文档介绍：摘要神经网络是一门发展十分迅速的交叉学科,它是由大量的处理单元组成非线性的大规模自适应动力系统。神经网络具有分布式存储、并行处理、高容错能力以及良好的自学****自适应、联想等特点。目前已经提出了多种训练算法和网络模型,其中应用最广泛的是前馈型神经网络。前馈型神经网络训练中使用最多的方法是误差反向传播(BP) 学****算法。但随着使用的广泛,人们发现 BP 网络存在收敛速度缓慢、易陷入局部极小等缺陷。于是我们就可以分析其产生问题的原因, 从收敛速度和局部极小两个方面分别提出改进的 BP网络训练方法。关键字:神经网络,收敛速度,局部极小, BP网络,改进方法 Abstract work isa cross discipline which now developing very rapidly, it is the nonlinearity adaptive power system which made upby abundant of the processing units . The work has features such as distributed storage, parallel processing, high tolerance and good self-learning, adaptive, associate, etc. Currently various training algorithm work model have been proposed , which the most widely used type is Feedforward work model. Feedforward work training type used in most of the method is back-propagation (BP) algorithm. But with the use of work, people find that the convergence speed is slow, and easy fall into the local minimum. So we can analyze the causes of problems, from the two aspects respectively we can improve the BP training methods of work. Keywords : work , convergence speed , local minimum , BP wor k improving methods 目录 1 神经网络概述........................................................................ 3 ......................................................................................... 3 ......................................................................................... 3 2 BP 神经网络........................................................................... 5 BP神经网络特点........................................................................................... 5 BP神经网络介绍........................................................................................... 6 3 BP 神经网络算法的改进...................................................... 8 BP神经网络训练过程介绍..................................................................................... 8 动态调节学****率的改进方法.................................................................................... 8 BP神经网络收敛速