摘要
回归分析中,回归系数的确定一般采用普通最小二乘法(OLS),而OLS估计参数一般要求数据满足一些性质,如正态性等等。而现实中的数据,往往不能满足那些性质。这样就导致常用的参数估计方法,如OLS,很难达到满意的预测精度。对于含有异常点的数据,引进了一种方法——M-估计法,分别采用最小二乘法和稳健回归的M-估计法估计参数,建立了用电量的回归模型。通过对比表明了稳健M-估计可以克服最小二乘估计受离群值影响较大的弊病,能保持较满意的预测精度,使模型参数更接近实际模型,具有稳健性。
It is difficult to achieve satisfactory precision by use of common parameter estimation method. For data containing outliers, this paper introduced one new method-M-estimator. In this paper, one regression model of power load based on least square and M-estimator of robust regression was built. By contrast, it showed that M- estimator has less influence of outliers than least square and can keep a certain forecasting precision. The parameter will be more close to the real mode and M-estimator has more strong robustness.
出处
《电力科学与工程》
2009年第4期25-27,共3页
Electric Power Science and Engineering