Adaptive neuro-fuzzy modeling for prediction of ambient CO concentration at urban intersections and roadways

This paper describes development of neuro-fuzzy models to predict 1-h average carbon monoxide (CO) concentrations. The neuro-fuzzy models are developed at two urban air quality control regions in the Delhi city, India-one, a typical traffic intersection (i.e., Income Tax Office, ITO); and second, an urban straight roadway (i.e., Siri Fort). ITO models have been validated at Siri Fort site and Siri Fort models have been validated at ITO site. The models have been developed for 'heterogeneous' traffic conditions and 'tropical' meteorology having three meteorological and one traffic characteristic variables as their inputs. Two-year data, from 1 January 1997 to 31 December 1998 has been used to train the model. The models have been tested using data from 1 January to 31 December 1999 at the same site and then validated using data set from 1 January 2004 to 30 June 2005 at different sites. Prediction performances of 1-h average neuro-fuzzy CO models have been found to be satisfactory with predictions accuracy varying from 89% to 93%.