Data-driven Intelligent Transportation Systems, Internet of Things (IoT), Big Data Integration, Mobility Analytics, Smart City Development

Reference: Multi-Agent Reinforcement Learning for Dynamic Traffic Signal Control in Urban Networks Urban traffic congestion remains a persistent issue in large-scale cities where static and rule-based signal control fails to adapt to dynamic traffic patterns. This study introduces a multi-agent reinforcement learning approach for real-time traffic signal control in urban networks. Each intersection is modeled as an autonomous agent that learns optimal signal policies through continuous interaction with the environment. The proposed framework integrates Graph Neural Network to capture spatial dependencies among intersections and improve coordination across the network. Traffic states are represented using spatio-temporal features derived from real-time data, enabling agents to respond to fluctuations in traffic demand. The learning process is formulated using reinforcement learning, where agents optimize cumulative rewards defined by delay reduction and throughput maximization. Experimental evaluation on simulated urban traffic scenarios shows that the proposed model outperforms conventional fixed-time and single-agent approaches in reducing average waiting time and improving traffic flow stability. The results indicate that incorporating spatial awareness and decentralized learning enhances scalability and robustness in complex traffic environments.