基于最近邻分析的空气质量时空数据异常点识别

发布时间：2018-05-16 15:51

本文选题：空气质量 + 时空数据　；参考：《统计研究》2017年08期

【摘要】：空气质量数据具有在时间上连续、空间上相关的特点,这提高了异常点识别的难度。本文提出在时间维度上运用移动平均法,在空间维度上运用反距离加权法对观测值进行预测并求残差的解决思路,从而将时空数据的异常点识别问题转化为二维残差值的异常点检测问题。通过仿真验证表明新方法具有良好的检出力。最后将新方法应用于北京市实际观测数据,取得了满意的识别效果。
[Abstract]:Air quality data have the characteristics of continuous time and spatial correlation, which makes it more difficult to identify outliers. In this paper, the method of moving average in time dimension and inverse distance weighting method in spatial dimension are put forward to predict the observed value and solve the residual error. Thus, the problem of identifying outliers in spatiotemporal data is transformed into the problem of detecting outliers in two dimensional residuals. Simulation results show that the new method has good detection power. Finally, the new method is applied to the actual observation data in Beijing, and satisfactory recognition results are obtained.
【作者单位】：天津大学管理与经济学部工业工程系;天津大学管理与经济学部;
【分类号】：X51

【相似文献】