【摘要】 本文针对用GA训练NN权值时 ,花费的代价随精度的提高而剧烈增加的缺陷 ,提出了一种利用IGA较强的全局搜索能力和IBPA较强的局部搜索能力的结合算法 ;先利用IGA优化多层前馈神经网络的权值 ,然后再用IBPA提高搜索精度 ,有效地避免了IBPA易陷入局部极小点和IGA过早收敛的缺点 ,实验结果表明 ,此算法是有效的更多还原

【Abstract】 In this paper, a combined algorithm of utilizing good global search of improved genetic algorithm and good local search of improved backpropagation algorithm is put forward aiming at the shortcoming of the cost rise steeply as the precision is improved when the weights of neural network are trained by use genetic algorithm; the weights of multi_layer feedforward neural network are optimized by using improved genetic algorithm and then search precision is increased by using improved backpropagation algorithm, thus, lost in local minimum of IBPA easily and premature convergence of IGA are avoided effectively. The algorithm is good through experiments.更多还原