State University of New York at Buffalo
A Machine Learning-Based Hurricane-Resilient City
Abstract
dc:description.abstractA city resilient to hurricanes would not break but will bend from their associated hazards (e.g., wind, rain and surge). This city bending and bouncing back (to some extent) result in increasingly significant losses due to the intensified hurricane wind-rain-surge extremes under changing climate and the continued escalation of coastal population density. Therefore, both long-term efforts (e.g., hurricane mitigation and climate adaptation) and short-term efforts (e.g., pre-hurricane preparedness, during-hurricane response and post-hurricane repair) are urgently needed to enhance city resilience by collectively improving the ability of a city to adapt to changing conditions, and to withstand and recover rapidly from disruptions caused by hurricane hazards. This study focuses on the short-term efforts to systematically advance hurricane resilience of urban built environment, where the problem of optimizing stakeholders’ action plans for resilience enhancement is mathematically formulated as a Markov decision process and approached by the deep reinforcement learning algorithms. The component-level preparedness in pre-hurricane phase is first implemented on individual hurricane-sensitive structures to reduce their failure probabilities and enhance city robustness. Specifically, a deep reinforcement learning-based framework of aerodynamic shape and control optimization is built as the foundation of future morphing tall buildings and long-span bridges to mitigate structural response under approaching hurricane wind. Then, the system-level response in during-hurricane phase is performed within an isolated critical infrastructure network to utilize its redundancy and hence allow the city to operate at a lower capacity. Specifically, a deep reinforcement learning-based traffic management system of interdependent transportation infrastructure components is constructed for coordinated relocation of vehicles to minimize travel time and injuries/fatalities under predicted hurricane damages caused by wind, rain and storm surge. Furthermore, the system-of-system-level repair in post-hurricane phase is conducted on interconnected critical infrastructure systems to utilize stakeholders’ resourcefulness and hence enhance restoration rapidity of city functionality. Specifically, a deep reinforcement learning-based repair scheduling system for coupled traffic-electric power networks is designed for planning collaborative actions to accelerate recovery for observed hurricane damages. To improve the training efficiency and model interpretability of deep reinforcement learning algorithms, the knowledge-enhanced machine learning methodology is adopted in this study for effective risk-informed decision making. Finally, a multi-phase multi-scale resilience enhancement framework is developed to investigate the contribution of the individual short-term efforts (involving different structure/infrastructure types in various hurricane life-cycle phases) to the overall resilience performance.
Degree
thesis:*- Grantor dc:publisher
- State University of New York at Buffalo
- Year dc:date.issued
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Li, Shaopeng
- Contributors dc:contributor
-
- Wu, Teng
- Civil, Structural and Environmental Engineering
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- Copyright retained by author. Users of works found in University at Buffalo Institutional Repository (UBIR) are responsible for identifying and contacting the copyright owner for permission to reuse. University at Buffalo Libraries do not manage rights for copyright-protected works and cannot assist with permissions.
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/10477/88694