人工智能培训

搜索

人工智能论文:基于动态时空图的CNN用于流量预测(Dynamic Spatio-temporal Graph-based CNNs for Traffic Pr

[复制链接]
dabiao2008 发表于 2018-12-6 12:00:12 | 显示全部楼层 |阅读模式
dabiao2008 2018-12-6 12:00:12 229 0 显示全部楼层
人工智能论文:基于动态时空图的CNN用于流量预测(Dynamic Spatio-temporal Graph-based CNNs for Traffic Prediction)由于大规模的问题规模,以及交通流的时空依赖性的复杂和动态性质,准确的交通预测是一个具有挑战性的问题。大多数现有的基于图形的CNN试图捕获静态关系,同时在很大程度上忽略了连续数据的动态。在本文中,我们通过学习表达特征来呈现动态时空图基于CNN(DST-GCNN)来表示时空结构并预测来自历史交通流的未来交通。特别地,DST-GCNN是双流网络。在流预测流中,我们提出了一种新颖的基于图的时空卷积层,以从交通流的图表表示中提取特征。然后将几个这样的层堆叠在一起以预测未来的流量。同时,随着交通条件随时间变化,图中节点之间的邻近关系通常是时变的。为了捕获图形动态,我们使用图形预测流来预测动态图形结构,并将预测结构馈送到流动预测流中。对实际交通数据集的实验表明,与其他最先进的方法相比,所提出的模型实现了竞争性能。
Accurate traffic forecast is a challenging problem due to the large-scaleproblem size, as well as the complex and dynamic nature of spatio-temporaldependency of traffic flow.Most existing graph-based CNNs attempt to capturethe static relations while largely neglecting the dynamics underlyingsequential data.In this paper, we present dynamic spatio-temporal graph-basedCNNs (DST-GCNNs) by learning expressive features to represent spatio-temporalstructures and predict future traffic from historical traffic flow.Inparticular, DST-GCNN is a two stream network.In the flow prediction stream, wepresent a novel graph-based spatio-temporal convolutional layer to extractfeatures from a graph representation of traffic flow.Then several such layersare stacked together to predict future traffic over time.Meanwhile, theproximity relations between nodes in the graph are often time variant as thetraffic condition changes over time.To capture the graph dynamics, we use thegraph prediction stream to predict the dynamic graph structures, and thepredicted structures are fed into the flow prediction stream.Experiments onreal traffic datasets demonstrate that the proposed model achieves competitiveperformances compared with the other state-of-the-art methods.人工智能论文:基于动态时空图的CNN用于流量预测(Dynamic Spatio-temporal Graph-based CNNs for Traffic Prediction) mQNZsPcWGzSLUm4j.jpg
URL地址:https://arxiv.org/abs/1812.02019     ----pdf下载地址:https://arxiv.org/pdf/1812.02019    ----人工智能论文:基于动态时空图的CNN用于流量预测(Dynamic Spatio-temporal Graph-based CNNs for Traffic Prediction)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

dabiao2008当前离线
新手上路

查看:229 | 回复:0

快速回复 返回顶部 返回列表