A STACKING ENSEMBLE MODEL TO PREDICT CRITICAL HEAT FLUX (CHF)

Abstract

The critical heat flux (CHF) corresponding to the nucleate boiling crisis (DNB) is critical for the design, operation, and safety of nuclear facilities. The development of an accurate and robust CHF prediction model remains the primary goal of the thermal-hydraulic engineering community.

The goal of this study is to use a stacked ensemble learning model to achieve superior capabilities for CHF prediction. Our approach, in particular, is based on a stacking ensemble learning scheme, in which the predictions produced by nineteen base learning and linear regression are used as meta-learner (super learner) in the top-level method to produce final predictions.

We tested the proposed scheme on an experimental datasets that reported CHF in a variety of ranges and geometries. The results show that a CHF prediction approach based on ensemble learning combines predictions produced by weaker learning methods and then feeds them to a meta-learner to achieve superior results. More importantly, this case study demonstrated that using an improved stacking ensemble model can result in very accurate predictions with low errors, making it a suitable approach for addressing the CHF prediction problem.