Application of unsupervised clustering model based on graph embedding in water environment

Abstract

Abstract Surface water monitoring data has spatiotemporal characteristics, and water quality will change with time and space in different seasons and climates. Data of this nature brings challenges to clustering, especially in terms of obtaining the temporal and spatial characteristics of the data. Therefore, this paper proposes an improved TADW algorithm and names it RTADW to obtain the spatiotemporal characteristics of surface water monitoring points. We improve the feature matrix in TADW and input the original time series data and spatial information into the improved model to obtain the spatiotemporal feature vector. When the improved TADW model captures watershed information for clustering, it can simultaneously extract the temporal and spatial characteristics of surface water compared with other clustering algorithms such as the DTW algorithm. We applied the proposed method to multiple different monitoring sites in the Liaohe River Basin, analyzed the spatiotemporal regional distribution of surface water monitoring points. The results show that the improved feature extraction method can better capture the spatiotemporal feature information between surface water monitoring points. Therefore, this method can provide more potential information for cluster analysis of water environment monitoring, thereby providing a scientific basis for watershed zoning management.