Adaptive kernel fuzzy weighted particle swarm optimized deep learning model to predict air pollution PM2.5.

Abstract

Precise prediction by deep learning algorithms will help regulate the pollutant PM2.5 (particulate matter, inhalable particles with diameter 2.5µm micrometres or smaller) a global threat to the entire environment by adulterating the air causing respiratory, cardiovascular, numerous other diseases and long term exposure to the pollutantin turn leads to human mortality in the globe. A combination of both 1D CNN and BIGRU model used to forecast the menacing pollutant PM2.5. For extemporizing proposed model accuracy, precise choice of hyper parameter selection is inevitable. Hyper parameters of ID CNN and BIGRU are automatically selected by a novel Adaptive kernel fuzzy weighted particle swarm optimization (AKFWPSO).Both PM2.5 and meteorological hourly data set of Beijing from UCI Machine learning repository is exploited for this analysis. For measuring model performance three measurement approaches such as RMSE, MAE and SMAPE are used. The model accuracy is considered superior comparing theexisting model with estimation of error metrics.This model can be applied not only to oversee and regulate the PM2.5but also alert the public when the amount of the pollutant level goes beyond the level prescribed.