A model for predicting gaseous air pollutants in Tabriz using neural network

Abstract

Air pollution is still a serious environmental challenge in the world. The problem is much more severe in urban areas of developing countries, where it affects quality of life and public health It is usually caused by industrial activities, energy production by power plants, residential heating, fuel burning vehicles and natural disasters. Therefore, forecasting of air pollutants has become a popular topic of environmental research today. There are many methods for the forecasting of air pollutant concentrations. In recent years, there has been considerable progress in the development of neural network models for air pollution forecasting. In the current study, the artificial neural network technique is widely use as a reliable method for forecasting air pollutant in urban areas. On the other hand, the evolutionary polynomial regression(EPR) model has recently been used models to forecasting tool in some environmental issue. In this research, we compared the ability of these models to forecast carbon monoxide, sulfur dioxide and nitrogen oxides concentrations in the urban area of Tabriz city. The dataset of CO, SO2 and NOx concentrations used in this study were obtained from five fixed monitoring stations including S1, S2, S3, S4 and S5 operated by Tabriz Air Quality Control Co (TAQCC) from 2008 to 2012. A feed forward multi layer perceptron network was used to forecast CO, SO2 and NOx concentrations at the all stations mentioned above. The data were divided into three subsets, namely training (70-80% of all), validation (10- 15% of all) and a test set (10-15% of all). In this thesis the number of hidden layers varied between 1 and 4, too number of neurons in the hidden layers varied between 8 and 25. The results showed the performance of ANN is more reliable in comparison with EPR. using the ANN model, the correlation coefficient values at all monitoring stations were obtained 0.91, whereas using the EPR model the correlation coefficients of the polynomial evolution model 0.45.