基于特征差异的仿真模型验证及选择方法

Validation and Selection of Simulation Model Based on the Feature Differences

为了实现在系统存在多个数据类型各异的输出时,多个备选仿真模型的验证和择优,提出了基于特征差异的仿真模型验证及选择方法.首先,将系统输出分为静态、缓变和速变三类数据,并分别给出了每类数据的特征差异度量模型;然后,采用主成分分析法从多个具有相关性的特征差异中提取出少数几个独立的主成分变量;再者依据主成分数据,采用K-均值聚类分析方法将多个备选仿真模型的输出划分为K类;最后,基于Fisher判别分析法判定参考输出是否属于其中的某类,进而实现对多个备选仿真模型的验证和选择.通过实例应用,表明了该方法的有效性.

To validate alternative models and select the most credible one when the models have many outputs of different kinds, a validation and selection method of simulation model based on feature differences is proposed. The outputs are divided into three kinds： static data, gradual data and fast data, and the measure models of feature differences for each kind data are given. The correlation among the feature differences is eliminated via principal component analysis, and several independent principal components are gained, b-~rthermore, the outputs of simulation models are divided into K kinds of clusters based on K-means clustering according to the principal components. Which cluster the output of actual system belongs to is judged by Fisher discriminant analysis. So the validation and selection of alternative models are realized. Finally, the method is validated in an application.