| In recent years,with the rapid growth of China’s economy,a large number of energy consumption and pollution emissions have been produced,which has caused serious environmental problems,especially air pollution.In 2018,239 cities with over standard air quality in China accounted for about 70%.Air pollution has a direct impact on human health,and also causes direct or indirect losses to all aspects of social economy.Therefore,many cities have established air quality monitoring stations to release air quality information regularly.However,due to the high cost of operation and maintenance,the number of urban air quality monitoring stations is limited,which can not meet the needs of people to understand the urban air quality in advance.Therefore,it is very important for people to accurately predict the air quality of cities for people’s production and life.The goal of urban air quality prediction is to predict the future air quality of specific areas based on historical data.Air quality is affected by many factors,such as meteorological factors,geographical factors,pollution sources and diffusion paths.Accurate air quality prediction needs to consider complex factors such as time and space.In this paper,a spatiotemporal prediction model of urban air quality is proposed by integrating regional clustering,Granger causality analysis and deep long-term and long-term prediction methods.Urban air quality is not only related to local historical conditions,but also affected by neighboring cities.In this paper,firstly,regional clustering is carried out,and then Granger causality analysis method is used to identify the causality of air quality between cities.Furthermore,considering that the air quality is affected by the numerical value in the past few hours and has certain regularity periodicity,the prediction of air quality is a complex nonlinear problem.In this paper,long-term and long-term memory networks are selected to study the time dependence of air quality evolution.Considering that the prediction performance of time series decreases rapidly with the increase of input sequence length,a two-stage attention mechanism is used to optimize the modeling.The results of Granger causality analysis,namely,the urban air quality data that have a great impact on the air quality of the target city,the historical air quality data and meteorological data of the target city are input into the air quality prediction model,and the two-stage attention mechanism is used to optimize the modeling The input attention layer of the attention model selects the corresponding time series,encodes the attention weight,and then selects the corresponding hidden layer state by the temporal attention layer,decodes the importance weight that affects the final prediction,and inputs it to the long-term and short-term memory network layer to obtain the final air quality prediction value.In this paper,urban air quality data from January 2015 to December 2019 were obtained from China environmental monitoring station and National Climate Data Center(NCDC).A total of 344 cities with air quality monitoring records were clustered into 7 clusters.This paper focuses on the key pollution urban agglomeration formed by regional cluster 7(including 58 cities in the North China Plain),and carries out Granger causality analysis on this urban agglomeration region,and obtains the causality network within the region and the pollution source cities of key cities.Compared with the traditional machine learning model and deep learning model,the model proposed in this paper has greater advantages in different feature combinations,different sequence lengths and different target cities.At the same time,compared with the spatiotemporal prediction model of air quality without considering Granger causality,it is found that the introduction of Granger causality not only reduces the amount of calculation,greatly improves the time efficiency,but also reduces the influence of unrelated factors and slightly reduces the error. |
PDF Full Text Request