An Atmospheric Particulate Matter Forecast Model Based on Time Series Data

In recent years, air pollution becomes an increasing concern globally because it directly affects people’s health and daily life. Especially the particulate matter with diameters less than 2.5 micrometers (PM2.5), which is one of the most common air pollutants. Air quality forecast helps to give early warnings and prevent the effects of air pollution. Effective air quality forecast has become one of the hot research issues. In this paper, the advantages of the existing prediction algorithms are analyzed and compared, and then the Long Short Term Memory (LSTM) network was applied to the research of atmospheric particulate matter forecast, and an atmospheric particulate matter prediction model based on time series data was constructed. The model was implemented based on the TensorFlow deep learning framework, Keras neural network library and Python language. It was tested on 365 daily mean concentration data and 22,287 hourly concentration data, and the prediction results were visualized. Finally, the established model was evaluated by mean square error (MSE). The experimental results demonstrate that the proposed prediction model can achieve better prediction performance for the PM2.5 concentration data even if it has a simple network structure and sample data of single factor.