项目名称: 递推局部多项式回归估计及其应用
项目编号: No.61203118
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 自动化学科
项目作者: 陈性敏
作者单位: 大连理工大学
项目金额: 24万元
中文摘要: 非参数方法不依赖于模型的较多先验信息,且可以对付随机性、时变性和非线性,在系统辨识与控制的各种方法中起着重要的作用。局部多项式方法是非参数统计中最重要的工具之一。研究递推局部多项式方法及其在非线性系统辨识与自适应控制中的应用,有重要的意义。 目前对该方法的研究大都是基于离线算法,且其统计性质的获得需要很强的条件。相对而言递推算法有着独特的优势。我们研究递推的局部多项式回归估计,探讨算法收敛的持续激励条件,并证明收敛性。进而将之应用于非线性ARX(NARX)系统、仿射NARX和非线性自回归条件异方差模型的辨识,同时考虑量测带误差的情形,并证明算法的收敛性。最后,从控制项增益已知的仿射NARX系统着手,研究其在自适应控制中的应用。 从递推算法的角度研究局部多项式回归估计并将之应用于非线性系统辨识与控制,是以往工作中几乎没有考虑过的。由于它相比经典核回归估计的优势,有许多深刻的问题值得去研究。
中文关键词: 局部多项式回归;递推辨识;核估计;随机优化;分布式优化
英文摘要: In existing various approaches to identification and control of nonlinear systems, the nonparametric method plays an important role, because it depends not much on a priori information of the system and is capable to cope with randomness, time-varying, and nonlinearity. Local polynomial regression is one of the most important tools in nonparametric statistics. Thus, the research on recursive local polynomial regression and its applications in identification and adaptive control of nonlinear systems is of great significance. However, almost the previous works are based on nonrecursive algorithms and strong conditions are required for various statistical results. The recursive algorithms have its unique advantages compared with the nonrecursive ones. Recursive local polynomial regression estimation and it applications are considered in the project. First, the recursive formulae and the persistent excitation condition for convergence of the nonparametric estimates are to be derived, and then convergence of the estimates is to be proved. Then, its applications to identification of the nonlinear ARX (NARX) system, affine NARX system and the nonlinear autoregressive conditional heteroskedasticity model are to be investigated. The errors-in-variables cases are also to be considered. Convergence of the estimates is to
英文关键词: local polynomial regression;recursive identification;kernel estimation;stochastic optimization;distributed optimization