数学物理学报 ›› 2020, Vol. 40 ›› Issue (5): 1381-1392.

• 论文 • 上一篇    下一篇

线性回归模型参数估计方法的分辨率

张晶(),余旌胡*()   

  1. 武汉理工大学理学院 武汉 430071
  • 收稿日期:2019-12-18 出版日期:2020-10-26 发布日期:2020-11-04
  • 通讯作者: 余旌胡 E-mail:sdsgzj@163.com;yujh67@126.com
  • 作者简介:张晶, E-mail:sdsgzj@163.com

Parameter Resolution of Estimation Methods for Linear Regression Models

Jing Zhang(),Jinghu Yu*()   

  1. Department of Mathematics, School of Science, Wuhan University of Technology, Wuhan 430070
  • Received:2019-12-18 Online:2020-10-26 Published:2020-11-04
  • Contact: Jinghu Yu E-mail:sdsgzj@163.com;yujh67@126.com

摘要:

为给出算法对一般参数的区分能力,提出参数分辨率的概念.该文结合聚类思想,给出参数分辨率的定义和计算方法,并分别用最小二乘法与全最小一乘法对一元线性回归模型进行参数分辨率分析.实验结果表明两种算法均具有性质:参数分辨率的精度随信噪比的增大而增大;局部参数分辨率与整体参数分辨率保持一致;噪声的标准差与最小二乘参数分辨率满足线性关系,并利用区间估计理论给出证明.最后,对两段相似的音频信号,分别采用最小二乘法和全最小一乘法进行参数分辨率分析,实验结果表明了参数分辨率概念的合理性和有效性.参数分辨率是衡量两个相近信号能否被分开的一个标准,是评价模型及算法准确度的有效指标.

关键词: 参数分辨率, 最小二乘法, 全最小一乘法, 聚类, 区间估计

Abstract:

In order to give the algorithm's ability to distinguish general parameters, this paper proposes the concept of algorithm parameter resolution. This paper combines the idea of clustering to give the definition and calculation method of parameter resolution. The least squares estimation and the total least absolute deviations estimation method are used to analyze the parameter resolution of unary linear regression model. Experimental results show that both algorithms have properties:as the SNR increases, the accuracy of the parameter resolution is higher; the local parameter resolution is consistent with the overall parameter resolution; the standard deviation of noise and parameter resolution of least squares satisfy a linear relationship which has been proved by using interval estimation theory. Finally, the least squares and the total least absolute deviations are used to estimate the parameter resolution of two similar audio signals. The experimental results illustrate the rationality and effectiveness of the definition of parameter resolution. Parameter resolution is a criterion for measuring whether two similar signals can be separated, it is also an effective indicator for evaluating the accuracy of models and algorithms.

Key words: Parameter Resolution, Least Squares, Total Least Absolute Deviations, Clustering, Interval Estimation

中图分类号: 

  • O224