Adaptação do algoritmo genético NSGA-DO à problemas de otimização multiobjetivo estáticos e dinâmicos

Abstract

Evolutionary Algorithms (EAs) are useful in solving Multi-Objective Optimization Problems (MOOPs) because they allow finding different solutions with different compensations for the objectives. One class of EAs are Genetic Algorithms (GAs), which use parallel search and optimization techniques based on natural selection and genetic reproduction. A GA commonly applied in the resolution of MOOPs, both artificial and in the real world, is the NSGA-II, which is sometimes used as a basis for the development of other algorithms, such as the NSGA-DO. The field of Multi-objective Optimization (MOO) is consolidated, we currently have different benchmarks, performance metrics and efficient AEs. However, regarding the latter, what is observed is that the performance of the algorithms is proportional to their complexity, which induces researchers from other fields to continue to prefer the NSGA-II. Furthermore, interest in Multi-objective Dynamic Optimization (DMOO), in which the environment changes over time, has intensified only in recent years and there are many challenges in this emerging field of research. Regarding the NSGA-DO, it proposes modifications in part of the NSGA-II, and even having shown superior performance in other fields, the algorithm does not present satisfactory results when applied to continuous MOOPs. In this context, recognizing the simplicity and potential of the recent algorithm, as well as the need for advances in the field of DMOO, the objective of this research was the development of improvements to NSGA-DO, as well as the elucidation of important issues related to the field of DMOO. The methodology adopted here was divided into two phases partially interspersed. In the first phase, classified as a descriptive bibliographical research, review studies published in the field of DMOO were identified, described and analyzed. In the second phase, classified as an explanatory experimental research, the evolutionary strategy of the NSGA-DO was investigated and improvements were applied. As a result of the analysis of the studies, it can be seen that the main challenges in the field of DMOO revolve around detecting changes and responding to changes. In this process, a DMOA (Dynamic Multi-objective Algorithm) faces difficulties related to the preservation of diversity, convergence considering the new environment and recovery of possible unfeasible solutions. On experimentation, the modifications applied to NSGA-DO resulted in a new GA, Modified NSGA-DO (MNSGA-DO), which i surpasses NSGA-DO and even NSGA-II in problems with different characteristics . Also, a dynamic variant of MNSGA-DO was proposed, the Dynamic MNSGA-DO (D-MNSGA-DO), which achieved satisfactory performance, managing to track and respond to changes in the environment. With the results obtained, it can be concluded that the present study achieved its objectives by proposing a new GA with a simple strategy and able to solve MOOPS and DMOPs, as well as presenting a compilation of review studies published over the years, these in the field from DMOO