Incorporating Twinkling in Genetic Algorithms for Global Optimization
Genetic algorithms have been extensively used as a reliable tool for global optimization. However these algorithms suffer from their slow convergence. To address this limitation, this paper proposes a two-fold approach to address these limitations. The first approach is to introduce a twinkling process within the crossover phase of a genetic algorithm. Twinkling can be incorporated within any standard algorithm by introducing a controlled random deviation from its standard progression to avoiding being trapped at a local minimum. The second approach is to introduce a crossover technique: the weighted average normally-distributed arithmetic crossover that is shown to enhance the rate of convergence. Two possible twinkling genetic algorithms are proposed. The performance of the proposed algorithms is successfully compared to simple genetic algorithms using various standard mathematical and engineering design problems. The twinkling genetic algorithms show their ability to consistently reach known global minima, rather than nearby sub-optimal points with a competitive rate of convergence.