Application of neural networks to the prediction of gas pollution of air

Abstract

The issue of projecting the air pollution levels is quite essential from the viewpoint of the necessity to adopt specific prevention measures intended to reduce the pollution concentration in the air. One can apply certain machine learning methods, including neural networks, to build pollution concentration models. Neural networks are characterised by the fact that they can be used to solve the relevant problem when we face shortage of data, or we do not know the analytical relationship between input and output data. Consequently, neural networks can be applied in a number of problems. This paper discusses a possibility to apply neural networks to the prediction of selected gas concentrations in the air, based on the data originating from the measurement networks of the Polish State Environmental Monitoring System, combined with local meteorological data. Forecast results have been presented here for SO2, NO, NO2, and O3 in various locations. The author also discusses the accuracy of the respective forecasts and indicates the relevant contributing factors.