École Nationale Supérieure de
Techniques Avancées

Résumé

Critical heat flux (CHF) is a critical parameter in thermal–hydraulic phenomena. Over the decades, an accurate CHF prediction has been pursued to optimize nuclear reactors’ design and safety. The 2006 Groeneveld Lookup table (LUT) is a credit way to predict CHF based on a large amount of database. Now, the critical heat flux (CHF) model based on the SVR machine learning model is established, along with the model based on the Xgboost model, as well as a back-propagation (BP) neural network model. The hyper parameters are tuned to get the best performance model. Compared with the experimental tests from the KTH Lab, the SVR model achieved the highest 0.95 R2 score, while xgboost and neural network get 0.9 R2 score and 0.49 R2 score, respectively. The comparisons show that the SVR model can achieve better accuracy and generalization than the xgboost, BP NN (neural network) model, and LUT table.