The effect of neighborhood structure on simulated annealing

Abstract

This thesis investigates the effect of neighborhood structure on simulated annealing, a random search algorithm that has the ability to search through a sequence of local optimal solutions and return the globally optimal solution. Neighborhood sizes of one, two, three, four and N (equal to the size of the problem) decision variable alterations have been experimentally tested on the following problem types: Quadratic Assignment problem; Quadratic Selection problem; Stochastic Optimization problem; and Traveling Salesman problem. Empirical results indicate that a smaller neighborhood size is better. However, for some instances, a neighborhood size of one larger than the smallest defined neighborhood size performed best.