Design Experiments For The Construction Of Simulation Metamodels

Metamodels are used as analysis tools for solving optimization problems or as surrogates used as building blocks in larger scale simulations. The metamodel replaces the simulation model by a simplified input-output relationship, frequently a mathematical function with customized parameters. The construction of a metamodel is based on the simulation results for a set of design points. In order to collect statistical information each design point may be simulated repeatedly. This paper explores the precision of the resulting metamodel based on the tradeoff between higher number of design points versus a higher number of replications at a smaller number of design points.