A new population-based simulated annealing algorithm

In this paper, we propose sequential Monte Carlo simulated annealing (SMC-SA), a population-based simulated annealing algorithm, for continuous global optimization. SMC-SA incorporates the sequential Monte Carlo method to track the converging sequence of Boltzmann distributions in simulated annealing, such that the empirical distribution will converge weakly to the uniform distribution on the set of global optima. Numerical results show that SMC-SA is a great improvement of the standard simulated annealing on all test problems and outperforms the popular cross-entropy method on badly-scaled objective functions.