融合拓扑信息与气象信息的空气质量预测网络  

Air Quality Prediction Network Integrating Topological and Meteorological Information

作　　者：严梦瑶 林宣雄[1] 张旭 YAN Mengyao;LIN Xuanxiong;ZHANG Xu(School of Software,Department of Electronic and Information Science,Xi an Jiaotong University,Xi an 710049,China;Xi an Jointsky Software Holding Co.,Ltd.,Xi an 710065,China)

机构地区：[1]西安交通大学电子与信息学部软件学院,陕西西安710049 [2]西安长天长软件股份有限公司,陕西西安710065

出　　处：《中国环境监测》2023年第S01期80-90,共11页Environmental Monitoring in China

基　　金：国家重点研发计划项目(2019YFB2103001)。

摘　　要：通过空气质量监测数据对正在形成或即将到来的空气污染进行预测是一项具有重要意义的工作,而空气质量监测站只能检测其周围一定范围内的空气污染情况。为了衡量整个城市的空气污染情况,获取任意时间、任意位置的空气质量信息,结合交叉注意力机制,提出了一种融合拓扑信息与气象信息的空气质量预测网络(CGMIM)。将西安市空气质量监测数据与气象数据转换为图像拼接起来,作为输入信息。在高阶非线性时空动态神经网络(MIM)的基础上引入注意力机制,并增加拓扑图编码器模块,提高模型提取能力以及对空气质量监测数据中的空间特征的利用率。最后,使用时空损失函数替代传统的均方误差损失函数,提高模型对空间关系的关注。结果表明:CGMIM网络模型能够在准确预测的同时,对位置区域合理填充,能够有效提升空气质量监测数据的空间分辨率。Predicting emerging or impending air pollution from air quality monitoring data is an important task,and air quality monitoring stations can only detect air pollution within a certain range of their surroundings.In order to measure the air pollution of the whole city,obtain air quality information at any time and any location,combininge the cross-attention mechanism,an air quality prediction network(CGMIM)that integrates topological information and meteorological information is proposed.The air quality monitoring data and meteorological data in Xi an are converted into images and stitched as input,and the attention mechanism is introduced on the basis of the high-order nonlinear spatiotemporal dynamic neural network(MIM),and the topology map encoder module is added to improve the model extraction ability and the utilization rate of spatial features in the air quality monitoring data.Finally,the spatiotemporal loss function is used to replace the traditional mean squared error loss function to improve the model s attention to spatial relationships.The results show that the CGMIM network model can accurately predict and reasonably fill the location area,which can effectively improve the spatial resolution of air quality monitoring data.

关 键 词：空气质量预测 循环神经网络 图卷积神经网络 

分 类 号：X823[环境科学与工程—环境工程]