摘要
为对交通流进行准确预测,提出一种将图优化与预测相结合,在单管路中面向交通流的时空混合图卷积预测模型,用于边缘环境下物联网的城市交通流预测。首先对关联图进行预处理,以去除城市交通数据原始道路网中的噪声;再用LOF删除不相关的模型和噪声;最后将得到的图扩展成图卷积神经网络,估算城市的交通流。另外,采用基于分支定界的优化技术对超参数进行精确调整。结果表明:所提模型在交通流预测方面效果更优,当图中节点数较多时,预测的精准性明显优于其他基准模型。
In order to accurately predict traffic flow,a spatiotemporal mixed graph convolutional prediction model is proposed,which combines graph optimization and prediction,and is oriented towards traffic flow in a single pipe road.It is used for predicting urban traffic flow in the edge environment of the Internet of Things.The correlation graph is preprocessed to remove the noise in the original road network of urban traffic data.LOF is used to remove irrelevant models and noise.Then the obtained graph is extended into a graph convolutional neural network to estimate the traffic flow of the city.In order to accurately adjust the hyperparameters,an optimization technique based on branch and bound is adopted.The results show that the proposed model has achieved better results in traffic flow prediction.When the number of nodes in the graph is large,the accuracy of prediction is obviously better than other benchmark models.
作者
王静潇
王辛岩
周禹彤
张越
WANG Jingxiao;WANG Xinyan;ZHOU Yutong;ZHANG Yue(College of Engineering,Tibet University,Lhasa 850000,China)
出处
《现代电子技术》
2023年第12期153-158,共6页
Modern Electronics Technique
基金
国家自然科学基金项目:基于西藏地域特色的道路景观形态研究(51868068)。
关键词
智能交通系统
混合图卷积神经网络
交通流预测
分支定界算法
深度学习
预处理
超参数优化
intelligent transportation system
hybrid graph convolutional neural networks
traffic flow prediction
branch and bound algorithm
deep learning
pretreatment
hyperparameter optimization
