Physical-Virtual Collaboration Modeling for Intra- and Inter-Station Metro Ridership Prediction

Due to the widespread applications in real-world scenarios, metro ridership prediction is a crucial but challenging task intelligent transportation systems. However, conventional methods either ignore topological information of systems or directly learn on physical topology, and cannot fully explore patterns evolution. To address this problem, we model system as graphs with various topologies propose unified Physical-Virtual Collaboration Graph Network (PVCGN), which can effectively complex from tailor-designed graphs. Specifically, graph built based realistic topology studied system, while similarity correlation are virtual under guidance inter-station passenger flow correlation. These complementary incorporated into Convolution Gated Recurrent Unit (GC-GRU) for spatial-temporal representation learning. Further, Fully-Connected (FC-GRU) also applied capture global evolution tendency. Finally, develop Seq2Seq GC-GRU FC-GRU forecast future sequentially. Extensive experiments two large-scale benchmarks (e.g., Shanghai Metro Hangzhou Metro) well demonstrate superiority our PVCGN station-level prediction. Moreover, apply proposed online origin-destination (OD) experiment results show universality method. Our code available at https://github.com/HCPLab-SYSU/PVCGN.