非定常输运基于香农熵的自动调整样本数策略  

A Shannon entropy-based strategy for adjusting history number of time-dependent transport problem automatically

作　　者：上官丹骅[1] 邓力[1] 李刚[1] 张宝印[1] 

机构地区：[1]北京应用物理与计算数学研究所,北京100094

出　　处：《强激光与粒子束》2018年第1期144-147,共4页High Power Laser and Particle Beams

基　　金：能源局重大专项(2015ZX06002008);中国工程物理研究院科学技术发展基金项目(2014B0202029)

摘　　要：对于非定常输运问题提出了一种基于香农熵的自动调整样本数策略。将每一计算步的总样本数划分为若干批并逐步模拟每批中的粒子,可以在每批粒子模拟结束后通过计算得到该时间步幸存粒子属性分布对应的香农熵值。采用在线收敛性诊断方法,一旦通过香农熵值序列判断对应的幸存粒子属性分布已经收敛,则可以提前结束本时间步的计算。对一个空间一维非定常输运模型的计算结果表明,该策略可以显著减少每一计算步的实际样本数且保持最终的结果基本不变,从而减少了计算时间,提高了计算效率。This paper proposes a Shannon entropy-based strategy for adjusting history number of time-dependent transport problem automatically.By dividing the total history number of each step into many batches and simulating all batches one by one,we calculate the Shannon entropy of the survival particle distribution after each batch.If the on-the-fly diagnostic of convergence of entropy shows the survival particle distribution has converged,the calculation of the current step will be stopped in advance and the next step will be activated immediately.Test for a one-dimensional model shows this strategy has decreased the calculation time greatly while keeping the results almost unchanged simultaneously.

关 键 词：非定常输运 香农熵 在线收敛性诊断 

分 类 号：O242.1[理学—计算数学]