A novel reduced-order model and two-stage optimization strategy for high-efficiency predicting thermal-hydraulic behavior in a 6×6 fuel bundle

Abstract

Achieving fast and low-cost computation of the thermal-hydraulic flow field inside a reactor is of great significance for understanding the reactor's thermal-hydraulic characteristics and ensuring its safety. Based on the Fluent scripting technique, a total of 216 cases with varying inlet velocities, inlet temperatures and power densities were simulated, forming the dataset for model training. A reduced-order model (ROM), referred to as proper orthogonal decomposition (POD)-surrogate model (SM)-ROM, was constructed by combining the POD method with SM. The model takes inlet velocity, inlet temperature, and power density as inputs and outputs the corresponding POD mode coefficients. The proposed POD-SM-ROM can enable rapid prediction of flow fields and temperature fields under non-sample points operating conditions by reconstructing the flow fields and temperature fields from the predicted POD mode coefficients and POD mode. In addition, a novel two-stage optimization strategy for the POD-SM-ROM was proposed to further reduce computational time while preserving accuracy. The ROM and two-stage optimization framework developed in this study enables efficient and rapid prediction of thermal-hydraulic behavior in nuclear reactors. This advancement not only enhances computational efficiency but also supports timely and reliable reactor safety assessments. All the codes in this paper can be found at our GitHub link: (https://github.com/guangyunmin/A-novel-data-driven-reduced-order-model).