Atomic Energy Science and Technology ›› 2021, Vol. 55 ›› Issue (5): 954-960.DOI: 10.7538/yzk.2020.youxian.0426

Previous Articles    

Application of Artificial Intelligence Method in Layout Optimization of 3He Pipe in AWCC

WANG Duan;LI Duohong;WEI Zihao;HOU Cheng;LIN Hui;LI Da;LIU Likun;WU Zhaohui   

  1. Nuclear Industry University, Beijing 102413, China;State Nuclear Security Technology Center, Beijing 102401, China;China Institute of Atomic Energy, Beijing 102413, China
  • Online:2021-05-20 Published:2021-05-20

人工智能方法在AWCC中3He管排布设计的应用

王端;李多宏;韦子豪;侯丞;林辉;李达;刘立坤;武朝辉   

  1. 核工业大学,北京102413;国家核安保技术中心,北京102401;中国原子能科学研究院,北京102413

Abstract: When the other parameters of active well coincidence counter (AWCC) were determined, artificial intelligence algorithm was used to find the best way to arrange the 3He tubes and improve the efficiency of neutron detection. Firstly, the whole space of the layout scheme was uniformly sampled, and Monte Carlo method was used to simulate the detection process of AWCC to calculate the neutron detection efficiency of each scheme, and generate sample data for artificial intelligence algorithm. Then the relationship between 3He tube layout scheme and neutron detection efficiency was quickly fitted by deep neural network (DNN). Finally, the optimal layout of 3He tube in AWCC was searched by genetic algorithm. The error between the neutron detection efficiency of this method and Monte Carlo calculation result is acceptable, and the optimal efficiency is higher than that of the original equipment. This method can also be used to optimize other parameters and solve mult-objective optimization problem. It opens up a new way to improve the intelligence of AWCC design.

Key words: Monte Carlo simulation, deep neural network, genetic algorithm, layout optimization

摘要: 在有源井型符合中子计数器(AWCC)其他参数确定的情况下,使用人工智能算法寻找3He管的最佳排布方式,以提高设备的中子探测效率。首先对排布方案整体空间进行均匀抽样,再使用蒙特卡罗方法模拟AWCC探测过程,计算每种排布方式下的中子探测效率,从而生成数据样本供人工智能算法学习,然后使用深度神经网络(DNN)快速拟合出3He管排布与中子探测效率的关系,最后结合遗传算法寻找AWCC中3He管的最优排布方式。该方法的中子探测效率与蒙特卡罗计算结果的误差在可接受范围内,且筛选出的最优效率高于原设备的效率。该方法还可用于其他参数的优化设计,并拓展至解决多目标优化问题,为提高AWCC设计的智能性开辟了新途径。

关键词: 蒙特卡罗模拟, 深度神经网络, 遗传算法, 排布优化