Session: 07-10: Simulations and Predictions - II

Paper Number: 134988

134988 - Prediction of Thermal-Hydraulic Parameters for Autonomous Load Following Lfr Core Using Surrogate Model 

Abstract: 

The concept of autonomous load following refers to the reactor's ability to adjust core power through inherent reactivity feedback, without the need for control rod operation. One promising candidate for this concept is the lead-based fast reactor, known for its high inherent safety characteristic. Ensuring the safety of the reactor core is a primary concern during the load following process. The quasi-static reactivity balance (QSRB) method can be used to quickly calculate the various steady states of the system under load following scenarios. Using the reactivity feedback coefficients, the QSRB method can calculate the relationship between power, mass flow rate and temperature. By applying these boundary conditions, the maximum temperatures of the fuel and cladding can be determined, allowing for an evaluation of core safety from a thermal-hydraulic analysis perspective. However, during the load following process, there are multiple system states, and it would be time-consuming and computationally intensive to use the CFD method for each state. Given the fixed geometry and similar boundary conditions, surrogate modeling is a perfect and suitable approach for this application. In this paper, a parametric batch study is conducted using Computational Fluid Dynamics (CFD) to analyze fuel assembly with different inlet temperature, mass flow rate, and power. The parameter matrix is generated using the Latin hypercube sampling method. The maximum temperature of the fuel and cladding are extracted as the output to generate the training dataset. To verify the accuracy of the CFD results, comparisons are made with Nu empirical relations and other TH code. Various surrogate modeling methods are employed for training the model, including the Kriging method, the Gaussian Process method, the Radial basis function method, the Response surface method, and the neural network method. The impact of different kernel function types is also considered. The K-folder cross-validation method is used due to the relatively small size of the dataset. The root mean square error (RMSE) and maximum error of surrogate models are evaluated and compared. Among them, the fourth-order response surface model demonstrates the best accuracy. Its RMSE is 1.58 K for maximum fuel temperature and 0.46 K for maximum cladding temperature, while the maximum error is 7.21 K for maximum fuel temperature and 1.88 K for maximum cladding temperature. As a result, the response surface method is chosen as the surrogate modeling method. It is then applied in the quasi-static autonomous load following analysis scenario to calculate the maximum temperature of fuel and cladding with near accuracy, but in significantly less time compared to CFD analysis.

Presenting Author: Kefan Zhang University of Science and technology of China

Presenting Author Biography: Doctorial student of School of Nulcear Science and Technology, Unviersity of Science and Technology of China

Authors:

Kefan Zhang University of Science and technology of China
Wenshun Duan University of sicence and technology of Chian
Junjia Zhang University of Science and technology of China
Hongli Chen University of Science and Technology of China