科学网基金客服电话
010-62580792
联系我们

国家自然科学基金项目查询

基于核矩阵的柔性系数回归模型及其在风速时序间歇性建模中的应用研究

批准号61673155 学科分类过程控制系统 ( F030210 )
项目负责人甘敏 负责人职称副教授 依托单位福州大学
资助金额61.00
万元
项目类别面上项目 研究期限2017 年 01 月 01 日 至
2020 年 12 月 31 日
中文主题词系统辨识;非线性建模;回归模型;间歇性
英文主题词system identification;nonlinear modeling;regressive model;intermittent

摘要

中文摘要 时间序列建模是一个学科交叉研究方向,在自动控制、经济等众多领域有重要应用,对其的研究有着普适性的现实意义。为使实践者可以针对数据的特性选择或设计某些核函数,来更好地刻画数据的动态性,本项目拟在前期研究基础上提出一种基于核矩阵的柔性系数回归模型。项目将围绕模型结构机理、辨识方法、实际应用等方面展开研究:①通过分析现存一些非线性模型的结构特点,提出一类以非线性核、延迟变量、线性参数三者乘积为元素的柔性系数回归模型;②研究不同核函数与模型性质的关系,如选取何种核函数可描述“非对称”现象,何种核函数可刻画”间歇性”;③研究模型参数估计的正则化问题,通过分析可分离的非线性最小二乘问题目标函数性质,提出迭代与非迭代的正则化优化算法,并比较其性能;④考虑模型中随机噪声为非高斯的情形,并基于贝叶斯方法和蒙特卡罗粒子滤波算法辨识模型; ⑤针对风速时间序列的间歇性,基于提出的方法进行建模与预测分析。
英文摘要 Time series modeling is an important interdisciplinary field of research, which has many important applications in automatic control, economics, etc. The research on it has universal theoretical and practical significance. This project intends to propose a class of kernel matrix based flexible coefficient regressive model, and study its structure, parameter optimization, extensions and applications: ①By analyzing the structures of some existing nonlinear models, we will propose a kernel matrix based flexible coefficient regressive model which consists of the product of nonlinear kernel, delay variables and linear parameters, and consider its probabilistic properties and universal approximation capability. ②We will study the properties of models with different kernel matrix. For example, which kernel functions can describe the “asymmetry” and which kernel functions can produce the “intermittent”? ③ We will investigate the regularization problem of the parameter estimation algorithm, and proposed iterative and non-iterative algorithms by analyzing the properties of the objective functions. Performance comparisons of the algorithms will also be studied. ④ We will consider the situation when the model noise is non-Gaussian, and give the representation of the model by self-organizing state-space model. The identification algorithm based on Bayesian method and Monte Carlo particle filtering will also be proposed. ⑤ To cope with the intermittent, the proposed methodologies will be applied to wind speed time series modeling.
结题摘要

成果

序号 标题 类型 作者

  • Copyright @ 2007-
  • 中国科学报社
  • All Rights Reserved
  • 地址:北京市海淀区中关村南一条乙三号 电话:010-62580783