diplomska naloga
Štefan Markič (Author), Vlado Stankovski (Mentor), Janko Logar (Thesis defence commission member), Iztok Peruš (Co-mentor)

Abstract

V diplomskem delu preverjamo možnosti za uporabo namenskega programa za strojno učenje enačb Lagramge na specifičnem inženirskem problemu napovedi največjega pospeška tal, ki se zgodi ob potresu. Lagramge pri svojem delovanju uporablja formalizem kontekstno neodvisne gramatike, ki vsebuje pravila za izgradnjo enačb in omeji prostor hipotez - mogočih enačb. V študiji smo razvili tri gramatike, vsako s svojo ravnijo uporabe specifičnega znanja področja potresnega inženirstva, ki ga je sestavljalo 68 enačb iz literature. Za poskuse smo imeli na voljo bazo 3550 zapisov o močnejših potresih, ki smo jo za namene navzkrižne validacije desetkrat naključno razbili na 90-odstotno učno in 10-odstotno testno množico. Algoritem je med izvajanjem hevristino in/ali izčrpno preiskal vse tri prostore hipotez in ovrednotil enačbe s funkcijo srednjega kvadratnega odklona na učni in testni množici. V vsakem poskusu smo določili tri najboljše enačbe glede na kvantitativni kriterij in jih primerjali med seboj ter z modeli študij Next Generation Attenuation in modelom evropskih avtorjev Akkarja in Bommerja. Ugotovili smo, da vključitev specifičnega znanja, ki ni niti preohlapna niti preveč specifična, pozitivno vpliva na kvaliteto rezultatov. Poleg tega smo preučili še vpliv vrednosti vhodnih parametrov programa, ki usmerjajo potek iskanja najboljše enačbe. Rezultati kažejo, da bi Lagramge lahko uporabili tako pri podobnih problemih v potresnem inženirstvu kot tudi na drugih inženirskih področjih.

Keywords

gradbeništvo;diplomska dela;UNI-B;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL FGG - Faculty of Civil and Geodetic Engineering
Publisher: [Š. Markič]
UDC: 621.396.9:691.5:692.2(043.2)
COBISS: 6256481 Link will open in a new window
Views: 1530
Downloads: 438
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Predicting peak ground acceleration of an earthquake by using the machine learning program Lagramge
Secondary abstract: This bachelor’s thesis deals with testing the equation discovery system Lagramge when applied to a specific engineering problem of modelling the earthquake’s peak ground acceleration. The Lagramge system uses context-free grammar formalism which contains rules for building equations and limits the hypothesis space of possible equations. We developed three different grammars, each incorporating a different level of domain specific knowledge, which included 68 published equations. In the experiments a database of 3550 strong motion earthquake recordings was used and for the purpose of cross validation split 10 times into 90 % learning and 10 % testing sets. The algorithm employed exhaustive and/or heuristic search methods in all three hypothesis spaces and evaluated the equations on the learning and testing datasets using the mean squared error criterion. From each of 4 experiments three best equations were selected on the basis of quantitative criterion and compared with each other and with the equations from the Next Generation Attenuation study as well as with the equation developed by the European authors Akkar and Bommer. We found out that inclusion of the domain specific knowledge which is neither too specific nor too general improves the quality of the results. In addition to this, influences of other input parameters guiding the process of equation discovery were examined. The results of this study show that the Lagramge system could also be applied to similar problems in earthquake engineering as well as to other fields of engineering.
Secondary keywords: graduation thesis;civil engineering;
File type: application/pdf
Type (COBISS): Bachelor thesis/paper
Thesis comment: Univ. v Ljubljani, Fak. za gradbeništvo in geodezijo
Pages: XII, 28 str., 52 str. pril.
ID: 8312777