An Enhanced Evolutionary Algorithm with a Surrogate Model

In this paper we present an enhanced evolutionary algorithm (EA) to solve computationally expensive design optimization problems. In this algorithm we integrate a genetic algorithm (GA) with a local search method to expedite convergence of the GA. We first use a GA to generate a population of data by evaluating real functions, then we construct computationally cheap surrogate models based on the available data. Thereafter, we perform gradient-based local searches on the surrogate models in lieu of the real functions. We apply the GA and gradientbased method alternatively until an optimum is reached. To guarantee convergence to the original problem, we use a trust region management to handle surrogate models. We investigate the effects of number of points used to construct the surrogate model, number of surrogate model constructed, and number of local search performed. Our numerical results, based on two single-objective problems and one multi-objective optimization problem, demonstrate the advantages of the hybrid GA over pure GAs.