Multi-step GA for better results
GA (Genetic Algorithm) returns a little different optimization results each time GA runs, caused by its stochastic nature. How do we get better results which means the optimization result as close to the actual minimum/maximum as possible by GA? The paper presents a trial method, as a practical optimizing strategy for GA, called multi-step GA to get the better optimization result than that result calculated by GA running once. The multi-step GA introduces the two variable 'maxStep' and 'maxRound' as the new stopping conditions to specify the number of iterations so that GA can get better results in limited trial steps. It is proved that the method is an effective method for GA to increase accuracy generally by three different experiments regardless of optimization problems and varieties of GA. The method presented in this paper improves the accuracy of GA in whole or externally instead of improving the component parts of GA internally. It let the evolution of the world of the genome finish completely, and replays the world again and again to find the optimal result of that world.