Comparison of Two Modeling Methods for the Prediction of Carbon Monoxide Concentration Using Neuro-Fuzzy System

Monitoring and forecasting of air quality Parameters in urban areas is considered as one of the challenges of the human environment. It depends on several factors such as topography, climate, population and transport network which the interaction of these spatial factors has been as a dynamic phenomenon, non-linear and ambiguous. In this study, two models suggested to predict and modeling concentrations of carbon monoxide (CO) pollutant using neuro-fuzzy system and GIS. In the first model the training data which is created by Kriging. An area was considered for each station and the data in that area was used for training. Fuzzy rules were extracted for each area and applied to each pixel of the region for the concentration estimation of the pollutant. In the second model, each station is trained with its own data separately. Fuzzy were extracted for each station and pollutant concentration was estimated as well. Having concentration predictions at station points, Kriging was used to model the spatial concentration. The data gathered from different meteorological stations in Tehran is used to train the neural network. In the first model, average RMSE of all stations for ANFIS is 1.613 ppm and for M-ANFIS is 1.484 ppm and in the second model average RMSE of all stations for ANFIS is 1.445 ppm and for M-ANFIS is 1.374 ppm. The results showed that both models have a good capability of concentration prediction of the pollutant.