Using statistical Design of Experiment to decide the effective parameter values for a new hybrid ant colony algorithm

Abstract

This paper introduces a newly developed hybrid algorithm for solving Quadratic Assignment Problems (QAPs) and presents the results of a Design of Experiment (DOE) conducted to determine the effective values for all the important parameters used in the algorithm. This hybrid algorithm (called GenANT) is a combination between Ant Colony Optimisation (ACO) and Genetic Algorithm (GA). Although ACO and GA maintain their original parameters in GenANT, new values for their parameters are required so that each of them can work hand-in-hand to achieve optimal performance in solving QAPs. These parameters are: pheromone persistence factor, ρ; factor of Minimum Pheromone Threshold (MPT) adjustment, θ; interval for MPT adjustment, π; genetic crossover rate. In order to determine their effective values, a fractional factorial DOE was run and the outputs obtained were then analysed. The main advantage of using DOE is its ability to provide values for good parameter setting, based on statistical analysis.