Using dynamic recurrent neural network NAR for predicting monoxide carbon concentration

Air pollution is one of the most important problems in big cities. One of the goals of urban managers is their awareness on air pollution in the future. For prediction of air quality, air pollutant must be modeled first. Carbon monoxide is one of the most toxic air pollutants that has harmful effect on human health.
In this paper, modeling carbon monoxide concentration and 24-h prediction by ARMA and NAR neural network have been studied. Then, the results of the two methods are compared. For this purpose, data is collected on 29 November until 31 December 2009 in Azadi air quality monitoring station: belonged to Tehran department of environment.
Findings: The results of the two methods showed that, NAR is more accurate than ARMA for modeling and prediction of carbon monoxide. NAR neural network had MSE=1.6 and a correlation coefficient of 0.84 while ARMA had MSE=5.46 and correlation coefficient=0.72 for 24 hours prediction.
Discussion and Finally, the predicted values can be used and published in internet for public awareness. Also urban managers can use the results of modeling and prediction for a better management. Result of this paper showed NAR neural network has sufficient ability to model and predict time series of monoxide carbon