BP神经网络隐含层节点数确定方法研究

摘要

在BP神经网络的众多参数中,隐含层节点数是其中一个非常重要的参数,它的设置对BP神经网络的性能影响很大,而且是导致"过拟合"现象的直接原因.目前理论上还不存在一种科学普遍的用于确定隐含层节点数的方法,应用时只是凭借设计者以往的经验以及借助多次实验进行确定,因此无法高效地获得隐含层节点数.针对BP神经网络隐含层节点数的确定问题,提出一种"三分法"算法,用于快速确定BP神经网络的隐含层节点数的最优解.在Wine-data数据集上,通过Matlab仿真实验验证了"三分法"算法比传统方法获取隐含层节点数的效率提高了1.8倍,是一种行之有效的方法.%In the many parameters of BP neural networks,the number of nodes in the hidden layer is a very important one,which has a great influence on the performance of BP neural networks,and it is the immediate cause of the phenomenon of over-fitting.At present, there is no scientific and universal method to determine the number of nodes in the hidden layer.In application it is determined just by the designer's experience and the help of many experiments,therefore the number of hidden layer nodes can't be obtained efficiently.Ai-ming at the problem,we propose a ternary algorithm to rapidly determine the optimal solution of the number of hidden layer nodes in BP neural networks.In the data set of Wine-data,it is proved that the efficiency of the proposed method which is effective is 1.8 times high-er than the traditional method through Matlab simulation.