粒子流事例重建中簇射区分算法研究

粒子流算法是当前LHC实验以及未来正负电子对撞机实验中重要的数据分析工具。与传统事例重建算法相比，粒子流算法可明显改善喷注能量分辨率。但是，现有粒子流重建算法基本都仅利用了量能器击中的位置信息，通过邻近聚类方法进行簇射重建，对空间上重叠的粒子簇射区分能力不够。本研究提出一种综合考虑簇射击中的位置、能量、密度和时间等信息区分重叠簇射的方法。此方法利用簇射的多维变量通过流形学习算法进行粒子重建，并且使用关于簇射结构特征的先验知识作为对重建质量的反馈，以达到高准确率区分重叠簇射的目标。通过此方法可进一步改善现有粒子流算法在高能区的喷注能量分辨率，提高物理实验测量值的精度。同时，粒子流算法喷注能量分辨率性能的提高也有利于优化探测器硬件设计。

The Particle Flow Algorithm (PFA) is a crucial data analysis tool for the experiments on the current LHC and the future electron-positron collider. Comparing with the traditional reconstruction algorithms, the PFA can significantly improve the jet energy resolution. However, the existing PFAs only use the position information of hit in calorimeter to reconstruct the shower by the nearest neighbor clustering method, so the separation power of the spatially overlapped showers is still not competent enough. A method is proposed to take the position, energy, density and time information of shower hits into account for the separation of the overlapped showers. This method utilizes the multi-dimensional variables of shower to reconstruct particles by the manifold learning algorithm, and the prior knowledge about the properties of shower structure is also used as the feedback of reconstruction quality. The target of separation for the overlapped showers with a high accuracy can be achieved via this new method. The jet energy resolution from the current PFA at high energy scale can be further improved, and the precision of observables at physics experiment upgrades accordingly. Meanwhile, the performance enhancement of PFA on jet energy resolution is also beneficial for the optimization of the detector hardware design.