Performance Comparative of Surrogate Models as Fitness Functions for Metaheuristic Algorithms

Real-world optimization problems can use a high number of computational resources to be solved because of the complexity of the fitness function or objective function. This situation negatively impacts the performance of any Heuristic or deterministic solver since most of the computational time is invested into processing the fitness function. Surrogate models are used to reduce the computational time of processing complex functions by replacing extensive numerical simulations with approximate functions that are faster to evaluate. Surrogate models are constructed by evaluating the original model at a set of points (training), and then using the corresponding evaluations to construct an approximate model based on given mathematical functions. The main idea is to have a faster function that can guide an optimization solver with a minimum error or difference against the original fitness function. In this work, several surrogated models are constructed and compared in the context of a heuristic optimization solver to analyze their performance as fitness functions.