Coordinated optimization of speed, parking, and turns in an integrated urban traffic system
[EN] Urban traffic congestion remains a major challenge for city mobility. This study addresses this issue by presenting an intelligent traffic control system that unifies speed control, parking management, and turn permissions using Deep Reinforcement Learning with Proximal Policy Optimization (DRL...
| Autores: | , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2026 |
| País: | España |
| Institución: | Universitat Politècnica de València (UPV) |
| Repositorio: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia |
| Idioma: | inglés |
| OAI Identifier: | oai:dnet:riunet______::c6f2ed01c4035f7c9a51f41239fd0a34 |
| Acceso en línea: | https://riunet.upv.es/handle/10251/235238 |
| Access Level: | acceso abierto |
| Palabra clave: | Urban traffic control Conditional turns Congestion reduction Parking management Real-time traffic management |
| Sumario: | [EN] Urban traffic congestion remains a major challenge for city mobility. This study addresses this issue by presenting an intelligent traffic control system that unifies speed control, parking management, and turn permissions using Deep Reinforcement Learning with Proximal Policy Optimization (DRL + PPO). The approach is applied to a Manhattan-style grid of 16 intersections across 4 traffic scenarios and evaluated under three control methods (static, rule-based, and PPO) over approximately 4 M training steps. Using a SUMO-Python simulation environment, the system dynamically adjusts speed limits, parking zones, and intersection turns based on real-time input variables to prevent bottlenecks. The system analyzes multiple traffic indicators to make coordinated decisions and evaluates its effectiveness across diverse congestion patterns. Results show that the PPO system improves traffic performance across all scenarios, demonstrating scalability in terms of traffic demand (within the tested topology), achieving: CO 2 (CO 2 Emissions) down arrow 19-31%, QL (Queue length) down arrow 18-35%, ATT (Average Travel Time) down arrow 8-18%, and TF (Traffic Flow) up arrow 5-7%, compared with rule-based speed adjustment and static baselines. Furthermore, joint management of all three controls outperforms the best individual strategy, achieving additional improvements of 17-32% depending on the case. These findings demonstrate a clear coordination dividend, where unified control of speed, parking, and turns yields superior performance compared to operating them independently, offering a modular and potentially adaptable alternative for the development of more sustainable cities and the intelligent modification of traffic rules. |
|---|