BP算法的改进方向  

The Improving Direction of BP Algorithm

作　　者：王冰山[1] 蔡婷[1] 王志辉[1] 

机构地区：[1]武汉理工大学机电工程学院,湖北武汉430070

出　　处：《孝感学院学报》2005年第6期86-88,共3页JOURNAL OF XIAOGAN UNIVERSITY

摘　　要：BP是对非线性可微分函数进行权值训练的多层前向网络。BP算法利用了神经网络既能考虑定量和定性因素,又具有自学习、自组织功能的特点。它在控制行业中的应用很有效果,但也有其致命的缺点。通过JAVA语言实现BP算法,并通过实例说明了BP算法的特点,且提出了改进方向。BP is a forward broadcasting multiply-layer-network that trains the weight of the differentiable non-linear functions. BP algorithm uses the the characteristic of considering of the quantitative and qualitative factor, and the self-study and self-organization function of neural network. Its application in controlling industry is effective, but it also has fatal drawbacks. The article actualizes BP algorithm with JAVA and illustrates the features of BP. It also puts forward the improving direction.

关 键 词：人工神经网络 BP算法 改进方向 初始权值 学习步长 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]