基于KPCA-MPA-ELM的n/γ甄别方法研究  

Neutron/gamma(n/γ)discrimination method based on KPCA-MPA-ELM

作　　者：胡万平 张贵宇 张云龙 庹先国 李户林 HU Wanping;ZHANG Guiyu;ZHANG Yunlong;TUO Xianguo;LI Hulin(School of Automation&Information Engineering,Sichuan University of Science&Engineering,Yibin 644000,China;Artificial Intelligence Key Laboratory of Sichuan Province,Sichuan University of Science&Engineering,Yibin 644000,China;School of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,China;Chengdu Huilite Automation Technology Co.,Ltd.,Chengdu 610000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,宜宾644000 [2]四川轻化工大学人工智能四川省重点实验室,宜宾644000 [3]西南科技大学信息工程学院,绵阳621010 [4]成都惠利特自动化科技有限公司,成都610000

出　　处：《核技术》2024年第4期73-82,共10页Nuclear Techniques

基　　金：国家自然科学基金(No.42004151,No.42074218);四川轻化工大学研究生创新基金(No.Y2022117)资助。

摘　　要：中子/伽马(n/γ)甄别在γ射线辐射存在的中子探测中至关重要。为了解决传统n/γ脉冲形状甄别方法存在的甄别精度不稳定的问题,提出了一种结合核主成分分析(Kernel Principal Component Analysis,KPCA)、海洋捕食者算法(Marine Predator Algorithm,MPA)和极限学习机(Extreme Learning Machine,ELM)的机器学习鉴别方法,以提升n/γ甄别效率。KPCA用于对中子和γ射线的脉冲信号特征进行降维;考虑到ELM输入层权重和隐藏层偏置的随机性,将MPA用于优化ELM的输入层权重和隐藏层偏置,提高ELM的n/γ甄别效率。通过对未降维和KPCA降维的数据进行训练和测试,实验结果表明:在测试集中KPCA-MPA-ELM的平均甄别准确率高达99.07%,分别高出ELM、MPA-ELM、KPCA-ELM模型12.19%、2.52%、1.56%;相较于电荷比较法和脉冲梯度分析法,精度也分别提高了1.80%和5.91%。该模型结构简单,稳定性好,能够处理高维数据,具有较好的甄别效果和泛化能力。[Background]Neutrons/Gamma(n/γ)discrimination is critical for neutron detection in the presence ofγradiation and traditional pulse shape discrimination methods suffer from unstable discrimination accuracy.[Purpose]This study aims to implement a machine-learning method that combines the kernel principal component analysis(KPCA),marine predator algorithm(MPA),and extreme learning machine(ELM)is proposed to improve the n/γdiscrimination efficiency and accuracy against the traditional pulse shape discrimination methods.[Methods]The KPCA was used to reduce the dimensionality of the pulse signal characteristics of neutrons and gamma rays.Owing to the randomness in the ELM input layer weight and hidden layer bias,the MPA was employed to optimize the foregoing factors to improve the n/γdiscrimination accuracy of the ELM.Finally,experimental data of Pu-C neutron source using BC-501A liquid scintillator detector were applied to effectiveness comparison of training and test with and without KPCA dimensionality reduction.[Results]Comparison results reveal that the average discrimination accuracy of the KPCA-MPA-ELM is as high as 99.07%,which is 12.19%,2.52%,and 1.56%higher than those of the ELM,MPA-ELM,and KPCA-ELM models,respectively.Compared with the charge comparison method and pulse gradient analysis method,the accuracy is improved by 1.80%and 5.91%,respectively.[Conclusions]The proposed model has a simple structure,exhibits good stability,hence be applied to handling high-dimensional data with good discrimination and generalization ability.

关 键 词：n/γ甄别 机器学习 核主成分分析 海洋捕食者算法 极限学习机 

分 类 号：TL812[核科学技术—核技术及应用]