基于多源城市数据的空气质量预测模型
Air quality prediction model based on multi-source urban data
毕乐 1冯春芳 1陈湘国 1魏忠诚 1赵继军1
作者信息
- 1. 河北工程大学信息与电气工程学院,河北邯郸 056038;河北工程大学河北省安防信息感知与处理重点实验室,河北邯郸 056038
- 折叠
摘要
针对目前空气质量预测研究方法只考虑单一站点时序特性的缺陷问题,提出一种基于多源城市数据的STRGNN模型.将相邻站点之间的空间相关性表示为距离图和相似图,根据领域类别对特征分组;将分组特征与构建的距离图和相似图成对组合输入至模型.通过收集邻近信息捕获空气质量的空间相关性,利用叠加多层的卷积捕获空气质量的时间相关性.仿真结果表明,该模型与GRU、Seq2Seq等4个基准模型相比,在l h-6 h预测的MAE和RMSE分别降低了 3％和1％,预测效果有所提升,验证所提方法与现有方法相比具有优越性.
Abstract
To solve the problem that the current research methods of air quality prediction only consider the time series characte-ristics of a single station,a STRGNN model based on multi-source city data was proposed.The spatial correlation between adjacent sites was expressed as distance graph and similarity graph,and features were grouped according to domain categories.After that,the grouped features were combined with the constructed distance map and similarity map in pairs and input into the model.The spatial correlation of air quality was captured by collecting neighborhood information,and the temporal correlation of air quality was captured by overlapping multi-layer convolution.The simulation results show that compared with GRU,Seq2Seq and other four benchmark models,the MAE and RMSE predicted through the proposed model is decreased by 3％and 1％respectively in 1 h-6 h,and the prediction effect is improved,confirming that the proposed method is superior to the existing methods.
关键词
空气质量预测/多源城市数据/距离图/相似图/空间相关性/时间相关性/图卷积网络Key words
air quality forecast/multi-source city data/distance graph/similarity graph/spatial correlation/temporal correla-tion/graph convolutional network引用本文复制引用
基金项目
河北省级研究生示范课程建设基金项目(KCJSX2022090)
河北省高等学校科学技术研究基金项目(QN2020193)
邯郸市科学技术研究与发展计划基金项目(21422031288)
出版年
2024
NETL
NSTL
万方数据