Air pollution modelling from meteorological parameters using artificial neural network

Abstract

The aim of this study is to develop neural network air quality prediction model for PM10 (particle whose diameter is less the 10 µm), NO2 and SO2. A multilayer neural network model with a hidden recurrent layer is used to predict pollutant concentrations at four monitoring sites in Belagavi city of Karnataka State, India. The Levenberg Marquardt algorithm is used to train the network. A combination of input variables were investigated taking into the predictability of meteorological input variables and the study of model performance. The meteorological variables air temperature, wind speed, wind direction, rainfall and relative humidity were considered as input variables for this study. The results show very good agreement between measured and predicted pollutant concentrations. The performance of the developed model was assessed through performance index. The models developed have good prediction performance (>85%) for all the pollutants. The proposed models were predicted pollutant concentration with relatively good accuracy and outputs were proven to be satisfactory by measuring of the goodness of fit and by mean absolute percentage error.