基于图注意力网络和时间卷积网络的空气污染物浓度预测方法
Method for air pollutant concentration prediction based on graph attention network and temporal convolutional network
陈伟洪 1杨茹 1王浩 2郑中华3
作者信息
- 1. 上海师范大学信息与机电工程学院,上海 200234
- 2. 上海新致软件股份有限公司,上海 200127
- 3. 安徽博约信息科技股份有限公司,合肥安徽 230601
- 折叠
摘要
提出了 一种融合图注意网络(GAT)和带注意力机制的时间卷积网络(ATCN)的创新模型——GAT-ATCN,旨在提高空气污染物浓度预测的精度和效率.在通过GAT捕捉监测站点间的复杂空间依赖关系,利用注意力机制,自适应地加强重要节点之间的连接,从而提取空间特征.ATCN被用来处理时间序列数据,通过学习时间维度上的长期依赖关系,捕获污染物浓度随时间变化的动态特性.选取中国江浙沪地区7个城市2018-2020年的实际空气质量监测和气象数据,构建数据集并进行实验,验证了 GAT-ATCN模型的有效性.实验结果显示:GAT-ATCN模型在多个评价指标上均表现优异,能够更准确地预测空气污染物浓度.
Abstract
An innovative model that integrated graph attention networks(GAT)and attention-based temporal convolutional networks(ATCN),named GAT-ATCN was proposed to improve the accuracy and efficiency of air pollutant concentration prediction.Firstly,the complex spatial dependencies between monitoring stations through GAT were captured,using an attention mechanism to adaptively strengthen the connections between important nodes,thereby extracting spatial features.Secondly,the ATCN part was used to process time series data,learning long-term dependencies in the time dimension to capture the dynamic characteristics of pollutant concentration changes over time.Finally,actual air quality monitoring data and meteorological data from seven cities in the Jiangsu-Zhejiang-Shanghai region of China from 2018 to 2020 were selected to build a dataset and conduct experiments,which verified the effectiveness of the GAT-ATCN model.Experimental results showed that the GAT-ATCN model performed excellently across multiple evaluation metrics and could predict air pollutant concentration more accurately.
关键词
空气污染物浓度预测/图注意网络(GAT)/带注意力机制的时间卷积网络(ATCN)/深度学习Key words
air pollution concentration prediction/graph attention network(GAT)/attention-based temporal convolutional network(ATCN)/deep learning引用本文复制引用
基金项目
国家自然科学基金(62302306)
国家自然科学基金(62372300)
国家自然科学基金(62201350)
出版年
2024
NETL
NSTL
万方数据