J0110205 DAKOTAとJAGUARによる最適化([J011-02]解析・設計の高度化・最適化(2))

Abstract

DAKOTA (Design Analysis Kit for Optimization and Terascale Applications) toolkit is developed by Sandia National Laboratories, and available under a GNU Lesser General Public License (LGPL). DAKOTA provides the following 4 capabilities; (1) optimization with gradient and nongradient-based methods, (2) uncertainty quantification with sampling, reliability, stochastic expansion and epistemic methods, (3) parameter estimation with nonlinear least squares methods, and (4) sensitivity/variance analysis with design of experiments and parameter study methods. JAGUAR (JAva GUi for Applied Research) is a Java software tool providing an advanced text editor and graphical user interface to manipulate DAKOTA input specifications. In this paper, we report the results of the capabilities of DAKOTA, especially optimization. DAKOTA is applied to some famous optimization examples and is verified that outputs are correct. DAKOTA is applied to the gradient-based local methods with constraints, the derivative-free global method, and multi-objective problem, to confirmative its optimal capabilities. Additionally, DAKOTA is compared with other optimal system.