A Test Bed for Method Development and Validation for Reactor Fuel Loading Optimization

One of the key challenges in in-core nuclear fuel management for a typical light water reactor is the vast number of unique core configurations that make up the design space. While various methodologies have been developed and used to enhance fuel management optimization, their performance evaluation largely remains in relative terms, primarily in comparison with existing optimization methods. For the same reason, the high-dimensional design space prohibits the evolution of the fuel loading strategy from being visualized, thus obscuring the optimization dynamics from reactor core designers.This study addresses these limitations by creating a test bed that features an exhaustive core design database and associated design assistance tools. The database contains the entirety of an integrated pressurized water reactor (IPWR) design space with approximately 1.6 million loading patterns, encompassing all possible first-cycle IPWR core configurations using six unique assembly types. This allows for a comprehensive parametric study to be conducted using the genetic algorithm (GA), a optimization approach widely utilized for solving such problems.The optimal core design, determined by a specific set of hyperparameters, was compared against the known best solution to identify the most effective hyperparameters for future optimization problems. Furthermore, two dimensionality reduction techniques, namely t-distributed stochastic neighbor embedding and uniform manifold approximation projection, were employed to generate low-dimensional embeddings and visualizations of the design space. The entire search process driven by GA was then projected onto a two-dimensional space to visualize the evolution of traits and reveal the optimization trajectories.The results suggest that these design assistance tools offer novel capabilities that not only improve the efficiency of the optimization solvers, but also enhance the explainability of the otherwise opaque optimization process. This holds potential for developing more advanced methods to address complex nuclear engineering design problems.