Colônias de abelhas como modelo para otimização multimodal em espaços contínuos: uma abordagem baseada em alocação de tarefas

Abstract

Nowadays, it is common to use optimization techniques in several areas of application. Regardless the application, the modeling of real-world problems rarely takes into account all possible fixed or temporary constraints that may eventually arise. This scenario motivates the proposal of mechanisms dedicated to the generation and maintenance of diversity during the search for the optimal solution, so that alternative high quality candidate solutions can be found in a single execution of an optimization algorithm. This avoids the need for new modeling and executions, given the possibility of choice, by the user, of a solution among those returned that meets the constraints or that he/she deems to be the most feasible to be implemented. In this context, this thesis proposes a new Swarm Intelligence algorithm, called Opt- Bees, to treat mono-objective optimization problemas in continuous spaces, inspired by the foraging behavior of bees and by mechanisms (their models) involved in the self-organizing process of task allocation in insect societies. The OptBees was designed to have the inherent ability to generate and maintain diversity, finding distinct local optima of the problem, so that multiple high-quality solutions can be returned in a single run. The algorithm was evaluated by conducting experiments based on the set of test problems proposed for the Competition on Real-Parameter Optimization of the CEC Special Session on Real-Parameter Optimization, which occurred in IEEE Congress on Evolutionary Computation (CEC) in 2005. The performance of OptBees was compared to that of several tools based on different approaches, including algorithms inspired by the behavior of bees. The results obtained demonstrated that the proposed tool, can maintain diversity throughout its execution whilst still being competitive in the search for the global optimum.