Water environment index spatio-temporal correlation simulation and prediction method based on graph deep learning

Abstract

The application relates to the technical field of water environment monitoring and discloses a water environment index space-time correlation simulation and prediction method based on graph deep learning. The method comprises the following steps: acquiring water environment monitoring data of each monitoring station at the current time, updating feature vectors of each node in a space-time graph structure of a basin water environment monitoring station, inputting a graph attention network module for multi-layer space message transmission, obtaining a space representation sequence of each node at each time step, inputting a time sequence attention module, capturing long-term time dependence, and generating a space-time embedding vector of each node; simultaneously outputting prediction sequences of multiple water environment indexes by using a multi-task prediction head; the prediction sequence is a prediction result optimized based on a water environment substance migration conservation physical constraint; a Monte Carlo random inactivation layer is enabled, a prediction mean value and a confidence interval are calculated according to a normal distribution assumption, and multi-step rolling water environment prediction is carried out. The above scheme improves the water environment prediction accuracy.

Images