论文部分内容阅读
人工神经网络BP模型是一种常用的建模方法,但仍存在很多问题.通过对BP算法的应用发现,当固定学习率η大于0～1内某一值,将导致网络算法不收敛.本文从数学理论上分析了这一现象产生的内在原因.最后提出两种有效对策:方法是第一次η取小值,随后取较大值,最后取小值;方法二是调整传递函数f(x).通过这两种方法,解决了固定学习率不收敛的问题,并对改进方法进行了实证检验.
BP neural network model is a common modeling method, but there are still many problems.Through the application of BP algorithm found that when the fixed learning rate η is greater than a certain value within 0 ~ 1, the network algorithm will not converge. From the mathematical theory, we analyze the internal causes of this phenomenon.Finally, we propose two effective countermeasures: the first one takes a small value of η, then takes a larger value, and finally takes a small value; the second method adjusts the transfer function f x) By these two methods, we solve the problem that the fixed learning rate does not converge, and empirically test the improvement method.