A deep neural network model with GCN and 3D convolutional network for short‐term metro passenger flow forecasting

Rail transit has many advantages, such as large passenger capacity, convenience, safety, and environmental protection, making it the preferred travel mode for most passengers. Deep learning become an effective method short-term rail flow prediction. A deep model which combines a graph convolutional network three-dimensional Convolutional Neural Network improved by residual module attention mechanism (ARConv-CGN) is proposed. First, historical inflow outflow data are aggregated into three patterns: recent pattern, daily weekly pattern separately. The GCN applied to capture spatiotemporal topological information of flows in each pattern. Second, neural used deeply integrate patterns information. Additionally, increase number layers prevent gradient from disappearing. Finally, also introduced adjust importance different so that improving performance model. Training with automatic fare collection on weekdays Beijing Xiamen provides good demonstration superiority ARConv-CGN metro forecasting.