Accurate departure from nucleate boiling ratio (DNBR) prediction using SIMCA’s partial least squares regression and clustering

Abstract Data prediction and classification are the cornerstones of modern data analysis, especially when it comes to enhancing operational safety in critical systems. The SIMCA software offers robust, scalable tools for these tasks, leveraging the partial least squares (PLS) method for both classification and regression. This study applied regression and classification analyses to a thermal-hydraulic dataset generated using the COBRA-IV code, focusing on the calculation of the Departure from Nucleate Boiling Ratio (DNBR). The dataset was categorized into two subsets: a steady-state random dataset and a transient dataset, incorporating scenarios such as Rod Cluster Control Assembly (RCCA) operations and Loss of Flow Rate Accidents (LOFA). The steady-state dataset was developed through the variation of five key features – power, flow rate, enthalpy rise peaking factor (F ΔH ), temperature, and pressure – and computing the corresponding DNBR for each case. Hyper parameter optimization was conducted to identify the optimal number of PLS components, ensuring comprehensive information retention across datasets. For each dataset, the PLS regression metrics, including variable importance in projection (VIP), R 2 X, R 2 Y, Q 2 Y, and PLS coefficients, were analyzed to evaluate the predictive accuracy and interpretability of the model. The results demonstrated a strong similarity with COBRA-IV outputs and confirmed the findings of prior work which employed artificial neural networks (ANNs). These outcomes highlight the reliability of PLS as a predictive tool and its potential for advancing the analysis of thermal-hydraulic datasets.