An Improved Kriging Assisted Multi-Objective Genetic Algorithm

Abstract

Although Genetic Algorithms (GAs) and Multi-Objective Genetic Algorithms (MOGAs) have been widely used in engineering design optimization, the important challenge still faced by researchers in using these methods is their high computational cost due to the population-based nature of these methods. For these problems it is important to devise MOGAs that can significantly reduce the number of simulation calls compared to a conventional MOGA. We present an improved kriging assisted MOGA, called Circled Kriging MOGA (CK-MOGA), in which kriging metamodels are embedded within the computation procedure of a traditional MOGA. In the proposed approach, the decision as to whether the original simulation or its kriging metamodel should be used for evaluating an individual is based on a new objective switch criterion and an adaptive metamodeling technique. The effect of the possible estimated error from the metamodel is mitigated by applying the new switch criterion. Three numerical and engineering examples with different degrees of difficulty are used to illustrate applicability of the proposed approach. The results show that, on the average, CK-MOGA outperforms both a conventional MOGA and our developed Kriging MOGA in terms of the number of simulation calls.