Deep Spatiotemporal Network GC-SCI: A Model for Short-Term Forecasting of Subway Passenger Flow

Accurate short-term passenger flow forecasting is crucial for the organization and management of subway systems. However, passenger flow in subway networks exhibits high nonlinearities and complex patterns due to unstructured spatiotemporal dependencies. To address these challenges, this paper proposes a novel deep learning approach, the Graph Convolutional Sample Convolution and Interaction Network (GC-SCI), for forecasting subway passenger flow. GC-SCI mainly consists of two components, Graph Convolutional Neural Network (GCN) and Sample Convolution and Interaction Network (SCINet), which are used to model graph-based spatial and temporal dependencies, respectively. The proposed method is validated using a real-world case of the Shenzhen Metro in China, and the experimental results demonstrate that GC-SCI can well capture the graph-based spatial and temporal dependencies and outperform state-of-the-art baselines in terms of forecasting accuracy. Furthermore, considering the inter-station correlation driven by the relationship between the inflow and outflow of passengers enhances the model’s forecasting accuracy.