Link To Latest Report : Coming Soon.
Background :
While some recent studies have analyzed interdependencies within and between infrastructures and developed integrated policies for their concurrent restoration , they do not specifically address the effects of pre-disaster fortification – particularly bridge fortification – on post-disaster restoration resilience. Furthermore, prior works mainly rely solely on optimization techniques , limiting their capacity to handle uncertainties that arise during fortification and restoration processes, nonlinear relationships in fortification operations, as well as their ability to manage large-scale, real-world infrastructure.
Objectives :
The research objective is to develop a novel decision-making framework, called RL-empowered Optimizer for Bridge Fortification, that combines and leverages optimization, Markov Decision Processes (MDP), and Reinforcement Learning (RL) to optimize bridge fortification process in road infrastructure of disaster-prone communities. This framework determines the optimal level, technology, and sequence or concurrency for fortifying bridges in a community over a planning period to maximize the post-disaster resilience of road infrastructure.
Scope :
To design the proposed framework, the following tasks will be accomplished:
- Task 1: The inner RL development. A mathematical structure, combining the strengths of day-to-day traffic simulation, optimization, and RL, will be used to model reactions of travelers to road and bridge restoration activities, optimize restoration process for damaged roads and bridges in the road infrastructure, and estimate post-disaster traffic acceleration that can be caused by pre-disaster bridge fortification decisions.
- Task 2: The outer RL development. To design optimal bridge fortification policies for road infrastructures in disaster-prone communities, an RL-based framework will be developed. Considering the resource constraints that necessitate sequential decision-making and the stochastic nature of bridge fortification (e.g., uncertain fortification duration and effectiveness), the agent’s decision-making process will be formulated as an MDP. The inner RL developed in Task 1 will constitute the learning environment of this framework.
- Task 3: Computational speed enhancement. To accelerate the computational capabilities of the decision-making framework, neural networks (NNs) will be trained and embedded to speed up the learning process, increasing the scalability of the framework for application to large-scale transportation infrastructures with high number of bridges.
- Task 4: Framework validation and verification. The framework’s effectiveness will be evaluated through a case study focusing on bridge fortification in the transportation infrastructure of Sioux Falls, South Dakota. The impact of these fortification decisions on the restoration of the city’s transportation infrastructure, disrupted by various tornado scenarios, will be assessed.
- Task 5: Computational platform development. In addition to mathematical foundation,a computational platform (including software code and documentations of the developed computing models) will be developed for the framework, making it accessible to researchers and practitioners with limited optimization and AI expertise.
Research Team :
Principal Investigator : Shabnam Rezapour, Ph.D.
Co-Principal Investigator : Mohammadhadi Amini, Ph.D.