The Parameter-less Evolutionary Search for real-parameter single objective optimization

Abstract

A parameter-less algorithm allows optimal solutions to be found without the need for setting the control parameters. Namely, finding an appropriate parameter setup for an evolutionary algorithm is a challenging research problem, and the setup optimality is crucial for algorithm's good performance. Therefore, the approaches that are able to solve any problem without any human intervention to set suitable control parameters are particulary interesting. The Parameterless Evolutionary Search (PLES) algorithm, with its real-value and combinatorial version, is based on a basic genetic algorithm, but it does not need any control parameter to be set in advance. It is able to find optimal, or at least very good, solutions relatively quickly, and without the need for a parameter-setting specialist. The last of these is a very important issue when used by engineers that do not have a detailed background knowledge: neither about optimization algorithms, nor about the settings of their control parameters. The efficiency of the proposed parameter-less algorithm was already evaluated using theoretical and real-world problems, being either real-valued or combinatorial. It was shown that the presented, adaptive, parameter-less algorithm has a faster convergence than comparable algorithms. Furthermore, it demonstrates its search ability by finding the solution without the need for predefined control parameters.