PSO based Deep Learning Model for Forecasting PM2.5.

Abstract

Entire eco system and all the living organisms in the earth are affected by the toxic air pollutants to a great extent day by day. It is essential to forecast air pollution in order to limit the pollutant concentration and maintain the Air Quality standards prescribed by the government. In this paper PSO optimized 1D CNN and BIGRU are applied to predict accurately the fine particulate matter (PM2.5) pollutant which triggers various death causing diseases when its level exceeds the prescribed limit. The influence of meteorological parameters on air pollution cannot be omitted when doing the prediction analysis. UCI Machine Learning Repository Beijing PM2.5 time series dataset along with meteorological attributes are taken for this analysis.  Proposed  PSO based CNN-BIGRU model prediction results achieves perfect prediction performance than the existing deep learning convolutional-based bidirectional gated recurrent unit short term forecast of PM2.5 model. The proposed models RMSE, MAE, SMAPE are relatively low with the existing model.