铁道科学与工程学报2025，Vol.22Issue(7)：2932-2945,14.DOI:10.19713/j.cnki.43-1423/u.T20241543
特殊路网拓扑解构下的时空异质化交通流预测
Spatio-temporal heterogeneous traffic flow prediction under special road network topology deconstruction
摘要
Abstract
In the urban road network,the overall general road network traffic flow usually has the temporal heterogeneity of morning,midday and evening and the spatial heterogeneity of road network association differences,but most of the local special road networks present Y-shaped or ring-shaped topology,and their traffic flow breaks the regular spatio-temporal heterogeneity pattern of the overall road network,showing atypical temporal patterns and spatial association distributions.However,most of the existing studies have modelled the road network as a whole,ignoring the impact of local special road networks.In view of this,in order to solve the problem of insufficient consideration of the influence of Y-shaped and circular road networks and the time-varying spatial association relationship of various types of road network nodes in the existing studies,a spatio-temporal heterogeneous traffic flow prediction model under the deconstruction of special road network topology was proposed,which could make use of the dynamic graph generation module under the influence of time lag to construct a graph structure reflecting the spatial association relationship of the road network at the current time step.On this basis,the special road network deconstruction and dynamic mapping module was utilised to isolate the Y-shaped and ring-shaped road network timing features and their time-lag dynamic maps.Following this,the overall road network,Y-shaped and circular road networks were modelled independently using the spatial feature extraction module under the influence of special road networks.The experiments were based on the public high-speed road network dataset.The results of the study show that the Emae and Ermse of the proposed model improve the performance of the PEMSD4,PEMSD8,and Chengdu-DDT datasets by 4.907 4%,4.340 4%,3.229 5%,0.166 7%,1.267 7%,and 1.186 1%,respectively,when compared with the current state-of-the-art models.Meanwhile,compared with modelling the road network as a whole,the performance of Emae and Ermse of the proposed model is improved by 8.6514%and 6.5366%on the PEMSD8 dataset,which further can prove the effectiveness of considering the local special road network.In summary,the proposed model can fully consider the influence of local special road networks on the overall traffic network,and provide a new way of thinking for the spatio-temporal heterogeneous traffic flow prediction.关键词
交通流预测/图卷积网络/门控循环单元/特殊路网/时空异质性Key words
traffic flow predict/graph convolutional network/gate recurrent unit/special road network/spatio-temporal heterogeneity分类
交通工程引用本文复制引用
侯越,张鑫,袭著涛,王甜甜,马宝君..特殊路网拓扑解构下的时空异质化交通流预测[J].铁道科学与工程学报,2025,22(7):2932-2945,14.基金项目
国家自然科学基金资助项目(62063014,62363020) （62063014,62363020）
甘肃省自然科学基金资助项目(22JR5RA365) （22JR5RA365）
