Short-term air quality forecasting model based on hybrid RF-IACA-BPNN algorithm

Abstract

Despite the apparent improvement in air quality in recent years through a series of effective measures, the concentration of PM2.5 and O3 in Chengdu city remains high. And both the two pollutants can cause serious damage to human health and property; consequently, it is imperative to accurately forecast hourly concentration of PM2.5 and O3 in advance. In this study, an air quality forecasting method based on random forest (RF) method and improved ant colony algorithm coupled with back-propagation neural network (IACA-BPNN) are proposed. RF method was used to screen out highly correlated input variables, and the improved ant colony algorithm (IACA) was adopted to combine with BPNN to improve the convergence performance. Two datasets based on two different kinds of monitoring stations along with meteorological data were applied to verify the performance of this proposed model and compared with another five plain models. The results showed that the RF-IACA-BPNN model has the minimum statistical error of the mean absolute error, root mean square error, and mean absolute percentage error, and the values of R2 consistently outperform other models. Thus, it is concluded that the proposed model is suitable for air quality prediction. It was also detected that the performance of the models for the forecasting of the hourly concentrations of PM2.5 were more acceptable at suburban station than downtown station, while the case is just the opposite for O3, on account of the low variability dataset at suburban station.