小波神经网络结构设计新算法

Novel Algorithm of Wavelet Network Structural Design

依据训练数据集，构造多分辨率的小波节点库，再根据输入给出小波节点的输出向量．在此基础上，把一种非线性动态系统模型结构确定和参数估计方法与小波网络相结合，提出一种新的小波网络学习算法．该算法权衡网络的规模和精度两方面因素，自动地确定网络的节点数目，可以得到在ＢＩＣ准则下最优的小波神经网络．仿真结果表明，用本算法设计得到的小波神经网络具有较小的网络规模。

According to the input of sample set, a function set of multiresolution wavelet node was obtained and the output vectors of each node were computed. Based on it, combining the wavelet networks with a kind method of model structure design and parameters estimation of nonlinear dynamic system, a novel wavelet networks training scheme considering both scale and precision of the networks was proposed. The scheme determined the node number of the networks automatically and resulted in an optimal network structure under BIC rule. The two numerical examples show that the wavelet networks are of small scale and good generalization performance.