Artificial Neural Network Model to Forecast the Concentration of Pollutants Over Delhi: Skill Assessment of Learning Rules

Abstract

Air pollution has been reported to persuade climate as well as health significantly and is a matter of concern. The scientific endeavour should thus, be to develop forecast or warning system to predict the concentration of pollutants with considerable accuracy so that the calamities associated with pollution can be minimized, if not eradicated. The purpose of the study is to develop an Artificial Neural Network (ANN) model with different learning rules to predict the concentration of pollutants over Delhi (28° 38'N, 77° 12'E), India for the year 2009. Two types of learning rules are implemented in this study to forecast the concentration of different pollutants. The result reveals that the forecast accuracy of a particular pollutant depends on the type of the learning rule of the ANN model. The result of the study further reveals that the non-linear perceptron is better for forecasting the concentration of sulphur dioxide (SO $_2$ ), carbon monoxide (CO), suspended particulate matter (SPM) and ozone (O $_3$ ) whereas delta learning is better for forecasting nitrogen dioxide (NO $_2$ ). The percentage errors in forecast with different learning rules of the ANN model are compared for all the pollutants. The result shows that the concentration of SO $_2$ can be predicted over Delhi with maximum accuracy using nonlinear perceptron.