Comparative studies on design of experiments for tuning parameters in a genetic algorithm for a scheduling problem
Arif Arin, Ghaith Rabadi, Resit Unal · International Journal of Experimental Design and Process Optimisation · 2011
Metaheuristic algorithms have shown to work well with large scale optimisation problems as they obtain optimal or near-optimal solutions. However, most metaheuristics have several parameters that need to be tuned before they can reach good results. Design of experiments (DoE) methods offer practical approaches to tune the parameters effectively. In this paper, we seek the best parameter setting for a genetic algorithm (GA) that is developed to solve the single machine total weighted tardiness problem. To tune the GA parameters, multiple DOE methods are employed and their results are compared. According to their fitness performances in both single and multiple runs, D-optimal and S/N ratio designs found the best parameter settings among DOE methods presented. Additionally, S/N ratio design showed quite robust results for different problem sizes.