Research On Air Quality Prediction Method Based On Spatio-temporal Graph Neural Network

There is a close bind between air quality and human living standards.Persistent and serious air pollution not only threatens people’s life safety,but also overdrafts the potential of national development.Due to deal with it,China proposes a concept of harmonious symbiosis and green development,and the work in this research also comes from the actual prediction project of cooperation with environmental protection scientific research institutions.As the main air pollutant in China is PM2.5,an air quality prediction method based on spatio-temporal graph neural network is proposed in this thesis.This method combines a variety of methods of feature analysis,spatial analysis,and time series analysis to predict PM2.5.The main contents of the thesis are as follows:1.To make full use of features in the datasets of meteorological monitoring stations and air monitoring stations,and guarantee the model can be applied in the datasets.A data preprocessing method combined with feature selection and data filling is used to process two real-world air quality datasets.2.In order to obtain the graph structure that can fully capture the relationship inter atmo- spheric monitoring stations,a graph search and optimization method based on evolution- ary multi-objective algorithm is proposed in this thesis.This method can combine the analysis methods in the field of geographic information,machine learning,and time series analysis to search a variety of graph structures.Then,through a compression reduction method,the evolutionary multi-objective algorithm is adopted for iterative optimization.Combined with the prediction model,a high-quality graph structure with outstanding per- formance and less redundancy is obtained.3.An Adaptive Scalable spatio-temporal Graph Convolutional Network(ASGCN)model is proposed to better analyze spatio-temporal data.Several modules of the model are used to optimize the global graph structure,dynamically construct the local graph structure,and analyze the characteristics of time series period according to the long and short periods,which solves the problem that the previous model can not optimize the bad graph structure or dynamically construct the graph structure suitable for the current data; It also solves the problem of parallel analysis of long and short periods of time series.Two real Chinese air quality datasets are preprocessed and the PM2.5 are analyzed and predicted in this thesis.Through a large number of experiments,The accuracy and feasibility of the air quality prediction method based on spatio-temporal graph neural network model is verified in experiments.The method proposed by this thesis is a novel and effective prediction method of spatio-temporal series data with great development trends.The research results of this thesis can provide effective ideas for the prediction and early warning of air quality.And there will be opportunities to provide effective methods and ideas for air pollution prevention and planning in the future.