Methods to reduce dimensionality and identify candidate solutions in multi-objective signal timing problems

Open Access
Hitchcock, Owen Martin
Graduate Program:
Civil Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 27, 2018
Committee Members:
  • Vikash Gayah, Thesis Advisor
  • Eric Donnell, Committee Member
  • S. Ilgin Guler, Committee Member
  • multi-objective optimization
  • traffic signal timing
  • optimal solution
Adjusting signal timings at signalized intersections is a practical way for transportation agencies to manage traffic without the need for significant infrastructure upgrades or additions. Traffic signal timings are directly related to the delay vehicles experience at signalized intersections. Typically, traffic engineers select signal timings to minimize the total delay experienced by vehicles at the entire intersection. At first glance, this is a fairly reasonable thing to do—minimizing the total delay at the intersection reduces the negative impacts imparted onto cars. However, there are many other possible objectives to consider when selecting signal timings at a signalized intersection. An engineer may wish to consider the approach or movement delay separately and assign a weight of importance to each. For example, one approach may experience heavy bus traffic, and thus its delay might need to be weighted more heavily to reduce the total delay experienced by all passengers served by the intersection. Or, by increasing the delay of one approach by a small amount, the delay of another approach may decrease by a large amount thus providing a more equitable distribution of delay at the intersection. These and other cases are generally not considered under current signal timing standard practice. Even when they are, no methodology exists to incorporate multiple objectives into signal timing optimization. In light of this, the goal of this research is to develop methods that traffic engineers can use to optimize signal timings while considering multiple, potentially competing, objectives. The methods proposed rely on the application of a well-known multi-objective optimization (MOO) genetic algorithm, NSGA-II, to obtain a set of signal timings that consider multiple objectives that may be relevant when selecting signal timings. The set of possible signal timings obtained by the MOO represents a Pareto frontier that defines the optimal tradeoffs that exist between the unique objectives. The research applies relatively new MOO visualization techniques to easily explore the tradeoffs between objectives that exist within this Pareto frontier and proposes new techniques to identify and remove objectives that might not be necessary. Additionally, methods are proposed to select the best solution in the Pareto frontier based on how a user values each of the potentially competing objectives. These methods will allow transportation agencies to obtain signal timings that provide the best tradeoff between objectives that are defined for any particular location.