Atmospheric pollutant prediction method and system based on deep learning model

Abstract

The invention provides an atmospheric pollutant prediction method based on a deep learning model. The atmospheric pollutant prediction method comprises the following steps: converting unstructured input data based on a monitoring station into graph structure data containing structure information; learning spatial interactivity between monitoring stations by using a GNN model; inputting the monitoring station data completing the spatial relationship interaction into a GRUA model to capture the time dependence of PM2.5; and in combination with the to-be-predicted characteristic factors at the future moment k, completing the prediction of the PM2.5 concentration at the future moment k by using the full-connection network. According to the invention, sparsity and non-Euclidean distribution characteristics of environment monitoring stations are considered, and spatial correlation modeling prediction is carried out by combining the advantages of GNN on a spatial interaction relationship; in combination with the capture performance of the GRU on the time correlation, an attention mechanism is fused to capture global information, so that the modeling prediction capability on the time correlation is improved; and in combination with the land utilization type data in the multi-source data, the capture of the model on the spatial correlation is enhanced, so that the accuracy of model prediction is improved.

Description

technical field [0001] The invention relates to the technical field of air pollutant prediction, in particular to an air pollutant prediction method and system based on a deep learning model. Background technique [0002] Atmospheric pollution has a major impact on human beings and the ecological environment on which they depend. Among them, PM 2.5 It is the main component of smog, which endangers human health and increases the morbidity and mortality of cardiovascular, cerebrovascular and respiratory diseases. The timely and accurate prediction of the concentration of air pollutants is helpful for scientific prevention and effective reduction of losses caused by pollution events. [0003] In recent years, with the rapid increase of observational data, it is difficult for traditional numerical forecasting and statistical methods to make full use of massive data modeling. Deep learning methods that can be adapted to big data analysis have become a research hotspot. Early d...

AI Technical Summary

Problems solved by technology

[0006] In general, the current deep learning prediction method of air pollution combines the concentration of air pollutants with meteorological data and simple geographic information, and considers the spatial interaction dynamics and spatial correlation modeling between cities, and lacks the complexity of the site itself. In addition, in terms of capturing the time dependence of pollutant concentration, although the existing methods extract the time correlation characteristics of historical data to a certain extent, they ignore the different moments in the past. The influence of the characteristic state of the future on the concentration of pollutants

Images