南京师范大学学报（工程技术版）2025，Vol.25Issue(2)：14-27,14.DOI:10.3969/j.issn.1672-1292.2025.02.002
基于观测与机理数据融合的多站点空气质量联合预测方法
Multi-site Air Quality Collaborative Prediction Method Based on Fusion of Observational and Mechanism Data
摘要
Abstract
This paper proposes a multi-site AQP model based on fusion of observation and mechanism data(MFOM).MFOM considers the multi-site AQP issue with dual-source data of observation and mechanism simulation,aiming to mine the complementary knowledge in dual-source data and the dynamic spatiotemporal dependencies among multiple sites to enhance the AQP effect.Specifically,MFOM adopts an encoding-decoding framework:the encoding part mainly extracts the evolution patterns contained in the observational data,and the decoding stage introduces mechanism data to mitigate the long-term decay of forecast performance.To model the dynamic spatiotemporal dependency of multiple sites,MFOM employs an adaptive dynamic graph generation and convolution method,enhancing collaborative prediction performance through dynamic propagation and aggregation of multi-site information.Additionally,MFOM proposes a spatiotemporal sequence contrastive pre-training module to further enhance the encoder's ability to encode the spatiotemporal patterns of data.Experiments on two years of real data in North China show that MFOM outperforms several state-of-the-art baselines in terms of the prediction of 24 hours and 72 hours.关键词
空气质量预测/时空数据挖掘/动态图神经网络/多源数据融合/对比学习Key words
air quality prediction/spatiotemporal data mining/dynamic graph neural networks/multi-source data fusion/contrastive learning分类
计算机与自动化引用本文复制引用
许碧荷,李清勇,彭庆杰,耿阳李敖..基于观测与机理数据融合的多站点空气质量联合预测方法[J].南京师范大学学报（工程技术版）,2025,25(2):14-27,14.基金项目
中国气象局雷电重点开放实验室课题(2023KELL-B002). （2023KELL-B002）
