MODELLING OF ROADSIDE AIR QUALITY USING ARTIFICAL NEURAL NETWORK

Abstract

Over the past decade rapid urbanisation and development has led to steady increase in number of vehicles on road in Delhi, which has led to alarming increase in level of air pollution. Air pollution in Delhi has become a serious environmental problem in recent years. The concentration of pollutants like particulate matter (PM10,PM 2.5), nitrogen oxides (NOx) and carbon monoxide (CO) in ambient air have continuously exceeded the threshold limits specially in areas near arterial roads. The high concentration of toxic pollutants in ambient air are a silent and lethal killer .The poor air quality causes serious health ailments such as respiratory diseases, increase in risk of developing cancer, heart diseases and other serious ailments . This leads to tremendous loss of financial resources in form of medical expenses for treatment of affected people. According to data out of the total pollutants discharged in air in Delhi every day, vehicular emissions from transportation sector is the major contributor in the total air pollutant load .Air quality monitoring and prediction systems are the need of the hour for controlling the pollutant concentration to improve urban air quality of Delhi . The main objective of this study is to develop non-parametric Artificial Neural Network models to predict roadside air quality in Delhi for pollutants like PM10 and PM2.5,NOx and CO .Air quality prediction is usually carried out by soft computing techniques, fuzzy logic and generic algorithms . Artificial Neural Network (ANN), a soft computing technique has been steadily gaining popularity as an optimised air quality prediction tool among researchers over the past few years. The motivation for using Artificial Neural Network for modelling in this study stems from the capability and efficiency of ANN in computation of highly complex non-linear dynamic systems with large dimensional data. The proposed ANN models considered meteorological parameters like wind speed, gust wind, relative humidity, pressure, temperature and traffic characteristics as inputs and concentration of various pollutants as outputs. Different models were developed for short term prediction and their performance was evaluated on the basis of statistical parameters like MAPE, MAE, RMSE and coefficient of determination (R2). The weights established after training process were used to find out the relative influence of different vehicles on the pollutant concentrations.ANN modelling can be used as a pre-warning mechanism for air pollution episodes and help the policy makers in formulating suitable mitigation measures to curb pollution arising from vehicular exhaust emissions .