Abstract：Practical analysis and forecasting of PM2.5 concentrations is complex and challenging owing to the volatility and non-stationarity of PM2.5 series. Most previous studies mainly focused on deterministic predictions, whereas the uncertainty in the prediction is not considered. In this study, a novel uncertainty analysis–forecasting system comprising distribution function analysis, intelligent deterministic prediction, and interval prediction is designed. Based on the characteristics of PM2.5 series, 16 hybrid models composed of various distribution functions and swarm optimization algorithms are selected to determine the exact PM2.5 distribution. Subsequently, a hybrid deterministic forecasting model based on a novel decomposition–ensemble framework is established for PM2.5 prediction. Regarding uncertainty analysis, interval prediction is established to provide uncertain information required for decision–making based on the optimal distribution functions and deterministic prediction results. PM2.5 concentration series obtained from three cities in China are used to conduct an empirical study. The empirical results show that the proposed system can achieve better prediction results than other comparable models as well as provide meaningful and practical quantification of future PM trends. Hence, the system can provide more constructive suggestions for government administrators and the public.

Keywords：artificial intelligence algorithm, interval prediction, PM2.5 prediction, swarm optimization algorithm, uncertainty analysis

基金资助：This work was supported by the National Natural Science Foundation Project of China (72004086), Fundamental Research Funds for the Central Universities (22lzujbkydx011) and Ministry of Education philosophy and social science research major project (21JZD037).

原文链接：https://onlinelibrary.wiley.com/doi/epdf/10.1002/for.3005