项目名称: 实际复杂系统不确定量化中的降阶建模理论
项目编号: No.91330104
项目类型: 重大研究计划
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 明炬
作者单位: 北京计算科学研究中心
项目金额: 70万元
中文摘要: 在复杂系统的研究中,当完整的实验数据缺失的时候,利用数学模型进行数值模拟则是唯一的分析和研究工具。这时候由于信息的不完整和模型带来的误差所造成的不确定性进行量化分析则成为模型成功与否的核心挑战。这其中的主要困难之一在于在进行数值模拟的时候,由于刻画不确定性的随机空间的高维数所带来的计算上的巨大困难。本项目将利用随机降阶建模,针对各种模型,在保留原有模型主要特性的同时,降低计算所需的自由度,从而达到降低计算复杂度的目的。事实上,随机降阶模型的有效性与信息的过滤直接相关,即需要找到保留的信息和能够容忍的误差之间的平衡,本项目希望通过利用不确定量化问题所关注的统计信息对误差的容忍度相对较大的特点,研究利用POD/CVT,PCE和压缩感知等理论,减少计算所需的空间的自由度和用于刻画不确定性的随机空间的维数,对各种不确定量化问题建立高效而准确的低维逼近方法。
中文关键词: 不确定性量化;降阶建模;多重蒙特卡洛方法;随机控制;随机微分方程
英文摘要: Simulation is often the only analysis and research tool for verifying the accuracy of model outputs for complex systems when experimentation is neither feasible nor possible. A successful simulation of complex system requires characterizing and quantifying the uncertainties and errors caused by the incomplete knowledge of the inputs and the mathematical models. Due to the high-dimensional space involved, the computation for predicting the outputs based on uncertain inputs or deriving the inputs from the statistics of outputs is often a formidable work. For the purpose of lessening the complexity of such computation, this project will establish different stochastic reduced-order models for various types of systems. More specifically, considering the effectiveness of the stochastic reduced-order modeling is tightly connected with how to “filtrate” the information of systems, i.e., preserve the essential and discard the useless information, and the fact that the error tolerance for the concerned statistics of the system inputs or outputs is relative large, we will apply the methods including polynmial chaos expansion, centroidal Voronoi tessellation, proper orthogonal decomposition and compressed sensing, etc, to reduce the degrees of freedom of the spacial space or the dimensionality of the stochastic space that
英文关键词: uncertainty quantification;reduced-order modeling;multilevel Monte-Carlo Method;stochastic control;stochastic partial differential equations