自组织前向神经网络与非线性动态系统模化

Self Organizing Feedforward Neural Network and Modeling of Nonlinear Dynamical System

将自组织学习过程引入到前向网络的训练中 ,提出了一种新的三层前向神经网络的训练方法 .训练过程首先利用自组织分簇算法确定隐含层结点的数目以及权值 ,然后通过求解线性最小二乘问题估计输出层权值 .自组织过程产生的激活权值对输入数据具有一种特征变换的功能 .利用该方法训练的网络可以称之为自组织前向网络 (SOFN) .文中通过实际非线性动态系统建模的例子 ,说明了SOFN网络具有良好性能 .

In this paper a new learning procedure of MLP is presented which named as self organizing feedforward neural Network (SOFN). The optimization of weights is implemented layer by layer. At the stage of training hidden weights, an unsuperivsed self organizing clustering is introduced, then the weights of output layer are estimated by supervised least square algorithms. With self organizing stage, the number of hidden nodes can be determined automatically, furthermore, the hidden layer weights created by clustering work as a feature transformation matrix for input vectors. Two examples are given to show the feasibility and advantages of the approach, which is particularly suitable for modeling of nonlinear dynamical system.