Analysis And Prediction Of Air Pollution Data Mining Based On Association Rules

In recent years,China has achieved rapid economic growth and made great progress in the development of industrialization,motorization and urbanization.Heavy air pollution occurs frequently.PM2.5 and O3 have become important obstacle factors limiting the improvement of air quality in China.In order to investigate the causes of heavy air pollution and improve air quality,China has been committed to air environment monitoring and has accumulated a large amount of air quality monitoring data.Mining spatial and temporal variation patterns of air pollution,analyzing potential source areas and transmission paths of pollution,exploring potential information within the data and predicting pollution are particularly important to efficiently use massive data and avoid data wastage.This study will focus on Changzhi,Shanxi Province,to explore the characteristics of urban air pollution,influencing factors,transmission characteristics,and then predict pollution.In this paper,multi-source spatio-temporal data of atmospheric environment and meteorology in Changzhi were collected.The missing values were supplemented by tensor decomposition for subsequent data analysis.First,the study analyzed the temporal variation patterns and transmission characteristics of air pollutants in Changzhi.Based on long time,multi-point and complete pollutant monitoring data,the study derived the spatial and temporal distribution characteristics of PM2.5 and O3 pollution changes in Changzhi in different seasons.The backward trajectory model was applied to analyze the pollutant transport paths and potential source areas.Second,hidden association rules between meteorological factors and atmospheric pollutants were identified.Based on the maximum information coefficient calculation method,the correlation between meteorological factors and pollutants was quantitatively analyzed,and the association rules between meteorological factors and pollutants under PM2.5 and O3 pollution in different seasons were excavated.Finally,a comparative study of predicting PM2.5 concentrations under multiple ensemble learning methods was completed.Considering that the real atmosphere was a complex system with multi-scale correlations and multi-factor influences,this study used the characteristics of various factors related to PM2.5,such as meteorological conditions,concentrations of other pollutants at the study station,PM2.5 concentration data from neighboring stations,and own historical time series data,to build PM2.5 ensemble learning prediction models and conduct a comparative study of the prediction results.Mining and analysis of monitoring data within the implied characteristics of air pollution changes and association rules and then on this basis to predict the pollution.It can systematically understand the causes of air pollution and provide data support and methodological basis for the refined control of urban air pollution and timely prediction of future pollution.This also can provide information reference for people’s production and life to reduce the impact of pollution on human health and property.