基于递归滤波算法的非线性状态空间模型的统计推断-应用数学专业论文.docxVIP

摘要 摘要 摘要 状态空间模型是一种动态时域模型，以时间为自变量，是分析时间序列数据 的有力工具，能完全表征系统的所有动力学特征，目前的研究方兴未艾，其应用 也非常广泛． 本文将处理纵向数据的混合效应模型引入到状态空间建模中，并利用该模型 研究了 HIV 动力学特征. 首先我们基于 HIV 动力系统的非线性微分方程模型，通 过欧拉离散得到了该系统的非线性状态空间模型；继而考虑实际临床病人的个体 差异性，将混合效应模型嵌入其中，由此建立的模型称为非线性混合效应状态空 间模型（NLMESSM）．基于该模型，我们利用非线性状态空间模型统计推断中的 粒子滤波（PF）算法和极大似然递归滤波（MIF）算法以及混合效应模型中处理纵 向数据的 EM 算法对模型的状态变量和参数进行估计．模拟结果显示我们的模型 能够很好地刻画这一非线性系统的动力学特性，参数估计可以收敛至真实值，状 态预测结果合理，达到了预期要求，具有一定的实际意义和应用价值． 鉴于状态空间模型应用的广泛性，NLMESSM 及其相关算法亦可扩展应用于 生物、工程、经济、金融等其它领域． 关键词：HIV 动力系统 状态空间模型 混合效应模型 粒子滤波 EM 算法 基于递归滤波算法的非线性状态空间模型的统计推断 Abst Abstract Abstract State space models are a kind of dynamic models which act as powerful tools in time series data analysis. Such models have widely used in many fields. This paper draw the idea of state space modelling into the research of HIV dynamics, which are usually characterized by a system of ordinary differential equations(ODEs). Based on the ODEs models of HIV dynamics, nonlinear state space models with mixed-effect (NLMESSM) are proposed to analyze the HIV clinical longitudinal data. Under the framework of NLMESSM, filtering based algorithms such as particle filtering(PF) and maximum likelihood via iterated filtering(MIF) which are usually implemented for the statistical inference in state space models and EM algorithm which is of great importence to longitudinal data analysis in mixed-effects models are used here to estimate the state variables and parameters respectively. Simulation results show that for both states estimation and parameters estimation, NLMESSM and the proposed algorithms perform well. In view of the widely use of state space models, the proposed models and methods may also have potential applications in other fields, such as biology, HYPERLINK /s?wd=engineering engineering, economics and finance. Keywords: HIV dynamics state space models mixed-effects models particle filtering EM algorithm Abstract 目录 目录 目录 HYPERLINK \l _bookmark0 第一章 绪论 1 HYPERLINK \l _bookmark1 1.1 状态空间模型概述 1 HYPERLINK \l _bookmark2