Banca de DEFESA: CLAUDIO LUCIO DO VAL LOPES

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : CLAUDIO LUCIO DO VAL LOPES
DATE: 30/11/2022
TIME: 13:30
LOCAL: Videoconferência
TITLE:

MULTI- AND MANY-OBJECTIVE OPTIMIZATION: SOME ADVANCES TOWARDS THEORETICAL ASPECTS IN PERFORMANCE QUALITY INDICATORS AND EVOLUTIONARY FRAMEWORKS

KEY WORDS:

Multi- and many-objective optimization, quality indicators, evolutionary algorithms.

PAGES: 221
BIG AREA: Outra
AREA: Multidisciplinar
SUMMARY:

Many-Objective Optimization (MaO) refers to optimization problems having four or more objectives, the increase in objective dimensionality brings some complex issues, such as the ineffectiveness of the Pareto dominance relation, quality indicators calculation, solution sets visualization, balancing convergence and diversity, and others. A key issue in many- and multi-objective optimization is comparing and assessing solution sets obtained by optimization algorithms. This is not a simple task; the outcome of many-objective optimization algorithms is typically a set of incomparable solutions. Using a quality indicator to reflect the inner Pareto front characteristics requires careful design/selection of such indicators. In the first part of this thesis, we deal with the Dominance move (DoM) quality indicator. We propose novel approaches to calculate DoM using mixed-integer programming (MIP) models and an approximate method using machine learning techniques. In general, our attempts uses the dominance move quality indicator as a suitable way to measure, compare, and assess many-objective problems. Another challenge in MaO is to provide a true representative set with the desired number of Pareto-optimal solutions in a reliably well-distributed set. In the second part of this work, we propose a multi-stage framework involving reference-vector-based evolutionary multi- and many-objective algorithms that attempt to rectify previous stages’ shortcomings by careful executions of subsequent stages so that a prescribed number of well-distributed and well-converged solutions are achieved. The results presented in this thesis come from the attempts to address challenges in evolutionary many- and multi-objective optimization. This research has not only analyzed and systematically evaluated the existing approaches but has also extended them in innovative directions related to quality indicators and improvements in multistage approach for convergence and diversity balance in evolutionary algorithms.

BANKING MEMBERS:
Externo à Instituição - KALYANMOY DEB
Interno - ADRIANO CHAVES LISBOA - IFSC
Externo à Instituição - CARLOS MANUEL MIRA DA FONSECA - UC
Interna - ELISANGELA MARTINS DE SA
Interna - ELIZABETH FIALHO WANNER
Presidente - FLAVIO VINICIUS CRUZEIRO MARTINS
Externo à Instituição - RICARDO HIROSHI CALDEIRA TAKAHASHI - UFMG