Air quality data series estimation based on machine learning approaches for urban environments

AbstractAir pollution is one of the main environmental problems in residential areas. In many cases, the effects of air pollution on human health can be prevented by forecasting the air quality in the next day. In order to predict the 1  day in advance air quality index (AQI) of Orumiyeh city, the hybrid single decomposition (HSD) and hybrid two-phase decomposition (HTPD) models were used. In the first step, the AQI data were decomposed by complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and was hy bridized with general regression neural network (GRNN) and extreme learning machine (ELM) as HSD models. In the second step, using variational mode decomposition (VMD) technique the results of the first intrinsic mode functions (IMFs) of CEEMDAN model were decomposed into nine VMs and were predicted by GRNN and ELM models to obtain IMF1. Finally, in the third step, GRNN and ELM were used again to predict the IMFS as HTPD models. Results showed that in predicting AQI series data by HSD models both CEEMDAN-ELM and CEEMDAN-GRNN models were similarly accurate. Among all the models used, the accuracy of CEEMDAN-VMD-GRNN as the HTPD model was the highest in the training phase (R2 = 0.98, RMSE = 4.13 and MAE = 2.99) and in the testing phase (R2 = 0.74, RMSE = 5.45 and MAE = 3.87). It can be concluded that HTPD models have more accurate results to predict AQI data compared with HSD models.
Source: Air Quality, Atmosphere and Health - Category: Environmental Health Source Type: research