Por favor, use este identificador para citar o enlazar este ítem:
http://cimat.repositorioinstitucional.mx/jspui/handle/1008/1164| THE IMPORTANCE OF DIVERSITY IN EVOLUTIONARY ALGORITHMS | |
| Joel Chacón Castillo | |
| Acceso Abierto | |
| Atribución-NoComercial | |
| CIENCIAS DE LA COMPUTACIÓN | |
| Population-based algorithms are flexible methods that can be effectively applied to complex optimization problems. As part of their design, several aspects have to be taken into consideration. One of them, perhaps the most important, is the early loss of diversity, which leads to premature convergence of the population. However, several works have shown that this situation can be alleviated by taking into account mechanisms that attempt to control the diversity of the population. The aim of this work is to show that the quality of current population-based algorithms can be enhanced by integrating mechanisms to explicitly manage diversity. The key is to consider the stopping criterion and elapsed period in order to dynamically alter the importance granted to the diversity. In this dissertation, enough evidence is collected to conclude that this strategy benefits population-based algorithms mainly for long runs. Even more, this strategy is empirically analyzed in both single-objective and multi-objective problems as well as for continuous and discrete domains. In the first part of this document, single-objective optimization problems are studied. At the beginning, a Differential Evolution variant with enhanced diversity maintenance is proposed. The main novelty is the use of a dynamic balance between exploration and exploitation to adapt the optimizer to the requirements of the different optimization stages. For this section, the experimental validation is carried out with several benchmark tests proposed in competitions of the "IEEE Congress on Evolutionary Computation" and with the top-rank algorithms of each competition, as well as other diversity-based schemes. The new method avoids premature convergence and significantly improves further the results obtained by state-of-the-art algorithms. Thereafter, regarding single-objective optimization, the Menu Planning Problem is addressed. This optimization problem is a complex task that involves finding a combination of menu items taking into account several types of features, such as nutritional, economical, and level of repetition, among others. To deal with the menu planning problem, some of these features are transformed into constraints. Several variants of this problem have been defined, and memetic algorithms have been quite successful in solving them. Specifically, two formulations based on the transformation of menu planning into a single-objective constrained optimization problem are studied. Each | |
| 11-10-2022 | |
| Trabajo de grado, doctorado | |
| OTRAS ESPECIALIDADES MATEMÁTICAS | |
| Versión aceptada | |
| acceptedVersion - Versión aceptada | |
| Aparece en las colecciones: | Tesis del CIMAT |
Cargar archivos:
| Fichero | Descripción | Tamaño | Formato | |
|---|---|---|---|---|
| TE 865.pdf | 2.49 MB | Adobe PDF | Visualizar/Abrir |