A Self-Tuned Simulated Annealing Algorithm using Hidden Markov Model

Simulated Annealing algorithm (SA) is a well-known probabilistic heuristic. It mimics the annealing process in metallurgy to approximate the global minimum of an optimization problem. The SA has many parameters which need to be tuned manually when applied to a specific problem. The tuning may be difficult and time-consuming. This paper aims to overcome this difficulty by using a self-tuning approach based on a machine learning algorithm called Hidden Markov Model (HMM). The main idea is allowing the SA to adapt his own cooling law at each iteration, according to the search history. An experiment was performed on many benchmark functions to show the efficiency of this approach compared to the classical one.