Learning Heuristics for Transit Network Design and Improvement with Deep Reinforcement Learning

The authors use deep reinforcement learning with graph neural networks to learn low-level heuristics for an evolutionary algorithm, which improves the algorithm's results on benchmark synthetic cities and obtains state-of-the-art results when optimizing operating costs. The learned heuristics also improve upon a simulation of the real transit network in the city of Laval, Canada, by as much as 54% and 18% on two key metrics, and offer cost savings of up to 12% over the city's existing transit network.