A Spatial-Temporal Neural Network Model Based on Meta Learning for Bus Arrival Time Prediction

Accurate bus arrival time prediction is of great practical significance for improving passenger travel experience and optimizing bus scheduling. However, buses typically exhibit significant temporal pattern differences across different road segments. To address this issue, we propose a spatial-temporal neural network model based on meta learning, named MetaSTNN. Specifically, MetaSTNN first constructs Meta Graph Convolutional Network (MetaGCN) and Meta Temporal Convolutional Network (MetaTCN) based on a meta network for modeling diverse spatial-temporal dependencies. Second, MetaSTNN models higher-order spatial-temporal dependencies by stacking multiple layers of Meta Spatial-Temporal Feature Extraction Block (Meta STBlock) composed of a MetaGCN and a MetaTCN. Finally, MetaSTNN utilizes an information fusion layer to integrate intrinsic spatial-temporal representation with external information for accurate single-line bus arrival time prediction. Experimental verification on bus line 72 and bus line 406 in Beijing demonstrates the effectiveness and superiority of MetaSTNN.