Análise comparativa dos algoritmos genéticos many-objective em problemas de otimização discreta

Abstract

This research aims to show the performance of some genetic multi-goals algorithms, especially those known as many-objective, applied on a discrete problem, the knapsack problem.

In addition, a study of the evolution of genetic algorithms, wich are a non-deterministic approach to solving problems, will be written. The study will go through multi-objectives algorithms, that solve 2 or 3 objective functions, and ending in the many-objectives, which are able to solve problems with various objective functions, ie starting on 4. It is established a comparitive performance analysis of the NSGA-II algorithms (Nondominated Sorting Genetic Algorithm II) and SPEA2 (Strength Pareto Evolutionary Algorithm 2), considered multi-objective, and the most recent ones, categorized as many-objectives like MOEA/D (Multi-objective Evolutionary Algorithm Based on Decomposition), AEMMT (Multiobjective Evolutionary Algorithm With Many Tables) and AEMMD (Multiobjective Evolutionary Algorithm with Multiple Dominances). The comparison will take, as criteria, the execution time and other performance metrics applied to four or more objective functions.

This work, therefore, proves a greater effectiveness of the many-objective algorithms in detriment of multi-goals when applied to a higher amount of objective functions. It also shows that, as the complexity of the problem increases, the many-objectives algorithms perform better than SPEA2 and NSGA-II.