Machine Learning-Based Methodology for Assessment of Doppler Reactivity of Sodium-Cooled Fast Reactor

被引:0
作者
Petrovic, Dorde [1 ]
Mikityuk, Konstantin [1 ]
机构
[1] Paul Scherrer Inst, Lab Sci Comp & Modelling, Grp Adv Nucl Syst, Forsch Str 111 OHSA-D12, CH-5232 Villigen, Switzerland
来源
JOURNAL OF NUCLEAR ENGINEERING AND RADIATION SCIENCE | 2021年 / 7卷 / 04期
关键词
sodium-cooled fast reactor; Doppler reactivity; Monte Carlo; machine learning; artificial neural network;
D O I
10.1115/1.4050216
中图分类号
TL [原子能技术]; O571 [原子核物理学];
学科分类号
0827 ; 082701 ;
摘要
In order to close nuclear fuel cycle and address the problem of sustainability, advanced nuclear reactor systems of the fourth generation are in the focus of the research for many years. With a simple goal of supporting this research, machine learning-based methodology for the assessment of the Doppler reactivity has been developed and applied to the European Sodium Fast Reactor (ESFR) in the frame of the ESFR-Safety Measures Assessment and Research Tools (SMART) Horizon-2020 project. In the scope of this study, a database of the precise Monte Carlo (MC) calculations was prepared and used to train artificial neural network (ANN) as a surrogate model to assess the Doppler reactivity across the range of reactor conditions that could occur throughout the life of the reactor core, in fast, yet accurate manner. The database was generated for all the combinations of several core parameters carefully predefined in order to account for both operational and accidental states of the core. Subsequently, Doppler reactivity change as a function of the above-mentioned parameters was assessed by herein developed methodology, as well as by widely used logarithmic dependence of the Doppler reactivity on the fuel temperature and compared to the results of the precise MC simulations. This study proves that, if certain computational resources are allocated to the database generation and ANN training, newly developed methodology yields similar or even more accurate results compared to the classical methodology and at the same time provides a tool for parameterization and interpolation of Doppler reactivity not only on the fuel temperature but also on the other parameters characterizing core of the sodium-cooled fast reactor (SFR).
引用
收藏
页数:6
相关论文
共 4 条
[1]  
Krepel J., 2017, P INT C FAST REACT R
[2]   The Serpent Monte Carlo code: Status, development and applications in 2013 [J].
Leppanen, Jaakko ;
Pusa, Maria ;
Viitanen, Tuomas ;
Valtavirta, Ville ;
Kaltiaisenaho, Toni .
ANNALS OF NUCLEAR ENERGY, 2015, 82 :142-150
[3]  
Rineiski A., 2018, HORIZON 2020 ESFR SM, DOI [10.5281/zenodo.1990703, DOI 10.5281/ZENODO.1990703]
[4]  
The MathWorks, 2019, MATLAB DEEP LEARNING