一种优化多层前馈神经网络中隐节点数的算法  

An improved algorithm on hidden nodes in multi-layer feed-forward neural networks

作　　者：周红晓[1] 蔡俊[2] 任德官[2] 

机构地区：[1]金华职业技术学院经贸分院,浙江金华321007 [2]浙江师范大学计算机科学与工程学院,浙江金华321004

出　　处：《浙江师范大学学报（自然科学版）》2002年第3期268-271,共4页Journal of Zhejiang Normal University:Natural Sciences

摘　　要：多层神经元网络中隐含层的层数及其每一层的神经元数的多寡 ,至今仍无法用一个解析式精确求得 .提出了一种基于黄金分割法的算法 ,用于求解三层前馈神经网络中的隐层节点数 ,所得结果有助于提高三层前馈神经网络整体品质 .So far there is still no analytic formula that can precisely solve the quantity of the neural elements and the numbers of hidden layers in the multi-layer neural networks. The author in this essay put forward an algorithm based on the golden section to solve the numbers of hidden nodes in three-layer BP network. The outcome is conductive to improve the whole quality of three-layer BP network.The result of simulation indicates that people can find the suitable quantity of hidden layers nodes quickly using the algorithm.

关 键 词：多层前馈神经网络 BP算法 隐层节点数 黄金分割法 MATLAB 神经网络工具箱 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]