Antifragile perimeter control: Anticipating and gaining from disruptions with reinforcement learning

Abstract

The optimal operation of transportation networks is often susceptible to unexpected disruptions, such as traffic incidents and social events. Many established control strategies rely on mathematical models that struggle to cope with real-world uncertainties, leading to a significant decline in effectiveness when faced with substantial disruptions. While previous research works have dedicated efforts to improving the robustness or resilience of transportation systems against disruptions, this paper applies the cutting-edge concept of antifragility to better design a traffic control strategy for urban road networks. Antifragility sets itself apart from robustness and resilience as it represents a system's ability to not only withstand stressors, shocks, and volatility but also thrive and enhance performance in the presence of such adversarial events. Hence, modern transportation systems call for solutions that are antifragile. In this work, we propose a model-free deep Reinforcement Learning (RL) scheme to control a two-region urban traffic perimeter network. The system exploits the learning capability of RL under disruptions to achieve antifragility. By monitoring the change rate and curvature of the traffic state with the RL framework, the proposed algorithm anticipates imminent disruptions. An additional term is also integrated into the RL algorithm as redundancy to improve the performance under disruption scenarios. When compared to a state-of-the-art model predictive control approach and a state-of-the-art RL algorithm, our proposed method demonstrates two antifragility-related properties: (a) gradual performance improvement under disruptions of constant magnitude; and (b) increasingly superior performance under growing disruptions.