ANÁLISIS Y DISEÑO DE ALGORITMOS EVOLUTIVOS UTILIZANDO MÚLTIPLES ENFOQUES HÍBRIDOS

Abstract

Evolutionary Computation Techniques (ECT) are part of artificial intelligence discipline concerned with the design of optimization algorithms for solving complex optimization problems through the search of the optimal solution. Traditionally, several ECT have been conceived by the abstraction of natural, biological or even social phenomena as search strategies to improve the location and allocation of global optima while decreasing the possibility of being stagnated on suboptimal solutions. Although such methodologies are designed to meet the requirements of generic optimization problems, no single evolutionary algorithm can solve all the problems competitively. Therefore, researchers have been devoted to find novel optimization strategies to achieve better performance indexes. This thesis presents some new ECT designed considering different persectives than traditional ECT. The proposed ECT adopt the developmental approach, which is based on fuzzy logic, clustering, multi-agent consensus and knowledge-based paradigms to generate novel search strategies in the optimization process. Fuzzy logic emulates the human reasoning in the use of imprecise information to generate decisions. Unlike traditional approaches, fuzzy logic comprises an alternative way of processing, which permits modeling complex systems through the use of human knowledge. Clustering is defined as the process of dividing a set of elements into disjoint and homogenous sets, called clusters and it is commonly used in classification problems. On the other hand, multi-agent systems involve the cooperation of agents through information sharing mechanisms to accomplish certain tasks. These mechanisms connect several agents using simple local behaviors to generate complex interaction models among agents to solve certain tasks. Finally, the presented knowledge-based approach is a field closely related to data mining and machine learning which consider the process of identifying novel, significant, potentially useful information in the data to find hidden relationshps among the data.