基于PCA-GA-SVM的n/γ甄别方法研究  

Study on n/γDiscrimination Method Based on PCA-GA-SVM

作　　者：宋海声[1] 马通达 麻林召 吕柏阳 刘鹏浩[2] 马佳宁 秦秀波[2] SONG Haisheng;MA Tongda;MA Linzhao;LYU Boyang;LIU Penghao;MA Jianing;QIN Xiubo(College of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730000,China;Beijing Engineering Research Center of Radiographic Techniques and Equipment,Institute of High Energy Physics,Chinese Academy of Sciences,Beijing 100049,China;School of Nuclear Science and Technology,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]西北师范大学物理与电子工程学院,甘肃兰州730000 [2]中国科学院高能物理研究所北京市射线成像技术与装备工程技术研究中心,北京100049 [3]中国科学院大学核科学与技术学院,北京100049

出　　处：《原子能科学技术》2021年第6期1114-1123,共10页Atomic Energy Science and Technology

基　　金：国家重大科学仪器设备开发专项资助项目(2013YQ03062902);中国科学院战略性先导科技专项(A类)资助项目(XDA14020502);国家自然科学基金资助项目(11705224,11805213)。

摘　　要：由于常用中子探测器对中子和γ射线均呈现敏感性,所以消除γ射线对中子测量的影响很有必要性。考虑到支持向量机(SVM)能实现二分类器功能,本文结合主成分分析法(PCA)、遗传算法(GA),将SVM应用在混合场n/γ的甄别工作中。通过PCA对特征值进行降维,避免SVM出现过拟合现象,同时通过GA迭代方式寻找SVM关键参数惩罚因子C和核函数参数g的最优值。对PCA-GA-SVM网络在n/γ甄别中的准确性进行验证后与电荷比较法及频域梯度分析法甄别结果进行对比。结果表明,经过PCA与GA优化后的SVM网络甄别精度提升显著,该方法可为混合场n/γ提供有效的甄别。The commonly used scintillator detectors are sensitive to neutron andγray,and it is necessary to eliminate the influence ofγray on neutron measurement.The support vector machine(SVM)can realize the function of two classifiers.In this paper,the SVM combining principal component analysis(PCA)and genetic algorithm(GA)was applied to particle discrimination in n/γmixed field.PCA was used to reduce the dimensionality of eigenvalue to avoid SVM over-fitting.At the same time,GA found the optimal value of SVM key parameter penalty factor C and kernel function parameter g through iteration.The accuracy of PCA-GA-SVM in the discrimination of n/γpulse waveform was verified and compared with the discrimination result of charge comparison method and frequency gradient analysis method.The results show that the discrimination accuracy of the SVM network optimized by PCA and GA is improved significantly.This method can provide effective discrimination for n/γmixed field.

关 键 词：n/γ甄别 支持向量机 遗传算法 主成分分析法 电荷比较法 频域梯度分析法 

分 类 号：TL812.1[核科学技术—核技术及应用]