摘要
为解决忽略了不同邻近区域之间存在的空间依赖关系导致预测空气质量结果缺乏可靠性这一问题,将图卷积神经网络(graph convolutional networks,GCN)应用于空气质量指数(air quality index,AQI)的预测,基于GCN特征聚合的特点,充分考虑邻近区域之间的空间相关性。文中以安徽省16个地级市为例,构建了城市之间的依赖关系图,预测各个城市的AQI。实验结果表明,相比传统的神经网络模型(ARIMA、RNN、LSTM、GRU),提出的GCN模型表现突出,具有更高的预测准确性和可靠性。
To address the issue of unreliable air quality prediction results due to the neglect of spatial dependency be⁃tween different neighboring regions,Graph Convolutional Networks(GCN)is applied to predict the Air Quality Index(AQI).Based on the characteristics of GCN's feature aggregation,the spatial correlation between adjacent regions is fully considered.Using 16 prefecture-level cities in Anhui Province as an example,a dependency graph is constructed to capture the inter-city relationships and predict the AQI for each city.Experimental results demonstrate that the proposed GCN model outperforms traditional neural network models such as ARIMA,RNN,LSTM,and GRU,exhibiting higher prediction accuracy and reliability.
作者
史雨梅
余晓美
SHI Yumei;YU Xiaomei(Mathematics and Finance School,Chuzhou University,Chuzhou 239000,China)
出处
《三明学院学报》
2024年第3期58-66,共9页
Journal of Sanming University
基金
安徽省自然科学基金一般研究项目(KJ2020B04)
安徽省高等学校科学研究项目(2022AH05009)。
关键词
图卷积神经网络
空气质量指数
空间相关性
graph convolutional networks
air quality index
spatial correlation
