Multiobjective in-core fuel management optimisation for nuclear research reactors

Schlunz, Evert Barend (2016-12)

Thesis (PhD)--Stellenbosch University, 2016.

Thesis

ENGLISH SUMMARY : The efficiency and effectiveness of fuel usage in a typical nuclear reactor is influenced by the specific arrangement of available fuel assemblies in the reactor core positions. This arrangement of assemblies is referred to as a fuel reload configuration and usually has to be determined anew for each operational cycle of a reactor. Very often, multiple objectives are pursued simultaneously when designing a reload configuration, especially in the context of nuclear research reactors. In the multiobjective in-core fuel management optimization (MICFMO) problem, the aim is to identify a Pareto optimal set of compromise or trade-off reload configurations. Such a set may then be presented to a decision maker (i.e. a nuclear reactor operator) for consideration so as to select a preferred configuration. In the first part of this dissertation, a secularization-based methodology for MICFMO is pro- posed in order to address several shortcomings associated with the popular weighting method often employed in the literature for solving the MICFMO problem. The proposed methodology has been implemented in a reactor simulation code, called the OSCAR-4 system. In order to demonstrate its practical applicability, the methodology is applied to solve several MICFMO problem instances in the context of two research reactors. In the second part of the dissertation, an extensive investigation is conducted into the suitability of several multiobjective optimization algorithms for solving the constrained MICFMO problem. The computation time required to perform the investigation is reduced through the usage of several artificial neural networks constructed in the dissertation for objective and constraint function evaluations. Eight multiobjective metaheuristics are compared in the context of a test suite of several MICFMO problem instances, based on the SAFARI-1 research reactor in South Africa. The investigation reveals that the NSGA-II, the P-ACO algorithm and the MOOCEM are generally the best-performing metaheuristics across the problem instances in the test suite, while the MOVNS algorithm also performs well in the context of bi-objective problem instances. As part of this investigation, a multiplicative penalty function (MPF) constraint handling technique is also proposed and compared to an existing constraint handling technique, called constrained-domination. The comparison reveals that the MPF technique is a competitive alternative to constrained-domination. In an attempt to raise the level of generality at which MICFMO may be performed and potentially improve the quality of optimization results, a multiobjective hyperheuristic, called the AMALGAM method, is also considered in this dissertation. This hyperheuristic incorporates multiple metaheuristic sub-algorithms simultaneously for optimization. Testing reveals that the AMALGAM method yields superior results in the majority of problem instances in the test suite, thus achieving the dual goal of raising the level of generality and of yielding improved optimization results. The method has also been implemented in the OSCAR-4 system and is applied to solve several MICFMO case study problem instances, based on two research reactors, in order to demonstrate its practical applicability. Finally, in the third part of this dissertation, a conceptual framework is proposed for an optimization-based personal decision support system, dedicated to MICFM. This framework may serve as the basis for developing a computerized tool to aid nuclear reactor operators in designing suitable reload configurations.

AFRIKAANSE OPSOMMING : Die doeltreffendheid en doelmatigheid van brandstofverbruik in 'n tipiese kernreaktor word deur die spesieke rangskikking van beskikbare brandstofelemente in die laaiposisies van die reaktor beinvloed. Hierdie rangskikking staan bekend as 'n brandstof herlaaikongurasie en word gewoonlik opnuut bepaal vir elke operasionele siklus van 'n reaktor. Die gelyktydige optimering van veelvuldige doele word dikwels tydens die ontwerp van 'n herlaaikongurasie nagestreef, veral binne die konteks van navorsingsreaktore. Die doelwit van meerdoelige binne-kern brandstofbeheeroptimering (MBKBBO) is om 'n Pareto optimale versameling van herlaaikongurasieafruilings te identiseer. So 'n versameling mag dan vir oorweging (deur byvoorbeeld 'n kernreaktoroperateur) voorgele word sodat 'n voorkeurkongurasie gekies kan word. In die eerste gedeelte van hierdie proefskrif word 'n skalariseringsgebaseerde metodologie vir MBKBBO voorgestel om verskeie tekortkominge in die gewilde gewigverswaringsmetode aan te spreek. Laasgenoemde metode word gereeld in die literatuur gebruik om die MBKBBO probleem op te los. Die voorgestelde metodologie is in 'n reaktorsimulasiestelsel, bekend as die OSCAR-4 stelsel, geimplementeer. Om die praktiese toepasbaarheid daarvan te demonstreer, word die metodologie gebruik om 'n aantal MBKBBO probleemgevalle binne die konteks van twee navorsingsreaktore op te los. In die tweede gedeelte van die proefskrif word 'n uitgebreide ondersoek ingestel om die geskiktheid van verskeie meerdoelige optimeringsalgoritmes vir die oplos van die beperkte MBKBBO probleem te bepaal. Die berekeningstyd wat vir die ondersoek benodig word, word verminder deur die gebruik van kunsmatige neurale netwerke, wat in die proefskrif gekonstrueer word, om doelfunksies en beperkings te evalueer. Agt meerdoelige metaheuristieke word binne die konteks van verskeie MBKBBO toetsprobleemgevalle vergelyk wat op die SAFARI-1 navorsingsreaktor in Suid-Afrika gebaseer is. Toetse dui daarop dat die NSGA-II, die P-ACO algoritme en die MOOCEM oor die algemeen die beste oor al die toetsprobleemgevalle presteer. Die MOVNS algoritme presteer ook goed in die konteks van tweedoelige probleemgevalle. 'n Vermenigvuldigende boetefunksie (VBF) beperkinghanteringstegniek word ook voorgestel en vergelyk met 'n bestaande tegniek bekend as beperkte dominasie. Daar word bevind dat the VBF tegniek 'n mededingende alternatief tot beperkte dominasie is. 'n Poging word aangewend om die vlak van algemeenheid waarmee MBKBBO uitgevoer word, te verhoog, asook om potensieel die kwaliteit van die optimeringsresultate te verbeter. 'n Meerdoelige hiperheuristiek, bekend as die AMALGAM metode, word in die nastreef van hierdie twee doelwitte oorweeg. Die metode funksioneer deur middel van die gelyktydige insluiting van 'n aantal metaheuristieke deel-algoritmes. Toetse dui daarop dat the AMALGAM metode beter resultate vir die meerderheid van toetsprobleme lewer, en dus word die bogenoemde twee doelwitte bereik. Die metode is ook in the OSCAR-4 stelsel ge mplementeer en word gebruik om 'n aantal MBKBBO gevallestudie probleemgevalle (binne die konteks van twee navorsingsreaktore) op te los. Sodoende word die praktiese toepasbaarheid van die metode gedemonstreer. In die derde deel van die proefskrif word 'n konseptuele raamwerk laastens vir 'n optimeringsgebaseerde persoonlike besluitsteunstelsel gemik op MBKBB, voorgestel. Hierdie raamwerk mag as grondslag dien vir die ontwikkeling van 'n gerekenariseerde hulpmiddel vir kernreaktoroperateurs om aanvaarbare herlaaikongurasies te ontwerp.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/100070
This item appears in the following collections: