Συμβολή στην επιτάχυνση σύγκλισης των εξελικτικών αλγορίθμων μέσω μεταπροτύπων και ανάλυσης σε κύριες συνιστώσες. Εφαρμογές στην αεροδυναμική

Κουκούλης, Ιωάννης; Koukoulis, Ioannis

dc.contributor.author	Κουκούλης, Ιωάννης	el
dc.contributor.author	Koukoulis, Ioannis	en
dc.date.accessioned	2017-09-20T06:38:28Z
dc.date.available	2017-09-20T06:38:28Z
dc.date.issued	2017-09-20
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/45649
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.14487
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc/3.0/gr/	*
dc.subject	Βελτιστοποίηση	el
dc.subject	Εξελικτικός αλγόριθμος	el
dc.subject	Ανάλυση σε κύριες συνιστώσες	el
dc.subject	Μεταπρότυπα	el
dc.subject	Αεροδυναμική	el
dc.subject	Optimization	en
dc.subject	Evolutionary algorithm	en
dc.subject	Principal component analysis	en
dc.subject	Metamodels	en
dc.subject	Aerodynamics	en
dc.title	Συμβολή στην επιτάχυνση σύγκλισης των εξελικτικών αλγορίθμων μέσω μεταπροτύπων και ανάλυσης σε κύριες συνιστώσες. Εφαρμογές στην αεροδυναμική	el
dc.title	Acceleration of evolutionary algorithms using metamodels and principal component analysis. Αpplications in aerodynamics	en
heal.type	bachelorThesis
heal.classification	Engineering optimization	en
heal.classification	Βελτιστοποίηση	el
heal.language	el
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2017-07-14
heal.abstract	Στόχος της διπλωματικής εργασίας είναι η παρουσίαση, ανάλυση και πιστοποίηση μεθόδων βελτιστοποίησης, οι οποίες βασίζονται στους Εξελικτικούς Αλγορίθμους (ΕΑ), και αξιοποιούν Ανάλυση σε Κύριες Συνιστώσες (Principal Component Analysis), για να αυξήσουν την αποτελεσματικότητά τους.	el
heal.abstract	The aim of this diploma thesis is to present and validate optimization methods based on Evolutionary Algorithms (EAs), which utilize Principal Component Analysis (PCA) in order to increase their efficiency. EA-based optimization has been adequately developed over the last years at the Parallel CFD & Optimization Unit (PCOpt) of the Laboratory of Thermal Turbomachines (LTT) of the NTUA. During the last decades, the EASY optimization software has been developed. EASY is an EA-based, flexible all-purpose optimization tool. Coupled with the appropriate problem-specific evaluation program, it can compute the optimal solution to a wide range of optimization problems. The major weakness of EA’s is the large number of candidate solutions that must be evaluated, in order for the evolution process to take place.This difficulty is accentuated in large-scale industrial optimization problems,where the evaluation of a single candidate solution is very expensive, in terms of both wall-clock time and computational cost. For this purpose, Metamodel-Assisted Evolutionary Algorithms (MAEAs) have been developed. The role of MAEAs is to construct approximate models of the objective function(s)to be optimized. These models, called metamodels or surrogate evaluation models, are computationally cheap, and allow for the detection of the most promising candidate solutions, without resorting to the expensive problem- specific evaluation program. The time and computational resources required for the optimization process can be greatly reduced in this way.The use of metamodels significantly improves the performance of EAs,however their major setback remains; therefore, new techniques are being re-searched that will allow for an EA with minimal usage of the costly evaluation software. In this diploma thesis, this is accomplished through the use of Principal Component Analysis (PCA), applied to an appropriate dataset. PCA is a statistical tool used for analyzing large sets of data. Its function is to define a new set of variables, which describe data in a statistically uncorrelated way. This can facilitate the separation between noise and signal and the detection of patterns in the dataset. Two separate PCA methods are presented and used in this diploma thesis: Linear PCA, in which the new set of variables is the result of an orthogonal transformation of the original variables, and Kernel PCA. The latter is a generalization of the former, utilizing properties of a non-linear transformation in order to detect a great variety of characteristics in the dataset, a technique known as “the kernel trick”.Principal component analysis is performed on the set of offspring pro-duced by the EA in each generation, providing information about the nature of the problem’s design space. The results of the analysis can be used to increase the EA’s performance in two different ways: by applying the evo- lutionary operators of mutation and crossover to parents expressed in terms of the new variables, thus producing candidate solutions of higher quality (a variant that is called EA(PCA) ), or by training metamodels of lesser dimen-sionality (a variant called M(PCA)AEA). This reduction in dimensionality produces metamodels of improved predictive ability that are also computa- tionally cheaper to train.The above mentioned methods can be used separately or in conjunction with each other. This leads to several variants of the conventional EAs and MAEAs, which can be named accordingly, depending on the chosen PCA method and its use: EA(L), EA(K), M(L)AEA, M(K)AEA, M(L)AEA(L),M(K)AEA(K), where the letter L or K is used for a linear or kernel method respectively.The proposed methods are implemented using the EASY software and applied to optimization problems: specifically three mathematical problems,two cases of pseudo-engineering optimization exercises, and a case of aero- dynamic optimization are studied. Each problem is solved using the variants mentioned above and their performances are compared to those of the con-ventional EA and MAEA.	en
heal.advisorName	Γιαννάκογλου, Κυριάκος	el
heal.committeeMemberName	Αρετάκης, Νικόλαος	el
heal.committeeMemberName	Γιαννάκογλου, Κυριάκος	el
heal.committeeMemberName	Μαθιουδάκης, Κωνσταντίνος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Μηχανολόγων Μηχανικών. Τομέας Ρευστών. Εργαστήριο Θερμικών Στροβιλομηχανών	el
heal.academicPublisherID	ntua
heal.numberOfPages	107 σ.
heal.fullTextAvailability	true