| An innovative conditional transit signal priority algorithm, based on the concept of Reinforcement Learning (RL), is used to optimize transit operations through headway control. The proposed algorithm has the ability to mimic an intelligent agent which can determine the best combination and duration of each signal phase. The objective of the single agent is to maintain a regular headway between streetcars, while the multiple agents are trained to deal with the onset of streetcar bunching. The microscopic traffic simulation software Paramics was employed to simulate transit and traffic operations along the King Streetcar route. Simulation results show that the control policy learned by the single agent could effectively reduce the transit headway deviation and causes smaller disruption to crossroad traffic compared with the existing unconditional transit signal priority algorithm. Furthermore, the multiple agents could split up a streetcar bunch and prevent it from forming again with a high success rate. |
