基于优化极限学习机模型的反应堆中子通量与k_(eff)预测方法研究  

Prediction method of reactor neutron flux and k_(eff) based on the optimized extreme learning machine model

作　　者：陈镜宇 刘喜洋 赵鹏程[1] 刘紫静[1] 李卫 CHEN Jingyu;LIU Xiyang;ZHAO Pengcheng;LIU Zijing;LI Wei(School of Nuclear Science and Technology,University of South China,Hengyang 421001,China)

机构地区：[1]南华大学核科学技术学院,衡阳421001

出　　处：《核技术》2024年第10期178-187,共10页Nuclear Techniques

基　　金：装备预研教育部联合基金(No.8091B032243)资助。

摘　　要：通过模拟和扩展人类智能,人工智能能够解决预测反应堆k_(eff)和中子通量等问题。本研究选用国际原子能机构(International Atomic Energy Agency,IAEA)反应堆作为研究对象,以稳态时的堆芯中子通量和k_(eff)为预测量,通过堆芯物理分析程序ADPRES生成数据样本后,利用极限学习机(Extreme Learning Machine,ELM)构建中子通量和k_(eff)的基础神经网络模型,随后通过随机森林(Random Forest,RF)评估特征值重要程度以建立各模型最佳输入特征子集,采用遍历方法确定隐藏层最佳神经元数目,最后使用鲸鱼优化算法(Whale Optimization Algorithm,WOA)对其初始权值与阈值进行优化,进一步提高了模型的精度。研究结果显示:经降维优化处理后,神经网络的预测能力有较大提升,其中k_(eff)的预测精度提高了两个量级,中子通量的预测误差降低了87.24%,并且减少了模型训练时间。本文构建方法对快速评估堆芯物理特性有一定参考意义。[Background]By simulating and augmenting human intelligence,artificial intelligence can address challenges such as predicting k_(eff) and neutron flux of a reactor.[Purpose]This study aims to apply the optimized extreme learning machine model to the prediction of reactor neutron flux and k_(eff).[Methods]First of all,a three-dimensional IAEA reactor was selected as the research object,with the steady-state neutron flux and k_(eff) as the predictive variables.and the core physics analysis program ADPRES was employed to generate data samples.Then,the basic neural network models for neutron flux and k_(eff) were constructed using Extreme Learning Machine(ELM),and the importance of feature values was evaluated using Random Forest(RF)to establish the optimal input feature subset for each model.Subsequently,the optimal number of neurons in the hidden layer was determined using a traversal method.Finally,the Whale Optimization Algorithm(WOA)was used to optimize the initial weights and thresholds for further improvement of the model accuracy.[Results]The evaluation results show that after dimensionality reduction and optimization processing,the predictive accuracy of k_(eff) has improved by two orders of magnitude,and the prediction error of neutron flux has decreased by 87.24%,and the model training time is also reduced.[Conclusions]The model method constructed of this study has certain reference significance for solving reactor k_(eff) and neutron flux.

关 键 词：极限学习机 鲸鱼优化算法 中子通量 k_(eff) 参数预测方法 随机森林 

分 类 号：TL433[核科学技术—核技术及应用]