Research On PM2.5 Concentration Prediction Algorithm Based On Temporal And Spatial Features

The severe haze weather in the Beijing-Tianjin-Hebei region has attracted people’s attention to PM2.5.Accurate prediction of PM2.5concentration can provide effective guidance for air pollution prevention and related policy formulation.However,due to the long transmission distance and long residence time of PM2.5 in the atmosphere,it is difficult to further improve the accuracy of PM2.5 concentration prediction.This thesis builds a PM2.5 flow graph based on geographic and meteorological data,and combines graph neural network and recurrent neural network to build a PM2.5 concentration prediction model that integrates spatiotemporal features to achieve accurate prediction of future PM2.5concentration in urban agglomerations.The main research work of this thesis is as follows:(1)Aiming at the problem that the existing PM2.5 spatial correlation graph is difficult to effectively simulate the spatial flow and transmission of PM2.5,this thesis proposes an adjacency matrix construction method based on altitude and distance information,and an edge weight calculation method based on geographic and meteorological data.By introducing geographic features such as city altitude,latitude and longitude,and meteorological features such as wind speed and wind direction into the graph structure,a PM2.5 flow graph based on geographic and meteorological data was constructed.(2)Aiming at the low accuracy of multi-step prediction of PM2.5concentration in the existing spatiotemporal prediction model,this thesis proposes a PM2.5 spatial correlation feature extraction method based on WGAT and a PM2.5 prediction method based on AGRU.The two methods are combined to construct a PM2.5 concentration prediction model that integrates spatiotemporal features.By inputting the PM2.5 flow graph into the prediction model,the multi-step prediction of PM2.5 concentration is realized.This thesis conducts a comparative experiment on the KnowAir data set to predict the PM2.5 concentration of urban agglomerations in multiple steps.The experimental results show that the prediction accuracy and air pollution prediction ability evaluation index of the WGAT-AGRU prediction model proposed in this thesis are better than other predictions.The model verifies the validity of the PM2.5 concentration multi-step prediction model proposed in this thesis,and can provide effective help for air pollution control decision-making in practical applications.