PM2.5 forecasting using SVR with PSOGSA algorithm based on CEEMD, GRNN and GCA considering meteorological factors

• Published in: Atmospheric environment (1994) Vol. 183; pp. 20 - 32
• Main Authors: Zhu, Suling, Lian, Xiuyuan, Wei, Lin, Che, Jinxing, Shen, Xiping, Yang, Ling, Qiu, Xuanlin, Liu, Xiaoning, Gao, Wenlong, Ren, Xiaowei, Li, Juansheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2018
• Subjects: air pollution; air quality; algorithms; atmospheric chemistry; China; cities; climate; Gravitational search algorithm; Grey correlation analysis; landscapes; meteorological parameters; Particle swarm optimization; particulates; PM2.5 concentrations; pollutants; pollution control; respiratory tract diseases; Support vector regression; China; PM2.5 concentrations; Support vector regression; Gravitational search algorithm; Grey correlation analysis; Particle swarm optimization
• ISSN: 1352-2310; 1873-2844
• DOI: 10.1016/j.atmosenv.2018.04.004

The PM2.5 is the culprit of air pollution, and it leads to respiratory system disease when the fine particles are inhaled. Therefore, it is increasingly significant to develop an effective model for PM2.5 forecasting and warnings that informs people to foresee the air quality. People can reduce outdoor activities and take preventive measures if they know the air quality is bad ahead of time. In addition, reliable forecasting results can remind the relevant departments to control and reduce pollutants discharge. According to our knowledge, the current hybrid forecasting techniques of PM2.5 do not take the meteorological factors into consideration. Actually, meteorological factors affect the concentrations of air pollution, but it is unclear whether meteorological factors are helpful for improving the PM2.5 forecasting results or not. This paper proposes a hybrid model called CEEMD-PSOGSA-SVR-GRNN, based on complementary ensemble empirical mode decomposition (CEEMD), particle swarm optimization and gravitational search algorithm (PSOGSA), support vector regression (SVR), generalized regression neural network (GRNN) and grey correlation analysis (GCA), for the daily PM2.5 concentrations forecasting. The main steps of proposed model are described as follows: the original PM2.5 data decomposition with CEEMD, optimal SVR selection with PSOGCA, meteorological factors selection with GCA, residual revision by GRNN and forecasting results analysis. Three cities (Chongqing, Harbin and Jinan) in China with different characteristics of climate, terrain and pollution sources are selected to verify the effectiveness of proposed model, and CEEMD-PSOGSA-SVR*, EEMD-PSOGSA-SVR, PSOGSA-SVR, CEEMD-PSO-SVR, CEEMD-GSA-SVR, CEEMD-GWO-SVR are considered to be compared models. The experimental results show that the hybrid CEEMD-PSOGSA-SVR-GRNN model outperforms other six compared models. Therefore, the proposed CEEMD-PSOGSA-SVR-GRNN model can be used to develop air quality forecasting and warnings. [Display omitted] •Meteorological factors are studied for PM2.5 forecasting.•Effective information of PM2.5 is extracted by CEEMD.•The proposed CEEMD-PSOGSA-SVR-GRNN model is effective for PM2.5 forecasting.•The proposed theory can be used to effectively forecast other pollutions.