Qualitative Robust估计建模时异常数据检测的一种新算法

A New Algorithm for Building Models and Detecting Outliers with Qualitative Robust Estimation

本文研究了用 Qualitative Robust 估计建立随机过程 ARMA(p,q)模型时检测异常数据的新算法。构造了新的Ψ-函数,使之具有较宽的适用范围,既能使数据逐层压缩,又有利于检测异常数据;导出了迭代过程中异常数据估计量及其统计特性;锐化了异常数据对判决的影响;构造了最小距离判决规则;给出了迭代算法。Monte Carlo 模拟显示了良好的效果。

A new algorithm for outlier detection is introduced during estimating the ARMA(p,q)model(with qualitative robust estimation).Huber's favorite ψ-fun- orion is modified.A new ψ-function is constructed.The applied range of the rew function is wider and is suitable for detecting outliers.Its robustness is approximate to Huber's ψ-function during iteration.Some estimators and their statistical characteristics are derived.The influence of outlier to decision is increased. Minimum distance decision criterion are established.An iterative algorithm for modeling and detecting with qualitative robust estimation is given.The results of Monte Carlo simulation demonstrate that this algorithm is efficient.