Air Pollution Forecast Model Based on LSTM

Air pollution becomes a severe obstacle which restricts the development of urbanization in developing countries. In order to predict the concentration of PM2.5 accurately, in this research, a PM2.5 prediction model, named Air-S2S-LSTM, is constructed based on Seq2seq. LSTM, that is long short-term memory neural network, is imported into this model to study the historical meteorological data. Experiments show that the model performs better accuracy in different time span prediction task. The RMSE and MAE of s2s-LSTM model are reduced by 57.49%, 53.35%, 32.835% and 60.52%, 50.95%, 58.16% while comparing with ARIMA, Holt-Winters, and SVR respectively. What’s more, results of Pearson correlation coefficient and principal component analysis indicated that dew point, wind speed, air temperature, air pressure, rainfall in all characteristics have a predominant influence upon the concentration of PM2.5.