h1

TY  - THES
AU  - Xu, Yu
TI  - Machine learning application in low energy liquid scintillator neutrino experiment
PB  - RWTH Aachen University
VL  - Dissertation
CY  - Aachen
M1  - RWTH-2020-12064
SP  - 1 Online-Ressource (ii, xv, 153 Seiten) : Illustrationen, Diagramme
PY  - 2020
N1  - Veröffentlicht auf dem Publikationsserver der RWTH Aachen University 2021
N1  - Dissertation, RWTH Aachen University, 2020
AB  - Neutrinos are the keys for physics beyond the Standard Model, since neutrino oscillationis the directly observed new physics phenomena. We use PMNS matrix to describe this phenomena. In the matrix, there are six free parameters: three mixing angles θ<sub>12</sub>, θ<sub>23</sub>, θ<sub>13</sub>; two difference of the squared neutrino masses ∆m<sup>2</sup><sub>21</sub>, ∆m<sup>2</sup><sub>32</sub>; and CP violation angle δ<sub>CP</sub>. Nowadays, there are one and half unknown parameters: sign of ∆m<sup>2</sup><sub>32</sub> and δ<sub>CP</sub>.The Jiangmen Underground Neutrino Observatory (JUNO) Experiment was designed to measure the sign of ∆m<sup>2</sup><sub>32</sub>, and we aim to achieve 3 σ confidence level with 6 years data. To achieve this goal, we construct a 20 kt liquid scintillator (LS) detector to measure the ―ν<sub>e</sub> energy spectrum from Yangjiang and Taishan Nuclear Power Plant(NPP). Machine Learning (ML) is becoming more and more popular in data analysis. It can simplify the process of reconstruction with faster speed and better performance. In this thesis I will discuss the application of ML application in JUNO experiment, including waveform reconstruction, particle identification, and vertex/energy reconstruction. Waveform reconstruction is the base for other reconstructions. In this thesis, I show the potential of ML on waveform reconstruction: time reconstruction of each single hit, which is impossible with traditional method. Particle identification (PID) can reduce background/signal (B/S) ratio and improve the sensitivity of the experiment. In this thesis, we study the potential of particle identification (PID) with machine learning method. We can receive 95% signal with 5% background for alpha/beta discrimination, and 50% signal with 5% background for electron/positron discrimination. We also perform a research on vertex and energy reconstruction with machine learning method, and find that we can meet the requirement of 3% energy resolution.
LB  - PUB:(DE-HGF)11
DO  - DOI:10.18154/RWTH-2020-12064
UR  - https://publications.rwth-aachen.de/record/808442
ER  -