Research On The Premature Convergence And Improved Politics Of Genetic Algorithm

Genetic Algorithm is a self-adaptive and self-organizational artificial intelligence which simulates nature evolution to obtain best solution as possible. It has a series of operation, such as selection, crossover and mutation. With these operations, it keeps generating new era population and makes population evolved. When evolution stops, populations can cover best solution. Because algorithm is easy to be realized and application effect is outstanding, Genetic Algorithm is applied in a lot of domains which contain self- adaptive control, combinatorial combination, pattern recognition, machine learning, and life.Some experimental data show that Genetic Algorithms still has some problems. Premature Convergence and ebb local searching is still in presence. People research Genetic Algorithm from point of schema, but blank of theory is still in existence. Now, theory of schema regeneration is little. So theory of schema regeneration is researched and result which Simple Genetic Algorithm hardly regenerates lost scheme is given in paper.For the case of that Simple Genetic Algorithm hardly regenerates lost schema, population loses population diversity and premature Convergence occurs once. So population stops research. To improve capacity of regeneration schema, some policy must be introduced into algorithm. But the policy effect the pace that algorithm convergence. To improve the pace that algorithm converge, other policy should be introduced into algorithm. So resolved Premature Converge, the pace that algorithm converges and capacity of regeneration schema should be improved. Consult other algorithms and some realism phenomena, three kinds of policy are given in paper. Theoretic analysis and experimental data show that three kinds of policy improve capacity of regeneration schema and pace that algorithm converge. So performance of algorithm is improved.Optimization of the neural networks is a complicated process. The result of the neural networks is not most excellent while conventional algorithm optimizes neural networks. So Genetic Algorithm is used to optimize neural networks. Experimental data show that effect of optimization is better.