Abstract: Aiming at the balance of search results and search speed of evolutionary algorithm, we proposed a search strategy to classify the individuals by the similarity of their fitness. This differentiated respective function of individuals in search process. Nevertheless,premature convergence was one of GA-difficulties. So,an improved selection mechanism in GA was used to deal with the mentioned drawback. On the one hand,a new parameter named success ratio which is higher setting causes higher selection pressure. It could keep the algorithm from premature convergence. On the other hand, another parameter T the same as the simulated annealing algorithm's temperature was recommended. While mutation and crossover happened,a virtual population could be generated according to this parameter for enlargeing the search space and keeping the diversity of generation. Finally, experimental results on benchmark problems of TSP show that the new method is capable of producing highest quality solutions and preventing premature convergence efficiently.

Key words: Classification, Genetic algorithm, Self-adaptive, Selection mechanism