一种时空注意力网络的交通预测模型

交通预测在智能交通中有着重要的意义和应用.由于交通数据的复杂性和高度的非线性,精确的交通预测的核心挑战在于如何对复杂的空间相关性和时间动态建立模型.在现实生活中,我们发现:1)区域间的空间依赖是动态的;2)时间依赖有日和周的模式,但由于有动态时间变化,它不具有严格周期性.为了解决这两个问题,我们提出了一个新的时空注意力网络(STAN),该模型的主要思想是区域间的动态相似性用一个门控机制学习,长期周期性时间转移现象由一个周期性注意力转移机制来学习,并考虑交通道路、天气状况等外部因素.通过与不同的方法在两个数据集上进行评估,实验结果表明,我们提出的模型有更好的准确性.

Traffic Prediction Model Based on Spatiotemporal Attention Network

Traffic prediction has important significance and application in intelligent transportation.Due to the complexity and high degree of non-linearity of traffic data,the core challenge of accurate traffic prediction is how to model complex spatial correlations and time dynamics.In real life,we find:1)the spatial dependence between regions is dynamic;2)the time dependence have day and week patterns,but it does not have strict periodicity due to dynamic time changes.In order to solve these two problems,we propose a new spatiotemporal attention network(STAN).The main idea of the model is that the dynamic similarity between regions is learned by a gated mechanism,and the long-term periodic time shift phenomenon is learned by a periodic attention shift mechanism,and external factors such as traffic roads and weather conditions are considered.By evaluating on two data sets with different methods,the experimental results show that our proposed model has better accuracy.