Research On PM2.5 Concentration Prediction Method Of Hybrid Spatiotemporal Model

With the continuing advancement of industrialization,the air quality of cities in China is deteriorating rapidly.In particular,the frequent occurrence of haze weather seriously affects people’s lives and health,and PM2.5 is the main factor causing haze weather.Exploring the change in PM2.5 concentration can better formulate air pollution prevention and control plans.At present,many deeper learning methods are used in PM2.5 concentration prediction.Existing PM2.5 concentration prediction methods cannot effectively capture the complex nonlinearity of PM2.5 concentration,and most of these methods cannot accurately simulate the spatiotemporal characteristics of PM2.5 concentrations.The main content of this thesis is as follows:1.In the prediction of PM2.5 concentration in the time dimension,the existing prediction methods of time series decomposition usually use the prediction of each subsequence obtained after decomposition,and then add the prediction results of each subsequence to obtain the final result,which in turn causes errors to accumulate and consume a lot of time.In order to reduce the generation of errors and save time,and combine the advantages of multi-scale components of sequence decomposition,the CEEMDAN-Time2Vec-GRU model is proposed.The model uses CEEMDAN to capture the nonlinearity of PM2.5 concentration series,and uses Time2Vec to replace the original series,captures the periodicity and aperiodicity of the series,and optimizes prediction effects of GRU.Experiments show that the model can be effectively applied to relatively long prediction tasks.2.In the prediction of PM2.5 concentration in the spatiotemporal dimension,by adopting a spatiotemporal modeling method,the convolution operation in the three-dimensional CNN is used to extract the spatial features of the data,and then use the GRU network to fit the time characteristics of the data to complete the prediction of PM2.5 concentration.3.In the existing prediction methods for time series decomposition,they often only use the historical data of the target site without considering the spatial characteristics of PM2.5 concentration,and cannot make full use of the spatial characteristics of air pollutant concentration.A spatiotemporal modeling approach is proposed.In this way,the CEEMDANTime2Vec-GRU model and the CNN-GRU model are fused using adaptive weights to obtain the CEEMDAN-Time2Vec-GRU-CNN hybrid spatiotemporal model.