Initial Study on Machine Learning Predictive Models for Lattice-Level Neutronic Analysis in High-Temperature Gas-Cooled Reactors

High-temperature gas-cooled reactors (HTGRs) are being developed not only for electricity generation but also to produce hydrogen, leveraging the high outlet coolant temperature up to 950°C from the reactor core. To achieve this condition safely, the power density distribution within the core should be optimized through effective in-core fuel management. Depending on the reactor’s intended application and core design, the neutronic parameters should be calculated for each core configuration. In this present study, high-fidelity simulations based on the Monte Carlo (MC) method were performed to predict the key neutronic parameter of multiplication factors kinf. Although it produces highly accurate results, the MC method is time-consuming and computationally intensive, particularly for large-scale parametric studies. This initial study proposes using surrogate models developed through machine learning techniques to predict important neutronic parameters more efficiently. The current stage focuses on developing a predictive model at the lattice level of HTGRs. The widely used ensemble method Random Forest regression was adopted. A dataset consisting of approximately 640 samples was prepared and used for model training and validation. The model’s performance was evaluated using the coefficient of determination R2 and the root-mean-square error, yielding values of 0.998 and 0.011, respectively. Furthermore, when applied to predict the kinf values, the model demonstrated good agreement with MC simulation, with prediction errors ranging from 0.9% to 5.5%, depending on the uranium enrichment and the fuel packing fraction.