Prediction of Air Quality in Smart Cities Using Deep Learning Techniques

Abstract

Air pollution has developed as a serious and potentially fatal anxiety in various nations throughout the world in recent decades, owing mostly to human activity, industry, and urbanization. Substantial Particulate Matter having a Diameter of 2.5m (PM2.5) is a particularly dangerous component of air pollution that causes major health hazards, including respiratory and cardiovascular disorders. As a result, precisely forecasting PM2.5 levels is critical in order to protect people from the negative effects of air pollution. PM2.5 levels are impacted by a number of factors, such as climatic conditions and the quantity of other contaminants in metropolitan areas. In this study, we used a DL (Deep Learning) technique, especially (CLARP) CNN-LSTM-Attention Mechanism-Recurrent Mechanism-Pooling Mechanism, to anticipate the hourly PM2.5 concentration in Beijing, China. Our model includes data fusion approaches that involve the merging of many data sources, such as historical pollutant data, meteorological data, and PM2.5 values, to provide more accurate estimates or projections. We compared the performance of numerous LSTM, Bi-LSTM, GRU, Bi-GRU, PM-GRU, RM-LSTM and a hybrid CLARP model. Based on experimental data, the CLARP Model technique outperformed all standard models tested, giving 99% R2 and improved results for RMSE & MAE emphasizing its expanded predictive capabilities.