Spatiotemporal Distribution Characteristics Of PM_2.5 Concentration In Sichuan Province And Its Prediction Based On LSTM Network

With the rapid development of urbanization and industrialization in my country,air pollution has gradually become a widely discussed issue in the current society.As the most important air pollutant,PM_2.5has a huge negative impact on people’s travel,life,and health.Sichuan Province is one of the areas with serious PM_2.5pollution in China.Analyzing its temporal and spatial distribution characteristics and influencing factors of PM_2.5,and establishing a good PM_2.5prediction model can provide a theoretical basis for air pollution forecasting and control.Based on those,the main research work of this paper is as follows:（1）We used ArcGIS to draw annual and quarterly PM_2.5concentration distribution maps of 21 cities（prefectures）in Sichuan Province,and conduct spatial autocorrelation analysis of PM_2.5to explore its temporal and spatial distribution characteristics of PM_2.5from 2016 to 2019.The results show that:from the perspective of time,over time,the PM_2.5concentration in various regions of Sichuan Province has decreased,and the pollution range has also shrunk,the concentration of PM_2.5in different seasons from high to low is as follows:Winter,Spring,Autumn,Summer,which is related to the seasonal characteristics of meteorological conditions.From the point of view of spatial distribution,the topography of Sichuan Province is characterized by high west and low east.The eastern basin has low terrain and poor air mobility,which is not conducive to the diffusion of pollutants and is likely to lead to their accumulation.Therefore,the PM_2.5concentration in the study area varies from high to low:Eastern Sichuan Basin,Southwest Sichuan Mountain Area,West Sichuan Alpine Plateau Region.Spatial autocorrelation analysis shows that there is significant spatial aggregation of PM_2.5.The basin area forms a high-high cluster;Guangyuan City and Aba Prefecture form low-high clusters with their surrounding areas;Ganzi Prefecture forms a low-low cluster with its surrounding areas.（2）We analyzed the influencing factors of PM_2.5in Sichuan Province relied on the pearson correlation coefficient,stepwise regression and panel data model.Firstly,the relevant analysis and stepwise regression analysis were conducted by using the average daily air quality data and meteorological data of Chengdu from 2016 to 2020;secondly,the monthly average air quality data and meteorological data of 21 cities（prefectures）in Sichuan Province from 2016 to 2019 were used for Panel data analysis.The results show that among other air pollution factors,PM₁₀,CO,SO₂,and NO₂have a positive impact on PM_2.5,and O₃has a negative impact on PM_2.5;among meteorological factors,temperature,precipitation,and wind speed have a negative impact on PM_2.5,relative humidity and air pressure have a positive impact on PM_2.5;the individual fixed effect model also reveals the impact of different regions on PM_2.5concentration:Zigong,Yibin,Luzhou,etc.,their fixed effect values are positive;Guangyuan,Aba Prefecture,Panzhihua,etc.,their fixed effect values are negative.（3）Taking Chengdu for example,based on time series analysis methods and LSTM network,we established PM_2.5concentration prediction models.Firstly,for the monthly average PM_2.5concentration and daily average PM_2.5concentration in Chengdu from 2016 to 2020,we respectively established a traditional time series model,a univariate LSTM model,and a multivariate LSTM model for prediction.Finally,according to the model evaluation indicators,we compared the prediction results.The results show that the multivariate LSTM model has the best effect on the prediction of monthly average PM_2.5concentration and daily average PM_2.5concentration,with higher prediction accuracy and better overall effect.