Research On The Analysis And Prediction Of Urban Air Pollution Through Spatio-temporal Data Mining

The frequent regional air pollution has contributed to the massive investment in environmental information,but also directly led to the explosive growth of environmental data.The public is increasingly pressing for sharing urban air quality data and getting helpful knowledge and information services for daily travel.Meanwhile,the government is also increasingly pressing on the need for the policy-making as the output of large environmental investment.As to the high-dimensional and massive environmental data,how to mine the hidden values and provide management services for decision-making and public travel is a great challenge.This thesis focuses on exploring urban air pollution associations,transport path and prediction methods based on spatio-temporal data mining.The first step is to collect multi-source spatio-temporal data and do the data pre-processing.Four types of data are shown in this thesis,including urban air quality data,weather conditions,traffic,and urban point of interest?POI?.The first task of air quality data mining is to develop a region segmentation approach based on the spatio-temporal characteristics of air pollution.The air quality tensor model was proposed with third orders from site,day and time dimensions.Five pollution patterns were divided through the Candecomp/Parafac?CP?decomposition.An extra POI distribution model was developed for helping understanding the formulation and variation of air pollution pattern.Meanwhile,the functional types of the contaminated areas were also identified with the combined information of the POI feature model.This tensor model avoids the division of air pollution patterns by tensor decomposition,making up for the lack of a large amount of correlation information loss caused by the separation of the patterns in previous methods.Secondly,an Ambient air quality-Traffic-Meteorology Maximum Information Entropy model and the method for discovering associated rules of atmospheric compound pollution under different meteorological conditions were proposed.The combined effects of traffic and meteorology on environmental pollution were quantitatively analyzed.The associated rules for the compound heavy pollution of PM_2.5 and O₃ under different meteorological conditions were clearly shown.These rules provide support for ambient air pollution control strategies.Thirdly,a conventional urban air pollution transport path and pollution source discovery method were conducted.An air pollution graph was developed,and frequent sub-image mining methods was used to achieve urban internal pollution transport path.This approach perfectly models the transport path of air pollution.Finally,a Sequence-to-Sequence model for predicting the concentration of air pollutants was proposed.This method makes full use of the advantages of long and short-term memory of LSTM and the ability of the time series characterization by the combination of coding and decoding process.It also takes the multiple pollution of atmospheric compound pollution into account.The effect of gas pollutants and the meteorological conditions on the concentration of particles is totally considered.Both the prediction accuracy of the air pollutants concentration and the operating speed are improved based on the sequence-to-sequence model of multi-source data.The research focused on the innovative data-driven approaches for urban air pollution associations,pollution transport path and pollution prediction methods based on spatio-temporal data mining.It provides new ideas and new methods for the analysis of environmental big data mining and air pollution problems.The results could also provide scientific basis for the targeted treatment of urban air pollution.Simultaneously,it gives guidance support for urban construction,industrial planning,traffic organization,residents'travel decisions,and environmental emission reduction decisions.