International Symposium on Symbiotic Nuclear Power (ISSNP2024)

The Critical Heat Flux (CHF) is a pivotal thermodynamic parameter in the design and safety assessment of water-cooled reactors. Accurate CHF prediction is essential for reliable reactor operation, as it enables the anticipation of thermal limits that, if exceeded, could lead to system instability and potential failure. Research into predicting CHF using machine learning methods has emerged as a promising avenue to enhance prediction accuracy and efficiency. However, conventional machine learning models suffer from challenges related to interpretability and extrapolation due to their data-driven and opaque nature. This study proposes a Physics-informed machine learning-aided framework to predict CHF. By integrating diverse physical methods for CHF prediction and leveraging machine learning to address biases in training data, the study evaluates this hybrid model against traditional models and standalone physics-based methods. The results demonstrate that the hybrid machine learning models offer superior stability in CHF predictions, effectively managing anomalies present in physical methods. Moreover, the models exhibit robust extrapolation capabilities, with errors maintained within an acceptable threshold of 20%. The findings of this study contribute to the advancement of CHF prediction methodologies, providing a more reliable and efficient tool.
 

Critical heat flux,Machine learning,Physics-informed machine learning-aided,Back propagation neural network,Support vector machine