中文摘要：

提出利用中位数法与集成经验模态分解（EEMD）相结合的方法对时间序列数据的异常值进行检测,首先通过中位数法对明显异常的数据进行初步筛选,再用EEMD对剩余数据进行分解,通过叠加低频分量可以拟合出大多数数据的整体变化趋势,而不受异常值的影响,从而根据偏差比率可有效检测出异常值。然后根据异常值检测后的时间序列数据的凹凸性变化趋势,用分段曲线拟合对异常值校正。最后,以H1自来水厂的日取水量数据为例进行实证分析。结果表明：提出的中位数法与EEMD相结合的方法能够有效地检测异常值,校正后得到的数据能够真实反映该水厂取用水情况,可为后续分析提供更加真实可靠的数据。

英文摘要：

In order to improve the availability and accuracy of online monitoring data of water resources, it is very important to detect and correct the outliers of monitoring data. The water resources monitoring data are non-linear and non-stationary time series data, the outlier detection method of the conventional time series did not take into account the convexity and concavity of time series. A combining median and ensemble empirical mode decomposition （EEMD） method was presented for outlier detection. Firstly, the outliers were preliminarily detected by the median method. And then the remaining data were decomposed by EEMD. The overall trend of most of the data can be fitted by superimposing the low- frequency components, but not affected by outlier, and the outlier can be detected effectively according to the deviation rate. Then, according to change of convexity and concavity of time series data after outlier detection, the method of piecewise curve fitting was used to correct the outliers. Finally, taking the daily water intake data of H1 waterworks as an example, the results showed that the method of combining median and EEMD can detect outliers effectively. The data obtained after correction can truly reflect the actual situation of water intake of waterworks. It can also provide more reliable data for subsequent analysis.