Weather condition-based hybrid models for multiple air pollutants forecasting and minimisation. (10th June 2022)

• Record Type: Journal Article
• Authors: Wang, Chang; Zheng, Jianqin; Du, Jian; Wang, Guotao; Klemeš, Jiří Jaromír; Wang, Bohong; Liao, Qi; Liang, Yongtu
• Is Part Of: Journal of cleaner production. Volume 352(2022)
• Publication Date: 2022-06-10
• Subjects: Air pollutants forecasting -- Hybrid model -- Model fusion -- Spatiotemporal characteristics; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2022.131610
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: With the deterioration of air quality in recent years, the establishment of accurate and efficient forecasting models for pollutants has become the top priority. Due to the imperfect internal mechanism, the traditional numerical model, Weather Research and Forecast - Community Multiscale Air Quality (WRF-CMAQ), whose performance is limited in predicting the concentration of pollutants. To solve that issue presented, this study proposed two hybrid models for pollutant concentration forecasting based on weather conditions of various monitoring points. The hybrid model I applies long and short-term memory neural networks (LSTM) to extract the temporal characteristics and random forest (RF) to extract the non-line characteristics. Then, a fusion layer is built to combine them, which is optimised by the particle swarm optimisation (PSO) algorithm. Based on hybrid model I, hybrid model II also considers the regional synergy of different monitoring points to capture the spatial correlation of weather conditions. Taking a certain region of China as an example, the performance of these two hybrid models is proved. The results and discussions indicate that not only do the hybrid models achieve higher accuracy than other comparable models such as LSTM, convolutional neural network (CNN), and WRF-CMAQ, but they also prove that the regional synergy can significantly improve the effectiveness of air pollutants forecasting. The root mean squared error (RMSE) of the hybrid model II for predicted six pollutants concentration dropped to 1.781, 6.630, 5.556, 4.154, 49.558, 4.074 compared with the RMSE values of the hybrid model I and WRF-CMAQ, which are 1.972, 6.734, 6.731, 4.937, 63.487, 5.422 and 7.98, 38.175, 29.511, 21.077, 78.479, 22.810. This work provides the high-precision prediction and comprehensive evaluation of primary pollutants, which provides a targeting option to deal with the highest predicted pollutants.