Center-point-based Simulated Annealing

The S-metaheuristic algorithms work with a single candidate-solution during the search process. That is why they are prone to be trapped in local optima. Many research has being conducted to speed up and also minimize their premature convergence. The Center-point Sampling was introduced by Rahnamayan and Wang in 2008. Based on their experiments, it has shown increase in probability of closeness of the unique point in the center of the search space, to an unknown solution, as the dimensionality of the problem increases. It means, the center is an exceptional point to be used as initial point, specially during solving large-scale black-box problems. In this paper, we investigate this phenomena on Simulated Annealing (SA). The purpose is to accelerate the convergence speed of the algorithm by using the center point as an initial point for SA algorithm. This modified version, called Center-Point-Based SA (CSA), is a very simple and effective idea to enhance SA. The experimental verifications are provided on seven shifted large-scale (i.e., D=300) benchmark functions to show improvements achieved by the CSA algorithm. Using the shifted version of the functions ensures there is no bias towards the center, and so towards CSA algorithm. The results confirm that CSA outperforms parent SA algorithm in overall.