GCN for Multistep Speed Prediction

Neighbors Shape Tomorrow's Speed

A road does not evolve in isolation. Speed on one link can be affected by traffic arriving from upstream, queues spilling back from downstream, and conditions on neighboring links. Forecasting several steps ahead therefore requires more than extending each road's own recent trend.

This video uses a graph convolutional network (GCN) to represent the road network as a graph. Each road segment holds a current traffic state, while connections determine which other segments can influence it. Information moves across those connections, combines with recent history, and produces a near-future speed estimate. The final sequence makes the longer forecast visible: a predicted network state becomes part of the information used for the next prediction. The central insight is the joint treatment of space and time. A model that understands both the road connections and how congestion changes over successive moments can anticipate the movement of slowdowns better than a collection of independent link forecasts.