Desenvolvimento de um algoritmo de otimização evolutivo auto-adaptativo para a resolução de problemas de otimização com variáveis mistas

Abstract

In last decades, evolutionary algorithm have been received considerable attention in different fields of science and engineering, with emphasis in engineering systems design. Despite at the large number of applications, these algorithms consider that parameters are constant during the search process, which may result in a search less efficient in design space. In addition, realistic problems are more complex in relation to the nature of project variables involved, i.e., these problems may present mixed variables (real (or continuous), integer, binary and discrete). In this contribution, it is proposed a new strategy to update the Firefly Algorithm (FA) parameters, as well as a new methodology to generate candidates. It is also proposed a strategy for the treatment of problems with mixed variables. An analysis of the parametric sensitivity is performed to understand how the choice of the values of each parameter of the FA could influence the method optimization process and still compromise the quality of the solution. The results obtained with the application of proposed methodology in mixed and real problems demonstrate that the performance of FA can be improved, in terms of convergence and computational cost, in relation to canonical algorithm.