近年来，新一代反应堆以及大型科学装置飞速发展，对粒子输运模拟要求也更加精细。蒙特卡罗方法能够精确描述几何、复杂源项与物理过程，因此逐渐成为辐射屏蔽计算的重要工具。但是蒙卡方法收敛速度较慢，计算较为耗时，这一缺点成为其在实际工程应用中的关键障碍。针对这一问题，本文从软件（减方差技巧）和硬件（GPU计算）两个方面进行加速。屏蔽计算问题根据统计目标的不同，可以分为源-探测器问题、区域问题和全局问题。因为统计目标不同，所以适用不同的减方差技巧。自动重要抽样方法（Auto-Importance Sampling Method，AIS）是清华大学提出的针对深穿透问题的减方差技巧。AIS方法在区域问题中取得了比较好的加速效果，但对源-探测器问题和全局问题有一定的局限。本文针对这两种问题对AIS方法做了改进。针对源-探测器问题，将AIS方法和指向概率方法相结合，提出了小探测器自动重要抽样方法（Small Detector Auto-Importance Sampling Method）。针对全局问题，提出了网格化AIS方法，在虚拟面上划分细网格，实现均匀的全空间粒子密度分布。使用测试例题对以上两种方法进行了验证。验证结果表明，相比于传统AIS方法，计算效率有约1个数量级的提升；相比于直接模拟，计算效率有约2个数量级的提升。基于伴随计算生成源偏倚和权窗参数的耦合减方差技巧在源-探测器问题中有着较广泛的应用，但是传统伴随蒙卡方法在深穿透问题中收敛速度较慢，有一定的局限性。针对这一问题，本文提出了基于AIS伴随蒙卡的耦合减方差方法，将AIS方法引入伴随蒙卡计算，提高了伴随蒙卡方法在深穿透问题中的收敛速度，可以生成更高质量的偏倚源项和权窗参数。同时相比于基于确定论伴随的耦合减方差方法，因为仅使用一套蒙卡程序，避免了确定论伴随计算带来的耦合问题。GPU是新型计算机硬件计算平台，而点通量是一种在源-探测器问题中应用广泛的蒙卡减方差技巧，但有较为耗时的缺点。本文针对点通量计算耗时的问题，提出了点通量CPU-GPU耦合加速方法，对点通量计算进行GPU局部加速。使用NUREG/CR-6115 PWR基准题进行了验证，结果表明该方法计算效率比单核CPU计算高约50倍。

In recent years, the new generation of reactors and large-scale scientific devices have developed rapidly, and the requirements for particle transport simulation are also more elaborate. The Monte Carlo (MC) method can accurately describe geometric, complex source terms and physical processes, so it has gradually become an important tool for radiation shielding calculation. However, the MC method has a slower convergence rate and is more time consuming to calculate. This shortcoming has become a key obstacle to its application in practical engineering. Aiming at the slow convergence of the MC method, this paper studies the two aspects of the variance reduction technique and using computer hardware to accelerate the calculation.Shielding calculation problems can be divided into source-detector problems, regional problems, and global problems according to different objectives. Because of the different objectives, the applicable variance reduction technique are also different. The Auto-Importance Sampling Method (AIS) is a variance reduction technique proposed by Tsinghua University for deep penetration problems. In this paper, it is implemented in the self-developed shielding calculation program MCShield (Monte Carlo Radiation Shielding), and the geometric modeling of the virtual surface and parallel method are improved. The AIS method has achieved good acceleration effects in regional problems, but has limitations on source-detector problems and global problems. Aiming at the source-detector problem, this paper combines the AIS method with the pointing probability method, and proposes the small detector AIS method. Aiming at the global problem, a grid-AIS method is proposed to make the particle density in the whole space as uniform as possible. The above two methods were validated using the benchmark questions. The results show that the computational efficiency is about one order of magnitude improvement compared to the traditional AIS method. Compared with the anglog simulation, the computational efficiency is improved by about two orders of magnitude.In the source-detector problem, the source biasing and weight window method based on the adjoint transport has a wide range of applications. However, the deterministic adjoint code needs to be coupled with the MC code, and there are problems such as geometric matching. However, the traditional adjoint MC calculation converges slowly in the deep penetration problem. In this paper, the local variance reduction technique method based on AIS is proposed. The AIS method is introduced into the adjoint MC calculation, which improves the convergence speed of the adjoint MC calculation in the deep penetration problem, and can generate higher quality biased source parameters and weight window values. At the same time, because only a set of MC programs is used, the coupling problem caused by deterministic code is avoided.GPU is a new computer hardware computing platform. Point flux is a widely used variance reduction technique for source-detector problems, but it has the disadvantage of being time consuming. In this paper, the problem of time-consuming calculation of point flux in MC calculation is proposed. The CPU-GPU coupling acceleration method is proposed to accelerate the point flux calculation locally. A particle cluster GPU calling method is proposed, and the parallel method of MPI parallel at CPU and CUDA parallel at GPU is studied. The NUREG/CR-6115 PWR benchmark was used to verify the results. The results show that the computational efficiency of this method is about 50 times higher than that of the single-core CPU.