Real-time dynamic route generation algorithm of DRT with deep Q-learning

Demand-responsive transportation (DRT) offers a flexible solution for transportation issues in areas with insufficient public transit. Unlike fixed-route buses, DRT dynamically adjusts routes based on user demand. However, without effective route optimisation, DRT can underperform compared with fixed-route buses. This study proposes a method to reduce passenger waiting and boarding times by applying a deep Q-network (DQN) algorithm to establish dynamic routes for semi-dynamic DRT systems in urban residential areas. A simulation was conducted to compare the performance of fixed-route and dynamic-route systems, analysing how changes in passenger demand affect waiting and travel times. Results indicate that dynamic routes optimised by the DQN algorithm achieved higher boarding and alighting rates across all demand levels compared with fixed routes. In addition, even under high demand, dynamic-route DRTs reduced waiting and travel times, demonstrating superior efficiency. These findings confirm that dynamic-route DRTs enhance service quality and operational performance in residential areas.