ST-VGBiGRU: A Hybrid Model for Traffic Flow Prediction With Spatio-Temporal Multimodality

Traffic flow prediction is a key step to realizing the effective guidance and control of intelligent transportation systems. For using short-term non-stationarity spatio-temporal correlation presented in traffic flow, hybrid model, which called ST-VGBiGRU, based on improved Variational Modal Decomposition (VMD), Graph Attention Network (GAT), Bidirectional Gated Recurrent Unit (BiGRU) network proposed. First, sequence decomposed into series relatively stationary modal components VMD algorithm reduce its non-stationarity. The high-frequency are noise-reduced Fuzzy Entropy (Fuzzy En) method improve accuracy decomposition. After that, GAT used capture different attention levels node their neighboring nodes, obtain more spatial characteristics flow. Then, each component containing features fed BiGRU separately temporal correlation. Each model parameter trained optimum RMSProp algorithm, improves model’s while speeding up convergence algorithm. In order illustrate performance ST-VGBiGRU RTMC dataset conduct ablation experiments module, module. Meanwhile, we combined PeMS baseline multi-step with other six models. results show that our better than all