Similarity Mining of Hydrological Time Series Based on Semantic Similarity Measures
YU Dazheng · Journal of China Hydrology · 2011
It is now one of the hot spots in scientific research to use data mining technology over long-term observations of time-series to find the regular pattern. Similarity mining is the basis for other tasks of time series data mining. This paper proposed a semantic similarity-based similarity search method over hydrological time-series data. Firstly, wavelet transform was employed to smooth the time-series data; then the extreme points were extracted and the segments were symbolized, each of which represents a semantic pattern. The semantically similar sequences were as the candidate set, so as to get the similar sub-sequences precisely from the candidate set by using dynamic time warping. Experiments on the water level of Taihu Lake show that the method can find the similar sequences accurately while the time complexity is significantly reduced.