Research On Kinetics Simulation And Heterogeneous Parallelization With Reactor Monte Carlo Code RMC

With the rapid development of computer hardware and improvements of nuclearreactor core designs, neutron transport Monte Carlo methods have played anincreasingly important role in reactor analysis, due to their flexibility of core geometryand energy spectrum treatments. However, recent years, the application scope of MonteCarlo methods has been mainly confined to reactor steady state analysis, the advantageof Monte Carlo methods can not be fully utilized in reactor transient analysis.Meanwhile, Monte Carlo methods have a poor computational efficiency, which makesMonte Carlo methods not suitable for routine reactor physics analysis. Because of theabove reasons, based on the reactor Monte Carlo code RMC, this thesis focuses on theresearch of Monte Carlo transient simulation methods and Monte Carlo accelerationtechniques by GPUs.There are two types of methods for Monte Carlo transient simulation, oneis the direct simulation method, the other is the quasi-static method. The MonteCarlo direct simulation method is based on the first principle, thus it is accurate,but this method is time-consuming, not suitable for practical transient problems.This thesis performs an in-depth research on the Monte Carlo quasi-staticmethods. First, the Monte Carlo improved quasi-static method is studied, and itscapability to analysis the continuous-energy transient problems is discussed, theresult is that this method is not suitable for continuous-energy problems. Thisthesis mainly focuses on the predictor-corrector quasi-static method, andproposes a continuous-energy full Monte Carlo predictor-corrector quasi-staticmethod. The difficult points of this new method, including estimation of pointkinetics parameters, solution of the fixed-source equation and generation of thefixed-source, are overcome. Based on the new method, the transient simulationmodule of RMC is developed and verified by a large amount of multi-groupbenchmarks and the continuous-energy PWR assembly transient problems. Thecomputation time of RMC transient simulation module is made controllable, andcompared to the direct simulation method, the acceptable duration of thetransient procedure can be extended. In order to improve the computational efficiency of Monte Carlo codes,this thesis resorts to the latest research achievements in the field of computerhardware, performing research on Monte Carlo acceleration techniques by GPUs.First, as to the continuous-energy Monte Carlo method, the local accelerationmethod is proposed, according to the SIMT parallel mode of GPUs. Thegeometry module of RMC is selected and accelerated by GPUs, with theproposed event-driven acceleration technique and the concept of neutron vectors.After local acceleration, the simulation efficiency of RMC is improved. Inaddition, the GPU global acceleration method for the whole multi-grouptransport module of RMC is studied. The research is focused on the load balancealgorithm for parallelism. Based on the fine-grained parallelism feature of GPUs,the neutron transport step based load balance algorithm is proposed and appliedfor GPU acceleration of RMC multi-group transport module. Also RMC isaccelerated.Based on the reactor Monte Carlo code RMC, considering the deficienciesof Monte Carlo methods in application scope and computing performance, thethesis has deeply studied on the Monte Carlo transient simulation methods aswell as the GPU acceleration techniques. Not only the application scope of RMCis expanded, but also the computing efficiency is improved. Meanwhile, theGPU acceleration study in this thesis offers important lessons for future researchon hardware heterogeneous acceleration of Monte Carlo codes.