PREDICTION OF AMBIENT AIR POLLUTANTS CONCENTRATIONS by ARTIFICIAL NEURAL NETWORK IN KOTA CITY

Abstract

In many cities, the level of air pollutants has already crossed the tolerance newlinelimit of human exposure of individual pollutants. To keep the situation in control newlinethere is an urgent need for careful pollution control measures in major megacities of newlinethe world not only for local environmental protection but also for regulation of newlinegreenhouse gas emission. This can be done only by air quality management which newlineworks on knowledge about pollution trends, spatial trends, sources trend, temporal newlinetrends and control trends. It is imperative to model the level of pollutants. The newlinemodeling of an atmospheric pollutant for a specific site involves the development of newlinefunctional relationships among input data and target data of prime pollutants that are newlineneeded to be modeled. As these parameters are site-dependent and vary from site to newlinesite arises the need for a site-specific model. Sulphur-dioxide, nitrogen dioxide and newlineparticulate matter are the pollutants which are monitored in almost all the places. newlineTherefore the trend and prediction of these pollutants in various cities could be newlineaccounted in study so far. The predictor models were developed for the prediction newlineof prime pollutants by taking the known sets of input and target datasets that evolves newlinepredictions for the response of input and output data. Daily data of three years from newline2012-2014 of a specific site of Kota city is taken into account to train, test and newlinevalidate four different topologies of the neural networks namely Feed Forward newlineNeural Network (FFNN), Layer Recurrent Neural Network (LRNN), Nonlinear newlineautoregressive Exogenous (NARX) and Radial Basis Function Neural Network newline(RBFN). A meaningful comparison between these topologies discovered that RBFN newlineis a suitable topology for prediction. .The worst prediction is performed by LRNN newlinetopologies as the predicted values by this topology are placed quite far. For better newlineprediction and accurate results the meteorological parameters have also been taken newlineinto account as they play significant roles in air pollution.