AR模型参数的抗差估计研究

Robust estimation of AR model parameters

实时校正一般以实测洪水流量为校正依据.研究实测洪水流量过程出现异常值时,采用抗差递推最小二乘法代替传统递推最小二乘法估计AR模型参数,能获得更稳健的参数结果.将闽江七里街流域的洪水资料人工生成异常值,对采用抗差递推最小二乘法和传统递推最小二乘法所得的校正结果进行比较,结果表明抗差递推最小二乘法具有更强的容差能力,是一种稳健的参数估计方法.

The real-time rectification is generally based on the measured data of flood discharge. The parameters of AR model obtained by the robustified recursive least square method are more robust than those obtained by the standard recursive least square method when outliers occur in the measured data of flood discharge. By artificial generation of outliers among the flood data of the Qilijie Basin of the Minjiang River, the rectified results of parameters by the two approaches were compared. The result shows that the robustified recursive least square method has potential to reduce estimation bias in the presence of noise, and that it is superior to the standard recursive least square method.