Múltiplos enxames combinados com métodos de arquivamento, pontos de referência e topologias na otimização com muitos objetivos

Abstract

Multi-Objective Optimization Problems can be classified as a set of problems that have more than one conflicting objective function. In these problems, the objective functions to be optimized have performance indexes that are conflicting, that is, usually when one value of an objective function has an improvement, a value of another objective function tends to worsen. With this, it is necessary to obtain a set of better solutions, where the values of the objective functions are simultaneously acceptable. It can be emphasized that in this class of problems the number of best solutions increases exponentially as the number of objectives increases. In this sense, this increase in the number of solutions causes a deterioration in the search for better solutions, making progress towards optimum solutions difficult. Despite the successful application of several Multiobjective Evolutionary Algorithms to these types of problems, most studies focus on problems with a small number of objectives. In addition, these algorithms suffer from search deterioration when the number of optimized objective functions is greater than three. Thus, recently there is the search for new techniques and algorithms that seek to reduce the deterioration of multiobjective algorithms. The area that studies these new techniques is called Many-Objective Optimization and multiobjective problems that have more than three functions are classified as Many-Objective Optimization Problems. Recently, different approaches have been proposed to improve the performance of these algorithms in optimization problems with many objectives. One of these approaches is the use of multiple populations in multi-objective particle swarm optimization, which we call Multiple Swarms. Multiple Swarms are techniques for exploring parallel populations to decompose the problem and optimize it in a collaborative way. In this work we developed algorithms that incorporate the characteristics of multiple swarms with topologies, archiving methods and reference points to solve Many-Objective Optimization Problems. The first algorithm developed involved archiving methods and topologies, another explored reference points, and the latter worked on modifications of reference points in order to achieve good convergence and diversity in these types of problems. A set of experiments is done to evaluate the proposed algorithms and seek to identify the best configuration of each algorithm. In addition, a performance analysis of the algorithms is performed comparing them to the literature methods.