A Surrogate-Based Intelligent Variation Operator for Multiobjective Optimization

Abstract

Evolutionary algorithms are meta-heuristics that have shown flexibility, adaptability and good performance when solving Multiobjective Optimization Problems (MOPs). However, in order to achieve acceptable results, Multiobjective Evolutionary Algorithms (MOEAs) usually require several evaluations of the optimization function. Moreover, when each of these evaluations represents a high computational cost, these expensive problems remain intractable even by these meta-heuristics. To reduce the computational cost in expensive optimization problems, some researchers have replaced the real optimization function with a computationally inexpensive surrogate model. In this paper, we propose a new intelligent variation operator which is based on surrogate models. The operator is incorporated into a stand-alone search mechanism in order to perform its validation. Results indicate that the proposed algorithm can be used to optimize MOPs. However, it presents premature convergence when optimizing multifrontal MOPs. Therefore, in order to solve this drawback, the proposed operator was successfully hybridized with a MOEA. Results show that this latter approach outperformed both, the former proposed algorithm and the evolutionary algorithm but without the operator.