Periodic Identification of Runoff in Hei River Based on Predictive Extension Method of Eliminating the Boundary Effect of the Wavelet Transform

Abstract

In order to eliminate the wavelet boundary effect caused by the limitation of data length, a univariate predictive extension method based on phase space reconstruction and least squares support vector machine (LSSVM) is proposed. Compared with the traditional extension methods, whether in large scale or small scale, the predictive extension method can effectively eliminate the boundary effect of the wavelet transform. The predictive extension method is used to extend the average annual flow of Zamashik, Qilian, Yingluoxia, and Zhengyixia hydrological stations in the Hei River. Then, wavelet transform multiscale analysis is performed on the extended time series. The wavelet variance curve of the runoff time series after extending is obviously smoother than the unprocessed wavelet variance curve, and to a certain extent, the pseudo period caused by the wavelet boundary effect is eliminated. Analysis results show there are 10-year and 36-year cycles at Zamashik station, 10-year and 25-year cycles at Qilian station, 16-year and 38-year cycles at Yingluoxia station, and 9-year, 21-year, and 30-year cycles at Zhengyixia station.