A Surrogate-Assisted Cooperative Co-evolutionary Algorithm for Solving High Dimensional, Expensive and Black Box Optimization Problems

Many research efforts have been recently focus to solve large-scale global optimization (LSGO) problems by means of evolutionary algorithms. Cooperative co-evolution has been proposed to solve such problems depending on thousands of variables. This methodology has proved very efficient in solving a wide range of LSGO problems. Nevertheless, it often requires an extremely large number of function evaluations to reach a suitable solution. This is somewhat problematic when the function evaluation is computationally expensive. A globally effective approach to high-fidelity optimization problems based on such expensive analyses lies in the exploitation of surrogate models. They act as cheap-to-evaluate alternatives to the original high-fidelity models reducing the computational cost, while still providing improved designs. This kind of optimization process, referred to as surrogate-assisted optimization, has proved very efficient on small-dimensional problems but suffers from the curse of dimensionality to solve LSGO problems. In this paper, cooperative co-evolution was combined with surrogate-assisted optimization in order to efficiently solve high dimensional, expensive and black-box problems. Experimental results are provided on a wide set of benchmark problems and show promising results for the proposed algorithm.