PI: Arun Sathanur

Networked critical infrastructure systems such as in power, transportation, airline, and other systems are vital to the nation’s well-being, economy, and national security. However, these networks are prone to cascading failures resulting in large-scale electrical blackouts, traffic backups, etc. With advances in sensing, communication, and computation, there is a clear opportunity for automated decision-making in the form of AI algorithms to sense the failure sequence early and take appropriate action to mitigate cascading failures. This project uses the road transportation network as an exemplar system to investigate the efficacy of AI methods in reducing traffic delays (and hence energy consumption and emissions) due to accidents and other planned events in a systematic manner.

Specifically, in this project, we are utilizing realistic traffic data on freeway networks in the Seattle area, modeled through the transportation simulator SUMO (Simulation of Urban Mobility) along with AI algorithms such as combinatorial optimization and deep reinforcement learning to study their efficacy in congestion reduction. We have developed a novel OpenAI-gym interface to the SUMO traffic models of interest. This enables us to determine the state of the network and simulate the effect of actions such as variable speed limits and controlled rerouting actions, in the process allowing us to benchmark state-of-the-art deep reinforcement learning models. Building on optimistic early results on small networks, we plan to extend the prototype to further study larger systems that include freeways and arterials and address multiple simultaneous incidents.

A successful proof-of-concept of the efficacy of AI algorithms in mitigating congestion reduction in simulated road networks will pave the way for more detailed investigation into real-world scenarios and potential applications to other networked critical infrastructure systems.