| 一种面向细粒度空气质量分指数(IAQI)预测的时空因果卷积模型 |
| |
| 引用本文： | 张羽民,赵俊杰,梅强,刘希亮,陈卓栋,李建强,王少华,石宇良,柴金川,高雨瑶,井小倩,杨念迪,马小焱.一种面向细粒度空气质量分指数(IAQI)预测的时空因果卷积模型[J].地球信息科学,2023,25(1):115-130. |
| |
| 作者姓名： | 张羽民 赵俊杰 梅强 刘希亮 陈卓栋 李建强 王少华 石宇良 柴金川 高雨瑶 井小倩 杨念迪 马小焱 |
| |
| 作者单位： | 1.北京工业大学软件学院，北京 1001242.集美大学航海学院，厦门 3610213.中国石油审计服务中心 北京 1000284.中国科学院空天信息创新研究院 中国科学院数字地球重点实验室，北京 1000945.中国铁道科学研究院集团有限公司 国家铁道试验中心，北京 100015 |
| |
| 基金项目： | 国家重点研发计划项目(2020YFB2104400) |
| |
| 摘 要： | 精确、细粒度空气质量分指数（Individual Air Quality Index, IAQI）预测是空气质量指数（Air Quality Index, AQI）的基础，对于空气质量防治和保护人类身心健康均具有重要意义。目前传统时序建模、循环神经网络（Recurrent Neural Network, RNN）、图卷积网络（Graph Convolutional Network, GCN）等方法难以有效融合时空因素和气象因素，稳定提取监测站点间动态边缘关系。本文提出了基于时空因果卷积网络（Spatial-Temporal Causal Convolution Networks, ST-CCN）的空气质量分指数预测模型ST-CCN-IAQI。首先采用空间注意力机制分析多源空气污染物和气象因素的空间效应；其次利用堆叠膨胀卷积和时间注意力机制提取特征矩阵的时间依赖性特征；最后采用贝叶斯调优方法对膨胀卷积的多种参数进行了调优。本文采用上海市空气监测站空气质量分指数（IAQI-PM2.5）数据展开实验，并采用一系列基线模型（AR、MA、ARMA、ANN、SVR、GRU、LSTM和ST-GCN）与ST-CCN-IAQI效果进行对比。实验结果显示：① 在单测站测试中，ST-CCN-IAQI的RMSE和MAE值分别为9.873、7.469，相比基线模型平均下降了24.95%和16.87%；R2值为0.917，相比基线平均提升了5.69%；② 对全部站点的IAQI-PM2.5、IAQI-PM10和IAQI-NO2的预测，证明了ST-CCN-IAQI具有较强的泛化能力和稳定性。③ 采用Shapley分析方法论证了IAQI-PM10、湿度、IAQI-NO2对IAQI-PM2.5的预测具有较大程度的影响；通过不同数据抽样条件下的Friedman检验，证明了ST-CCN-IAQI对比基线模型有显著的性能提升。ST-CCN-IAQI方法为细粒度IAQI精准预测提供了一种鲁棒可行的解决方案。
|
| 关 键 词： | 细粒度空气质量分指数预测 多源影响因素 时空注意力 因果卷积网络 贝叶斯优化 Shapley分析 Friedman检验 上海市 |
| 收稿时间： | 2022-05-18 |
A Spatial-temporal Causal Convolution Model for Fine-grained Individual Air Quality Index (IAQI) Prediction |
| |
| ZHANG Yumin,ZHAO Junjie,MEI Qiang,LIU Xiliang,CHEN Zhuodong,LI Jianqiang,WANG Shaohua,SHI Yuliang,CHAI Jinchuan,GAO Yuyao,JING Xiaoqian,YANG Niandi,MA Xiaoyan.A Spatial-temporal Causal Convolution Model for Fine-grained Individual Air Quality Index (IAQI) Prediction[J].Geo-information Science,2023,25(1):115-130. |
| |
| Authors: | ZHANG Yumin ZHAO Junjie MEI Qiang LIU Xiliang CHEN Zhuodong LI Jianqiang WANG Shaohua SHI Yuliang CHAI Jinchuan GAO Yuyao JING Xiaoqian YANG Niandi MA Xiaoyan |
| |
| Institution: | 1. College of Software, Beijing University of Technology, Beijing 100124, China2. Navigation College, Jimei University, Xiamen 361021, China3. China National Petroleum Corporation Auditing Service Center, Beijing 100028, China4. Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China5. National Railway Track Test Center, China Academy of Railway Sciences Corporation Limited, Beijing 100015, China |
| |
| Abstract: | Accurate and fine-grained individual Air Quality Index (IAQI) prediction is the basis of Air Quality Index (AQI), which is of great significance for air quality control and human health. Traditional approaches such as time series modeling, Recurrent Neural Network (RNN) or Graph Convolutional Network (GCN) cannot effectively integrate spatial-temporal and meteorological factors and manage dynamic edge relationship among scattered monitoring stations. In this paper, a ST-CCN-IAQI model is proposed based on spatial-temporal causal convolution networks. Firstly, both the spatial effects of multi-source air pollutants and meteorological factors are considered via spatial attention mechanism. Secondly, time-dependent features in causal convolution network are extracted by stacked dilated convolution and time attention. Finally, multiple parameters in ST-CCN-IAQI are tuned by Bayesian optimization. In this paper, the Individual Air Quality Index (IAQI-PM2.5) data of Shanghai air monitoring station are used to carry out the experiment, and a series of baseline models (AR, MA, ARMA, ANN, SVR, GRU, LSTM, and ST-GCN) are employed to compare with ST-CCN-IAQI. Our results show that: (1) In the single station test, RMSE and MAE values of ST-CCN-IAQI are 9.873 and 7.469, respectively, which decreases by 24.95% and 16.87% on average, respectively; R2 is 0.917, about 5.69% higher than that of the baselines; (2) The prediction of IAQI-PM2.5, IAQI-PM10, and IAQI-NO2 of all stations proves that ST-CCN-IAQI has strong generalization ability and stability; (3) Shapley analysis shows IAQI-PM10, humidity, and IAQI-NO2 have a great impact on the prediction of IAQI-PM2.5. Friedman test under different data sampling conditions proves that ST-CCN-IAQI has significant performance improvement by comparisons with baselines. The ST-CCN-IAQI method provides a robust and feasible solution for accurate prediction of fine-grained IAQI. |
| |
| Keywords: | individual air quality Index prediction multi-source factors temporal and spatial attention causal convolution network Bayesian optimization shapley analysis Friedman test Shanghai |
|
| 点击此处可从《地球信息科学》浏览原始摘要信息 |
| 点击此处可从《地球信息科学》下载免费的PDF全文 |
|
