Evaluating Visibility Range on Air Pollution using NARX Neural Network

Abstract

Evaluating air visibility range is considered as one of the apparent criteria of air quality. Haze air as a conclusion of air pollution causes unpleasant breathing, psychological effects, and visibility restriction. In this study, NARX neural network applied to determine air visibility restriction factors. Data of air quality control stations of Baghshomal, Rastebazar, and Abresan in Tabriz City, Iran used which include PM2.5, PM10, NO2, SO2, O3, and CO for the duration of four years from 2013 to 2017 that considered as independent variables. NARX neural network created to find each pollutant relation to visibility restriction and networks used for simulation to analysis network results in conspectuses condition. The results showed that PM10 pollutant has the most influence on-air visibility with R=0.9 in the train, R=0.728 in the test, and R=0.75 in validation process. Also error results of the PM10 obtained as MSE=0.054. Moreover, simulation results demonstrated the least area integral between curves according to ascending order for six pollutant factors and verified PM10 accuracy in NARX network simulation. The total result as study conclusion verified NARX neural network efficiency to evaluate air visibility range while using air pollutant parameters.