Eventos Anais de eventos
ENCIT 2018
Brazilian Congress of Thermal Sciences and Engineering
ESTIMATION OF THERMAL PROPERTIES USING THE SOBOL SEQUENCE AND MERSENNE TWISTER WITH THE TOPOGRAPHICAL GLOBAL OPTIMIZATION
Submission Author:
Lucas Jardim , RJ , Brazil
Co-Authors:
Lucas Jardim, Diego Knupp, Antônio Silva Neto, Wagner Sacco
Presenter: Lucas Jardim
doi://10.26678/ABCM.ENCIT2018.CIT18-0600
Abstract
When one wishes to estimate parameters of a model from available experimental data, a traditional approach is the procedure of maximum likelihood, which results in an objective function to be minimized. A robust global optimization method must be employed to prevent stagnation in local minima. In this work, the technique known as Topographical Global Optimization (TGO) will be used as an optimization strategy. Fundamentally, TGO distributes random points in a search space and, through the topographic information of the objective function, selects considered topographical minimum points. These minima are then used as the initial solution for a local search method. The objective of the present work is to present investigations of how a random point generator can influence the final result of the method, to do so, the Wilcoxon Signed-Rank Test is performed. The Sobol Sequence was tested, a quasi-random sequence with low discrepancy, the pseudo-random Mersenne Twister, a generator that tends more to real randomness and the built-in routine RandomReal of Mathematica to represent a software package solution. The results obtained reinforce the efficiency of the TGO and, in addition, show the potential of the Sobol Sequence as initial point sampler.
Keywords
Topographical Global Optimization, Sobol Sequence, Mersenne Twister, Inverse problem, thermal properties
