Safety-Aware Reinforcement Learning Model for Adaptive Traffic Signal Optimization in Work Zone Environments

Work zones reduce roadway capacity and create unstable merging, queue spillback, and stop-and-go conditions that simultaneously degrade traffic operations and elevate crash risk. Conventional fixed-time, actuated, and adaptive signal controllers are poorly suited to these non-stationary conditions, and most reinforcement learning (RL) approaches optimize for mobility while treating safety only as a post-hoc evaluation measure. This study develops a safety-aware Deep Q-Network (DQN) framework for adaptive signal control at intersections operating near work zone activity areas. The framework embeds merge conflict risk, upstream spillback propagation, and stop-and-go instability directly into both the state representation and the reward formulation, alongside operational objectives such as delay, throughput, and speed. A merge-conflict model based on relative spacing, relative speed, and acceleration characterizes unsafe interactions in the merge region, and a Pareto-based multi-objective procedure samples reward-weight vectors to identify non-dominated policies balancing efficiency and safety. The framework was implemented and evaluated in a SUMO microscopic simulation of a signalized intersection under lane-closure conditions. The best policy increased throughput by 24.6%, 32.7%, 37.3%, and 29.7% for cars, trucks, buses, and mixed traffic relative to default timing (all p < 0.001; Cohen’s d = 0.53 - 1.29), with the largest gains for trucks and buses. Shockwave analysis showed a 39.1% reduction in maximum queue length and a 45.8% reduction in spillback distance, with faster queue dissipation. The results indicate that encoding surrogate safety indicators as learning objectives, rather than evaluation criteria, enables a single controller to jointly improve mobility and safety in work zones.