PERFORMANCE COMPARISON OF METAMODELING METHODS FROM THE PERSPECTIVE OF SAMPLE QUALITY MERITS

摘要

Although many metamodeling methods have been developed in the past decades to model the relationships between input and output parameters, selection of an appropriate or the optimal metamodel for solving a certain engineering problem is not a trivial task. Various performance measures of different metamodels are strongly influenced by the characteristics of sample data. This research focuses on the study of the relationships between sample data characteristics and metamodel performance measures considering different types of metamodeling methods. In this research, sample quality merits are introduced to quantitatively model the characteristics of sample data. In this work, four types of metamodeling methods, including multivariate polynomial model, radial basis function model, kriging model and Bayesian neural network model, three sample quality merits, including sample size, uniformity and noise, and four performance evaluation measures, including accuracy, confidence, robustness and efficiency, are considered to study the relationships between the sample quality merits and the performance measures of the metamodeling methods.