Font Size: a A A

Modified loss functions and artificial neural networks in nonlinear multi-response optimization problems

Posted on:2009-03-04Degree:M.EType:Thesis
University:University of Puerto Rico, Mayaguez (Puerto Rico)Candidate:Torres-Pizarro, IsmaelFull Text:PDF
GTID:2448390005956829Subject:Engineering
Abstract/Summary:
This work presents four case studies that are used to compare the suitability of several techniques to solve multiple response optimization problems. The approaches compared include linear regression using Lagrange optimization, linear regression using a "brute force" search optimization approach, and a neural network with the same "brute force" optimization method. Our approach uses the loss function technique described by Taguchi (1986) and modified by Artiles-Leon (1996) for multiple responses optimization. In general, the regression approach with the Lagrange optimization provided the best results with expected loss values up to 1.17 times the actual minimum loss, but the linear regression using a "brute force" optimization proved comparable with up to 1.31 times the actual minimum loss. Results using neural network were neither acceptable nor expected with expected loss values up to 2.4 times the actual minimum loss.
Keywords/Search Tags:Loss, Optimization, Neural, Linear regression using
Related items