Self-configuring multi-strategy genetic algorithm for non-stationary environments

Many real-world problems of design and control in a field of the aerospace lead to optimization problems. Such optimization problems are complicated and become a great challenge to many optimization techniques. Moreover, many real-world optimization problems are dynamic and changing over time. Changes occur in the parameters, objectives and/or problem constraints. In this case, search algorithms should have the capability to track moving optima and adapt to a new environment. In past years many approaches for non-stationary optimization were proposed. The best results are achieved using a stochastic population-based search such as evolutionary and genetic algorithms. Unfortunately, real-world non-stationary optimization problems include various types of changes and are poorly predictable, thus there is a problem of choosing a proper optimization technique and tuning its parameters. This study presents a novel approach for designing a multi-strategy genetic algorithm based on a hybrid of the island model, cooperative and competitive coevolution schemes. The approach controls interactions of different genetic algorithms and leads to the self-configuring solving of problems with a priori unknown structure. A short survey on non-stationary optimization problem and methods is presented. The results of numerical experiments for benchmark problems from the CEC competition are discussed. The proposed approach has demonstrated efficiency comparable with other well-studied techniques for non-stationary optimization. And it has significant advantage - it does not require the participation of the human-expert, because it operates in an automated, self-configuring way.