Comparison of Gene Expression Programming and Common Metamodeling Techniques in Engineering Design

Abstract

To reduce the tremendous computational expense of implementing complex simulation and analysis in engineering design, more and more researchers pay attention to the construction of approximation models. The approximation models, also called surrogate models and metamodels, can be utilized to replace simulation and analysis codes for design and optimization. Commonly used metamodeling techniques include response surface methodology (RSM), kriging and radial basis functions (RBF). In this paper, gene expression programming (GEP) algorithm in evolutionary computing is investigated as an alternative technique for approximation. The performance of GEP is examined by its innovative applications to the approximation of mathematical functions and engineering analyses. Compared to RSM, kriging and RBF, GEP is demonstrated to be more accurate for the small sample size. For large sample sets, GEP also shows good approximation accuracy. Additionally, GEP has the best transparency since it can provide explicit and compact function relationships and clear factor contributions. Overall, as a novel metamodeling technique, GEP exhibits great capabilities to provide the accurate approximation of a design space and will have wide applications in engineering design, especially when only a few sample points are selected for approximation.