Historical-Boundary-Condition-Based POD-MLP Surrogate Modeling for Transient Thermal-Hydraulic Prediction in Rod-Bundle Channels

Abstract

Fast and accurate prediction of transient thermalhydraulic parameters is essential for reliability assessment and safety-margin evaluation in safety-critical nuclear reactor components. However, high-fidelity computational-fluiddynamics simulations are computationally prohibitive for the many-query scenarios required by uncertainty quantification, risk analysis, and system safety assessment. To alleviate this computational burden, this study applies a proper orthogonal decomposition (POD) and multilayer perceptron (MLP) reducedorder surrogate to reconstruct target-time three-dimensional transient temperature parameters from recent historical boundary conditions. The emphasis is placed on historicalboundary encoding: five feature representations are systematically compared, including raw nodal input, slope-plusinitial-value representation, initial-value-plus-segment-integral representation, all-nodes-plus-slopes representation, and allnodes-plus-segment-integrals representation. The optimized model achieves an average coefficient of determination R2 of 0.999952 and a root-mean-square error (RMSE) of 15.13. Compared with a steady-state-equivalent strategy that uses only the terminal boundary condition, the proposed transient framework substantially reduces prediction error under strongly varying transient conditions, with steady-state peak relative errors exceeding 1.8 percent on the vertical section and transient errors remaining negligibly small. Among the tested encodings, the initial-value-plus-segment-integral representation yields the best overall performance with the lowest global RMSE of 14.983 and the highest global R2 of 0.999955. The proposed surrogate reduces the online computational cost of transient parameter prediction by orders of magnitude while preserving high-fidelity accuracy, thereby enabling feasible reliability-oriented many-query analyses for safety-critical thermal-hydraulic systems.