PM2.5 concentration prediction is one of the atmospheric environmental issues of great concern to the public, and specifically the long-term PM2.5 change prediction can provide scientific basis for the government’s energy conservation and emission reduction and industrial structure adjustment policies in advance. This paper proposes a new dynamics forecasting approach suitable for small samples and the approach transforms the time series prediction into a dynamics system through the ordinary differential equation theory, which overcomes the limitation of traditional statistical methods on sample size. What is more important is that it not only includes the time series itself, but also uses prior information in the modeling process. Based on the dynamics forecasting approach, the phase objectives analysis model for PM2.5 reduction is constructed. The simulation experiment takes 11 prefecture level cities in Shanxi Province as research sites, and uses the annual PM2.5 concentration from 2014 to 2018 to verify whether the proposed model can significantly improve the fitting accuracy compared with the single SVM. The experimental results show that the MAE (mean absolute error) of Taiyuan site is reduced by 73.24% from 1.6030 of SVM model to 0.4290 of our proposed method. Similar conclusions can be obtained from other data sets, which considerably demonstrates the better generalization ability of the proposed model. In addition, this paper presents the forecast results of annual PM2.5 concentration in different PGDP (energy consumption per unit of GDP) scenarios from 2019 to 2023, and analyzes the impact of PGDP reduction on PM2.5 concentration. Among the three scenarios, the PM2.5 reduction is the most significant in the scenario of PGDP with 10% decrease. We would argue that a larger PGDP reduction might lead to a greater PM2.5 reduction. Therefore, PGDP can be used as the basis for the Chinese authorities to tailor strategies to reduce PM2.5 concentration.
Full Text Finder