基于模糊预测系统的观测数据野值剔除方法

针对观测数据序列中影响数据处理和分析的野值,提出了一种在线辨识和剔除野值的方法。该方法利用梯度下降法构造最小均方准则下最优的观测序列模糊预测系统,从而获得预测值与观测值的残差序列,然后基于狄克松准则快速辨识并剔除观测数据中的异常值。对实测数据的仿真实验表明:该方法能够准确跟踪观测信号的变化,适合于各种观测信号单个性野值的辨识和剔除。

New method for outlier removal from observed data based on fuzzy forecasting system

To determine the outliers having an influence on data processing and analysis in observed data,a new on-line method for identifing and eliminating outliers based on fuzzy forecasting system of time series.Firstly,the optimal fuzzy forecasting system is designed according to the least mean square criterion by gradient descent algorithm,and residual error sequence between observed data and predicted data is obtained.Then outliers are identified and discarded according to exceptional value in residual error sequence based on Dickson criterion.Effectiveness of the new method is proved through experiments with real observed data.The on-line fuzzy forecasting system of time series can exactly track the signals,suitable to identify and discard outliers in various of signals with right initialization.