基于深度神经网络的放射性废物桶γ能谱解析方法  

Analytical method for γ energy spectrum of radioactive waste drum based on deep neural network

作　　者：王江玮 顾卫国[1] 杨桧 王德忠[1] WANG Jiangwei;GU Weiguo;YANG Hui;WANG Dezhong(Machinery and Power Engineering College,Shanghai Jiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《核技术》2022年第4期51-57,共7页Nuclear Techniques

摘　　要：在核电厂放射性废物桶测量中,为了解决传统γ能谱解析方法存在的核素误识别和峰面积计算精度较差的问题,提出了基于深度神经网络的γ能谱解析方法。深度神经网络以γ能谱全谱数据作为分析对象,无需传统方法的谱线平滑、寻峰等工作。利用蒙特卡罗模拟得到的γ谱线作为神经网络的数据集,在秦山核电一期200 L钢桶的三种不同介质内置5种γ源,实验测量获得的γ谱线用于验证。结果发现:神经网络方法能快速识别核素并计算峰面积,精度达到96.47%;对于多核素混合的复杂能谱不会产生核素误识别,对于能谱中弱峰的峰面积计算误差也控制在10%以内。整体而言,基于深度神经网络的γ能谱解析方法适用于放射性废物桶的能谱解析,且解谱精度优于传统方法。[Background] In the measurement of radioactive waste drums in nuclear power plants, the traditional analytical method of γ energy spectrum has the problems of nuclide misjudgment and poor accuracy of peak area calculation. [Purpose] This study aims to evaluate the performance of an analytical method for γ energy spectrum based on deep neural network. [Methods] The whole data of γ energy spectrum were taken by the deep neural network model as the analysis object, hence no need of the traditional methods such as spectral line smoothing and peak searching. First of all, combinations of five different γ sources placed in different positions in the 200 L steel drum from Qinshan nuclear power plant-phase I, filled three different media(air, water and sand) were experimental measured by using digital γ-ray spectrometer and high purity germanium(HPGe) detector. Then the γ energy spectra obtained by Monte Carlo simulation using the model of the same experimental measurement system were used as the data set of the neural network training. Finally, γ energy spectra obtained by experiments were compared with simulated for verification. [Results] The trained deep neural network converges quickly, both the nuclides identification and peak area calculation are fast with the accuracy of 96.47%. Hardly misidentification of nuclides is caused for the complex energy spectrum of multi-nuclide mixture, and less than 10% identification error for the peak area calculation of a weak peak in the γ energy spectrum. [Conclusions] The analytical method based on deep neural network is suitable for the analysis for γ energy spectrum of radioactive waste drums, and the spectrum resolution accuracy is better than the traditional method.

关 键 词：深度神经网络 蒙特卡罗 Γ能谱 核素识别 

分 类 号：TL81[核科学技术—核技术及应用]