基于元模型的多元输出仿真模型校准方法研究

为解决具有多元不同类型输出的仿真模型校准问题，提出一种基于优化和元模型的仿真模型校准方法.首先提出一种基于双层嵌套拉丁超立方抽样（LHS）的不确定性参数传播方法，获得系统同时含有认知和固有不确定性时的输出；其次，给出一种基于数据特征的仿真输出一致性度量方法，实现仿真多元异类输出的一致性度量；进而，利用随机Kriging模型拟合认知不确定性抽样样本与仿真输出一致性度量结果的元模型，并在该元模型上通过遗传算法实现校准过程.最后，通过实例验证了本文所提方法的有效性. 

A Calibration Method Based on Surrogate Model for Simulation Models with Multi-Variant Outputs

To solve the calibration problem of simulation model with multi-variant and different kinds of output data, a calibration method based on optimization and surrogate model was presented. To acquire the output of simulation model with both of aleatory and epistemic uncertainty, an uncertainty propagation method based on two stage nested latin hyper sample(LHS) was introduced. Then, a coherence measurement method based on data feature was used to measure the coherence of the simulation and reference outputs. Furthermore, a stochastic Kriging model was applied to build the data coherence surrogate model of the simulation output and epistemic uncertainty sample. And based on the surrogate model, the calibration results were obtained via the genetic algorithm. Finally, the method was validated in the application.