Author:

Keywords:

nuclear reactor, fast spectrum system, accelerator-driven systems (ADS), MYRRHA, loading pattern optimization problem, in-core fuel management, neutronics analysis, optimization, nuclear engineering, software, Genetic Algorithm (GA), Ant Colony Optimization (ACO), Metaheuristics, core reshuffling problem, constrained optimization

Abstract:

This dissertation develops a core management tool called RELOAD-M capable of optimizing reactor-core fuel loadings for MYRRHA, the future fast-spectrum research facility currently under development at SCK-CEN, Belgium. Such a tool is needed for designing highly efficient loading patterns that reflect various performance objectives of the multipurpose machine. RELOAD-M can solve the single-cycle loading pattern optimization problem, using different metaheuristic optimization methods and reactor analysis codes.Two iterative population-based metaheuristics are implemented to solve the loading pattern optimization problem: Genetic Algorithm (GA) (with or without elitism) and Ant Colony Optimization (ACO). Both methods are applied to a simple core-reload problem with a known global optimum and the optimization results are compared. It is found that the elitist GA gives the most consistent results and performs best.MYRRHA reactor-core models are described and used for the neutronics evaluation of different loading patterns by reactor analysis codes tailored to fast-spectrum systems. A simple thermal-hydraulics module is implemented for the calculation of the maximum fuel-cladding temperature. All employed models give results that are sufficiently accurate and fast enough for optimization purposes.A MYRRHA loading pattern optimization problem is solved that aims at maximizing the facility’s irradiation performance expressed in terms of the fast-neutron fluence achieved in reactor experimental channels. Three types of constraints are included in the problem: limited number of available fuel assemblies, maximum allowed fuel-cladding temperature, and end-of-cycle criticality condition. It is concluded that both the GA and ACO algorithms provide feasible solutions that outperform intuitively designed loading patterns. However, the resulting improvement is only marginal.