The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)
Daily PM2.5 concentration is very uncertain and unstable due to multiple factors of local and adjacent areas. The common PM2.5 real-valued and interval series reflect the daily and extreme value fluctuations respectively, while triangular fuzzy series combined with both advantages and include more effective information. Therefore, this paper proposes a multi-factor combination forecasting model of PM2.5 triangular fuzzy series based on Multiple Empirical Mode Decomposition and spatial hierarchical clustering. Firstly, the Pearson correlation coefficient is used to analyze the correlation between PM2.5 concentration and local pollutant concentration and meteorological elements, so as to select the local impact factors. Secondly, the Pearson correlation between PM2.5 and spatial pollutant concentration is calculated. Based on this, we obtain the urban agglomerations with core impact, general impact and remote impact by the K-means spatial clustering of neighboring cities. Thirdly, The comprehensive index of different pollutants in each urban agglomeration is obtained, which is used as the spatial impact factors. Then, the triangular fuzzy sequences of PM2.5 and influence factors are decomposed simultaneously by MEMD, and the high frequency, low frequency and trend sequences are reconstructed. Finally, BP, LSTM and LSSVR are employed to predict the subsequences respectively, and the predicted value of PM2.5 triangular fuzzy series is obtained by adding the above single prediction results. The simulation results show that the proposed model can effectively utilize the meteorological conditions and the spatial effects of various pollutants, and has strong predictive performance and good practicability.