A Long Short-Term Memory Neural Network for Daily NO2 Concentration Forecasting

Literature Review

The existing research methods for air prediction are mainly divided into physics-based and statistical methods. The physics-based models include dispersion models (Fallah-shorshani et al., 2017) and chemical transport models (Menut et al., 2012), among others. However, due to the uncertainty in the physicochemical properties of air pollutants (Kim et al., 2010; Pisoni et al., 2018), the quality and scale of emission values can still largely affect the accuracy of prediction results even if physics-based models can simulate some components of air (Xiao et al., 2015; Wei et al., 2013), it is difficult to predict air pollutant concentrations accurately. Statistical methods usually construct models for prediction based on a large amount of historical air quality data and statistical correlations between objective factors related to them (Yuval and Broday, 2013; Mihăiţă et al., 2019). Statistical methods can be divided into statistical models, machine learning models, and deep learning models.