基于数据预处理和支持向量回归的PM_2.5、NO₂和SO₂组合预测研究

Research on Combined Forecasting of PM_2.5, NO₂and SO₂ Based on Data Preprocessing Technology and Support Vector Regression

摘要: 由于空气污染能够引起多种呼吸道非传染性疾病，甚至造成生命质量受损或者过早死亡，所以空气污染已经成为全球死亡人数增加的第二大原因，因此，制定合理且有效的空气污染治理措施已经迫在眉睫。预测在空气污染预警过程中扮演着重要的角色，准确且科学的预测能够帮助人们有效规避空气污染的危害，因此，提高预测的精度与科学性也成为众多学者关心的问题之一。本研究为了提高空气污染预测的准确性，采用了数据预处理互补集合经验模式分解(complementary ensemble empirical mode decomposition, CEEMD)技术、支持向量回归(support vector regression, SVR)、广义回归神经网络(general regression neural network, GRNN)和粒子群优化(particle swarm optimization, PSO)算法建立组合模型。通过PM_2.5、NO₂和SO₂时间序列数据检验建立的组合预测模型的有效性，根据平均绝对百分比误差(mean absolute percentage error, MAPE)发现：组合预测模型能够提高PM_2.5、NO₂和SO₂指标的预测精度，如西安市的SO₂指标，最优单项模型的MAPE值为6.13%，而组合模型的MAPE值为5.86%。总之，组合预测模型能够为空气污染治理提供更准确的预测信息，为空气污染的防控提供理论支持。

Abstract: Because air pollution can cause a variety of respiratory non-communicable diseases, and even lead to impaired quality of life or premature death, the air pollution has become the second leading cause of the increase in deaths worldwide. Therefore, it is extremely urgent to formulate reasonable and effective air pollution control measures. Prediction plays an important role in the early warning process of air pollution. Accurate and scientific predictions can help people avoid the hazards of air pollution effectively. Therefore, how to improve the accuracy and scientificity of prediction has become one of the concerns of many scholars. To improve the accuracy of air pollution prediction, this study adopts data preprocessing technology complementary ensemble empirical mode decomposition (CEEMD), support vector regression (SVR), general regression neural network (GRNN), and particle swarm optimization (PSO) algorithms to establish a combined model. By PM_2.5, NO₂ and SO₂ index time-series data set checking the validity of combined forecasting model, according to the mean absolute percentage error (MAPE), it is found that the combined model can improve the prediction accuracy of PM_2.5, NO₂ and SO₂ index. For example, the MAPE of the optimal individual model and the combined model of the SO₂ index is 6.13% and 5.86% respectively in Xi’an. In short, the combined prediction model can provide more accurate prediction information for air pollution control and provide theoretical support for air pollution prevention and control.