Efficient surrogate-based optimization using reduced order models by moment matching coupled with kriging - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2018

Efficient surrogate-based optimization using reduced order models by moment matching coupled with kriging

Résumé

Structured Abstract. Purpose-Performing dynamic simulation and optimization of electromagnetic systems can be very time consuming and even prohibitively slow when used in optimization processes. This paper contains a new methodology that couples Model Order Reduction (MOR) with a Response Surface Methodology (RSM) to greatly reduce the computational time. The methodology is validated by optimizing the capacitors placement on a laminated bus bar. Design/methodology/approach-The proposed methodology uses Moment Matching (MM) as a Model Order Reduction technique to perform very fast computation of objective functions. These results are used as input by a Kriging interpolations algorithm to create an adaptive Response Surface (RS). Optimization is performed based on the Expected Improvement (EI). To guarantee that the objective functions computed by the Reduced Order Models are accurate, an adaptive reduction schema is developed. Findings-The methodology has been able to greatly reduce the optimization time. An adaptive Model Order Reduction schema has been successful in producing accurate models for different parameters demanded by the optimization process. Research limitations/implications-Not all systems are eligible to be used by the proposed methodology. They are assumed to be writable in the form of a generalized state space. Therefore, only linear systems are eligible. Originality/value-Two different techniques for accelerating computational time in optimization have been combined. A method to perform adaptive Model Order Reduction has been used to guarantee accuracy, allowing the connection of the methods.
Fichier principal
Vignette du fichier
hal-01881909, version 1 Efficient Surrogate.pdf (771.75 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01881909 , version 1 (21-04-2020)

Identifiants

  • HAL Id : hal-01881909 , version 1

Citer

Mateus A. O. Leite, Benoît Delinchant, Jean-Michel Guichon, João A. Vasconcelos. Efficient surrogate-based optimization using reduced order models by moment matching coupled with kriging. 15th International Workshop on Optimization and Inverse Problems in Electromagnetism, Sep 2018, Hall in Tirol, Austria. ⟨hal-01881909⟩
49 Consultations
21 Téléchargements

Partager

Gmail Facebook X LinkedIn More