Online transfer learning with an MLP-assisted graph convolutional network for traffic flow prediction: a solution for edge intelligent devices

• Proposes OTL-GM, an online transfer learning framework with an MLP-assisted GCN that transfers source-domain features to edge devices and bridges domain gaps via online learning. • Reduces convergence time by 24.77%–95.32% compared with non-OTL models across four traffic flow datasets. • Addresses critical practical constraints of intelligent edge devices, including limited computing resources, data sparsity, and external environmental influences. • Enhances timeliness and reliability of traffic flow prediction, supporting real-time route planning and intelligent transportation management.