【摘要】 简单遗传算法仅靠变异产生新的数值,故其搜索精确度不高、收敛速度慢,稳定性差,且易发生"早熟"现象。为解决上述问题,利用替代的方法确保每代中最适应的个体总是被传播到下一代,有条件地保留了最佳个体;并采用子种群迁移策略,引进新种群,解决种群中个体多样性的问题,既保证个体的多样性,也提高解的精确度。通过对多峰函数进行仿真,结果表明采用新的策略后,GA的收敛快速性和全局收敛性都有了明显的改善。将改进的遗传算法用于优化冷轧轧制力预报模型中BP神经网络的参数,仿真结果表明改进型遗传算法改善了标准遗传算法的局部搜索能力,同时,也加快了算法的收敛速度。经优化的BP网络预报精度(0%~6%)较传统理论计算精度(0%~44%)也有明显提高。更多还原

【Abstract】 Simple genetic algorithm(SGA) produces a new value only based on the mutation operator,it often obtains a solution without high precision.Moreover,SGAs instinct deficiency like the unusual slow convergence,bad stability and easily-oriented prematurity have become the biggest obstacle for its further application.To solve these problems,a method of replacement and migration to reserve the best individual is proposed,maintain the diversity of the population,and increase the precision of the solution.According to the simulation of multimodal function,and the optimization of BP neural network in prediction model of rolling force for tandem cold rolling mill,the results show that the convergent speed and global convergence are clearly improved.Besides,the predicted accuracy(0~6 %) of the optimized BP network is higher than the theoretical accuracy(0~44 %).更多还原