基于图神经网络的长江流域经济带PM_(2.5)浓度预测  

Prediction of PM_(2.5)Concentration in Yangtze River Economic Belt Based on Graph Neural Network

作　　者：蒋锋[1] 韩兴钰 王辉[2,3] JIANG Feng;HAN Xingyu;WANG Hui(School of Statistics and Mathematics,Zhongnan University of Economics and Law,Wuhan 430073,China;Wuhan Documentation and Information Center,Chinese Academy of Sciences,Wuhan 430071,China;Hubei Key Laboratory of Big Data in Science and Technology,Chinese Academy of Sciences,Wuhan 430071,China)

机构地区：[1]中南财经政法大学统计与数学学院,湖北武汉430073 [2]中国科学院武汉文献情报中心,湖北武汉430071 [3]中国科学院科技大数据湖北省重点实验室,湖北武汉430071

出　　处：《环境科学与技术》2023年第11期90-101,共12页Environmental Science & Technology

基　　金：教育部人文社科一般项目(22YJAZH038);科技大数据湖北省重点实验室项目(E3KF291001)。

摘　　要：文章基于长江流域经济带沿线99个城市的PM_(2.5)浓度监测数据,采用转移熵构建出了长江流域经济带大气污染空间交互影响网络,并从整体和局部2个角度分析了长江流域经济带污染物传导方向和传导强度。为了充分利用城市大气污染的空间关联信息,文章使用大气污染空间交互影响网络改进了T-GCN图卷积神经网络,并构建了基于T-GCN_(TE)的预测模型,对长江流域经济带99个城市的污染物浓度进行预测。研究发现长江流域经济带各个城市的大气污染表现出很强的紧密性,整体网络的信息传递以地区间的信息传递为主。另外,T-GCN_(TE)能够捕捉到大气污染的时空依赖性和影响方向,能获得更好的效果。基于上述研究结论,文章从建立大气污染多方联防联控机制、加强产业合作、完善生态补偿机制等方面对深入开展长江流域经济带大气污染协同治理提供了建议。Based on the monitoring data of PM_(2.5)concentration in 99 cities along the Yangtze River Economic Belt,the article constructed a spatial interaction network of air pollution in the Belt using transfer entropy,and analyzed the pollutant transmission direction and transmission intensity from both the overall and local perspectives.Then,in order to make full use of the spatial correlation information of urban air pollution,the article used the spatial interaction network of air pollution to improve the graph structure of T-GCN,and constructed a prediction model based on the T-GCN_(TE)to predict the pollutant concentrations in 99 cities of the Belt.It is found that the air pollution in each city shows strong compactness,and the information transfer of the overall network is dominated by inter-regional information transfer.Moreover,T-GCN_(TE)can capture the spatio-temporal dependence and the influence direction of the air pollution,and better results can be obtained.Based on the above conclusions,the article provides suggestions for the development of collaborative governance system of air pollution in the Belt,strengthening industrial cooperation and improving the ecological compensation mechanism.

关 键 词：图神经网络 网络分析 长江流域经济带 转移熵 

分 类 号：X513[环境科学与工程—环境工程]