Surrogate model-driven bio-inspired optimization algorithms for large-scale and high-dimensional problems

Complex engineering problems in the modern era are challenging to solve mainly due to their numerical modeling and analytical complexities. This chapter addresses the use of surrogate model–assisted bio-inspired optimization algorithms for handling the solution of large-scale and high-dimensional computation-intensive optimization problems. In most of the large-scale high-dimensional optimization problems, the majority of computation is involved in the repetitive function calls to evaluate the system response/behavior under consideration. The quality of solution depends on the system response estimation, and in most cases, high-fidelity models are used to obtain accurate results. Conventional techniques, such as evolutionary algorithms, require a great number of such high-fidelity function calls. To this end, the use of low-cost surrogate models or metamodels, which approximate the original model mathematically, significantly reduces the computation cost for desired accuracy level in an optimization solution. The surrogate model training requires a minimum amount of evaluations of the original model at support points, and efficient design of an experiment can be performed for that. In this chapter, the more popular surrogate models that are extensively used in real-life engineering simulation, polynomial regression models, support vector regression models, and Gaussian process regression models or Kriging surrogate models are discussed. Furthermore, multilevel bio-inspired optimization algorithms are covered, wherein the inner loop consists of surrogate model–based response quantification for objective and constraint functions. Finally, a hierarchical surrogate model–based particle swarm optimization algorithm, effective on a range of large-scale design optimization problems, is addressed.