摘要
As a kind of particulate matter which is very detrimental to human health, the prediction of PM2.5 has attracted increasing attention in air pollution forecasting and early warning. In this paper, density peaks clustering (DPC) is employed to select the individual models for combination forecasting, and a novel high-precision combination forecasting model named DPC-based combined model is proposed to predict PM2.5 concentrations. Firstly, CEEMD is introduced to decompose the original PM2.5 data into three intrinsic mode functions (IMFs) with different features. Then, 125 individual models are obtained by applying SVR and GRNN to these three IMFs, and the intelligent optimization algorithms are adopted to optimize the parameters of SVR. Finally, the 125 individual models are clustered by DPC based on five model evaluation indexes to select the individual model for combination forecasting. The experimental results of four Chinese cities demonstrate that the proposed model outperforms the optimal individual model and can greatly improve the prediction accuracy. Taking Dalian as an example, the MAE of the proposed model is 2.2 by clustering the 125 individual models into two groups, and the result reveals that the accuracy is increased by 65.73% compared with the minimum MAE (6.42) of the individual models.
学科与关键词
- A Combined Model Based on Feature Selection and WOA for PM2.5 Concentration ForecastingZhao, Fang; Li, Weide*Atmosphere, 2019, 10(4): 223.被引 21
- Short-term PM2.5 forecasting based on CEEMD-RF in five cities of ChinaLiu, Da; Sun, Kun*Environmental Science and Pollution Research, 2019, 26(32): 32790-32803.被引 35
- Hourly PM2.5 concentration forecasting based on mode decomposition-recombination technique and ensemble learning approach in severe haze episodes of ChinaSun, Wei; Li, Zhaoqi*JOURNAL OF CLEANER PRODUCTION, 2020, 263: 121442.被引 42
