This website requires JavaScript.

Boosting Urban Traffic Speed Prediction via Integrating Implicit Spatial Correlations

Dongkun WangWei FanPengyang Wang ...+3 Yanjie Fu
Dec 2022
摘要
Urban traffic speed prediction aims to estimate the future traffic speed forimproving the urban transportation services. Enormous efforts have been made onexploiting spatial correlations and temporal dependencies of traffic speedevolving patterns by leveraging explicit spatial relations (geographicalproximity) through pre-defined geographical structures ({\it e.g.}, regiongrids or road networks). While achieving promising results, current trafficspeed prediction methods still suffer from ignoring implicit spatialcorrelations (interactions), which cannot be captured by grid/graphconvolutions. To tackle the challenge, we propose a generic model for enablingthe current traffic speed prediction methods to preserve implicit spatialcorrelations. Specifically, we first develop a Dual-Transformer architecture,including a Spatial Transformer and a Temporal Transformer. The SpatialTransformer automatically learns the implicit spatial correlations across theroad segments beyond the boundary of geographical structures, while theTemporal Transformer aims to capture the dynamic changing patterns of theimplicit spatial correlations. Then, to further integrate both explicit andimplicit spatial correlations, we propose a distillation-style learningframework, in which the existing traffic speed prediction methods areconsidered as the teacher model, and the proposed Dual-Transformerarchitectures are considered as the student model. The extensive experimentsover three real-world datasets indicate significant improvements of ourproposed framework over the existing methods.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答