Research Of Urban Air Quality Index Prediction Based On Deep Learning

With the development of urbanization and the expansion of industrialization,the emission of various harmful substances leads to the problem of air pollution,which seriously affects the health and daily life of urban residents.Air quality monitoring stations have been set up in many cities,and people are increasingly concerned about urban air quality,as well as monitoring,there has also been an increase in the need to predict future air quality.Aiming at the phenomenon of missing values in air pollutant data and the single structure of prediction model,at the same time,the impact of rapid meteorological changes on air quality is considered.This paper mainly studies short-term(hourly)air quality index prediction,a combined deletion processing method is proposed and an air quality index prediction model based on deep recurrent neural network(GRU-BPNN)is constructed.The main work includes:1)Through an intensive study of the air quality index(AQI),the data used in this paper are no longer the concentrations of air pollutants obtained from air quality monitoring stations,but based on the definition of the air quality index and how it’s calculated,the concentration value of each pollutant is converted into the individual air quality index(IAQI)of each pollutant project through the environmental standard,such data transformation is beneficial for subsequent data processing.2)For some air quality monitoring stations,the data of a certain period of time will be missing to different degrees for some reason.The data characteristics of air quality index are ignored in the current data filling process and the lack of data processing method is relatively single,through the analysis of missing data attributes and the length of missing time,in this paper,a combined deletion algorithm of individual air quality index is proposed.3)On the basis of combined deletion processing algorithm,recurrent neural network is selected and a time series sample is generated from air quality data as input,the effects of meteorological factors on air quality are also considered,by analyzing the correlation between air quality index data and meteorological data,the meteorological sequence data with relatively high correlation coefficient is introduced as input to BP neural network to optimize the prediction results.In this paper,6 kinds of deep recurrent neural networks and9 kinds of single network structure models were constructed for training,so as to achievethe prediction of air quality index and the accuracy evaluation of the predicted results of the AQI grade.The experiment results show that the proposed combination of mean substitution and moving average method has achieved good results,the prediction results of the integrated deep recurrent neural network are better than that of the single network structure modle,among them,the prediction model based on MA-GRU-BPNN has the highest prediction accuracy for air quality index.