Session: 08-05: Computational Fluid Dynamics (CFD) and Applications - V

Paper Number: 135080

135080 - An Online State Estimation Method Based on Quantum Genetic Algorithm for Space Nuclear Reactors 

Abstract: 

Space nuclear reactors provide continuous, stable, and powerful energy supply for long-term space exploration missions, their operation being independent of sunlight, which is a key technology for deep space exploration. To overcome the issue of high computational costs associated with models based on traditional numerical algorithms, methods like machine learning have been proposed to accelerate the computation process. These methods, by simulating the complex physical processes, can effectively reduce the extensive iterations and computational resources required by traditional algorithms. However, despite the great potential of machine learning methods in improving computational efficiency, their accuracy and generalizability still pose challenges. These challenges mainly stem from the biases and uncertainties that may arise in machine learning models when dealing with highly complex and nonlinear systems. Especially for space reactors, there is limited experimental data and a significant amount of uncertainty in numerical models. Therefore, it is necessary to develop an online state estimation algorithm to obtain the true operating status of space reactors.

This paper introduces an online state estimation method for space nuclear reactors, designed to mitigate the limitations in accuracy and generalizability associated with machine learning-enhanced models. During the operation phase of the reactor, the machine learning accelerated model remains parallel to the real reactor in time. By using sensors and other measuring devices, the reactor transmits measurement data to the model within a certain time interval. After receiving measurement parameters, a quantum genetic algorithm is used for parameter optimization, adjusting key coefficients in the model to align computational results with actual measurements, thereby more accurately simulating the real operating state. The quantum genetic algorithm takes the L2-norm minimum value between the calculated and measured results as the optimization object, and uses a small population and multi-iteration method for training. Key coefficient determination involves a two-step process: First, a preliminary array of key coefficients is selected through sensitivity analysis of program parameters; Secondly, recursive methods are used to analyze the sensitivity of these key coefficients to model modifications and determine the final array of key coefficients. The study focuses on the improved TOPAZ-II model with heat pipes, employing a traditional numerical algorithm-based model as a benchmark. It then corrects this machine learning-enhanced models. The results show that the proposed method can effectively correct the results of the machine learning-enhanced model, and the corrected model can more accurately reflect the operating state of the reactor. This study has a wide range of applications, not only applicable to the state estimation of space reactors, but also to the state estimation of other reactors.

Presenting Author: Enping Zhu shanghai jiaotong university

Presenting Author Biography: Zhu Enping is a doctoral student at Shanghai Jiao Tong University in China, specializing in the design and technical development of digital twin systems for space reactors. He has studied flow distribution in lead based fast reactor cores and natural cycle stability system analysis. In Nuclear Engineering and Design, Annals of Nuclear Energy, and Computer Methods in Applied Mechanics and Engineering, having published five papers as the first author.

Authors:

Enping Zhu shanghai jiaotong university
Xiang Chai shanghai jiaotong university