稀疏组lasso罚向量自回归模型的大气污染物预测:京津冀案例研究

进行了大气污染物预测研究。针对传统的向量自回归模型方法所面临的过参数化问题,提出了稀疏组lasso罚向量自回归模型并应用近邻梯度下降法求解模型参数。为了验证模型的有效性,将其应用于2015年京津冀大气污染物数据中并对2016年1月1日北京6项大气污染物浓度进行预测。实验数据表明:基于稀疏组lasso罚模型的PM2.5预测归一化均方误差约为3.8%,预测精度高于向量自回归(VAR)模型、基于各种稀疏结构的向量自回归(VAR-L)模型、分层向量自回归(HVAR)模型。此外,京津冀不同城市对北京的空气质量影响程度不同,这可以通过组内稀疏模型参数进行解释。将凸优化概念与向量自回归模型结合应用于大气污染物浓度的预测中,对京津冀大气污染协同治理具有重要意义。

Sparse group lasso VARX for prediction of atmospheric pollutants: a case study of Beijing-Tianjin-Hebei

The prediction of atmosphere pollutants was studied .Aiming at the problem of over parameterization of the tra-ditional vector autoregressive ( VAR) model, a sparse group lasso penalized VAR model for atmosphere pollutant prediction was proposed .The model parameters are solved by the proximal gradient descent method .In order to prove the validity of the model , this model was applied to prediction of 6 indexes of air quality of January 1, 2016 in Beijing region by using the Beijing-Tianjin-Hebei air pollutant data of 2015.The experimental results show that the normalized mean square error of PM2.5 of the model based on the sparse group lasso penalty is about 3.8%, and its prediction accuracy is higher than that of the VAR model , the large VAR based on various sparse structures ( VAR-L) , and the hierarchical vector autoregression ( HVAR) .In addition, the impacts of different cities in Bei-jing-Tianjin-Hebei on the air quality of the Beijing region can be explained by the parameters of sparse lag group penalized VARX-L and Sparse Own/Other Group penalized VARX-L model.The application of the combination of the convex optimization and the VAR model to the predication of atmospheric pollutant concentration is of great sig -nificance to the synergistic control of air pollution in Beijing-Tianjin-Hebei .