This paper presents a schedule-driven signal priority control for modern trams using reinforcement learning. This self-learning adaptive system has the following three features: 1) In addition to getting the minimal delay of general vehicles, the schedule-driven concept was adopted by using reliability of the tram's schedule as the primary control goal. 2) Model-free reinforcement learning was used to find the optimal signal timing strategy instead of traditional model-based methods. 3) A dynamic phasing sequence was used to extend the flexibility of signal priority instead of fixed sequence in cyclic timing. Compared with three other traffic signal control strategies, extensive VISSIM simulation experiments under different traffic conditions have demonstrated the excellent punctuality and delay reduction of schedule-driven TSP for modern trams.
展开▼