Research And Implementation On Air Quality Prediction Method Deep Learning Based

In recent years,with the rapid urbanization,air pollution problems have occurred frequently,which has a great impact on people’s physical health and normal life.Therefore,accurate prediction of urban air quality is of great significance.In this thesis,deep neural networks are used to study and predict the concentrations of PM2.5 and PM10 in Beijing.Firstly,according to the spatial correlation between air pollutants,this paper proposes a space conversion method,which divides the area around the target detection station by spatial division,domain aggregation and domain interpolation,so that each area can be obtained.Air quality data and meteorological data in the same format ultimately transform spatially sparse air quality data into uniform consistent inputs,extracting features between spatial data.Then,based on RNN,this thesis constructs the basic structure of the prediction model using GRU network,and adds Relu activation function,random deactivation and batch normalization to optimize the deep learning model.Improve.On top of this,the spatial transformation method and the deep neural network are used to fuse the spatial data with the time domain data to obtain a new fusion model.Finally,this thesis constructs four models to compare the prediction results.The experimental results show that the proposed method not only combines the air correlation between air pollutants,but also weakens the error caused by regional factors on air quality prediction.Compared with other models,the accuracy is greatly improved.