Research On Joint Prediction Of Air Pollutants Based On Multi-modal Data

Air pollution problems have a severe effect on the natural environment and public health,how to use the existing data to effectively predict air pollutants is of great significance for preventing air ollution accidents,reducing environmental pollution and protecting human beings.At present,the prediction methods of air pollutants concentration still use traditional methods,such as statistical prediction and numerical prediction.Although the traditional forecasting method can produce relatively accurate prediction to some extent,it will consume a lot of manpower and material resources.In addition,the sources of air pollutants are more and more extensive,and the types of pollutants are more and more,which increases the complexity of air pollutant prediction.Current bottlenecks in air pollutant prediction include:(1)When predicting concentration of air pollutants,the effects of other air pollutants or meteorological data on pollutant concentration prediction were not fully considered.(2)the prediction of pollutant concentration of air pollutants fails to make full use of the big data of pollutants and meteorological history to mine the relationship and characteristics between the data so as to accurately predict air pollutants.(3)limited by data sources,data sources mainly rely on monitoring sites.In recent years,with the application of deep learning in various fields,the advantages of deep learning have been fully demonstrated.Deep learning technology can process massive data and fully explore the relationship between the data,so as to improve the accuracy of prediction,which solves the shortcomings of traditional methods in processing large amounts of historical data.In order to better predict the concentration of air pollutant,image can be used for auxiliary prediction,which is an important application of Internet of things technology and mobile Internet technology in the field of pollution prediction,thereby helping to get rid of the limitation of fixed monitoring sites.Aiming at the difficulties of traditional pollutant concentration prediction methods and the advantages of deep learning,this paper proposed a joint air pollutants prediction model based on the multi-modal data.The main research work is as follows:(1)In order to improve the accuracy of the numerical prediction model and explore the inherent relationship between pollutant data.Our proposed the numerical prediction model based on Auto Encoder and Bi LSTM.The numerical prediction model mainly predicts the concentration of air pollutants PM2.5 through the data obtained from monitoring stations.The input data includes not only air pollutant PM2.5 data,but also other pollutant data and meteorological data,so as to solve the problem of big data of pollutants,fully explore the internal relationship between pollutant data and improve the accuracy of air pollutant concentration prediction.(2)In order to solve the limitation of pollutant data source,image is used for auxiliary prediction to help get rid of the limitation of fixed monitoring sites.Our proposed the integrated convolutional neural network(CNN)and LSTM image prediction model was studied.It can not only effectively solve the problem of historical data sources,but also mine more useful information through images to improve the prediction of air pollutant concentration.(3)Combining the numerical prediction model with the image prediction model,our proposed a joint air pollutants prediction model based on the multi-modal data,improve the accuracy of air pollutant prediction,has a certain degree of innovation.