Open AccessNumerical method of optimization based on computer experiment planning.
Author(s) -
Lev I. Evelson,
Lev I. Evelson,
Елена Рыжикова,
Елена Рыжикова
Publication year - 2015
Publication title -
vestnik brânskogo gosudarstvennogo tehničeskogo universiteta
Language(s) - English
Resource type - Journals
ISSN - 1999-8775
DOI - 10.12737/22737
Subject(s) - mathematical optimization , quadratic equation , computer science , matrix (chemical analysis) , process (computing) , gauss , regression analysis , design matrix , algorithm , mathematics , machine learning , materials science , geometry , physics , quantum mechanics , composite material , operating system
The computer experiment methods are presented. Departure from traditional assumptions of regression analysis is taken into account: a big difference between real and Gauss statistical distribution, presence of inadmissible design matrix points and the points with a great amount of quasi-intolerable simulation runs including cases with only permissible run. Modification of optimization (with restrictions) numerical method is proposed. The method includes development and analysis of quadratic regression function. Effectiveness of the modified method for situations with inadmissible points and simulation runs is corroborated by results of the model problem solution connected with optimization of industrial process parameters.