项目名称: 基于部分线性模型的随机偏微分方程辨识方法研究
项目编号: No.11301544
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 宁瀚文
作者单位: 中南财经政法大学
项目金额: 22万元
中文摘要: 随机偏微分方程的系统辨识是利用随机分布参数系统的观测数据去重构描述这个系统的未知的随机偏微分方程,它可以看作是对随机偏微分方程的"反向"的研究。现有的辨识方法是在非线性自回归模型架构下将集中参数系统中的辨识建模方法平行的推广过来,因而无法解决随机偏微分方程系统辨识研究中一些特有的问题。我们在分析中发现:基于定性理论与数值理论,部分线性模型与随机偏微分方程有着紧密而有趣的联系。因此,本项目提出"随机偏微分方程系统辨识的部分线性模型分析法"来研究此领域中当前的一些热点问题:利用有限元方法和数理统计理论给出部分线性模型的逼近误差并建立判定模型复杂度的性能指标;提出"逆有限元方法"研究"数据采样点空间分布不均匀"情况下的辨识问题;利用极限分布理论,研究核学习方法下系统状态估计量的统计学性质;以状态估计量的置信区间为基础,提出适合的"信息更新"策略,进而发展有效的随机偏微分方程在线辨识方法。
中文关键词: 偏微分方程;部分线性模型;核学习方法;系统辨识;鲁棒最优控制
英文摘要: The objective of system identification of stochastic partial differential equation is to reconstruct the unknown stochastic differential equations based on the observation data of corresponding stochastic distributed parameter system.It can be considered as a kind of "inverse" research for stochastic partial differential equations.In the previous work, the identification problems are conventionally considered in the framework of nonlinear auto-regressive network with exogenous model, and the identification techniques for stochastic ordinary differential equations are directly employed,some particular problems for identification of stochastic partial differential equations could not be solved in this framework. We introduced the numerical theory of stochastic partial differential equations into the corresponding research, found that the unknown stochastic partial differential equations and partially linear model are theoretically correlated. Thus, this project proposes "The Partially Linear Model Analysis for System Identification of Stochasitc Differential Equations" to solve the problems of this research area that has been widely concerned: To obtain the approximation error of the partially linear model to the original stochastic partial differential equations and the corresponding complexity index; To investig
英文关键词: Partial differential system;Partially linear model;kernel learning method;System identification;Robust optimal control