Environmental Missing Data Imputation And Air Pollutant Concentration Prediction Based On Deep Learning

With the rapid development of industrialization and urbanization,the air pollution has become the major problem which hinders global development.The accurate and effective method of predicting air pollutant concentration can offer important information support for the treatment of urban environmental pollution and the urban construction.Nowadays,many cities have carried out air quality monitoring,and have accumulated massive high-dimensional environmental data,but data lacking and damaging problems are not well settled at all times.The purpose of the paper is to improve the accuracy of environmental missing data imputing and air pollutants concentration predicting.Considering the nonlinearity and isomer of high-dimensional environmental data,thesis uses the deep learning methods to analyze it,the main work as followings:(1)Focusing on the problem that the existing missing data imputation model lacks long-term dependence,the thesis introduces dual attentions mechanism based on the missing data imputation algorithm(E~2GAN),then adaptively extractions of time series features,captures potential and important characteristics in historical data.In addition,adding the gradient penalty to the loss function of the discriminator,and the environmental missing data is effectively imputed.(2)The zone division method of spatial correlation is proposed.Using the cosine similarity to calculate the air pollutant concentration and the spatial correlation of meteorological factors,the region division is performed by the KNN algorithm,thus filtering out the site data which shows lower influence.(3)Considering environmental factors,meteorological factors,and geographical factors,combined with residual network,a multi-modal input air pollutant concentration prediction model is proposed(MR-ALSTM).On the basis of space correlation zone division,adaptively integrates the time and space information from different air quality monitoring sites,and then enhances the accuracy of the concentration of air pollutants.(4)Experiments with collected environmental data from four cities in Shaanxi Province in the past three years show that the improved missing data imputation algorithm can effectively impute the environmental miss data,and the proposed multi-mode input air pollutant concentration prediction model effectively combined with multi-mode data,the prediction perform well.The research work of the thesis proves the feasibility of considering the missing data imputation and multi-mode input air quality prediction method,and provides a practical basis for the missing data imputation and the air pollutant concentration prediction.