Numerical experiments on variants and parameters of the simulated annealing method

The simulated annealing procedure, commonly used in process system engineering, requires potential solutions to be encoded, and involves several variants in its implementation as well as various control parameters. On the basis of six mathematical functions with known solutions, these points are studied in order to propose general guidelines for using the simulated annealing method. The stochastic acceptance procedure of a solution proposed by Metropolis et al. (1953), combined with the temperature decreasing geometric scheme of Kirkpatrick et al. (1982) are the most efficient variants. For the three codes used, the convergence rates are similar, so it can be noted that they are rather equivalent. The initial temperature, the length of temperature plateau, the decreasing multiplicative factor for the temperature and the stopping criterion are analysed through a design of experiments. The most significant parameter is the length of temperature plateau, the least significant one is the initial temperature value, and the decreasing temperature scheme and the number of temperature plateaux for stopping the search, have a similar influence. (C) 2004 Elsevier B.V. All rights reserved.