Optimal simulated-annealing method based on stochastic-dynamic programming

A new simulated-annealing method based on stochastic-dynamic programming is proposed to obtain a rapid convergence to a global minimum of multivariable optimization problems. A central concern is to provide appropriate temperature values that determine the convergence of the cooling schedule. By use of the stochastic-dynamic programming method, the present cooling schedule is derived by minimizing the time that the system requires to reach the global minimum from an initial state. Monte Carlo simulation shows that the present schedule gives a convergence 10 to 100 times faster than traditional schedules.