Large Scale Optimization of Computationally Expensive Functions: an approach based on Parallel Cooperative Coevolution and Fitness Metamodeling

In recent years, research on large scale global optimization (LSGO) provided metaheuristics able to effectively tackle real-valued objective functions depending on thousand of variables. Nevertheless, finding a suitable solution of LSGO problems often requires a significantly high number of fitness evaluations. Therefore, when the objective function is computationally expensive, metaheuristics-based solutions of LSGO problems can easily become infeasible or at least unattractive. In this paper, we address such an issue with a joint approach based on problem decomposition, fitness meta-modeling and parallel computing. We present a preliminary numerical investigation of the proposed methodology, which provided significant gains in terms of both exact evaluations of the objective functions and parallel speedup.