Application Of Spatio-Temporal Mixed Model In Air Quality Prediction

In recent years,China ’ s economic and social development has made remarkable achievements.With the continuous expansion of industry and urban scale,the resulting air pollution problem has become increasingly prominent.Effective prediction and assessment of air quality can reduce the negative impact of air pollution as much as possible,and help to formulate pollution prevention measures,which has important practical significance for regional atmospheric environmental governance.With the increasing popularity and continuous improvement of environmental monitoring network,atmospheric environmental monitoring has entered the era of big data.Big data analysis and processing technology represented by machine learning and deep learning,with its powerful information extraction and data fitting ability,makes it widely used in target recognition,classification,clustering and prediction,and provides a new method for atmospheric quality prediction.In this paper,the latest spatiotemporal data mining technology is introduced into the air quality prediction.Taking the air quality monitoring data of 12 cities in the Beijing-Tianjin-Hebei urban agglomeration for one year as the experimental sample,the time series prediction model,the spatial series prediction model and the spatio-temporal weighted hybrid prediction model are constructed respectively.The specific research contents and results are as follows:⑴ In the data preprocessing stage,aiming at the missing values in the original data,based on the traditional mean filling method,according to the characteristics of air quality data and the actual situation of regional air quality,a random forest regression filling algorithm is proposed.This method has a very good effect in filling the data set with multiple features and the correlation between the feature matrix and the label,which is especially suitable for the filling of air quality data.⑵ Single time series model(ARIMA)can not fully extract all the information in the sequence.Therefore,this paper proposes a combination model of ARIMA and support vector regression(SVR).ARIMA model is suitable for linear feature extraction,SVR model is suitable for nonlinear feature extraction,and the combination of the two can achieve complementary advantages.The experimental results show that the prediction accuracy of ARIMA model is significantly improved after the residual correction of SVR model.⑶ In order to study the spatial correlation between the air quality of the target site and the air quality of other sites in the surrounding area,a spatial series prediction model based on long-term and short-term memory neural network(LSTM)is proposed on the basis of the time series prediction model.The experimental results show that the spatial sequence prediction model constructed by LSTM can fully extract the spatial correlation of air quality data and give accurate prediction from a spatial perspective.⑷ In order to further improve the prediction accuracy,the time and space prediction models are integrated based on the Gaussian mixture model(GMM),and the spatio-temporal weighted aggregation prediction model is constructed.The experimental results show that the spatio-temporal weighted aggregation prediction model has good integration effect for time and space prediction models in different seasons,and its prediction accuracy is better than that of a single time or space series prediction model.There are 30 pictures,21 tables,and 71 references in this paper.