Genetic algorithms (GAs) and simulated annealing (SA) have emerged as the leading methodologies for search and optimization problems in high dimensional spaces. A simple scheme of using simulated-annealing mutation (SAM) and recombination (SAR) as operators use the SA stochastic acceptance function internally to limit adverse moves. This is shown to solve two key problems in GA optimization, i.e., populations can be kept small, and hill-climbing in the later phase of the search is facilitated. The implementation of this algorithm within an existing GA environment is shown to be trivial, allowing the system to operate as pure SA (or iterated SA), pure GA, or in various hybrid modes. The performance of the algorithm is tested on various large-scale applications, including DeJong's functions, a 100-city traveling-salesman problem, and the optimization of weights in a feedforward neural network. The hybrid algorithm is seen to improve on pure GA in two ways, i.e., better solutions for a given number of evaluations, and more consistency over many runs.
展开▼
