Divulgação - Defesa Nº 274

Aluno: João Paulo Martins Alcântara

Título: "Handling Constraints in FSS – Enhanced Fish School Search Algorithm for Constrained Optimization (e-rwFSS)".

Orientador: Fernando Buarque de Lima Neto

Coorientador: Marcelo Gomes Pereira de Lacerda

Examinador Externo: Hugo Valadares Siqueira (UTFPR)

Examinador Interno: Bruno José Torres Fernandes (PPGEC)

Data-hora: 31 de Julho de 2023 às 09:00h.
Local: Formato Remoto

Resumo:

Metaheuristics can be used to find a good set of parameters values in optimization problems in them industry. However, hazardous or unfeasible values combinations can arise throughout the search process, which can affect productivity and safety. One way to mitigate this problem is to make metaheuristic approaches incorporate constraints to prevent unsafe behaviour. In this work, we propose modifications on the restricted niching version of Fish School Search algorithm (rwFSS) to create a new version capable of better tackling constrained optimizations, referred as Enhanced Restricted Weight-Based Fish School Search (e-rwFSS). This algorithm exploits the search space in terms of fitness and feasibility and provide multiple solutions. e-rwFSS enables the constraints to be tackled gradually using adjustable tolerances and favours feasible individuals on sub swarm leader’s selection to enhance algorithm’s convergence. The proposed approach was compared to rwFSS and to other state-of-the-art algorithms with proven capabilities on solving constrained optimization problems (COPs). The benchmark functions for the CEC 2020 competition on real-world constrained optimization were used to evaluate the algorithms. Tests proved that e-rwFSS enhanced feasibility capabilities regarding equality constraints, but it still struggles to show competitiveness with other consolidated state-of-the-art algorithms on highly constrained search spaces.