Paper code: Pure model code.

The seamless integration of shared bikes and metro systems promotes green and eco-friendly travel, yet the supply-demand imbalance of shared bikes around metro stations remains a critical challenge, making accurate demand prediction particularly crucial. Targeting metro-adjacent areas, this study proposes a method to identify shared bike trips connecting to metro usage, effectively filtering out approximately 24% of non-connecting travel records within the buffer zones. A predictive model integrating a Spatiotemporal Attention Graph Convolutional Network (STAGCN), Long Short-Term Memory (LSTM) network, and Informer is developed to forecast shared bike demand for metro connectivity. Specifically, the Informer model incorporates STAGCN to capture spatial correlations in bike demand and introduces an LSTM module to learn long- and short-term temporal dependencies. The final demand prediction is generated through a multilayer perceptron. Experiments conducted on shared bike and metro datasets in Shenzhen demonstrate that the proposed model achieves a coefficient of determination (R2) of 0.893, outperforming baseline models by 6.7% in prediction accuracy. Additionally, it exhibits lower Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) compared to traditional time-series forecasting methods. The proposed demand prediction model can assist operators in optimizing the allocation of shared bike resources, which is of great significance for improving user experience.