Graph Neural Networks for Traffic Flow Prediction

Sammanfattning

Network reconfigurations, including closures, new road segments, capacity changes,and speed changes, can redistribute traffic flows. Predicting this redistribution isimportant for rapid assessment of infrastructure changes and operational interventions.Conventional Stochastic User Equilibrium (SUE) methods can compute theresulting equilibrium state, but they require Origin-Destination (OD) demand andmust be solved again for each reconfigured network. This thesis studies a more constrainedsetting: predicting post-edit equilibrium flows from the pre-edit network,old flows, and the reconfigured network, without OD demand as input.A physics-informed graph learning framework, called ST-PINN GatedGCN, is developedfor this task. First, edge alignment maps old-flow information onto the edgeset of the reconfigured graph. The GatedGCN then learns flow propagation andtopology changes through graph message passing. The training objective combinesedge-flow regression with flow-conservation regularization. It fits SUE-generated labelswhile reducing node-level physical inconsistency. Inference is completely ODfree as the demand is used only to generate labels.Experiments use the Sioux Falls and EMA networks, with 10,000 network-pair samplesfor each network. The model obtains 12.92% WMAPE on Sioux Falls and7.84% WMAPE on EMA, which is lower than the tested baselines. Retained edgesare predicted more accurately than newly added edges, since new edges have nohistorical flow and may introduce new route choices. Ablations show that old-flowinformation, residual physical injection, the global message channel, and recurrentpressure update affect error and conservation behavior. Inference is about 197 timesfaster than SUE on Sioux Falls and about 1896 times faster on EMA.These results suggest that old flows and graph structure contain useful informationfor fast scenario screening under network reconfiguration. The remaining errorsreflect the uncertainty caused by unobserved OD demand, especially for newly addedlinks.