295 0

An efficient hybrid metaheuristic algorithm for solving constrained global optimization problems

Title
An efficient hybrid metaheuristic algorithm for solving constrained global optimization problems
Author
박규병
Advisor(s)
최동훈
Issue Date
2016-02
Publisher
한양대학교
Degree
Doctor
Abstract
An efficient hybrid metaheuristic (eHM) algorithm is proposed to solve constrained global optimization problems. This eHM algorithm composes a combination of two metaheuristic algorithms and the addition of a new constraint handling technique called the feasibility boundary prediction (FBP) technique. To develop eHM algorithm, rank-iMDDE and the modified cuckoo search (MCS) are employed. rank-iMDDE is the newest variant of differential evolution and it is outstanding for exploitation; MCS is an improved version of cuckoo search and it is outstanding for exploration. Moreover, a crossover operator for rank-iMDDE is modified to solve various problems. Using the bi-population concept, current individuals are divided into two sub-populations according to their fitness values, so that each sub-population can perform independently. The top-ranked sub-population is updated using rank-iMDDE, and the low-ranked sub-population is updated using MCS. As there are two independently performing sub-populations, this process can be parallelled using two or more processors to increase its speed. In addition to the hybridization of rank-iMDDE and MCS, the FBP technique is also applied in the eHM algorithm. In the FBP technique, a quadratic regression model is employed to predict the boundary between feasible and infeasible regions. Because infeasible individuals can be replaced by feasible individuals or individuals in the less constraint violation region using the approximated boundary between feasible and infeasible regions, the FBP technique leads to fast convergence of the eHM algorithm. As the performance of the eHM algorithm is highly dependent on its parameter settings, the parameter values are determined using a proposed parameter tuning method, which employs an optimal Latin hypercube design and statistical analysis to determine recommended parameter ranges. To verify the performance of the eHM algorithm with recommended parameter values for solving constrained global optimization problems, it is compared to seven state-of-the-art algorithms using 22 mathematical problems, and to various state-of-the-art algorithms using five engineering problems. The eHM algorithm is found to be competitive in the performance of the mathematical test problems and the engineering problems.
URI
https://repository.hanyang.ac.kr/handle/20.500.11754/126758http://hanyang.dcollection.net/common/orgView/200000427903
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > MECHANICAL CONVERGENCE ENGINEERING(융합기계공학과) > Theses (Ph.D.)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE