SISCOG has just published a scientific paper addressing the optimization of public transport timetables with respect to passenger’s total travel times, combining several techniques that have never been applied in the field, including reinforcement learning. The approach applied to real-world data performed better than existing approaches and even outperforms current state-of-the-art algorithms.
In more detail, the paper “Solving periodic timetabling problems with SAT and machine learning” co-authored by SISCOGuians Gonçalo Matos, Luís Albino, Ricardo Saldanha and Ernesto Morgado, is a research work that used an approximation method based on SAT, reinforcement learning and multiagents, a combination of techniques which (to the authors' knowledge) has never been applied in this field before. In order to evaluate the approach, it was benchmarked against a set of periodic timetabling optimization problems publicly available, namely real-world railroads and bus periodic timetabling problems.
SISCOG Suite’s ONTIME product, for creating and managing timetables, already incorporates such an optimizer.
