Wan Dingsheng

A Novel Approach to the Similarity Analysis of Multivariate Time Series and Its Application in Hydrological Data Mining

There has been large amount of hydrological data collected by various sensors during the last years and how to discover the hidden knowledge among these data has caused more and more attention from diverse fields, such as hydrologist and researchers from data mining. This paper deals with similarity mining from hydrological time series and concentrates itself on the similarity analysis of multivariate time series (MTS). A novel similarity measure has been put forward, which is based on the well-known BORDA count in multiple classifier system. Firstly, dimension reduction is adaptively conducted according to the target data complexity; then the similarity of single time series is computed and lastly, the overall similarity of the MTS is obtained by synthesizing each of the single similarity based on BORDA count. Experiments on the similarity analysis of water level data of TAIHU Lake and historical flood data from YIFENG Basin have shown the feasibility and effectiveness of the proposed method.

Study on Hydrology Time Series Prediction Based on Wavelet-neural Networks

To improve the performance of the model of wavelet-neural network for complex time series, a novel multi-factor prediction model is proposed, which inherits the advantage of previous wavelet-neural network and makes full use of the characteristics of different time series on different time scales. In addition, this paper also put forwards a criterion for the selection of wavelet functions, which is based on the weighted sum of the linear correlation coefficients of different time series with the target peer series. At last, the proposed model is applied to the prediction of daily discharge of WANGJIABA station at HUAIHE River and the experimental results show that higher accuracy is achieved, compared with previous models.