Abordagem C-GRASP com adaptação automática para otimização global contínua

Abstract

The use of meta-heuristic is strongly recommended for solving optimization problems. They are usually used to solve discrete optimization problems. The metaheuristics aim to achieve good approximated solutions, and adaptive methods are implemented to achieve higher performance, thus improv ing the way the goal-heuristic works. Greedy Randomized Adaptive Search Procedure (GRASP), Variable Neighborhood Search (VNS) and Variable Neighborhood Descent (VND) have been used to solve continuous global optimization problems along with its implemented adaptations. The meta-heuristic Continuous GRASP is included in the class that have most been adapted in order to solve these problems. Some works are relevant to implement these adaptations Continuous GRASP. The Directed Continuous GRASP (DC-GRASP) is a proposed improvement to accelerate the convergence of the C-GRASP method by generating downward directions without derivative calculations using a local search based on the Adaptive Search Pattern(APS)method. An automatic adjustment is inserted into the DC-GRASP in order to define the parameters in high-dimensional functions, using other meta-heuristic, the Particle Swarm Optimization (PSO). For functions with few dimensions, a step size ampliation mechanism on APS and the use of inexact linear search has been proposed to improve the use of iteration methods. To validate the method implemented, the last adaptive implementations Continuous GRASP found in recent literature were used,as well as in original versions published in early articles using meta-heuristic. Some computational experiments were performed in a benchmark test of functions in an known global minimum, thus proving the efectiveness of the method for aiding in convergence.