Optimising the few-group energy structure for the full-core simulation of lead-cooled fast reators

The objective this thesis is the application of genetic algorithm techniques to optimize the few-group energy grid for the multiphysics simulation of Lead-cooled Fast Reactors (LFR). The identification of a proper energy group structure to collapse the group constants to be used in full-core simulations is a crucial step: such energy structure has to be knowledgeably chosen as it may strongly affect the accuracy and efficiency of the multigroup calculation results. The energy grid structure to be used must be selected with the intention of preserving the physics of the problem, which is at the basis of accurate results, and to minimize as much as possible the computational effort involved in the simulation. This is true in general, but the problem complexity increases as we move towards fast spectrum fission reactions. In this framework, optimization techniques play a key role, as they are applied to a wide range of engineering fields, and the nuclear reactor physics is one of them. Starting from a set of feasible solutions, optimization techniques aim at identifying the nearly-optimal solutions by using proper objective functions, which are usually maximized or minimized, according to their definition. Evolutionary algorithms, and in particular Generic Algorithms (GAs), have revealed themselves to be powerful tools in solving high-dimensional, non-linear optimization problems. Inspired by the working principle of Darwinian selection and evolution, in recent years, GAs have been already tested in specific field of nuclear engineering: for example, they have been applied to optimize the energy grid structure for Molten Salt Reactors (MSR), providing encouraging results. Based on this outcome, it is interesting to evaluate and explore the performances and the possibilities related to the use of such techniques for the identification of the optimal energy group boundaries, focusing the attention on its application to the new Advanced Lead Fast Reactor European Demonstrator (ALFRED) benchmark, proposed by the Nuclear Energy Agency (NEA). To perform this study, the use of the multigroup diffusion module of the Fast REactor NEutronics/Thermal-hydrauliCs (FRENETIC) multiphysics code is adopted. Developed during the last decade at Politecnico di Torino, the code aims at simulating coupled neutronic/thermal-hydraulic transients in liquid-metal-cooled fast reactors with hexagonal subassemblies, providing an approximated yet still reliable description of the physical behaviour of such technologies. Specifically, the code is used to evaluate the accuracy of few-group structures proposed by the GA, comparing them with a reference solution, provided by the Serpent2 Monte Carlo neutron transport code (which supplied quantities such as the effective multiplication factor, the radial and axial power distributions, etc.). The code is also exploited to retrieve the fine-grid nuclear data that need to be fed to neutronic module of the FRENETIC code as input parameters (e.g. macroscopic reaction cross sections, diffusion coefficient, etc.). To perform the comparison among the solutions, a set of physics-driven fitness functions (FFs) have been defined. Such FFs are related to the effective multiplication factor, to the power density, the neutron flux and the kinetic parameters, and turned out to be a suitable combination. The physical justification of the optimal solutions evaluated by the GA is confirmed through the evaluation of some significant figures of merit.