the straight-line distance between the starting point and the ending point. .

[0019] Furthermore, the comprehensive risk score The calculation method is as follows:

[0020]

[0021]

[0022] in, Divide into various basic fault categories, For each continuous risk score, These are the weighting coefficients assigned based on the importance of each risk indicator; The higher the value, the more likely the current scenario is to expose system defects.

[0023] Furthermore, a Gaussian process regression is used to fit a surrogate model, and a data acquisition function is constructed to find the next set of scene parameter samples most worthy of evaluation. Specifically: Based on comprehensive risk score As the optimization objective function With scene parameter set As an optimization variable, the scenario optimization problem is defined as a constrained maximization problem: , ; Using Gaussian process regression, based on existing evaluation data... Fitting the objective function The probability distribution model, also known as the surrogate model, can provide the probability distribution model for any point in the parameter space. X Predicted mean and forecast uncertainty ; Constructing the acquisition function using an upper confidence bound strategy ,in, To explore the weighting coefficients; by solving Find the next set of scenario parameter samples that are most worth evaluating. .

[0024] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario.

[0025] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario.

[0026] The advantages of this invention compared to the prior art are: (1) The intelligent guidance of test scenario generation is realized, which significantly improves the targeting and efficiency of defect discovery: Traditional random generation or manual design methods are like "finding a needle in a haystack", while this invention, through Bayesian optimization algorithm and guided by risk scoring, transforms the scenario generation process into "precision guidance". The system can automatically and directionally search for the stability boundaries and vulnerabilities of spacecraft intelligent software and actively generate high-risk scenarios that can trigger failure. This greatly improves the probability of discovering deep and hidden defects, concentrates test resources on the most effective areas, and thus discovers more key defects with fewer simulations, resulting in an order-of-magnitude improvement in test efficiency.

[0027] (2) A fully automated test optimization closed loop was constructed, significantly reducing reliance on manpower and improving process reliability: This invention achieves full-process automation from scenario generation, simulation execution, result evaluation to parameter optimization. It can autonomously decide the next test direction without requiring test engineers to repeatedly analyze the causes of failure and manually adjust scenario parameters. This not only frees experts from tedious trial-and-error work, but also avoids subjective biases and errors that may be introduced by human intervention, fundamentally ensuring the objectivity and repeatability of the test process, and laying a solid foundation for realizing an unattended automated test pipeline.

[0028] (3) A quantitative risk assessment system has been established, providing a scientific basis for test analysis: The comprehensive risk scoring system proposed in this invention transforms the vague, qualitative concept of "high risk" into precise, calculable quantitative indicators. This system comprehensively considers basic failures and continuous risks, and can comprehensively and meticulously assess the potential threat level of each test scenario. This provides a clear and explicit objective function for optimizing the testing process, making data-driven intelligent decision-making possible, and also provides a unified and scientific measurement standard for horizontal comparison and in-depth analysis of test results. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating an adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario according to the present invention. Figure 2 This is a framework diagram of an adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario according to the present invention. Detailed Implementation

[0030] To address the aforementioned technical problems, this invention provides an adaptive optimization method for virtual simulation testing scenarios of spacecraft intelligent software. The core of this method lies in constructing an automated intelligent closed loop of "simulation execution – risk quantification – parameter optimization – scenario regeneration," which continuously approximates and triggers the defective state of the spacecraft intelligent software through iterative feedback.

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] like Figure 1 As shown, the adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario of the present invention specifically includes the following steps: Step (1) Parametric scene modeling and initialization The key factors affecting the stability of spacecraft intelligent software are abstracted into a set of adjustable and optimizable scenario parameters. X And initialize the parameter space.

[0033] a) Parameter classification Obstacle layout parameters Such as the number, location, size, shape, and distribution of obstacles.

[0034] Environmental disturbance parameters Such as sensor noise intensity, lighting conditions, ground friction coefficient, communication delay, etc.

[0035] Task start condition parameters Such as the spacecraft's initial attitude (position, orientation), velocity, target point position, etc.

[0036] b) Constraint Definition Define the physical feasible domain of scene parameters using the syntax of a scene description language (such as Scenic). And task constraints, to ensure that the generated scene is physically plausible and task-relevant.

[0037] c) Initialize sampling Experimental design methods such as Latin hypercube sampling (LHS) were employed in the feasible region. Intrinsic generation Initial scene parameter samples This is to ensure uniform coverage of the initial samples in the parameter space.

[0038] Step (2): Simulation Execution and Risk Quantification Assessment For each scene parameter sample , i =1~ N It executes the corresponding simulation scenario and accurately quantifies its risks.

[0039] a) Simulation Execution and Monitoring Will The corresponding scenario scripts are loaded into the simulation engine (such as Webots), the spacecraft's intelligent software is run, and its operational status data (such as position, velocity, attitude, sensor data, mission status, etc.) are monitored and collected in real time.

[0040] b) Quantitative risk assessment Based on a predefined key scenario evaluation index system, the collected operational status data is analyzed in real time to calculate the comprehensive risk score for the scenario. .

[0041] The key scenario evaluation index system consists of two parts: Basic Fault Score (FS): Assigns a high fixed penalty score to catastrophic binary events such as collisions, boundary violations, mission failures, and loss of connection (e.g., Once this occurs, it signifies a high-risk scenario. The priorities of each event, from highest to lowest, are collision, task failure, loss of connection, and boundary violation; the corresponding basic fault scores are set from highest to lowest.

[0042] Collision: Highest priority. A physical collision with any object (celestial body, obstacle, other spacecraft) is an absolute failure. .

[0043] Offbound: A serious deviation from the pre-designated flight corridor or safe area, posing an extremely high risk. .

[0044] Task failure: Failure to complete the core task (such as failing to enter the target area, failing to collect samples, etc.). .

[0045] Loss: Communication interruption exceeding the maximum allowed duration means the task is out of control. .

[0046] The Continuous Risk Score (CRS) normalizes and weights continuous performance indicators such as maximum vehicle tilt angle, minimum obstacle distance, average slip rate, and path detour, quantifying potential performance degradation and hidden risks. The maximum vehicle tilt angle, minimum obstacle distance, and path detour indicators are applicable to all spacecraft types, such as aircraft or vehicles; the average slip rate indicator is only processed when the spacecraft is a vehicle.

[0047] Define the normalization function: .

[0048] Where value, min, and max are the calculated value, lower limit, and upper limit of a certain continuous performance index, respectively.

[0049] Vehicle camber: The greater the camber angle, the higher the risk. Maximum vehicle camber angle. ,in, Indicates the angle of the vehicle's left and right tilt. Indicates the angle at which the front of the vehicle is tilted up or down. Inclination fraction:

[0050] in, For a safe tilt angle, This is the maximum permissible tilt angle.

[0051] Minimum obstacle distance: The closer you are to the obstacle, the higher the risk. ,in, The distance from the spacecraft to the obstacle; obstacle distance fraction:

[0052] in, For safe distance.

[0053] Slip rate: Severe slippage means a low coefficient of friction with the terrain, posing a potential risk of getting stuck and wasting energy. Average slip rate ,in, To the expected speed of the probe, This represents the actual speed of the probe. Slippage rate score:

[0054] in, This indicates the upper limit of the slippage rate, which defaults to 1.0.

[0055] Path detour: A higher ratio indicates a less efficient planner or that it is forced to make numerous detours, wasting time and energy. ,in, This represents the actual distance traveled from the starting point to the ending point. Represents the straight-line distance between the starting point and the ending point. Detour score:

[0056] c) Scoring Summary Comprehensive Risk Score The calculation formula is:

[0057]

[0058]

[0059] in, Divide into various basic fault categories, For each continuous risk score, These are the weighting coefficients assigned based on the importance of each risk indicator. The higher the value, the more likely the current scenario is to expose system defects.

[0060] Step (3) Feedback-driven scene parameter optimization Based on comprehensive risk score As the optimization objective function With scene parameter set As optimization variables, a Bayesian optimization (BO) framework is used to drive the directed mutation and intelligent search of scene parameters, including the construction of a surrogate model and a data acquisition function, and the iterative optimization process. Specifically: a) Problem formalization Define the scenario optimization problem as a constrained maximization problem: , .

[0061] b) Constructing a proxy model Using Gaussian Process Regression (GPR), based on existing evaluation data... Fitting the objective function The probability distribution model (i.e., the surrogate model) that can provide any point Predicted mean ("Utilizing known high-risk areas") and predicting uncertainty ("Exploring the unknown parameter space").

[0062] c) Optimize the acquisition function Acquisition functions such as Upper Confidence Bound (UCB) are used. To weigh the options between "utilizing known high-risk areas" and "exploring the unknown parameter space." In order to explore the weighting coefficients, Larger areas tend to explore regions with high uncertainty. Smaller ones tend to utilize known high-mean regions. By solving... To determine the next set of scenario parameter samples that are most valuable for evaluation.

[0063] d) Iterative loop right Conduct a realistic simulation evaluation to obtain , to put the new data Add the dataset and update the surrogate model. Iterate repeatedly until the maximum number of iterations is reached or the comprehensive risk score converges. This process automatically and efficiently guides the search to approach the boundary of system stability.

[0064] Step (4) Archiving and Application of High-Risk Scenarios Throughout the entire optimization iteration process, all triggering faults ( or obtain a high comprehensive risk score ( Scene parameters The corresponding scripts and simulation process data (such as logs and recordings) are automatically saved to the high-risk scenario library.

[0065] The high-risk scenario database is used for: Defect reproduction and verification: After the software is modified, these scenarios are replayed to verify whether the defects have been fixed; Regression test suite: As the core high-pressure test case suite, it is used for automated testing in subsequent versions to prevent feature rollback; System hardening analysis: Analyze these scenarios to gain a deeper understanding of the system's vulnerabilities, thereby improving the robustness of intelligent software in a targeted manner.

[0066] This invention also provides an adaptive optimization system for aerospace test scenarios based on closed-loop feedback to implement the above method, such as... Figure 2 As shown, it includes: The parameter management module is used to define, store, and manage the set of optimizable scene parameters X and their constraints.

[0067] Simulation Execution and Monitoring Module: Used to load scenario scripts, drive the simulation engine to run, and collect spacecraft operational status data in real time.

[0068] Fault detection and risk assessment module: Built-in key scenario assessment index system for real-time data analysis and calculation of comprehensive risk score. .

[0069] Adaptive Optimization Engine: Integrates Bayesian optimization algorithm to manage optimization loops and generate new scene parameter suggestions based on historical evaluation results.

[0070] High-risk scenario library module: used to store, manage, and retrieve high-risk scenarios discovered during the optimization process.

[0071] Example 1 The following detailed description, in conjunction with embodiments, illustrates an adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios provided by the present invention. The specific implementation of this method may include the following steps: Step 1: Parametric Modeling and Initialization This step aims to transform the abstract testing intent into a mathematical problem that can be processed by computer-optimized algorithms. The specific implementation is as follows: (1) Abstraction of key parameters: Analyze the key scenario factors that affect the performance of spacecraft intelligent software and abstract them into a set of optimizable scenario parameter vectors. For example, parameters for obstacle avoidance testing of lunar rovers may include: Obstacle layout parameters ( Examples of obstacles include the number of obstacles, the distribution range of their location coordinates, and their size.

[0072] Environmental disturbance parameters ( ): Such as the intensity range of camera sensor noise ( (e.g., communication delay in milliseconds).

[0073] Task parameters ( Examples include the starting speed of the rover and the position tolerance of the target point.

[0074] (2) Define feasible domain and constraints: Using the syntax of a scenario description language (such as Scenic), define the physical and task-related feasible domains for the above parameters. (i.e., the range of parameter values ​​and constraints), for example, ensuring that obstacles do not overlap and that the light intensity is not lower than the sensor's minimum light sensitivity threshold.

[0075] (3) Initial sample generation: The Latin hypercube sampling (LHS) method is used in the feasible region. Generate a set (e.g., 20) of initial scene parameter samples. LHS can ensure that the sample is uniformly covered in the multidimensional parameter space, providing a good initial exploration basis for subsequent optimization.

[0076] Step 2: Simulation Execution and Risk Quantification Assessment This step is a closed-loop data collection and perception process, designed to quantify the risk level of each test scenario.

[0077] (1) Scene instantiation and simulation: Each parameter sample generated in step one This is transformed into a specific, executable simulation scenario script. The script is then loaded into a high-fidelity simulation engine (such as Webots or Gazebo), runs the intelligent software (such as navigation algorithms) of the spacecraft under test, and monitors its operational status in real time, collecting full data including position, velocity, attitude, sensor data, and mission status logs.

[0078] (2) Real-time fault detection: Based on a predefined fault dictionary, the collected data stream is matched in real time to detect whether a basic fault has occurred. Once a collision, boundary violation, task failure, or loss of connection event is detected, the corresponding binary fault flag is immediately triggered, and the occurrence time is recorded. For example, if a collision event is detected, the corresponding binary fault flag is set. .

[0079] (3) Comprehensive risk score calculation: The defined indicator system includes basic fault score ( ) and continuous risk segment ( After the simulation, the risk assessment module calculates the comprehensive risk score for the scenario based on a predefined indicator system. .

[0080] Step 3: Feedback-driven optimization of scene parameters This step is the system's intelligent decision-making center, which automatically decides the direction of the next test based on historical data.

[0081] (1) Constructing a proxy model: The Bayesian optimizer will use all parameter-score pairs evaluated in history. As training data, a surrogate model is fitted using Gaussian process (GP) regression. This model can predict any point in the parameter space. X Expected risk score and its uncertainty .

[0082] (2) Optimize the acquisition function: Construct the acquisition function using the upper confidence bound (UCB) strategy. The value of this function is high in regions with high prediction scores and high uncertainty. This is achieved by solving... Find the next set of scenario parameter samples that are most worth evaluating.

[0083] (3) Iterative loop: The sample is sent to step two for simulation evaluation to obtain a new score. and will apply the new data to Add historical datasets and update the proxy model. Repeat this process to form a closed loop, gradually guiding the search direction until termination conditions are met (such as finding high-risk scenarios exceeding a threshold, reaching the maximum number of iterations, or score convergence).

[0084] Step 4: Archiving and Application of High-Risk Scenarios This step marks the end of the closed loop and is also the point where value is output.

[0085] Automatic archiving: Throughout the optimization process, the system automatically archives all triggered faults ( (or comprehensive risk score) Scene parameters that exceed the set threshold, the corresponding scripts, and simulation data records (such as logs, keyframe images / videos) are stored in the high-risk scene library.

[0086] In summary, this invention transforms traditional static, open-loop testing into dynamic, intelligent closed-loop testing through closed-loop feedback and adaptive optimization. This fundamentally solves the industry problems of low testing efficiency and insufficient coverage of key scenarios, and significantly improves the verification capability and assurance level of the reliability of spacecraft intelligent software.

[0087] This application provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform... Figure 1 The method described.

[0088] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0089] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0091] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0092] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0093] The contents not described in detail in this specification are common knowledge to those skilled in the art.

Claims

1. An adaptive optimization method for virtual simulation test scenarios of spacecraft intelligent software, characterized in that, Includes the following steps: We analyze the key scenario factors affecting the performance of spacecraft intelligent software and abstract them into a set of optimizable scenario parameters. And define the feasible domain of scene parameters. The Latin hypercube sampling method is used in the feasible region. Generate N initial scene parameter samples. ; Sample each scene parameter , i =1~ N The simulation scenario script is converted into a simulation scenario script and loaded into the simulation engine to run the intelligent software of the spacecraft under test, and to monitor and collect the software operation status data in real time. The defined indicator system includes a basic fault score and a continuous risk score. After the simulation, the comprehensive risk score for the scenario is calculated using the collected state data based on the predefined indicator system. The basic faults are divided into collisions, boundary crossings, mission failures, and loss of contact events, which are assigned high fixed penalty points. The continuous risk score is normalized and weighted based on the vehicle's maximum tilt angle, minimum obstacle distance, average slip rate, and path detour index. Using all evaluated parameters-rating pairs As training data, a Gaussian process regression model is fitted to construct a sampling function to find the next set of scene parameter samples most worthy of evaluation. ,right A simulation evaluation was conducted to obtain a new score. and will apply the new data to Add historical datasets, update the agent model, and iterate repeatedly until the maximum number of iterations is reached or the comprehensive risk score converges. During the iteration process, sample all scenario parameters that trigger failures or whose comprehensive risk scores exceed the threshold. The corresponding scripts and simulation process data are automatically saved to the high-risk scenario library.

2. The adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios according to claim 1, characterized in that, The scenario parameter categories include: Obstacle layout parameters This includes the number, location, size, shape, and distribution of obstacles; Environmental disturbance parameters This includes sensor noise intensity, lighting conditions, ground friction coefficient, and communication delay. Task start condition parameters This includes the spacecraft's initial attitude, velocity, and target location.

3. The adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios according to claim 1, characterized in that, The software runtime status data includes position, velocity, attitude, sensor data, and task status logs.

4. The adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios according to claim 1, characterized in that, set up The basic fault categories for collision, boundary crossing, mission failure, and loss of connection are as follows: , , , ,but .

5. The adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios according to claim 4, characterized in that, The maximum tilt angle, minimum obstacle distance, and path detour indices are applicable to all spacecraft types, while the average slip rate index is only processed when the spacecraft is a vehicle type.

6. The adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario according to claim 5, characterized in that, The method for calculating the continuous risk score is as follows: Define the normalization function: Where value, min, and max are the calculated value, lower limit, and upper limit of a certain indicator, respectively; Maximum body camber ,in, Indicates the angle of the vehicle's left and right tilt. Indicates the angle at which the front of the vehicle is tilted up or down; the fractional angle is: in, For a safe tilt angle, This is the maximum permissible tilt angle; Minimum obstacle distance ,in, The distance from the spacecraft to the obstacle; the obstacle distance fraction is: in, To maintain a safe distance; Average slip rate ,in, To the expected speed of the probe, The actual speed of the probe; the slippage rate fraction is: in, This indicates the upper limit of the slippage rate, which defaults to 1.

0. Path detour ,in, This represents the actual distance traveled from the starting point to the ending point. The detour score represents the straight-line distance between the starting point and the ending point. 。 7. The adaptive optimization method for spacecraft intelligent software virtual simulation test scenarios according to claim 6, characterized in that, The comprehensive risk score The calculation method is as follows: in, Divide into various basic fault categories, For each continuous risk score, These are the weighting coefficients assigned based on the importance of each risk indicator; The higher the value, the more likely the current scenario is to expose system defects.

8. The adaptive optimization method for a spacecraft intelligent software virtual simulation test scenario according to claim 7, characterized in that, The method employs Gaussian process regression to fit a surrogate model and constructs a data acquisition function to find the next set of scene parameter samples most worthy of evaluation. Specifically: Based on comprehensive risk score As the optimization objective function With scene parameter set As an optimization variable, the scenario optimization problem is defined as a constrained maximization problem: ， ； Using Gaussian process regression, based on existing evaluation data... Fitting the objective function The probability distribution model, also known as the surrogate model, can provide the probability distribution model for any point in the parameter space. X Predicted mean and forecast uncertainty ; Constructing the acquisition function using an upper confidence bound strategy ,in, To explore the weighting coefficients; by solving Find the next set of scenario parameter samples that are most worth evaluating. .

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 8.

Images