神经网络权值和阈值的优化方法  

Optimum Method About Weights and Thresholds of Neural Network

作　　者：侯祥林[1] 陈长征[2] 虞和济[1] 王铁光[1] 纪盛青[1] 

机构地区：[1]东北大学理学院,沈阳110006 [2]沈阳工业大学建筑工程系,沈阳110021

出　　处：《东北大学学报（自然科学版）》1999年第4期447-450,共4页Journal of Northeastern University(Natural Science)

基　　金：国家自然科学基金

摘　　要：为确定多层神经网络权值和阈值建立了真实的最优化求解方法，即将网络总体平均误差建立为目标函数，以权值和阈值作为设计变量，采用梯度法和共轭梯度法对网络权值和阈值进行优化计算·通过 Ｂ Ｐ 算法、梯度法和共轭梯度法对相同实例网络权值和阈值计算，验证了所提出的优化方法的有效性，实现了权值和阈值的快速准确计算·Real optimum method was built to calculate the weights and threshods of multi layer neural network. Through defining network average error as optimum objective function, weights and thresholds as design variable, network weights and threshods were computed quickly by terraced and conjugate terraced optimizations. Comparing terraced and conjugate terraced optimum methods with BP algorithm, validity of the methods was testified, which results in rapid computing of weights and thresholds.

关 键 词：最优化方法 共轭梯度法 神经网络 权值 阈值 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]