函数系数部分线性模型的B样条估计-概率与数理统计专业论文.docxVIP

中文摘要摘 中文摘要 摘 要 非参数方法为统计工作者建立一个更加接近现实的模型提供了更宽松更自由的工 具，由于非参数模型的结构假设很少，尽量让数据为自己说话，具有稳健的优点，近 来受到人们的普遍关注。对于非参数回归，现今研究比较多的估计方法包括：核估 计，局部多项式估计，级数估计，样条估计(包括光滑样条、惩罚样条、B样条等)。 函数系数部分线性模型是一个比较广泛的模型，但该模型由于常数项函数和系数 函数具有不同的自变量给模型的估计带来了不小的挑战。目前对它的研究仅限于用局 部线性方法两步估计系数函数，虽然局部线性方法在数据的局部表现了很好的估计效 果但该方法是分别对常数项函数部分和系数函数部分做估计且对窗宽的选择比较困 难。B样条作为一种全局的估计方法则弥补了局部线性方法的不足，为函数系数部分 线性模型的估计提供了很好的理论准备。 本论文将利用B样条方法对函数系数部分线性模型常数项部分和系数函数同时 进行估计，讨论了两部分估计的收敛速度达到了非参数估计方法的最优收敛速度，同 时给出了估计的相合性和渐近正态性，并且给出了它们的证明过程。最后通过模拟验 证，说明了B样条方法对函数系数部分线性模型的估计是有效的。 关键词：函数系数部分线性模型，B样条，最小二乘法，渐近正态性 英文摘要Abstract 英文摘要 Abstract Nonparametric methods are developed as more flexible tools for statistician to build more accurate models．With the advantages of very few restrictions on model struc— ture，strong robustness，and data-driven property,nonpaxametric models axe extensively discussed recently．Regarding nonpaxametric regression models，the most widely used ap- proaches include：kernel method，local polynomial technique，series estimation and spline method(such as smoothing spline，penalized sprine and B—spline)． Functional-coefficient partially linear regression model is a generalized model by com- bining nonparametric and functional-coefficient regression model．It is proposed aiming to fit the model using the real data of Yingluo River．The most challenging part is that the smoothing variable of the constant part is different from that of the coefficient functions which brings difficulty to estimation．So far the local linear technique and the integrated method have been introduced to estimate all functions in FCPLR model．Although local linear method shows a good statistical property to the local area while in the global area it still needs improvement．B—spline with the appearing statistical properties as a global technique is provided to improve the estimator． This dissertation plans to employ B-spline to estimate all the coefficients of Functional-coefficient partially linear regression model．The least square estimation method and asy