DQN-Based Coordinated Traffic Flow Control for Adjacent Intersections

Abstract

Urban traffic congestion remains a major challenge in intelligent transportation systems, particularly in urban cor-ridors containing closely spaced signalized intersections. Conventional fixed-time control schemes operate with predetermined phase durations and do not adapt effectively to dynamic traffic patterns, queue spillback, or interactions between neighboring intersections. Reinforcement Learning (RL) has emerged as a promising approach for adaptive signal control because it enables a controller to learn from direct interaction with the traffic environment. However, a large part of the existing work focuses on isolated intersections, while real-world urban traffic networks require coordinated decision-making across connected nodes. This paper presents a research-based framework for coordinated traffic flow control at adjacent intersections using a Deep Q-Network (DQN) in a SUMOTraCI simulation environment. The study formulates the corridor control problem as a Markov Decision Process in which a coordinated state representation captures queue conditions and active signal phases across neigh-boring intersections. A DQN agent is designed to learn adaptive signal switching policies that account for local congestion as well as downstream traffic conditions. The paper emphasizes methodology, system design, and evaluation strategy rather than reporting fixed numerical claims. The proposed framework is intended to support comparative study against conventional fixed-time control and independent local decision strategies using metrics such as delay, queue length, throughput, and travel-time behavior. The work contributes a structured, simulationdriven, and academically grounded basis for multi-node traffic signal research that can be extended toward larger corridors, multiagent control, and sustainability-aware traffic management.