RT info:eu-repo/semantics/doctoralThesis
T1 Técnicas de optimización paralelas :
esquema híbrido basado en hiperheurísticas
y computación evolutiva
A1 Segura González, Carlos
K1 Investigación operativa
AB Optimisation is the process of selecting the best element from a set of availablealternatives. Solutions are termed good or bad depending onits performance for aset of objectives. Several algorithms to deal with such kindof problems have beendefined in the literature. Metaheuristics are one of the mostprominent techniques.They are a class of modern heuristics whose main goal is to combine heuristics ina problem independent way with the aim of improving their performance. Meta-heuristics have reported high-quality solutions in several fields. One of the reasonsof the good behaviour of metaheuristics is that they are defined in general terms.Therefore, metaheuristic algorithms can be adapted to fit the needs of most real-lifeoptimisation. However, such an adaptation is a hard task, andit requires a highcomputational and user effort.There are two main ways of reducing the effort associated to the usage of meta-heuristics. First, the application of hyperheuristics andparameter setting strategiesfacilitates the process of tackling novel optimisation problems and instances.  Ahyperheuristic can be viewed as a heuristic that iteratively chooses between a setof given low-level metaheuristics in order to solve an optimisation problem.  Byusing hyperheuristics, metaheuristic practitioners do not need to manually test alarge number of metaheuristics and parameterisations for discovering the properalgorithms to use. Instead, they can define the set of configurations which mustbe tested, and the model tries to automatically detect the best-behaved ones, inorder to grant more resources to them. Second, the usage of parallel environmentsmight speedup the process of automatic testing, so high quality solutions might beachieved in less time.This research focuses on the design of novel hyperheuristics and defines a set ofmodels to allow their usage in parallel environments. Different hyperheuristics forcontrolling mono-objective and multi-objective multi-point optimisation strategieshave been defined.  Moreover, a set of novel multiobjectivisation techniques hasbeen proposed. In addition, with the aim of facilitating theusage of multiobjectivi-sation, the performance of models that combine the usage of multiobjectivisationand hyperheuristics has been studied.The proper performance of the proposed techniques has been validated with aset of well-known benchmark optimisation problems. In addition, several practicaland complex optimisation problems have been addressed.  Some of the analysedproblems arise in the communication field. In addition, a packing problem proposedin a competition has been faced up.  The proposals for such problems have notbeen limited to use the problem-independent schemes. Instead, new metaheuristics,operators and local search strategies have been defined. Such schemes have beenintegrated with the designed parallel hyperheuristics with the aim of accelerating theachievement of high quality solutions, and with the aim of facilitating their usage.In several complex optimisation problems, the current best-known solutions havebeen found with the methods defined in this dissertation.
PB Universidad de La Laguna. Servicio de Publicaciones
SN 978-84-15910-59-6
YR 2014
FD 2014
LK http://riull.ull.es/xmlui/handle/915/57
UL http://riull.ull.es/xmlui/handle/915/57
LA en
DS Repositorio institucional de la Universidad de La Laguna
RD 03-may-2024