Efficient decomposition method for the stochastic optimization of public transport schedules

Sofia Zaourar-Michel

Oct 15, 2016 (modified: Oct 15, 2016) NIPS 2016 workshop MLITS submission readers: everyone
  • Abstract: We propose a new method to optimize public transport schedules by minimizing the waiting time during transfers. Using ticket validation data to construct a realistic scenario-based model of the waiting times, our goal is to design shifts of the current schedules that reduce the overall expected waiting time. We propose a parallel local search heuristic that exploits the structure of the problem to efficiently explore a large number of possible schedules. Compared to previous approaches, our algorithm should allow to both treat more transfers (bigger cities) and more scenarios (ensuring a better generalization). We provide promising preliminary results on transit data collected from Nancy, France.
  • Conflicts: xerox.com