Authors: Mohammed Talat Khouj; Abdullah Alsubaie; Khaled Alutaibi; Haitham Magdi Ahmed; Sarbjit Sarkaria; José R. Martí
Addresses: Department of Industrial Engineering, University of Business and Technology, Dahban-North Jeddah, Saudi Arabia ' Energy and Water Research Institute, King Abdulaziz City for Science and Technology, Riyadh, Saudi Arabia ' Department of Electrical Engineering, University of British Columbia, 2332 Main Mall, Vancouver, BC, V6T1Z4, Canada ' Department of Mining Engineering, King Abdulaziz University, Jeddah, Saudi Arabia ' Department of Electrical Engineering, University of British Columbia, 2332 Main Mall, Vancouver, BC, V6T1Z4, Canada ' Department of Electrical Engineering, University of British Columbia, 2332 Main Mall, Vancouver, BC, V6T1Z4, Canada
Abstract: Disaster mitigation of severe catastrophic events depend heavily on effective decisions that are made by officials. The goal of disaster management is to make decisions that properly reallocate and redistribute the scarce resources produced by the available interconnected-critical infrastructures (CI's). This paper investigates the application of Monte Carlo (MC)-based policy estimation in reinforcement learning (RL) to mount up experience from a massive number of simulations. This method, in conjunction with an optimised set of RL parameters, will help the RL agent to explore and exploit those trajectories that lead to an optimum result in a reasonable time. It shows that a learning agent using MC estimation policy, through interactions with an environment of simulated disastrous scenarios (i2Sim-infrastrucuture interdependency simulator) is capable of making informed decisions for complex systems in a timely manner.
Keywords: artificial intelligence; critical infrastructure; disaster management; i2Sim real-time simulator; reinforcement-learning agent; responsive crisis management; Monte Carlo policy estimation; decision support system; agent based modelling; machine learning.
International Journal of Critical Infrastructures, 2018 Vol.14 No.4, pp.375 - 399
Received: 28 Oct 2017
Accepted: 30 May 2018
Published online: 03 Oct 2018 *