Loading...

Optimization of Post-disaster Recovery: Integration of Bayesian Network and Agent-based Simulation

Gerami, Borhan | 2024

6 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56930 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Mahsouli, Mojtaba
  7. Abstract:
  8. This thesis proposes a probabilistic framework for evaluating and enhancing the community recovery process following an earthquake occurrence. This framework serves as a recovery decision support tool, aiming to enhance community resilience through the optimal allocation of limited available resources following a disaster occurrence. To this end, Bayesian network and agent-based simulation are integrated, each of which forming one of the modules of the proposed framework. The first module, namely, the Bayesian network, leads to situational awareness of the status of the community after the earthquake by fusing data from various sources. The data, such as the earthquake event characteristics including the magnitude and hypocenter location, as well as infrastructure components damage state and functionalities, can be collected from various sources such as social sensors, physical sensors, aerial imaging, and crowdsourcing. The Bayesian network is a directed acyclic graph that models causal relationships amongst phenomena, such as the relationship between the earthquake intensity, components damage, and other consequences. The Bayesian network fuses heterogeneous and uncertain data collected from various sources through Bayesian updating. As a result, it gradually reduces the uncertainty in the status of the community. Prior to the occurrence of the earthquake, the Bayesian network is constructed by creating nodes for different phenomena, such as the hypocenter location, magnitude, component damage states and functionality thereof. Connections between these nodes are established by arcs that represent the causal relationships. Subsequently, the Bayesian network uses learning techniques to establish conditional probability tables for each node by utilizing random sampling on hazard and risk models. Following the occurrence of an earthquake, the aforementioned data from various sources are fed into to the Bayesian network nodes as evidence, and the posterior probability of components damage and functionality is obtained using Bayesian inference. The inference is conducted using sampling methods that generate samples from the posterior distribution of the nodes in the network, such as the earthquake intensity or component damage states. The second module of the proposed framework, namely, the agent-based simulation, evaluates the community resilience given the posterior status of the community as provided by the Bayesian network. The difference between the simulation of the second module and the simulation in a pre-earthquake resilience analysis is that, in the former, the module uses the posterior samples, which reflect the status of the community given the collected evidence, instead of predicting that status using hazard and risk probabilistic models, i.e., predicting the earthquake intensity and infrastructure component damage and functionality. The agent-based module simulates the process of the community recovery to the normal state, utilizing agent-based modeling within the framework of discrete event simulation, and estimates the distribution of the total cost incurred by the community due to the earthquake event. In this simulation, agents represent the institutions responsible for the community recovery, which make decisions on the priority of various operations, such as inspection, repair, and search and rescue. The proposed framework is capable of evaluating various prioritization strategies and determining the best strategy based on the least cost incurred by the community. The proposed framework is applied as a case study to a virtual city entitled “INSURER City,” which is named after the Center for Infrastructure Sustainability and Resilience Research (INSURER). This city has a population of over 125,000 and comprises more than 17,000 buildings. It also features water, power, healthcare, and transportation infrastructure. In this case study, information enters the decision support system in three periods of time after the earthquake: immediately after the earthquake occurrence, including event information; 32 days after the earthquake occurrence, including evidence from damage and functionality of infrastructure components that is collected from inspection operations; and 500 days after the earthquake occurrence, including information on the status of the progress of various recovery operations. The first module of the proposed framework converts the information to situational awareness and the second module, evaluates 12 different recovery strategies and determines the best strategy through simulation. The proposed framework in this research serves as a decision support tool for emergency managers, enabling them to select the most effective community recovery strategies. This leads to reducing costs, fatalities, recovery duration, and ultimately enhances the resourcefulness component of the community resilience
  9. Keywords:
  10. Bayesian Network ; Decision Making Support System ; Emergency Management ; Probabilistic Modeling ; Infrastructure ; Resilience Index ; Recovery ; Agent Based Simulation ; Situational Awareness

 Digital Object List

 Bookmark

No TOC