Applicability study of Bayesian optimization to neutronic design of a homogeneous two-region core

Kuwagaki, Kazuki  ; Yokoyama, Kenji   

At the Japan Atomic Energy Agency (JAEA), a design support tool for advanced nuclear reactors is currently under development. This tool is called ARKADIA-Design, and is expected to support the integrated design evaluation of reactors from the viewpoints of safety, economy, and sustainability as a carbon-free energy source by utilizing the newest analysis/evaluation technologies such as AI technology, and the accumulated knowledge of fast reactor development. One development task of the ARKADIA-Design is to build a system that automatically identifies optimized design parameters by which an objective function specified by core performance is minimized (or maximized). In the present study, we set up a single objective optimization example problem with multiple constraints for a homogeneous two-region core, and showed that the optimal solution of this example problem can be automatically obtained by the Bayesian optimization method, which is a candidate optimization algorithm for the system. In addition, we also demonstrated how the system would assist the core design procedure in future, by indirectly solving a three-variable optimization problem of the core design. From these results and demonstrations, we confirmed that the system to be developed has the potential to be a useful support tool for the designers, enabling them to obtain optimal core designs efficiently.