Development and application of data-driven CHF model in system analysis code

In nuclear reactor systems, when the fuel rods reach the critical heat flux (CHF), a sharp increase in fuel temperature occurs due to a drastic reduction in heat transfer capacity, thus posing a considerable risk to the reactor's safety. Consequently, accurate CHF prediction holds paramount importance in accurately simulating accident scenarios and enhancing the overall safety of reactor systems. To tackle the challenge of limited prediction accuracy in existing CHF models, this study initially established a comprehensive database by utilizing available experimental data and lookup tables. Subsequently, various methodologies, including the Back Propagation Neural Network (BPNN), Random Forest (RF), and Physics-Informed Machine Learning (PIML), were employed to develop multiple CHF prediction models, and their performance was thoroughly evaluated. Furthermore, the optimal CHF model was integrated into the self-developed analysis code ARSAC, which was then validated using the ORNL-THTF experiment. The results indicated that the BPNN-based model not only demonstrated exceptional prediction accuracy but also exhibited rapid calculation speeds. Notably, the average relative error between the experimental data points and the calculation results of the modified code is 3.64%, while for the original code, it is 22.84%. This study effectively leverages the strengths of data-driven approaches, providing a robust technical solution for high-precision, efficient, and adaptive numerical prediction and analysis of reactor accidents.