Long Short-Term Memory based PM2.5 Concentration Prediction Method.

Air pollution is now a serious problem in China and elsewhere, makes a significant impact on human life. Air quality prediction is one of the research fields in environmental protection, which helps people to plan their lives reasonably and avoid pollution exposure. Considering that the atmospheric pollutants and meteorological data are ultimate time-series data, this paper proposed a prediction model based on LSTM which has low implementation complexity and an open cost of production. We selected atmospheric pollutants (SO2, PM10, O3, etc.), as well as meteorological parameters (temperature, atmospheric pressure, rainfall, etc.) as input of training model, these factors from the first few hours are used to forecast PM2.5 concentration in the next few hours. We evaluated our model on a large dataset of air pollution records and grid meteorological data in Qingdao. The results show that the proposed Long Short-Term Memory (LSTM) network generally has the marginal error compared with baselines, which indicates that the proposed LSTM network approach for PM2.5 forecast is effective and robust. [ABSTRACT FROM AUTHOR]