论文部分内容阅读
本文从非线性函数逼近的角度讨论了小波网络的优越性,其主要特点是具有局部学习和在多个尺度上学习的功能。一般情况下,由延迟序列重构的系统吸引子在重构相空间内的分布是不均匀的,对此,小波网络较全局学习网络(例如,反向传播(BP)网络)和局部学习网络(例如,径向基函数(RBF)网络)有理论及实际上的优势。简要讨论了小波网络的构造和小波函数的选取,最后用于上证指数的预测,取得令人满意的效果。
This paper discusses the superiority of wavelet networks from the perspective of approximation of nonlinear functions. Its main feature is the function of local learning and learning on multiple scales. In general, the distribution of the system attractors reconstructed by the delay sequence is nonuniform in the reconstructed phase space, and the wavelet network is more non-uniform than the global learning network (for example, back propagation (BP) network) and the local learning network (Eg Radial Basis Function (RBF) networks) have both theoretical and practical advantages. The structure of wavelet network and the selection of wavelet function are briefly discussed. Finally, it is used in the prediction of Shanghai Stock Index, and the satisfactory result is obtained.