A framework for memetic optimization using variable global and local surrogate models

摘要: We propose a framework of memetic optimization using variable global and local surrogate-models for expensive functions. The employs the trust-region approach but replaces quadratic models with more general RBF ones. It makes an extensive use accuracy assessment to select used improve them if necessary. also several efficient stable numerical methods its performance. Rigorous performance analysis shows proposed significantly outperforms existing surrogate-assisted evolutionary algorithms.