用于多层前馈神经网络学习的快速遗传算法  

Rapid Genetic Algorithm Used for Multilayer Feedforward Neural Network Training

作　　者：韩明红[1] 李凡[1] 邓家 

机构地区：[1]北京航空航天大学

出　　处：《航空制造技术》2003年第7期31-34,共4页Aeronautical Manufacturing Technology

基　　金：国家自然科学基金项目

摘　　要：提出了一种用于多层前馈神经网络学习的快速遗传算法 ,论述了其算法的进化过程并对多层前馈神经网络的权值进行了优化 ,对权值的初始化范围以及输入层节点进行了优化选择。仿真试验结果表明 ,该算法收敛速度快 ,网络逼近精度高 ,克服了BP算法易于陷入局部极小的问题。A rapid genetic algorithm used for multilayer feedforward neural network training is put forward. The evolution process of the algorithm is described and the weight of the multilayer feedforward neural network is optimized. The initiation scale of the weight and input layer node point are optimally selected. The test results of simulation show that this algorithm has the advantages in fast convergent speed and high approximation accuracy of the network. And this algorithm overcomes the disadvantage that BP algorithm is easy to being trapped in local optimums.

关 键 词：快速遗传算法 多层前馈神经网络学习 收敛速度 网络逼近精度 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]