基于OMLHD的仿真模型验证方法

当仿真因子数目较多,而真实系统观测数据相对稀少时,仿真模型的验证工作难以进行.针对仿真因子空间庞大,对应于每一组仿真因子配置,有且仅有对真实系统在相应工作条件下的一次观测这一极端情形,提出一种基于最优拉丁超立方设计(OMLHD, Orthogonal-Maximin Latin Hypercube Designs)的仿真模型验证方法.利用仿真验证试验的优化设计技术对仿真因子配置进行筛选,合理的选择、安排验证试验;采用p值分析方法解决验证试验中的数据一致性分析问题;基于逆概率分布定理的思想,对各组验证试验一致性分析结果进行综合分析.以某导弹制导仿真模型验证为例,对方法的性能进行了讨论.结果表明该方法能够实现在整个因子空间上对仿真模型的可信性进行验证的目的.

Simulation model validation based on OMLHD method

When simulation factors are numerous while real-world observed data are sparse, the issue of validating the simulation models is problematic. An extreme case was focused that limited real-world observations were available cross the factor space, and only a single replicate was obtained on per simulation factor setting. A method based on orthogonal-maximin Latin hypercube designs (OMLHD) was proposed by which the validation experiments could be well arranged across the factor space through optimal design. The p-value test and the inverse cumulative distribution function(CDF) theorem were introduced to evaluate the statistical consistency of the simulation/observation data, and combine the analysis results to make an overall validation study on the entire factor space. An example of validation of a guided missile simulation was taken, which demonstrates that the method is useful.