基于LSTM回归算法的缪子定位模拟研究  

Simulation study of muon localization based on the LSTM regression algorithm

作　　者：张海峰 罗思远[1,2] 刘相满 何列 肖万成 尹隆乡 王克 邹雨辰 刘雨晨 王晓冬 ZHANG Haifeng;LUO Siyuan;LIU Xiangman;HE Lie;XIAO Wancheng;YIN Longxiang;WANG Ke;ZOU Yuchen;LIU Yuchen;WANG Xiaodong(School of Nuclear Science and Technology,University of South China,Hengyang 421001,China;Key Laboratory of Advanced Nuclear Energy Technology Design and Safety,Ministry of Education,Hengyang 421001,China;Institute of Modern Physics,Chinese Academy of Sciences,Lanzhou 730000,China;School of Nuclear Science and Technology,University of Chinese Academy of Sciences,Beijing 100049,China;School of Nuclear Science and Technology,Lanzhou University,Lanzhou 730000,China)

机构地区：[1]南华大学核科学技术学院,衡阳421001 [2]先进核能技术设计与安全教育部重点实验室,衡阳421001 [3]中国科学院近代物理研究所,兰州730000 [4]中国科学院大学核科学与技术学院,北京100049 [5]兰州大学核科学与技术学院,兰州730000

出　　处：《核技术》2025年第2期44-56,共13页Nuclear Techniques

基　　金：国家自然科学基金(No.12275120);中国科学技术部基金(No.2020YFE0202001);湖南省自然科学基金(No.2021JJ20006);湖湘青年英才(No.2022RC1202)资助。

摘　　要：利用宇宙线缪子对物体成像需要确定缪子的径迹,而对缪子的击中点进行精确定位是缪子径迹重建的关键。当前主流的缪子径迹探测系统需要搭配多路电子学通道才能对缪子的击中点进行精确定位,此类探测系统的构造复杂且成本高昂。为实现简便、低成本且高精度的缪子径迹探测系统设计,本研究基于Geant4软件,对无切割式的方形和圆形塑料闪烁体耦合硅光电倍增器(Silicon Photonmultipliers,SiPMs)的探测器进行模拟研究,使用SiPM收集的光子数和触发SiPM响应的时间作为特征参数,采用人工智能回归算法作为缪子定位的方法。模拟结果表明:以光子数作为特征参数的回归算法中,长短时间记忆(Long Short Term Memory,LSTM)算法在三种回归算法中的精度最高;在LSTM算法下,探测器上表面耦合12个SiPM的位置分辨率可达到厘米级别;当使用光子数和触发时间作为特征参数时,在探测器侧边仅耦合6个SiPM的位置分辨率同样能达到厘米级别,且与大面积塑料闪烁体四角耦合光电倍增管(Photomultiplier Tube,PMT)的探测器在实验中对缪子定位得到的结果吻合。本研究使用LSTM回归算法作为缪子定位算法,提出的在塑料闪烁体侧边耦合6个SiPM的探测器系统结构简便、制造成本低且定位精度达到厘米级别。[Background]Facilitating object imaging through the utilization of cosmic-ray muons mandates the precise delineation of muon trajectories,where the pinpoint localization of muon impact points assumes paramount importance for effective muon track reconstruction.Existing muon track detection systems necessitate the integration of multifaceted electronic channels to attain meticulous positioning of muon impact points.The construction of such detection systems is distinguished by its intricacy and entails substantial associated costs.[Purpose]This study aims to achieve a design for a muon track detection system that is characterized by simplicity,low cost,and high precision.[Methods]The Geant4 software was applied to the simulation of detectors comprising square and circular plastic scintillators coupled with silicon photon multipliers(SiPMs)without segmentation.The SiPMs was used to collect the number of photons and the time triggering SiPM responsed as characteristic parameters in the simulation,and a uncut square and circular plastic scintillator detector with an area of 200 mm×200 mm was constructed,with a thickness of 10 mm.The surface was coated with a TiO2 reflective coating with a thickness of 0.11 mm and a reflectivity of 95%.Then,three types of artificial intelligence regression algorithms,i.e.,extreme gradient boosting(XGBoost),multilayer perceptron(MLP)and long short-term memory(LSTM),were employed as the method for muon localization.[Results]The simulation results demonstrate that LSTM algorithm achieves the highest accuracy among the three regression algorithms when photon number is considered as the characteristic parameter.Specifically,under the LSTM algorithm,the position resolution of a configuration comprising 12 SiPMs coupled to the upper surface of the detector can attain a resolution at the centimeter level.Furthermore,by employing photon number and trigger time as characteristic parameters,the position resolution of a setup involving only 6 SiPMs coupled to the side of the detector also reache

关 键 词：缪子探测器 长短时间记忆算法 硅光电倍增器 缪子定位 

分 类 号：TL99[核科学技术—核技术及应用]