The Use of Artificial Neural Network (ANN) for Prediction of Some Airborne Pollutants Concentration in Urban Areas

The quality of atmospheric air has become a major priority for world wide health. The rise of human illness, because of pollution, leads to the gathering of new strategies for the levels of integrated systems of environmental management, which consider a gradually control, analyzes and evaluation of the noxious airborne compounds. The present paper subscribes to this concerti and proposes a model for concentration prediction of inorganic airborne pollutants: H(2)S-SO(2), NO-NO(2)-NO(x), CO (CO(2)) and PM(10) (particle matter with an aerodynamic diameter of 10 mu m or less) front a risk area two industrial area - IA) and an urban area (UA) from Constanta. The model uses one of the newest methods of nonlinear function approximation, the Artificial Neural Network (ANN). ANNs are used for the systems of phenomenon for which the linearity between different variables can not be determined or approximated. Actually the ANN can simulate this phenomenon, and in the present paper, values of the pollutants concentration from AU will be provisioned. The present ANN uses for training a small number of variables and a large number of data (measured values). For the development and validation of the model it is necessary to have an adequate and continuous monitoring system for data, for the analyzed chemical phenomenon. The main sources of emissions for different airborne pollutants in the analysed urban area are: road transport, stationary combustion processes and industrial processes from the specific geographic area. Inorganic airborne pollutants concentrations were measured with different instruments (Chemiluminescence NO-NO(2)-NO(x), analyzer, Pulsed fluorescence H(2)S-SO(2) analyzer, gas filter correlation CO analyzer and EPAM 5000 instrument - Portable Environmental Particulate Air Monitor for PM(10)) between January 2006 and August 2007. The comparison between the results from real measured data from urban area and the result of simulated values provided a small maximum absolute error of 0.42. 10(-4) that demonstrates the efficiency and validity of the proposed method in the evaluation of different airborne pollutants emissions.