摘要
基于梯度下降的神经网络训练算法易于陷入局部最小,从而使网络不能对输入模式进行准确分类。本文提出综合遗传算法和BP算法的杂交算法GA-QP,它结合遗传算法的全局搜索特性和BP的局部收敛特性,实现对神经网络的有效训练。实验表明该算法优于BP算法,实验结果令人满意。
Neural network BP training algorithm based on gradient descend technique may lead to entrapment in local optimum so that the network inaccurately classifies input patterns. This paper presents a hybrid training algorithm GA-QP combined to genetic algorithm with BP algorithm. Experiments show that the hybrid algorithm outperforms BP algorithm. Satisfactory experimental results are obtained.
基金
国家教委博士点专项基金9361403号
中科院自动化所国家模式识别重点实验室基金
关键词
神经网络
杂交训练算法
遗传算法
多层感知机
Neural network, Hybrid training algorithm, Genetic algorithm, Quickprop algorithm, BP algorithm
