Cloud-Edge collaborative adaptive dual-graph spatio-temporal passenger flow forecasting model for smart high-speed railway services

Passenger flow in rapidly developing high-speed railway networks varies with temporal fluctuations, spatial interactions among stations, and regional travel dependencies. Accurate forecasts support the management of station operations, resource allocation, and service optimization, yet methods that impose fixed spatial structures or rely on centralized computing architectures have difficulty representing dynamic passenger flow correlations while satisfying the real-time and large-scale requirements of railway service applications. We therefore propose a cloud-edge collaborative adaptive dual-graph spatio-temporal passenger flow forecasting model for smart high-speed railway services. The dual-graph representation combines physical connectivity among railway stations with an input-window-adaptive correlation graph derived from recent passenger flow observations. A spatio-temporal learning module captures multi-scale temporal variations and complex spatial dependencies, while the cloud-edge collaborative mechanism integrates global knowledge from the cloud side with local real-time features from the edge side to improve prediction accuracy while maintaining a lightweight edge-side inference branch. Experiments on a real-world large-scale rail transit passenger flow dataset demonstrate that the proposed model achieves better forecasting performance than the compared baselines. Additional analyses verify the effectiveness of each component and the advantages of the proposed cloud-edge architecture in practical smart railway service scenarios.