College of Resources and Environment, Shanxi University of Finance and Economics, Wucheng Road, Taiyuan, 030006, Shanxi, People's Republic of China.
School of Economics and Management, Shanxi University, Wucheng Road, Taiyuan, 030006, Shanxi, People's Republic of China.
Environ Sci Pollut Res Int. 2023 Jul;30(34):82878-82894. doi: 10.1007/s11356-023-26834-4. Epub 2023 Jun 19.
In recent years, the frequent occurrence of air pollution incidents has seriously affected people's health and life. Therefore, PM[Formula: see text], as the main pollutant, is an important research object of air pollution at present. Effectively improving the prediction accuracy of PM[Formula: see text] volatility makes the PM[Formula: see text] prediction content perfect, which is an important aspect of PM[Formula: see text] concentration research. The volatility series has an inherent complex function law, which drives the volatility movement. When machine learning algorithms such as LSTM (Long Short-Term Memory Network) and SVM (Support Vector Machine) are used for volatility analysis, a high-order nonlinear form is used to fit the functional law of the volatility series, but the time-frequency information of the volatility has not been utilized. Based on EMD (Empirical Mode Decomposition) technique, GARCH (Generalized AutoRegressive Conditional Heteroskedasticity) model and machine learning algorithms, a new hybrid PM[Formula: see text] volatility prediction model is proposed in this study. This model realizes time-frequency characteristic extraction of volatility series through EMD technology, and integrates residual and historical volatility information through GARCH model. The simulation results of the proposed model are verified by comparing the samples of 54 cities in North China with the benchmark models. The experimental results in Beijing showed that MAE (mean absolute deviation) of hybrid-LSTM decreased from 0.00875 to 0.00718 compared with LSTM, and hybrid-SVM based on the basic model SVM also significantly improved generalization ability, and its IA (index of agreement) improved from 0.846707 to 0.96595, showing the best performance. The experimental results show that the hybrid model is superior to other considered models in terms of prediction accuracy and stability, which verifies that the hybrid system modeling method is suitable for PM[Formula: see text] volatility analysis.
近年来，空气污染事件频发，严重影响了人们的健康和生活。因此，PM[Formula: see text]作为主要污染物，是当前空气污染的重要研究对象。有效提高 PM[Formula: see text]挥发度的预测精度，使 PM[Formula: see text]预测内容更加完善，是 PM[Formula: see text]浓度研究的重要方面。挥发度序列具有内在的复杂函数规律，驱动着挥发度的运动。当使用 LSTM（长短期记忆网络）和 SVM（支持向量机）等机器学习算法进行挥发度分析时，采用高阶非线性形式来拟合挥发度序列的函数规律，但并未利用挥发度的时频信息。基于 EMD（经验模态分解）技术、GARCH（广义自回归条件异方差）模型和机器学习算法，本文提出了一种新的混合 PM[Formula: see text]挥发度预测模型。该模型通过 EMD 技术实现了挥发度序列的时频特征提取，并通过 GARCH 模型集成了残差和历史波动率信息。通过与基准模型比较华北 54 个城市的样本对所提出模型的模拟结果进行验证。北京的实验结果表明，与 LSTM 相比，混合-LSTM 的 MAE（平均绝对偏差）从 0.00875 下降到 0.00718，基于基本模型 SVM 的混合-SVM 也显著提高了泛化能力，其 IA（一致性指数）从 0.846707 提高到 0.96595，表现出最佳性能。实验结果表明，混合模型在预测精度和稳定性方面优于其他考虑模型，验证了混合系统建模方法适用于 PM[Formula: see text]挥发度分析。
