Prediction Of Air Pollutant Distribution Based On Deep Neural Network

Air pollution is one of the most serious threats to human health,so it is imperative to provide more accurate air quality forecasts.With the development of data-driven modeling technology,the neural network is used to simulate the transport process of atmospheric pollutants.In this paper,a hybrid model based on the deep learning method is constructed,which integrates Graph Convolutional network(GCN)and Long Short-Term memory network(LSTM)to predict the spatial and temporal changes of urban air po llutant concentration.To different sites of historical observation data structure for spatiotemporal graphic sequence,the ground air quality monitoring station data,the meteorological factor,space factor and time attribute is defined as a graphic signa l,Graph convolutional network is used to extract the spatial correlation between the observed values of different monitoring stations,and LSTM is used to capture the temporal correlation between the observed values at different times,the simulation of urban air pollution pollutants characteristics of space and time.In order to assess our hybrid model the performance of the concentration prediction of atmospheric pollutants in the city,we chose to use the Root Mean Square Error(RMSE),Mean Absolute Error(MAE),Recall Rate(RR)and False Alarm Rate(FAR)as evaluation scale,at the same time,Multiple Linear Regression(MLR)model and Feed-Forward neural network(FNN)model and taking into consideration the LSTM model as control group,The Gc-Lstm model in this paper shows that MAE value 13.82 and RSME value 21.45 are the optimal values,and the upper quartile and lower quartile of the prediction deviation value are within 9?g/m~3.The recall rate of 81.2%and the error rate of 2.2%based on the 115?g/m~3 threshold also prove the accuracy and feasibility of the model.