Search for a command to run...
With the acceleration of urbanization, public transit systems face prominent challenges, including insufficient passenger flow prediction accuracy and low scheduling efficiency. This study analyzes passenger flow variation patterns from both spatial and temporal dimensions, constructs spatiotemporal matrices, and employs matrix dimensionality reduction methods to extract key features. We propose a passenger flow prediction model based on GCN-AM-LSTM and a dynamic real-time intelligent scheduling strategy. For passenger flow prediction, the model first utilizes Graph Convolutional Networks (GCNs) to extract spatial features of the transit network, then employs Attention Mechanism-enhanced Long Short-Term Memory networks (AM-LSTM) to perform weighted extraction of temporal features, and finally integrates external factors such as weather conditions to generate prediction outputs. For scheduling optimization, a dynamic real-time scheduling mode is adopted: the foundational framework optimizes dynamic departure timetables using a multi-objective particle swarm optimization algorithm, which is then combined with real-time passenger flow data to adjust departure intervals at the route level and implement stop-skipping strategies at the station level. Validation was conducted using Xiamen BRT Line 1 as a case study. Experimental results demonstrate that the proposed GCN-AM-LSTM prediction model reduces Mean Absolute Error (MAE) by 14% and 22% compared to CNN and LSTM models, respectively, achieving significantly improved prediction accuracy. Regarding scheduling optimization, the number of departures decreased by 15.24%, passenger waiting time costs were reduced by 3.7%, and transit operating costs decreased by 3.19%, effectively balancing service quality and operational efficiency.