SHIFTING OF LHS DESIGN FOR SURROGATE MODELING

Abstract eng:
Surrogate modeling (Meta-modeling) is a commonly used approach for analysis of complex systems' behavior. Time and computing demands of analytical models describing such systems are usually very high and in cases of need of multiple evaluations (for example in Monte Carlo based reliability analysis) they cannot be used. Instead, a model of the original model called surrogate model can be used. The purpose of the surrogate model is to approximate an original model's response in an arbitrary point of the design domain while constructed on a very limited and thus computationally cheap training data. The training data consist of the Design of Experiments (DoE) and corresponding responses of the original model. The choice of the DoE is crucial for the quality of the surrogate's approximation and therefore the LHS design is often used for its convenient properties. The contribution proposes a procedure of shifting of a part of the design of experiments in cases where the area of interest is located after some original model's evaluations were performed. The goal is clear: to use the already computed training data while not deteriorate the quality of the DoE.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket