Prediction of Carbon Monoxide Concentration in Tehran using Artificial Neural Networks

Nowadays, air pollution is one of the most important problems almost all over the world. There are many strategies to control and reduce air pollution, one of which is prediction of this event and getting ready to deal with the negative effects of it. The aim of this study is to provide a multi-layer structure of artificial neural networks (ANN) for predicting of carbon monoxide pollution at subsequent 24 hours in Tehran metropolis.
To predict the amount of CO emissions in near future (subsequent 24 hours), wind speed and direction, temperature, relative humidity, and barometric pressure characteristics are used as meteorological data, and concentration of carbon monoxide is considered as a pollution parameter. To eliminate the noise of data, wavelets transform method and determining the threshold with normal distribution are used before training the ANN. Finally, two neural networks as two general models are proposed and used for modelling.
Findings: The results show that the correlation coefficient, index of agreement, accuracy of prediction, and root mean square error for model no. 1 with duplicate data are 0.9012, 0.915, 0.848, and 0.1012 and for model no. 2 with duplicate data are 0.9572, 0.978, 0.963, and 0.0385 respectively. Moreover, the results of listed parameters for model no. 1 with new data are 0.9086, 0.89, 0.885, and 0.0825 and for model No. 2 with new data are 0.8678, 0.928, 0.932, and 0.1163 respectively.
Results showed that there is a good agreement between predicted and observed values, hence the proposed models have a high potential for air pollution prediction.