摘要
普通最小二乘法将所有的数据点对预测值(控制值)的影响作用等同看待。实际不应该如此,数据点的相互联系是有所区别的。以时间序列数据来说,近期的数据点往往更能说明待预测值,而远离预测期的早期数据点关联作用较小。基于这种看法,这里提出了最小二乘加权法,对不同的数据点在离差平方和算式Q中给予不同的权数,推导出参数估计的公式;提出了最小二乘加权法权重设置的机理,并就时间序列数据提出指数权重法。
In traditional Least Squares, all data points are deemed to have the same influence on forecasted point. However, it isn＇t the fact. Different data points should have different status. Taking timing serial data for example, recent data points usually have more influence on forecasted point, while forepart data away from expected period have less. So here we bring forward Weighted type Least Squares（WTLS）. It gives different weight to different data point in the Q equation. Then the parameters of the equation are estimated. Put forward the mechanism to establishment weights in the method of weighted type Least Squares. Exponential weight is designed to serve timing serial data.
出处
《东华大学学报（自然科学版）》
CAS
CSCD
北大核心
2006年第6期51-55,共5页
Journal of Donghua University(Natural Science)
关键词
数据点
最小二乘加权法
权数
指数权重法
data point, weighted type least squares, weight, exponent weight
