Cross-Attention Fusion Based Spatial-Temporal Multi-GraphConvolutional Network for Traffic Flow Pre

 Lin M . Cross-Attention Fusion Based Spatial-Temporal Multi-Graph Convolutional Network for Traffic Flow Prediction[J]. Sensors, 2021, 21.  SCI三区
论文的详细梳理有个博主整理的很好: (85条消息) [论]Cross-Attention Fusion Based Spatial-Temporal Multi-Graph Convolutional Network for Traffic Flow_panbaoran913的博客-CSDN博客icon-default.png?t=M5H6https://blog.csdn.net/panbaoran913/article/details/125236994

下面是我的阅读笔记:作者花了很大篇幅总结关于交通流预测的图卷积模型,文章指出以往文章只关注静态的时空特征没有考虑动态的。

文章提出:Cross-Attention Fusion Based Spatial-Temporal Multi-GraphConvolutional Network for Traffic Flow Pre_第1张图片

 节点1发生堵塞时,很快就会影响到节点2,但是节点5不受影响,就是说节点具有相似的空间联系却不一关联,而距离远但在一定时间可达的节点却关联。

文中提出的多图卷积是指:三种不同时间间隔的图:current,daily,weekly,这和ASTGCN定义的一样。三种不同的空间联系图:proximity,connectivity,regional similarity 

Cross-Attention Fusion Based Spatial-Temporal Multi-GraphConvolutional Network for Traffic Flow Pre_第2张图片

 Cross-Attention Fusion Based Spatial-Temporal Multi-GraphConvolutional Network for Traffic Flow Pre_第3张图片

 在多图卷积模块运用的是最基本的图卷积操作,不过加入了超参数K,决定可学习参数矩阵W的阶数,公式如下:

Cross-Attention Fusion Based Spatial-Temporal Multi-GraphConvolutional Network for Traffic Flow Pre_第4张图片

 Cross-Attention Fusion Based Spatial-Temporal Multi-GraphConvolutional Network for Traffic Flow Pre_第5张图片

 

 

 

 

 

你可能感兴趣的:(大数据,人工智能,交通物流)