The Optimization of Traffic Flow and Resource Allocation in Urban Infrastructures Using Predictive Analytics and Multi Agent Systems

This research work presents a novel Multi-Agent Reinforcement Learning (MARL) model for optimising urban traffic flow (UTF) and resource allocation. The proposed MARL integrates specialised agents for traffic prediction, intersection control, resource allocation, and incident detection, all of which are coordinated through a centralised command structure. This work formulates the traffic management problem as a dual-objective optimisation task, simultaneously minimising congestion and optimising resource allocation. The model employs Proximal Policy Optimization (PPO) for training agents, enabling efficient real-time decision-making and adaptation to dynamic traffic conditions. This MARL's performance was evaluated using the Simulation of Urban Mobility (SUMO) traffic simulator, which features a network comprising 25 road segments and 8 resource allocation regions. Simulation results validate significant improvements over traditional methods, including a 47.4% reduction in average travel time, a 40.2% increase in network throughput, and a 35.7% improvement in resource utilisation. During peak hours, the model achieved a 51.2% reduction in queue lengths and maintained a 92.4% resource utilisation rate under increased demand. The Incident Detection System (IDS) proved 94.5% accuracy with an average response time of 4.3 minutes, significantly outperforming baseline approaches. The empirical results of this study indicate that the proposed MARL provides a robust and scalable solution for UTF, effectively balancing traffic flow optimisation with resource allocation efficiency. The system's ability to maintain performance under varying traffic conditions and incident scenarios recommends its viability for real-world implementation in smart city traffic management systems.