No abstract

The genetic algorithm is the intelligence computational method widely used in solving the optimization problems, however many current genetic algorithms has premature convergence and stagnation behavior while solving complex question. This may because the algorithm had lost the ability of discovering that new or potential genetic material before convergence limited its search to solution space with wide range. Enlightened by hierarchical fair competition widely existed in nature, this paper has studied a model based on Pyramid model of hierarchical fair competition which uses pipeline structure divided sub-population rank according to fitness gradient, guarantee the recent discovery potential genetic material to obtain full development by maintaining a certain global selection pressure under reducing the local selection pressure of sub-population.Unlike traditional genetic algorithm's strategy which attempts to jumps out local optima region from high evolutional population included a high similarity "building block", the Pyramid model continuously maintain the sub-population with the medium fitness value to ensure keeping the multiplicity of the population by the strategy which the new local optima is bottom-up bred and processed unceasingly.Finally, this paper confirms the validity of the algorithm and the multiplicity of the population by HIFF128/256 problems based on binary encoding, and compared the improved algorithm from the solution quality and standard deviation of the population with traditional genetic algorithm, proves this improved algorithm can effectively alleviate premature convergence problem of the traditional genetic algorithm.