An experimental study of the multi-objective Go with the Winners algorithm on the biobjective QAP with correlated flow matrices
Everardo Gutiérrez-López, Carlos A. Brizuela · 2007
This paper studies the multi-objective version of the Go With The Winners algorithm over the well known Quadratic Assignment Problem. The analyzed multi-objective problem is the variant for which we have one distance matrix and several flow matrices, each one associated to a different objective function. The analysis is focused on the influence of flow matrices correlation on the algorithm's behavior. We analyze the algorithm performance as a function of its two design parameters: the random walk length and the number of particles. A brief comparison between our experimental results and the best known up to date results for a set of instances is presented. The main outcome of this study is that for instances with positive flow matrices correlation the parameter controlling the algorithm performance is the random walk length, while the main parameter for negative correlation instances is the number of particles. Additionally, we found a clear relation between the mean random walk effort per particle and the mean solution quality generated by the algorithm.