Bayesian Ranking and Selection Model for Multi-Objective Discrete Network Design Problem with Uncertainties

The design of sustainable transportation systems calls for multi-objective models which accounts for congestion, environmental and energy objective in parallel. In this study the authors present a Bayesian Ranking and Selection (R&S) model for the Multi-Objective discrete Network Design Problem with Uncertainty (MONDPU). In this formulation, each solution to the MONDPU problem represents an “alternative”. The expected objective values of each alternative represents a vector “reward” which the authors estimate and maintain through parametric Bayesian beliefs. Uncertainties are modeled by independent normal distributions on each alternative. They iteratively update their belief about the objective functions and select the next sample based on information from previous iterations. The authors define a multi-objective version of the Knowledge Gradient sampling policy and apply a surrogate-assisted approach with a crowding distance metric to ensure the computational efficiency and diversity of the final solution set. Case studies are conducted on the Sioux Falls network and Anaheim network. Results showed that the authors' Bayesian R&S model is able to identify a very diverse set of highly optimal solutions under very limited computational budget and high levels of uncertainty, significantly outperforming the bench-marking NSGA-II algorithm in both convergence speed and coverage of the Pareto optimal set. The model provides a highly practical framework for network designers/policy makers to make informed multi-criteria decisions. It also extends the Bayesian R&S model and the knowledge gradient sampling policies to generic multi-objective cases, which provides valuable insights for a large class of similar optimization and learning problems.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ADB30(7) Paper Review Group #3.
• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Wang, Xun (Richard)
  • Gao, H Oliver
• Conference:
  • Transportation Research Board 93rd Annual Meeting
  • Location: Washington DC
  • Date: 2014-1-12 to 2014-1-16
• Date: 2014

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: 18p
• Monograph Title: TRB 93rd Annual Meeting Compendium of Papers

Subject/Index Terms

• TRT Terms: Bayes' theorem; Case studies; Design; Networks; Ranking (Statistics); Sustainable transportation; Uncertainty
• Geographic Terms: Anaheim (California); Sioux Falls (South Dakota)
• Subject Areas: Planning and Forecasting; Public Transportation; I72: Traffic and Transport Planning;

Filing Info

• Accession Number: 01520112
• Record Type: Publication
• Report/Paper Numbers: 14-5026
• Files: TRIS, TRB, ATRI
• Created Date: Mar 26 2014 10:13AM