Tuned annealing for optimization

Abstract

The utility and capability of simulated annealing algorithm for generalpurpose engineering optimization is well established since introduced by Kirkpatrick et. al1. Numerous augmentations are proposed to make the algorithm effective in solving specific problems or classes of problems. Some proposed modifications were intended to enhance the performance of the algorithm in certain situations. Some specific research has been devoted to augment the convergence and related behavior of annealing algorithms by modifying its parameters, otherwise known as cooling schedule. Here we introduce an approach to tune the simulated annealing algorithm by combining algorithmic and parametric augmentations. Such tuned algorithm harnesses the benefits inherent in both types of augmentations resulting in a robust optimizer. The concept of ‘reheat’ in SA, is also used as another tune up strategy for the annealing algorithm. The beneficial effects of ‘reheat’ for escaping local optima are demonstrated by the solution of a multimodal optimization problem. Specific augmentations include handling of constraints, fast recovery from infeasible design space, immunization against premature convergence, and a simple but effective cooling schedule. Several representative optimization problems are solved to demonstrate effectiveness of tuning annealing algorithms.