Surrogate Model-Driven Evolutionary Algorithms: Theory and Applications

Engineering optimization problems are challenging to solve mainly due to their numerical modeling and analysis complexities. This chapter deals with the efficient use of surrogate model-driven evolutionary algorithms, built hierarchically for the solution of large-scale computation intensive optimization problems. In most optimization problems, the majority of computation is involved in repetitive function calls to evaluate the system response/bahaviour under consideration. The quality solutions depends on the system response estimation, and in most cases high fidelity models are used to get accurate results. Conventional evolutionary algorithms require a great number of such high fidelity function calls. Here, we use low cost surrogate models or metamodels, which approximate the original model mathematically, but significantly reduce the computation cost for a desired accuracy level. The surrogate model training requires a small amount of evaluations of the original model at support points. The hierarchical surrogate model-based PSO algorithms we propose are tested on a range of large-scale design optimization problems and compared with other well-known surrogate modeling techniques.