A Self Controlled Simulated Annealing Algorithm using Hidden Markov Model State Classification

Abstract

The Simulated Annealing (SA) is a stochastic local search algorithm. Its efficiency involves the adaptation of the cooling law. In this paper, we integrate Hidden Markov Model (HMM) in SA to adapt the geometric cooling law at each iteration, based on the history of the search. This approach allows to controls the cooling of SA during the run, based on sequence of state generated from a set of rules. The HMM parameters are calculated and updated at each cooling step. The Viterbi algorithm is then used to classify the observed sequence as an exploration or exploitation or an escape from the local minimum. An experiments was performed on many benchmark functions and compared with others SA variants.