Improving The Research Capability Of Multi-Conding Genetic Algorithm By Update Of Pattern Base

Genetic Algorithm as once the heuristic optimization method at random of good commonality, strong rude stick is broadly applying to Automation, combination spend especially, the image handles , the manpower life, machine study , fields such as artificial intelligence and engineering design. Especially thinking that searched space is more bigger, the problem is very complicated or priori knowledge of problems region is few, use classic methods for an implement to search the solution question under no proper condition Genetic algorithm has provided one kind of the rational method efficient and having effect's find the solution question.Development of multi-coding genetic algorithm has not been long, but for continuous type variable numerical value optimization problem and large-scale optimization grouping problem, using the reality number as expression gene is especially natural and direct. It's being different from tradition genetic algorithm (binary coding), the chromosome is that a reality value is a vector, and the chromosome length is this vector reality value size. At present, multi-coding genetic algorithm has already aroused more and more experts and scholars' attention, and has got and large amount of application study achievement.Though multi-coding genetic algorithm having a lot of merits, multi-coding genetic algorithm crossing operation can not produce the new gene beyond the parents, whose mutate rate is not likely very big, such genes ate easy to lose. Maintaining certain population diversity is very difficult. The algorithm only is capable to run into local solutions. For keeping population diversity, the algorithm needs enhance population scale, then with population scale expansion, algorithm searching is to be deferred and algorithmic convergence speed is to be affected. Therefore, blindly maintaining population diversity has not been able to improve an algorithm's effectively searches. The paper bases on patter theorem of binary coding genetic algorithm, further analyses pattern processing property of binary coding and multi-coding, according to the analysis, proposes a new pattern gene extraction coefficient to extract excellent gene, then applies immune theory to genetic algorithm, builds a pattern base with the excellent pattern picked from population. Using pattern base to spread excellent pattern gene and update pattern. Above ways lead to orientation of algorithm' searching, and accelerate pattern propagation and repair destroy by crossing and mutation operation, improve evolution capability of genetic algorithm. Finally the simulation results for Multiple Choice Knapsack problem and Multi-Knapsack problem validate the proposed method effectiveness.