基于时空注意力机制的GCN-LSTM地铁短时OD客流预测方法OA

Under subway network operation conditions,passenger flow exhibits highly complex spatio-temporal dynamics.Accurate short-term origin-destination(OD)passenger flow prediction is a fundamental requirement for transportation organization optimization and congestion mitigation.However,existing prediction methods still show limited capability in joint spatio-temporal dependency modeling of passenger flow.To address this issue,this paper proposed a novel short-term OD passenger flow prediction model.It introduced a spatio-temporal attention mechanism,and deeply integrated graph convolutional networks(GCN)-long short-term memory networks(LSTM)modeling advantages.The model adaptively assigned historical time-step importance and dynamically identified key stations or regions influencing target OD pairs.This allowed for a more precise capture of the nonlinear spatio-temporal passenger flow propagation in complex subway networks.Validation was performed using real subway network data.The results show that the proposed spatio-temporal attention GCN-LSTM model significantly improves prediction accuracy compared with baseline models.Spatio-temporal passenger flow fluctuations are captured more accurately.This provides reliable data support and a decision-making basis for subsequent fine-grained passenger flow control,capacity allocation,and collaborative optimization in subway network systems.