A Surrogate-Based Optimization Using Polynomial Response Surface in Collaboration with Population-Based Evolutionary Algorithm

Abstract

The evaluation of system design is undoubtedly a time-consuming process with limited computational budget especially when some criteria such as reliability maximization or cost minimization are introduced as main objectives. This attracts many attentions to utilize the effectiveness of meta-models (surrogate-based methods) in the context of optimization. In this study, a collaboration between the population of Evolutionary Algorithms (population-based) and a polynomial surrogate model leads to reach global optimal points. As a population is formed to search the design space for the best solution, a response surface formation is intended in light of the fitness evaluation of population simultaneously. The accuracy of the response surface then can be increased by making beneficial use of original function evaluation of the population in the next iteration. To be more precise, construction of the surrogate model occurs from the first optimization iteration by means of population values (using original fitness function) and updating of this surrogate model is possible using the population cost of the next iterations. Meanwhile, the best solution of the surrogate model has to be injected into the population as a new member to empower the optimization search engine. The proposed creativity brings about promising results of global optimal solution with fewer function evaluations.