改进的BP神经网络算法及其应用  

Optimized BP Neural Networks Algorithm and Its Application

作　　者：冯立颖[1] 

机构地区：[1]燕山大学图书馆,河北秦皇岛066004

出　　处：《计算机仿真》2010年第12期172-175,共4页Computer Simulation

基　　金：国家自然科学基金资助项目(50675189);河北省自然科学基金(F2006000267)

摘　　要：研究提高神经网络算法的快速性和稳定性问题,针对BP算法收敛速度慢且易陷入局部极小值的缺点,分析了一般改进算法在神经网络结构优化过程中存在的问题,并根据遗传算法的特点,提出了一种改进的基于压缩映射遗传的BP神经网络优化方法。算法通过引入泛函分析中的压缩映射原理,不但解决了BP算法收敛速度慢且易陷入局部极小值的缺陷,加快了BP网络的收敛速度,而且还弥补了BP神经网络在学习过程中与网络连接权值初值选择密切相关的不足。与传统的BP算法相比,训练步数减少了17.4387%,训练时间节省了8.2774%。实验结果表明改进的方法取得了良好的效果,可应用于实践中。Aiming at the insufficiency of slow convergence rate and easily immerging in local minimum frequently in the BP algorithm,the problems that exist in the optimization process of neural networks were analyzed.According to the characteristics of genetic algorithms,a method of BP neural networks optimization based on contractive mapping genetic algorithm is proposed.The optimized method by using contractive mapping principle in functional analysis,not only resolves the insufficiency of slow convergence rate and easily immerges in local minimum frequently in the BP algorithm and raises the training speed,but also effectively compensates the defect that the BP neural network and the initial network connection weights closely related in the learning process.It can save the training epochs about 17.4387% and the training time about 8.2774% in the experimentations than the traditional BP algorithm.Experimental results demonstrate that the method is effective.

关 键 词：神经网络 反向传播算法 压缩映射 优化 适应度 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]