MiGSA: A new simulated annealing algorithm with mixture distribution as generating function
Seyed Hanif Mirhosseini, Hasan Yarmohamadi, Jahanshah Kabudian · 2014
One of the optimization algorithms for multi-dimensional functions is simulated annealing. In this paper, a modified simulated annealing (SA) is proposed which utilizes a memory to keep best-so-far met (visited) states/solutions. One of the worst flaws of standard SA is its tendency of oblivion and the chance of losing good points. For avoiding this defect, we use a mixture probability distribution function based on saved previous good solutions (memory) to elect next state. The best-so-far solutions are center (mean vectors) of the mixture probability distribution. So we name this approach MiGSA (Mixture Generating function Simulated Annealing). Our experiments indicate that this approach can improve convergence and stability and avoid delusive areas in benchmark functions better than SA. Each element of mixture generating function can be of Gaussian type (in Boltzmann Annealing case), Cauchy type (in Fast Annealing case) or any other type of distribution.