混合递阶遗传算法的自适应小波神经网络优化设计

Optimum Design of Self-Adaptive Wavelet Neural Networks based on Hybird Hierarchy Genetic Algorithm

在研究自适应小波神经网络学习算法的基础上,提出了一种混合递阶遗传算法,与标准遗传算法相比,该算法不仅可以同时确定网络参数(连接权、尺度参数和平移参数),而且解决了网络拓扑结构的优化训练问题。仿真结果表明,该算法可以准确地搜索到自适应小波网络的网络参数和最优结构,并能大幅度提高学习效率,是切实可行的。

Based on the study of self-adaptive wavelet neural networks, a hybrid hierarchy genetic algorithm is proposed to training network. Compared with standard genetic algorithm, this method trains not only network parameters such as scale factor, transform factor parameter and connection weights, but also solves configuration problem of self-adaptive wavelet neural networks. The result of simulation indicates that the algorithm can efficiently determinate the parameter and structure of self-adaptive wavelet neural networks, and has better higher training efficiency and forecasting precision.