Traffic forecasting is an important issue in Intelligent Transportation Systems (ITS). Recently, Graph Convolution Network (GCN) based traffic forecasting methods have been proposed and achieved competitive performance for natural graph representation of traffic data. However, most GCN-based methods focus on exploring the spatio-temporal characteristics of traffic data and ignore the effects of external cross-modal data such as Point of Interest (POI). Although some works have tried to use POI data to improve traffic forecasting by including POI data in the construction of the adjacency matrix of GCN, the problem of representing POI data is far from being solved, and the correlation between cross-modal traffic data and POI data is not well exploited. As a result, the performance improvement is not obvious despite incorporating POI data. In this paper, a novel POI Correlation Learning GCN (PCLGCN) is proposed to bridge the heterogeneous gap between the traffic data and POI data for traffic forecasting. Here, the POI data and its category are encoded and embedded into the GCN by the proposed cross-modal correlation learning network and gate mechanism. The proposed PCLGCN method is evaluated in detail using two real city traffic speed datasets. Experimental results show that it demonstrates significant performance improvements over the current GCN-based traffic forecasting approaches. Moreover, the proposed POI correlation learning network is also valid when integrated with different GCN-based traffic forecasting methods.
Bridging Heterogeneous Gap: A POI Correlation Learning Graph Convolution Network for Traffic Forecasting
Lect. Notes Electrical Eng.
International Conference on SmartRail, Traffic and Transportation Engineering ; 2024 ; Chongqing, China October 25, 2024 - October 27, 2024
Advances and Applications in SmartRail, Traffic, and Transportation Engineering ; Chapter : 8 ; 75-87
2025-07-19
13 pages
Article/Chapter (Book)
Electronic Resource
English
Contrastive learning for traffic flow forecasting based on multi graph convolution network
DOAJ | 2024
|Contrastive learning for traffic flow forecasting based on multi graph convolution network
Wiley | 2024
|Spatio-Temporal Graph Attention Convolution Network for Traffic Flow Forecasting
Transportation Research Record | 2024
|