Research on Anomaly Pattern Detection in Hydrological Time Series

Abstract

The abnormal patterns in hydrological time series play an important role in the analysis and decision-making. Aiming at the problems that the amount of hydrological data is large and there is a lot of “noise” in this data, which lead to the high time complexity of traditional anomaly detection algorithm, we propose anomaly pattern detection based on density for hydrological time series. Firstly, this method makes a piecewise linear representation of the sequence through the important feature points, then extracts the slope, length and mean of the pattern, and maps them to the three-dimensional space. Finally, it calculates the local outlier factor of each pattern. The selection of important feature points and parameters in the algorithm are discussed and verified by the actual data which are historical water level of Jin-niu mountain reservoir. Experimental results show that the algorithm has low complexity and it has full mining results, which can meet the requirements of large-scale time series.