OPTIMIZATION USING VARIATIONS OF SIMULATED ANNEALING

Simulated annealing (SA) was originally developed for determining global optima for combinatorial optimization problems, but it has been shown to be amenable to continuous functions as well. Difficulties in SA arise when determining proper operating cooling schedule variables for particular optimization problems. Generalized simulated annealing (GSA) reduces many of these complications. However, GSA usually locates near-global optima and not exact optima. If GSA is set up interactively, convergence to the exact global optimum becomes achievable. In order to attain automatic convergence to the exact global optimum, the variable step size generalized simulated annealing (VSGSA) algorithm tunes its operation variables based on data collected during optimization. This tutorial describes the basic operating principles for SA, GSA, and VSGSA, and incorporates worked examples from the literature.