A simulation study of ρ(PM2.5)in Lanzhou City proper based on deep learning methods
In order to simulate the ρ(PM2.5),an important pollutant,more accurately and quickly,three deep learning models were constructed,i.e.deep neural network(DNN),long and short-term memory recurrent neural network(LSTM),and convolutional neural network;at the same time,the monitoring da-ta of Lanzhou City meteorological station,inventory of air pollutant emissions and the conventional state-controlled stations of ambient air quality monitoring were used.The pollutant monitoring data,the hour-by-hour ρ(PM2.5)in Lanzhou City proper,were simulated.The results showed that the DNN model was the most effective of the three models at the yearly scale,and the LSTM model performed better than the other two in simulating data,with larger measured values.On the seasonal scale,the simulation of differ-ent seasons was better than simulation using year-round data,and the LSTM model performed better than the other three models.On the whole,the simulation results were better in spring,summer and fall,but worse in winter.
deep learningρ(PM2.5)emission inventorymeteorological factor
万方数据
