Research On Simulation Model Validation And Calibration Methods Under Uncertainty

Model validation and calibration is important in simulation credibility evaluation. The relevant theories and methods have been received wide attentions and achieved fruitful results. With the improvement of the accuracy of the simulation model, the uncertainty of modeling and simulation has been paid more and more attention, model validation and calibration should also consider the impact of uncertainty. How to implement the model validation and calibration process under uncertainty is the key problem. This thesis focuses on this key problem, and carries out related research.First, simulation uncertainty description and propagation methods are studied. A unified propagation method based on probability sampling for different kinds of uncertainty is proposed on the foundation of the existing probability and nonprobability uncertainty description methods. Random variable based on probability framework and interval variable, fuzzy variable, imprecise random variable, fuzzy random variable based on nonprobability framework are proposed as the uncertainty description method. The probability framework transformation model is constructed because the uncertainty description method based on nonprobability framework has the difficulty in the propagation. The unified propagation method based on probability sampling is proposed to achieve the unified propagating for different kinds of uncertainty variables described by different uncertainty description methods.Second, simulation results validation method under uncertainty is studied. To validate the simulation results under aleatory uncertainty, a model validation method based on data features is introduced. Through the feature extraction, feature validation matrix is built to measure the uncertainty and reduce the dimension of the simulation results. Bayesian factor is used to achieve the validation result. To validate the simulation multi-outputs results under both aleatory and epistemic uncertainty, while the reference data contain uncertainty, a validation method based on evidence distance is proposed. The validation data is transferred into the evidence framework with the unified evidence modeling and description method for simulation results, a fusion method for multiple evidences based on weighted average operator is proposed to fuse the results data, and evidence distance is used to achieve the validation result.Third, simulation model calibration method under uncertainty is studied. In order to solve the complex systems model calibration problem under uncertainty with various types and large amounts of data, a calibration method based on metamodel for simulation models with multivariant outputs is proposed. A multiple output consistency model based on features and Mahalanobis distance is proposed to obtain the consistency of outputs. Stochastic Kriging and Genetic Algorithm are used to speed up the calibration process. Meanwhile, in order to guarantee the efficiency and accuracy of the meta-model fitting, a meta-modeling method based on sequential experimental design is proposed.Finally, a software platform for simulation model validation and calibration is developed. In order to solve the problem that the simulation model validation and calibration work under uncertainty involves lots of simulation runs, mass of simulation result data and complicated data processing methods, the simulation model validation and calibration software platform is developed with functions of uncertainty modeling and description, uncertainty propagation, result validation and model calibration. The validation and calibration of an aircraft terminal guidance model is applied to validate the effectiveness and practicability of the platform.