Banca de DEFESA: Rodney Oliveira Marinho Diana

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : Rodney Oliveira Marinho Diana
DATE: 27/08/2021
TIME: 14:00
LOCAL: Videoconferência
TITLE:

**

KEY WORDS:

Job scheduling, Unrelated parallel machines, Metaheuristics, Just-in-Time scheduling, Local search methods

PAGES: 203
BIG AREA: Outra
AREA: Multidisciplinar
SUMMARY:

This thesis addresses an important class of scheduling problems named unrelated parallel machines with sequence and machine-dependent setup times scheduling problems. Firstly, a literature review is performed, which raises three gaps: (i) the contribution of components used in local search has received little attention in the metaheuristics convergence validation. Consequently, there is little qualitative information to determine patterns that can enhance the construction of local search operators; (ii) the implemented metaheuristics for solving the scheduling problems incorporate many features of the own optimization criteria in the search process. As a result, these implemented metaheuristics are hard to adapt to the optimization criteria with little theoretical visibility; (iii) the literature review did not found studies for the class of scheduling problems under study guided by a Just-in-Time (JIT) policy considering time windows. This thesis has as main objective to present three contributions regarding these gaps. In the first study, a methodology for the local search operators design for metaheuristics is proposed. This methodology is evaluated in three metaheuristics previously proposed in the literature. The study shows that the use of the proposed methodology leads to a significant increase in the performance of the metaheuristics. The redesigned metaheuristics even outperform the state-of-theart approaches for the evaluated problem. Furthermore, through the analysis of the local search components, some patterns were found that can be used to build local search operators in other scenarios. The second study is aimed at building a hybrid metaheuristic approach to the evaluated scheduling problem. This approach must reduce the use of optimization criteria features in the search process. At the same time, the approach must found results with the same quality of the state-of-the-art approaches designed for a specific optimization criterion. It is shown, using four case studies, that the proposed approach finds superior or similar results to the state-ofthe-art approaches in most of the evaluated scenarios. Finally, we conducted a study about the importance and difficulty of the design of metaheuristics for the studied problems class, when guided by a Just-in-Time policy, allowing the insertion of idle times together with time windows. In this study, it is proposed to adapt, for unrelated parallel machines, a method of optimal insertion of idle times in single-machine environments. This method entails a significant increase in the order of complexity of the metaheuristics approaches. For this reason, a local search method is proposed with reduced neighborhood structures. The proposed methods are integrated into four metaheuristics previously proposed in the literature and to the hybrid metaheuristic proposed in this thesis. The metaheuristics behaviors are evaluated to the problem solution. The results show that the hybrid metaheuristic outperforms the four other approaches in most of the evaluated scenarios. Furthermore, the influence of the size of the time windows on the scheduling results is evaluated. The results indicate the existence of a linear correlation between the size of the time window and the costs arising from jobs earliness and tardiness.

BANKING MEMBERS:
Interna - ELISANGELA MARTINS DE SA
Externo à Instituição - LUIZ SATORU OCHI - UFF
Interno - MARCONE JAMILSON FREITAS SOUZA - UFOP
Externo ao Programa - MOACIR FELIZARDO DE FRANCA FILHO
Presidente - SERGIO RICARDO DE SOUZA