Unraveling High-Granularity Day-to-day Network Impact of Cascading Disruptions to Transportation Infrastructure and Services

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

The objective of this project is to support research on a novel paradigm for estimating, predicting, and managing traffic in response to cascading failures. Cascading failures occur when disruptions in critical components (e.g., bridge collapses, highway failures, or major public transit disruptions) trigger widespread ripple effects across inter-connected infrastructure and service networks. Leveraging low-cost, ubiquitous, system-level data, this project seeks to understand how and why travelers alter their routes and modes during extreme events. The theories and models are further validated with case studies in Pittsburgh, Pennsylvania, and Baltimore, Maryland. This project has the potential to enhance resilience of the nation’s critical infrastructure and lifeline services, ultimately saving lives and reducing economic losses. It supports development of efficient strategies for emergency response, daily operations, and long-term planning. The project promotes interdisciplinary education by integrating findings into undergraduate and graduate curricula, offering an online short course, and mentoring researchers. Project outcomes are broadly disseminated through open-source software, tutorials, international conferences, and policy briefs - ensuring benefits for government agencies, industry stakeholders, and the public. Research funded by this project seeks to develop a theoretical and computational framework to model spatiotemporal vehicular and passenger flow under cascading disruptions. It leverages high-granularity, ubiquitous, multi-year, system-level data (e.g., 5-min traffic speeds, transit vehicle locations, passenger counts, and high-resolution satellite imagery) to develop travel behavior models that capture travelers’ stochastic choices in mode, route, and departure time. By integrating both within-day dynamics and day-to-day adaptations, the project looks to fuse behavioral models with mesoscopic, multimodal network flow simulations. This integration is enabled by a novel computational graph (CG) design, which learns high-dimensional parameters to replicate observed flows, across large-scale networks and different disruption stages - immediate impact, partial re-opening, and full recovery. This framework seeks to facilitate the discovery of fundamental knowledge in how travelers respond to large-scale infrastructure and service failures, establishing a methodological foundation for traffic management under cascading disruptions. Replicable and transferable machine-learning based computational tools are provided to inform public agencies to design and implement mitigation measures. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.