Please use this identifier to cite or link to this item: http://cimat.repositorioinstitucional.mx/jspui/handle/1008/1161
IMPROVEMENT DIRECTION MAPPING METHOD FO RMULTI-OBJECTIVE LOCAL SEARCH OPTIMIZATION
Salvador Botello-Aceves
Acceso Abierto
Atribución-NoComercial
CIENCIAS DE LA COMPUTACIÓN
Multi-objective Local Search Algorithms (MOLSAs) is an emergent specialization of Multi-Objective Optimization. Unlike Multi-Objective Evolutionary Algorithms (MOEAs), which are global optimization algorithms that apply stochastic operators in order to compute a set of non-dominated solutions, MOLSAs require local operators to further improve the non-dominated front by displacing each solution independently, applying local information. Line search MOLSAs compute directions in the search domain using first-order and second-order information. Several proposals introduced different strategies to calculate search directions within their corresponding descent cone, compound by all the directions that improve all objective functions. However, the main features of the non-dominated solution front, such as convergence toward the true non-dominated front, distribution of the solutions, and their spread on the objective space, are loosely controlled. The framework and the mathematical development of a novel MOLSA are introduced in this thesis. The Improvement Directions Mapping (IDM) is a line search MOLSA that proposes improvement directions in the objective space and transforms them to search directions in the variable space. The proposal presents a proper control from the aforementioned features of the non-dominated front. The method is decoupled into two tasks: 1) The computation of the improvement directions in the objective space; 2) the transformation function that relates the objective space and the variable space. This work delves into the independent study of each task, presenting their main features, advantages, and issues; on which research lines are being suggested, while proposing novel methodologies, attending to alleviate such predicaments. Experiments show the ability of the IDM as a stand-alone algorithm, shedding light on the behavior of the IDM under various combinations of improvement direction and transformation tensors, encouraging the coupling with different global search engines. The flexibility of the IDM to be coupled with different MOEAs presents the proposed method as an attractive scheme to steer the stochastic search indirectly. The last contribution of this work is the application of the IDM within the reproductive operator of a multi-objective estimation of distribution algorithm. The search direction from a particular solution, taken from the IDM method, is used to orientate a local probability distribution towards the Pareto fr
18-03-2022
Trabajo de grado, doctorado
OTRAS
Versión aceptada
acceptedVersion - Versión aceptada
Appears in Collections:Tesis del CIMAT

Upload archives


File Description SizeFormat 
TE 861.pdf10.46 MBAdobe PDFView/Open