A Robust Method for Change-Points Detection in Hydrological Time Series
Huihui Helen Wang · 2008
Traditional studies on the detection of change-points in hydrological time series considering little infection of noise always ignore the robust of the methods.In this paper,a highly robust Gaussian mixture density decomposition method is proposed for finding change-point of the mean value for hydrological time series.Given the observed hydrological data,different parts of the time series which are realized from different normal distributions are identified successively.Thus identifying change-point is transferred to clustering problemes based on the normal mixture models.In this method,the number and the position of change-points can be determined by analysising the different normal components in the mixture which have been arranged,so they can be gotten automatically.The computation results of the examples show that this method is robust and effective in change-point detection.And the robustness of the change-points detection for hydrological time series can be improved.