输入模式增强神经网络及应用  

INPUT PATTERN ENHANCING OF FORWARD NEURAL NETWORK

作　　者：李洪奇[1] 李莉[1] 刘文胜[1] 

机构地区：[1]中国石油大学计算机科学与技术系,北京102249

出　　处：《计算机应用与软件》2008年第7期213-215,共3页Computer Applications and Software

摘　　要：监督学习神经网络的学习收敛速度是网络的重要指标。以三元奇偶问题和沉积微相识别问题为例,分析了前向神经网络增强输入模式和常规输入模式的网络响应特点。分别在隐层数不变和隐层节点数不变的条件下,对常规输入模式网络和增强输入模式网络的学习速度进行了对比分析。数值实验和沉积微相应用实例说明输入模式增强网络学习迭代次数少,网络分辨率高。输入模式增强网络隐层节点数和隐层数的可选范围均比常规输入模式网络隐层节点数和隐层数的可选范围大,网络收敛稳定性好。输入模式增强网络的性能明显好于常规输入模式网络的性能。Conver speed is an important index of a neural network with surveillance training. The response characteristics of forward neural network, with normal input patterns or enhanced input patterns are analysed based on three-element odd and even number recognition and mi- crofacices identification. The difference of training speed between normal input pattern network and enhanced input pattern network is dis- cussed with no change of hidden layer number and node number of hidden layers. The iterative number and resolving power of network with en- hanced input patterns are better than those of the network with normal input patterns. The available range of node numbers and layer numbers of an enhanced input pattern network is larger than those of a normal input pattern network. The performance of an enhanced input pattern network is better than that of a normal input pattern network.

关 键 词：神经网络 输入模式增强 收敛速度 节点 隐层 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP311.56[自动化与计算机技术—控制科学与工程]