【作者】 王静；

【导师】 于龙昆； 夏长锋；

【作者基本信息】 南昌大学 ， 电子信息硕士（专业学位）， 2025， 硕士

【摘要】 高能宇宙辐射探测设施(High Energy cosmic-Radiation Detection,HERD)是用于空间天文和粒子天体物理研究的实验设施,计划于2028年发射并部署在中国空间站上。硅电荷探测器(Silicon Charge Detector,SCD)作为HERD的最外层探测器,旨在精确测量宇宙线中从氢(Z=1)到镍(Z=28)的粒子电荷量,其电荷测量能力直接影响HERD的粒子鉴别能力及整体科学产出。优化电荷重建算法有助于提高SCD的电荷测量能力。本研究将多种机器学习算法系统应用于HERD SCD电荷重建领域,打破传统基于统计分布标定的固有模式,为宇宙线探测领域的电荷重建研究开辟全新路径。SCD原理样机于2022年在欧洲核子研究中心(CERN)进行了离子束测试。本文基于所采集的数据,开发使用支持向量机(Support Vector Machines,SVM)和多层感知机(Multi-Layer Perceptron,MLP)的电荷重建算法。针对重核数据统计量不足的问题,引入迁移学习,将低电荷区训练得到的知识迁移至高电荷区,从而有效缓解数据稀缺问题。实验结果表明,与传统方法相比,机器学习方法利用多通道信息,使得从钠(Z=11)到镍(Z=28)的重核电荷分辨平均提升约14%。HERD SCD首次研制了全尺寸准电性件样机,并参与了2024年的束流实验。本文首先对硅微条探测器模块进行了基线噪声测试和宇宙线测试,以评估其基本性能。随后,将机器学习电荷重建算法应用于SCD准电性件样机,以提取其电荷测量能力。电荷重建结果显示,全尺寸样机的电荷测量范围为Z=3～28,且Z=6时SCD的平均电荷分辨约为0.074 c.u.,符合SCD的性能指标要求。更多还原

【Abstract】 The High Energy cosmic-Radiation Detection(HERD)facility is an experimental setup designed for space astronomy and particle astrophysics research,scheduled for launch and deployment on the China Space Station in 2028.The Silicon Charge Detector(SCD),as the outermost detector of HERD,aims to precisely measure the charge of cosmic ray particles ranging from hydrogen(Z=1)to nickel(Z=28).Its charge measurement capability directly affects HERD’s particle identification ability and overall scientific output.Optimizing the charge reconstruction algorithm helps improve the charge measuring ability of the SCD.This study systematically applies multiple machine learning algorithms to the field of charge reconstruction for HERD SCD.It breaks the traditional fixed pattern based on statistical distribution calibration and opens up a new path for the research on charge reconstruction in the field of cosmic-ray detection.A prototype of the SCD was tested with an ion beam at CERN in 2022.Based on the collected data,this paper develops charge reconstruction algorithms using Support Vector Machines(SVM)and Multi-Layer Perceptron(MLP).To address the issue of insufficient heavy nucleus data,transfer learning is introduced to transfer knowledge learned from the low-charge region to the high-charge region,effectively mitigating data scarcity.Experimental results show that,compared to traditional methods,machine learning approaches utilizing multi-channel information improve charge resolution by approximately 14%on average for heavy nuclei from sodium(Z=11)to nickel(Z=28).The HERD SCD developed a full-scale quasi-electric prototype for the first time and participated in the beam test in 2024.This paper first evaluates the fundamental performance of the silicon microstrip detector modules through baseline noise and cosmic-ray tests.Subsequently,the machine learning based charge reconstruction algorithm was applied to the SCD quasi-electrical prototype to extract its charge measuring capabilities.The charge reconstruction results indicate that the full-scale prototype achieves a charge measurement range of Z=3 to 28,and the average charge resolution of the SCD at Z=6 is approximately 0.074 c.u.,meeting the performance index requirements of SCD.更多还原