Consideration of Searching Mechanism and Parameters for the Distributed Probabilistic Model-Building Genetic Algorithm
|
Accession number;07A0176818
|
| Title;Consideration of Searching Mechanism and Parameters for the Distributed Probabilistic Model-Building Genetic Algorithm |
| Author;
SHIMOSAKA HISASHI
(Doshisha Univ., Grad. Sch.)
HIRAI SATOSHI
(Doshisha Univ., Grad. Sch.)
HIROYASU TOMOYUKI
(Doshisha Univ., Fac. of Eng.)
MIKI MITSUNORI
(Doshisha Univ., Fac. of Eng.)
|
Journal Title;IPSJ Transactions on Database
|
Journal Code:Z0778A
|
ISSN:0387-5806
|
|
VOL.48;NO.SIG2(TOM16);PAGE.114-124(2007)
|
| Figure&Table&Reference;FIG.15, TBL.3, REF.11 |
| Pub. Country;Japan |
| Language;Japanese |
| Abstract;In the previous research, we have proposed the Distributed Probabilistic Model-Building Genetic Algorithm (DPMBGA). DPMBGA is a distributed population model of probabilistic model-building GA and is a real-coded Genetic Algorithm. As it can search for the solution taking into consideration the correlation between solutions, the DPMBGA has high search ability. First, this thesis presents a discussion of the optimum parameter values for the DPMBGA. The DPMBGA requires many specific parameters and parameter setting experiments to realize high search ability. To resolve this problem, the DPMBGA has been improved. The improved DPMBGA can achieve the same performance as the conventional DPMBGA without setting the parameters. Second, this thesis discusses the search mechanism of the DPMBGA. The results of numerical experiments indicated that in a function with many subpeaks, it is important to maintain the diversity of the population using the island model. On the other hand, in functions with correlations among design variables, Principal Component Analysis (PCA) using the island model can be used to achieve effective search ability. In conclusion, the DPMBGA can achieve high search ability in both of these functions. (author abst.) |
|
|
|
Related Articles;
|
|
