Prediction of Nucleate Boiling and Burnout Heat Flux in Pressurized Water Reactor Using Machine Learning Algorithms

Abstract

The accurate prediction of both nucleate boiling and burnout heat flux is crucial for enhancing the security and stability of a water-cooled pressurized reactor. Burnout phenomenon occurs when the heat transfer rate surpasses the critical heat flux (CHF), rapidly increasing fuel rod temperature due to a substantial drop in the convective heat exchange coefficient. This event can cause severe overheating and potential damage to the reactor core. Since no deterministic theory exists for predicting both heat fluxes, the process of obtaining accurate predictions is complex and not straightforward. To overcome this complexity we developed various machine learning (ML) algorithms for predicting boiling and burnout heat flux (BHF) in a reactor core. In this paper, three ML algorithms, k-nearest neighbors (KNN), Decision Tree (DT), and Artificial neural network (ANN) are developed, trained, and compared. The data were analyzed and processed by implementing an R programming environment by using different packages. The demonstrated model performance was evaluated using k-value, accuracy, cross-validation result, and confusion matrix. By comparing  prediction efficiency and others parameter of three developed algorithms we finalize that ANN algorithm shows better performance than the DT and KNN algorithm regarding classification heat flux prediction problem. These findings are encouraging for the potential future implementation of machine learning techniques for heat flux along with reactor core diagnostics.

Journal of Engineering Science 16(1), 2025, 11-20