基于神经网络的球床堆在线核材料衡算方法  

Online Nuclear Material Accounting Research Based on Neural Network for Pebble Bed Reactor

作　　者：刘若晴 马涛 张立国 LIU Ruoqing;MA Tao;ZHANG Liguo(Institute of Nuclear and New Energy Technology,Collaborative Innovation Center of Advanced Nuclear Energy Technology,Key Laboratory of Advanced Reactor Engineering and Safety of Ministry of Education,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学核能与新能源技术研究院,先进核能技术协同创新中心,先进反应堆工程与安全教育部重点实验室,北京100084

出　　处：《原子能科学技术》2020年第10期1885-1891,共7页Atomic Energy Science and Technology

基　　金：国家科技重大专项资助项目(2018ZX06906011);国防科工局核能开发项目资助。

摘　　要：10 MW高温气冷实验堆在线测量系统可通过γ谱数据分析测得部分核素的活度,这些核素不包含绝大部分超铀元素同位素。出于核安保的需要,为确定超铀元素同位素的活度,提出一种基于深度学习的超铀元素含量计算方法。该方法采用引入反向误差传播的深度神经网络模型,以易测核素活度为输入,输出不易测核素活度。采用反应堆核素生成和耗减程序跟踪10 MW高温气冷实验堆的运行功率历史,产生核素活度数据样本,对神经网络模型进行训练和测试。研究结果表明:深度神经网络估算核素含量具有较高的准确性,在球床式高温气冷堆辐照后燃料中超铀元素在线估算方面具有较大发展潜力。The online measurement system of 10 MW high temperature gas cooled test reactor can obtain the activity of some nuclides by analyzingγspectra excluding the vast majority of the transuranic isotopes.For the purpose of nuclear security,a deep learning with method of transuranic calculation was proposed to determine the activity of transuranic isotopes.The method employed a deep neural network model with back-propagation of error that took in the activity of easy-to-measure nuclides and output the activity of those difficult to measure.In this paper,the historical working capacity of 10 MW high temperature gas cooled test reactor was tracked with the nuclide generation and depletion program of the reactor,from which data samples of nuclide activity were produced and the neural network model was trained and tested.According to the research results,the deep neural network model was of higher accuracy in nuclide content estimating and therefore promising in the online estimation of transuranic elements in fuels after irradiation of pebble bed high temperature gas-cooled reactor.

关 键 词：超铀元素同位素 核素活度 深度神经网络 燃耗测量 

分 类 号：TL99[核科学技术—核技术及应用]