基于滑动窗口采样技术和DTWCorr距离度量的多元时间序列分割方法研究OA

To improve the accuracy of multi-time series event detection in watershed water resource management,in response to the problem that traditional synchronous segmentation methods ignore the time lag and asynchrony between variables,a multi-time series asynchronous segmentation method based on sliding window sampling technology and composite metric(DTWCorr)was proposed.The method first obtained initial segments of each variable using univariate time series segmentation,then integrated multi-variable segment information through time correlation segmentation to capture asynchronous characteristics among variables.Multi-scale sliding windows were subsequently applied to resample the segments,enhancing data robustness.The DTWCorr distance metric was constructed by combining DTW with the Pearson correlation coefficient,and the multi-segment graph shortest path algorithm was used to achieve globally optimal asynchronous segmentation.Experiments based on water quality data from the Taihu Lake watershed demonstrated that this method outperformed traditional synchronous segmentation in segmentation accuracy,inter-variable correlation,and noise robustness.It more accurately reflected the asynchronous variation patterns of multivariate time series.This study provides reliable data support for hydrological event identification and intelligent watershed water resource management.