Métodos de otimização para alocação de recursos em redes ópticas OCDMA

• Autor(a) principal: Martinez, Cristiane Aparecida Pendeza
• Data de Publicação: 2020
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))
• Texto Completo: http://repositorio.utfpr.edu.br/jspui/handle/1/5310
• topic: Otimização matemática; Programação não-linear; Redes neurais (Computação); Controladores elétricos; Mathematical optimization; Nonlinear programming; Neural networks (Computer science); Electric controllers; CNPQ::CIENCIAS EXATAS E DA TERRA; Engenharia Elétrica

This thesis deals with a study on resource allocation in optical code division multiple access networks and aims to analyze optimization techniques for solving three resource allocation problems: the minimum power allocation problem, the problem of maximizing energy efficiency and the spectral-energy efficiency tradeoff problem. For the minimum power allocation problem, we propose the implementation of the modified Hopfield artificial neural network approach and also the use of nonlinear programming methods such as sequential quadratic programming and the augmented Lagrangian method, which are compared in performance-complexity, algorithmic robustness and quality of the solution. For the problem of maximizing energy efficiency in optical code division multiple access, four non-linear programming methods were proposed and analyzed: the augmented Lagrangian method, the sequential quadratic programming method, the majorationminimization technique and the Dinkelbach method. All the methods used were able to achieve convergence for the adequate solution of the problem. However, in some specific yet realistic configurations of the multiple access communication system with a high number of users, the Dinkelbach method is inefficient in terms of full convergence. Another proposed strategy for solving the problem of maximizing energy efficiency combines the augmented Lagrangian with evolutionary heuristic methods, namely optimization by particle swarm, as well as the hybridization of the particle swarming technique with the gravitational search algorithm, also combined with the Quasi-Newton method. The structure of the augmented Lagrangian optimization allows a better adjustment to the problem while providing, more information about the solution. For the problem of the energy-spectral efficiencies tradeoff, namely resource efficiency, two multiobjective optimization techniques were used, the weight sum method and the Є-restricted method, both associated with the augmented Lagrangian method, which was employed in solving the generated subproblems. As a result, the quality of the solution towards the Pareto limit was checked.