Primjena regresijskog modela s potpornim vektorima u proračunu reaktorskog štita

doctoral thesis

• Title (english): Application of Support Vector Regression Model in Reactor Shielding Calculations
• Author: Krešimir Trontl
• Mentor: Dubravko Pevec (mentor)
• Mentor: Tomislav Šmuc (komentor)
• Committee member: Nikola Čavlina (predsjednik povjerenstva)
• Committee member: Dubravko Pevec (član povjerenstva)
• Committee member: Tomislav Šmuc (član povjerenstva)
• Committee member: Mile Baće (član povjerenstva)
• Committee member: Nenad Debrecin (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Applied Physics); Zagreb
• Defense date and country: 2008-11-26, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering
• Universal decimal classification (UDC): 621.3 - Electrical engineering

Abstract

Za preliminirane inženjerske proračune brzine doza gama zračenja često se koristi aproksimativna metoda Greenovih funkcija, čija je preciznost uvelike određena točnošću faktora nakupljanja upotrijebljenih u proračunu. Faktori nakupljanja za jednoslojne štitove dobro su definirani i poznati, no prisutan je problem određivanja faktora nakupljanja višeslojnih štitova, koji se najčešće i susreću u praksi. Stoga je analizirana mogućnost primjene metode potpornih vektora za izgradnju regresijskog modela s potpornim vektorima (SVR modela) za određivanje faktora nakupljanja brzine doze gama zraka za višeslojne štitove, a s ciljem povećanja točnosti njihovog određivanja čime se povećava i točnost proračuna brzina doza. Izgradnja regresijskog modela za određivanje faktora nakupljanja provedena je kao klasični računarski eksperiment u kojem je složeni računarski program upotrijebljen za generaciju parova ulaz/izlaz na kojima algoritam metode potpornih vektora „uči“. Kako bi proces izgradnje bio optimalan definirana je procedura optimalnog razvoja modela, s naglaskom na metodama aktivnog učenja. Dobiveni regresijski model ugrađen je u računarski program QAD-CGGP i testiran na dva problema proračuna brzina doza gama zračenja. Rezultati testiranja pokazuju djelomično poboljšanje kvalitete proračuna, odnosno povećanje točnosti izračunatih brzina doza u usporedbi sa klasičnim pristupom koji se standardno primjenjuje kod QAD-CGGP programa.

Abstract (english)

The point-kernel method is a widely used practical tool for preliminary engineering gamma ray shielding calculations. The accuracy of the method is highly dependent on the accuracy of the buildup factors used in the calculations. Buildup factors for single-layer shields, comprised of a single material, are well known. On the other hand, buildup factors for stratified shields, commonly encountered in practical problems, represent a complex physical problem difficult to model. Therefore, the possibility of applying a support vector regression model for the determination of the multi-layer shield buildup factors has been analysed. Development of the regression model for the determination of buildup factors has been conducted as a classical computer experiment in which a complex computer code has been employed for the generation of input/output pairs on which support vector machine algorithm “learns“. In order to perform optimal development of the model, optimal development procedure has been defined, with particular attention placed on active learning methods. Obtained regression model has been incorporated into QAD-CGGP code and tested on two gamma radiation shielding problems. Testing results suggest partial improvement of the calculational quality, i.e., higher accuracy of the calculated dose rates compared to ones obtained by classical QAD-CGGP approach.

• Keywords: faktor nakupljanja; SVR model; računarski eksperiment; optimalni razvoj; aktivno učenje
• Keywords (english): buildup factor; SVR model; computer experiment; optimal development; active learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:607342
• Project: Number: 036-0361590-1579; Title: Gospodarenje gorivom standardnih i naprednih nuklearnih reaktora; Leader: Dubravko Pevec; Jurisdiction: Croatia; Funder: MZOS; Funding stream: ZP
• Study programme: Title: Doctoral study programme "Electrical Engineering and Computing"; Study programme type: university; Study level: postgraduate; Academic / professional title: doktor/doktorica znanosti, po-dručje tehničkih znanosti (doktor/doktorica znanosti, po-dručje tehničkih znanosti)
• Catalog URL: http://lib.fer.hr/cgi-bin/koha/opac-detail.pl?biblionumber=33168
• Type of resource: Text
• Extent: 108 str.
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-03-19 11:14:50