Development of a novel outlier index for real-time detection of water level outliers for open-channel water transfer projects

Abstract

Abstract Real-time detection of water level outliers is critical for real-time regulation of gates or pump stations in open-channel water transfer projects. However, this remains a challenging task because of the lack of definition of water level outliers and the imbalance of flow monitoring data. In this study, we define the water level outliers and then propose a highly accurate outlier index for real-time detection of water level outliers based on the water level-flow relationship, and the thresholds for water level outliers are determined based on the order of magnitude of flow and water level differences. A case study is performed with the South-to-North Water Diversion Project of China. A random noise is added to 15 randomly selected non-adjacent monitoring datasets to verify the accuracy of the index, and the noise is increased from 4 to 9 cm at a step of 1 cm. The results show that a total of 159 outliers are detected out of 180 outliers with an accuracy rate of 88.3%.